Multimedia Communications over Next Generation Wireless Networks

Transcription

1 EURASIP Journal on Wireless Communications and Networking Multimedia Communications over Next Generation Wireless Networks Guest Editors: Liang Zhou, Athanasios V. Vasilakos, Laurence T. Yang, and Naixue Xiong

2 Multimedia Communications over Next Generation Wireless Networks

3 EURASIP Journal on Wireless Communications and Networking Multimedia Communications over Next Generation Wireless Networks Guest Editors: Liang Zhou, Athanasios V. Vasilakos, Laurence T. Yang, and Naixue Xiong

4 Copyright 21 Hindawi Publishing Corporation. All rights reserved. This is a special issue published in volume 21 of EURASIP Journal on Wireless Communications and Networking. All articles are open access articles distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

5 Editor-in-Chief Luc Vandendorpe, Université catholique de Louvain, Belgium Associate Editors T. D. Abhayapala, Australia Mohamed Hossam Ahmed, Canada Farid Ahmed, USA Carles Antón-Haro, Spain A. C. Boucouvalas, Greece Yuh Shyan Chen, Taiwan Pascal Chevalier, France C.-C. Chong, Republic of Korea Nicolai Czink, Austria Daniel Benevides da Costa, Brazil S. Dasgupta, USA Rodrigo C. De Lamare, UK Ibrahim Develi, Turkey Petar M. Djuric, USA Abraham Fapojuwo, Canada Michael Gastpar, USA A. B. Gershman, Germany Wolfgang H. Gerstacker, Germany David Gesbert, France Zabih F. Ghassemlooy, UK Jean-marie Gorce, France Fabrizio Granelli, Italy Christian Hartmann, Germany Stefan Kaiser, Germany George K. Karagiannidis, Greece Chi Chung Ko, Singapore Nicholas Kolokotronis, Greece Richard Kozick, USA Sangarapillai Lambotharan, UK Vincent Lau, Hong Kong D. I. Laurenson, UK Tho Le-Ngoc, Canada Tongtong Li, USA Wei Li, USA Lingjia Liu, USA Zhiqiang Liu, USA Steven McLaughlin, UK Sudip Misra, India I. Moerman, Belgium Marc Moonen, Belgium Sayandev Mukherjee, USA Kameswara Rao Namuduri, USA AmiyaNayak,Canada Monica Nicoli, Italy Claude Oestges, Belgium Ashish Pandharipande, The Netherlands Jordi Pérez-Romero, Spain Phillip Regalia, France George S. Tombras, Greece Athanasios V. Vasilakos, Greece Ping Wang, Canada Weidong Xiang, USA Xueshi Yang, USA Kwan L. Yeung, Hong Kong Fei Richard Yu, Canada W. H. Zhuang, Canada

6 Contents Multimedia Communications over Next Generation Wireless Networks, Liang Zhou, Athanasios V. Vasilakos, Laurence T. Yang, and Naixue Xiong Volume 21, Article ID 89641, 2 pages Cooperative Coding and Caching for Streaming Data in Multihop Wireless Networks,DanWang, Jiangchuan Liu, Qian Zhang, and Fajun Chen Volume 21, Article ID , 1 pages Converged Wireless Networking and Optimization for Next Generation Services,J.Rodriguez, V. Monteiro, A. Gomes, MarcoDi Renzo, Jesús Alonso-Zárate, Christos Verikoukis, Ainara Gonzalez, OscarLázaro, Ahmet Akan, Julian Pérez Vila, George Kormentzas, David Boixade, and Silvia de la Maza Volume 21, Article ID 92472, 11 pages New Trends on Ubiquitous Mobile Multimedia Applications,JoelJ.P.C.Rodrigues,MarcoOliveira, and Binod Vaidya Volume 21, Article ID , 11 pages Power-Aware DVB-H Mobile TV System on Heterogeneous Multicore Platform, Yu-Sheng Lu, Chin-Feng Lai, Chia-Cheng Hu, Han-Chieh Chao, and Yueh-Min Huang Volume 21, Article ID , 1 pages Embedding Protection Inside H.264/AVC and SVC Streams, Catherine Lamy-Bergot and Benjamin Gadat Volume 21, Article ID , 11 pages Quality-Assured and Sociality-Enriched Multimedia Mobile Mashup, Hongguang Zhang, Zhenzhen Zhao, Shanmugalingam Sivasothy, Cuiting Huang, and Noël Crespi Volume 21, Article ID , 11 pages Packet-Scheduling Algorithm by the Ratio of Transmit Power to the Transmission Bits in 3GPP LTE Downlink, Jungsup Song, Gye-Tae Gil, and Dong-Hoi Kim Volume 21, Article ID , 8 pages Cross-Layer Handover Scheme for Multimedia Communications in Next Generation Wireless Networks, Yuliang Tang, Chun-Cheng Lin, Guannan Kou, and Der-Jiunn Deng Volume 21, Article ID 3976, 1 pages A Dynamic Utility Adaptation Framework for Efficient Multimedia Service Support in CDMA Wireless Networks, Timotheos Kastrinogiannis and Symeon Papavassiliou Volume 21, Article ID 37541, 2 pages Adaptive Reliable Routing Based on Cluster Hierarchy for Wireless Multimedia Sensor Networks, Kai Lin, Min Chen, and Xiaohu Ge Volume 21, Article ID , 1 pages COSR: A Reputation-Based Secure Route Protocol in MANET, Fei Wang, Furong Wang, Benxiong Huang, and Laurence T. Yang Volume 21, Article ID , 1 pages

7 A Multimedia Application: Spatial Perceptual Entropy of Multichannel Audio Signals, Shuixian Chen, Ruimin Hu, and Naixue Xiong Volume 21, Article ID , 13 pages Novel Approaches to Enhance Mobile WiMAX Security, Taeshik Shon, Bonhyun Koo, Jong Hyuk Park, and Hangbae Chang Volume 21, Article ID , 11 pages A Speed-Adaptive Media Encryption Scheme for Real-Time Recording and Playback System,ChenXiao, Shiguo Lian, Lifeng Wang, Shilong Ma, Weifeng Lv, and Ke Xu Volume 21, Article ID , 9 pages ESVD: An Integrated Energy Scalable Framework for Low-Power Video Decoding Systems,WenJi, Min Chen, Xiaohu Ge, Peng Li, and Yiqiang Chen Volume 21, Article ID , 14 pages

8 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID 89641, 2 pages doi:1.1155/21/89641 Editorial Multimedia Communications over Next Generation Wireless Networks Liang Zhou, 1 Athanasios V. Vasilakos, 2 Laurence T. Yang, 3 and Naixue Xiong 4 1 Technical University of Munich, 8333 Munich, Germany 2 University of Western Macedonia, 51 Kozani, Greece 3 St. Francis Xavier University, Antigonish, NS, Canada B2G 2WS 4 Georgia State University, Atlanta, GA 332, USA Correspondence should be addressed to Liang Zhou, liang.zhou@ieee.org Received 16 December 21; Accepted 16 December 21 Copyright 21 Liang Zhou et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Recent advances in communication technologies have witnessed a growing and evolving multimedia-content-delivery market based on information gathering, manipulation, and dissemination. Unlike traditional communication systems, a fundamental challenge for next generation wireless networks is the ability to transport multimedia content over a variety of networks at different channel conditions and bandwidth capacities with various requirements of qualityof-service. There are many open problems for this challenging issue such as signal processing, collaborations, power management, flexible delivery, specialization of new content, dynamic access. This exciting special issue has received large submissions that covered all topics of wireless multimedia technology. Due to page budget and timing constraints, many good quality works have been turned away, and 15 papers have finally been selected after a careful and highly competitive review process. These papers are organized into three sections in this special issue, namely multimedia service, multimedia communications and multimedia information processing. The first set of six papers describes the emerging multimedia service technology. The first paper, Cooperative coding and caching for streaming data in multihop wireless Networks by D. Wang eta l. studies the distributed caching managements for the current flourish of the streaming applications in multihop wireless networks. The second paper, Converged wireless networking and optimization for next generation services by J. Rodriguez et al. provides the key achievement that has been tunneled into first prototypes for showcasing next generation services for operators and process manufacturers. The third paper, New trends on ubiquitous mobile multimedia applications byj.p.c. Rodrigues et al. tackles several important challenges such as communication cost and device limitations for development of ubiquitous multimedia applications. The fourth paper, Power-aware DVB-H mobile TV system on heterogeneous multicore platform by Y.-S. Lu et al. proposes a mobile TV system on a heterogeneous multicore platform which utilizes a DVB-H wireless network to receive the TV program signal. The fifth paper, Embedding protection inside H.264/AVC and SVC streams by C. Lamy-Bergot and B. Gadat, describes a backward compatible error protection mechanism embedded into the H.264. The sixth paper, Quality-assured and sociality-enriched multimedia mobile mashup byh.zhang et al. presents a metadata-based mashup framework in next generation wireless networks, which guarantees the quality and supports social interactions. The next set of five papers deals with multimedia communications problems. The first paper, Packet scheduling algorithm by the ratio of the transmit power to the transmission bits in 3GPP LTE downlink by J. Song et al. proposes a novel minimum transmit power-based packet scheduling thatcanachieve power-efficient transmission to the UEs. The second paper, Cross-layer handover scheme for multimedia communications in next generation wireless networks byy. Tang et al. combines session initiation, fast mobile IPv6 and media independent handover protocols. The third paper, A dynamic utility adaptation framework for efficient multimedia service support in CDMA wireless networks by K. Timotheos and S. Papavassiliou, introduces a novel generic framework that enables the dynamic adaptation of real-time multimedia

9 2 EURASIP Journal on Wireless Communications and Networking utilities. The fourth paper, Adaptive reliable routing based on cluster hierarchy for wireless multimedia sensor networks by K. Lin et al. proposes an adaptive reliable routing based on clustering hierarchy, which includes energy prediction and power allocation mechanism. The fifth paper, COSR: A reputation based secure route protocol in MANET byf.wang et al. proposes a cooperative on-demand secure route against malicious and selfish behaviors. The final set of four papers addresses multimedia information processing issues. The first paper, A multimedia application: spatial perceptual entropy of multichannel audio signals by S. Chen et al. builds a binaural cue physiological perception model on the ground of binaural hearing which represents spatial information in the physical and physiological layers. The second paper, Novel approaches to enhance mobile WiMax security by T. Shon et al. investigates the current Mobile WiMax security architecture focusing mainly on pointing out new security vulnerabilities. The third paper, A speed adaptive media encryption scheme for real-time recording and playback system byc.xiaoetal. proposes a novel adaptive media encryption scheme. The fourth paper, ESVD: an integrated energy scalable framework for low power video decoding systems by W. Ji et al. is dedicated to developing an energy-scalable video decoding strategy for energy limited mobile terminals. Acknowledgments The guest editorial team would like to thank all authors for submitting their quality work to this special issue and to the numerous reviewers expert contributions. Finally, special thanks go to Professor Luc Vandendorpe, Editor-in-Chief for approving and making this issue possible and to Professor Mariam Albert for her priceless guidance along the whole process of this special issue. Liang Zhou Athanasios V. Vasilakos Laurence T. Yang Naixue Xiong

10 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 1 pages doi:1.1155/21/ Research Article Cooperative Coding and Caching for Streaming Data in Multihop Wireless Networks Dan Wang, 1 Jiangchuan Liu, 2 Qian Zhang, 3 and Fajun Chen 4 1 Department of Computing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong 2 School of Computing Science, Simon Fraser University Burnaby, British Columbia, Canada V5A 1S6 3 Department of Computer Science, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong 4 Key Laboratory of Science and Technology for National Defence of Parallel and Distributed Processing, School of Computer, National University of Defense Technology, Changsha, Hunan, China Correspondence should be addressed to Dan Wang, csdwang@comp.polyu.edu.hk Received 25 February 21; Accepted 15 May 21 Academic Editor: Liang Zhou Copyright 21 Dan Wang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. This paper studies the distributed caching managements for the current flourish of the streaming applications in multihop wireless networks. Many caching managements to date use randomized network coding approach, which provides an elegant solution for ubiquitous data accesses in such systems. However, the encoding, essentially a combination operation, makes the coded data difficult to be changed. In particular, to accommodate new data, the system may have to first decode all the combined data segments, remove some unimportant ones, and then reencode the data segments again. This procedure is clearly expensive for continuously evolving data storage. As such, we introduce a novel Cooperative Coding and Caching (C 3 ) scheme, which allows decoding-free data removal through a triangle-like codeword organization. Its decoding performance is very close to the conventional network coding with only a sublinear overhead. Our scheme offers a promising solution to the caching management for streaming data. 1. Introduction Multihop wireless networks as wideband Internet access solutions have been widely researched nowadays, and promote some real deployments for communities [1 4]. Since then, the requirement for supporting streaming applications in such infrastructures becomes more and more imperative [5, 6]. Path caching is a common used technology in wired networks for delivering media streaming efficiently, which can reduce the client-perceived access delays as well as server/network loads [7]. In the wireless paradigm, for the broadcast nature of wireless transmission, it is more direct and intuitive to cache streaming data on the way. Many works have been done to exploit the caching benefits in multihop wireless networks [8 1]. In the caching-enhanced multihop wireless networks, single wireless node usually has limited space for caching and can only save part of streaming data. It is common for a client to fetch data segments from multiple relay nodes. In this way, the caching-enhanced multihop wireless networks as a whole can be treated as distributed storage systems. However, previous works spend little attentions to the caching management, and data are unevenly distributed in the networks. Usually, a sophisticated caching searching algorithm is needed [1]. Recently, network coding, in particular, random linear coding, has been suggested as an elegant solution for distributed storage managements [11 13]. In a network-coding-based storage system, the original data segments (say N) are combined through linear operations with independent coefficient vectors. The combined data segments are of the same size as the original segments and are equivalent in decodability. Each relay node can therefore record a subset of the combined data segments, and a client is able to decode all the original data segments as long as N combined data segments are retrieved. This combination process however makes the caching storage inflexible to change. More explicitly, since media data are usually coded as different important segments

11 2 EURASIP Journal on Wireless Communications and Networking before transmission for providing scalable streaming abilities [14], if unimportant data segments are to be removed to accommodate new data, the system needs to first decode all the data segments, remove the unimportant ones, and then re-encode the data segments again. This operation is timeand resource consuming [15]. Even worse, given that a node can only store a partial set of the data segments, it is generally impossible for each single node to carry out the decoding operation. To effectively solve this problem, we introduce a novel Cooperative Coding and Caching (C 3 )scheme.c 3 extends the conventional random network coding [16], and enables decoding-free data removal through a triangle-like codeword organization. Its decoding performance matches that of the conventional network coding with only a sublinear overhead. It also enables retrieval of only a subset of the data. As such, it offers a promising solution to the caching management of streaming data in multihop wireless networks. In this paper, we present the theoretical foundations of C 3 and a set of general rules for applying C 3 in the caching management. We then demonstrate the actual benefits of C 3 for streaming applications in multihop wireless networks. Again, we show that, while conventional network coding is capable to achieve high throughput, it cannot easily manage streaming data. The remainder of the paper is organized as follows. We first present the system model, and demonstrate the superiority and problems when directly applying network coding to the system in Section 2. InSection 3, weoffer the theoretical foundations of cooperative coding and caching. We discuss the design issues of C 3 -based cache management in Section 4. Our preliminary simulation results are shown in Section 5. In Section 6, we discuss the related work. Finally, Section 7 concludes the paper and presents some future directions. 2. Preliminaries 2.1. Model and Notation. We now give the a formal description of the system. Our caching model is quite similar with the caching model of Ditto [9]. The main difference is that we apply our novel coding schema to manage cached data. We consider a multihop wireless network of N nodes, each with a buffer of size B. Assume that we only want to cache totally N data segments for one session for the limited caching space. In general, we have N>B, that is, no single node can store all the data of a session. A media server located in the Internet which can be accessed by the clients of the multihop wireless network through gateway (GW). Media files are split into equally sized data segments. When a media file is first requested, there is no information about the media file in the network before; then the request will be sent to the Internet media server directly. A second or later requests will benefit from previous transmissions through caching. Requesting for the same media file in a community is a quite common user behavior as suggested by recent progresses in the traffic analysis and social networks. Wireless node in the network will cache all successfully received data segments. Data segment receiving can happen either when the node is on the routing path or when it is beside the path but can overhear the transmission. The system is continuously evolving with new streaming data coming and existing data being obsolete, though the total number of useful data segments is always N. Since user requests are probably random from all parts of the wireless network, it is reasonable to assume that data segments are randomly distributed inside the network. We do not assume any indexing service (no matter centralized or distributed) in the network. For a client to retrieve the N data segments stored in the wireless network, it will simply broadcast the request to the nearest M neighbor nodes, which are not necessarily exactly nearest for performance tradeoff. It can be easily realized by sending out M requests and one request can be processed by only one node. Every node has a proxy module just like what they have in the study by Ditto. Every proxy can serve the data to its previous hop, either from its local cache or by requesting it from its nexthop proxy [9]. Without loss of generality, we assume that each node will provide one data segment from its storage space. Clearly, to obtain all the N data segments, we must have M N, and even so, not all the data segments are necessarily obtained in such kind of retrieving scheme. Therefore, we define a success ratio, which serves as the major evaluation criterion in our study, as follows Definition 1 (successratio). The success ratio is the probability that a data retrieval scheme successfully retrieves all the N data segments. The default settings of M and B are M = N and B = 1, which are their lower bounds for valid schemes Network-Coding-Based Caching: Superiority and Problems. We now show that network coding, in particular, random linear coding, can significantly improve the success ratio over a codingless caching system. With random linear coding, all data segments are stored in a combined fashion. More specifically, assume that the original data segments are c j, j = 1, 2,..., N; a coded data segment f i (also referred to as a combined data segment) is generated as N 1 j= β j c j, where β = (β, β 1,..., β N 1 )isacoefficient vector, each item of which is randomly generated from a finite field F q.itis worth noting that the size of each f i remains equal to c j.we define the cardinality of f i to be the number of original data segments it contains, and the full cardinality of the system is the highest possible number, that is, N. To obtain the N original data segments, one can simply collect any N combined data segments and then decode through solving a set of linear equations. Here, a necessary condition for successful decoding is that the coefficient vectors must be linearly independent. This is generally true if the coefficient is generated from a large enough field size q [12]. As shown in [17], the probability of linear independency is over 99.6% for q = 2 8, and this is almost independent of N. As such, for the network-coding-based data storage and collection scheme, the success ratio with M = N and B = 1 is close to 1%.

12 EURASIP Journal on Wireless Communications and Networking 3 Besides the combined data segments, the coefficient vectors also have to be collected by the client for decoding. Such overheads, generally of a few bytes, are much lower than the data volume, and the benefit of network coding thus well dominates these costs. Unfortunately, while in conventional network coding it is easy to combine new data segments to existing data segments and increase the cardinality, the reverse operation is very difficult. Specifically, to remove an original data segment, we have to first decode the combined data segments, remove the unnecessary original data segments, and then recombine the remaining data segments. This is time and resource consuming. Even worse, it is often impossible to perform these operations in a single node given that B < N,as decoding requires N combineddata segments. As such, for the streaming application in multihop wireless networks, caching storage systems are continuously evolving; if we keep unimportant or obsolete data segments in the system, then the clients have to download more and more unnecessary data segments to successfully decode the expected useful data segments. Eventually, the buffer will overflow, and the system simply crashes. This becomes a key deficiency for applying conventional network coding in continuous data management. A related problem is that network coding has no flexibility in retrieving partial data sets only, for example, a set of the most important m original data segments that comprise the most important frames of a media file, where m<n. 3. Cooperative Coding and Caching: Basic Idea In this section, we show a new coding scheme that conveniently solves the problem of data removal, thus facilitating caching management for streaming data. Our coding scheme enables the combination of only part of the original data segments, and we refer to it as Cooperative Coding and Caching (C 3 ); compre for example Network Coding (NC)and no coding at all (Non-NC ) Overview of Cooperative Coding and Caching Code words. In C 3, instead of having full cardinality only, the combined data segments may have different cardinalities, from 1 to N. In addition, for each original data segment c i,wherei denotes the time sequence, there is a weight w i associated to this data. The definition of weight in our streaming application combines the time stamp of the data segment and its relative importance to the media playback performance. Figure 1 shows an example of our weight assignment schema for a video clip. In the figure, data segments are classified into I, P, B three categories, which correspond to the I-frame, P-frame, and B-frame of the video to be streamed. Generally speaking, I-frame is the most important data for video playback, then the P- frame, and then the B-frame. So, in a batch, data segments corresponding to the I-frame are assigned with the highest weights. w x + I 11 B B 1 P 8 B 2 B 3 Figure 1: Demo of weight assignment for streaming data starting from w x. P 9 B 4 f = [c 3, c 2, c 1, c ], f 1 = [c 3, c 2, c 1 ], f 2 = [c 3, c 2 ], f 3 = [c 3 ]. Figure 2: An example of C 3 for N = 4 and weight w 3 w 2 w 1 w. The time index is not shown in this example. It is worth noting that the weight assignment schema can be application specific, and our C 3 solution can be applied to all kinds of weight definitions which respect the following constraint. The only constraint is an operator, which should follow the following. (1) (Transitive) forany w i, w j, w k,ifw i w j and w j w k, then w i w k ;and(2) for every w i, w j, either w i w j or w i w j. We have the following convention in this paper: w j w j if j>j and we say that the weight of c j is higher than c j if j>j,and c j is a more recent data segment than that c j if j>j.for simplicity, we will use only the subscript or superscript for a data segment if the context is clear. Thus, for a data segment. with weight index i and time index j,weusec i = c j to denote that they are the same data segment, that is, c j i. For original data segments c, c 1,..., c N 1, we have a coding base B = { f k f k = N 1 j=k β j c j, k [,..., N 1], β j F q }.We omit β j in the following and use f k = [c N 1, c N 2,..., c k ]for ease of exposition. The coding base for N = 4isillustratedin Figure 2. The storage for each node is S ={fi k fi k B, i B 1}. Intuitively, each node stores a few combined data segments selected from the coding base B. Notice that, if k denotes the cardinality of a combined data segment, then the cardinality of f k can be calculated by k = N k. The cardinality difference of f k1 and f k2 is k 1 k 2. We may drop the superscript and use f i provided that k is clear in the context to represent the ith combined data segment in this node. A salient feature of this triangle-like coding scheme is the decoding-free data removal; that is, to remove the current least important original data segment c in the system, we can simply drop the combined data segments containing c. Intuitively, c is included in the combined data segments B 5 P 1 B 6 B 7 I

13 4 EURASIP Journal on Wireless Communications and Networking with the highest cardinality. The amount of these combined data segments consists of only a fraction of the combined data segments in the system and a removal of them will not adversely affect the success ratio of the system (recall that, with conventional network coding, all combined data segments have to be deleted in this case). Figure 3 illustrates a concrete example, where the new data segment c 4 is to replace data segment c. To simplify our example, we assume. that c 4 = c 4, that is, c 4 also has the highest weight. We see that, after data replacement, all f = [c 3, c 2, c 1, c ] are replaced by f = [c 4 ], and the cardinalities of other combined data segments increase by one. This data replacement operation is decoding free and the system remains stable. A general observation can be drawn as follows. Observation 1. For original data segments c i and c j,wherei> j (i.e., w i w j ), c i will be deleted no earlier than c j from the system. This is true irrespective to any data removing scheme. Once we have data cached in the wireless subnet, two kinds of data request routing protocols can be applied depending on whether the data segments have been cached or not. First, when the requested segments can be retrieved inside the wireless network, then m data requests will be broadcasted to its neighbors. Every neighbor can only reply to one request, and has to decide whether to forward the request or just reply to it depending on its history record. Second, when the requested segments cannot be retrieved inside the wireless network, any unicast routing protocol in multihop wireless network can be used. We use the OSLR routing protocol in our model in this case. Though nodes on the route will cooperatively cache data segments when the traffic comes from outside of the wireless network, they will only respond to data requests on demand of the sink node. And packets will travel along the reverse route path of the request Distribution of Cardinality. Cooperative coding and caching intrinsically manages the shape (i.e., cardinality) of the combined data segments. It is not difficult to see that, for a client retrieval, if the contacted nodes provide combined data segments with high cardinalities, then the success ratio will be higher. We summarize this in the following two observations. Observation 2. The success ratio is maximized if every node provides the client the combined data segment with the highest cardinality from its buffer. Observation 3. Consider time instances t 1 and t 2.Ifatt 1 the probability for each node to provide high-cardinality data segments is greater than at t 2, then success ratio for a data retrieval at t 1 is higher than that at t 2. Generally speaking, in each particular time instance, it is ideal for the system to have combined data segments of cardinalities as high as possible. In an extreme case, if all the combined data segments in the system have cardinality N, the success ratio is 1% but the system is essentially reduced s : { f = [c 3, c 2, c 1, c ], f 1 = [c 3, c 2 ] }, s 1 : { f = [c 3, c 2, c 1 ], f 1 = [c 3 ] }, s 2 : { f = [c 3, c 2, c 1 ], f 1 = [c 3, c 2 ] }, s 3 : { f = [c 3, c 2 ], f 1 = [c 3 ] }, s 4 : { f = [c 3, c 2, c 1, c ], f 1 = [c 3 ] }, s 5 : { f = [c 3, c 2, c 1 ], f 1 = [c 3, c 2 ] }. (a) C 3 before using c 4 to replace c s : { f = [c 4 ], f 1 = [c 4, c 3, c 2 ] }, s 1 : { f = [c 4, c 3, c 2, c 1 ], f 1 = [c 4, c 3 ] }, s 2 : { f = [c 4, c 3, c 2, c 1 ], f 1 = [c 4, c 3, c 2 ] }, s 3 : { f = [c 4, c 3, c 2 ], f 1 = [c 4, c 3 ] }, s 4 : { f = [c 4 ], f 1 = [c 4, c 3 ] }, s 5 : { f = [c 4, c 3,c 2, c 1 ], f 1 = [c 4, c 3, c 2 ] }. (b) C 3 after using c 4 to replace c Figure 3: Data cached in 6 wireless nodes (s through s 5 )each with two buffer units. We omit the coefficients for each combined data segment for ease of exposition; however, it should be known that f s is not the same as f s 4 as they are coded with different coefficient vectors. (a) shows C 3 before insertion of c 4,and (b) shows C 3 after the insertion of c 4 and removal of c. After this data replacement operation, the system is fairly stable in terms of the cardinalities of the combined data segments. to the case of conventional network coding. Once a new data segment arrives, all the combined data segments will have to be deleted to make room for the new segment. For subsequent client retrievals, no data segment but the newest one can be answered. As said, client requests could come from any part of the wireless networks, so it is reasonable to assume that cached data are uniform distributed inside the network storage. We now consider the performance of C 3 with a uniform cardinality distribution throughout the lifetime of the system. This distribution does not favor data arrivals or client retrieval at any particular time, and can serve as a baseline for further application-specific optimizations Performance Analysis of C 3. With the uniform distribution, we can focus on the success ratio of a single client data retrieval. Since the cardinality of each data segment is not necessarily N in C 3, it is possible that the success ratio is less than 1%. Let B = 1 for each node and the success ratio for obtaining i data segments by collectingi data segments using C 3 be F(i). We quantify the success ratio in this scenario as follows. Theorem 2. Define F() = 1, then the success ratio ( ) 1 N N 1( ) N F(N) = F(i)i N N i i. (1) i=

14 EURASIP Journal on Wireless Communications and Networking 5 Proof. The basic idea of our proof is to find the sets of combined data segments that are decodable, referred to as valid sets. For example, a set of combined data segments with cardinality 1 through N are decodable, and the set of combined data segments with all cardinalities being 1 is not decodable. The success ratio is given by the number of valid sets over the total number of possible sets of combined data segments; the latter is N N. A valid set can be constructed as follows: pick N i combined data segments of cardinality N and i combined data segments with cardinality less or equal to i. These i combined data segments should be a valid set (decode-able) in terms of i. For example, if N = 4, a valid set may consist of two combined data segments of cardinality 4 and two combined data segments with cardinality less than or equal to 2. The number of the latter is F(i)i i. Since the retrieval can be of any sequence,we need to fit all these combined data segments into N pickup sequences. For those N i combined data segments with cardinality N, there are ( ) N N i locations to fit in. Notice that we do not need to shuffle the i combined data segments with smaller cardinalities as F(i)i i has already contained all possible shuffles. Therefore, the total number of valid sets is N 1( N ) i= N i F(i)i i after summing up all i [, N 1], and the theorem follows. We further derive an upper bound and lower bound for C 3. Theorem 3. The upper and lower bounds of the success ratio are 1 ((N i)/n) N and N 1 i= ((N i)/n),respectively. Proof. The success ratio is upper bounded by successfully obtaining the combined data segments with the highest cardinality. This probability is 1/N due to the uniform distribution. The probability of not getting it in N picks is ((N i)/n) N, giving an upper bound 1 ((N i)/n) N. The success ratio of getting a set of combined data segment with cardinality 1 through N is (1/N) N N!, which is equal to N 1 i= ((N i)/n). This is clearly a lower bound. For comparison, we also calculate the success ratio of Non-NC, as follows. Observation 4. If the distribution of the data segments is uniform, the success ratio for obtaining all N data segments by randomly collecting N data segments (Non-NC) is N 1 i= ((N i)/n). A preliminary comparison of the success ratios can be found in Figure 4. Detailed comparisons as well as practical enhancements which substantially improve the success ratio will be presented in the following sections. The probabilistic nature of success ratio only provides us a rough idea of how many combined data segments should be retrieved. If decoding cannot be carried out after retrieving a certain number of combined data segments, then the clients have to retrieve additional data segments. This may not be acceptable for delay-sensitive applications. An ideal case is thus to inform the client of exactly how many combined data Success ratio Non-NC C 3 Figure 4: Success ratio as a function of N (in default value M = N and B = 1). segments should be retrieved so that they are guaranteed to decode out the necessary information. Before we give a concrete solution, we first illustrate the basic idea to achieve this. As shown in Observation 2, if each node sends a combined data segment with the highest cardinality in its buffer, the success ratio will be improved. In addition, if a node can always provide a combined data segment that has sufficiently high cardinality, then the success ratio will also be improved. We quantify these observations as follows. For each node, if it has a buffer size of N + 1, then it is able to store the data segments with cardinality starting from 1 to N + N, that is, the cardinality difference of f i and f i+1 is N for all i [, N]. Notice that we have extended the maximum cardinality of the system to N + N. At anytime the node can provide a data segment with cardinality in [N, N + N]. Consequently, if the client queries N + N nodes, and each node provides its combined data segment with cardinality no less than N, there will be N + N linear equations with the rank of the coefficient matrix being at least N and at most N + N. Thus, the system can guarantee a successful decoding of all N original data segments. Formally speaking, we have the following. Theorem 4. The success ratio of C 3 with B = N +1and M = N + N is identical to the success ratio of NC with B = 1 and M = N, that is, 1%. Proof. A rigorous proof can be found in [18]. This theorem shows that the success ratio of C 3 is identical with NC, with only a sublinear sacrifice of buffer size and retrieval cost. Yet it achieves continuous data management. This theorem also gives us important information; that is, to improve the success ratio of C 3, it is also important to have a balanced cardinality distribution within each single node, as each node is able to provide a combined data segment with fairly high cardinality at any time. N

15 6 EURASIP Journal on Wireless Communications and Networking 3.4. Maintaining Uniform Distribution of Cardinality. We have shown the performance of C 3 under a uniform cardinality distribution. It remains to show how a uniform distribution can be achieved and maintained in this dynamically evolving system. At first, we need to initial the caching system to have uniform distributed cardinality. And this can be done easily in a centralized way. For example, we take the gateway node as the initialization control node. If there are P nodes in the network willing to cache the application data, then each of them will provide its reserved capacity to the gateway. Say that if we have B P units for caching, then the gateway will use a random generator to uniformly distribute cardinality vector to each node. And the overhead could be very low, since we only have to send P packets, each of which has the payload of B log 2 N bits. The overhead can be further decreased, if we piggyback those cardinality initialization vectors to some real data packets. After initialization, we will apply a simple data replacement strategy to maintain this uniformity.. Assume that a new data segment c i = c j is generated to replace c. Our data replace procedure is as follows. (1) Remove f = [c N 1,..., c ]. (2) For each node containing f i+1 = [c N 1,..., c i+1 ], duplicate f i+1 and combine f i+1 with c i to produce new f i = [c N 1,..., c i ]. (3) For each node containing f i k, k = 1...i 1, combine c i with f i k.(4)send f i to the nodes which just removed f. duplicate additional f i to send if there are more nodes with f.andfinally(5) delete unnecessary f i if there are fewer nodes with f to send. Theorem 5. If the cardinality is uniformly distributed, then after the above data replacement procedure, the distribution of the cardinality remains uniform, and the number of combined data segments stored in each node remains unchanged. Proof. The above procedure uses the new combined data segment to replace the deleted f. Therefore, the set of nodes which removed f was refilled by f i. The set of nodes which duplicated combined data segment f i either sent this data segment out or discarded it. Thus, the number of combined data segments stored in each node is unchanged. If the distribution of the cardinality is uniform, then the probability that a combined data segment has cardinality k is 1/N for all k = 1...N. After the data replacement with new data segment c i, the probability of a combined data segment having cardinality 1 < k < i remains unchanged. The probability of a combined data segment having cardinality i is the same as that having cardinality N before data replacement, which is 1/N. The probability of a combined data segment having cardinality i + k, 1 k N i is the same as that having cardinality i + k +1beforedata replacement, which again is 1/N. Thus the distribution of cardinality remains uniform. Depending on applications, one may choose other, maybe simpler, data replacement schemes to maintain the uniformity. For example, if the new data segment is always the one with the highest weight, then new data segment can Algorithm Data Replacement (c j ) c j : new data segment; estimate w(c j. ); c i = c j if kw(c j ) <w(c k ), delete c j for l = 1...B if cardinality ( f l ) <N, if ( f l ) contains c i 1, f l = β i c i + f l ; else if fl i+1 = [c N 1,..., c i+1 ], duplicate fl i and save as fl i+1 fl+1 i = f l i + β i c i if f i whose cardinality is greater than N delete f i Algorithm 1: Pseudocode for data replacement. automatically replace f and no transmission is necessary (seethe example infigure 3). We have yet to show how the uniform/balanced cardinality in each node can be maintained. In a naive case, the nodes just exchange the combined data segments with others to make their cardinality balanced. This introduces high transmission overhead. Although in some applications the paramount objective is high success ratio for each client access [13], we will show some optimization schemes to substantially reduce this overhead in the next section. 4. C 3 for Caching Management We now detail the caching management through C 3 in multihop wireless networks. The management behavior mainly contains two operations: one is the data replacement, and the other is the data retrieval Data Replacement. When a new data segment c j is successfully received by a node, the node will perform the data replacement algorithm, as shown in Algorithm 1. This algorithm maintains the uniform distribution of cardinality in the entire network while replacing the less important original data segment by new data segment c j if there is no available space. As mentioned previously, we need to distribute some combined data segments to balance the distribution of cardinality in node level. Consider that we need a strictly balanced distribution; that is, the cardinality difference between two neighboring combined data segments f i and f i+1 is strictly N in a node. Since combining the new data segment will make some of the data segments have a difference of N + 1, a naive scheme will have to adjust the data segments in the entire system. This overhead is incurred each time a new data segment arrives. To avoid it, we propose the following alternative scheme. In this new strategy, a valid cardinality difference of f i and f i+1 is not strictly N, but in between (1/2) N and 2 N. Therefore, to balance the cardinality distribution in each node, we only need to transmit data segments in two scenarios given as follows. (1) If the cardinality difference

16 EURASIP Journal on Wireless Communications and Networking 7 between two combined data segments is greater than 2 N, then the node just needs to obtain one combined data segment that can fit in the gap from other node. And (2) if the cardinality difference between two combined data segments is less than (1/2) N, then the node just need to send out the data segment of cardinality in the middle. Since the cardinality difference of the data segments is doubled, if each node has a buffer size of 2 N, it is easy to redefine the cardinality difference and still guarantee that the cardinality difference in each node is at most N using the new scheme. Formally speaking, we have the following. Success ratio Corollary 6. Given a buffer space 2 N,andthenumberof nodes contacted to be N + N, C 3 with the new caching scheme in each node still achieves the same success ratio Number of node queried (M) The gaps of (1/2) N and 2 N are not unique. They can be generalized to (1/k) N and k N or other numbers depending on the buffer size on each node. In a practical heterogenous system, if O( N)buffer size cannot be guaranteed, several nodes can collaborate as a cluster to form a combined buffer and provide data segments with reasonably high cardinality Partial Data Retrieval. While we have focused on retrieving N data segments, our C 3 is indeed flexible in retrieving a subset of the segments and with different quality of services. Specifically, for a client that wants to retrieve the most important m data segments (m N), there are two possible services available. (1) Guaranteed Service. The client contacts m + m nodes and send a request message to each of them. This request message includes the type of the service (guaranteed service) and m. Each queried node will send back a data segment with the highest cardinality no greater than m + m.(2) Probabilistic Service. The client sends retrieval request message to m nodes. This request message includes the type of service (probabilistic service). The nodes queried will send back combined data segment with cardinality no less than m. While downloading the combined data segments, the client simultaneously performs decoding operations. If it cannot decode out the original data segments, it will send additional requests to other nodes for additional combined data segments. This will help with optimizing the cost if there is no stringent delay requirement. 5. Simulation Results In this section, we present our preliminary simulation results for C 3 -based caching data management. We deploy 1 static wireless nodes in the system. The default number of data segments to be cached for a source is N = 5 and the default buffer size B allocated for that source is 1. We examine other possible values in our simulation as well. The linear equations in network coding are solved using the Gaussian Elimination [19], and the coefficient field is q = 2 8,whichcanbeefficiently implemented in a 8-bit or more advanced microprocessor [2]. To mitigate C 3 Non-NC Figure 5: Success ratio as a function of M for C 3 and Non-NC. Number of data segement collected Number of required data (N) Non-NC C 3 Figure 6: Number of combined data segment retrieved as a function of N. randomness, each data point in a figure is an average of 1 independent experiments. Figure 5 shows the success ratio as a function of the number of data segments retrieved (N). Not surprisingly, the success ratio increases when N increases for both C 3 and Non-NC, but the improvement C 3 is more substantial. For example, if 1 data segments are to be retrieved, the success ratio is about 8% for C 3 ; for Non-NC, after retrieving 2 data segments, the success ratio is still 4% only. In our next experiment, the client will first randomly retrieve 5 data segments, and if some original data segments are missing (for Non-NC) or the combined data segments cannot be decoded (for C 3 ), then the client will send additional requests one by one, until all the 5 data segments are obtained by successful decoding. The results are shown in Figure 6. It is clear that C 3 outperforms Non-NC for different

17 8 EURASIP Journal on Wireless Communications and Networking Success ratio.6.4 Success ratio Number of node queried (M) B = 1 B = 2 B = 3 Figure 7: Success ratioas afunctionof M withdifferent buffer size Ratio between number of collections and cardinality N = 2 N = 5 N = 1 Figure 8: Success ratio as a function of λ = M/N. N s. It can be seen that the number of data segment collected is linear with respect to the number of required data (N), but the slope for C 3 is much smaller. As a result, when N is greater than 5, the cost with Non-NC is 3 to 4 times higher than that with C 3. We then increase the buffer size from B = 1to2and 3 to investigate its impact. We require the nodes to provide the data segment of the highest cardinality for each client access. By carefully managing the buffer of each node, the cardinality of the combined data segments each node could provide is no less than N/2 and2n/3 forb = 2andB = 3, respectively. The results are shown in Figure 7, where a buffer size expansion from 1 to 2 has a noticeable improvement in success ratio, and a buffer of 3 segments delivers almost optimal performance. This is not surprising because there is a higher degree of freedom for storing and uploading data in a larger buffer space. Notice that, on the contrary, the performance of Non-NC will not improve with the buffer size expansion as the nodes are unaware of which original data segments other nodes will provide to the client. We further explore the impact of the cardinality N. In Figure 8, we depict the decoding ratio for different number of original data segments (N = 2, 5, and 1). The x-axis denotes the ratio between the number of data collected and the cardinality, that is, λ = M/N. WecanseefromFigure 8 that their differences are insignificant, and generally reduced when M increases. Recall that the performance of Non-NC decreases sharply when N increases, while NC is marginally affected by N only. These simulation results thus reaffirm that C 3 inherits the good scalability of NC. 6. Related Work Proxy caching for media streaming over wired networks has been exploited for a long history [7]. Nowadays, as the rapid developmentofmultihopwirelessnetworkssuchaswireless mesh networks [1], proxy caching for media streaming in multihop wireless networks has also been discussed [7, 9, 1]. Different caching mechanisms have been proposed. However, previous works seldom pay attention to the caching data management. This paper takes advantages of coding to facilitate the caching management for streaming data in multihop wireless networks. Network coding was first introduced in [21] to improve multicast throughput. To maximize the benefit of network coding, the linear codes should be constructed carefully such that the output at each destination is solvable. Randomized network coding was introduced in [16], which adopts randomly generated coefficient vectors, thus making the calculation of alphabet decentralized. There are numerous recent studies applying conventional network coding and/or random linear coding in piratical systems. Examples include network diagnosis [13], router buffer management [22], energy improvement in wireless networks [23], data gossiping [24], as well as data dissemination in sensor networks [11] and in peer-to-peer networks [15]. For storage systems, various erasure codes have long been employed [25, 26]. Yet most of them require a central server for code block calculation; random linear coding is thus suggested for distributed storage [12]. A careful comparison between no coding, erasure codes, and random linear codes can be found in [12]aswell. While many of the studies have faced the problem of continuous data management, for example, in [2, 22, 27], their common solution is to cut the data flow in generations, that is, time periods, and combine all the original data segments in one generation. The length of a generation depends on the application and the choice is often experience based. Our study on cooperative coding and caching extends the network coding design from a different aspect, namely, decoding-free data removal. We solve this problem by a novel triangle-like coding method, which largely inherits the power of network coding, and yet well matches the demands

18 EURASIP Journal on Wireless Communications and Networking 9 from continuous data management. While the cardinalitymaximized combination process in network coding is the major source that improves its efficiency, our results show that the opposite direction, encoding only partial set of data segments, is worth consideration as well. Our initial study on C 3 with an application on data collection in sensor networks was presented in [17]. This paper extends [17] in the following aspects. First, we introduce the weight factor for the original data segments. This offers great flexibility in managing nonhomogeneous data, and in particular, data in a time series as shown in this paper. Second, this paper offers a more in-depth discussion on the effect of the cardinality distribution, which is an important factor for the system to evolve efficiently and effectively. Finally, we apply the C 3 in the caching management of streaming data in multihop wireless networks. 7. Conclusion and Future Work In this paper, we introduced a novel solution for caching management of streaming data in multihop wireless networks, Cooperative Coding and Caching (C 3 ), which effectively solves the problem of removing obsolete information in coded data segments. The problem is a major deficiency of the conventional network coding. We provided general design guidelines for C 3 and presented a theoretical analysis of its performance. We then addressed a series of practical issues for applying C 3 in the streaming applications under multihop wireless networks. Our C 3 offers a new look into the random linear combination process in conventional network coding. In network coding research, it is known that the higher the cardinality is, the more the benefits one may expect. Therefore, many existing schemes have focused on achieving a full cardinality in data combination. For example, the proposals in [11, 12, 15, 24] generally increase the cardinality by combining as much data as possible in intermediate nodes and then forward to others. Our work on cooperative coding and caching, however, shows that the opposite direction is worth consideration as well. Nevertheless, there are still many unsolved issues for C 3. Theoretically, we suspect whether the overhead of N reaches the potential limit of C 3? Practically, we need more measurement works to examine the performance of C 3, such as estimating the expected delay and buffering time for node that is sending the request. At least, considering the recent flourish of data streaming in numerous fields, we believe that C 3 may be applied in many related applications beyond caching management for steaming in wireless networks. Acknowledgments D. Wang s work is supported by Grants Hong Kong PolyU/G- YG78, A-PBR, A-PJ19, 1-ZV5W, and RGC/GRF PolyU 535/8E. References [1] I. F. Akyildiz, X. Wang, and W. Wang, Wireless mesh networks: a survey, Computer Networks, vol. 47, no. 4, pp , 25. [2] M. Conti and S. Giordano, Multihop ad hoc networking: the reality, IEEE Communications Magazine, vol. 45, no. 4, pp , 27. [3] M. Conti and S. Giordano, Multihop ad hoc networking: the theory, IEEE Communications Magazine, vol. 45, no. 4, pp , 27. [4] A. Aziz, A. El Fawal, J. Y. Le Boudec, and P. Thiran, Azialanet: deploying a scalable multi-hop wireless testbed platform for research purposes, in Proceedings of the ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc 9), pp , New York, NY, USA, 29. [5] H.-P. Shiang and M. van der Schaar, Multi-user video streaming over multi-hop wireless networks: a distributed, cross-layer approach based on priority queuing, IEEE Journal on Selected Areas in Communications, vol. 25, no. 4, pp , 27. [6] D. Wu, S. Ci, H. Luo, H. Wang, and A. K. Katsaggelos, Application-centric routing for video streaming over multihop wireless networks, in Proceedings of the 6th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON 9), pp. 1 9, June 29. [7] J. Liu and J. Xu, Proxy caching for media streaming over the internet, IEEE Communications Magazine, vol. 42, no. 8, pp , 24. [8] Y. Zhu, W. Zeng, H. Liu, Y. Guo, and S. Mathur, Supporting video streaming services in infrastructure wireless mesh networks: architecture and protocols, in Proceedings of the IEEE International Conference on Communications (ICC 8), pp , May 28. [9] F. R. Dogar, A. Phanishayee, H. Pucha, O. Ruwase, and D. G. Andersen, Ditto: a system for opportunistic caching in multihop wireless networks, in Proceedings of the 14th Annual International Conference on Mobile Computing and Networking (MobiCom 8), pp , ACM, New York, NY, USA, September 28. [1] F. Xie and K. A. Hua, A caching-based video-on-demand service in wireless relay networks, in Proceedings of International Conference on Wireless Communications and Signal Processing (WCSP 9), pp. 1 5, November 29. [11] A. G. Dimakis, V. Prabhakaran, and K. Ramchandran, Ubiquitous access to distributed data in large-scale sensor networks through decentralized erasure codes, in Proceedings of the 4th International Symposium on Information Processing in Sensor Networks (IPSN 5), pp , April 25. [12] S. Acedanski, S. Deb, M. Médard, and R. Koetter, How good is random linear coding based distributed networked storage, in Proceedings of the Workshop on Network Coding, Theory and Applications, 25. [13] C. Wu and B. Li, Echelon: peer-to-peer network diagnosis with network coding, in Proceedings of the 14th IEEE International Workshop on Quality of Service (IWQoS 6), pp. 2 29, June 26. [14] M. Ouaret, F. Dufaux, and T. Ebrahimi, Error-resilient scalable compression based on distributed video coding, Signal Processing: Image Communication, vol.24,no.6,pp , 29. [15] C. Gkantsidis and P. R. Rodriguez, Network coding for large scale content distribution, in Proceedings of the 24th

19 1 EURASIP Journal on Wireless Communications and Networking IEEE Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM 5), vol. 4, pp , Miami, Fla, USA, March 25. [16] T. Ho, M. Medard, J. Shi, M. Effros, and D. Karger, On randomized network coding, in Proceedings of the 41st Annual Allerton Conference on Communication, Control, and Computing, pp. 11 2, 23. [17] D. Wang, Q. Zhangt, and J. Liu, Partial network coding: theory and application for continuous sensor data collection, in Proceedings of the 14th IEEE International Workshop on Quality of Service (IWQoS 6), pp , June 26. [18] D. Wang, J. Liu, and Q. Zhang, Partial network coding for continuous data manage- ment in storage networks, Tech. Rep., School of Computing Science, Simon Fraser University, August 26. [19] J. Gentle, J. Chambers, W. Eddy, and S. Sheather, Numerical Linear Algebra for Applications in Statistics, Springer, New York, NY, USA, [2] J. Widmer and J.-Y. Le Boudec, Network coding for efficient communication in extreme networks, in Proceedings of the ACM Workshop on Delay Tolerant Networking and Related Networks (WDTN 5), pp , Philadelphia, Pa, USA, August 25. [21] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung, Network information flow, IEEE Transactions on Information Theory, vol. 46, no. 4, pp , 2. [22] S. Bhadra and S. Shakkottai, Looking at large networks: coding vs. queueing, in Proceedings of the 25th IEEE International Conference on Computer Communications (INFOCOM 6), pp. 1 12, April 26. [23] Y. Wu, P. A. Chou, and S.-Y. Kung, Minimum-energy multicast in mobile ad hoc networks using network coding, IEEE Transactions on Communications, vol. 53, no. 11, pp , 25. [24] S. Deb, M. Médard, and C. Choute, Algebraic gossip: a network coding approach to optimal multiple rumor mongering, IEEE Transactions on Information Theory, vol. 52, no. 6, pp , 26. [25] R. Blauht, TheoryandPracticeofErrorControlCodes, Addison Wesley, London, UK, [26] D. Costello and S. Lin, Error Control Coding: Fundamentals and Applications, Prentice Hall, Englewood Cliffs, NJ, USA, [27] P. Chou, Y. Wu, and K. Jain, Practical network coding, in Proceedings of the 41st Annual Allerton Conference on Communication, Control, and Computing, pp. 4 49, 23.

20 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID 92472, 11 pages doi:1.1155/21/92472 Research Article Converged Wireless Networking and Optimization for Next Generation Services J. Rodriguez, 1 V. Monteiro, 1 A. Gomes, 2 Marco Di Renzo, 3 Jesús Alonso-Zárate, 4 Christos Verikoukis, 4 Ainara Gonzalez, 5 Oscar Lázaro, 5 Ahmet Akan, 6 Julian Pérez Vila, 7 George Kormentzas, 8 David Boixade, 9 and Silvia de la Maza 1 1 Instituto de Telecomunicações, Campus Universitário de Santiago, Aveiro, Portugal 2 Portugal Telecom Inovação, S.A., Rua Eng. José Ferreira Pinto Basto, Aveiro, Portugal 3 Laboratory of Signals and Systems (LSS), French National Center for Scientific Research (CNRS), École Supérieured Électricité(SUPÉLEC), rue Joliot-Curie 3, Gif-sur-Yvette, Paris, France 4 Centre Tecnològic de Telecomunicacions de Catalunya (CTTC), Parc Mediterrani de la Tecnologia (PMT), Building B, Av. Carl Friedrich Gauss 7, 886 Castelldefels, Barcelona, Spain 5 CBT Communication and Multimedia, 4893 Getxo (Vizcaya), Spain 6 Turkcell Iletisim Hizmetleri A.S., Telco Product Strategy and Research, Turkcell Plaza Mesrutiyet Cad. 71, 3443 Tepebasi, Istanbul, Turkey 7 Telefonica I+D, C/ Emilio Vargas 6, 2843, Madrid, Spain 8 NCSR Demokritos, Terma Patriarchou Grigoriou, 1531 Aghia Paraskevi, Athens, Greece 9 IQUADRAT, Passeig Sant Joan 89, local 3, 89 Barcelona, Spain 1 TRIMEK, Camino de la Yesera no. 2, 1139 Altube-Zuia, Spain Correspondence should be addressed to J. Rodriguez, jonathan@av.it.pt Received 26 February 21; Accepted 6 June 21 Academic Editor: Liang Zhou Copyright 21 J. Rodriguez et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Next Generation Network (NGN) vision is tending towards the convergence of internet and mobile services providing the impetus for new market opportunities in combining the appealing services of internet with the roaming capability of mobile networks. However, this convergence does not go far enough, and with the emergence of new coexistence scenarios, there is a clear need to evolve the current architecture to provide cost-effective end-to-endcommunication. The LOOP project, a EUREKA- CELTIC driven initiative, is one piece in the jigsaw by helping European industry to sustain a leading role in telecommunications and manufacturing of high-value products and machinery by delivering pioneering converged wireless networking solutions that can be successfully demonstrated. This paper provides an overview of the LOOP project and the key achievements that have been tunneled into first prototypes for showcasing next generation services for operators and process manufacturers. 1. Introduction The NGN vision is tending towards a diverse wireless networking world where scenarios define that the user will be able to effectively attain any service, at any time on any network that is optimized for the application at hand. An important architectural issue is that of defining a next-generation wireless system, which acts as a networkof-wireless-networks accommodating a variety of radio technologies and mobile service requirements in a seamless cost-effective manner. The convergence of internet and mobile services is currently being addressed by the IMS (IP Multimedia Subsystems) platform, driven mainly by the operators and service providers to address market opportunities in combining the appealing services of internet with the roaming capability of mobile networks. But this convergence does not go far enough, and with the emergence of new coexistence scenarios, there is a clear need to evolve

21 2 EURASIP Journal on Wireless Communications and Networking the current architecture to provide cost-effective end-toend communications. This will raise significant research challenges and, undeniably, system coexistence solutions to address WAN (Wireless Area Networks), and LTE (Long- Term Evolution RAN) interoperability (Figure 1), and their impact on the 3GPP SAE (System Architecture Evolution) and IMS architectures require further innovation to align with future wireless trends and deliver new market opportunities for all players in the supply chain. Under the umbrella of converged services and networks, LOOP technology is targeting potential applications in the wireless market for process manufacturing. This market is expected to grow at a pace neighbouring 3% per year; faster than the wired contingent. Nevertheless, adoption of wireless technology is still low and most managers are reluctant to introduce radio solutions; key impediments being latency and performance issues. In LOOP, these challenges have been addressed for delivering virtual metrology services in the automotive industry as a case study. In this paper, we provide an overview of the main achievements emanating from the LOOP project (EUREKA- CELTIC call 4: an instrument that aims to strengthen Europe s competitiveness in telecommunications through short- and medium-term collaborative R&D projects) that have led to potential innovative products for operators and wireless process manufacturers. This paper is organized as follows: Section 2 presents the LOOP case studies and the associated technical challenges; Section 3 provides an overview of the key technical achievements; Section 4 presents the product innovations born from LOOP; the conclusion is in Section LOOP Scenarios and Technical Challenges In order to better understand the technical challenges faced by the convergence of wireless networks, herein we provide the description of the two major scenarios identified within the scope of the project targeting the telecoms industry and process manufacturing. The first scenario targets potential new services and energy-efficient networks for the operators in order to anticipate the deployment of NGNs in an era where spectral resources are at premium. The deployment of NGN aims at a global infrastructure where several systems can coexist to support transparent end-to-end communications in a costeffective manner. An important issue for next generation wireless systems will be coexistence and optimization to provide a network-of-wireless networks accommodating a variety of radio technologies and mobile services in a seamless and cost-effective manner. To address these issues, the main focus of LOOP was to explore innovative solutions targeting the following. (i) Network discovery, session management and roaming allowing the end-user to maintain session continuity whilst roaming between operators and heterogeneous wireless technologies. (ii) Ad-Hoc networking for relay-based cell coverage extension to extend wireless and mobile coverage providing enhanced QoS delivery and extended service delivery to remote and fringe users. (iii) Dynamic spectrum allocation for heterogeneous networks to investigate the opportunistic use of licensed spectrum by secondary systems for optimized utilization of scarce spectral resources. (iv) Intra-system optimization to maximize network utilization by exploring the application of a crosslayered protocol architecture. The second scenario is directed towards the car manufacturing industry and focuses on the deployment of metrology services on the factory floor to allow quality control production engineers to analyze and process large volumes of 3D multimedia information in real-time anywhere and anytime, as shown in Figure 2. A major problem for manufacturing companies is the maintenance of their cost-intensive, production-critical assets, such as machines, tools, and equipment. These assets constantly suffer from aging and wear, which often lead to functional loss and breakdown of machines, and ultimately, complete standstill of production with costly consequences. One promising approach is to anticipate and address the problems before they occur. The LOOP project provides a solution for ubiquitous monitoring and management of Coordinate Measurement Machines (CMM) based on closed-loop approaches through wireless and nomadic sensors placed around the mechanical robot in a car production line. The LOOP application would inform the maintenance engineering team of potential problems in order to ensure their prevention and to manage their repair with a sustainable plan in mind. The targeted solution is based on the need for NGN solutions enabling physical and semantic interoperability of required sensors, devices, services and systems. LOOP builds on relevant ongoing progress in wireless routing protocols based on cross-layer design to ensure that a fast and proactive communication path is established on the factory floor back to a Maintenance Service Centre (MSC)/Central Decision Point (CDP) in order to mobilize the resources for repair or to stand by for further updates and information. 3. LOOP Achievements 3.1. Suitability-Based RAT Selection Algorithm. Networks of the future will explore cooperative platforms in a bid to provide cost-effective communications to the end user. In a bid to address this challenge, interworking architectures have been proposed by ETSI/BRAN [1] and 3GPP [2] such as the loose and tight coupling approach for WiFi and UMTS/HSDPA (High-Speed Downlink Packet Access). Moreover, several solutions have been proposed within international research projects that worked on architectures and platforms for cooperation schemes between heterogeneous Radio Access Networks (RANs) [3 6], mainly focusing on interworking architectures between UMTS/HSDPA and WiFi. In [7], the requirements and algorithms for cooperation of several RANs are presented. Cooperation can also be

22 EURASIP Journal on Wireless Communications and Networking 3 Internet WT WT WR Ad-hoc/mesh network WR WL WiMAX/UMTS-LTE BS AN AN AN AN AN Test MMT WiMAX/UMTS-LTE connection WLAN connection AN: Ad-hoc node WR: wireless router Figure 1: LOOP scenario. WLT: WLAN terminal WT: WiMAX/UMTS LTE terminal Technical assistance department Wireless router 1 Point cloud storage server Wireless router 2 Video WLAN Wireless router 5 Point cloud 5 Wireless router 3 Point cloud 3 Wireless router 4 Point cloud 4 AD-HOC network Figure 2: TRIMEK scenario. achieved by means of a CRRM (Cooperative Radio Resource Management Entity) entity that is able to direct traffic through different networks according to operator-specific requirements and based on cross-system information. More specifically, the CRRM is responsible for (i) gathering system- and user-specific information, (ii) processing this information according to operator specific criteria, and (iii) triggering a new handover event according to the load balancing criteria and position. It is assumed that either a common operator deploys both systems or the system operators share a service-level agreement (SLS). Reference [8] investigated a CRRM-type cooperation based

23 4 EURASIP Journal on Wireless Communications and Networking on the load-suitability for delay-constrained services. The notion of suitability is based on the most preferred access system to accommodate the service, but the suitability factor can change as load increases in order to maintain the quality of service across the networks. In LOOP, we extend this notion of suitability cooperation for RAT (Radio Access Technology) selection to optimize the choice between WiFi and HSDPA. The suitability cooperative algorithm for RAT selection is expressed by S ( L ( cell i,j )) 1, if L ( ) cell i,j LTh j, = 1 ) L( 2 cell i,j, if L ( ) cell i,j > LTh j, 1 LTh j where cell i,j represents the cell/ap i pertaining to RAT j ; L(cell i,j ) is the normalized load in cell i,j ;LTh j is the load threshold for RAT j ; S (L(cell i,j )) is the suitability value for accepting a new user in cell i,j. The algorithm was testing the use-case scenario involving HSDPA partially overlapped by WiFi indoor hotspots, assuming high-priority NRTV (Near Real-Time Video) traffic at 64 kbps characterised by the 3GPP model [9]. Figure 3 provides the simulation results for CRRM goodput (bits that are received correctly and within the QoS delay threshold) versus offered load. LOOP results show that the CRRM system throughput gain introduced by service suitability is significant and sensitive with regards to the service suitability threshold. The optimal load threshold was determined to be LTh =.6 where the potential observable gain is around 1.2 Mbps in contrast to the stand-alone HSPDA scenario; the use of smaller load thresholds is not advised, since it causes the WiFi system to overload faster-causing problems to the existing background traffic Ad hoc Networking for Relay-Based Cell Coverage Extension. In the LOOP project, we have studied mechanisms to extend the cell coverage through cooperative mechanisms based on the use of relays. In particular, we have focused on Cooperative Automatic Retransmission Request (C-ARQ) schemes [1] which allow for the transmission of data even when the channel conditions are poor, and errors are frequent, by enabling spontaneous relays to retransmit upon the occurrence of a transmission error from the source. In LOOP, we have designed a new MAC protocol to coordinate the retransmissions from these helpers or relays called Persistent Relay Carrier Sensing Multiple Access (PRCSMA) protocol, and it represents an extension of the IEEE Standard [11] to operate in C-ARQ schemes. A comprehensive description, theoretical analysis, and performance evaluation of the protocol can be found in [12]. When using PRCSMA, all the stations must listen to every ongoing transmission in order to be able to cooperate if required. Whenever a data packet is received with errors at the destination station, a cooperation phase can be initiated (1) Goodput (kbps) Total number of user NRTV offered load CRRM goodput.5 CRRM goodput.6 CRRM goodput.7 CRRM goodput.8 Goodput without CRRM Figure 3: System total throughput with CRRM entity exploring the diversity gain for cell radius = 5 m. by broadcasting a Call for Cooperation (CFC) packet. Upon the reception of the CFC, all the stations willing and able to cooperatebecomeactiverelaysandgetreadytoforwardthe original packet. To do so, they use the MAC rules specified in the IEEE Standard [11] considering the two following modifications. (1) There is no expected ACK associated to each transmitted cooperation packet. (2) Those active relays which do not have an already set back-off counter (from a previous transmission attempt) set it up and initiate a random back-off period before attempting to transmit for the first time. Those relays which already have a non-zero back-off counter value keep the value upon the initialization of a cooperation phase. A cooperation phase is completed, either when the destination station is able to decode the original data packet by properly combining the different retransmissions from the relays, or when a certain maximum cooperation timeout has elapsed. In the former case, an ACK packet is transmitted by the destination station. In the latter case, a negative ACK (NACK) is transmitted by the destination station. In any case, all the relays pop out the cooperative packet from their queue upon the end of a cooperation phase. The performance of PRCSMA has been analytically modeled by applying Markov Chain Theory [12]. We have focused on the evaluation of the average packet transmission delay when cooperation is required, which is defined as the average amount of time elapsed from the moment a packet is transmitted for the first time until it can be decoded without errors at the destination upon the reception of an arbitrary number K of retransmissions received from the relays. We plot in Figures 4 and 5 the value of the average packet transmission delay for the case when the relays can

24 EURASIP Journal on Wireless Communications and Networking 5 Average packet transmission delay (ms) CW = 32; active relays = 1; basic access Number of required retransmissions (K) Traditional ARQ 1 54 C-ARQ 1 54 sim C-ARQ 1 54 model Traditional ARQ 6 54 C-ARQ 6 54 sim C-ARQ 6 54 model Figure 4: Average Packet Transmission Delay (relay low rate regime). transmit at 54 Mbps to the destination while the source can do it at 1, 6, 24, or 54 Mbps. The data transmission rates are represented in the legend of the plots indicating the transmission rate of the source and the transmission rate of the relays separated by a dash. The control transmission rate has been fixed in all cases to 6 Mbps. In addition, we consider in all cases that the C-ARQ is executed by means of the PRCSMA basic access, that is, without RTS/CTS handshake. The traditional ARQ curve represents the case when the retransmissions are only requested from the original source (there are no relays). Finally, it is worth mentioning that we have included in the plots both the results obtained through computer simulation and by means of the derived theoretical model. The perfect match between the two cases shows the accuracy of the developed model. The ratio between the transmission rate of the source and that of the relays determines how efficient the C-ARQ mechanism is in comparison to the traditional non-cooperative ARQ approach, where the retransmissions are only requested from the source at the best available transmission rate between the source and the intended destination station and without contention between consecutive retransmissions. For example, in the case of using the transmission rate set 1 54 (source-relays transmission rate), when K = 5, the C-ARQ reduces the average packet transmission delay by a factor 4 compared to the traditional ARQ scheme. On the other hand, at the limit where the relay stations transmit at the same rate as the source station, the average delay in the C-ARQ scheme is higher due to the cost of coordinating the set of relays. It is worth mentioning that, as can be expected, if K is very low, then the efficiency of the C-ARQ scheme becomes similar to that of a traditional non-cooperative ARQ scheme. This is due to the fact that, despite the faster relay retransmissions, the overhead associated to the protocol does not pay off the reduction of the actual data retransmission time. In the case of networks where the data transmission Average packet transmission delay (ms) CW = 32; active relays = 1; basic access Number of required retransmissions (K) Traditional ARQ delay Distributed ARQ delay sim Distributed ARQ delay model Traditional ARQ delay Distributed ARQ delay sim Distributed ARQ delay model Figure 5: Average Packet Transmission Delay (relay high rate regime). rate of each station is selected as a function of the channel state between source and destination stations, as in IEEE WLANs, the behavior of PRCSMA shows that C-ARQ schemes would be especially beneficial for those stations located far away, in radio-electric terms, that is, at the cell boundaries from a transmitting station. Note that these stations will be prone to transmit at very low transmission rates and therefore they could benefit from faster and more reliable retransmissions performed by intermediate relay stations on the path from the source station. In addition, the whole network, that is, the rest of the stations, will benefit from this scheme in the sense that faster transmissions will occupy the channel for shorter periods of time Cooperative Spectrum Sensing for Cognitive Radio- Enhanced Heterogeneous Networks. As wireless technologies continue to grow, more and more spectrum resources will be needed. However within the current spectrum regulatory framework, all of the frequency bands are exclusively allocated to specific services, and no violation from unlicensed users is allowed. A recent survey of spectrum utilization made by the Federal Communications Commission (FCC) has indicated that the actual licensed spectrum is largely underutilized in vast temporal and geographic dimensions [13]. Spectrum utilization can be improved significantly by allowing a Secondary User (SU) to utilize a licensed band when the Primary User (PU) is absent. Cognitive Radio (CR), as an agile radio technology, has been proposed to promote the efficient use of the spectrum [14]. By sensing and adapting to the environment, a CR is able to fill in spectrum holes and serve its users without causing harmful interference to the licensed user. To do so, the CR must continuously sense the spectrum it is using in order to detect the reappearance of the PU. Once the PU is detected, the CR

25 6 EURASIP Journal on Wireless Communications and Networking PU (H /H 1 ) 1 Sensing channel 1 1 Pm SU 1 SU 2 SU L 1 2 Reporting channel H / H 1 Fusion center Figure 6: Multilayer distributed/cooperative spectrum sensing ρ SR =, ρ RD =, ρ SD = ρ SR =.2, ρ RD =, ρ SD = ρ SR =, ρ RD =.2, ρ SD = ρ SR =.2, ρ RD =.2, ρ SD = ρ SR =.2, ρ RD =.2, ρ SD =.2 L 1 should withdraw from the spectrum so as to minimize the interference it may possibly cause. However, a very important challenge of implementing spectrum sensing is the hidden terminal problem, which occurs when the CR is shadowed, in severe multipath fading or inside buildings with a high penetration loss while a PU is operating in the vicinity. Cooperative communications are an emerging and powerful solution that can overcome the limitation of wireless systems [15]. The basic idea behind cooperative transmission rests on the observation that, in a wireless environment, the signal transmitted or broadcast by a source to a destination node is also received by other terminals. These latter nodes can process and retransmit the signals they receive. The destination then combines the signals coming from the source and the partners, thereby creating spatial diversity by taking advantage of the multiple receptions of the same data at the various terminals and transmission paths. By allowing multiple CRs to cooperate in spectrum sensing, the hidden terminal problem can be addressed [16]. Indeed, cooperative spectrum sensing in CR networks has an analogy to a distributed decision in wireless sensor networks, where each sensor makes a local decision and those decision results are reported to a fusion centre to give a final decision according to some fusion rule. The main and fundamental difference between these two applications lies in the wireless environment. Compared to wireless sensor networks, CRs and the fusion centre (or common receiver) are distributed over a larger geographic area. This difference brings out a much more challenging problem to cooperative spectrum sensing because sensing channels (from the PU to CRs) and reporting channels (from the CRs to the fusion centre or common receiver) are normally subject to fading or heavy shadowing. In LOOP [17], we have analyzed, for the first time in the literature, the fundamental problem of cooperative spectrum sensing over wireless environments characterized by realistic Figure 7: Probability of not detecting a PU (P m ) against the number of cooperating CRs (L). The curves are obtained for different values of the correlation coefficient of the shadow fading over the sensing channel (ρ SR ), the reporting channel (ρ RD ), and pairs of links on the sensing and reporting channels (ρ SD ). propagation conditions, that is, heavily and spatially correlated shadowing environments. More specifically, we have proposed an advanced framework for performance analysis and optimization of a general multilayer decentralized data fusion problem for application to cooperative spectrum sensing, which includes realistic sensing/reporting channels and correlated Log-Normal shadowfading in all wireless links of the cooperative network. The analyzed system setup is sketched in Figure 6. The analysis of the scenario in Figure 6 has revealed an important result: even though always overlooked in typical cooperative spectrum sensing analysis, shadowing correlation on the reporting channel can yield similar performance degradations as shadowing correlation on the sensing channel. So, our performance study has revealed that further importance should be given to the role played by the reporting channel for a sound analysis and design of distributed detection problems with data fusion, especially when the system is expected to be deployed in realistic propagation environments targeted for CR applications. An example of the obtained results is shown in Figure Cross-Layer Packet Scheduling for WiMAX. The IEEE standard [18, 19] provides specification for the Medium Access Control (MAC) and Physical (PHY) layers for WiMAX (Worldwide Interoperability for Microwave Access). A critical part of the MAC layer specification is the scheduler, which resolves contention for bandwidth and determines the transmission order of users; it is imperative for a scheduler to satisfy QoS requirements of the users,

26 EURASIP Journal on Wireless Communications and Networking 7 maximizing system utilization and ensuring fairness among the users. The basic approach for providing the QoS guarantees in the WiMAX network [2, 21] considers that the BS performs the scheduling for both the uplink and downlink directions; an algorithm at the BS has then to translate the QoS requirements of SSs into the appropriate number of slots. The IEEE 82.16d/e standards [18, 19] do not specify scheduling techniques for MAC layer in WiMAX networks, and the existing NS-2-based simulation platforms [22], implement only QoS-aware scheduling based on Service class prioritization. We propose a simple, efficient solution for the WiMAX scheduler that is capable of allocating slots based on the QoS Service class, traffic priority and the WiMAX network and transmission parameters. To test the proposed solution, the QoS model for the IEEE 82.16d/e MAC layer in the NS-2 simulator [23] developed by the WiMAXForum [22, 24, 25] was taken as a reference. We propose the Enhanced Round Robin (err) scheduler (cf. Figure 8). It is based on the simple round robin solution, but introduces more elements in the decisionmaking process for packet allocation within each radio frame. The proposed scheduler algorithm has in fact two objectives. (i) The first, that was already mentioned, maps the user traffic to the available radio resources according to the service class and radio channel quality. (ii) The second allows user differentiation/priority within each service class and thus enables the network operator to implement new business models, the concept of gold, silver and bronze users, guaranteeing at the same time the subscribed QoS. In practice, the algorithm initially performs the same round robin procedure as explained in the previous models, that is, serving first connections in the following order: UGS, rtps, nrtps, and BE. From the list of existing connections inside the same class, a priority is also established taking into account the RSSI (Received Signal Strength Indication) value for the given node; where highest priority is given to users with highest signal strength. This approach will provide a trade-off between optimizing spectral efficiency and guaranteeing QoS. Simulations were realized using a point-to-multipoint topology with three services running on the same terminal, conveying differentiated traffic in the uplink direction, namely, the configured traffic sources for UGS (Unsolicited grant service), rtps (Real time polling service) and BE (Best Effort). In this scenario, we have defined the relevant PHY layer simulation parameters. The key simulation parameters are summarized in Table 1. Figure 9 shows the slight gain difference that can be achieved in throughput using the enhanced Round Robin solution, in contrast to the observed earlier service class differentiation. In this particular case, the traffic prioritywas assumed to be equal among the same classes, the priorities here are based on the service class and RSSI of the respective terminal. Metric Start UL allocation UGS rtps nrtps BE User priority H RSSI H RSSI L User priority L Figure 8: Enhanced Round Robin algorithm. Table 1: Simulation parameters. RSSI H RSSI L Quantity Frequency GHz Bandwidth 2 MHz Frame duration 5 ms Downlink ratio.3 Modulation 16 QAM Channel model Cost 231 Fading model ITU PDP PED A Cyclic prefix.25 Queue length 1 packets Services Parameters Traffictype Bit rate (kbps) Packet size (bytes) BE to 124 UGS 2 3 rtps 2 2 to 98 Concerning delay, as shown in Figures 9 and 1, the proposed scheduler reduces the overall packet delay and either equals or slightly outperforms the existing Round Robin based on the WMF (WiMAX Forum) model. Figures 11 and 12 illustrate the scenario consisting of terminals supporting the rtps and BE classes, respectively, and different traffic priorities inside each service class, that is, rtps1 has lower priority than the rtps connection and BE1, also in respect to BE. The results show the priorities in the scheduling decision as both classes are distinguished in terms of throughput and delay (better values are observed for rtps classes than BE ones) and traffic prioritization inside each particular class (improved performance for rtps and BE in relation to rtps1 and BE1, resp.). In Summary, the err algorithm provides a new innovative scheme to implement new business models based on the application of the cross-layer paradigm for RR scheduling. Numerical results show that the proposed scheme can

27 8 EURASIP Journal on Wireless Communications and Networking Throughput (kbps) Delay (ms) Throughput versus Nr. of MN (WMF/eRR) Number of subscribers UGS WMF UGS err RTPS WMF RTPS err BE WMF BE err Figure 9: Throughput RR/eRR Number of subscribers UGS WMF UGS err RTPS WMF Delay versus Nr. of MN (WMF/eRR) RTPS err BE WMF BE err Figure 1: Delay RR/eRR. increase the system throughput by up to 11%, reduce traffic delayby27% Cross-Layer Optimized Routing Strategies. The benefits of cross-layer system design are mainly being applied in the area of mobile and wireless operators. In recent years, the area of communications in the manufacturing process is gaining importance. Traditional Ethernet and PROFIBUS [26, 27] factory systems are being enhanced to facilitate new means of automation through attractive wireless solutions: to provide added flexibility and self-configuration in processing machines to reduce production costs. LOOP tackles machine automation by investing communication challenges related to remote management of Coordinate err service class and traffic priority Throughput (Kb/s) rtps rtps1 BE BE1 Figure 11: Throughput. err service class and traffic priority Delay (ms) rtps rtps1 BE BE1 Figure 12: Delay. Measurement Machines (CMMs) for future manufacturing environments. Specifically, we aim to investigate routing strategies to provide fast and efficient data management on the factory floor that is highly dynamic in nature. LOOP specifically addresses cross-layer enhancements to both flat and hierarchical routing strategies. In HOLSR (Hierarchical Optimized Link State Routing), there are two levels of hierarchy according to our network design as shown in Figure 13, where Level-1 hierarchy corresponds to connection among backbone network nodes, while Level-2 hierarchy corresponds to connection among mesh routers in access networks. Regarding the second cross-layer enhancement, Cross Layer Link Layer Notification, the basis is to utilize link break information gathered at the MAC layer to impose OLSR [28] routing table recalculation. More specifically, the MAC layer detects the link break and sends an indication to the protocol layer. Upon receiving such an indication which is treated as a topology or neighbor change, OLSR conducts routing table recalculation immediately. Finally, note it is of great importance to understand which approach is more effective, namely, cross-layer or hierarchical, to deploy the correct solution based on the dimension addressed. Simulation results (Figure 14) provide the performance of video communication over the network, as the transmission of the virtual part was taking place. Such multimedia stream would be directed to experts in assisting the manufacturing decisions all over the plants that are normally very

28 EURASIP Journal on Wireless Communications and Networking 9 Cluster level-1 Nodes type-1 Cluster level-2 Nodes type-2 Throughput (kbps) Number of CBR connections HOLSR OLSR Throughput Figure 13: Hierarchical network design. large. The video connection has a bit rate of 128 kbps and a QCIF format. The results obtained suggest that in small deployment areas, intrasystem optimization based on cross-layer approaches is more effective than the hierarchical counterparts. 4. LOOP Products Carrying out research at European level is primarity important when facing the global market, since gaining knowledge on novel technologies and system integration may provide the needed competitive advantage at industrial level, that is, better European products or better European-based networks deployed around the world. In this context, LOOP has transferred engineering know-how to meet the shortterm market requirements to allow industry to anticipate the commercial deployment towards NGNs in terms of delivering potential products that include the following. OptiMaX (Portugal Telecom Inovação). Radio access network planning and deployment is a complex process that can be divided into two key stages. In this first stage, the optimization goals (capacity, coverage and QoS) are defined, the network is dimensioned and the radio planning and optimisation loop are initialized. In this process, sophisticated planning and optimisation tools are used which resort to complex cost functions to perform various trade-offs. The output from the iterative optimization stage results in BS parameters corresponding to the Radio Resource Management (RRM) algorithm under test. In the second stage, after network deployment, network performance and quality characteristics are monitored. In this stage, monitoring tools are used to collect the geo-referenced radio measurements (e.g., SNR) in order to evaluate the difference between what was planned and what is in fact implemented. Based on monitoring results and on the RNP (Radio Network Planning) simulations, the radio network parameters are Figure 14: Throughput Hierarchical OLSR versus enhanced OLSR. tuned which usually include both hard (e.g., antenna tilts) and soft parameters (RRM mechanisms). Despite the widespread deployment of WiMAX (IEEE 82-16d) networks, there is no radio monitoring tool on the market to support network operators in the optimization task; hence this provided the impetus for the OptiMax tool proposed in the scope of the LOOP project. OptiMax is a new tool that allows the network operator to perform network analysis and planning for the WiMAX (IEEE82.16d) system. The monitoring phase not only constitutes collecting and storing the radio signal quality for coverage measurements, but can also sniff-out essential network information pertaining to the IEEE82.16 protocol. Moreover, the monitoring capabilities of the tool can also estimate the maximum bit rate per location for a particular bandwidth. In order to obtain this network-related data, specific CLI (Command-Line Interface) requests are made to the SS (Subscriber Station). The replies are parsed to XML format resultant from the monitoring phase. Each monitoring session is attached with the potential locations of each WiMAX Base Station so that it can be overlaid on a geographical map, where each position is represented by a coloured circle ranging from red (low RSSI, Received Signal Strength Indication) to green (higher RSSI). The hardware needed to execute each test is shown by Figure 15. It constitutes: (1) laptop, with OptiMax application installed, for mobility testing, (2) GPS system, the used GPS system connects via Bluetooth, (3) WiMAX omnidirectional antenna, (4) UPS system and a PoE (Power over Ethernet) unit providing energy for the SS, (5) a WiMAX SS (Subscriber Station), connected to the laptop using an Ethernet cable. The equipment used, was chosen by its flexibility, low cost and easy loading in an automobile for field testing.

29 1 EURASIP Journal on Wireless Communications and Networking UPS GPS Application PoE Antenna SS WiMAX Figure 16: Router performance before versus after LOOP technology. Figure 15: Portugal Telecom Inovação s OptiMaxtool. The OptiMax tool collects the geo-referenced radio measurements (e.g., SNR) in order to evaluate the difference between what was planned and what is in fact implemented. Based on the monitoring results and on RNP (Radio Network Planning) tool simulations, the radio network is optimized: antenna tilts and azimuth, transmitted power level, and so forth. WiMAX System Experimental Platform (Turkcell). Mobile WiMAX is an access technology that promises high-data rates and wide coverage at low cost. Mobile WiMAX is based on which specifies the air interface including the physical layer (PHY) and medium access layer (MAC) for broadband wireless systems. To achieve high throughput and very good spectral efficiency, mobile WiMAX combines orthogonal frequency division multiple access (OFDMA) and multiple input multiple output (MIMO) with link adaptation and hybrid automatic repeat request (ARQ) algorithms. However, wireless communication technologies, and how to exploit better spectral efficiency improve dayby-day. Toward this end, we were interested in developing the WiMAX system level simulator to act as an experimental platform to test new algorithms/protocols for enhancing system efficiency through augmenting cell capacity. The main challenges in the implementation of such a simulator were the selection of parameters and assumptions. To overcome this challenge, we developed a systemlevel simulator compliant with the m Evaluation Methodology [29]. The simulator test-bed was validated for different network configurations such as antenna numbers, frequency reuse patterns, user densities, and mobility of users. Management Tools for Wireless Process Manufacturing (TRIMEK). The wireless market in Process Manufacturing is expected to grow at a pace neighbouring 3% per year. It is growing faster than wired market. Nevertheless, adoption of wireless technology is still very low and most of managers in industry are reluctant to introduce radio solutions when wired alternatives exist. The primary impediments to wireless penetration are latency and performance issues, as well as reliability and security of sensitive information. In the scope of LOOP, these challenges have been addressed by applying autonomous wireless networks with crosslayer routing strategies to provide remote management capabilities for the flexiblity and self-configuration of measurement machines. Remote management is a key business opportunity for process manufacturers to provide technical assistance towards the offering of complete automation solutions allowing industrial automation and control providers, as well as major system integrators, to manage or visualize complete and self-contained control components, including software-based functionality, command and control, configuration, diagnostics, and documentation. TRIMEK is a CMM manufacturer, as well as a service provider for this kind of machinery. These are complex systems composed by a large number of parts (mechanical, electrical, electronic, IT, etc.). Therefore, any service regarding them might require the knowledge of professionals from different fields. Unfortunately it is not possible to forecast in advance the needs of both the machine and the service demanded by the client. For this reason, the help of new technology in this field will have the potential to provide an internal tool to satisfy unexpected problems in a short-time basis and with the accuracy required by this type of systems. TheroleofLOOPhasbeentoprovidemorewireless flexibility and self-configuration by integrating ubiquitous monitoring and management tools on TRIMEK in-house 3D CMMs for the automotive industry. Therefore, based on the LOOP cross-layer routing strategy and traffic rules (Section 3.5), a wireless ad-hoc link was established on the factory floor resulting in highly autonomous CMM machines with high flexibility. Moreover it was established that the use of traffic rules improved the bandwidth assigned to the prioritised traffic while maintaining the quality of the video streams at the desired level. It has also been demonstrated that dynamic queue management, based on adaptive priority handling, is a key factor when trying to offer a specific quality to the provided services. Figure 16 shows the remote scan before and after LOOP technology. 5. Conclusions Even though converged NGNs are still in their early stages, the impacts of NGN are expected to be significant to the ICT market on two levels: firstly NGN will provide the vehicle for enhancing access to communication services, and more

30 EURASIP Journal on Wireless Communications and Networking 11 innovative and personalised services and applications; and secondly NGN would be a basis for the UNS (Ubiquitous Network Society), where easy-to-use networks are connected at anytime, anywhere, with anything and for anyone. LOOP is one piece in the jigsaw, however more investment is required to help this vision to become a reality and to address new emerging challenges that include energy-efficient and secure communications. Acknowledgments The work presented in this paper was carried out in the scope of the LOOP project (CP4-16), a nationaly funded project supported by the EUREKA-CELTIC cluster ( and also partially funded by the FP7 ICT-WHERE with contract no References [1] Broadband Radio Access Networks (BRAN); HIPERLAN Type 2; Requirements and Architectures for Interworking between HIPERLAN/2 and 3rd Generation Cellular Systems, Tech. Rep. ETSI TR , 21, V [2] Third Generation Partnership Project, 3GPP, [3] Evolving systems beyond 3G, IST MIND, [4] IST Project CAUTION, Capacity Utilizations in Cellular networks of present and Future Generation, [5] Evolutionay stategies for RRM, IST project EVEREST, [6] Advanced Resource Management Solutions for Future All IP Heterogeneous Mobile Radio Environments, IST Project AROMA, [7] A. Mihovska, E. Tragos, E. Mino et al., Requirements and algorithms for cooperation of heterogeneous radio access networks, Wireless Personal Communications, vol. 5, no. 2, pp , 29. [8] V. Monteiro, O. Cabral, J. Rodriguez, F. Velez, and A. Gameiro, HSDPA/WiFi RAT selection based on load suitability, in Proceedings of the ICT Mobile and Wireless Communications Summit (ICT-MobileSummit 8), Stockholm, Sweden, June 28. [9] 3GPP, Feasibility Study for Orthogonal Frequency Division Multiplexing (OFDM) for UTRAN enhancement, Tech. Rep. TR , 3rd Generation Partnership Project, Technical Specification Group Radio Access Network, 24, v6... [1] M. Dianati, X. Ling, K. Naik, and X. Shen, A nodecooperative ARQ scheme for wireless ad hoc networks, IEEE Transactions on Vehicular Technology, vol. 55, no. 3, pp , 26. [11] IEEE, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, IEEE Std , August [12] J. Alonso-Zárate, E. Kartsakli, C. Verikoukis, and L. Alonso, Persistent RCSMA: a MAC protocol for a distributed cooperative ARQ scheme in wireless networks, EURASIP Journal on Advances in Signal Processing, vol. 28, Article ID 81741, 13 pages, 28. [13] FCC, Spectrum Policy Task Force, ET Docket 2-135, November 22. [14] J. Mitola III and G. Q. Maguire Jr., Cognitive radio: making software radios more personal, IEEE Personal Communications, vol. 6, no. 4, pp , [15] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, Cooperative diversity in wireless networks: efficient protocols and outage behavior, IEEE Transactions on Information Theory, vol. 5, no. 12, pp , 24. [16] D. Cabric, S. M. Mishra, and R. W. Brodersen, Implementation issues in spectrum sensing for cognitive radios, in Proceedings of the 38th Asilomar Conference on Signals, Systems and Computers, pp , Pacific Grove, Calif, USA, November 24. [17] M. Di Renzo, L. Imbriglio, F. Graziosi, F. Santucci, and C. Verikoukis, Cooperative spectrum sensing for cognitive radios: performance analysis for realistic system setups and channel conditions, in Mobile Lightweight Wireless Systems, vol. 13 of Lecture Notes of the Institute for Computer Sciences, pp , Springer, Berlin, Germany, 29. [18] IEEE Working Group, IEEE Standard for Local and Metropolitan Area Networks. Part 16: Air Interface for Fixed Broadband Wireless Access Systems, IEEE Std , October 24. [19] IEEE Working Group, IEEE Standard for Local and Metropolitan Area Networks. Part 16: Air Interface for Fixed Broadband Wireless Access Systems. Amendment 2: Physical and Medium Access Control Layer for Combined Fixed and Mobile Operation in Licensed Bands, IEEE Std e, December 25. [2] WiMAX Forum, WiMAX End-to-End Network Systems Architecture Stage 2: Architecture Tenets, Reference Model and Reference Points, Release 1.1., June 27. [21] WiMAX Forum, WiMAX End-to-End Network Systems Architecture Stage 3: Detailed Protocols and Procedures, Release 1.1., June 27. [22] The Network Simulator NS-2 MAC+PHY Add-On for WiMAX (IEEE 82.16): ns2 release 2 Documentation, WiMAX Forum, August 27. [23] The Network Simulator ns-2, 22, nsnam/ns/. [24] Application Working Group, about/board/working Group Organizations/AWG. [25] NS-2 NIST, [26] PROFIBUS, [27] Hans Gerlach-Erhardt, Real Time Requirements in Industrial Automation, PNO TC2/WG12, ETSI Wireless Factory Starter Group Meeting, October 29. [28] T. Clausen and F. Jacquet, Optimized Link State Routing Protocol (OLSR), IETF RFC 3626, October 23. [29] IEEE82.16 TGm, 82.16m Evaluation Methodology, IEEE 82.16m-8/4r5, January 29.

31 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 11 pages doi:1.1155/21/ Research Article New Trends on Ubiquitous Mobile Multimedia Applications Joel J. P. C. Rodrigues, 1, 2 Marco Oliveira, 2 and Binod Vaidya 1 1 Instituto de Telecomunicações, UBI, Rua Marquês D Avila e Bolama, Covilhã, Portugal 2 Department of Informatics, University of Beira Interior, Rua Marquês D Avila e Bolama, Covilhã, Portugal CorrespondenceshouldbeaddressedtoJoelJ.P.C.Rodrigues,joeljr@ieee.org Received 2 March 21; Accepted 1 July 21 Academic Editor: Liang Zhou Copyright 21 Joel J. P. C. Rodrigues et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Mobile devices present the opportunity to enhance our fast-growing and globally connected society, improving user-experience through novel approaches for information dissemination through mobile communication. The research community is developing new technologies, services, and applications to enable ubiquitous environments based on mobile technology. This paper tackles several important challenges such as communication cost and device limitations for development of ubiquitous multimedia applications. And we propose a system for news delivery using a set of wireless multimedia applications. For this purpose, we have performed a case study with Apple iphone s platform, featuring two multimedia application contexts, namely, Web and native applications. The multimedia mobile applications draw on iphone s assets, enabling context-awareness to distribute news, improving communication efficiency and setting-up viewing optimizations, thus enhancing user-experience. The proposed system is evaluated and validated through a series of real-life experiments on real devices, with online full availability. Moreover, due to the Web application availability, the system is not restrained to Apple s iphone platform, but can also benefit users with other devices. 1. Introduction Mobile devices have fulfilled the true aim of Internet by offering full connectivity anytime anywhere. The trend of going wireless goes beyond the walls of homes, university buildings, or hotels and reaches the open spaces of nature or the mobile spaces of trains and buses. The freedom of movements is used to speak everywhere without the need to log in a local wireless network, and to extend it to other Internet services such as Web surfing, checking, reading news, listen to online radios, or even watching video streaming and television. The mobile devices market is an emerging mass market with little data usage research available. Consumers are changing their habits, the Internet players are adapting their contents to adjust the new needs, and operators are maintaining a high cost and network restrictions to avoid massive usage. In this market, the business model is restricted to cost-per-use, so data porting to mobile devices have not really taken off. With embedded technology, mobile devices have many features, so people tend to use mobile devices in order to access to Internet contents more frequently. In this paper, we want to explore and take advantage of this technology, study people habits, and understand their perspectives to this recent trends for the future to create new ways of content distribution. This paper also proposes a mobile system that tries to gather the above-mentioned characteristics for iphone platform from Apple. We refer to iphone assuming that mobile device incorporates iphone, iphone 3G, iphone 3GS, and ipod Touch functionalities. This work studies ubiquitous Internet services in two types of applications Web applications and native applications. The delivery system incorporates context-awareness, industry standards, and users predefined settings to evaluate how the information is processed, transmitted and displayed. The proposed system improves mobile devices, communication, server side computation processes, user-experience, and usability. The benefits for clients are focused on what and how they want to see, speed, and Internet lower data transfers costs. The benefits for mobile devices will affect battery, memory, processor, and Internet connectivity performance [1]. Furthermore, service providers should improve network availability, bandwidth, and reduce storage costs.

32 2 EURASIP Journal on Wireless Communications and Networking The native application resolves some of the wireless networks limitations such as low-bandwidth and unreliability, understands the user definitions, and reads some mobile devices context in order to adjust the dynamic content downloaded from the Web Service. This solution combines ubiquitous information in both sides of the provided system. On the server side, it calculates image quality, number of items to download, and text format. On the mobile device (client) side, it decides what, how and how many news information or images will be downloaded. This ubiquitous collaboration between mobile device and Web Service is a new approach of intelligent applications that brings out the betterment of two worlds the server-side power and the client-side context, location, and sensor awareness [2]. Moreover, for journalists and all the news related professionals, it is important to know if a mobile device is just a Web page extension or a new way of communication, more specifically the seventh way of communication [3]. The rest of the paper is organized as follows. Section 2 reviews the related literature and the background for mobile and ubiquitous applications. Section 3 focuses on technology, specifications, tools, and methods used to create the proposed applications. Section 4 describes, in detail, the four proposed Web applications for news delivery and presents a comparison study between Mobile, News, RSS, and Edition applications. Section 5 elaborates on the native application and addresses ubiquitous technology incorporated on the system. Section 6 concludes the paper and points out directions for further research works. 2. Related Work Mobile devices, which maybe also referred to as handheld, portable devices, or wearable devices, such as mobile phones and personal digital assistants (PDAs), are small and lightweight equipments that can be fit into a suit pocket, hand, or briefcase. For this work, we have considered mobile devices such as mobile phones and PDAs with Internet connection capabilities and a small-size screen. These gadgets can provide not only regular phone calls but also other features like electronic mail ( ), gaming, infrared, Bluetooth and Wi-Fi connectivity, photo camera, video recording, music player, radio, and global positioning system (GPS). Even though mobile devices have a screen limitation in terms of size, all the content available to other kind of displays (cinema, television, and personal computers) are being produced accordantly to mobile characteristics. The content from other types of media such as recordings, print, radio, and Internet access, can also be imported to these devices. In 29, the estimated number of mobile phones subscriptions around the world was about 4,6 billion for a population of 6.8 billion people. And the penetration of mobile phone in the industrialized world was about 133%. In the same period, the broadband Internet access penetration (in mobile devices) was 2 times bigger than in 25. This makes the mobile phone the most widely spread technology and the most common electronic device in the world. We can predict that it will be the device of the future. According to [4], the enterprise mobile phones will replace desktop phones in North America by 211. As the number of people that accessed the Internet via their portable devices have increased by 25% in the second and third quarters of 28, among which the most audience is young generation, it can be predicted that the contents for these gadgets will continue to grow in future as well. Furthermore, in 27, for the generation of USA Internet users between 18 and 24 years, the preferred consumer electronics was the mobile phone (47%) over the computer laptop or desktops (38%). Indeed, as a 28 Nielsen Media Research report highlighted, mobile devices have increased traffic by an average of 13% across several popular websites [5]. The emergence of Web 2. has transformed the web into a more dynamic and interactive environment, offering a set of tools that enhance contact and collaboration between users. Several applications including online social networks, wikis and blogs, support such Web vision [6]. Currently, the interconnectivity and interactivity of Web-delivered content, which were born for the desktop computers, have been extensively applied to mobile devices as well. This new vision of Web is gathering the best that these devices have to offer portability and connection everywhere, and the Internet providers are creating new and innovative services for this market [7]. In general, the integration techniques used to combine Web Services and mobile devices are Socket communication and messaging techniques. Web Services uses extensible markup language (XML) and simple object access protocol (SOAP) to provide a mechanism that facilitates the data exchange over the Internet. They are being widely developed to enable quick and cheap integration with existing services, by combining multiple services in a single workflow. This facilitates interoperability across different hardware and software implementations, as it will be discussed on next sections. Nowadays, the application programming interfaces (APIs) created by several companies, such as Google, Amazon, and ebay, have robust Web server integration with desktops but when they migrate to mobile environment, new challenges should be addressed. The following aspects are identified: client-server data transfer optimization performance of the mobile application memory management security issues and user interface design with such display limitations. To overcome these issues, new techniques for caching, large data set handling, information on demand, data compression, paging, filtering, and performance improvement of network protocols are proposed for mobile computing. The number of news and media content downloaded through Internet and portable devices are increasing [8], therefore, we can predict that the number of these kinds of media Web Services will also increase. We also have the perception that typical iphone owners are bigger infotainment consumers. They are visiting, at least three times more than average, to several popular social, communication, and entertainment sites. The major newspapers and media groups of the world have already used the iphone application for delivering news in this new format. So, at this point,

33 EURASIP Journal on Wireless Communications and Networking 3 several questions may be raised. What mobile Web Services can we choose? What about the design guidelines for mobile devices? Or, there are used mobile development best practices? These issues were considered and kept in mind when planning our approach for the mobile Web Services. Furthermore, ubiquitous computing is an omnipresent relationship in terms of connections, management, and information interaction. This technology must create calm, and act as a quiet, invisible servant. It should help humans to extend their unconsciousness and intuition. This calm technology has the ability of going from the periphery of our attention, to the center, and back again [8]. Computing, communications with other devices, relationship management, are empowering our periphery but not moving much to the center of our attention. The impact on everyday life is already huge and becoming so commonplace likes writing or electricity. Mobile devices are getting smaller and powered up with add-ons, speed, and battery life-time. Combining those features and the growth of short-range ad hoc networks, the possibility of accomplishing the vision of ubiquitous computing, that was sketched out in the early nineties, is getting closer [9, 1]. With these significant improvements in the network and devices, the software is trying to catch up. Between them, the ubiquitous systems are still in their early phase. The impact of this technological wave will alter the place of technology in our lives. By now, every mobile device has Internet connection capabilities. We can interconnect them with other devices and systems, creating a computational relationship between them in a calm and ubiquity context. With increadingly use of mobile devices with GPS capabilities, location-awareness devices such as the iphone have changed our daily life. We will be capable of pinpointing our location on a Google Map, tracking friends, finding the nearest place to eat and shop, finding information of the particular area and so on. The first application for the iphone that uses the faux-gps feature (which used cell tower information to triangulate your position) is Google Maps. Now, there are a large number of applications that use real GPS features and services. Some examples of locationaware applications that can be found in Apple App Store are the following: Loopt Friend-finder application with virtual earth display, which allows user to share his location to the community; Whrrl From Pelago s, is a friendfinder, business applications with browsing functionality; Urban Spoon is a restaurant picker based on your location; NearPics is a location-aware photo browser and uses Google s Panoramio service; Weatherbug is a location aware weather service with predefined cities; StreetFlow; Yelp; Twinkle (Twitter); BrightKite. The list is growing every day. Mobile devices are equipped with wireless capabilities and users can go through several contextual changes as they move around. These changes are related to the movement of the user in his physical and social surroundings. By sensing their environment, mobile devices are capable of communicating and delivering ubiquitous services adequate to the situation. The dynamic nature of the system implies that as device context changes, delivered information can also change, due to an interoperation with the content server. The server will find the information, adapt it to the user context and format it for delivering. The history of the user is also taken in consideration, hence intelligent handling of the data. Mobile Internet is about functionality opposed to entertainment and e-commerce on the screen-based systems. We also believe that mobile experience merits its own design, customized to their needs, having the best practices, efficiency, and accessibility. We know that a small screen size doesn t match a 22 liquid crystal display (LCD). People use the portable devices when the information or functionality they need cannot wait, so they go to a computer screen. Therefore, developing for these screens and devices also brings more new issues, paradigms, and semantics to the world of mobile devices applications. 3. Developing for Mobile Devices Web development involves the creation of optimized Web pages for mobile devices. Standard Web programming languages, such as, hypertext markup language (HTML), cascading style sheets (CSS), JavaScript, and hypertext preprocessor (PHP) may be combined with available tools provided by companies or Web developers. These Web pages run on the mobile device browsers. iphone uses a mobile version of Apple s Safari. Main advantages of these developing tools are the following: ease and fast development; ease of user-access; ease updating; access to dynamic data; and offline server access. Native applications have more functionalities than Web applications. Therefore, a native application is the best choice for iphone users. iphone native applications have four distinct layers. The first layer includes the source code, the compiled code, and the software development kit (SDK) frameworks. The second has the Nib files, which contains the user interface (UI) elements and other objects (the design), and details about how objects relate each other. The third contains resource files (images, sounds, string, video), and finally, the fourth includes the Info.plist. This file saves details about application configuration. The proposed system assumes that all applications run natively on iphone. The programs were created using iphone SDK and Objective-C language. These tools offer several advantages in comparison with others. They include a more complete development environment, improved language depth, integration with SDK frameworks, iphone emulator, and software debugging. In terms of development aspects, Native application completely differs from a Web application optimized for the iphone excluding some similar tools in the SDK. For instance, Safari Web browser limits web applications while native applications are limited by the iphone operating system. In terms of price and business model, the differences tend to get bigger. Native applications are sold and distributed through the App Store. They can be downloaded directly to iphone or using itunes desktop version. Native applications follow the Apple software license agreement, keeping 3% of the price (if payment is required). This fee is paid for keeping the store clean and for support

34 4 EURASIP Journal on Wireless Communications and Networking Table 1: Main differences between iphone Web and native applications. Web applications Native applications Technology HTML, CSS, JavaScript Cocoa, Objective-C Deployment Web server App Store Frameworks Safari, limited iphone OS iphone OS Limitations Memory, JavaScript Runtime Cache, DB Installation Just online access Download or sync Findability URL/Web itunes/web Access Through Safari framework App Store download Offline usage No Yes transactions, server storage, helpdesk, and quality control. Main differences are summarized in Table 1. Communications speed and the fast information process need to be considered since the wireless communication tends to drain out the battery. Furthermore, assuming that access networking is limited in several locations, Web applications cease functioning. Web applications do not have a repository store like the itunes App Store. All the existing applications are on the Internet without any common reference. The advantages of the App Store could be reduced if Apple creates a website to store and control Web applications instead of being dispersed on the Web. Web applications have another drawback relying on Safari browser. Safari, like other piece of software, owns flaws, bugs, memory leaks, and, in a future upgrade, Safari App could jeopardize the Web application. Furthermore, Xcode tool offers an easy environment to create native applications in comparison with tools and resources available for Web developers [11]. The major advantage for Web applications is the programming language because it is not limited to Objective-C or object oriented programming. Moreover, contents are always updated and synchronized with Web server. Regarding native applications, they can only be downloaded through itunes while Web applications uses the user-friendly uniform resource locator (URL) entry in Safari. In terms of applications acquisition, buying natives is easier when compared with Web because the first uses itunes and the later needs to use a credit card each time an item, service or paid access is purchased. iphone SDK can be used to create both kinds of applications. Table 2 summarizes a comparison between developing Web Apps and Native Apps, in the developer perspective. 4. Web Applications Toolkits This section describes the four proposed Web applications such as Mobile, News, RSS, and Edition and the corresponding Web server created to deliver news from the Urbi et Orbi, which is online newspaper at University of Beira Interior, Portugal. These Web applications use generic libraries for structural support of the mobile devices browsers. The Mobile version uses the iwebkit free toolkit created for anyone wanting to create iphone websites. Versions News and RSS are based on Apple s Table 2: Comparison of main technical characteristics between Web and native applications in the developer perspective. Features/access Web applications Native applications Installation Add to main screen option Through App Store Initializing Open Safari bookmark or insert Click installed icon URL App Frameworks JavaScripted Custom iphone Frameworks Limited Full SDK Sandbox Safari sandbox App sandbox Cache Safari cache Sandbox files File system None Sandboxed OS Memory Page shared iphone OS (128/256 MB) Customizing User Web login App and iphone settings Sensors Limited Yes Location Yes Yes Accelerometers Limited Yes Cocoa Touch Limited Yes Network Auto Custom Multimedia Embedded Custom Database No SQLLite Offline No Yes UIKit that is the equivalent of AppKit for traditional OS X applications. The Edition version uses a JavaScript framework called WebApp.Net, which allows working with asynchronous JavaScript and XML (AJAX). Due to screen sizes and browser limitations, the proposed ubiquitous mobile multimedia applications also addresses design concerns. The user interface is designed to avoid horizontal scroll bars, and the most important news is located on top of the screen. The content design follows a top-down approach according to the importance and a left-right disposition, per levels, according to the intended detail of the news content. Furthermore, another important concern kept in mind regarding the design of user interface is Web accessibility on mobile devices, improving usability and user experience. Then, the Web content accessibility guidelines (WCAG) 2. is followed [12]. This study analyses the differences between the four proposed versions of Web applications related to their frameworks in order to determine betterment for the ultimate client. The Mobile Web application was initially used to create software specifically for the iphone because it has optimized code to work with this device. This version was also tested in different browsers outside iphone and the results have shown a complete website with no information lost. Two additional Web pages were created for displaying news through images ( Urbi in thumbnails and Urbi on images ), in order to navigate through the touch control with slide effect from iphone. The Web application

35 EURASIP Journal on Wireless Communications and Networking 5 structure is based on a top-to-down item table to display the most important information on the page beginning. This structure is also organized in left-to-right navigation to access detailed information on the right, as recommended by the UI rules that Apple uses for Web applications. Figure 1 depicts the website organization. Level (1) area is the front-page of the newspaper edition. This part is structured in Latest, Categories, and Others sections ( Urbi on Images and Urbi on Miniatures ). Level (3) presents the page with news details and it can be accessed from levels (1), (2), (2 ), and (2 ). Specifically, the detail page on level (3) contains all information regarding the new item information title, super-lead, picture, corpus, journalists, and published date. From here, users can follow more pictures, or other detailed pages, other websites, and multimedia contents. Level (2) illustrates an example of a item list of that category and it is ordered by the latest item publishing date. Each item is also linked to a detail page previously described as level (3). Part (2 ) is the Urbi on images Web page containing the images and captions that forms the list of news, and users can navigate through them horizontally using the Flick control action. Part (2 ) is the list of thumbnails containing all the images from the news edition. Each thumbnail also is linked to the referring detail page known as level (3). The News Web application presents a table with the list of news contained in the latest edition of newspaper. It was built on Apple Dashcode tool and uses the UIKit, a framework specifically created for iphone Safari. The major difference to previous framework is the inclusion of JavaScript functions to enable content delivery through XMLHttpRequest methods. This asynchronous request to the Web service is made while a browser renders the page in order to save time on page loading. The retrieved file of this request includes XML elements that will be used to populate the main HTML tags of the Web page. The RSS is another version of the Web application and it uses the really simple syndication (RSS) Web service provided by the same newspaper. This Web application gets a RSS feed (asynchronously) containing XML elements that will be parsed with HTML elements in the main page. The buttons on the detailed information page are linked to the Desktop version allowing users to continue his reading. Authors modified the template in order to include JavaScript functions to retrieve the time required to conclude the page display and the number of items included on the feed file. The RSS version does not requests any kind of multimedia files as opposed with other Web applications in order to compare how much JavaScript computation is needed to complete the information display. When these Web applications (both News and RSS ) are activated through the dashboard icon, the JavaScript included in the application hides both top and bottom navigation bars of Safari, making these Web applications almost similar to a native one. Then, the table list is displayed with maximum pixel height, providing to users the same experience of a similar native application. The last version, called Edition, uses WebApp.Net framework, which is an open source Web application framework, created by Chris Apers and it was designed to mimic the current iphone graphic UI. All content of this application is dynamically loaded through AJAX requests. On the contrary of versions News and RSS, Edition performs an asynchronous request only when the user needs it. Each request of detailed item information creates a different AJAX request. Image files with lower quality are created for this application in order to reduce costs-perdownload. The CSS file retrieved with the first page of this version is also different regarding the accesstimestamp. This version also includes search possibility by providing a form to input queries. The versions News, RSS, and Edition use XML- HttpRequest object to connect directly to XML data for feed updates without reloading the page. Normally two JavaScript functions are used to provide AJAX requests: loadxmldoc, processreqchange. These generic functions include object creation, event handler assignment, and submission of a GET request. After creating the object through an ActiveX constructor, several other methods (abort, getallresponse- Headers, getresponseheader, open, send, setrequestheader) and properties (onreadystatechange, readystate, responsetext, responsexml, status, and statustext) can be used to manage the connections. The XML data is then converted (parsed) into standard HTML content. A system prototype (testbed) was created to test and validate the proposal and to evaluate the performance in terms of speed and size. The procedure consisted of loading the Web application from different systems and platforms. To measure the size, requests, and loading speed, the following three different clients were used: (i) iphone 3G connected through Wi-Fi; (ii) iphone simulator; (iii) Safari browser running both on a Microsoft Windows XP machine and on Mac OS Leopard. Those latest clients were connected through cable network at 1 Mbps. Table 3 presents a comparison among the four versions of the proposed Web applications, evaluating the following downloads: the homepage of a given website, the corresponding first option webpage (called detail page), and the whole website. As may be seen, Edition version obtains best values for downloading the homepage and the website as a whole. The Mobile version is the smallest version when it comes to the homepage, but the time consumed to satisfy all the requests is bigger because it spends more time to download images. This version performs better on the parser time to display the homepage. Regarding the download of the detail page and the website as a whole, RSS performs better because does not download images. It does not show the best performance on the homepage because JavaScript files from the UIKit have greater size than the other version. News version shows the worst performance in the homepage download scenario because request all the images of the newspaper edition. It can be concluded that taking into account the main characteristics of each version and the above-mentioned considerations, Mobile performs better than other versions. Therefore, this version of the Web application is selected as a default version of the system when a Web application is requested.

36 6 EURASIP Journal on Wireless Communications and Networking Level 1 Level 2 Left to right level disposition 3 2 Level Tap to down information importance Figure 1: Mobile Web application structure and screenshots. Table 3: Performance evolution of the of Web applications taking in account size and speed of the transferred content between server and client. Mobile News RSS Edition First page access Page size 59 KB 368 KB 15 KB 9 KB Items requested Communications (seconds) 1.8 to to to.6.7 to 1.9 Runtime parsing (seconds).2.6 to Detail page Page size 81 KB 2 KB 5 KB 23.5 KB Communications (seconds) Whole website Levels Number of files Full site size 215 KB 183 KB 148 KB 149 KB After describing and evaluating the Web applications, we focus on the Web server. The proposed model for server is based on standard Web-based client/server architecture. Web standards and simplified models were used to design the system architecture in order to improve portability and scalability. Designing with Web standards offers a major benefit because once designed, it can be published everywhere [13]. The Web server is based on the Linux operating system, Apache HTTP server, PostgreSQL database management system, and PHP programming language, constituting the LAPP architecture. This server was designed in the perspective of ubiquitous computing integration on the solution. In this sense, a ubiquitous Web service was created to process and answer the client requests, illustrated in Figure 2. It can be seen that phase (1) of the process focuses on collecting and filtering information from Apache server and news database, phase (2) applies the templates to the data gathered on phase (1), and phase (3) adds the specific design files (CSS s and JavaScript) according to the ubiquitous results. Client requests from Edition, RSS, and News versions are received and handled by three Web services, one per Web application. For the Mobile version, ubiquitous computing is performed on each page request taking into account that version requests page by page. Each Web service uses specific classes and rules to handle client requests in a ubiquitous perspective. When a Web server receives a client request, ubiquitous decisions influence the images treatment in order to create and deliver images and thumbnails. A rule to change the image size and quality was created to reduce the file size and drop cost per download. This rule applies different compression algorithms according to the Web application version request. Upon reception of a request, another rule is invocated to deliver a specific CSS file, according to the client s time stamp, trying to improve contrast on the mobile device. Domain name system (DNS) reverse lookup to give the location of the client is also used. The time stamp is calculated with client location and server local time. The application can choose specific news for

37 EURASIP Journal on Wireless Communications and Networking 7 Web app stages HTTP Mobile browser Templates HTTP Desktop browser Storage CSS, javascript, images, multimedia files HTTP Other client apps Figure 2: Ubiquitous Web service model diagram. eachregion, offering the content in a different text language. Furthermore, if the IP address belongs to the user intranet, maximum quality of images is applied. A transmission of a big file is more efficient than a transmission of several small files [14]. This approach is considered on News and RSS versions. The other versions use a single request for each webpage. Therefore, News and RSS have better performance in comparison with the others, as may be seen in Table 3, taking into account the number of files downloaded. The server also stores other multimedia files not processed by ubiquitous rules, such as portable document format files (PDFs), audio and video files. 5. iphone Native Application The news for mobile devices (N4MD) is a simple application for viewing the weekly news produced in the Urbi et Orbi newspaper. This application runs natively on iphone devices and the primary difference from the previous Web applications is that the content is available for offline reading. This section describes the native application called N4MD.ap. It has been tested and validated. And it has the foundation for a worldwide quality application to be distributed through Apple App Store. The design of the native application follows the same approach as described for Web applications in Section 4. User interface has a top-down list of news and a left-toright navigational interface. The default primary view is the list-table-view Urbi et Orbi News. This window has four distinct areas, as shown in Figure 3. From the top to bottom, the user interface contains the following elements: (i) the status bar iphone s grey bar at the top with (from left to right): the cell signal and the carrier name, the network connection type (Wi-Fi, EDGE, or 3G icon), the clock information, and the battery status; (ii) top Bar Navigation Table s title Urbi et Orbi News and reload button on the right; (iii) the Table View List of news with thumbnails (default option) for each item, and at the bottom in orange, the table contents information (date and number of items); and (iv) the Navigation Bar The bottom bar in black with two options: Urbi et Orbi News and Settings. The Detail View appears when a row item on the first window is clicked. This is a subview of the primary view. Therefore, the Top Bar and relative content are related, which has the following three objects: the back button with the title of the table information Urbi et Orbi News ; the new information index and total of news; and the segmented buttons with navigation through items details. The Detail View appears in the white area below the orange Top Bar. This white area is a vertical scrollable object and with similar behavior to the above described in Web applications detail page. A user can find (from top to bottom) the following news elements: title, description, image, image caption, published date, category, author info, and full new corpus text. On the settings tab, the user can choose (through the slider button on the top) if he wants to see images or not. In the same view, user can find some control information that may be collected and sent to the server. These data (the device unique identification, the name, the system version, and the battery information) will be used to perform ubiquity. The iphone OS Library at iphone Dev Center [15] wasusedto create the application. The usual touch screen of iphone controls the navigation actions. The finger movements supported are Flick for scrolling and Tap for action or selecting. The N4MD user interface uses the Apple s suggested left-to-right navigation approach to go from top level to detail levels. It also uses

38 8 EURASIP Journal on Wireless Communications and Networking Web app on mobile safari Native iphone app Title, URL bar, reload button and search bar Top navigation bar Top navigation bar andreloadbutton Items list Items list Buttons for navigation, options, bookmarks and page controller Edition details Tab bar controller (item list and settings) Figure 3: Web versus native applications differences and characteristics. a Table View and a Scrollable View to display more information in a top-down structure without zooming or panning. The N4MD application does not support Rotation View because we primarily tried to mimic some native Apple iphone applications like Phone.app, Clock.app, or itunes.app, which do not support Rotation View as well. Normally, we wanted to use the iphone s features and frameworks integrated in the device operating system, and available to the developer, in order to ubiquitously connect, download, manage, and display the news content. So, the content will have to be downloaded and saved in iphone file system, allowing the application to access them offline. However, the application provides some form of user control to override the application and server ubiquitous decisions [16]. Ubiquitous computing is an important matter for mobile computing. Therefore, two types of ubiquitous decisions were created, one for the server and other for the native application. The server considers two types of applications, the native and the above-described Web applications. Windows Mobile, iphone or Google Android devices post different requests. Therefore, answers must be adequate and specific for each kinds of system. Users have different needs and there are other types of applications. Thus, the parameters sent to server, in order to make those ubiquitous decisions are the following: user agent, client device type, network type, screen width, battery, categories, sections, items, and edition date (as shown in Figure 4). Communications with the server are also important for the application speed and for the device battery life. So, one of the requirements for the applications should be related with network type (Wi-Fi, 3G, or EDGE) and images. To ubiquitously avoid image downloading, the user can choose an option to override images or not. The system also have more rules to define the default option for image downloading, to seamlessly create image thumbnail and asynchronous downloading. Other rules for avoiding images downloading, such as the following: if the application uses the EDGE network connection or if images are already cached. When there is no image to show, the application adapts the Detail View and do not present Image Caption and image placeholders. The N4MD application started with the creation of a default Xcode template The iphone OS Window-base Application [11]. By doing that all the files and documents were automatic created, and some of them had a few lines of codes allowing us to Build and Go, even without changing a line. Then we had to modify some of the existent bundle files (png images, visual elements on the Xib, and the application default.h and.m files). In the init stage, while launching, the application gets user preferences and starts to build the window with components like the Navigation Bar. Then, the application checks if it has a network connection. If it has network connection, it sends a request for server statistics. The next stage is getting the articles (the news). If there is a network connection, the application makes a XML http request, if not, it tries to load a previous saved plist file. In neither case, the application continues by parsing the information. Then, the display stage happens when the information is loaded and we can see the Table View with article title, description and thumbnail. After the first display stage, the application enters in a cycle (waiting for a user s input that change status in order to make new display changes) or quits. The user-default feature is used to save three state variables for user and application settings in the applications bundle directory. This information is useful as it seamless shows the last viewed window before exiting. Therefore, the user does not have to navigate through the application all over again. The data from the XML http request in a form of a plist file which is also a XML format is saved within the application sandbox. The downloaded images are saved on Documents directory. When user click on the reload button and the flag for Internet connectivity is on, the applications erase all the downloaded data from the bundle

39 EURASIP Journal on Wireless Communications and Networking 9 Ubiquitous mobile multimedia apps Mobile device Server 1-Connectivity and speed 2-Location, timestamp 3- User info and preferences 4-Device info (name, model, OS version, unique ID, battery status) Async message Message delivery 1-HTTP request, timestamp 2-DNS reverse lookup 3-Find and select information 4-Adapt multimedia files 5-Compress and package 1-Loading 2-Parsing Figure 4: Sequence diagram model with context-awareness information and process variables to ubiquitously perform output results. directory and starts the asynchronous communication with the server to get refreshed information. Moreover, screen size forces a top-down design for information display, the Table View is one of the most used components for the iphone regarding its vertical characteristics. Mobile devices do not have characteristics that a desktop computer has, such as the following: all the mouse movements or click events; not all views have page zooming; text size adjustments, options for fonts sizes types, form controls or form saved information. Therefore, those characteristics cannot be used in the ubiquitous system. The iphone OS recognizes some special links like: itunes:, phone:, mailto:, making communication between applications more dynamic. For improvement of speed, the application depends on the network speed and number of items to be downloaded. It only occurs when threaded url connections are used to speed up the application, and it is provided by the asyncimageview class. The application contacts the server in two distinct stages. The first contains information about the device (name, model, localized model, system name, system version, and unique identifier). The second communication occurs when the application makes a request for data (news content feed), and at this point the information to server contains battery status and client type. These stages are important for the success of the ubiquitous system. Other requests are motivated by references of thumbnails and images included in the XML file. This request just happens if the user allows image loading on settings, and if the used network is 3G or Wi-Fi. In order to create ubiquity in N4MD application, the communication between server and client was forced. The client has to collect context-aware information and send it back to the server [17]. Then, the server must determine the location-awareness of the device (so far we do not use GPS), save the user statistics, and use the information received in order to deliver the desired information content in a specific format. 6. Conclusions and Future Work Applications for mobile devices are gaining their own market as the seventh way of communications by being less equal to small online versions of the bigger brothers, and taking advantage of the opportunities as they popup. Web applications and native applications must co-exist due to connectivity issues and offline reading. In terms of design, they are similar in many ways and also share architectural and structural models. But their final purposes make them unique to each other and have usefulness on their own way. Moreover, Web applications are oriented for cloud computing, peer collaboration and synchronous connectivity. On the other hand, Native applications take advantages of mobile devices characteristics and access to more frameworks. Applications for mobile devices must use ubiquitous computing techniques in more effective way. By addressing fundamental topics, in a quest to unleash the full potential of data consumption, the usage of location and contextawareness in mobile devices are changing our life quality for the betterment. Those above described systems combine the portability of a Web Service with the mobility of users to overcome the limitations of mobile devices. This paper proposed and described in detail a system for delivering news using wireless multimedia applications and transmission techniques to mobile devices, using proposed platform to Apple s iphone. For study purposes, four

40 1 EURASIP Journal on Wireless Communications and Networking different versions of a University online newspaper were created. These versions were produced to specifically provide an important parameter of their target mobile device. While testing size, runtime speed, connections, design, and usability, our study shows that each version has its own advantages as we have expected. Ultimately, the server decides the best version for the specific client and delivers the corresponding application. But, the standard default (to be used with all around mobile devices) is the Mobile version. This version was developed with the iwebkit, that proved to be, the best all around accessibility platform for general mobile devices multimedia Web applications. The decisions for ubiquity occur in the server-side and in the application itself. They use the information of each other to decide the best for several parameters, and, ultimately, the best for the user client. This new level of ubiquitous collaboration brings out the best of two worlds the server-side power, and, the client-side context, location and sensor awareness, making the delivering of news seamless, visually effective, communication efficient, configurable, or ubiquitously personal. The application resolves some of the wireless networks limitations, reads the context-awareness of mobile devices, communicates with the server and understands the information received. The proposed applications are conceptually simple as they proved to be the best way. Like the open software, the usage of standards in mobile development is a necessity ( must have ). It opens doors to new applications or services, improves compatibility with other devices, and all is under control of the programmer. The news delivery to all kinds of the readers was also studied. The proposed ubiquitous computing, software design engineering, service architecture, and news content, bring innovation and contribute to improve communication in this modern world. In a near future, the proposed applications can be improved in several ways. The following items are suggested during tests and debugging stages and can be found in other applications studied. It seems worthy to make it for the application. In the Web applications, the following may be performed: (i) create advanced client information client history and online statistics that would be added on the ubiquitous system to create more personal, seamless, and user-oriented news content; (ii) add multimedia capabilities to the Web application and Web server (streaming server) enlarge compatibility with automatic conversion on codecs, containers, sizes, and file formats; (iii) add user registration access to post comments, news, suggestions, uploading files, images, and slideshows; (iv) add options like Send content to ; and (v) add them to Twitter, Facebook, LinkedIn, Hi5, or other social networks. For Native applications, some features for the server side of the Web applications type may also be proposed. We can also perform the following: (i) add location-awareness given by GPS to ubiquitously choose news, language and design for a specific region; (ii) improve native application ubiquity by using more information from sensors, location services, settings, connectivity, and specific device characteristics; (iii) embeded multimedia elements such as video, audio, photo slideshow, and other Web pages without quitting from the application; and (iv) as recommended by Apple improve the software design. Make it iphone, by bringing innovation on design, more information on display, new features or services, usability, and accessibility. Acknowledgments Part of this work has been supported by Instituto de Telecomunicações, Next Generation Networks and Applications Group (NetGNA), Portugal, and by Online Communications Lab (LabCom), University of Beira Interior, Portugal. References [1]C.A.DaCosta,A.C.Yamin,andC.F.R.Geyer, Toward a general software infrastructure for ubiquitous computing, IEEE Pervasive Computing, vol. 7, no. 1, pp , 28. [2] G.OrtizandA.G.DePrado, Mobile-awarewebservices, in 3rd International Conference on Mobile Ubiquitous Computing, Systems, Services, and Technologies, (UBICOMM 9), pp. 65 7, October 29. [3] T. T. Ahonen and A. Moore, Communities Dominate Brands: Business and Marketing Challenges for the 21st Century, Futuretext, 22. [4] C. Pettey and H. Stevens, Gartner Says Enterprise Mobile Phones Will Replace Desktop Phones in North America by 211, Gartner, 29, it/page.jsp?id= [5] N. Covey, Mobile Internet Extends the Reach of Leading Internet Sites by 13%, Nielsen Media Research, 28, [6] T. O Relly, What is Web 2., O Reilly Media Inc., 25, [7] J. Zhang, A. Hämäläinen, and J. Porras, Addressing mobility issues in mobile environment, in Proceedings of the 1st Workshop on Mobile Middleware: Embracing the Personal Communication Device, Co-located with ACM/IFIP/USENIX International Middleware Conference (MobMid 8), Leuven, Belgium, December 28. [8] M. Weiser and J. S. Brown, The coming age of calm technology, in Beyond Calculation: The Next Fifty Years of Computing, P. J. Denning and R. M. Metcalf, Eds., Springer, New York, NY, USA, [9] J. Steele, comscore Reports 6.5 Million Americans Watched Mobile Video in August, comscore, Inc., 28, Events/Press Releases/28/ 1/Mobile Video. [1] S. K. Mostefaoui, Z. Maamar, and G. M. Giaglis, Advances in Ubiquitous Computing: Future Paradigms and Directions, IGI Publishing, New York, NY, USA, 28. [11] E. Sadun, The iphone Developer s Cookbook: Building Mobile Applications with the iphone SDK, Addison-Wesley, Boston, Mass, USA, 29. [12] B. Caldwell, M. Cooper, L. G. Reid, and G. Venderheiden, Web Content Accessibility Guidelines (WCAG) 2., W3C, 28, [13] D. Hazaël-Massieux, Return of the mobile style sheet, A List Apart, no. 275, 29, return-of-the-mobile-stylesheet.

41 EURASIP Journal on Wireless Communications and Networking 11 [14] S. J. Zilora and S. S. Ketha, Think inside the box! optimizing web services performance today, IEEE Communications Magazine, vol. 46, no. 3, pp , 28. [15] D. Mark and J. LaMarche, Beginning iphone 3 Development: Exploring the iphone SDK, Apress, New York, NY, USA, 29. [16] P. Tarasewich, Designing mobile commerce applications, Communications of the ACM, vol. 46, no. 12, pp. 57 6, 23. [17] G. D. Abowd, G. R. Hayes, G. Iachello et al., Prototypes and paratypes: designing mobile and ubiquitous computing applications, IEEE Pervasive Computing, vol. 4, no. 4, pp , 25.

42 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 1 pages doi:1.1155/21/ Research Article Power-Aware DVB-H Mobile TV System on Heterogeneous Multicore Platform Yu-Sheng Lu, 1, 2 Chin-Feng Lai, 1 Chia-Cheng Hu, 3 Han-Chieh Chao, 4 and Yueh-Min Huang 1 1 Department of Engineering Science, National Cheng Kung University, No.1, University Rd., Tainan 71, Taiwan 2 Business Customer Solutions Laboratory, Chunghwa Telecom Laboratories, No. 12, Lane 551, Min-Tsu Rd. Sec.5 Yang-Mei, Taoyuan 326, Taiwan 3 Department of Information Management, Naval Academy, No. 669, Junxiao Rd., Zuoying District, Kaohsiung 813, Taiwan 4 College of Electrical Engineering & Computer Science, National ILan University, No. 1, Sec. 1, Shen-Lung Rd., I-Lan 26, Taiwan Correspondence should be addressed to Yueh-Min Huang, huang@mail.ncku.edu.tw Received 19 March 21; Accepted 15 June 21 Academic Editor: Liang Zhou Copyright 21 Yu-Sheng Lu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In mobile communication network, the mobile device integrated with TV player is a novel technology that provides TV program services to end users. As TV program is a real-time video service, it has greater technical difficulties to overcome than a traditional video file download or online streaming, especially when TV programs are played on handheld devices. A challenge is how to save power in order to provide users with longer TV program services. To address this issue, this study proposes a mobile TV system on a heterogeneous multicore platform, which utilizes a Digital Video Broadcasting-Handheld (DVB-H) wireless network to receive the TV program signal, thus, saving power according to the features of DVB-H TV signal and heterogeneous multi-core. 1. Introduction Along with the progressive digital TV broadcasting technology, TV viewing is no longer restricted by time or space; the new trend is to watch digital TV programs through wireless mobile devices. At present, watching TV on a mobile phone device can be performed in two ways. Service providers can transmit TV program data to mobile phone users by 3G network, or a base station can transmit TV programs through a Digital Video Broadcasting network [1 3]. The main difference between 3G and DVB network is that 3G network transmits data through on demand wireless network communication, hence, there will be transmission rates and bandwidth limits if too many users access this network at the same time. As to DVB-H, it transmits TV programs through the broadcasting transmission of TV base station; hence, there will be no transmission network congestion. Three issues have been studied regarding the mobile TV system: (1) TV signal transmission technology and how to enhance TV signal fault-tolerance or increase signal transport efficiency in order to improve display quality of TV programs; (2) mobile TV application developments and provision of personal context aware services, recommending suitable TV programs according to user habits and preferences of watching TV; (3) how to enhance display quality, provide smooth TV programming if delays occur, and reduce power consumption in mobile TV players [4 7]. Concerning power-saving issues, two parts are discussed: (1) components of receiving TV signals, how to design receiver startup schedule while receiving a TV program signal to save receiver power; (2) design a power-saving play mechanism according to TV program signal features, after received TV signal is converted to digital data by the demodulator (Figure 1) [8 11]. Therefore, this study proposes a power-aware DVB-H mobile TV system on a heterogeneous multicore platform. This system is implemented in two major parts: a front-end buffer control mechanism and a parallel DVB-H TV signal decoding model. When receiving a DVB-H TV program signal from a base station, signal is demodulated to generate video and audio data. As video bit rate, quality, and resolution are directly related to content complexity, running too many buffers will consume power, while too few buffers will cause the program to fail to be played successfully. Hence, this

43 2 EURASIP Journal on Wireless Communications and Networking Broadcast content provider Broadcast network operator Mobile TV MEPG2 TV service MEPG2 TV service. MEPG2 TV service MUX TS DVB-T modulator 8k 4k 2k DVB-H TPS Broadcast content provider. Broadcast content provider DVB multiplexer DVB-H DVB-H transmitter Audio AAC MPEG2 section Video H.264 Service information Program specific information DVB-H IP encapsulator MPE MPE Time -FEC slicing 8k 4k RF DVB-H IP encapsulator 2k DVB-H TPS Channel TS RF DVB-H IP encapsulator Existing Time MPE slicing -FEC MPE New to DVB-H IP Figure 1: DVB-H Mobile TV Workflow. Figure 2: DVB-H/T System Archietcture. paper proposes a front-end buffer control mechanism to configure appropriate buffers according to the TV program video features, in order to utilize buffers and save power. The parallel DVB-H TV signal decoding model uses a data partition processing method to run parallel DSP decoding of DVB-H videos on a heterogeneous multicore platform. It also schedules videos according to the DVB-H video features, in order to reduce data dependency among the frames on a multicore platform. The remainder of this paper is organized as follows. Section 2 introduces DVB-H specification and the parallel decoding technique; Section 3 presents the overall architecture of the DVB-H mobile TV system, the front-end buffer control mechanism, and the processes and methodology of the parallel DVB-H TV signal decoding model; Section 4 discusses implementation and result, and Section5 gives conclusions. 2. Related Work 2.1. Digital Video Broadcasting-Handheld. Digital Video Broadcasting-Handheld (DVB-H) is based on Digital Video Broadcasting-Terrestrial (DVB-T) specification and provides a solution to lower receiver power consumption and improves mobile receiving performance [12 18]. Figure 2 shows the outline of the DVB-H/T system specifications for common TV broadcasting programs using the DVB- T signal transfer mode. Senders can use an A/D converter to convert the analog video and audio signals to a digital signal, respectively, and use a Moving Picture Experts Group 2 (MPEG-2) codec technique to convert TV program data into MPEG-2 format. DVB-H service data are compressed and encapsulated into an IP packet then encapsulated into the transmission stream through a Multiprotocol Encapsulation (MPE) mechanism. Meanwhile, the time slicing data stream is added. Along with other DVB-T TV services, the multiplexer multiplexes it into a larger transmission stream (or multiple program transmission stream) before sending the data in a DVB wireless network. At the receiver, if a client wants to receive certain services, the receiver front-end circuit must run continuously in order to obtain the complete transmission stream. Then, the demultiplexer extracts the video, audio, and data information streams of the selected programs and delivers this information to the video decoder, audio decoder, and other applications for processing. The sender Multi-Protocol Encapsulation-Forward Error Correction (MPE-FEC) and time slicing mechanisms are collectively called the DVB-H IP-Encapsulator, while the receiver reverse recovery portion is called the DVB-H IP- Decapsulator. The overall DVB-H container format is shown in Figure 3. The IP data container format for each layer of DVB-H is shown in Figure 3, as an IP packet in the MPE section and redundant data in the FEC section. After Section format encapsulation, the MPE and FEC sections are connected end to end according to the encapsulating sequence to form a section data string. Then, it begins to slice the first and all of the other 184 bytes of each section data string. A 4-byte transmission stream header is added to the front of the 184-byte data length in order to complete a transmission stream encapsulation or MPEG2 transmission stream packet. Its data length is 188 bytes, with two major parts. The first is a data front-end header that occupies a 4- byte length with the available information, including a Sync. Byte = 47 hex for synchronizing the emitter and receiver, error indications, and stream packet recognition. The second part is the data transfer payload, which length is 184 bytes. In Figure 3, above the IP packet is the User Datagram Protocol (UDP) and the Real-time Transport Protocol (RTP). The top layer is compressed video data, where IP and UDP packets add their packet headers. The RTP packet is encapsulated and used to bear the H.264 images and AAC compressed voice, as RFC3984 specification Parallel Decoding. One ideal parallel process could double the system processing efficiency; however, when coding/decoding a picture, there exists a data dependency problem [19 25]. As the video image format contained in the DVB-H TV signal is an H.264 baseline format, this section

44 EURASIP Journal on Wireless Communications and Networking 3 DVB-H protocol stack TV player Encoded data Entropy decoding IQ/IT Deblocking Decoded frame H.264 video AAC+ audio ESG EPG Intra prediction MC prediction Buffer UDP IP Figure 5: Functional Partition Decode Workflow. Sequence decode Parallel decode Waiting MPE/MPE-FEC MPEG2-section MPEG2-TS DVB-H bearer PSI SI Figure 3: DVB-H Protocol. I P1 P2 P3 P1 I Time P3 P2 Figure 4: Sequence/Parallel Video Decode. Waiting introduces the H.264 image feature. In H.264 decoding, pictures are divided into I frame, P frame, and B frame, where P frame is decoded according to the I frame picture data, and the B frame refers to the picture data of the I frame and the P frame. Unless there is good parallel processing, data collision will occur, as shown in Figure 4. When decoding two interdependent pictures, even when both pictures are simultaneously processed, the other picture must wait for a decoded reference before decoding. Therefore, how to utilize parallel processing to shorten the operation waiting time is the focus of many studies. Parallel decoding is divided into two orientations, a function partition, and a data partition, detailed as follows Function Partition. The H.264 decoding process can be roughly divided into entropy decoding (ED), inverse quantization and inverse transform (IQ/IT), intra- or interprediction (PPC), and deblocking filter (DF). As shown in Figure 5, the function partition divides the entire H.264 decoding process into independent tasks. The main purpose is to use a balanced processing concept to configure the tasks required by each processor so that each processor can share the load, and processing can be accelerated. The advantage of this mechanism is that it can easily and extensively eliminate data dependency, while its disadvantage is that its task division is subject to a number of processors Data Partition. Data partitioning divides decoding data into partitions, which are computed by different processors. Each processor performs the same data operations, but process different data units. Previous literatures have studied how to divide data while avoiding data dependency; the partition includes groups of picture (GOP) levels, frame levels, slice levels, and macroblock levels. Their features are detailed below. GOP level [21, 22]: it divides the video segments in GOP, allocates each GOP to each processor to decode, as each section of the GOP can independently run decoding. Decoding in this manner can increase processing quantities at linear speed [23]; however, applying this technique requires large memory space to save the decoded GOP fragments. Frame level: this parallel decoding method allocates each single picture to a respective processor to operate, where preanalysis sorting operations or a tournament algorithm is adopted in order to process picture allocation. Primarily, two pictures without data dependency are found and simultaneously operated. Flierl and Girod [24] proposed a B Frame parallel decoding method, which is suitable for traditional coding methods, and because the B frame is not referred to by other picture, no data dependency will occur. B frame is analyzed first and allocated to different processors for decoding. However, regarding H.264 coding, B frame can be a reference for other pictures, therefore, is not suitable here. Slice level [25, 26]: in H.264, the Slice is the smallest independent decoding unit, meaning that a single Slice can independently run decoding. Therefore, similar to GOP level partitioning, various Slices are allocated to various processors for decoding. As compared with the GOP Level, this parallel method is more favorable to memory utilization without extra analysis sorting. However, its main disadvantage is that, slice divisions can range from a macroblock to one entire frame, where memory use, scalability, and balance

45 4 EURASIP Journal on Wireless Communications and Networking Inter prediction Encoded data Front-end buffer Intra pred Intra DF Intra DF Decoding MB Intra pred Decoding MB Parallel DVB-H TV signal decode model MPU DVFS controller Encoded data DSP1 Encoded data DSP2. Multi-core Decoded frame Figure 6: Macroblock Decoding References. Figure 7: System Architecture. remain unsatisfied. Roitzsch [27]proposed a Slice-Balancing algorithm, which can improve its scalability; however, it is for the coder end and cannot be applied in the decoder end. Macroblock level: a frame consists of many picture macroblocks, and each macroblock is allocated to a variable processor to operate. This scheme has the best scalability and balance; however, it requires the most directions for solutions, such as data dependency. As shown in Figure 6,in H.264 decoding, each macroblock must refer to its neighboring macroblock to operate. The main concept of macroblock level parallel decoding is to locate two macroblocks, which are without data dependency in order to speed the rate decode of decoding. Van der Tol et al. [28] proposed an echelon sorting process to solve the data dependency problem. Although this sorting method can improve the speed of decoding, it is subject to a number of processors. In cases of high resolution pictures, this algorithm requires complex operations and several processors. Chong et al. [29] proposed scheduling with parse, render, and filter, locating the dependency relation of each macroblock prior to sorting. Azevedo et al. [3] proposed a 3-D-Wave method to integrate frame and macroblock levels in order to locate decodable macroblocks across all frames, which solves scalability and data dependency. However, the algorithm applied in a super multicore algorithm remains difficult to implement. Although many papers have addressed the parallel decoding problem, few studies have focused on the power consumption for DVB-H TV program. To this question, this paper presents a novel architecture to obtain a front-end buffer control mechanism and a parallel decoding model to increase the speed of decoding TV program and reduce the power consumption according to picture complexity. 3. System Architecture Figure 7 shows the system architecture proposed in this study. After accessing the system, DVB-H streaming data are saved to an external memory; the main processor unit (MPU) initializes and starts the digital signal processor (DSP), and then, the streaming data is moved to the internal memory of DSP to decode. The system waits for completely compressed data before entropy decoding (ED), and the H.264 baseline profile is similar to common compression specifications, as per context-adaptive variablelength coding (CAVLC). Inverse quantization and inverse transform (IQ/IT) are done on compressed pictures, then intraprediction or interprediction is made according to the video format. Finally, a deblocking Filter (DF) is carried out on the imaged pictures to eliminate boundary effects and improve image quality. After decoding, if the picture is referred to, then it is saved to internal memory. Otherwise, the direct memory access unit (DMA) will move the data to an external memory, and the MPU will transfer the picture data to a frame buffer for displaying. As to H.264 specifications, it adopts the concept of a network abstraction layer (NAL) to adapt to a progressive streaming application. The NAL Unit is the streaming unit, which includes a header and the contents of compressed video data or decoded auxiliary information (e.g., resolution, display time, and video information). The MPU locates the compressed video data and decodes it according to the auxiliary information. The streaming data are parallel sorted. Dynamic voltage and frequency scaling (DVFS) system decoding prediction is performed according to sorted video dependency and video formats [31 33]. In most cases, the DSP system codes/decodes videos through a heterogeneous multicore platform. Therefore, this paper focuses on parallel decoding of a single MPU, coupled with a multi-dsp-core platform. The MPU controls parallel planning, DVFS prediction, and settings of system. With the front-end processing design, parallel processing of the DVFS system can be performed without changing the DSP decoding process. The results of the multicore platform could also be applied to a parallel decoding design on another platform Front-End Buffer Control Mechanism Group of Picture Unit of DVB-H Video Content. DVB- H TV programming is composed of H.264 formatted images, and each H.264 stream consists of numerous group of picture (GOP). Each GOP consists of I-frames, B-frames, and P- frames, where I-frame is used for DCT-based compressed digital video frame, and B-frame and P-frame are used as backup frames to enhance compression ratio. Due to picture interdependency, which comes from the motion vectors of the I, P, and B frames in GOP and compensation coding, when the DVB-H TV signal accesses a system to decode, parallel decoding of DVB-H TV signal is performed according to GOP features. Since H.264 GOP size is subjected to the complexity of a picture in its content, this study

46 EURASIP Journal on Wireless Communications and Networking 5 DVB-H TV program bitstream DVB-H TV program bitstream No Read DVB-H TV program bitstream No No No No Start code SPS PPS SEL IDR Yes Yes Yes Yes Yes Get SPS information Get PPS information Get SEL information Get IDR information Figure 8: DVB-H TV Signal Reference Parse. Select scalability Reference parse information Select coded slice Figure 9: GOP Divide Workflow. GOP regulates the size of the front-end buffer according to the information provided by each GOP header in order to save power. First, an entire DVB-H bitstream is received from a receiver, then the entire bitstream is analyzed according to its streaming manner, and 1 is the start code of the NAL unit, and then moves to next byte and determines the NAL unit type. If the NAL unit type is SPS, PPS, or SEI, it collects the sequence, start position, data size, instantaneous decoder refresh (IDR) sequence, percentage of IDR sequence over entire coded video sequence, and the start position. The entire process is called reference parse, as shown in Figure 8. According to the parse result, the size of each GOP and its number of frames contained can be known. Then, the IDR feature is used to select the proper number of intrakey pictures, and each intrakey picture is converted into an IDR picture, as required. This study adds IDR pictures for original reference and changes its slice header syntax element, in case of decompression failure or incomplete state as each GOP unit changes its tunable combination (Figure 9) Front-End Buffer Configuration. One ideal parallel processing can increase the speed of system processing; however, data dependency often occurs when coding/decoding multimedia data. This study designs a simple parallel decoding architecture that could solve data dependency, as shown in Figure 1. The architecture utilizes a buffer mechanism to I1 B2 B2 B3 I1 P4 B3 P4 I5 B6 B7 B6 I5 P8 Parallel model P8 B7 I1 P2 I1 P3 P5 P7 P3 P4 P P P P5 P P P P6 P7 P2 P4 P6 P P P8 Figure 1: Parallel Decoding Architecture. store stream data and then waits for the next independent picture before decoding together. A buffer must be set for GOP to be decoded, and the buffer size is also a focus of smooth TV programming, as excessive buffers may occupy too much system memory, and insufficient buffers may have too few fragments leading to image delay. However, in a prolonged dependent video format, this may cause latency and buffer size problems. It is because prolonged latency is unacceptable to the streaming process, and buffer size is subject to hardware size, which influences power consumption. Thus, this study designs a dynamic allocation mechanism and defines a front-end buffer (FEB), which is used to store streaming data for parallel processing. Since FEB is not overflowed, independent parallel decoding is adopted when the system locates the next independent picture. However, when streaming data overflows FEB and no independent picture appears, dependent parallel decoding is used to meet data streaming features, and the frame and macroblock level for the integrated parallel decoding concept are applied to synchronize decoding. This decoding method is performed by the transfer of a synchronous signal. The steps for choosing FEB to satisfy a streaming system are defined as follows: Suppose that T acp denotes a time coefficient acceptable to the end user, then, the time spent on the entire computing process is P8 T = FEB S + T proc + D, (1) where, T proc denotes the DVB-H data streaming speed, and D denotes the FEB derived latency. To comply with a streaming system, the following equations must be satisfied: T acp = FEB acp + T proc + D>T, (2) S ( Tacp T proc D ) FEB acp = S. (3) As the system is a frame-based case, suppose that T proc denotes the processing time required for a picture, hence, FEB acp is the preset size.

47 6 EURASIP Journal on Wireless Communications and Networking DVFS table DVB-H TV program bitstream Reorder bitstream Reorder bitstream 1 Reference frame buffer DDR DDR SD SD1 SD2 SD SD1 SD2 AXI 5 Local memory 5 Local memory MPU 7 DSP core DSP core 7 AXI DSP1 DSP2 Figure 11: Data Parallel Architecture Parallel DVB-H TV Signal Decoding Model. As to a multimedia decoding system, the two most important methods for reducing energy consumption are (1) elimination of time slack, (2) prediction of processing quantity required for the next picture and preseting the voltage/frequency. To achieve these two objectives, this study uses a simple, yet practical, concept of using front end and back end buffers to construct the entire DVFS architecture, realize the parallel mechanism, and eliminate the time slack. The constant decoding time of each picture is defined as a deadline. However, rather than changing a deadline schedule, a power approximation method is used to predict system voltage and frequency and corrects the system voltage and frequency according to the weight of each task in the decoding program. The power consumed by a processor during the CMOS manufacturing process is defined as P = C eff V dd 2 f, (4) where, C eff denotes an effective switch capacitance, V dd denotes working voltage, and F denotes working frequency, and the frequency versus voltage relation can be expressed by the following: f = K (V dd V t ) a V dd, (5) where, K is a constant, V t denotes the threshold voltage, and a = denotes the electron coefficient [25]. The time spent in executing a task is called the workload and is defined as T Proc, which can be calculated from the following: T Proc = C f, (6) where, C is the number of cycles required for this task during system computing, and by substituting (2) for(3), we can obtain V dd T Proc = C F = C K (V dd V t ) a, (7) according to the energy formula E = P T Proc. (8) The architecture of data parallel implementation is shown in Figure 11. In this architecture, MPU analyzes the complexity of TV bit streams firstly, sets voltage and frequency for each TV bit stream according to the materials of DVFS Table, and then transfers TV bit streams to DSP core for decoding. After decoding TV bit streams, the decoded data is delivered to Reference Frame Buffer for playing TV programs. We can reduce the power consumption via DVFS adjustments. Different from a program-oriented parallel architecture, which evenly allocates one identical picture to various DSPs to decode, the data-oriented parallel architecture sends a picture to different DSP for separate decoding. Therefore, when a primary data stream arrives, the MPU analyzes its information, sorts, reallocates the pictures, then sends it to the DSP for decoding. The DVFS mechanism adopts a frame-based direct mechanism in order to dynamically tune the DSP voltage and frequency, as per the ratio of picture data size/decoding seconds. The difference is dynamic tuning performed during the workload period. The tuning equations are expressed by (7) and (8). Data-oriented parallel processing has another problem, when simultaneously decoded pictures are dependent, and then there is data synchronization problem. This study adopts a picture coordinate synchronizing method, where a macroblock coordinate of the picture being decoded is

48 EURASIP Journal on Wireless Communications and Networking 7 (N) frame size decoded time DVFS value (N) frame size decoded time DVFS value (N) frame size decoded time DVFS value DVFS database DVFS manager DVFS event Parallel DVB-H TV signal decode model DVFS controller Figure 12: Prediction Mechanism for DVFS. Table 1: Testing Sample. Video Name Type Resolution Frame type Mobile Foreman Highway Silent Football Speaker Motion Silent Motion Silent Motion Silent QCIF ( ) QCIF ( ) QVGA (32 24) QVGA (32 24) VGA (64 48) VGA (64 48) 1I3P6B (IPBBPBB) 1I3P6B (IPBBPBB) 1I3P6B (IPBBPBB) 1I3P6B (IPBBPBB) 1I3P6B (IPBBPBB) 1I3P6B (IPBBPBB) Frame number (X cur, Y cur ), and the coordinate of its reference picture is ((X ref, Y ref ), then, their relation formula is as follows: Y ref Y cur + N mac Y ref, Y cur Y Max, (9) X ref X cur + N mac X ref, X cur X Max, where, X Max and Y Max denote the numbers of boundaries of the divided picture, and N sub denotes the number of macroblock referenced by the macroblock Prediction of System Cycle Number C. In this paper, an offline come online mechanism is adopted in order to lower the prediction error rate. The method is shown in Figure 12. DVFS is divided into two parts in order to implement the entire design architecture, and the offline mechanism is implemented on the MPU. On the DSP, the online mechanism is realized through the dynamic tuning of voltage and frequency the decoding process. Prior to a system decoding process, the DVFS Model determines the initial voltage and frequency, set according to the picture format, previously decoded picture size, and decoding time. In the decoding process, the DSP end decides the dynamic tuning of voltage and frequency, set according to time spent and the dependency data of each function in the decoding process. Without knowing the exact cycles consumed by the next decoding picture, this study first predicts through offline statistics and builds a database in the MPU part, which records time, type, voltage, and frequency required for analyzing a video. Based on the analysis data of the previous picture of the same type, this study defines two formats: (1) the averaged cycles, C avg ; (2) the cycles of a reference picture, C prev and video content variation rate, α C j C j 1 α =. (1) C j The predicted cycles vary with the type of picture in the variable format. In a more static neighboring video, prediction by the previous format would be better. In a dynamic video, the average prediction would be more accurate. Therefore, different prediction effects are designed for each picture format. As shown in (11), as I frame is an

49 8 EURASIP Journal on Wireless Communications and Networking Deadline miss (%) Power cumsuntion (J) Figure 13: Experimental Environment. Mobile Foreman Highway Silent Football Speaker Sequence Sequence with DVFS Parallel Parallel with DVFS Error rate (ms/frame) Mobile Foreman Highway Silent Football Speaker Sequence Sequence with DVFS Figure 15: Deadline Miss. Parallel Parallel with DVFS Figure 14: Power Cumsuntion. Mobile Foreman Highway Silent Football Speaker independent picture without a reference picture, the average prediction mode is adopted. In B frame, the picture refers to both a previous and a next picture, thus, the cycles of the B frame are estimated from the average of the reference pictures. As to P frame, the average forecast or reference picture forecast is adopted according to α: C avg, for I frame C avg, for P frame && α 15% C est = C ref, for P frame && α<15% N C ref, for B frame. N 4. Experiment and Result (11) The experimental environment is shown in Figure 13. This study uses Fluke 8846A to measure the power consumed by each DSP during decoding and displays the data on a computer through FlukeView software. First, for dynamic or static variable resolution TV program in Table 1, the common DVB-H TV signal decoding and parallel DVB- H TV signal decoding methods are used to measure the consumed power. The results are shown in Figure 14, where Sequence Sequence with DVFS Figure 16: Error Rate. Parallel Parallel with DVFS 39% of the total power loss is saved. This parallel DVFS architecture shows that a Data parallel architecture can perform better for power control of a dynamic picture. After decoding, this study lists the decoding seconds, data size, and error occurrence of each picture, and calculates its error rate according to following formula: (Test T dec ) 2 err rate =, (12) Num frame err miss = Num miss 1%, (13) Num frame where, T est denotes the estimated decoding time, T dec denotes the actual decoding time, Num frame denotes the number of pictures, and Num miss denotes the number of pictures with missed deadlines. The statistical results are shown as Figures 14 and 15. When decoding a dynamic or high resolution picture, the use of the data parallel architecture has a high error rate, presumably because the data parallel

50 EURASIP Journal on Wireless Communications and Networking 9 architecture does not forecast at regular periods. Upon a missed deadline, continuous deadline misses will occur due to data dependency. 5. Conclusions This study proposed a power-aware DVB-H mobile TV system on a heterogeneous multicore platform and established a front-end buffer control mechanism and a parallel DVB-H TV signal decoding model. Applying a parallel architecture could increase the speed for the decoding of a DVB-H TV program. Dependent on picture complexity, the DVFS system can be used for dynamic tuning of the system voltage and frequency to lower power consumption. The experimental results confirmed that applying a DVFS system couldsaveasmuchas39%power,whichcouldincreasethe service time of some mobile TV devices. If coupled with a receiving scheduler mechanism at the Receiver, even more energy could be saved. When decoding dynamic or high resolution pictures, a high error rate is occurred in the data parallel architecture, since the data parallel architecture is not forecasted at regular periods probably. Upon a missed deadline, continuous deadline misses will occur due to data dependency. Those issues are interesting challenges in the parallel video decoding architecture. Acknowledgment This paper was supported by the Sustainable Growth Project (S) (98WFA9277) of the Department of Engineering and Application Science, National Science Council (NSC), Taiwan. References [1] S. Parkvall, E. Englund, M. Lundevall, and J. Torsner, Evolving 3G mobile systems: broadband and broadcast services in WCDMA, IEEE Communications Magazine,vol. 44,no.2,pp , 26. [2] S. Buchinger, S. Kriglstein, and H. Hlavacs, A comprehensive view on user studies: survey and open issues for mobile TV, in Proceedings of the 7th European Conference on European Interactive Television Conference, Leuven,Belgium, 29. [3] E. Kaasinen, M. Kulju, T. Kivinen, and V. Oksman, User acceptance of mobile TV services, in Proceedings of the 11th International Conference on Human-Computer Interaction with Mobile Devices and Services (MobileHCI 9), September 29. [4] H.FuchsandN.Färber, Optimizing channel change time in IPTV applications, in Proceedings of IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB 8), April 28. [5]S.Lee,J.Koo,andK.Chung, Content-awareratecontrol to improve the energy efficiency in mobile IPTV services, IJCSNS International Journal of Computer Science and Network Security, vol. 1, no. 5, pp , 21. [6] Y. Solomon, The Economics of Mobile Broadcast TV, Mobile DTV Alliance Whitepaper, January 27. [7] J. Valerdi, A. González, and F. J. Garrido, Automatic testing and measurement of QoE in IPTV using image and video comparison, in Proceedings of 4th International Conference on Digital Telecommunications, Colmar, France, 29. [8] S. Yaldiz, A. Demir, and S. Tasiran, Stochastic modeling and optimization for energy management in multicore systems: a video decoding case study, IEEE Transactions on Computer- Aided Design of Integrated Circuits and Systems, vol.27,no.7, pp , 28. [9] P. Francesco, P. Antonio, B. Davide, B. Luca, and M. Poletc, Energy-E cient multiprocessor systems-on-chip for embedded computing: exploring programming models and their architectural support, IEEE Transactions on Computers, vol. 56, no. 5, pp , 27. [1] M. Hefeeda and C.-H. Hsu, On burst transmission scheduling in mobile TV broadcast networks, IEEE/ACM Transactions on Networking, vol. 18, no. 2, pp , 21. [11] M. Rezaei, I. Bouazizi, and M. Gabbouj, Joint video coding and statistical multiplexing for broadcasting over DVB-H channels, IEEE Transactions on Multimedia,vol.1,no.8,pp , 28. [12] M. Kornfeld and G. May, DVB-H and IP datacast broadcast to handheld devices, IEEE Transactions on Broadcasting, vol. 53, no. 1, pp , 27. [13] ETSI EN 32 34, Digital Video Broadcasting (DVB): transmission system for handheld terminals (DVB-H), European Standard, v.1.1.1, 24. [14] ETSI TR v1.3.1, Digital Video Broadcasting (DVB); DVB-H Implementation Guidelines, May 27. [15] ETSI TS v1.2.1, IP Datacast over DVB-H: Content Delivery Protocols, December 26. [16] Digital Video Broadcasting - Handheld (DVB-H) home page, 28, [17] D. Gómez-Barquero and A. Bria, Forward error correction for file delivery in DVB-H, in Proceedings of IEEE Vehicular Technology Conference, pp , Dublin, Ireland, April 27. [18] M. Kornfeld and G. May, DVB-H and IP datacast - Broadcast to handheld devices, IEEE Transactions on Broadcasting, vol. 53, no. 1, pp , 27. [19] C. Meenderinck, A. Azevedo, B. Juurlink, M. Alvarez Mesa, and A. Ramirez, Parallel scalability of video decoders, Journal of Signal Processing Systems, pp. 1 22, 28. [2] A. Azevedo, C. Meenderinck, B. Juurlink, et al., Parallel H.264 decoding on an embedded multicore processor, in Proceedings of the 4th International Conference on High Performance and Embedded Architectures and Compilers, pp , January 29. [21]T.Olivares,F.J.Quiles,P.Cuenca,L.Orozco-Barbosa,and I. Ahmad, Study of data distribution techniques for the implementation of an MPEG-2 video encoder, in Proceedings of Parallel and Distributed Computing Systems, pp , November [22] A. Bilas, J. Fritts, and J. P. Singh, Real-time parallel MPEG-2 decoding in software, in Proceedings of the 11th International Parallel Processing Symposium (IPPS 97), pp , April [23] D. Farin, N. Mache, and H. N. Peter, SAMPEG, a scene adaptive parallel MPEG-2 software encoder, in Visual Communications and Image Processing, vol. 431 of Proceedings of SPIE, pp , San Jose, Calif, USA, January 21.

51 1 EURASIP Journal on Wireless Communications and Networking [24] M. Flierl and B. Girod, Generalized B pictures and the draft H.264/AVC video-compression standard, IEEE Transactions on Circuits and Systems for Video Technology, vol. 13, no. 7, pp , 23. [25] A. Rodriguez, A. Gonzalez, and M. P. Malumbres, Hierarchical parallelization of an h.264/avc video encoder, in Proceedings of International Symposium on Parallel Computing in Electrical Engineering, pp , September 26. [26] Z. Zhao and P. Liang, Data partition for wavefront parallelization of H.264 video encoder, in Proceedings of IEEE International Symposium on Circuits and Systems (ISCAS 6), pp , May 26. [27] M. Roitzsch, Slice-balancing H.264 video encoding for improved scalability of multicore decoding, in Proceedings of the 7th ACM and IEEE international conference on Embedded software, pp , September 26. [28] E. B. van der Tol, E. G. T. Jaspers, and R. H. Gelderblom, Mapping of H.264 decoding on a multiprocessor architecture, in Proceedings of the 6th International Conference on Advances in Mobile Computing and Multimedia, pp. 4 49, November 28. [29] J. Chong, N. Satish, B. Catanzaro, K. Ravindran, and K. Keutzer, Efficient parallelization of H.264 decoding with macro block level scheduling, in Proceedings of IEEE International Conference onmultimedia and Expo (ICME 7), pp , July 27. [3] A. Azevedo, C. Meenderinck, B. Juurlink, A. Terechko, et al., Parallel H.264 decoding on an embedded multicore processor, in Proceedings of the 4th International Conference on High Performance and Embedded Architectures and Compilers, pp , January 29. [31] W. Lee, K. Patel, and M. Pedram, GOP-level dynamic thermal management in MPEG-2 decoding, IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 16, no. 6, pp , 28. [32] I. Yeo, H. K. Lee, E. J. Kim, and K. H. Yum, Effective dynamic thermal management for MPEG-4 decoding, in Proceedings of IEEE International Conference on Computer Design (ICCD 7), pp , October 27. [33] M. Mehendale, Socs for portable video applications: architecture level considerations, in Proceedings of IEEE Electronic Design Processes Workshop, pp , April 27.

52 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 11 pages doi:1.1155/21/ Research Article Embedding Protection Inside H.264/AVC and SVC Streams Catherine Lamy-Bergot and Benjamin Gadat THALES Land and Joint Systems, RCP/DT and EDS/SPM Departments, 9274 Colombes, France Correspondence should be addressed to Catherine Lamy-Bergot, Received 3 March 21; Revised 2 July 21; Accepted 26 August 21 Academic Editor: Liang Zhou Copyright 21 C. Lamy-Bergot and B. Gadat. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. A backward compatible error-protection mechanism embedded into the H.264 (AVC or SVC) syntax is described. It consists of the addition into the H.264 bitstream of supplementary network abstraction layer (NAL) units that contain forward error-correction (FEC) data generated by a block error-correction code. The proposed mechanism allows to leave the original information bits and NAL units intact and does not rely on any side information or extra signalling coming from lower layers, ensuring backward compatibility with the standard syntax. Simulation results obtained with Reed-Solomon and Low-density parity check errorcorrecting codes show significant improvements for both erroneous and lossy transmission channel configurations. 1. Introduction The H.264 standard, both in its nonscalable (AVC) [1] and in its more recent scalable version (SVC) [2] has been established to offer enhanced coding efficiency when compared to previous video-coding standards such as the still widely used MPEG-2 one. The aim of the standardization effort has been to establish a solution enabling transmission of more video (or of video of better quality) over as diverse conditions as Internet/LAN, TV broadcasting, or mobile wireless networks. To cope with the loss/error conditions that may occur on those various networks type, the H.264 video-standard includes error resilience tools, such as picture segmentation, intra-placement on various levels, reference picture selection, data partitioning, flexible macroblock ordering, and so forth, [3]. These tools may however remain insufficient to offer a complete recovery of a corrupted stream, leading to degraded video rendering when in presence of very erroneous transmission conditions, such as the one occurring over wireless channels. Indeed, wireless channels rely on physical (PHY) layer protection by means of forward error-correction (FEC) and possible retransmissions (ARQ for Automatic Repeat request) to provide reliable transmission over their unreliable communication medium, but bandwidth limitations or eventual real-time constraints can prevent the transmission to be fully reliable. There have been different research efforts to overcome this issue, in particular via the introduction of error resilient coding mechanisms in the H.264 standard or via the usage of additional data transmission, by means of ARQ and/or FEC above the physical layer. The first type of mechanism consists in efficiently concealing the losses or errors due to the transmission over the channel, by embedding of data useful for error concealment into the video stream [4], or by intelligently encoding or decoding the stream [5, 6]. Those approaches however can only minimize the impact of the wireless channel, and not actually correct the errors or losses. For a perfect or almost perfect rendering of the information transmitted, one needs to rely on the second type of mechanism, namely the FEC and/or ARQ approaches. Beside the original approach where application and radio were considered directly connected, allowing various declinations of error-correction codes applications after compression and before transmission over the channel (e.g., in [7]), one finds in the literature and recent standards solutions integrating the protocol layers. Some rely on the introduction of retransmissions (ARQ or hybrid-arq) at the data link level [8, 9], while other solutions propose to introduce error correcting capability in transport layer (as promoted by IETF FecFrame [1] group or with RTP-FEC approaches such as [11], or with the fountain codes AP-FEC approach promoted by 3GPP [12, 13]). A more general approach, belonging

53 2 EURASIP Journal on Wireless Communications and Networking to the joint source and channel coding field, which aims at finely adapting the transmitted content as well as the protection applied to the considered source and transmission conditions, has also been considered, with eventually a combination of FEC applied at higher and PHY levels [14]. Those different solutions show interesting performance for multimedia delivery over erroneous or lossy channels, but present the drawback of requiring modifications below the application level, which may be difficult to implement in real life if only for fear of backward compatibility issues for already deployed networks. In this paper, we propose to introduce error-correction capability inside the video stream itself, transparently to the lower layers by embedding it in supplementary network abstraction layer units. This FEC capability introduced at the emission side will allow an aware receiver to correct the eventual losses and errors remaining after the transmission. As such, the approach is valid first in the case of packet losses due to packets drops in not reliable transport protocols such as UDP, or due to timeout for more reliable ones such as TCP, and second in the case of both packet losses and errors due to partially CRC-protected transport protocols such as UDP- Lite and DCCP. Interestingly, this protection is not transport or transmission channel dependent: it can be applied for wired or wireless transmissions, and is compatible with any transport protocol. Furthermore, this protection can be applied either directly together with the compression operation (for immediate or delayed transmission of the stream) or generated later, as a separate operation, typically within a transcoder. The interest of implementing the redundancy insertion in a transcoder module is that the operation can then be performed both over precompressed streams, or on the fly in a proxy or in a relay node if the transmission conditions necessitate it. Similarly, the decoding process can be either embedded directly into the video decoder or performed by a transdecoder module that performs the reverse operation to the transcoder one. For sake of simplicity, in the paper we will describe the case where the operation is collocated to the compression and decompression operations. This paper is organised as follows: Section 2 presents first the H.264 standard network abstraction layer organisation and its syntax, and describes the proposed new redundancy NAL units syntax and their functionality. Section 3 details the corresponding system processing for insertion of redundancy to protect H.264 streams. Section 4 then presents the simulation conditions used and the corresponding simulation results obtained. Finally some conclusions are drawn in Section 5 and perspectives are presented. 2. H.264 Standard NAL Syntax 2.1. NAL Structure. H.264 has been designed to be as network independent as possible. This is made possible by the introduction of encapsulation by means of a network adaptation layer (NAL) which contains the video-coding layer (VCL), as illustrated by Figure 1. The VCL consists of the result of the compression engine, which is the NAL NAL 7 NAL 8 NAL 5 NAL 1 VCL Sequence parameter set Picture parameter set (IDR) coded slice (Inter) coded slice NAL Figure 1: H.264 layer organization. compressed video data itself. The NAL adapts this video data to various network conditions with a transport oriented approach Inserting Redundancy by Means of Supplementary Redundancy NAL Units. The objective being to obtain a stream compliant with the H.264 specifications [1, 2] after the insertion of the redundant information, in order to have any standard decoder still decode the stream, it is necessary to first keep the original data information untouched, and secondly to place the protection in such a way a standard decoder will not try to interpret it. We propose to reach this goal by (i) using systematic codes for FEC protection (ii) inserting the redundancy into specific standard compatible NAL units. As a matter of fact, introducing redundancy data in the stream by embedding it in specific NAL units will allow to respect the H.264 video-standard structure; which means that any standard compliant decoder will merely discard the supplementary information added for protection. FEC aware decoders will on the contrary extract the redundancy information in order to obtain corrected useful data. Naturally, depending on the actual H.264 choice made, that is, either H.264/AVC or H.264/SVC, the NAL unit default header differs, which means that the implementation of the following mechanism must be attuned to the standard. In the following, we will detail the approach for H.264/AVC, with which it has been originally fully tested and validated. Nevertheless, except for the change of header (which is extended from the one-byte value in H.264 AVC to a four-bytes header including identifications but also error indication and importance information), the process of supplementary NAL units carrying redundancy information is identical. In order to protect the carried information data, we propose to consider the case were several (N 1)

54 EURASIP Journal on Wireless Communications and Networking 3 information data NAL units are used to generate several (M 1) redundancy NAL units. While the simple case of each unique information data NAL unit being followed by a unique redundancy data NAL unit is also being considered in our numerical results (see Section 4), only the more generic N/M case will allow to deal with bursts of losses or errors, but also will ensure that the overhead introduced by the supplementary NAL units is not too costly when high-rate protection is used. As illustrated by Figure 2, the chosen data organisation is based on a matrix, composed of a first part which is filled line by line with the information data of the N considered video NAL units, and a second part which is filled by the redundancy information generated by reading the information data in columns. This setting allows for a line/column interleaving of the information data, the redundancy corresponding to a given information NAL unit being spread over up to M different redundancy NAL units. In the case where block codes such as Reed-Solomon codes [15] or Low-Density Parity Check (LDPC) codes [16] are considered, the matrix row and column dimensions are, respectively, N (with K lines corresponding to the N information NAL units completed by eventual padding), and J, whose value is determined by dividing the overall size of the N video NAL units by the chosen code K value. Then, J FEC encoding operations are done, resulting in J (N K ) generated redundancy symbols that constitute the redundancy data to be transmitted. This redundancy constitutes the pseudo VCL information of the supplementary redundancy NAL units, whose headers must then contain information allowing the decoder to regenerate the matrix even when some of the NAL units, whether information or redundancy ones, are lost Syntax for Redundancy NAL Units. The proposed syntax for redundancy NAL units carries information necessary to allow the decoder to recover the original N data NAL units. In particular, the position indication information for each data NAL unit is provided, to ensure that the redundancy NAL units can be exploited even if one NAL unit is lost (meaning that in practice its size, varying by nature, is also lost). The proposed syntax, for a new NAL unit type that we have in our system fixed to the value 3 then contains (i) information on the type of redundancy NAL unit format (several can be considered, as illustrated later), (ii) error protection code used (possibly an index from a predefined table), (iii) additional (could be optional) information to allow for differentiation of frames corresponding to different matrixes (e.g., first video frame number), (iv) numbers of data and redundancy NAL units: N, M, (v) position of the N data NAL units carried by the matrix (this being stored in a (line, column) address format corresponding to the beginning of each NAL unit), (vi) position of the M redundancy NAL units in the matrix (this being stored in a (line, column) address format corresponding to the beginning of each redundancy NAL unit), (vii) a checksum (CRC) covering the whole NAL unit header. Based on this list, it has been observed in practice, that depending on the size of the considered matrix, it was interesting to either place in each of the M redundancy NAL unit the whole description information, or separate the description information to reduce the cost in terms of used bits. The first solution, illustrated by the format (denoted type 1 ) and proposed in Figure 3, allows to easily deal with potential NAL unit losses, as all position and synchronisation information are repeated in each redundancy NAL unit, but leads to a prohibitive cost in terms of bit-rate if M is too large. The second solution, proposed to reduce the number of signalling bits, places the signalisation relative to the information data in a first NAL unit type (denoted type ), as illustrated by Figure 4, and then in each following redundancy NAL unit carrying the redundancy information place only the redundancy signalisation and possibly a reminder on the error protection code used, as illustrated by Figure 5, denoted type 1. For comparison purpose, and to evaluate the interest of the interleaving matrix with respect to the header cost it introduces, we also considered a simplified 1/1 case where each data NAL unit is directly followed by a redundancy NAL unit. As a consequence, the headers have in this case been reduced to the minimal information of start code, standard NAL unit header and a very reduced extension including an index for the used error-correction code, the number of the video data frame protected (for simple loss detection) and a checksum, asillustratedbyfigure System Processing for Insertion of Redundancy and Protection of H.264 Streams As stated before, we will consider here the case where the protection mechanism is applied together with the compression process. Let us explain the proposed mechanisms both at the encoding and decoding side Encoding. Beside its traditional tasks, the video encoder takes the information data (i.e., the video data NAL units) and feeds them into the systematic error-correction encoder to generate redundancy data, and then generates accordingly redundancy NAL units headers accordingly to the format given in Section 2.3, to produce the M redundancy NAL units. The process is detailed in Figure Decoding. Beside performing its traditional tasks, the video decoder aware of possible redundancy NAL units presents also looks for such supplementary information. As detailed in Figure 8, the decoder first reeds NAL units in the bitstream and store them into its NAL units buffer up until finding a redundancy NAL unit, or reaching the buffer

55 4 EURASIP Journal on Wireless Communications and Networking J NAL n 1 NAL #2 NAL #3 K NAL #N Padding N K Red. NAL #1 Red. NAL #M Redundancy bits generated by the error correction code Direction for reading data to generate the redundancy Direction for filling the matrix and reading the redundancy Figure 2: N/M redundancy matrix organisation. Code and redundancy matrix parameters Signalisation for the considered information data Signalisation for the generated redundancy Start code (3 to 4 bytes) Standard NAL header (1 byte) NAL Red type 1 Used code (index) K N K J Video data number (1st one) Number of N column) data N NALs Redundancy counter column) M CRC Redondance data CRC coverage area Figure 3: N/M NAL unit 3 proposed format, type 1. Code and redundancy matrix parameters Signalisation for the considered information data Start code (3 to 4 bytes) Standard NAL header (1 byte) NAL Red type Used code (index) K N K J Video data number (1st one) Number of N video data column) N CRC CRC coverage area Figure 4: N/M NAL unit 3 proposed format, type. Code and redundancy matrix parameters Signalisation for the generated redundancy Start code (3 to 4 bytes) Standard NAL header (1 byte) NAL Red type 1 Used code (index) K N K J Redundancy counter column) CRC Redondance data CRC coverage area Figure 5: N/M NAL unit 3 proposed format, type 1.

56 EURASIP Journal on Wireless Communications and Networking 5 Start code (3 to 4 bytes) Standard NAL header (1 byte) NAL 3 specific header (variable size) Index for the used error correction code (ex: 4 to 8 bits) Video data frame number (frame num) (ex: 4 bits) Checksum (ex: 4 bits) Figure 6: Simplified syntax of the supplementary NAL unit in case of 1/1 redundancy insertion. Considered sequence (including considered Reed-Solomon code) Table 1: Backward compatibility tests for H.264 redundancy enhanced streams. PSNR after decoding with JM 16.2 PSNR after decoding with ffmpeg.5 PSNR after decoding with error correcting JM 12.1 Foreman QCIF, 128 kbps with an RS code (128, 64) Mobile QCIF, 255 kbps with an RS code (255, 232) Hall Monitor QCIF, 128 kbps with an RS code (255, 25) Akiyo QCIF, 128 kbps with an RS code (255, 25) Foreman CIF, 512 kbps with an RS code (255, 232) Mobile CIF, 248 kbps with an RS code (128, 64) HallMonitor CIF, 512 kbps with an RS code (255, 232) Akiyo CIF, 255 kbps with an RS code (255, 25) maximal size (to cope with possible losses of NAL units or temporary absence of redundancy NAL units). With the first redundancy NAL unit received, the signalling information present in the NAL unit header allows to create the matrix at the decoding side, and then to launch the process of fetching all redundancy NAL units and corresponding bits. The redundancy decoding operation is performed when last redundancy NAL unit is received (or a data information NAL unit, which then leads to detecting loss of last(s) redundancy NAL unit and leads to loss of NAL unit processing step), allowing to generate the corrected information data, which is then used to update the data information NAL units, that are then send to the standard video decoder. 4. Numerical Results 4.1. Considered FEC Codes and Transmission Channel. The generation of the redundancy carried by the supplementary NAL units of type is obtained through error-correction codes, whose role is to provide forward error-correction capabilities in error-prone environments. Different families of codes exist, that can be more attuned to error or loss corrections, adapted to random impairments or bursty channels, or also more efficient for shorter or longer sizes of code blocks. The choice of the used error-correction code will consequently have to be made with respect to the characteristics of both the data to be transmitted and the transmission channel. The approach proposed in this paper was consequently made generic, to allow application to various error-correction codes, with the limitation of them needing to be systematic, to ensure that the information NAL units are transmitted unmodified. In our tests and simulations, we chose two different error-correction codes to illustrate the versatility of our solution. The first and primary example considered is one of the most well-known family of error-correction codes: Reed-Solomon (RS) codes [15], over Galois Field GF(2 8 ) to easily manipulate bytes. The codes will be referred to as RS(N, K ) in the following, where N is the codewords symbol length and K the number of information symbols. The second example considered is the family of LDPC codes [16], built according to the progressive edge growth (PEG) algorithm [17]. Both RS andldpccodes are applied in the following to any K bytes of the upper part of the redundancy matrix to generate the N K redundancy bytes, that will then be used to generate the data part of the supplementary NAL units, as illustrated in Figure 2. It is interesting to point out that the FEC approach can be used in conjunction with exiting robustness enhancement techniques already foreseen or used for H.264. Typically, even redundant slices, that are different from our supplementary slices in the sense where they operate duplication operations, and not error-correction, can be protected with the NAL unit type 3. Similarly, the FEC can be used in conjunction with concealment techniques [18, 19] to deal with possible remaining errors by concealing them Backward Compatibility Compliance Tests. One of the interest of the presented approach is that the redundancy information is inserted in standard NAL units, using reserved numbers. As a consequence, an H.264 decoder that is not aware of the possibility to use the redundancy information contained in the supplementary NAL units, or that does not wish to use it (for instance due to complexity limitations considerations) will be in principle able to discard the supplementary packets and decode the stream. To validate in practice the backward compatibility of oursystem,wehavedecidedtotestthiscapacitytoskip supplementary NAL units, and for that have selected two

57 6 EURASIP Journal on Wireless Communications and Networking SNR on the BSC Table 2: Probability of decoding success for Foreman SVC encoded sequence (three layers) over a BSC channel. Observed BER Decoding probability without Reed-Solomon encoding Decoding probability with Reed-Solomon RS (128,12) encoding Decoding probability with Reed-Solomon RS (128,12) encoding and header protection % % % % % % % % 14% % 1% 71% % 93% 1% % 99% 1% % 1% 1% Read information data Encode one NAL Store the NAL into matrix No which presents the obtained PSNR for sequences including redundancy NAL units show that no difference can be observed between the three decoders in absence of perturbation. The impact of inserting redundancy by means of specific NAL units is consequently null in terms of backward compatibility. End of GOP or max NAL number reached? Yes Compute headers redundancy parameters (wrt matrix structure) Yes Generate the redundancy NALs and output them Number of redundancy NAL (M) reached? Yes Output redundancy NAL units No Figure 7: N/M redundancy encoding process. different decoders beside our own modified decoder. As our decoder, is based on the H.264 verification model version 12.1, the first reference decoder we chose is the current latest version of the verification model (JM 16.2). The second reference we selected is another well-known and well-used decoder of the online community: ffmeg (in its latest current version, ffmpeg-.5), which include resilience tools, which may try to interpret the supplementary NAL units. Simulations have been done with four different ITU-T reference sequences: Foreman, Akiyo, Mobile Calendar and Hall Monitor, and using different Reed-Solomon codes, as detailed in Table 1. The results of this table, 4.3. H.264/AVC Tests. Again, simulations have been done with the four different ITU-T reference sequences: Foreman, Akiyo, Mobile Calendar and Hall Monitor. In order to propose fair comparaisons, the different simulations were systematically done for the same overall throughput for compressed (in monoslice mode) and protected video. On the other hand, in order to offer diversity, various compression rates and temporal and spatial resolutions were considered. Similarly, two types of channel models have been considered, corresponding to either packet losses or errors. The first is a packet erasure channel (PEC), working at the NAL unit level, that emulates losses over wired channel as Internet or caused by an imperfect wireless channel. The second is a binary symmetric channel (BSC) corresponding to the case where the different protocol layers interconnecting application and radio are accepting bit errors thanks to partially CRC-protected transport and data link protocols. The results presented in Figure 1 to Figure 15 have been obtained with QCIF ( pixels) spatial resolution considered at 15 Hz with I 1 P 14 format and CIF ( pixels) spatial resolution considered at 3 Hz with I 1 P 29 format. Total throughput (including redundancy) for QCIF sequences was set to 128 kb/s except for Mobile Calendar which used 255 kb/s. Total throughput for CIF sequences was 512 kb/s except for Mobile Calendar which used 248 kb/s and Akiyo which used 255 kb/s. Finally, the comparison presented in Figure 16 has been made with QCIF spatial resolution at 3 Hz, with I 1 P 14 format, for an overall throughput of 128 kb/s for Foreman sequence and 255 kb/s for Mobile Calendar sequence. Beside Reed- Solomon codes, we also have been using, in the case of the BSC channel, an irregular LDPC code of rate 1/2, operating over 512 input bits, that is, with K = 64 and N = 128. This irregular LDPC code has been designed with the PEG algorithm [17] using a maximal degree of 15, our purpose being to favour the frame Error-Rate which is of interest

58 EURASIP Journal on Wireless Communications and Networking 7 ReadaNAL Store NAL in buffer Nb data NAL ok? Yes No Processing the loss of NAL No No NALtypeis redundancy? Yes CRCok? Yes Fetch redundancy bits No Processing the errors in NAL Last redundancy NAL? Yes Matrix error/loss decoding Corrected info. data Updating of NALs with corrected data Video decoding Figure 8: N/M redundancy decoding process. Error rate 1. E + 1. E 1 1. E 2 1. E 3 1. E 4 1. E 5 1. E 6 1. E Bit error rate Frame error rate Uncoded Eb/N (db) Figure 9: Error-Rate performance curves for the considered (128,64) irregular LDPC code. PSNR (db) Loss probability rate RS (128, 12) RS (255, 232) RS (255, 25) No RS Figure 1: PSNR evolution for Foreman sequence in CIF resolution over a packet erasure channel for different protection levels with RS codes. in our application, rather than the bit Error-Rate. Figure 9 presents the bit and frame Error-Rate performance curves of the obtained LDPC code over an Additive White Gaussian Noise (AWGN) channel, for a maximal of 8 iterations (in practice an average of 9 iterations are sufficient in the bottom of the waterfall region, i.e., after 2 db). When considering the different curves presented in Figure 1 to Figure 14, one clearly see appear different areas where the optimal protection level differs. In all cases, with either packet erasure channel (PEC) or binary symmetric channel, it can be observed that almost perfect channels will be greatly helped even by a very low redundancy.

59 8 EURASIP Journal on Wireless Communications and Networking PSNR (db) SNR (db) PSNR (db) SNR (db) RS (128, 12) RS (255, 232) RS (255, 25) No RS RS (128,12) RS (255, 232) RS (255, 25) No RS (a) (b) Figure 11: PSNR evolution for Foreman and Akyio sequences in QCIF resolution over BSC channel for different protection levels with RS codes. PSNR (db) SNR (db) RS (128, 12) RS (255, 232) (a) RS (255, 25) No RS PSNR (db) SNR (db) RS (128, 12) RS (255, 232) (b) RS (255, 25) No RS Figure 12: PSNR evolution for HallMonitor and Mobile Calendar sequences in QCIF resolution over BSC channel for different protection levels with RS codes. PSNR (db) SNR (db) PSNR (db) SNR (db) RS (128, 12) RS (255, 232) RS (255, 25) No RS RS (128, 12) RS (255, 232) RS (255, 25) No RS (a) (b) Figure 13: PSNR evolution for Foreman and Akyio sequences in CIF resolutions over BSC channel for different protection levels with RS codes.

60 EURASIP Journal on Wireless Communications and Networking 9 Typically, over the PEC channel, introducing protection with an RS(255,232) code allows to combat easily up to 5% loss rate, at almost no degradation of the original image. Similarly, over the BSC channels, we see that an RS(255,25) code, which introduces less than 5% of redundancy already allows to gain noticeably. However, it is interesting to note that pursuing always greater redundancy level is not always the best choice: typically the performance obtained with the RS(128,12) code will tend to be more interesting than the one obtained with the RS(232,255) over the BSC channel when one considered the actual region of interest of PSNR, namely for values above 25 or 3 db. This necessary balance can be explained by two reasons: first, as the global throughput is kept constant, using an higher redundancy rate implies to compress more the original data, and consequently to degrade the maximal quality attainable. This explains why at very high signal-to-noise ratio (SNR) the unprotected curve performs the best. The second reason is less obvious: simulations have shown that for lower SNR, the strong PSNR degradation comes from the fact that redundancy NAL units header is more often corrupted (corruption detected by the header checksum), leading to the impossibility to use the redundancy information. For that reason, increasing exaggeratedly the redundancy will not help in the proposed framework. This issue disappears in the case where the NAL unit headers are protected, as could for instance be achieved when using length variable transport checksums (with UDP-Lite protocol for instance) with an RTP packetization taking only a single NAL unit per RTP packet. Figure 15 illustrates this effect, and shows that when the redundancy NAL unit is not corrupted, the correct decoding occurs even in presence of a much more degraded channel (up to 6 db earlier in terms of channel SNR). In such a case, it becomes interesting to envisage error-correction codes operation with protection rate greater than the Reed-Solomon traditional range, for instance LDPC codes which are known to perform well for correction rates as low as 1/2 or 1/3. Figure 15(b) shows results obtained with a LDPC redundancy NAL units using the aforementioned (128,64) irregular LDPC code, showing an improvement due to the better performance of the LDPC of about 1.5 db SNR when compared to the usage of an (128,64) RS code. In all cases, one see appear different areas in which the optimal level of protection varies. When having a feedback link able to transmit the channel state quality at the transmission side, simple rules could be applied to allow an optimisation of the approach. For instance, QCIF sequences over a BSC channel could follow this simple rule (i) for BER < use RS (255,25) FEC code, (ii) for BER > use RS (128,12) FEC code. This very simple unequal error protection approach with time could also be broadened by taking into account the type of slices present in the matrix: more bandwidth could be provided for one matrix containing high-importance frames, while another matrix with less important frames could be more compressed or less protected to recover the previously over-used bitrate. Interestingly, it should be noted that the capability to change the FEC protection in time is not limited to change the coding rate: the decoding process detailed in Section 3.2 making sure to check for each new redundancy NAL unit the used code (with the used code index) and its parameters, it is also possible to change in the middle of a sequence the type of protection, typically going from a low protection level with Reed-Solomon codes to a higher protection level with LDPC codes. Finally, the results presented in Figure 16 allow to compare the performance obtained with 1/1 and N/M configurations, again in the case of a BSC channel. One finds that a noticeable gap is observed in favor to N/M approach when error probability is larger than 1 4. In practice, it is between 1 4 and to 1 2 that the largest additional gain is observed with the N/M method, ranging from 2 to 7 db over the BSC, which justifies the interest of the N/M approach H.264/SVC Tests. As mentioned in Section 2.2, one of the interest of the proposed approach is its validity for the scalable extension of the H.264 standard [2]. To illustrate this, and demonstrate the interest of our protection embedding approach, we have tested our system with the Foreman sequence encoded in SVC format with three layers corresponding to QCIF, 15 Hz for the base layer, CIF, 15 Hz with additional enhancement layer 1 and CIF, 3 Hz with additional enhancement layer 2. Due to the very low resistance to errors of the current SVC verification model (called JSVM), we have chosen to report in Table 2 the decoding success probability of the complete stream (full resolution) in three configurations: without Reed-Solomon protection (here using an RS(128,12) code), with Reed- Solomon protection as detailed in Section 3.1 and finally with Reed-Solomon protection with redundancy NAL unit header uncorrupted. One observes with the Reed-Solomon code protection a very good decoding rate up to a BER of 1 5, to be compared with a necessity of almost perfect channel when the scalable decoder is used alone. When the redundancy NAL unit header is protected, one can go up to bit Error-Rates of 1 4 with 71% chances of decoding success, with is becoming really interesting over error-prone channels. 5. Conclusions This paper proposes a backward compatible mechanism to embed error protection inside H.264 (AVC or SVC) streams. The introduced solution relies on the insertion of supplementary network abstraction layer units that carry redundancy information generated by a systematic errorcorrection code. The proposed syntax for these supplementary NAL units is presented, together with tests results carried out with different error correcting codes (Reed-Solomon or LDPC codes) for H.264 AVC and H.264 SVC video streams, that show a noticeable performance gain for the video decoder over lossy or erroneous channels. The possible use of this approach to realise an unequal error protection over time-varying channels has also been

61 1 EURASIP Journal on Wireless Communications and Networking PSNR (db) SNR (db) PSNR (db) SNR (db) RS (128, 12) RS (255, 232) RS (255, 25) No RS RS (128, 12) RS (255, 232) RS (255, 25) No RS (a) (b) Figure 14: PSNR evolution for HallMonitor and Mobile Calendar sequences in CIF resolution over BSC channel for different protection levels with RS codes. PSNR (db) SNR (db) PSNR (db) SNR (db) RS (128, 12) RS (255, 232) RS (255, 25) No RS RS (128, 64) RS (128, 12) RS(255, 232) RS (255, 25) No RS RS (128, 64) LDPC (128, 64) (a) (b) Figure 15: PSNR evolution for Mobile Calendar sequence in CIF resolution and Foreman sequence in QCIF resolution over BSC channel for different protection levels (RS or LDPC codes) and redundancy NAL unit headers protected. 4 4 PSNR (av MSE) (db) PSNR (av MSE) (db) E 1 1.E 2 1.E 3 1.E 4 1.E 5 BER 1 1 RS (128, 12) N M RS (128, 12) (a) 1.E 1 1.E 2 1.E 3 1.E 4 1.E 5 BER N M RS (128, 12) 1 1 RS (128, 12) (b) Figure 16: Comparing 1/1 and N/M redundancy approach performance for Foreman and Mobile QCIF sequence with RS(128,12) protection, BSC channel.

62 EURASIP Journal on Wireless Communications and Networking 11 pointed out. Further investigations on this point and more generally the usage of this standard compatible FEC feature in the context of adaptive schemes are foreseen. Acknowledgments This paper was achieved with support partly of the French Industry Department through the POLE POSEIDON Project (no ) and partly of the European Community s Seventh Framework Programme through Grant agreement ICT OPTIMIX (no. INFSO-ICT ). C. Lamy-Bergot and B. Gadat want to gratefully acknowledge the support of their colleague Erwann Renan for his contribution of an efficient SVC codec for the simulations. The contributions of Karim Saada and Elodie Bounmy who participated to the development of the program during their internships at THALES Land and Joint Systems, EDS/SPM, France are also gratefully acknowledged. References [14] M. G. Martini, M. Mazzotti, C. Lamy-Bergot, J. Huusko, and P. Amon, Content adaptive network aware joint optimization of wireless video transmission, IEEE Communications Magazine, vol. 45, no. 1, pp. 84 9, 27. [15] F. G. MacWilliams and N. J. A. Sloane, The Theory of Error Correcting Codes: Part 1, North-Holland Publishing Company, New York, NY, USA, [16] R. G. Gallager, Low-Density Parity-Check Codes, MIT Press, Cambridge, Mass, USA, [17] X.-Y. Hu, E. Eleftheriou, and D. M. Arnold, Regular and irregular progressive edge-growth tanner graphs, IEEE Transactions on Information Theory, vol. 51, no. 1, pp , 25. [18] S. Valente, C. Dufour, F. Grolière, and D. Snook, An efficient error concealment implementation for MPEG-4 video streams, IEEE Transactions in Consumer Electronics, vol. 47, no. 3, pp , 21. [19] D. Agrafiotis, D. R. Bull, and N. Canagarajah, Optimized temporal error concealment through performance evaluation of multiple concealment features, in Proceedings of the International Conference on Consumer Electronics (ICCE 6), pp , January 26. [1] ITU-T Rec. H.264 ISO/IEC , 23. [2] ITU-T Rec. H.264 ISO/IEC annex G, 29. [3] S. Wenger, H.264/AVC over IP, IEEE Transactions on Circuits and Systems for Video Technology, vol. 13, no. 7, pp , 23. [4] L.-W. Kang and J.-J. Leou, An error resilient coding scheme for H.264 video transmission based on data embedding, vol. 3, pp [5] D. Kim, S. Yang, and J. Jeong, A new temporal error concealment method for H.264 using adaptive block sizes, in Proceedings of the IEEE International Conference on Image Processing (ICIP 5), vol. 3, pp , September 25. [6] T. Keranen, J. Vehkapera, and J. Peltola, Error concealment for SVC utilizing spatial enhancement information, in Proceedings of the 4th International Mobile Multimedia Communications Conference (MobiMedia 8), Oulu, Finland, July 28. [7] M. M. Ghandi, B. Barmada, E. V. Jones, and M. Ghanbari, Wireless video transmission using feedback-controlled adaptive H.264 source and channel coding, IET Communications, vol. 3, no. 2, pp , 29. [8] D.J.C.CostelloJr.,J.Hagenauer,H.Imai,andS.B.Wicker, Applications of error-control coding, IEEE Transactions on Information Theory, vol. 44, no. 6, pp , [9] S. Soltani, K. Misra, and H. Radha, Delay constraint error control protocol for real-time video communication, IEEE Transactions on Multimedia, vol. 11, no. 4, pp , 29. [1] M. Watson, Forward Error Correction (FEC) Framework, draft-ietf-fecframe-framework-7 (work in progress: expires Sept. 21), March 21. [11] IETF RFC 551, Reed-solomon forward error correction (FEC) schemes, J. Lacan, V. Roca, J. Peltotalo, and S. Peltotalo, Eds., 29. [12] V. Sgardoni, M. Sarafianou, P. Ferré, A. Nix, and D. Bull, Robust video broadcasting over 82.11a/g in time-correlated fading channels, IEEE Transactions on Consumer Electronics, vol. 55, no. 1, pp , 29. [13] S. Ahmad, R. Hamzaoui, and M. Al-Akaidi, Adaptive unicast video streaming with rateless codes and feedback, IEEE Transactions on Circuits and Systems for Video Technology, vol. 2, no. 2, pp , 21.

63 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 11 pages doi:1.1155/21/ Research Article Quality-Assured and Sociality-Enriched Multimedia Mobile Mashup Hongguang Zhang, Zhenzhen Zhao, Shanmugalingam Sivasothy, Cuiting Huang, andnoël Crespi Wireless Networks and Multimedia Services Department, Institut Telecom, Telecom SudParis, 9 Rue Charles Fourier, 91 Evry, France Correspondence should be addressed to Hongguang Zhang, hongguang.zhang@gmail.com Received 1 April 21; Revised 3 June 21; Accepted 14 August 21 Academic Editor: Liang Zhou Copyright 21 Hongguang Zhang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Mashups are getting more complex with the addition of rich-media and real-time services. The new research challenges will be how to guarantee the quality of the aggregated services, and how to share them in a collaborative manner. This paper presents a metadata-based mashup framework in Next Generation Wireless Network (NGWN), which guarantees the quality and supports social interactions. In contrast to existing quality-assured approaches, the proposed mashup model addresses the quality management issue from a new perspective through defining the Quality of Service (QoS) metadata into two levels: fidelity (user perspective) and modality (application perspective). The quality is assured from quality-aware service selection and quality-adaptable service delivery. Furthermore, the mashup model is extended for users to annotate services collaboratively. The annotation occurs in two ways, social tagging (e.g., rating and comments) and QoS attributes (e.g., device type and access network, etc.). In order to apply this network-independent metadata model into NGWN architecture, we further introduce a new entity named Multimedia Mashup Engine (MME) which enables seamlessly access to the services and Adaptation Decision Taking (ADT). Finally, our prototype system and the simulation results demonstrate the performance of the proposed work. 1. Introduction The evolution of Web 2. has brought a significant impact on the Internet service provisioning by encouraging the contribution from end user for contents and services creation. This phenomenon, termed User-Generated Content (UGC) or User-Generated Service (UGS), aim to enlarge user personalization through the Do IT Yourself (DIY) manner. Mashup, as a general term in the UGC/UGS domain, is an application that incorporates elements coming from more than one source into an integrated user experience [1]. Meanwhile, in Telecom there is an ongoing process of transformation and migration from so-called legacy technology to an IP-based Next Generation Networking (NGN), or Next Generation Wireless Network (NGWN), which enables people to access multimedia anytime and anywhere. With the advantage of an All-over-IP network, the opportunity for integration and convergence is amplified, where the most prominent example is the Web-NGN convergence. Toward the convergence of Web and NGN, mobile mashup is promising for the next generation user-driven multimedia delivery [2, 3]. With the proliferation of services available on the Internet and the emergence of user-centric technologies, millions of users are able to voluntarily participate in the development of their own interests and benefits by means of service composition [4]. The concept of composition is to create a new service by combining several existing elementary services. A number of composition mechanisms have been proposed, such as workflow technique and Artificial Intelligence (AI) Planning [5]. However, as most of the existing solutions are still professional developer inclined, the arduous development task always discourages users to contribute themselves to the service creation process. In this context, mashup, which is well known with its intrinsic advantages of easy and fast integration, is a promising choice

64 2 EURASIP Journal on Wireless Communications and Networking for the user-driven service composition issues. Generally, the mashup mechanism is provided to combine nonrealtime Web services such as translation, search, and map. by leveraging the programming Application Programming Interfaces (APIs). With the proliferation of mobile devices and wireless networks, real-time and resource-consuming multimedia services have been ubiquitous and all pervasive. Thus, in this paper we consider mashup as userdriven multimedia aggregation. We argue that the userdriven multimedia delivery is more challenging than the provider-driven model. Firstly, to nonexpert users it is desirable to have a mashup model which hides the backend complexity and simplifies the aggregation process. Moreover, the emerging mashups are getting more and more complicated when the rich-media and real-time services are aggregated. Nevertheless, the diverse terminals, heterogonous networks as well as various user requirements constrain multimedia mashup to low quality, especially in the mobile network environment. The third challenge is raised from the sociality point of view. Since the great success of social networking has shown that user experiences are enriched by sharing, aggregating, and tagging collaboratively, the social phenomena behind mashup are worth being explored. Our paper presents a NGWN-based mashup framework, which is featured by an intermediate metadata model with the guarantee of quality and the support of sociality. The metadata-based framework brings the benefits in three aspects. Firstly, the human-readable metadata is the higher level description language compared with programming APIs, which can hide the programming complexity from nonexperts. Secondly, the scalable quality management can be enforced by Quality of Service (QoS) metadata. The concept of scalability in this paper means that the aggregated media can be tailored and adapted to diverse terminals and heterogeneous networks with the assured quality, which aims to provide the best user experiences across aggregated multimodal services. Thirdly, these metadata entities can be further enriched collaboratively by end users through social annotation. In this paper, we propose to extend the CAM4Home metadata as our mashup model. CAM4Home is an ITEA2 project enabling a novel way of multimedia provisioning by bundling different types of content and service into bundles on the level of metadata [6]. In our solution, rich-media services including video, audio, image, and even text can be encapsulated as Collaborative Aggregated Multimedia (CAM) Objects, which can be then aggregated into CAM Bundles. We further propose to integrate MPEG-21 metadata within the CAM4Home model. We enforce QoS by two ways, quality-aware service selection at design-time, and quality-adaptable service delivery at run time. The human-readable part of QoS metadata facilitates service selection firstly. Meanwhile it will enable adaptable delivery. Prominently, our system supports collaborative annotation. The annotation occurs in two ways, social tagging (e.g., rating and comments), and QoS tagging (e.g., device type and access network etc.). The former may facilitate service selection, while the latter will enhance QoS-aware mashup consumption. The rest of the paper is organized as follows. Section 2 reviews the background and related works. In Section 3, we describe a scenario and present the metadata-based model, in which we illustrate QoS management and social metadata. Section 4 discusses the approach to apply the metadata model into the NGN-based service architecture. A prototype system and the performance evaluation are described in Section 5. Section 5 concludes the paper and presents some issues for future research. 2. Related Work The past few years have witnessed the great success of userdriven models, such as Wikipedia, Blog, and YouTube, which are known as UGC. The next big user-driven hype will happen in the service arena, that is, UGS. Considerable researches have been conducted on mashup and service composition, most of which utilize Web-based programming technologies (e.g., Web Service Description Language (WSDL) and Representational State Transfer (REST)) for the implementation. In order to facilitate the creation of mashup, some Web platforms have been proposed by different communities, among which Yahoo Pipes [7] and Microsoft Popfly [8] are well-known examples. Nevertheless, these platforms are far from being popularized for the ordinary users due to their complexity. It is desirable to have a mashup model which hides the backend complexity from user, simplifies the service creation interface, and satisfies the service creation variety requirements. Unlike traditional data services, multimedia services face more challenges in the heterogeneous environments. A lot of research works have been conducted in this area. Z. Yu et al. proposed a context-aware multimedia middleware which supports multimedia content filtering, recommendation, and adaptation according to changing context [9, 1]. The article in [11] described an approach for contextaware and QoS-enabled learning content provisioning. L. Zhou et al. presented a context-aware middleware system in heterogeneous network environments, which facilitates diverse multimedia services by combining an adaptive service provisioning middleware framework with a context-aware multimedia middleware framework [12]. The scheduling and resource allocating issues were discussed for multimedia delivery over wireless network [13, 14]. However, these systems or solutions usually targeted one type of media. When more and more rich-media services are aggregated or composed, the quality issue is getting more challenging. In addition, the social phenomena between users are ignored by the past research works. Typically, a mashup process can be divided into three steps: service selection, service aggregation, and service execution. The quality issue is across these three steps, among which research efforts are firstly made to QoS-aware service selection. A composite service can be constructed and deployed by combining independently developed component services, each one may be offered by different providers with different nonfunctional QoS attributes. A random selection may not be optimal for its targeted execution environment and may incur inefficiencies and

65 EURASIP Journal on Wireless Communications and Networking 3 costs [15]. Therefore, a selection process is needed to identify which constituent services are to be used to construct a composite service that best meets the QoS requirements of its users. To formally define the QoS level required from the selected provider, the provider and the user may engage in negotiation process, which culminates in the creation of a Service Level Agreement (SLA). The management of QoS-based SLAs has become a very active area of research, including the QoS-aware service description, composition, and selection [16]. However, QoS-aware service selection is just the initial step to guarantee the quality. The other two steps may also bring a lot of impacts to the final quality. Most prominently, the context of service creation could be different to that of service execution, especially in the mobile environment, where the diverse terminals, heterogonous networks as well as various user requirements constrain the multimedia access to low quality. This problem is getting more and more complicated when the rich-media services are aggregated. As a result, a scalable model with QoS management is significantly important for mashups, especially for the mobile mashups in a highly dynamic service environment. Since the mechanism of mashup is to combine data from different sources, it is desired to have an overall quality model across aggregated services. T.C. Thang et al. [17 19] have intensively studied the quality in multimedia delivery. They identified the quality from two aspects: perceptual quality and semantic quality. The former known as fidelity refers to a user s satisfaction, while the latter is the amount of information the user obtains from the content. The former is sometimes referred to as Quality of Experience (QoE), while the latter is as Information Quality (IQ). In some cases, the perceptual quality of a media service is unacceptable or its semantic quality is much poorer compared with that of a substitute modality. A possible solution for this problem is to convert the modality. For example, when the available bandwidth is too low to support the video streaming service for a football match, the text-based statistics service would be more appropriate than the adapted video with poor perceptual quality. This is a typical case of video-totext modality adaptation. Apparently, the combination of fidelity and modality can enhance user experiences. Dynamic adaptation is seen as an important feature enabling terminals and applications to adapt to changes in access network, and available QoS due to mobility of users, devices, or sessions [2]. The previous research works on multimedia adaptation are more concerned with the perceptual quality from the aspect of end user. However, the intensive studies in [17 19] state that the semantic quality should be considered in some cases. They argue that modality conversion could be a better choice than unrestricted adaptation on fidelity. The Overlapped Content Value (OCV) model is introduced in [17] to represent conceptually both quality and modality. Unfortunately, a quality model for mashup has never been mentioned in the literature. In this paper, we propose to apply both fidelity and modality into the quality of mashup. We argue that both perceptual quality and semantic quality need to be considered in order to provide quality-assured mashup. Considering video as the most prominent media, we take video as the example for quality adaptation. There are some issues that cannot be ignored for video adaptation, such as complexity, flexibility, and optimization. In this regard, Scalable Video Coding (SVC) has emerged as a promising video format. SVC is developed as an extension of H.264/MPEG-4 Advance Video Coding (AVC) [21]. SVC offers spatial, temporal, and quality scalabilities at bit stream level, which enables the easy adaptation of video by selecting a subset of the bit stream. As a result, the SVC bit streams can be easily truncated in spatial, temporal, and quality dimensions to meet various constraints of heterogeneous environments [19]. The three-dimensional scalability offers a great flexibility that enables customizing video streams for a wide range of terminals and networks. SVC can thus allow a very simple, fast, and flexible adaptation to the heterogeneous networks and diverse terminals. M. Eberhard et al. have developed an SVC streaming test bed, which allows dynamic video adaptation [22]. It is desired to apply the advantages of SVC into mashup coping with the quality issue. The ubiquitous multimedia results in the overwhelming multimedia services where it has become difficult to retrieve specific ones. Semantic metadata is a solution to the overwhelming resources. The lack of semantic metadata is becoming a barrier for the in-depth study and wide application. Recently, the great success of social networking has shown that user experiences are enriched by sharing, aggregating, and tagging collaboratively. Under this trend, folksonomy also known as social tagging or collaborative annotation draws more and more attention as a promising source of semantic metadata. Several works have been launched to exploit the knowledge of the mass in order to improve the composition process by considering either social networks or collaborative environments [23 25]. However, they only make use of sociality for service selection or recommendation. The sociality across the process of mashup should be further explored, especially for the quality issue. In this paper, we present a mashup framework as illustrated in Figure 1. We enforce the quality by two ways, quality-aware service selection, and quality-adaptable service delivery. The proposed quality model considers both fidelity and modality to meet QoS requirements in the diverse terminals, heterogeneous networks as well as dynamic network conditions. We concentrate on both the user level by specifying user perceivable service parameter and the application level by adapting multimedia services according to the resource availability of terminal and network. Furthermore, we extend the mashup model allowing users to annotate the services collaboratively. 3. Mashup Model This section firstly describes the concept of metadata-based mashup model through an example scenario, followed by the illustration of the model decomposition. The mashup model is further decomposed into three essential parts: multimodal service aggregation, metadata-based QoS management, and metadata-based social enrichment.

66 4 EURASIP Journal on Wireless Communications and Networking Media Annotation CAM bundle Quality model Metadata model CAM Object CAM element CAM element metadata Mashup flow Selection Aggregation Execution User CAM object CAM element CAM element metadata Figure 1: Mashup Model. Figure 2: Conceptual view of CAM object and CAM bundle Concept of Mashup Model. Let us take Sports Live Broadcasting service as an example. The scenario is the last round of the football league where more than one team has the chance to win the champion. All teams start playing at the same time. Fans are watching the live TV broadcasting of their team. At the same time, they may also want to be updated on the information (e.g. goal, penalty, and red card, etc.) of other simultaneous matches. We assume that there are two relevant services from different providers. The first one is an Internet Protocol TV (IPTV) program delivering a live football game. The IPTV service component can be configured by a set of offered alternative operating parameters (e.g., frame sizes, frame rates and bit rates etc.), by which IPTV can be adjusted dynamically according to user context. The second one is a real-time literal broadcasting service delivering statistics data synchronized to all football matches. A user composes the Sports Live Broadcasting mashup containing above two services. Before multimedia session, the quality model firstly selects the service version according to the static capabilities of terminals or networks. During session, this service element of IPTV can be adapted according to dynamic network condition or user preferences. Moreover, if the adapted IPTV service cannot provide the expected user-perceived quality, a cross-modal adaptation from IPTV to Text may occur. Besides the quality adaptation, the fan can share the metadata-based mashup with friends like file sharing and annotate it by comment, rating as well as user-generated QoS parameters CAM4Home Metadata. The essential part of mashup model is the multimodal service aggregation. In this paper, we use CAM4Home framework as the metadata model for multimodal service aggregation. The CAM4Home is an ITEA2 project implementing the concept of Collaborative Aggregated Multimedia (CAM) [6]. The concept of CAM refers to aggregation and composition of individual multimedia contents into a content bundle that may include references to content-based services and can be delivered as a semantically coherent set of content and related services over various communication channels. This project creates a metadata-enabled content delivery framework by bundling semantically coherent contents and services on the level of metadata. The CAM4Home metadata model supports the representation of a wide variety of multimedia content and service in CAM Element as well as its descriptive metadata. CAM Object is the integrated representation of CAM Element and CAM Element Metadata on the association rule ismetadataof. CAM Bundles are the aggregation of two or more CAM Objects on the association rule containscamobjectreference. CAM Object and CAM Bundle can be uniquely identified by camelementmetadataid and cambundlemetadataid. Figure 2 illustrates a conceptual view of CAM Bundle and CAM Object. Moreover, some complicated rules such as spatial and synchronization are also defined for enhanced aggregation. The taxonomy of CAM Element has two subclasses, Multimedia Element and Service Element. The Multimedia Element is the container of a specific multimedia content, which is further divided into four types, document, image, audio, and video. The Service Element is the container of a specific service. The physical content in CAM Element is referred by the attribute EssenceFileIdentifier which is a Universal Resource Locator (URL). The Service Element includes the other attribute ServiceAccessMethod indicating the methods used to access the service. With the instinctive of CAM, we use the metadata-based approach for the content and service delivery. The service capabilities are described by a CAM object containing Service Element and related metadata, while the converged service is described by a CAM bundle containing several CAM objects of service capabilities. For instance, the attribute EssenceFileIdentifier can be used to indicate the Public Service Identity (PSI) of the service capability. And the other attribute ServiceAccessMethod indicates the SIP methods (e.g., INVITE) accessing the service. However, the described services are not limited to SIP based. This model can be used to encapsulate any types of services. In this paper, the CAM4Home metadata model is adopted as the richmedia aggregation model. Figure 3 shows an example for the aforementioned Sports Live Broadcasting service QoS Metadata. It is necessary to provide a qualityguaranteed and interoperable mashup delivery across various devicesandheterogeneousnetworksaswellasanoptimized use of underlying delivery network bandwidth and QoS characteristics. Generally, it is a computing intensive process for adapting decision-taking involved for choosing the right set of parameters that yield an adapted version. The computational efficiency of adaptating can be greatly enhanced if this process could be simplified, in particular by using metadata

67 EURASIP Journal on Wireless Communications and Networking 5 CAMElementMetadata CAMElement 1 1 +ID +Version +CreationDateTime +Title Service +EssenceFileIdentifier +ServiceAccessMethod Multimedia +EssenceFileIdentifier IPTV Text Figure 3: CAM4Home metadata example. that conveys precomputed relationships between feasible adaptation parameters and media characteristics obtained by selecting them [26]. Moreover, the development of an interoperable multimedia content adaptation framework has become a key issue for coping with this heterogeneity of multimedia content formats, networks, and terminals. Toward this purpose, MPEG-21 Digital Item Adaptation (DIA) specifying metadata for assisting adaptation has been finalized as part of the MPEG-21 Multimedia Framework [27]. MPEG-21 DIA aims to standardize various adaptation related metadata including those supporting decision-taking and the constraint specifications. MPEG-21 DIA specifies normative description tools in syntax and semantic to assist with the adaptation. The central tool is the Adaptation QoS (AQoS) representing the metadata supporting decisiontaking. The aim of AQoS is to select optimal parameter settings that satisfy constraints imposed by a given external context while maximizing QoS. The adaptation constraints may be specified implicitly by a variety of Usage Environment Description (UED) tool describing user characteristics (e.g. user information, user preferences, and location), terminal capabilities, network characteristics, and natural environment characteristics (e.g., location, time). The constraints can also be specified explicitly by Universal Constraints Description (UCD). Syntactically, the AQoS description consists of two main components: Module and Input Output Pin (IOPin). Module provides a means to select an output value given one or several input values. There are three types of modules, namely, Look-Up Table (LUT), Utility Function (UF), and Stack Function (SF). IOPin provides an identifier to these input and output values. The mashup QoS management is proposed on two levels: fidelity and modality. The fidelity is to adapt one of the aggregated service component adjusting QoS parameters, that is, multimedia adaptation with the perceptual quality from the perspective of end user. The modality is to select the most appropriate modality among aggregated multimodal services components, that is, modality conversion with the semantic quality from the application point of view. The overallqualitymodelisillustratedinfigure 4.Weproposeto integrate MPEG-21 DIA into CAM4Home model enabling QoS management. Originally in MPEG-21 DIA, the output values are utilized by Bitstream Syntax Description (BSD) for content-independent adaptation. However, in the proposed mashup model the adapted target is altered to CAM Bundle. Specifically, the AQoS is embedded in each CAM Object for quality adaptation as well as for modality adaptation. In this regard, for quality adaptation, the output values (e.g., bit rate, frame rate, resolution) are utilized to yield an adapted version on a single service component Social Metadata. Collaborative tagging describes the process by which many users add metadata in the form of keywords to shared content [28]. Social metadata is data generated by collaborative tagging, such as tags, ratings, and comments, added to content by individual users other than content creators. Examples can be found everywhere on the web, ratings and comments on YouTube, and tagging in

68 6 EURASIP Journal on Wireless Communications and Networking Fidelity Perceptual quality Quality of experience (QoE) User level Overall quality model Combination Figure 4: Mashup quality model. Modality Semantic quality Information quality (IQ) Application level Digg. The social metadata can help users navigate to relevant contents even quicker because members can use them to provide context and relevant description to the content. The proposed model takes advantage of social metadata to enrich the sociality of mashup from two aspects, service discovery, and QoS management. Accordingly, users are allowed to annotate the services collaboratively in two ways: social tagging (e.g., rating and comments), and QoS attributes (e.g., device type and access network etc.). For example, Bob can tag a CAM entity that is relevant to him and choose the tags he believes best to describe the entity. The keywords Bob chooses help organize and categorize the service element in a way that is meaningful to him. Later, Bob or other members can use those tags to locate data using the meaningful keywords. In order to introduce ambiguous social tagging into structured metadata, the CAM4Home metadata framework defines some attributes of social metadata which include social tag, user comment, and user rating. As mentioned above as the second point, the QoS metadata can also be generated by users. For example, Bob can tag a CAM entity indicating the relevant service inside is not suitable for a mobile device with limited bandwidth. Usually, it is the service provider in the value chain of service delivery to take the responsibility on specifying these QoS parameters. However, it is cost-inefficient and time consuming. The user-generated QoS metadata could be complementary to the provider-generated. 4. Mobile Mashup Architecture In this section, we firstly describe the mashup framework in detail. Then we propose the extension of session negotiation NGN-Based Mobile Mashup Framework. IP Multimedia Subsystem (IMS) has been widely recognized to be the service architecture for NGN/NGWN, offering multimedia services and enabling service convergence independent to the transport layer and the access layer. The IMS architecture is made up of two layers: the service layer and the control layer. The service layer comprises a set of Application Servers (ASs) that host and execute multimedia services. Session signaling and media handling are performed in the control layer. The key IMS entity in this layer is the Call Session Control Function (CSCF) which is an SIP server responsible for session control. There are three kinds of CSCF, among which Serving CSCF (S-CSCF) is the core for session controlling and service invocation. Home Subscriber Server (HSS) is the central database storing the subscriber s profile. Regarding the media delivery, the key component is Media Resource Function (MRF) that can be seen as media server for content delivery. The IMS-based mashup framework firstly supports the combined delivery of multimodal services based on CAM4Home model. Further, the QoS management enforced by MPEG-21 DIA metadata is applied into IMS service architecture. Especially, the cross-modal adaptation is implemented as service switching among aggregated services. AS also interacts with MRF in order to ensure the adaptive delivery of media. Figure 5 illustrates the conceptual mashup framework in IMS. The essential component in the proposed mashup platform is Multimedia Mashup Engine (MME) shown in Figure 5. MME provides the controlled network environment between the mashup clients and the service repository. MME enables easy and seamless access to the service repository, and supports the delivery of qualityassured experiences, across various devices, heterogeneous access networks, and multiple service models (e.g., Webbased, Telco-based). Aforementioned mashup is a userdriven model for service delivery. Therefore, MME is firstly proposed as a generic component of Service Deliver Platform, responsible for service-related functionalities, such as service registration and service discovery. Services represented as CAM metadata entities (e.g. object or bundle) are registered in MME. To end users, the rich semantic information may facilitate service composition and service discovery. The service repository holds both service objects and service bundles. To be noted that the service repository can be in MME or in an external database alternatively. For instance, the CAM4Home project provides a web service platform for metadata generating, storing, and searching. In this case, MME needs to access the external platform through Web service interfaces. Besides above functionalities, the vital role of MME is service routing. MME provides the address resolution decision-making on ASs. As shown in Figure 5, MME is located between S-CSCF and AS. For the consideration of scalability and extensibility, we collocate MME in a SIP AS behaving as Back-to-Back User Agent (B2BUA). On one hand, MME is configured to connect with IMS. On the other hand, MME interfaces with SIP ASs which host those aggregated service elements. In order to enable quality-assured mashup, we extend MME mainly from three aspects: Adaptation-Decision Taking Engine (ADTE), UED collecting, and social metadata interface. ADTE either selects appropriate content modalities among the aggregated service components or to choose adaptation parameters for a specific media service. Additionally, MME needs to collect UED as inputs of ADTE. For modality selection, MME

69 EURASIP Journal on Wireless Communications and Networking 7 Service layer Control layer Transport layer Access layer AS 1 AS 2 AS 3 I-CSCF HSS P-CSCF IP MME S-CSCF MANE Figure 5: Conceptual mobile mashup framework. MRF can act on the incoming requests and route them to AS according to the outputs of ADTE. Thanks to MPEG-21 QoS management, it is more intelligent compared with the routing criterion in [29] where it is based on the userrequested service element. Secondly, MME supports the social metadata interface, through which end users may enrich the original CAM metadata collaboratively. For quality adaptation, we hereafter take video as the target considering video that is the most challenging media type. We introduce the Media Aware Network Element (MANE), as shown in Figure 5. The concept of MANE is defined as network element, such as a middlebox or application layer gateway that is capable of adapt video in real time according to the configuring parameters. It is desirable to control the data rate without extensive processing of the incoming data, for example, by simply dropping packets. Due to the requirement to handle large amounts of data, MANEs have to identify removable packets as quickly as possible. In our solution, the objective of MANEs is to manipulate the forwarded bit stream of SVC according to the network conditions or terminal capabilities. The target configurations of video that can be generated include bit rate, resolution, and frame rate that in fact come as the outputs of ADTE Session Negotation Extension. The scalability we describe in this paper relies on the information exchange between client and server, which includes both static capabilities (e.g. terminal or network) and dynamic conditions (e.g. network or user preference). It allows participants to inform each other and negotiate about the QoS characteristics of the media components prior to session establishment. SIP together with Session Description Protocol (SDP) is used in IMS as the multimedia session negotiation protocol. However, the ability is very limited for SDP to indicate user environment information such as terminal capabilities and network characteristics. The User Agent Profile (UAProf) [3] is commonly used to specify user terminal and access network constraints. It is also not enough, because UAProf contains only static capabilities. Although RFC 384 [31] specifies mechanisms by which an SIP user agent can convey its capabilities and characteristics to other user agents, it is not compatible with MPEG-21-based ADTE. It is important to reach interoperability between IETF approaches for multimedia session management and the MPEG-21 efforts for metadata-driven adaptation, in order to enable personalized multimedia delivery [32]. In our model, UCD and UED serve as the input of ADTE. These input values are in the format of XML document with a known schema. UCD includes the constraints imposed by service providers. We can assume that UCD is available for ADTE. However, UED should be collected for dynamic multimedia session in real time since it is the constraint imposed by external user environment. Therefore, there should be a way to query and monitor UED, particularly terminal capabilities and network characteristics. In order to collect UED, we propose to extend the Offer/Answer mechanism. According to [33], SDP negotiationmayoccurintwoways,whicharereferredtoas Offer/Answer and Offer/Counter-Offer/Answer. In the first way the offerer offers an SDP, the answerer is only allowed to reject or restrict the offer. In the latter way, the answer makes a Counter-Offer with additional elements or capabilities not listed in the original SDP offer. We slightly modify the latter way to put querying information in the Counter-Offer. DIA defines a list of normative semantic references by means of a classification scheme [34], which includes normative terms for the network bandwidth, the horizontal and vertical resolution of a display, and so on. For instance, the termid describes the average available bandwidth in Network Condition. Table 1 show some examples of the semantic references. To indicate these normative terms in SDP, we define a new attribute/value pair as shown in Table 2. Offer and Answer are distinguished by recvonly and sendonly, respectively. The value in Offer means the threshold set by offerer, which is optional. The value in Answer is mandatory as return. In the adaptation framework, MME extracts the semantic inputs of AQoS and format them into SDP formats. During the Offer/Answer session negotiation procedure, the requested parameters are sent to UE in SDP. We assume that there is a module in User Equipment (UE) responsible for providing answers and monitoring dynamic conditions if necessary (e.g. presented by [35]). Accordingly, the answering values are also conveyed in SDP sending back to MME activating adaptation. The proposed adaptation process is divided into three phrases: session initiation, session monitoring, and session adaptation. In the session initiation phrase, the party who

70 8 EURASIP Journal on Wireless Communications and Networking Table 1: Examples of semantic termid in DIA. termid Semantic References The horizontal resolution of Display Capability The vertical resolution of Display Capability The max capacity of Network Capability The minimum guaranteed bandwidth of Network Capability The average available bandwidth in Network Condition Method Offer (for query) Answer (as reply) Table 2: SDP extension. Syntax q = (termid) a = (recvonly :< value >) q = (termid) a = (sendonly :< value >) invokes the service offers the default parameters in SDP by an SIP signaling message, normally SIP INVITE. Besides those well-known parameters as answer, MME extracts input parameters in AQoS and offers them again as request. Some input parameters can be answered immediately such as terminal capabilities and network capabilities, which is enough for modality selection. However, some of them need to be monitored in real time, for example network conditions. In case that any parameter varies out of the threshold set by AQoS, an SIP UPDATE with the specific SDP is feedback to MME. Once ADTE in MME receives the inputs and makes a decision, the adaptation starts with session renegotiation. In case of quality adaptation, MME commands the MANE with the new parameters. 5. Prototype and Evaluation To verify the proposed approach, we develop a prototype system to demonstrate the scenario mentioned in Section 3. The prototype system is the integration of several open source projects as illustrated in Figure 6. On the server side, Open IMS Core [36] is deployed as IMS testbed. We make use of UCT Advanced IPTV [37] to provide IPTV service. MME and Text AS is set up by Mobicents SIP Servlet [38] and configured to connect with Open IMS Core. The client is simulated in the signaling plane and in the media plane separately. The CAM4Home metadata are central to the proposed mashup model. Aforementioned, the CAM4Home project provides a web service platform for metadata generating, storing, and searching. In order to enable our client to access the service, we have deployed a gateway between IMS and CAM4Home. For metadata generating, a minimal set of data is required, such as title, description, and essence file identifier. In our case, CAM objects with QoS metadata (e.g. IPTV and Text) are generated by service providers and deposited in the platform. End users can search, aggregate, share, or annotate these multimedia resources through the gateway. Table 3: Terminal, access network, and settings. Terminal Resolution Access network Bandwidth Mobile Phone QCIF GPRS 1 kbps Smart Phone CIF UMTS 5 kbps Laptop 4CIF WiMAX 2 kbps The system performance is analyzed in the signaling plane and in the media plane, respectively. In the signaling plane, we emulate IMS signaling client by SIPp [39]. The prototype system demonstrates that the proposed SIP/SDP extension works compatibly with the standardized IMS platform. We observe that there are notably two kinds of latency: UED collecting and ADTE. The first one is more related to the characteristics of UED themselves. For instance, if the screen size is considered in UED, it could be retrieved immediately by UE. But in terms of available bandwidth, it depends on the time for sampling. Without considering UED, we further observe that ADTE-incurred delay is 1ms averagely. To some extent, this result confirms that the metadata-based adaptation is efficient, because the precomputation saves significant time over parameter selection. The media plane is correlated with quality adaptation. We simulate three types of terminal with various resolutions: mobile phone, smart phone, and laptop. These terminals are assumed to be connected with three kinds of access networks, General Packet Radio service (GPRS), Universal Mobile Telecommunications System (UMTS), and Worldwide Interoperability for Microwave Access (WiMAX), respectively. The terminal settings are listed in Table 3. The quality adaptation is simulated under the constraints of network bandwidth and terminal resolution. The SVC reference software JSVM 9.18 [4] is used as the video codec. The test sequence is ICE which is encoded with three spatial layers (QCIF, CIF, and 4CIF), five temporal layers (1.875, 3.75, 7.5, 15, and 3 fps), and two quality layers. The supported bitrates at various Spatial Quality and Temporal Quality are summarized in Table 4. Figure 7 shows the average bitrates of adapted videos. Figure 8 presents the output Peak Signal to Noise Ratio (PSNR) curves of adapted videos. It can be seen that the average bitrates of adapted videos are consistent with the settings. And the adapted videos have different qualities, measured by means of PSNR. Obviously the bitrates corelate with the values of PSNR. As we can see, SVC with the support ADTE is very suitable for qualityassured mashup. Considering this plane is more related to user experience, we plan to run usability tests in our future work. 6. Conclusion This paper presented a metadata-based multimedia mashup framework in NGWN. It is not only provided scalable QoS management but also enhanced the sociality of mashup. To achieve that, we proposed a flexible framework using

71 EURASIP Journal on Wireless Communications and Networking 9 Cx HSS Cx Sh MME ISC CAM4home IMS gateway Text AS CAM4home web service server I-CSCF Mw Mw S-CSCF Mw IPTV AS P-CSCF Gm Media server IMS client MANE Figure 6: Prototype system. Table 4: Average Bitrate. Bitrates (kbps) QCIF CIF CIF (kbps) 15 1 (db) QCIF CIF 4CIF Figure 7: Output bitrate of adapted video. 1 5 QCIF CIF 4CIF Figure 8: Output Y-PSNR of adapted video. the CAM4Home metadata model as a bundle of multimodal media. MPEG-21 DIA was further integrated into CAM4Home model to meet end-to-end QoS requirements. We addressed the issues in supporting QoS from two aspects, namely, fidelity and modality, in order to tailor and adapt multimedia to the diverse terminals and the heterogeneous networks, as well as dynamic network conditions. The social annotations were used to enrich CAM4Home metadata collaboratively. Finally, a prototype system was developed on IMS architecture to validate the proposed model. With the use of rich metadata, context awareness, and personalization could be challenging topics in the future. Acknowledgments This paper was supported in part by the projects of SERVERY and CAM4Home. The authers would like to thank all partners for their contributions and thank Hui Wang and Mengke Hu for their simulation work.

72 1 EURASIP Journal on Wireless Communications and Networking References [1] D. Benslimane, S. Dustdar, and A. Sheth, Services mashups: the new generation of web applications, IEEE Internet Computing, vol. 12, no. 5, pp , 28. [2] A. Brodt and D. Nicklas, The TELAR mobile mashup platform for Nokia internet tablets, in Proceedings of the 11th International Conference on Extending Database Technology (EDBT 8), pp. 7 74, March 28. [3] B. Falchuk, K. Sinkar, S. Loeb, and A. Dutta, Mobile contextual mashup service for IMS, in Proceedings of the 2nd International Conference on Internet Multimedia Services Architecture and Application (IMSAA 8), December 28. [4] X. Z. Liu, G. Huang, and H. Mei, A community-centric approach to automated service composition, Science in China, Series F, vol. 53, no. 1, pp. 5 63, 21. [5] J. Rao and X. Su, A survey of automated Web service composition methods, in Proceedings of the 1st International Workshop on Semantic Web Services and Web Process Composition, July 24. [6] CAM4Home Official Website, [7] Yahoo Pipes, [8] Microsoft Popfly, [9] Z. Yu, X. Zhou, Z. Yu, D. Zhang, and C.-Y. Chin, An OSGIbased infrastructure for context-aware multimedia services, IEEE Communications Magazine, vol. 44, no. 1, pp , 26. [1] Z. Yu, X. Zhou, D. Zhang, C. Chin, X. Wang, and J. Men, Supporting context-aware media recommendations for smart phones, IEEE Pervasive Computing, vol. 5, no. 3, pp , 26. [11] Z. Yu, Y. Nakamura, D. Zhang, S. Kajita, and K. Mase, Content provisioning for ubiquitous learning, IEEE Pervasive Computing, vol. 7, no. 4, pp. 62 7, 28. [12] L. Zhou, N. Xiong, L. Shu, A. Vasilakos, and S. S. Yeo, Context-aware middleware for multimedia services in heterogeneous networks, IEEE Intelligent Systems, vol. 25, no. 2, pp. 4 47, 21. [13] L. Zhou, X. Wang, W. Tu, G. Mutean, and B. Geller, Distributed scheduling scheme for video streaming over multichannel multi-radio multi-hop wireless networks, IEEE Journal on Selected Areas in Communications, vol.28,no.3,pp , 21. [14] L. Zhou, B. Geller, B. Zheng, A. Wei, and J. Cui, Distributed resource allocation for multi-source multi-description multipath video streaming over wireless networks, IEEE Transactions on Broadcasting, vol. 55, no. 4, pp , 29. [15] Q. Wu, A. Iyengar, R. Subramanian, I. Rouvellou, I. Silva-Lepe, and T. Mikalsen, Combining quality of service and social information for ranking services, in Proceedings of the 7th International Joint Conference on Service-Oriented Computing, vol. 59 of Lecture Notes in Computer Science, pp , 29. [16] V. Cardellini, E. Casalicchio, V. Grassi, and F. Lo Presti, Scalable service selection for web service composition supporting differentiated QoS classes, Tech. Rep., [17] T. C. Thang, Y. J. Jung, and Y. M. Ro, Modality conversion for QoS management in universal multimedia access, IEE Proceeding Vision, Image Signal Process, vol. 152, no. 3, pp , 25. [18] T. C. Thang, Y. J. Jung, and Y. M. Ro, Semantic quality for content-aware video adaptation, in Proceedings of the IEEE 7th Workshop on Multimedia Signal Processing (MMSP 5), November 25. [19] T. C. Thang, J.-G. Kim, J. W. Kang, and J.-J. Yoo, SVC adaptation: standard tools and supporting methods, EURASIP Signal Processing: Image Communication, vol.24,no.3,pp , 29. [2] M. Eberhard, L. Celetto, C. Timmerer, E. Quacchio, H. Hellwagner, and F. S. Rovati, An interoperable multimedia delivery framework for scalable video coding based on MPEG-21 digital item adaptation, in Proceedings of the IEEE International Conference on Multimedia and Expo (ICME 8), pp , June 28. [21] H. Schwarz, D. Marpe, and T. Wiegand, Overview of the scalable video coding extension of the H.264/AVC standard, IEEE Transactions on Circuits and Systems for Video Technology, vol. 17, no. 9, pp , 27. [22] A.Szwabe,A.Schorr,F.J.Hauck,andA.J.Kassler, Dynamic multimedia stream adaptation and rate control for heterogeneous networks, Journal of Zhejiang University: Science A, vol. 7, no. 1, pp , 26. [23] D. Schall, H.-L. Truong, and S. Dustdar, Unifying human and software services in web-scale collaborations, IEEE Internet Computing, vol. 12, no. 3, pp , 28. [24] A. Maaradji, H. Hacid, J. Daigremont, and N. Crespi, Social composer: a social-aware mashup creation environment, in Proceedings of the ACM Conference on Computer Supported Cooperative Work, February 21. [25] M. Treiber, K. Kritikos, D. Schall, D. Plexousakis, and S. Dustdar, Modeling context-aware and socially-enriched mashup, in Proceedings of the 3rd International Workshop on Web APIs and Services Mashups, October 29. [26] D. Mukherjee, E. Delfosse, J.-G. Kim, and Y. Wang, Optimal adaptation decision-taking for terminal and network qualityof-service, IEEE Transactions on Multimedia,vol.7,no.3,pp , 25. [27] Information Technology Multimedia Framework (MPEG- 21) Part 1: Vision, Technologies, and trategy, 22. [28] S. A. Golder and B. A. Huberman, Usage patterns of collaborative tagging systems, Journal of Information Science, vol. 32, no. 2, pp , 26. [29] H. Zhang, H. Nguyen, N. Crespi, S. Sivasothy, T. A. Le, and H. Wang, A novel metadata-based approach for content and service combined delivery over IMS, in Proceedings of the 8th Conference on Communication Networks and Services Research, May 21. [3] OMA-UAPROF, User Agent Profiling Specification (UAPROF) 1.1, Open Mobile Alliance, December 22. [31] IETF RFC 384, Indicating User Agent Capabilities in the Session Initiation Protocol (SIP), August 24. [32] A.Kassler,T.Guenkova-Luy,A.Schorr,H.Schmidt,F.Hauck, and I. Wolf, Network-based content adaptation of streaming media using MPEG-21 DIA and SDPng, in Proceedings of the 7th International Workshop on Image Analysis for Multimedia Interactive Services, 26. [33] IETF RFC 3264, An Offer/Answer Model with Session Description, June 22. [34] A. Vetro, Ch. Timmerer, and S. Devillers, Information Technology Multimedia Framework Part 7: Digital Item Adaptation, ISO/IEC JTC 1/SC 29/WG11/N5933, October 23. [35] T. Özçelebi, I. Radovanović, and M. Chaudron, Enhancing end-to-end QoS for multimedia streaming in IMS-based networks, in Proceedings of the 2nd International Conference on Systems and Networks Communications (ICSNC 7), August 27.

73 EURASIP Journal on Wireless Communications and Networking 11 [36] T. Magedanz, D. Witaszek, and K. Knuettel, The IMS FOKUS an open testbed for next generation network multimedia services, in Proceedings of the 1st International Conference on Testbeds and Research Infrastructures for the Development of Networks and Communities (Tridentcom 5), February 25. [37] UCT Advanced IPTV, [38] Mobicents, [39] SIPp, [4] SVC Reference Software (JSVM Software), imagecom G1/savce/downloads/SVC-Reference-Software.htm.

74 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 8 pages doi:1.1155/21/ Research Article Packet-Scheduling Algorithm by the Ratio of Transmit Power to the Transmission Bits in 3GPP LTE Downlink Jungsup Song, 1 Gye-Tae Gil, 2 and Dong-Hoi Kim 1 1 School of Information Technology, Kangwon National University, Hyoja-dong, Chuncheon 2-71, Republic of Korea 2 Central R&D Laboratory, Korea Telecom (KT), 463-1, Jeonmin-dong, Yuseong-gu, Daejeon , Republic of Korea Correspondence should be addressed to Dong-Hoi Kim, donghk@kangwon.ac.kr Received 22 February 21; Accepted 13 July 21 Academic Editor: Neal N. Xiong Copyright 21 Jungsup Song et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Packet scheduler plays the central role in determining the overall performance of the 3GPP long-term evolution (LTE) based on packet-switching operation. In this paper, a novel minimum transmit power-based (MP) packet-scheduling algorithm is proposed that can achieve power-efficient transmission to the UEs while providing both system throughput gain and fairness improvement. The proposed algorithm is based on a new scheduling metric focusing on the ratio of the transmit power per bit and allocates the physical resource block (PRB) to the UE that requires the least ratio of the transmit power per bit. Through computer simulation, the performance of the proposed MP packet-scheduling algorithm is compared with the conventional packet-scheduling algorithms by two primary criteria: fairness and throughput. The simulation results show that the proposed algorithm outperforms the conventional algorithms in terms of the fairness and throughput. 1. Introduction The 3GPP long-term evolution (LTE) standard, which is a subset of the upgraded specifications of 3G network system, aims at the goals such as peak data rate of 1 Mbps in downlink and 5 Mbps in uplink, throughput increase at the cell boundary, spectral efficiency improvement, and scalable bandwidth [1 3]. As the 3GPP LTE was developed under the assumption of a packet-switching operation, the packet scheduler plays the central role in determining the overall system performance. Several packet schedulers, focusing on fairness and throughput maximization, were introduced in [4 9] based on the round robin (RR), proportional fair (PF), and maximum throughput (MT) algorithms. To reduce the complexity, most schedulers operate in two phases: time domain packet scheduler (TDPS) followed by frequency domain packet scheduler (FDPS) [4, 5]. The efficient FDPS in [6] showed drastic increase in system throughput and cell coverage. In [7, 8], the authors proved significant improvement of spectral efficiency in 3GPP LTE downlink. Reference [9] showed that the PF algorithm provides fairness improvement but shows little decrease of throughput. Packet scheduling algorithms for mixed traffic system were also been proposed and evaluated in [1, 11], but only the data rate was adopted in the scheduling metric. In this paper, we propose a novel minimum transmit power-based (MP) packet scheduling algorithm that can achieve power-efficient transmission to the UEs while providing both system throughput gain and fairness improvement. The proposed algorithm is based on the ratio of the transmit power to the number of transmission bits. Thus, the proposed MP scheduler allocates the physical resource block (PRB) to the UE that requires the least ratio of the transmit power per bit. For this, it is assumed that the channel quality indication (CQI) information for all UE channels is available at the evolved Node B (enb), with which the modulation and coding scheme (MCS) level and the UE transmit power are determined. The performance of the proposed MP algorithm is compared with the conventional algorithms through computer simulation, considering real-time and non-real-time traffic in multicell environments. The rest of this paper is organized as follows. Sections 2 and 3 briefly describe the packet-scheduling model

75 2 EURASIP Journal on Wireless Communications and Networking L2 level data buffer Classifier FDPS Packet scheduler Link adaptation L1 TDPS... CQI HARQ QoS Figure 1: The structure of RT and NRT traffic packet scheduler in enb. and algorithms, respectively. Section 4 explains the simulation environment. The simulation results are discussed in Section 5, and we conclude this paper in Section Packet Scheduling Models The basic structure of downlink packet scheduler for RT and NRT traffics in enb of the 3GPP LTE is depicted in Figure 1. The packet scheduler is divided into two phases: TDPS and FDPS. In the TDPS, a small group of UEs are chosen as the scheduling candidate set (SCS) based on diverse metrics: buffer size, delay, CQI reports, and so forth. The TDPS does not allocate PRBs to the UEs, but it conveys the information of the UEs becoming scheduling candidates to the FDPS. In the FDPS, the PRBs at Layer 1 are directly allocated to the SCS received from the TDPS. In a mixed traffic system, a classifier is necessary for the efficiency of packet scheduling. The classifier sets independent queues based on traffic types, and each queue is given its own priority. Thus, each traffic type can be independently handled. With the classifier, the packet scheduler cooperates with the CQI manager, hybrid automatic repeat request (HARQ), link adaptation, and QoS manager. The link adaptation decides a proper MCS level for respective UE and PRB combinations based on the CQI which acts as the primary criterion [12]. The PRBs with good channel conditions are given a MCS level sending a lot of data [13]. The QoS manager checks UEs QoS requirements and the packet scheduler calculates packet scheduling metrics Classifier. In Figure 1, the classifier classifies the mixed traffic atlayer2databuffer according to the type of traffic. Because each traffic type has its own QoS requirement, the classifier is necessary in a mixed traffic systemforefficient packet scheduling. In this paper, we assume that RT and NRT traffics exist at the same time. The classifier is provided with traffic statements from L2 buffer and sets two independent queues for RT and NRT traffics assigning different priorities to the queues. Under the consistent traffic environment, the most efficient resource allocation scheme is divided into two adaptations. First of all, voice streaming and WWW data service exemplify RT and NRT traffic in real systems.rt traffics such as voice streaming have constant bit rate (CBR) feature. Margin adaptation for OFDMA systems [14] is considered as the most efficient resource allocation scheme for power minimization of RT traffics. On the other hand, NRT traffics like WWW data service have best effort (BE) characteristic. Rate adaptation [14] is known as the most efficient resource allocation scheme for throughput maximization of NRT traffic with a power constraint. Therefore, in order to maximize the system throughput and to minimize the transmit power of mixed RT as well as NRT traffics at the same time, efficienttransmitpower consumption becomes a key issue. Generally, RT traffics need to deal with a delay constraint, so the higher priority is essential [15]. Different priorities and power constraint influence the PRB allocation during one transmission time interval (TTI). Because the RT traffic features a delay constraint and CBR, the PRBs are firstly allocated to RT traffic UEs. After PRB allocation for the RT traffic, the NRT traffic UEs, having BE characteristic, consume the remaining transmit power for PRB allocation, aiming at bit rate maximization [15] Time Domain Packet Scheduling. The main purpose of the TDPS is to set the SCS. The TDPS does not directly allocate the PRBs, but it restricts the number of UEs for the FDPS to reduce the scheduling complexity. The SCS is chosen based on a computed metric such as the CQI, throughput, delay, and so forth. The SCS information is conveyed to the FDPS and only the UEs restricted by the TDPS are qualified as the FDPS candidates. The TDPS should concern the data in L2 buffer and HARQ, simultaneously. When retransmission is requested through HARQ, UEs requesting HARQ are automatically comprised in the SCS Frequency Domain Packet Scheduling. In the FDPS phase, the PRBs are directly allocated to the UEs and their data are transmitted. It delivers the allocated data after packet scheduling to physical level (L1) devices, and then the L1 devices send the data by modulated signal through physical channel. The FDPS considers only the SCS during one TTI. The FDPS is completed when all transmit power is consumed. A UE can load the information on the plural PRBs, but a PRB cannot be shared by more than one UEs at the same time. 3. Packet-Scheduling Algorithms 3.1. Conventional Packet-Scheduling Algorithms. Diverse packet scheduling algorithms were introduced and their performances were evaluated in terms of system throughput and fairness [16 18]. For the best fairness, the RR algorithm can be applied. In the RR algorithm, the scheduler at time t uses the information on the elapsed time since the latest

76 EURASIP Journal on Wireless Communications and Networking 3 scheduled time (t s )foreachues as the scheduling metric [1]: that is, ŝ = arg max t t s = arg min t s, s s (1) where ŝ denotes the selected UE index. The MT algorithm focuseson the spectralefficiency and achieves the best system throughput. In 3GPP LTE system, data rate to be transmitted is affected by the MCS level decided by the link adaptation based on the CQI reported from the corresponding UE. For the higher CQI, the link adaptation selects a higher MCS level with more bits per symbol. The data rate D s,n is calculated based on the recommended MCS level. Thus, the MT scheduler is expressed as (ŝ, n) = arg max D s,n = arg max Q s,n, s,n s,n (2) where n is the index of the selected PRB, and Q s,n denotes the CQI of the PRB n reported from the UE s. In other word, the UE with the highest data rate acquires the highest priority. The PF algorithm was introduced to solve monopolized situation in the MT algorithm. Scheduling metric is defined as the data rate divided by the past average user data rate. Thus, the scheduling metric is equal to the ratio of D s,n to the average past user data rate R s as (ŝ, n) = arg max s,n D s,n R s. (3) 3.2. Proposed MP Packet-Scheduling Algorithm. In order to improve the fairness and throughput, most of conventional algorithms including the MT and PF consider the instantaneous channel condition and throughput as key factors of scheduling metric. However, new factors should be considered to enhance the system performance. One of them is the ratio of the transmit power per bit, which has not been considered yet for packet scheduling. The transmit power is insufficient when the radio resources are fully utilized, huge amount of data are required to be transmitted, and most UEs have poor channel conditions. In this case, if scheduling metric of a packet scheduling algorithm considers the ratio of the transmit power to the number of transmission bits, more improvement in the system performance is expected. For this reason, in a system with limited transmit power, it is the most efficient to allocate PRBs to the UEs that requires the least ratio of the transmit power to the number of transmission bits. Thus, in the proposed MP scheduling algorithm, the scheduling metric selects the UEs to be allocated in ascending order of the ratio of the transmit power P s,n to the number of transmission bits b s,n as follows: (ŝ, n) = arg min s,n P s,n f ( ) b s,n = arg min, (4) b s,n s,n g s,n b s,n where g s,n is the channel power of the PRB n of the UE s. In (4), assuming that the same MCS level is used for all subcarriers in a PRB, the minimum transmit power f (b s,n ) Higher MCS Level MCS Level 1 MCS Level 2 MCS Level 3. MCS Level n Lower MCS Level. CQI of UE k CQI of UE j CQI of UE i Larger M(s, n) formp. Upper bound for j MCS level n Lower bound for MCS level n Smaller M(s, n) for MP Figure 2: MCS levels and scheduling metric calculation in the proposed packet-scheduling algorithm. required for transmission of b s,n bits with the target BER of P e is given by [19] f ( ) σ 2 [ ( )] s,n b s,n = Q 1 Pe 2(2 b s,n 1 ), (5) 3 4 where σs,n 2 is the noise variance for the subcarriers in the PRB n at the UE s,andq(x) = 1/ 2π x e t/2 dt. Assuming that the link adaptation is employed and that the maximum transmit powers of the enb are large enough, (4) can be rewritten as (see the appendix) (ŝ, n) = arg max s,n M(s, n), (6) where the scheduling metric M(s, n) is expressed by M(s, n) = Δ s,n f ( b s,n ) /bs,n, (7) and Δ s,n denotes the excess channel gain defined by Δ s,n = g s,n g min (b s,n ); g min (b s,n ) is the minimum channel gain required for the successful transmission of b s,n bits; b s,n is the maximum positive integer that satisfies Δ s,n. From (7), the MP scheduler assigns the PRB n to the UE with larger excess channel gain compared to the required received power per bit. For the UEs with equal value of excess channel gain, the MP scheduler assigns the PRB to the UE with smaller received power per bit. For example, consider UE k, j, andi in Figure 2 ranked on MCS level 1,2, and 3, respectively. In the figure, the MCS level 1 sends the highest data rate while the MCS level n transmits the lowest data rate. According to the 3GPP LTE AMC scheme, UE k is able to transmit more bits than UE j but UE k requires lower transmit power per bit than UE j. It is because the CQI of UE k is much larger than the minimum required CQI for the MCS level 1 which may require small transmit power, while the CQI for UE j is close to the minimum value for the MCS level 2 which requires larger transmit power than the other cases. Meanwhile, UE i has almost the same i k

77 4 EURASIP Journal on Wireless Communications and Networking excesschannelgainasuek, but it requires less received power per bit, f (b s,n )/b s,n, than UE k because f (b s,n )/b s,n in (7) increases exponentially with b s,n ; hence, the value of f (b s,n )/b s,n for UE i is smaller than UE k having higher MCS level than UE i. Therefore, the MP scheduler selects the UEs to be allocated in order of UE i, UEk, anduej. After all, for the efficiency of power consumption, the MP algorithm considers the transmit power and the number of transmission bits at the same time. The implementation complexity of the MP scheduling rule in (4)canbereducedasfollows.Define ω ( [ ( )] ) 1 b s,n = Q 1 Pe 2(2 b s,n 1 ). (8) 3b s,n 4 Then, (7) canberewrittenas M(s, n) = g ( ) s,n g min bs,n σs,nω 2 ( ). (9) b s,n Because g min (b s,n )andω(b s,n ) can be precalculated for all possiblevaluesofb s,n, the calculation of the metric in (9) is much simpler than the metric in (4). 4. Simulation Environment The algorithm evaluation is based on the 3GPP LTE downlink specificationsdefined in [1] and the simulation scenario in [2]. The 19-cell model with wrap around is assumed, in which omnidirectional antennas are used and the UEs are uniformly distributed. Calls are generated based on Poisson arrival rate and a simple admission control is applied in order to prevent users from gathering in a few cells. The admission control blocks a new call into a cell when the number of users in the cell is equal to the limit. The other simulation parameters are described in Table 1. One TTI is one subframe duration of 1 millisecond, during which 14 symbols are transmitted. Our simulation assumes 5 MHz transmission bandwidth, thus 25 PRBs are available during one TTI. The link adaptation selects the modulation mode for a user based on the CQI. An infinite buffer model is applied. We assume two different traffic types: RT traffic and NRT traffic. RT traffic needs to guarantee a target CBR for successful transmission hence, we set the guaranteed bit rate (GBR) as 64 kbps. Moreover, RT traffic has higher priority than NRT traffic because RT traffic is vulnerable to delay constraint. On the other hand, even though NRT traffic does not need to be guaranteed and is not sensitive to delay constraint, the remaining power after the transmit power consumption for RT traffic is used for NRT traffic since all transmission power must be spent during one TTI at enbs in order not to waste spectrum. Note that the HARQ scheme is not applied in this paper since it is beyond the scope of this paper. 5. Simulation Results The proposed MP packet scheduling algorithm is compared with the conventional MT, RR, and PF packet scheduling algorithms. Among the conventional three algorithms, Table 1: Simulation parameters. Parameter Setting Simulaion time 1 TTIs OFDM symbolsper TTI 14 (4 symbols for control) Subframe Length 1ms (14symbols) Transmission Bandwidth 5MHz Number of Subcarriers 3 Number of PRBs 25 (12 subcarriers per PRB) Total enb transmit power 46 dbm Modulation Schemes QPSK, 16-QAM, 64-QAM Channel coding notused Network Synchronous Reuse factor 1 CQI measurement period 1ms Max scheduled users 5 Round Robin, Packet Scheduler Max Throughput, Proportional Fairness, Minimum Transmit Power-based Traffictypes Real time (CBR), Non-real-time traffic(be) Pas loss factor 3.5 Standard deviation of shadowing 6.5 Tx/Rx antenna type SISO the MT algorithm shows the best throughput and the RR algorithm the worst throughput. However, in terms of fairness, the RR algorithm achieves the best performance and the MT algorithm shows the worst performance. The worst fairness of the MT algorithm is attributed to the monopolization of spectrum resource by only a few UEs with good CQIs. On the other hand, UEs with poor CQIs can be given a higher priority in the PF algorithm by using a different metric from the MT algorithm as divided by the past average data rate. Therefore, in despite of the poor channel states, the UEs can precede other UEs having good channel conditions. Monopolizing UEs tend to be located near enbs at the center of the cells. By applying the PF and RR algorithms, user throughput at the cell edge can be increased. In the following figures, the paired labels of the packet scheduling algorithms are applied for TDPS and FDPS in order. For example, the labeled MT-MT refers the MT algorithm used for both of the TDPS and the FDPS Average User and Cell Throughput Performance. Figure 3 shows the average user throughput, which is defined as the ratio of the total throughput in a cell divided by the total number of UEs, with different maximum number of UEs in a cell. From Figure 3, we find that the MP- MP algorithm achieves even better average UE throughput than the MT-MT algorithm. The MP algorithm s spectral efficiency seems to be more efficient than the other packet

78 EURASIP Journal on Wireless Communications and Networking 5 Average user throughput (Kbps) Maximum number of UEs at a cell Average user throughput (Kbps) Maximum number of UEs at a cell MT-MT RR-RR (a) Mixed trafficues Average user throughput (Kbps) PF-PF MP-MP MT-MT RR-RR Maximum number of UEs at a cell (b) NRT trafficues PF-PF MP-MP MT-MT RR-RR (c) RT trafficues PF-PF MP-MP Figure 3: Average user throughput versus maximum number of UEs in a cell. scheduling algorithms as the maximum number of UEs in a cell increases. When maximum 25 users exist in a cell, the MP-MP algorithm achieves 18% increase of average user throughput compared to the MT-MT algorithm. It is also found that most of the gain of average user throughput of mixed traffic UEs in Figure 3(a) comes from the NRT traffic UEs in Figure 3(b). It is because NRT traffic UEs having BE feature can receive as many available data as possible, while RT traffic UEs do not receive more data than their target data rates. In Figure 3(c), the MP algorithm also shows the best capacity of RT traffic because more capacity is provided when the algorithm is applied. Under the same maximum number of UEs in a cell, the MP-MP algorithm shows the best throughput per UE. This result indicates that better average user throughput occurs with more UEs. It is because of efficiency of transmit power consumption. Under the saturation of a cell, the transmit power consumption becomes a more critical issue because power is a limited resource. Therefore, from the results, the packet scheduling algorithm by the ratio of the transit power to the number of transmission bits provides a great increase of the average user throughput. Average call throughput (Mbps) MT-MT RR-RR Call arrival rate (times/s) PF-PF MP-MP Figure 4: Average cell throughput in the whole cell with various call arrival rates. Figure 4 shows the average cell throughput (i.e., the average of the 19 cell throughputs) with the call arrival rate,

79 6 EURASIP Journal on Wireless Communications and Networking Cell throughput (Kbps) Cell throughput (Mbps) RR-RR PS-PS MP-MP MT-MT MT-MT RR-RR Call arrival rate (times/s) PF-PF MP-MP Figure 5: Average cell throughput at the cell boundary with various call arrival rates. Cell throughput (Mbps) Transmit power (dbm) MT-MT RR-RR PF-PF MP-MP Figure 6: Average cell throughput in the whole cell with various transmit power. assuming maximum 15 UEs in each cell. It shows that the MP-MP algorithm achieves the best average cell throughput. As call arrival rate increases, the MP-MP algorithm provides more eminent performance. For example, when call arrival rate is 1 2, the algorithm shows 6% gain in the average cell throughput for total UEs compared to the MT-MT algorithm. Figure 5 shows the average cell throughput at the cell boundary with call arrival rate. In the simulation, 2% of the the UEs were located at the cell boundary in which the power-efficiency is particularly important. Compared to the RR-RR algorithm, 7% gain of the MP-MP algorithm at the cell boundary is obtained for the call arrival rate of 1 2.The improved spectrum efficiency comes because the proposed MP scheduling algorithm considers the ratio of the transmit power to the number of transmission bits. Figure 6 shows the average cell throughput with the transmit power, where the maximum allowable transmit power is 46 dbm as given in the 3GPP LTE downlink specification [1]. From the figure, the MP-MP algorithm Fairness (%) MT-MT: 1UEs RR-RR: 1UEs PF-PF: 1UEs MP-MP: 1UEs MT-MT: 15UEs RR-RR: 15UEs PF-PF: 15UEs MP-MP: 15UEs MT-MT: 2UEs RR-RR: 2UEs PF-PF: 2UEs MP-MP: 2UEs MT-MT: 25UEs RR-RR: 25UEs PF-PF: 25UEs MP-MP: 25UEs Figure 7: Fairness and cell throughput. can sustain more than 1 Mbps average cell throughput with 3 dbm. In addition, the MP-MP algorithm can save the transmit power about 8 dbm than the MT-MT algorithm while sustaining the same cell throughput Fairness Performance. Figure 7 shows fairness and cell throughput. Here, the fairness is defined as the ratio of the best 5% UEs throughput to the total cell throughput. The MT-MT algorithm shows the worst fairness as expected. In the MT-MT algorithm, the best 5% UEs occupy approximately 2% out of the whole cell throughput. On the other hand, in the RR-RR and PF-PF algorithms, although cell throughput shows less than 1 Mbps, the best 5% UEs occupy less than 1%. However, by the MP-MP algorithm, the cell throughput is more than 1 Mbps and the best 5% UEs occupy less than 1% of the cell throughput. As a result, the MP-MP algorithm provides better performance in terms of not only cell throughput but also fairness than the other algorithms. Figure 8 shows the distribution of normalized throughput with respect to the UE index. Here, the normalized throughput is defined as the ratio of the throughput per UE to the total throughput in a cell. From the figure, it is found that large portion of normalized throughput is centralized in only a few UEs with good channel conditions by the MT- MT algorithm. However, the normalized throughput by the RR-RR, PF-PF, and MP-MP algorithms are fairly distributed. The normalized throughput by the MP-MP algorithm shows relatively equal transmission probabilities for all UEs. Figure 9 shows the distribution of the normalized throughput of a UE with the distance from the serving enb normalized by the cell radius. Because the distance is the most important factor which affects the channel condition, in the MT-MT and PF-PF algorithms, the normalized throughput is centralized and decreases as far from

80 EURASIP Journal on Wireless Communications and Networking 7 Normalised throughput Normalised throughput MT-MT RR-RR PF-PF MP-MP Index number of user equipments Figure 8: Normalized throughput distribution per UE Distance MT-MT RR-RR PF-PF MP-MP Figure 9: Normalized throughput distribution according to distance. the center. However, normalized throughput in the RR-RR and MP-MP algorithms randomly spreads over the all region. The reason is because the MP algorithm considers the ratio of the transmit power to the number of transmission bits. From Figures 7, 8, and9, we find that the MP algorithm provides improved performance in terms of fairness and throughput. Specially, Figure 9 shows throughput increase at the cell boundary. 6. Conclusion In this paper, we presented the decoupled packet scheduling algorithms in 3GPP LTE systems and proposed a novel algorithm which considers the efficiency of transmit power consumption. The performance of the proposed algorithm was evaluated by comparing with the conventional algorithms: the maximum throughput (MT), round robin (RR), and proportional fairness (PF). From the simulation results, the MP-MP algorithm applying the proposed minimum transmit power-based packet scheduling (MP) algorithm to the time domain packet scheduler (TDPS) and the frequency domain packet scheduler (FDPS) in 3GPP LTE systems showed better throughput performances than the other conventional algorithms. Moreover, the MP-MP algorithm showed significant improvement of the fairness performance, which comes from the different packet scheduling metric regarding the ratio of the transmit power to the number of transmission bits. Conclusively, from the results, we confirm that the proposed scheduling metric successfully improves the system performance such as the fairness and throughput. Further work includes CQI reporting scheme because the performance of the proposed downlink scheduling algorithm, as well as the conventional ones, depends on the accuracy of the CQI information. Appendix Let Ps,n max denote the maximum transmit power at the enb that can be assigned for the UE s and the PRB n. Then, the minimum channel gain required for successful transmission of b s,n bits through the PRB n is given by g min (b s,n ) = f (b s,n )/Ps,n max,where f (b s,n )isdefinedin(5). Since we have g s,n = f (b s,n )/P s,n, the excess channel gain Δ s,n is written as ( ) ( ) ( 1 Δ s,n = g s,n g min bs,n = f bs,n 1 ) P s,n Ps,n max. (A.1) From (A.1), we get 1 P s,n = Using (A.2)in(4), weget (ŝ, n) = arg min s,n Δ s,n f ( b s,n ) + 1 ps,n max 1 ( Δs,n /f ( ) b s,n +1/P max s,n. (A.2) ) bs,n, (A.3) and, when Ps,n max is large enough, (A.3)canberewrittenas(6). Acknowledgment This work was financially supported by the Grant from the Industrial technology development program (Project no. KI2143) of the Ministry of Knowledge Economy (MKE) of Korea. References [1] 3GPP, Evolved universal terrestrial radio access (E-UTRA); LTE physical layer general description, Tech. Spec v8.2., 28.

81 8 EURASIP Journal on Wireless Communications and Networking [2] 3GPP, Evolved universal terrestrial radio access (E-UTRA); physical channels and modulation, Tech. Spec v8.5., 28. [3] A. Toskala, H. Holma, K. Pajukoski, and E. Tiirola, Utran long term evolution in 3GPP, in Proceedings of the IEEE 17th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 6), pp. 1 5, Helsinki, Finland, September 26. [4] P. Kela, J. Puttonen, N. Kolehmainen, T. Ristaniemi, T. Henttonen, and M. Moisio, Dynamic packet scheduling performance in UTRA long term evolution downlink, in Proceedings of the 3rd International Symposium on Wireless Pervasive Computing (ISWPC 8), pp , Santorini, Greece, 28. [5] G.Mongha,K.I.Pedersen,I.Z.Kovacs,andP.E.Mogensen, QoS oriented time and frequency domain packet schedulers for the UTRAN long term evolution, in Proceedings of the 52nd Vehicular Technology Conference (VTC 8), pp , Marina Bay, Singapore, May 28. [6] A. Pokhariyal, T. E. Kolding, and P. E. Mogensen, Performance of downlink frequency domain packet scheduling for the utran long term evolution, in Proceedings of the IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 6), pp. 1 5, Helsinki, Finland, 26. [7] N. Wei, A. Pokhariyal, C. Rom et al., Baseline E-UTRA downlink spectral efficiency evaluation, in Proceedings of the IEEE 64th Vehicular Technology Conference (VTC 6), pp. 1 5, Québec, Canada, September 26. [8] Y. Ofuji, T. Kawamura, Y. Kishiyama, K. Higuchi, and M. Sawahashi, System-level throughput evaluations in evolved UTRA, in Proceedings of the 1th IEEE Singapore International Conference on Communications Systems (ICCS 6), pp. 1 6, Singapore, October 26. [9] A. Pokhariyal, K. I. Pedersen, G. Monghal et al., HARQ aware frequency domain packet scheduler with different degrees of fairness for the UTRAN long term evolution, in Proceedings of the IEEE Vehicular Technology Conference (VTC 7), pp , Dublin, Ireland, 27. [1] J. Puttonen, N. Kolehmainen, T. Henttonen, M. Moisio, and M. Rinne, Mixed traffic packet scheduling in UTRAN long term evolution downlink, in Proceedings of the IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 8), pp. 1 5, Cannes, France, September 28. [11] M. Rinne, M. Kuusela, E. Tuomaala et al., A performance summary of the evolved 3G (E-UTRA) for voice over internet and best effort traffic, IEEE Transactions on Vehicular Technology, vol. 58, no. 7, pp , 29. [12] T. E. Kolding, F. Frederiksen, and A. Pokhariyal, Lowbandwidth channel quality indication for OFDMA frequency domain packet scheduling, in Proceedings of the 3rd International Symposium on Wireless Communication Systems (ISWCS 6), pp , Valencia, Spain, 26. [13] A. Jalali, R. Padovani, and R. Pankaj, Data throughput of CDMA HDR: a high efficiency-high data rate personal communication wireless system, in Proceedings of the 52nd IEEE Vehicular Technology Conference (VTC ), pp , Tokyo, Japan, May 2. [14] J. M. Cioffi, Digital Communications, Stanford University, Calif, USA, 23, EE379 Course Reader. [15] M. Tao, Y.-C. Liang, and F. Zhang, Resource allocation for delay differentiated traffic in multiuser OFDM systems, IEEE Transactions on Wireless Communications, vol.7,no.6,pp , 28. [16] C. Wengerter, J. Ohlhorst, and A. G.E. Von Elbwart, Fairness and throughput analysis for generalized proportional fair frequency scheduling in OFDMA, in Proceedings of the IEEE Vehicular Technology Conference (VTC 5), vol.61,no.3,pp , Stockholm, Sweden, May 25. [17]J.Lim,H.G.Myung,K.Oh,andD.J.Goodman, Proportional fair scheduling of uplink single-carrier FDMA systems, in Proceedings of the IEEE 17th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 6), pp. 1 6, Helsinki, Finland, September 26. [18] K. I. Pedersen, G. Monghal, I. Z. Kovács et al., Frequency domain scheduling for OFDMA with limited and noisy channel feedback, in Proceedings of the IEEE Vehicular Technology Conference (VTC 7), pp , 27. [19] C. Y. Wong, R. S. Cheng, K. B. Letaief, and R. D. Murch, Multiuser OFDM with adaptive subcarrier, bit, and power allocation, IEEE Journal on Selected Areas in Communications, vol. 17, no. 1, pp , [2] K. Brueninghaus, D. Astélyt, T. Salzer et al., Link performance models for system level simulations of broadband radio access systems, in IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 5), vol. 4, pp , Berlin, Germany, 25.

82 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID 3976, 1 pages doi:1.1155/21/3976 Research Article Cross-Layer Handover Scheme for Multimedia Communications in Next Generation Wireless Networks Yuliang Tang, 1 Chun-Cheng Lin, 2 Guannan Kou, 1 and Der-Jiunn Deng 3 1 Department of Communication Engineering, Xiamen University, Fujian 3615, China 2 Department of Computer Science, Taipei Municipal University of Education, Taipei 148, Taiwan 3 Department of Computer Science and Information Engineering, National Changhua University of Education, Changhua, Taiwan Correspondence should be addressed to Der-Jiunn Deng, djdeng@cc.ncue.edu.tw Received 27 February 21; Accepted 14 August 21 Academic Editor: Liang Zhou Copyright 21 Yuliang Tang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In order to achieve seamless handover for real-time applications in the IP Multimedia Subsystem (IMS) of next generation network, a multiprotocol combined handover mechanism is proposed in this paper. We combine SIP (Session Initiation Protocol), FMIP (Fast Mobile IPv6 Protocol), and MIH (Media Independent Handover) protocols by cross-layer design and optimize those protocols signaling flows to improve the performance of vertical handover. Theoretical analysis and simulation results illustrate that our proposed mechanism performs better than the original SIP and MIH combined handover mechanism in terms of service interruption time and packet loss. 1. Introduction In next generation wireless system, access networks can be carried out by different technologies, such as WiFi, WiMAX and UMTS, while the core network infrastructure is established on an all-ip-based network. There have existed a variety of applications for next generation wireless system, in which the multimedia service is one of main applications [1]. However, the characteristics of wireless systems provide a major challenge for reliable transport of multimedia since it is highly sensitive to interference and link or channel change, which may cause delay, packet loss, and jitter. The wireless networks have to cope with this lack of Quality of Service (QoS) guarantees [2]. For improving the QoS, many studies investigate how to optimize the system scheduling scheme of utilizing available network resources [3, 4]. The IP Multimedia Subsystem (IMS) is an architectural framework for delivering IP multimedia services, which applies Session Initiation Protocol (SIP) to controlling multimedia communication sessions. SIP can provide IP mobility by the REINVITE signaling. However, SIP has a longer endto-end signaling delay which may cause frequent disruption for real-time applications in node motion. Therefore, as the nodes move among heterogeneous wireless networks, one of the greatest challenges is how to provide fast and seamless mobility support. Media Independent Handover (MIH) standard [5] was proposed for solving the above problem. Some researches (see, e.g., [6, 7]) had been done by using MIH to improve the SIP-based node mobility handover process in vertical handover. In addition, MIH is also used to assist the Mobile IP-(MIP-) based handover process. Fast Mobile IPv6 Protocol (FMIP) is an extension to MIPv6 designed for eliminating the standard MIP handover latencies [8] and is a combined SIP+MIH handover architecture by cross-layer design. A combined FMIP+SIP handover architecture can be found in[9]. However, in fact, the improvement of the handover performance for the previous approaches is limited when only one or two kinds of protocols are combined to solve the handover problem. Hence, a better way to solve the problem is to design a combination of more layer protocols by cross-layer design. In this paper, our interests focus on how to accelerate the handover process. Handover occurs when a communicating node moves from one network to another. It can be classified into two modes: make-beforebreak and break-beforemake, inwhich the former connects to the new network before the node tears down the current connected network, while the latter

83 2 EURASIP Journal on Wireless Communications and Networking just does the other way. The make-beforebreak mode is more complicated to be implemented, but have a better performance on end-to-end delay and packet loss. In this paper, an integrated scheme of combining FMIP, SIP and MIH signaling is proposed to optimize the performance of vertical handover on make-beforebreak mode. The rest of this paper is organized as follows. Section 2 introduces the relevant protocols and related work. The proposed handover mechanism is given in Section 3. Section 4 shows the simulation results and Section 5 concludes this paper. At the end, in order to facilitate the understanding of this paper, some key terminologies used in this paper are listed in the Abbreviations. Handover decision module SIP/applications MIP/FMIP MIH SAP MIHF MIH LINK SAP 82.11u 82.16g... UMTS Figure 1: Multiprotocol architecture in heterogeneous networks. 2. Background 2.1. Relevant Protocols. SIP is a signaling protocol, which is widely used for controlling multimedia communication sessions such as voice and video calls over IP. It supports terminal mobility when a mobile node (MN) moves to a different location before a session establishment or during the middle of a session. Before the REINVITE signaling of SIP, the correspondent node (CN) can send data to the MN prior to the registration of MIP. However, even though working with MIP, SIP still needs a new care of address (NCoA) whose configuration spends more time. IEEE 82.21, a.k.a., Media-Independent Handover (MIH), is designed to optimize the handover between heterogeneous networks so that the continuity of transparent services comes true. The MIH consists of a signaling framework and the triggers that make available information from the lower layers (MAC and PHY) to the higher layers of the protocol stack (network layer to application layer). Furthermore, MIH is responsible for unifying a variety of L2-specific technology information used by the handover decision algorithms so that the upper layers can abstract the heterogeneity that belongs to different technologies. The core idea of MIH is the introduction of a new functional module Media Independent Handover Function (MIHF), which operates as a glue of L2 and L3 (see Figure 1). MIHF accesses various MAC layers in heterogeneous networks, controls them through different service access points (MIH LINK SAP), and provides to up-layer users a media independent service access point (MIH SAP), such as FMIP, SIP. It is accomplished through three services: media-independent event service (MIES), media independent information service (MIIS), and media-independent command service (MICS). In MIES, the MIH user can be notified a certain event by the local or remote MIHF. The MIH events are made available to upper layers through the MIH SAP, such as MIH Link Up (the L2 connection is established, and the link is available for the user), MIH Link Going Down (the loss of the L2 connection is imminent), and MIH Link Down (the L2 connection is lost). The MIIS is a function for MIHF which discovers the information of available neighboring networks to facilitate the network selection and handover. It provides mostly static information. The MICS gathers the information on the status of connected links and the connectivity decision to the lower layers by offering commands to upper layers (e.g., scanning available networks). Therefore, the MICS commands control, manage, and send actions to lower layers, and can be issued by both local and remote MIH users. There is an IETF workgroup-mipshop, which addresses an L3 transport mechanism for the reliable delivery of MIH messages between different access networks. MIPv6 was designed to enable MNs to maintain connectivity when they move from one network to another. However, the latency caused by the MIPv6 operation is unacceptable for real-time applications. To overcome this problem, fast handovers for the Mobile IP protocol have been proposed by the Mobile IP working group of the IETF, which enables an MN to connect to a new point of attachment more rapidly. Fast Mobile IPv6 (FMIP) applies an unclearlydefined link layer event to triggering the mobile node s beginning handover process while the MN still connects to the previous link. The MN exchanges the RtsolPr/PrRtAdv (Router Solicitation for Proxy Advertisement and Proxy Router Advertisement) message with the previous access router (PAR) to obtain the target access router s MAC, IP addresses, and valid prefix. By using the retrieved information, the MN formulates a prospective new care of address (NCoA) and sends a fast binding update (FBU) to the PAR. The purpose of the FBU is to authorize the PAR to bind previous care of address (PCoA) to NCoA, so that arriving packets can be tunneled to the new location of the MN. The PAR sends a Handover Initiate (HI) message to carry the NCoA to the NAR which determines whether NCoA is unique on the new link interface or not by duplicate address detection (DAD). The PAR will return the available address in the FBack. After attaching to the new network, the MN sends an unsolicited neighbor advertisement (UNA) immediately, so that the buffered packets at NAR can be forwarded to the MN right away. The tunnel created between the two routers remains active until the MN completes the binding update with its correspondent node. Note that the buffer packet to NAR can extremely reduce the packet loss but the service will be interrupted between FBU and UNA. If the FMIP is triggered to begin the handover process timely, the handover delay can be reduced a lot, but the protocol is not specific to the trigger

84 EURASIP Journal on Wireless Communications and Networking 3 method. This problem can be overcome by introducing MIH. The MIH provides intelligence to the link layer such as the link going down triggers to wake up FMIP Related Work. In [1, 11], some schemes of integrating SIP and MIP have been proposed to optimize the mobility management. For achieving the fast handover procedure, cross-layer schemes have been investigated widely. Among them, some use MIH to facilitate handover while others do not. In [12], an integrated mobility scheme is proposed to combine the procedures of FMIP and SIP. But without MIH, the real-time requirement of L2 trigger is still an unresolved problem. The scheme in[13] suggests to combine MIH and SIP, but, even if it claims to make handover before breaking the link, it does not consider the packet loss while the old link quality becomes poor. The schemes in [14, 15] use existing MIH services to optimize the FMIP. MIH is used to reduce the time of discovering Access Router (AR) by using MIIS to retrieve necessary information of neighboring network without using the RtSolPr/PrRtadv messages. Especially in [15], ARs control the data forwarding (to MN) with the subscribed triggers of MIH events (MIH Link Up and MIH Link Down). However, additional MIHF operations in handover may increase the system signaling load. Without simulation, it is hard to say that these schemes indeed improve the performance of handover. In [16],a mechanism is proposed to combine SIP, FMIP and MIH. However, the work is limited in networks, and there is no comparable simulation result either The OSM. In next generation wireless networks, the network infrastructure is heterogeneous and all-ip. There are multiple protocols and functional modules to support the handover (see Figure 1). Note that the conventional approaches for improving the handover performance are combined by SIP and MIH, while our proposed handover approach is a combination of SIP, FMIP and MIH. For comparison, we briefly describe the original SIP and MIH combined handover mechanism (OSM), which is a makebeforebreak handover mechanism. The OSM provides the IP mobility between heterogeneous networks as illustrated in the message flow of Figure 2. Recalling that IP mobility is achieved by the REINVITE signaling of SIP, the MN sends the REINVITE signaling to its corresponding node (CN) to reestablish the communicational session with the new IP address. Before the handover process begins, the MN retrieves the prefix of the NAR through the IS in advance. In order to complete handover process before previous link down, the new IP address configuration and the REINVITE signaling of SIP are triggered by MIH s link going down event (LGD) in OSM. After exchange Router Solicitation (RS) and Router Advertisement (RA) signaling, the MN connects to the NAR. 3. Proposed Mechanism In order to achieve seamless handover for IP multimedia subsystem in heterogeneous networks, we propose the FMIPauxiliary SIP and MIH handover mechanism (FASM), which is based upon the architecture of Figure 1. The idea behind the FASM is to introduce the FMIP to the SIP and MIH combination architecture. In [17], a handover decision module (HDM) was proposed to handle the network management, which decides a handover target network. Through the MIH SAP interface, the HDM registers with the local MIHF to become an MIH user. When the link layer event happens, the HDM can obtain the event notification from MIHF. Different from [17], our main concern in FASM is on how to use the cross-layer information to achieve a fast handover, rather than how to select a handover target network any more. Therefore, it is assumed that the link layer handover decision is always valid and the HDM takes charge of choosing the target network Handover Process. In FASM, the fast handover process is achieved by the following three main steps. See also Figure 3 which illustrates the signaling process in FASM. In the first step, the LGD (MIH Link Going Down) event is used to trigger the handover action, while the MIIS is used to tackle the issues related to the discoveries of radio access discovery and candidate AR discovery. The second step is started after the HDM chooses out the target network. In the second step, the FMIP operation is triggered by the LUP (MIH Link Up) event. The operations of HI, HAck, FBack and UNA signaling are used not only for the MN to configure its NCoA in advance but also for the ARs to buffer the packets that are forwarded to NCoA. After the NAR receives the UNA signaling, it can serve the MN immediately. The third step is the MIP Bind Update operation mixed with SIP, including SIP REREGISTER and SIP REINVITE signaling. In FASM, the SIP proxy server and the MIPv6 home agent (HA) are mixed together as an integrated logical entity which is the SIP Server (SS) in Figure Details of Signaling Flows Event Registration. At the early beginning, the HDM registers an interesting MIH Event (i.e., L2 triggers) to the local MIHF. This task can be done by the MIH Event Subscribe.request/response primitives. According to different MIH Event triggers, the HDM will control FMIP and SIP in different ways as follows: LGD will trigger the HDM to turn on the interface to connect the target network; LUP will trigger the HDM to tell the FMIP to send FBU to PAR and begin the other FMIP handover process sequentially; LD (MIH Link Up) will tell the HDM that the make-beforebreak handover is over, and the previous interface can be closed Retrieval of Neighboring Network Information from the IS. In FASM, the functions of RtSolPr/PrRtAdv messages in the standard FMIP are replaced by the MIH Get Information request/response messages, so the RtSolPr/PrRtAdv messages can be deleted in FASM, and thereby the signaling load can be reduced. The MN obtains the network s neighboring information by the MIH Get Information request/response messages, and stores the information

85 4 EURASIP Journal on Wireless Communications and Networking IS SIP proxy server MN PAR NAR CN MIH Get Information request MIH Get Information response LGD Connect to new BS Pre-session New IP RS RA SIP BU Disconnect SIP-REINVITE DATA(2OK) Figure 2: Signaling flows of the OSM. IS SIP proxy server MN PAR NAR CN MIH Get Information request MIH Get Information response LGD Pre-session Turn on interface Connect to new BS LUP FBU HI FBack UNA HAck FBack SIP BU LD Turn off interface SIP-REINVITE DATA(2OK) Figure 3: Signaling flows of the FASM. Dotted line depicts buffering and forwarding packets.

86 EURASIP Journal on Wireless Communications and Networking 5 about the networks in its cache. The MIH Get Information request/response can be done much before the L2 trigger (i.e., MIH Link Going Down), unlike the original FMIP in which the RtSolPr/PrRtAdv only occurs after L2 triggers Network Selection and Switching Link. When the signal strength of Base Station (BS) becomes poor, the HDM will be notified that the current connecting link is going down (i.e., LGD event). Then the HDM chooses the target handover network by using the neighboring network information in the MN s cache, and turns on the corresponding interface. Therefore, the MN can connect to the target network rapidly in the L2 layer. After the L2 connection is completed, the HDM is notified by LUP. The target network information stored in the MN s cache will be used to autoconfigure the NCoA. In the FMIP protocol operation, the FBU is sent to the PAR from the prelink. After sending FBU, the MN waits to receive FBack from the prelink. As soon as the MN receives FBack, it sends UNA to the NAR. UNA can be sent successfully because this operation is done after the LUP trigger. After receiving FBU from the MN, the PAR completes the HI/HAck operation to obtain a valid NCoA, and sends it to the MN via FBack. The proposed mechanism implements a bicasting buffering and forwarding policy in which the PAR buffers and forwards the data packet to MN s PCoA and NCoA simultaneously. Note that a cost function approach to the network selection algorithm providing better performance to the multiinterface terminals in the integrated networks can be found in [18]. WiFi IF1 WiMAX IF2 WiFi IF1 WiMAX IF2 WLAN signal fading LGD LD Weak link quality cause packet loss SIP RE-INVITE WiMAX connect, get IP address SIP BU Interface receives data WLAN signal fading RA/RS Figure 4: Network switching in OSM. LGD FBU Buffer and forward no loss WiMAX connect, get IP address UNA 2OK LD SIP RE-INVITE SIP and MIP Bing Update. After sending UNA to the NAR for announcing its existence, the MN, as an SIP user client, will continue the handover procedure by sending an SIP REINVITE message to the CN. The REINVITE message carries the updated SDP (Session Description Protocol) parameters to the CN. As a result, call parameters are renegotiated on an end-to-end basis. Meanwhile, SIP BU is sent to the MN s SIP server to update the relation between URI and CoA (care of address) as well as the binding of CoA and HA Mechanism Analysis. In OSM, during LGD and 2OK signaling, the link quality of prelink is too poor to receive the packets (see Figure 4). Assume that the probability distribution of data packet loss is P(x), where x is the ratio of the receiving signal power to the BS s sending power of the prelink. During LGD to 2OK, the data packet loss is L loss, which can be determined as follows: R2OK L loss = P(x)dx, (1) R lgd where R 2OK is the ratio when the MN receives the 2OK signaling, and R lgd is the ratio when the MN receives the LGD event. Theaboveweaknesscanbeovercomebyourproposed mechanism (see Figure 5). During FBU and 2OK, the data packets arrived will be buffered and forwarded to both the MN and the NAR simultaneously, and thus the data packet LUP Interface receives data SIP BU 2OK Figure 5: Network switching in FASM. loss L loss is reduced. When the data packets are bicasted, the MN may receive some packets twice. But the duplicate packets can be handled by the higher layer, for example, the duplicate packets can be found out by a sequence number of the RTP in the higher layer. As soon as the PAR receives FBU, it sends HI to the target NAR specified in the FBU. The NAR does the DAD for the NCoA autoconfigured by the MN, and sends the available address to the PAR. The PAR delivers the available NCoA to the MN in the FBack signal. Therefore, in comparison to the OSM scheme, the probability of successfully using NCoA is improved. In Figure 3, as soon as the MN receives the FBack, it sends UNA to the NAR for announcing its existence in the new network. This operation makes the network accessing in the FASM faster than the MIPv6 s RA and RS mechanism which is used in the OSM. Note that the SIP REINVITE will be sent from new-link, so, if the L3 connection time is decreased by the UNA signaling, the total handover latency will be reduced. The NAR also sends the buffered data packet to the MN as soon as it receives the UNA. The service interruption time is the latency when the MN receives the last packet

87 6 EURASIP Journal on Wireless Communications and Networking.4.4 Jitter (ms).2 Jitter (ms) Packet sequence Packet sequence 7 Without error model Figure 6: Jitter without error model. Error model Figure 7: Jitter with error model. from the old link to the first packet from the new link. As compared with the RA/RS mechanism, the MN can receive data packets earlier in the FASM, so the service interruption time canbe reducedasshowninfigure Simulation 4.1. Simulation Design. The NIST seamless and secure mobility software module is used in the NS-2.29 simulator. Note that the NIST software module can support the vertical handover as well as the MIH protocol, but not SIP and FMIP. Hence, the SIP and FMIP modules are implemented in our simulator based on the NIST software module as well as the NIST WiMAX module. For evaluating the performance, we focus on the data packet loss and the service interruption time from CN to MN when the MN hands over between and networks. An error model is applied in the simulation, which expresses a relationship of the data packet loss and link quality.theimpactoftheerrormodelcanbeobservedin FASM in Figures 6 and 7, in which the handover occurs when the time of the MN s receiving the RTP packet sequence is 6 to 7, and the jitter means the time interval of successive received packets. Therefore, if there is no error model, there is still no packet loss when the quality of the previous link is poor. On the contrary, the simulation result with the error model added reveals the relationship of the packet loss and the link quality more practically. A larger packet sequence would lead to poorer previous link quality, and greater jitter would lead to more packet loss. The result in Figure 7 has something to cope with the packet loss probability distribution P(x) which is designed for the simulation program. The simulation topology is illustrated in Figure 8. To evaluate our proposed mechanism, we set up a 2 2 simulation area with a WiMAX BS and a WiFi BS. The WiMAX BS has a power radius of 1 m which covers the WiFi BS that has a power radius of 5 m partly (see Figure 8). The CN connects to the backbone with 1 Mbps data transmission rate. The WiMAX BS and the WiFi BS connect to the backbone also with 1 Mbps. The IS and the SIP proxy servers connect to the backbone with a 1 Mbps data transmission rate. Except that the link delay between BSs and the RT router is 15 ms, the other links delay is 3 ms. The MN is initialized in the BS and moves to the BS area in random at the beginning of the simulation. A RTP application data flow is built between CN and MN, and starts at 5 s and ends at 4 s with a rate of 1 Mbps. Some other parameters also affect the simulation results, such as t21 timeout, which has an effect on the WiMAX L2 handover latency and the maxradelay that impacts the RA/RS delay. Nevertheless, the aim of our simulation is to evaluate the difference between the FASM and the OSM. Therefore, the simulation program is carried out under the same parameters in FASM and OSM Simulation Results. To evaluate our proposed mechanism, the vertical handover processes of FASM and OSM are simulated, respectively. The simulation results focus on the aspects of the received packet jitter, the data packet loss, the service interruption time, as well as buffer size Jitter. In the simulation, the jitter indicates the time interval of two successive packets. As shown in Figure 9, a large jitter is caused by a large packet loss. The FASM scheme shows a remarkable improvement of performance in jitter, as compared to the OSM scheme (see Figures 9 and 1). The improvement is attributed to the FMIP buffering and forwarding mechanism. When the previous link quality is poor, the LGD trigger comes out indicting the beginning of the handover process. Then, the PAR receives FBU and forwards packets to the MN s NCoA in FASM. When the NAR receives UNA, it begins to forward packets to the MN.

88 EURASIP Journal on Wireless Communications and Networking 7 CN SIP server.4 IS Router (RT) Jitter (ms).2 PAR NAR BS BS 67 FASM Packet sequence 7 MN Figure 1: Jitter in FASM with error model. Jitter (ms) Figure 8: The simulation network model. OSM Packet sequence Figure 9: Jitter in OSM with error model. 7 The MN begins to receive packets when the packet sequence number is 684. In comparison, in the OSM scheme the MN receives the packets from the PAR until the SIP 2OK is received. Hence, some packets are lost, and the jitter is larger than that of the FASM scheme. The large jitter between 69 and 7 in Figure 1 is caused by the SIP REINVITE signaling Packet Loss. In Figure 11, the Pr is the ratio of the MN s receiving power of LGD to that of LD which indicates the time interval between LGD and LD. A larger ratio also implies better link quality. Smaller Pr leads to a larger packet loss, because smaller Pr implies that the handover process will begin under poorer link quality and there might be not enough time to complete the FMIP signaling in prelink. While the Pr is greater than or equal to 1.45, the data packet losses of OSM and FASM are almost same. This is because the handover begins when the prelink quality is so good that no packets will be lost. In Figure 11, the data packet loss is reduced from 11 to 41 when Pr is Figure 12 shows the effect of different RTP data rates on the data packet loss. The RTP data rate is varied from.1 Mbps to 3 Mbps. With an increasing RTP data rate, both OSM and FASM suffer an increasing packet loss. However, the OSM experiences more severe packet loss than the FASM, because the FASM employs the FMIP for reducing the packet loss when the handover begins. Figure 13 shows the influence of the movement speed of the MN on the data packet loss. The MN s speed is varied from 1 m/s to 2 m/s. The OSM scheme is severely affected by the increase in speed, whereas the FASM scheme suffers a relatively small change. When increasing the movement speed of the MN, the quality of the link becomes poor more quickly (see also Figure 13, in which the packet loss is increased from 48 to 141 in the OSM scheme). In FASM, the average packet loss is 1. This result is also attributed to the FMIP s buffer function. When the packet is buffered by the NAR, no matter how the movement speed is modified, packets will ultimately be forward to the MN, and thus the packet loss is avoided Service Interruption Time. The influence of the Pr on the handover service interruption time is investigated as follows. The Pr is varied from 1.5 to 1.5. Both FASM and OSM are severely affected by the increase in Pr. Smaller Pr leads to larger service interruption time, because smaller Pr implies that the handover process will begin under poorer

89 8 EURASIP Journal on Wireless Communications and Networking Packet loss 15 1 Packet loss Pr TheMNmovespeed(m/s) 2 OSM FASM Figure 11:PacketlossversusPr. OSM FASM Figure 13: Packet loss versus moving speed of the MN. Packet loss 1 5 Service interruption time (ms) Data transmission rate (Mbps) Pr OSM FASM OSM FASM Figure 12: Packet loss versus data transmission rate of the CN. Figure 14: Service interruption time versus Pr. link quality and there might be not enough time to complete the FMIP signaling in prelink. While the Pr is greater than or equal to 1.45, the service interruption time of OSM and FASM is almost same. This is because the handover begins when the prelink quality is very good. In Figure 14, the service interruption time is reduced from 77 ms to 3 ms when Pr is 1.1. It is obvious that the FASM reduces the service interruption time almost half than the OSM when Pr is smaller than 1.4. The FASM benefits from the FMIP s UNA signaling so that the MN can connect to NAR more quickly. Although the Pr is 1.5, the service interruption time is still less than 1 ms. The phenomenon is caused by not only the make-beforebreak handover mechanism but also by the imperfect of the simulation in NS Buffer Size. The NAR buffers the packets forwarded to NCoA before the MN gets connected to the NAR. The tunnel between PAR and NAR will exist until the MN reinvites CN to send packets to NCoA. The buffer size of the NAR needs to be concerned. Figure 15 shows the relationship of the RTP data rate and the NAR s buffer size. When the RTP data rate is increased, the NAR needs to buffer more packets. 5. Conclusion In this paper, an integrated handover mechanism, called FASM, combined with SIP, FMIP and MIH protocols in IMS (IP Multimedia Subsystem), has been proposed to achieve the seamless handover in heterogeneous networks.

90 EURASIP Journal on Wireless Communications and Networking 9 Buffer size 2 1 PAR: Previous access router PCoA: Previous care of address RTP: Real-time transport protocol RS: Router solicitation SIP: Session initiation protocol UNA: Unsolicited neighbor advertisement URI: Uniform resource identifier. Acknowledgment The work of this paper was partially sponsored by ROC NSC under Grant E-18-2-MY3 and Grant E MY3. FASM 2 Data rate (Mbps) Figure 15: Relationship between the RTP data rate and the NAR s buffer size. In this scheme, FMIP is introduced into the SIP and MIH combination architecture. By using FMIP, the NCoA can be obtained in advance, and data packets are buffered and forwarded to both NCoA and PCoA while the previous link quality is poor. Hence, our scheme can significantly reduce data packet loss as well as service interruption time. Moreover, our simulation results obtained by the NS2 simulator show that the proposed FASM has better handover performance than OSM, for example, the service interruption time is reduced by about 5 percent when the ratio of the receiving power of LGD to that of LD is 1.1. The proposed mechanism has the ability to achieve the handover of seamless end-to-end services in heterogeneous networks. Abbreviations BS: Base station DAD: Duplicate address detection FASM: FMIP-auxiliary SIP and MIH handover mechanism FBU: Fast binding update FMIP: Fast mobile IPv6 protocol HDM: Handover decision module IMS: IP multimedia subsystem IS: Information server LD: Link down event LGD: link going down event LUP: Link up event MIH: Media independent handover MIIS: Media independent information service NAR: New access router NCoA: New care of address OSM: Original SIP and MIH combined handover mechanism RA: Router advertisement 4 6 References [1] L. Zhou, N. Xiong, L. Shu, A. Vasilakos, and S. S. Yeo, Context-aware middleware for multimedia services in heterogeneous networks, IEEE Intelligent Systems, vol. 25, no. 2, pp. 4 47, 21. [2] D.-J. Deng and H.-C. Yen, Quality-of-service provisioning system for multimedia transmission in IEEE wireless LANs, IEEE Journal on Selected Areas in Communications, vol. 23, no. 6, pp , 25. [3] L. Zhou, X. Wang, W. Tu, G.-M. Muutean, and B. Geller, Distributed scheduling scheme for video streaming over multi-channel multi-radio multi-hop wireless networks, IEEE Journal on Selected Areas in Communications, vol. 28, no. 3, pp , 21. [4] L. Zhou, B. Geller, B. Zheng, A. Wei, and J. Cui, System scheduling for multi-description video streaming over wireless multi-hop networks, IEEE Transactions on Broadcasting, vol. 55, no. 4, pp , 29. [5] IEEE P82.21, IEEE Standard for Local and Metropolitan Area Network: Media Independent Handover Services, 29. [6] C.-M.Huang,C.-H.Lee,andP.-H.Tseng, MultihomedSIPbased network mobility using IEEE media independent handover, in Proceedings of IEEE International Conference on Communications (ICC 1), pp , IEEE Press, 21. [7] J.-J. Won, M. Vadapalli, C.-H. Cho, and V. C. M. Leung, Secure media independent handover message transport in heterogeneous networks, EURASIP Journal on Wireless Communications and Networking, vol. 29, Article ID 71648, 15 pages, 29. [8] R. Koodli, Mobile IPv6 Fast Handovers IETF, RFC 5568, 29. [9] D. S. Nursimloo, G. K. Kalebaila, and H. A. Chan, A twolayered mobility architecture using fast mobile IPv6 and session initiation protocol, EURASIP Journal on Wireless Communications and Networking, vol. 28, Article ID , 8 pages, 28. [1] K. Andersson, M. Elkotob, and C. Åhlund, A new MIP-SIP interworking scheme, in Proceedings of the 7th International Conference on Mobile and Ubiquitous Multimedia (MUM 8), pp , ACM Press, December 28. [11] R. Prior and S. Sargento, SIP and MIPv6: cross-layer mobility, in Proceedings of the 12th IEEE International Symposium on Computers and Communications (ISCC 7), pp , IEEE Press, July 27. [12] D. S. Nursimloo, G. K. Kalebaila, and H. A. Chan, ATwo- Layered Mobility Architecture Using Fast Mobile IPv6 and Session Initiation Protocol, Hindawi, New York, NY, USA, 28.

91 1 EURASIP Journal on Wireless Communications and Networking [13] K. N. Choong, V. S. Kesavan, S. L. Ng, F. de Carvalho, A. L. Y. Low, and C. Maciocco, SIP-based IEEE82.21 media independent handover a BT Intel collaboration, BT Technology Journal, vol. 25, no. 2, pp , 27. [14] Q. B. Mussabbir and W. Yao, Optimized FMIPv6 handover using IEEE82.21 MIH services, in Proceedings of the 1st ACM/IEEE International Workshop on Mobility in the Evolving Internet Architecture (MobiArch 6), pp , ACM Press, December 26. [15] M. Boutabia and H. Afifi, MIH-based FMIPv6 optimization for fast-moving mobiles, in Proceedings of the 3rd International Conference on Pervasive Computing and Applications (ICPCA 8), pp , IEEE Press, October 28. [16] H.-H. Huang, J.-S. Wu, and S.-F. Yang, A multiple crosslayers explicit fast handover control using MIH in 82.16e networks, in Proceedings of the 5th IEEE and IFIP International Conference on Wireless and Optical Communications Networks (WOCN 8), pp. 1 5, IEEE Press, May 28. [17] S. Yoo, D. Cypher, and N. Golmie, Timely effective handover mechanism in heterogeneous wireless networks, in Proceedings of the Wireless and Optical Communications Networks (WOCN 8), IEEE Press, 28. [18]K.Hong,S.Lee,L.Kim,andP.Song, Cost-basedvertical handover decision algorithm for WWAN/WLAN integrated networks, EURASIP Journal on Wireless Communications and Networking, vol. 29, Article ID , 11 pages, 29.

92 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID 37541, 2 pages doi:1.1155/21/37541 Research Article A Dynamic Utility Adaptation Framework for Efficient Multimedia Service Support in CDMA Wireless Networks Timotheos Kastrinogiannis and Symeon Papavassiliou Network Management and Optimal Design Laboratory (NETMODE), Institute of Communications and Computer Systems (ICCS), School of Electrical and Computer Engineering, National Technical University of Athens (NTUA), 9 Iroon Polytechniou Street, Zografou , Athens, Greece Correspondence should be addressed to Symeon Papavassiliou, papavass@mail.ntua.gr Received 3 March 21; Revised 14 July 21; Accepted 24 August 21 Academic Editor: Liang Zhou Copyright 21 T. Kastrinogiannis and S. Papavassiliou. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In this paper, the problem of channel-aware opportunistic resource allocation for the downlink in code division multiple access wireless networks supporting simultaneously real-time multimedia and non-real-time data services is addressed. In order to treat different types of services with diverse QoS prerequisites through common optimization formulation a utility-based power and rate allocation framework is adopted. Emphasis is placed on real-time services strict short-term QoS prerequisites, the fulfillment of which requires a significantly more different treatment than the use of static utility functions, traditionally used to address longterm QoS or fairness prerequisites of delay-tolerant data services. To that end, we introduce a novel generic framework that enables the dynamic adaptation of real-time multimedia users utilities as the system evolves, with respect to the corresponding short-term throughput service performance variations. The corresponding nonconvex network utility maximization (NUM) problem is then formulated and solved to obtain optimal downlink power and rate allocation. Via simulation and analysis, it is demonstrated that significant performance improvements are achieved in terms of real-time user s short-term throughput requirement satisfaction, without any considerable loss in the total system throughput. Finally, an essential tradeoff between efficiently fulfilling real-time services short-term QoS prerequisites and maximizing overall system performance, under an opportunistic scheduling wireless environment, is revealed and quantified. 1. Introduction With the growing demand for high data rate and support of multiple services with various quality of service (QoS) requirements, the scheduling policy plays a key role in the efficient resource allocation process in future wireless networks. Moreover, users time and location-dependent channel conditions limit the system s available resources and hence its ability to satisfy their QoS properties. Therefore, a flexible power and rate allocation scheme is essential for optimizing the system s performance. Considerable research efforts have been devoted to the combined problem of power and rate allocation for the downlink of a code division multiple access (CDMA) system (e.g., [1 3]) aiming at the exploitation of multiuser diversity (i.e., users time-varying channel conditions) towards optimizing the system s performance, while satisfying various QoS constraints [4 9]. Moreover, due to the heterogeneity of the wireless environment and the need for the support of diverse QoS requirements, the concept of utilities from the field of economics has been adopted for devising proficient opportunistic resource allocation algorithms. A utility function reflects a users degree of satisfaction with respect to their service performance in a normalized and transparent way, allowing services with assorted QoS prerequisites to be represented, by forming appropriate utilities [1 15], under a common utility-based optimization framework. Hence, Network Utility Maximization (NUM) theory provides the foundations and mathematical tools for setting and treating such problems. In a typical NUM formulation, user s utilities are static, predetermined functions, associated to specific services or

93 2 EURASIP Journal on Wireless Communications and Networking service classes, emphasizing mainly on the support of nonreal-time users long-term requirements. Therefore, users utilities mainly define a continuous relationship between their service performance and their actual achieved throughput (i.e., goodput that reflects the number of reliable bits transmitted) over the wireless opportunistic CDMA paradigm, while considering long-term user s fairness issues [16], minimum performance requirements [17] and/or appropriate constraints imposed by the devices physical limitations [18]. On the other hand, the delay-sensitive nature of real-time multimedia services poses additional demands on accessing the system resourses within short time intervals. Therefore, the adoption of probabilistic short-term delay [19, 2] or throughput [21, 22] constraints have been proposed towards efficiently expressing real-time services QoS prerequisites over a time-varying wireless environment. However, the conventional use of static predefined users utilities does not permit the efficient integration of the latter probabilistic short-term constraints within a NUM problem formulation and thus in the system s resource scheduling policy. In this paper, we study the problem of jointly scheduling multiple services, that is, delay-sensitive, real-time, and delay-tolerant high-throughput non-real-time services, over a heterogeneous CDMA wireless system via NUM optimization. Towards achieving our goal, this paper makes the following contributions. (a) We design real-time users utilities that dynamically adjust with respect to users service short-term QoS satisfaction levels fulfillment, enabling them to efficiently reflect real-time services strict, instantaneous resource demands at the scheduling policy. We refer to the above novel approach as Dynamic Utility Adaptation (DUA) framework. DUA serves as an extension to current NUM theory in order to mainly treat and overcome issues that arise from the use of static utilities, when aiming at introducing users short-term goals or prerequisites under a NUM optimization setting. (b) We adopt and exploit probabilistic short-term throughput constraints, instead of myopic probabilistic delay constraints, in order to introduce the essential requirements of real-time users (requesting multimedia services) in the resource allocation process of a CDMA wireless network. (c) Through the proper use of static and dynamic utilities according to the respective service types, we aim at:(a) meeting various types of user services QoS requirements, namely, real-time and non-real-time, under a common optimization framework and (b) exploiting the benefits emerging by the scheduler s opportunistic character, not only individually per type of users but also in a collaborative manner as well. In this way, a scheduling policy is devised that avoids the problem where the optimization of the performance of users of a specific type of service leads to the corresponding degradation of the performance of other types of services. Thus, several inherent system and users limitations in satisfying services short-term QoS requirements, caused by the corresponding physical hardware constraints, under an interference limited opportunistic wireless environment are highlighted and discussed. (d) Finally, two simple iterative algorithms are proposed. The first one, residing at the base station, attains an asymptotically optimal (in the number of users) power and rate allocation of systems non-convex optimization problem, which is continuously reset at the beginning of each time slot with respect to users utilities adaptation. The second one, residing at the mobile node, dynamically adapts a real-time user s utility by realizing a control loop which: (a) constantly monitors a user s service performance, (b) analyzes its current status with respect to QoS requirements, and (c) reacts to QoS triggering events via the dynamic alteration of the user s utility. It is demonstrated via modeling and simulation that our proposed scheme achieves to the fulfillment of real-time users short-term prerequisites without any considerable loss in the system s total achieved throughput. The obtained results allow to reveal and quantify the inherent tradeoff between efficiently fulfilling real-time services demanding short-term QoS prerequisites and maximizing overall system performance, under an opportunistic wireless environment. The rest of the paper is organized as follows. In Section 2, the system model and definitions are presented. In Section 3, the proposed dynamic users utility adaptation framework is first analyzed, and its application on real-time services is presented. Then, the corresponding utility-based optimization problem is formulated, and its solution is derived. In Section 4, real-time users self-adaptation mechanism in QoS-triggered events is described, and an enhanced power and rate allocation scheme is proposed. Numerical results and relevant discussions are provided in Section5, while Section 6 concludes the paper. 2. System Model and Definitions In this paper, we consider the downlink of a single cell time-slotted CDMA wireless system with N continuously backlogged users at time slot t. A time slot is a fixed interval of time and could consist of one or several packets. Userchannel conditions, which are affectedby shadow fading and long-time scale variations, are assumed to be fixed within the duration of a time slot. The scheduler is assumed to resign at the base station, and hence it can make decisions on users power and rate allocation at the beginning of each time slot. Let us denote by R i (t) the downlink transmission rate at which the base station transmits to user i in the slot under consideration and by R max i the maximum rate at which they can receive data (due to physical hardware limitations). Let us also denote by γ i (t) = E b (t)/i o (t) the bit energy-tointerference density ratio for user i at their mobile device receiver, by G i (t) the path gain from the base station to

94 EURASIP Journal on Wireless Communications and Networking 3 mobile user i, andbyp i (t) the transmission power allocated at a given slot to user i, which, however, is limited by the base station s maximum downlink power P max.thereceivedγ i (t) for each user i is given [16 18]by γ i ( Ri (t), P(t) ) = W G i (t)p i (t) R i (t) θg i (t) N j=1 P j (t) θg i (t)p i (t) + I i (t) = W P i (t) R i (t) θ N j=1 P j (t) θp i (t) + A i (t), where θ denotes the orthogonality factor, W is the system s spreading bandwidth, P(t) denotes the users power allocation vector, I i (t) includes the background noise and intercell interference at user i, G i (t) N j=1 P j (t) G i (t)p i (t) determines the intracell interference at user i and A i (t) = I i (t)/g i (t) denotes the transmission environment between user i and the base station. In our system, we consider two basic types of users, namely, non-real-time users (NRT) requesting delay-tolerant high-throughput services and real-time (RT) users with strict short-term QoS constraints. Throughout the rest of thepaperwedenotebyn NRT (N RT ) the number of nonreal-time users (real-time users) and by S NRT (S RT ) the corresponding set. Due to the variety of the supported services QoS prerequisites, each mobile user is associated with a proper utility function Ui which represents his degree of satisfaction in accordance to his actual expected downlink throughput and can be expressed as ( ) Ui ( ) Ri (t), P(t), a i, b i = R i (t) f i γi (t), a i, b i, (2) i = 1, 2,...,N, where f i represents a function for the probability of a successful packet transmission for user i and is an increasing function of their bit energy to interference ratio γ i (t) at any time slot. A user s function for the probability of a successful packet transmission at fixed data rates depends on the transmission scheme (modulation and coding) being used and can be represented by a sigmoidal-like function of their power allocation for various modulation schemes [18]. Therefore, a user i efficiency function f i has the following properties. (1) (1) f i is an increasing function of γ i (t). (2) f i is a continuous, twice differentiable sigmoidal function with respect to γ i (t). (3) f i () = toensurethatu i = when P i (t)=. (4) f i ( ) = 1. Moreover, we define as a i, b i the two tunable parameters of the sigmoidal function f i that determine function s f i steepness and unique inflection point, respectively [24] (generic definition: f (γ, a, b) = c{1/(1 + e a(γ b) ) d}, where c = (1 + e ab )/e ab and d = 1/(1 + e ab ). Intuitively, since parameter a controls the slope of the sigmoidal function, it determines a user s tolerance in power deviations (in the region of functions f inflection point), while parameter b, controls the relative place of the inflection point of function f (at the access of P), and thus the power level upon which a user s successful packet receive probability increases rapidly (for small deviations of the allocated power), following a concave form [18, 23]). Without loss of generality, we assume that all users have the same value for their parameter a i, (i.e., a i = a for i = 1,..., N). The validity of the above properties has been demonstrated in several practical scenarios with reasonably large packet sizes M (i.e., M 1 bits) [25, 26]. Observing a user s utility as defined in (2), we can point out that the main factors that affect its values are a user s transmission environment (A i ), transmission rate (R i ) and transmission scheme (parameter b i of function f i ). For delay-tolerant non-real-time users, the maximization of their utility corresponds to their desired goodput maximization, and as a result, the corresponding utility is suitable for reflecting their desired throughput maximization at the system s resource allocation optimization problem. On the other hand, real-time users degree of satisfaction does not increase in a linear or concave way along with their throughput maximization (as in the case of NRT users), but according to their fixed data rate expectation fulfillment, as well as their short-term QoS requirements satisfaction due to their delay sensitive nature (e.g., sigmoidal form) Real-Time Services QoS Requirements. A real-time user s requirements consist mainly of a constant downlink rate and short-term delay and throughput guaranties [21, 22]. Therefore, we consider as a real-time user s performance indicator, the achieved probability of receiving an amount of service, in terms of data units, smaller than a predefined threshold within successive short observation time intervals, which is expressed as follows: Pr [ β RT,i (t) B RT,i ]W i t(slot) i S RT, (3) where W i denotes a RT user i observation time interval in terms of slots, B RT,i his predefined data units threshold, and β RT,i (t) the amount of data they received within a specific time interval from slot (t W i + 1) to slot t. The smaller the achieved value of an RT user s short-term throughput probability (3), the greater is their degree of satisfaction. Given a real-time user i requiring downlink rate, R RT,i,we can estimate their data units threshold as B RT,i = R RT,i W i t s, (4) where t s denotes the duration of a time slot. It has been shown in [22] that short-term throughput constraints can more efficiently and comprehensively reflect the essential requirements of RT users (i.e., both delay and throughput expectations) compared to myopic probabilistic delay constraints. This is due to the fact that the adoption of the latter over a time-varying wireless environment may often cause RT users throughput rates dissatisfaction, within either small or long time intervals, due to their potentially bad

95 4 EURASIP Journal on Wireless Communications and Networking channel conditions and variations, leading to their service QoS-aware performance degradation. The previous RT users QoS quarantines, as defined in (3) and(4), are suitable for Constant Bit Rate (CBR) real-time traffic (e.g., video conferencing, telephony (voice services), etc.). To incorporate the QoS prerequisites of other types of real-time services such as real-time Variable Bit Rate (VBR) traffic (e.g., compressed video streams) in the proposed probabilistic short-term throughput framework, the ability to dynamically adjust the requested downlink data rate, R RT,i (t), and thus their data units threshold B RT,i (t), at the corresponding RT user i should be provided. Therefore, in this case, we define B RT,i (t) = R RT,i (t) W i t s. The adopted short-term throughout prerequisites inherent attribute of fulfilling the requested data rate of an RT user within shortterm time intervals, instead of converging to it within longterm intervals (as in [16 18]), allowsthe efficient support of both CBR and VBR traffic. In order to guarantee short-term throughput requirements satisfaction for all RT users (i.e., achieve small values for their probabilities defined in (3)), we aim at providing them with the flexibility of dynamically affecting the priorities of being selected for receiving service according to their corresponding short-term throughput performance, through the introduction of an appropriate user-centric dynamic behavior which drives their ability to dynamically adapt their utility functions, as detailed in the following section. 3. Dynamic Utility Adaptation (DUA) Frame Work Problem Formulation and Solution In this section, we first detail and analyze a novel framework for reflecting users short-term QoS requirements at their utility functions under a NUM problem formulation. This is achieved via the dynamic alteration of the utilities properties in accordance to generic short-term time-varying QoS performance metrics we refer to this framework as Dynamic Utility Adaptation (DUA). Emphasizing on multimedia services and their corresponding QoS prerequisites, a methodology for dynamically adapting RT users utility parameters in accordance to their short-term throughput requirements is examined. Then, the overall utility-based optimization problem is formulated, considering both NRT and RT users performance expectations, and its solution is derived. Finally, following a pure optimization theoretic analysis, the design properties of the proposed DUA framework are examined by determining the way users utilities parameters deviations affect their priority of accessing system resources Dynamic Utility Adaptation Framework. Towards optimizing system s performance, a scheduling policy should allocate wireless network resources, in terms of transmission powers and corresponding rates, in a way that not only maximizes users utilities and hence their degree of satisfaction in each time slot, but also satisfies their QoS prerequisites. The use of fixed predefined utility functions enables the reflection of users long-term performance expectations at the scheduler and is in line with its opportunistic channelaware nature [24]. On the other hand, RT users short-term QoS demands require the scheduler s response within shorttime intervals in the light of short-term QoS violations; therefore, the latter should also be reflected in their utilities. With respect to the previous discussion and analysis, we introduce the dynamic adaptation of RT users utilities Ui (R i (t), P(t), a, b i )foralli S RT by allowing them to properly and dynamically adjust the values of their utility parameter b i for all i S RT. Moreover, we redefine RT users utility function as follows: ( Ri (t), P(t), a, b RT,i (t) ) i S RT, (5) U i where b RT,i (t) denotes a user s utility tunable parameter b i at time slot t andisdefinedas b RT,i (t) = b RT,i + b RT,i (t) i S RT b RT,i (t) [ b min,i (t), b max,i (t) ], where b min,i (t) b RT,i b max,i (t), where b RT,i is RT user i proper parameter in accordance to his transmission scheme (i.e., function s f i initial fixed b i parameter) and b RT,i (t) is the factor that dynamically adjusts parameter s b i overall value in accordance to user s short-term performance. Thus, b RT,i (t) is fixed within the duration of a time-slot. Let us underline that when RT users adjust their b RT,i (t) parameter does not actually select a different modulation scheme, defined only by the fixed part of (6) (i.e.,b RT, i ), but aim at reflecting in the scheduling policy (via their utility function) their expectations in system resources with respect to their current short-term QoS performance and thus, affecting their priority in accessing system s resources. In general, as b RT,i (t) decreases a user s lack of resources is mirrored to his utility and consequently their need for having high priority in accessing system resources is revealed, a desirable property that justifies its selection, as it is shown via the solution of the corresponding utility-based system optimization problem. Parameters b min,i (t) andb max,i (t) are the upper and lower bounds of a RT user s parameter b RT,i (t) in each time slot t, respectively. As it is analyzed later in this paper (Appendix A), the existence of these bounding parameters restricts a user s ability to self-optimize their QoS performance over a time-varying wireless environment, due to the potentially bad channel conditions or lack of available system radio resources Adjusting the Properties of Real-Time Users Utilities. In order real-time users to efficiently adjust their utility parameter b RT,i (t) for all i S RT at the beginning of each time slot t, according to their short-term throughput requirements, the introduction of their actual short-term throughput performance information into the tuning procedure of their utility b RT,i (t) parameter is essential. Therefore, let us define the actual amount of data units that a real-time (6)

96 EURASIP Journal on Wireless Communications and Networking 5 user i received within his observation time interval W i,from slot (t W i + 1) to slot (t 1) as follows: B Wi 1 W i 1 RT,i (t) = β RT,i (t k) i S RT, (7) k=1 where β RT,i (t) = Ui (R i (t), P i (t), a, b RT,i ) t s denotes the actual amount of data that a real-time users i received at time slot t and R i (t),p i (t) denote his corresponding transmission rate and power allocation at the under consideration time slot, respectively. By using the above information and comparing it with a portion of his predefined short-term data units threshold B RT,i, an RT users can adjust their utility b RT,i (t) parameter as b RT,i I RT,i (t) [ b RT,i b min,i (t) ] if B Wi 1 RT,i (t) (Tr +1)B RT,i, b RT,i (t) = b RT,i + G RT,i (t) [ ] b max,i (t) b RT,i if B Wi 1 RT,i (t) > (Tr +1)B RT,i, where I RT,i (t) [, 1] and G RT,i (t) [, 1] for all i S RT are two normalized indicators that reflect a real-time user s need for accessing the system resources at time slot t, when they have shortage or excess of data units received within their current short observation time interval, respectively. Furthermore, parameter Tr (Tr ), referred as the system s triggering parameter, determines the system s degree of preemption. Large values for the system s triggering parameter will make the scheduling policy react in a more preemptive way to real-time users short-term throughput performance deviations, and therefore the achieved probabilities of not satisfying their short-term QoS requirements values will decrease. In accordance to (8), when a real-time user i has received till time slot (t 1) less amount of data units than his predefined threshold, then in order to accomplish his shortterm throughput QoS requirements satisfaction, the value of his utility b RT,i (t) parameter decreases and thus, his probability of being selected at current slot t increases. Furthermore, the reduction of a real-time user s i utility b RT,i (t) parameter from its corresponding value b RT,i is determined by his normalized indicator I RT,i (t) at that time slot. A RT user s I RT,i (t) indicator reflects his need of accessing the system resources, according to the weighted distribution of his received data within his observation time interval, and therefore is defined as W i 1 ŵ RT,i (k, W i )β RT,i (t k) I RT,i (t) = 1 (Tr +1)B k=1 RT,i if B Wi 1 RT,i (t) (Tr +1)B RT,i, where ŵ RT,i (k, W i ) = W i k W i 1 for k = 1,..., W i i S RT. It is noted that ŵ RT,i (k, W i ) represents a weight related to each time slot within the last (W i 1) successive slots of (8) (9) an RT user i observation interval W i that determines the importance of the user i received amount of service at that slot (i.e.,time slot t k) on his estimated indicator. Moreover, the values of an RT user slots weights, as well as the importance of his information, are linearly inversely proportional to his distance k from the current slot t, since we want the information of the most distant slots to play a more important role on the degree of his need in accessing the system s resources. For instance, even if two real-time users i and j have received the same amount of data within the same observation time intervals (i.e., B Wi 1 Wj 1 RT,i (t) = BRT, j (t) when W i = W j, B RT,i = B RT, j, b RT,i = b RT, j and b min,i (t) = b min,j (t)), but user i has received service in slots more recent to the current than user j, then their indicator s I RT, i (t) value will be smaller than user j indicator value I RT, j (t) according to his slots weights, since his tolerance for not accessing the system s resources is greater. Thus, user j utility b RT, j (t) parameter will be smaller than user i corresponding parameter, and therefore they will have higher priority on accessing the system s resources at the current slot. On the other hand, when a real-time user has received a larger amount of data units than the predefined threshold within the last (W i 1) successive time slots of his observation interval W i, then their utility b RT,i (t) parameter increases according to (8), and their priority in being served decreases. The larger RT user i received amount of data within their observation interval W i is, the lower their selection priority should be, and therefore their normalized indicator can be defined as follows: G RT,i (t) = Wi 1 k=1 ŵ RT,i (k, W i )β RT,i (t k) Wi 1 k=1 ŵ RT,i (k, W i )R max i t s if B Wi 1 RT,i (t) > (Tr +1)B RT,i, (1) where the denominant W i 1 k=1 ŵ RT,i (k, W i )R max i t s denotes the maximum weighted amount of data units an RT user i can receive within any time interval of (W i 1) time slots, due to their downlink rate limitation R max i. Such a design attribute allows the reallocation of excess system resources to NRT users towards the desirable optimization of their throughput performance [21, 22]. Concluding this section s analysis, let us underline that the methodology expressed via (8), (9), and (1) applies in the most demanding case where the objective is to minimize RT users probabilistic throughput constraints (i.e., minpr[ β RT,i (t) B RT,i ] Wi for all i S RT ). Moreover, in the special case where an upper bound is set for RT users probabilistic prerequisites, that is, Pr[ β RT,i (t) B RT,i ] Wi q RT,i for all i S RT, then the two normalized parameters I RT,i, and B RT,i,aredefinedas Pr [ ] β RT,i (t) B RT,i I RT,i (t) = 1 q RT,i W i if Pr [ β RT,i (t) B RT,i ]W i q RT,i, (11)

97 6 EURASIP Journal on Wireless Communications and Networking G RT,i (t) = 1 q RT,i Pr [ β RT,i (t) B RT,i ] W i if Pr [ β RT,i (t) B RT,i ]W i >q RT,i, (12) towards reflecting RT user s need for accessing the system resources at time slot t, when they are accomplishing or not the requested bound q RT,i, respectively, under the assumption of the system s feasibility (i.e., there always exists at least one power and rate vector that leads to the satisfaction of all RT users probabilistic throughput constraints) Problem s Formulation, Transformation, and Solution. In order to optimize the overall system performance as well as users degree of satisfaction, the following utility-based power and rate allocation optimization problem must be solved at the scheduler at every time slot max R(t),P(t) s.t. N ( Ui Ri (t), P(t), a, b i (t) ) i=1 N P i (t) P max i=1 P i (t) P max i = 1, 2,..., N R i (t) R max i i = 1, 2,..., N b i (t) = b RT,i (t) i S RT, b i (t) = b NRT,i i S NRT, (13) where b NRT,i denotes NRT users fixed parameter of their f i function in accordance to the used modulation and coding scheme and b RT,i (t) is obtained via (8). Intuitively, (13) aims at jointly fulfilling and optimizing both NRT and RT users QoS-aware degree of satisfaction, via maximizing the actual achieved throughput of the first (i.e., expressed via utility (2)) and via fulfilling the probabilistic short-term throughput prerequisites of the second (i.e., expressed via utility (5) and(6)). In the rest of the paper, for simplicity in the presentation, we omit the notation of the specific slot t in the notations of the system s and users variables that remain fixed within the duration of a time slot. Following the approach in [24], the optimal solution of the above problem is achieved when the base station transmits with its maximum power level P max (i.e., N i=1 P i = P max ), and hence a user i utility defined in (2) or(5) is adjusted according to the following expression: U R i i ( R i (P i ), P i, a, b i ) =U R i i (P i, a, b i ) WP i γi (θp max θp i +A i ) f ( ) i γ i, if Pi Rmax i γi (θp max +A i ) W +θr max i γ ( ) i = γ i, A i R max i =P LIM i ( ( f i γi R max i )), P i, otherwise, (14) where γi = argmax γ 1 {(1/γ) f i (γ)}, Pi LIM (γi, A i ) is the break point of function U R i i (P i, a, b i )andr i = WP i /γi (θp max θp i + A i ) when P i Pi LIM (γi, A i ), or R i = R max i otherwise. For P i Pi LIM (γi, A i ), U R i i is a convex function of P i and for Pi LIM (γi, A i ) < P i P max is a sigmoidal function. Therefore, U R i i is a sigmoidal function of P i at his maximum transmission rate R max i, with inflection point denoted as Pi (specifically, 2 U R i i (P i, a i, b i )/ Pi 2 Pi=P i =, U R i i (P i, a i, b i )/ Pi 2 Pi<Pi > and 2 U R i i (P i, a i, b i )/ Pi 2 Pi>Pi < ). Furthermore, the optimization problem (13) can be transformed to the following: max P s.t. N U R i i=1 i (P i, a, b i ), N P i P max, i=1 P i P max i = 1, 2,..., N. (15) Towards solving the non-convex optimization (15), a pricing-based algorithm was developed in [18], and its asymptotic optimality, when the number of users is large, has been proven. Initially, the scheduler selects users to which nonzero power will be allocated by using the information of their parameters λ max i values. Parameter λ max i represents user i maximum willingness to pay per unit power { } λ max i = min { } λ max U R i i (P i, a, b i ) λp i = P P max. (16) In other words, λ max i is the price λ that maximizes user i net utility P(λ) = argmax P Pmax {U R i i (P, a, b i ) λp} (i.e., the utility minus the cost) and can be calculated as follows: U R i i (P i, a, b i ) P i if U R i i is a sigmoidal like λ max i where P i U R i i U R i i (P max, a, b i ) P max P=P is the unique solution of function and P exists otherwise, (17) U R i i (P (P i, a, b i ) i, a, b i ) P i = for Pi P i P max. P i (18) Moreover, if for any two users i and j U R i i U R j j for P P max, then λ max i λ max j, and therefore user i is more likely to be selected than user j. Hence, the scheduler selects users in a decreasing order of their maximum willingness to pay from 1 to T satisfying j T = max 1 j N ( ) P i λ max j P T (19) i=1

98 EURASIP Journal on Wireless Communications and Networking 7 Finally, for the selected users the base station updates and broadcasts λ max T till finding a unique equilibrium price λ that satisfies T i=1 P i (λ ) = P max [18]. Knowing λ, selected users transmission powers and rates can be easily derived. In accordance to the previous analysis, the price of a users willingness to pay λ max i plays a key role in their selection priority and, moreover, in the portion of total system s resources a user will finally occupy in the subsequent time slot. The following proposition shows that by allowing RT user to adapt their utility properties via adjusting his b i parameter, they gain the enhanced flexibility of controlling the priority of being selected in accessing system resources among the others, towards optimizing their service performance. Proposition 1. If A i b j λ max i >λ max j. Proof. seeappendix B. = A j and R max i = R max j, then if b i < Proposition 1 asserts that if all other conditions are equivalent, a user i with smaller parameter b i has a higher priority in being selected than a user j with larger value for his parameter b j. Moreover, if smaller values of an RT user s expected throughput within their observation time interval are observed, then lower values of his b i parameter will result to higher probability in being selected, and vise versa. Essentially, the above proposition can be generalized for more than two users, revealing not only a relational dependency among users utilities properties and his allocated resources, when the latter are derived through the solution of the system s utility-based optimization problem, defined in (13), but also the validity of the proposed DUA methodology expressed via (8), (9), and (1). 4. Proposed Scheduling Policy Towards Node s QoS-Aware Self-Optimization Nodes QoS-aware self-optimization refers to the ability of sensing his service performance variations as well as his environment changes, and then reacting to QoS triggering events towards optimizing his service performance. Such a behavior is revealed through the solution of the corresponding power and rate allocation optimization problem in CDMA networks when both NRT and RT services require access at system resources. Users requesting real-time services can monitor his services performance, analyze and compute their resource expectations in a normalized way according to their short-term QoS prerequisites, and then adapt their utility functions properties in order to affect their selection priority in the scheduling policy as well as the amount of anticipated resources. Moreover, at the base station, the system scheduler interacts with the mobile nodes towards solving the corresponding optimization problem, as defined in (13). In the rest of this section, we present a Dynamic Utility Adaptation-based Users Power and Rate Allocation (DUA UPRA) scheme, which is realized by the efficient collaboration of two low complexity algorithms residing at the mobile nodes and base station, respectively. From mobile nodes perspective, DUA UPRA introduces a control loop towards enabling their QoS-aware self-optimization, while at the base station, DUA UPRA realizes a flexible algorithm, executed at the end of each time slot, to obtain optimal users power and rate vectors for the subsequent time slot via obtaining the solution of (13). DUA UPRA Scheme At Mobile Nodes [A Control Loop] Step 1 (Information Monitoring). A user computes the actual amount of data units that has received within his current observation timeinterval according to (7). Step 2 (Information Analysis). Determines his need for accessing system resources with respect to his QoS prerequisites (3), inaccordanceto (9)or(1). Step 3 (Decision Making Towards Self-Optimization). Reflects his QoS requirements and resources expectation at the scheduler by adjusting his utility function following (8) and then, disseminates this information at the base station. At Base Station (A Resource Scheduler) Step 1. The scheduler requests users utility functions. Step 2. The non-convex power and rate optimization problem (13) is redefined with respect to the current users utilities (i.e., U RT,i(P, a, b RT,i )foralli,, S RT, and U NRT,j(P, a, b NRT,j )forallj S NRT. Step 3. Users selection is performed for the current optimization problem, according to the mobile selection procedure in (17) (19). Step 4. Users downlink allocated paower and throughput are estimated for non-real-time users from R R i NRT,i = Û R i NRT,i(P i, a, b NRT,i )foralli S NRT and for real-time users from R R i RT,i = Û R i RT,i(P i, a, b RT,i )foralli S RT, according to the power and rate allocation algorithm (PAA) in [18]. Let us underline, that a real-time user i actual downlink power and rate estimation is a function only of b RT,i parameter in Step 4 and hence of his corresponding transmission scheme. In the following, the complexity of DUA UPRA scheme is discussed. We initially place emphasis on DUA UPRA scheme at the mobile node, due to the low computational power of mobile devices. Specifically, the proposed control loop needs to perform the following computations to obtain: (a) a finite summation (7), (b) one normalized real number via (9) or(1) (a summation and a deviation), and then (c) an additional sum in (8). The latter requires the computation of the upper and lower bounds of b RT,i via the algorithms provided in Appendix A. The maximum upperbounded number v of iterations required to obtain the above

99 8 EURASIP Journal on Wireless Communications and Networking bounds is also justified in Appendix A. Apart from the time complexity, due to mobile devise hardware limitations, space considerations are also important. For the implementation of our proposed approach, the mobile device needs to store 2 W i + 1 (i.e., maintained on its memory) real numbers in order DUA UPRA scheme at the mobile node to operate (i.e., W i real numbers for computing its short-term throughput performance (7), W i real numbers for the corresponding slots weights and 1 for maintaining the value of b RT,i ). Finally, concerning DUA UPRA scheme at the base station, we adopt the low complexity algorithms provided in [18] towards obtaining the solution of non-convex optimization problem (14) (a simple shorting and a simple bisection algorithms with overall upper-bounded number of iterations to convergence). 5. Numerical Results and Discussions In this section, the operation and performance of the proposed dynamic utility adaptation-based users power and rate allocation scheme DUA UPRA is evaluated via modeling and simulation. In order to better illustrate the performance and the efficacy of the proposed scheme, in terms of average achieved actual downlink throughput and RT users short-term throughput constraints satisfaction, we compare it against the performance of a fundamental utility-based power and rate allocation scheme [24] (in the following, we refer to it as UPRA algorithm) which only aims at optimizing users actual throughput performance, without considering RT users QoS prerequisites; therefore serving the purpose of system s performance benchmark. Throughout our study, we considered a single cell timeslotted CDMA system. The duration of a slot is assumed to be 1.67 msec and the simulation lasts for 1, slots. We assume that the base station is located at the cell s center and that its maximum transmission power is P max = 1 (Watts). We model the path gain from the base station to user i, G i as G i = K i /di a (Rayleigh channels), where d i is the distance of user i from the base station, a is the distance loss exponent (a = 4), and K i is the log-normal distributed random variable with mean and variance σ 2 = 8 (db), which represents the shadowing [27]. We assumed that the system s spreading bandwidth is W = 1 8 and that the maximum downlink rate for all users is R max i = Kbps. The total number of continuously backlogged users in the system is N = 3, and we considered two types of users, namely, non-real-time users (N NRT )and real-time users (N RT ). Unless otherwise explicitly indicated, in the following, we consider that real-time users require constant downlink rates of R RT, i = 512 kbps for all i S RT (i.e., CBR real-time traffic) while their corresponding observation time intervals were set to W i = 2 slots for all i S RT, and therefore an RT user s short-term data units threshold is set equal to B RT,i = 172 bits. We consider saturated NRT users requesting best effort NRT services, aiming at maximizing the achieved actual downlink throughput. The system s triggering parameter is Tr =.3. Both types of users are assumed to have the same transmission scheme. Therefore, we considered that their f i (γ) functions parameters are a = 2[18] andb NRT,j = b RT,i = 3foralli S RT,forallj S NRT. In order to compute real-time users minimum and maximum values for their parameter b RT,i (t) in each time slot t, according to the algorithms proposed in Appendix A, we considered that ε = 1 5 and L max = 1 5. With the objective of better evaluating the performance of the proposed DUA UPRA scheme, we considered four basic scheduling scenarios. In the first scenario, referred to as SC1, in order to explore our scheme s behavior in terms of satisfying RT and NRT service QoS requirements and to gratify that the proposed dynamic users utilities adaptation framework DUA reflects correctly their corresponding degree of satisfaction, we assumed that all users have the same average channel conditions. In the second and the third scenarios (SC2 and SC3), we evaluate the performance of our proposed scheduler when users with different average channel conditions are served, considering, respectively, only RT users (SC2) and both NRT and RT users (SC3) at the system in order to demonstrate our schemes flexibility in adapting the resource allocation process according not only to users various QoS requirements but also to their average channel conditions, aiming at reducing the drawbacks emerging from the users near-far effect. Finally, the fourth scheduling scenario (SC4) aims at demonstrating and revealing DUA UPRA algorithms efficacy in supporting variablerate real-time traffic users Scheduling Scenario 1 (SC1). Figure 1 illustrates the total system s actual average throughput as a function of the number of RT users in the system (i.e., N RT = 5, 1,..., 3, and therefore RT users percentage in the system ranges from 16.67% to 1%, resp.), while Figure 2 presents RT users probabilities of not satisfying their short-term throughput requirements (i.e., Pr[ β RT,i (t) B RT,i ] Wi ) as a function of their number in the system under UPRA algorithm (black columns) and DUA UPRA scheme (blue columns). All users average channel conditions are similar (i.e., are placed at same distances from the cell s center), therefore, only their fast fading attribute affectstheir instantaneous values. We can clearly observe from Figure 2 that RT users probabilities of not satisfying their short-term throughput requirements are significantly reduced under DUA UPRA scheme, compared to the UPRA algorithm, even for large numbers of RT users in the system. Furthermore, our scheduling scheme s efficacy in satisfying RT users QoS requirements is obtained without any considerable loss in the system s average (per user) achieved throughput, since as shown in Figure 1 system s average achieved throughput under DUA UPRA scheme remains very close to the optimal one achieved by a pure opportunistic utility-based algorithm (UPRA). The observed loss in overall system s average throughput under DUA UPRA is due to RT users slight overprovisioning of available resources towards maintaining their strict short-term throughout prerequisites (i.e., fixed amount of data per short-term windows). On the other hand, the latter resources are allocated to NRT users under UPRA, which are purely opportunistically served and, therefore, obtain increased average actual throughout

100 EURASIP Journal on Wireless Communications and Networking 9 System s average throughput (kbps) RT users short-term throughput failure probabilities (%) UPRA DUA UPRA Number of RT users Figure 1: System s average throughput in SC1. 5 UPRA DUA UPRA Number of RT users Figure 2: RT users short-term throughput failure probabilities in SC1. (leading to better overall system throughput), at the expense of high RT user s short-term throughput failure probabilities (i.e., high RT users performance degradation). The latter tradeoff is revealed in more detail in the following scenarios as well. Moreover, by closely observing the allocation of system resources, in terms of actual average throughput for each one of the considered types of users individually, we can further see our scheme s property of exploiting the opportunistic nature individually for each type of users in order to optimize their diverse QoS requirements. Therefore, Figure 3 illustrates RT (NRT) users actual average throughput as a function of their number in the system under DUA UPRA. Specifically, it can be observed that an RT user s average received throughput remains almost constant, independent of their number in the system, due to DUA UPRA scheme s property of allocating system resources to RT users up to the point where their required streaming throughput is satisfied. It is noted that, as observed in Figure 3, RT users average achieved throughput is slightly larger than NRT and RT users actual average throughput (kbps) RT Users NRT Users Number of RT users Figure 3: RT and NRT user s actual average received throughput under DUA UPRA scheme in SC1. their predefined fixed downlink transmission rate, due to the system s preemptive nature when supporting RT services (determined by the values of triggering parameter Tr). On the other hand, an NRT user s average received throughput increases as the number of NRT users in the system decreases because the degree of competition among them for the excess system resources decreases as well, which is an inherent characteristic of any opportunistic scheduler. With the presentation of the following two figures (Figures 4 and 5), we focus on DUA UPRA scheme s performance under the most demanding case in SC1, in terms of RT users short-term throughput QoS requirements satisfaction, where all the users in the system are RT users (i.e., N RT = N = 3). We aim at demonstrating that DUA UPRA scheme s enhanced performance, with respect to RT services QoS properties, asserts and affects all RT users and not only a portion of them, despite the large fluctuations on their channel conditions (due to fast fading). Specifically in Figure 4, we present each RT user s average actual achieved throughput, for all thirty users in the system, whileinfigure 5 their corresponding short-term throughput failure probabilities under DUA UPRA scheme (black dots) and UPRA algorithm (gray square) are depicted. We observe that all users probabilities of not satisfying their short-term throughput constraints are very small (maximum value:.7% average:.19%) under DUA UPRA, while under UPRA they are high and diverse (maximum value: 2.7% average: 9.8%). Furthermore, under DUA UPRA realtime users average achieved throughput remains very high compared to the one achieved under UPRA that exploits optimally system s throughput abilities without, however, providing short-term throughput constraints. Thus, all realtime users average actual received throughput is almost the same under DUA UPRA Scheduling Scenario 2 (SC2). In the second scheduling scenario SC2, we also considered a system with 3 RT users (i.e., N = N RT = 3), however, separated into

101 1 EURASIP Journal on Wireless Communications and Networking RT users average actual throughput (kbps) UPRA DUA UPRA Users ID Figure 4: RT users average throughput under UPRA and DUA UPRA algorithms in SC1 (when N = N RT = 3). RT users short- term throughput failure probabilities (%) UPRA DUA UPRA Users ID Figure 5: RT users short-term throughput constraints failure probabilities under UPRA and DUA UPRA algorithms in SC1 (when N = N RT = 3). two classes: good users and bad users with good and bad average channel conditions, respectively. Good users average channel conditions are assumed to be 7 db larger than bad users. For each type of users in the system, we evaluate their average probabilities of not satisfying their short-term throughput constraints (illustrated in Figure 6), as well as their throughput performance (presented in Figure 7), as a function of the number of RT bad users in the system, under DUA UPRA (blue columns, solid for bad users and stripes for good users) and UPRA (black columns, solid for bad users and stripes for good users) algorithms. The corresponding results demonstrate that under UPRA algorithm (black solid columns) bad users are strongly unfavored, not only in terms of their short-term throughput constraints dissatisfaction (Figure 6) but also due to their low throughput performance (Figure 7), especially when their number in the system is low. This mainly occurs due to UPRA goal of maximizing the system s total utility. Bad users contribution to the maximization of the system s total utility is very low (i.e., they practically contribute only when their Average per class users short-term throughput failure probabilities (%) UPRA-bad UPRA-good Number of RT bad users DUA UPRA-Bad DUA UPRA-Good Figure 6: Users average short-term throughput failure probabilities in SC2. Mean per class users actual average throughput (kbps) UPRA-bad UPRA-good Number of RT bad users DUA UPRA-Bad DUA UPRA-Good Figure 7: Users average achieved throughput in SC2. channels are very good compared to good users average channel conditions), and therefore they are rarely selected by UPRA algorithm, which leads to their throughput performance degradation. On the other hand, under DUA UPRA scheme bad users short-term throughput performance is highly improved (Figure 7, solid blue columns). Especially, when their number in the system is small, the percentage of their short-term throughput dissatisfaction decreases even 75% compared with the corresponding one achieved under UPRA (solid black columns). Moreover, we observe that bad users average downlink throughput takes the same values independently of their number in the system under DUA UPRA scheme (Figure 7, solid blue columns). Observing good users performance metrics, we notice that their probabilities of not satisfying their shortterm throughput constraints are highly improved under DUA UPRA scheme (Figure 6 striped blue columns (last))

102 EURASIP Journal on Wireless Communications and Networking 11 Base station Real time users Non real time users d = 15 m d = 5 m Case 1 Case 2 Figure 8: Network varying topology (per Case #) under Scenario 3 (SC3). (i.e., always smaller than.18%), especially when their number in the system is high (number of bad users is small). On the other hand, under UPRA algorithm, due to the high competition, their short-term throughput constraints are still not satisfied (Figure 6 striped black columns). Finally, the downlink throughput performance reduction of good RT users under DUA UPRA scheme, when compared to the one achieved under UPRA algorithm, is not only harmless (Figure 7 striped blue and black columns) since good RT users required downlink rate is still achieved and satisfied, but rather desired since the excess system resources can be efficiently allocated to bad RT users in order to improve their short-term throughput requirements, as well as to other NRT users Scheduling Scenario 3 (SC3). With this scenario (SC3), we aim at studying DUA UPRA scheme s ability to efficiently treat near-far effect in a more pragmatic wireless setting, as well as quantifying the tradeoff between RT users shortterm throughput satisfaction fulfillment and system s overall achieved throughout. To that end, we consider N = 3 active users in the system, where five (N RT = 5) request real-time traffic (R RT,i =512 Kbps, W i = 2 slots, B RT,i = 172 bits) for all i S RT, while the rest are considered as NRT users (N NRT = 25). NRT users constantly maintain their position with respect to cell s center, placed in groups of five NRT users in the following distances d = 15 (m) d p+1 = d p + 5 (m) forp =,...,3 (Figure 8). On the other hand, the set of RT users in the system is gradually moving away from cell s center (per case), as shown in Figure 8, inorderto better simulate the fact that RT users experience Rayleigh fast fading channels with various average quality conditions, due to their corresponding distance to cell s base station. Thus, RT users distances from cell s centre per case are provided in Table 1. Figure 9, illustrates overall system downlink average throughput (black columns), as well RT and NRT users average throughput (gray and dotted columns, resp.) for each one of the simulated cases (horizontal axis). Furthermore, the corresponding RT users average short-term throughput failure probabilities are presented in Figure 1. The results show that the proposed DUA UPRA scheme efficiently overcomes the problem of near-far effect, since RT users QoS Table 1: RT users distances in (m) for cell center per Case # under Scenario 3 (SC3). Case # RT User ID Actual average throughput (kbps) Total system Real time users Non real time users Case # Figure 9: Total systems, RT and NRT users average achieved actual throughput in SC3. prerequisites are fulfilled, in terms of achieved average actual throughout larger than 512 kbps and short-term throughput failure probabilities less than.8%, even under the most demanding scenario that is, Case = 5. On the other hand, a significant tradeoff is revealed. As the average channel quality of RT users decreases (i.e., RT users are moving away from cell s center) then, the system increases the number of slots allocated to them, in order to balance between their short-term throughput requirements fulfillment and their inevitable actual throughput degradation (due to their bad channel quality). Therefore, RT users average throughput deceases (Figure 9, grey columns) but their short-term throughput failure probabilities remain very low (Figure 1). At the end, RT users QoS prerequisites are preserved, but, at the cost of low NRT users throughput as well as overall system throughout. That is due to the small number of system slots allocated to NRT users. The latter, leads only to a small increment of NRT users average throughput (as RT users are moving away from cell s center), even if their instantaneous achieved rates are increased, due to their good channel conditions (Figure 9, dotted columns) Scheduling Scenario 4 (SC4). In this final set of simulations (SC4), we explore the service performance of a RT user requesting variable rate traffic under the proposed

103 12 EURASIP Journal on Wireless Communications and Networking RT users short-term throughput failure probabilities (%) Case # Figure 1: RT users average short-term throughput failure probabilities in SC3. RT users short-term throughput failure probabilities Timeslots Figure 12: Short-term throughput failure probability (%) as the system evolves in SC4 for a user with variable rate real-time traffic. RT users actual throughput (kbps) Timeslots Figure 11: Actual achieved throughput as the system evolves in SC4 for a user with variable rate real-time traffic. DUA UPRA scheme, as the system evolves. To that end, we consider a scenario with N = N RT = 3 real-time users, with the same average channel conditions (with fast-fading Rayleigh channels). All users except one user (user j) are assumed to have the same QoS prerequisites (i.e., CBR traffic ofr RT,i = 512 Kbps, W i = 2 slots, B RT,i = 172 bits for all i S RT, i j). User j traffic is considered to be of variable rate as follows: from t = to t = 4 timeslot, R RT, j (t) = 512 (Kbps), from t = 41 to t = 16 timeslot, R RT, j (t) = 768 (Kbps) and from t = 16 to t = 2 timeslot, R RT, j (t) = 512 (Kbps). Figure 11 illustrates the variable rate RT user s instantaneous actual throughput as a function of time, while Figure 12 their corresponding short-term throughput probability (i.e., Pr[ β RT,i (t) B RT,i ] Wi, at timeslot t)asafunction of time, under DUA UPRA scheme. In both figures, the timeslots at which user j required throughput alters are marked with vertical red lines, while the corresponding requested rates are presented with gray horizontal lines in Figure 11. The results show that the dynamic adaptation of the under consideration user s requested actual throughput is fulfilled under DUA UPRA, and thus the timeframe required to complete a new request is less than 5 timeslots (i.e., less than 8.3 msec). Moreover, during the latter transition period (i.e., after a change of the requested throughput), the RT user s short-term throughput failure probabilities slightly increases (maximum value.4 = 4%) and then starts dropping again, due to algorithms adaptation to the new request (Figure 12). Finally, we can observe that the user s actual throughput always remains slightly higher that the required (Figure 11), since in order to always fulfill their short-term throughput requirements (per time slot), the system slightly overprovides them with resources. 6. Concluding Remarks In this paper, we studied the combined problem of allocating system resources, in terms of power assignment and transmission rate, in the downlink of a CDMA wireless system, where multiple services with various QoS requirements are simultaneously requested. We expressed users degree of satisfaction with respect to their QoS demands fulfillment (nonreal-time and real-time services) through a common utilitybased framework which provides us with the enhanced flexibility of effectively influencing the opportunistic scheduler to meet their various QoS prerequisites. Emphasis was placed on RT users probabilistic shortterm throughput requirements satisfaction. Specifically, in order to dynamically and accurately affect their selection priority with respect to their QoS prerequisites satisfaction, we introduced the information of their short-term received data distribution into the proposed methodology of tuning their utility functions properties. Through modeling and simulation under various scheduling scenarios, we demonstrated that significant performance improvements are achieved in terms of real-time user s short-term throughput requirement satisfaction and non-real-time users actual throughput maximization, without any considerable loss in the total system throughput. It should be noted that in this work, we considered linear relationship between users assigned data rates and their corresponding degree of satisfaction. However, the degree of a user s satisfaction with respect to their service quality can be more efficiently expressed by applying other than linear utility functions of their actual throughput rates [11]. The mathematical formulation and the analytical solution of the above utility optimization problem provide a first

104 EURASIP Journal on Wireless Communications and Networking 13 step towards the realization of autonomic wireless network where hybrid data flows are simultaneously supported, and therefore is an issue of a great importance and part of our current research work. Appendices A. Limitations on Controlling Users Selection Priorities In the following, we rigorously define the lower and upper bounds of real-time users b RT,i for all i S RT parameter (i.e., b i,min and b i,max, resp.) and justify their role in the proposed scheduling policy as well as the way they restrict users ability to self-optimize their services QoS performance. It is shown that even if RT users introduced self-optimization behavior enhances their ability to self-optimize their QoSaware performance, mobile nodes potential bad channel conditions, the system s lack of available resources, as well as their physical limitations may prevent the fulfillment of their short-term QoS requirements. Moreover, we provide low complexity algorithms for computing the above boundary values. As analyzed in the previous sections, a real-time user s utility function parameter b RT,i plays a key role in the selection priority of accessing system s resources at each time slot, since it affects the corresponding value of their willingness to pay λ max i. On the other hand, the appropriate values for b min,i and b max,i should be such that for any further decrement or increment in the value of b RT,i (with respect to these bounds) the corresponding value of parameter λ max i is not affected. Consequently, parameter b RT,i should be bounded among them. Therefore, we use the following condition for identifying an RT user i bounds of their b RT,i parameter λ max i ( b RT,i ) =, (A.1) b RT,i b RT,i<b min,i b RT,i>b max,i where λ max i λ max i ( b RT,i ) represents RT users maximum willingness as a function of his parameter b RT,i. Definitions of b min,i and b max,i for all i S RT. Following a pure functions theoretic analysis, the lower bound of an RT user s utility function parameter b RT,i can be formally defined as follows. Proposition 2 (Definition of b min,i for all i S RT ). We defines as b min,i for all i S RT the maximum value of a RT user i utility function parameter b RT,i at slot t, such that b min,i =max b RT,i :P LIM ( ) i γ i, A i brt,i >Pi brt,i where 2 U R i i (P i ) =, P=P i brt,i P i 2 ( ) U R i i P, b RT,i P UR i i ( ) P, b RT,i P > i (γi,a i) brt,i P=P LIM (A.2) where b min,i [, b RT,i ).Thus, when b RT,i < b min,i then λ max i ( b RT,i )/ b RT,i brt,i<b min,i =. Proof. SeeAppendix C. The existence of such a lower bound reveals an inherent users limitation on controlling their services short-term QoS requirements when operating over a time-varying wireless environment; the reason is twofold. On the one hand, due to its opportunistic nature, the scheduler in the sight of low available resources may potentially unfavor some user s towards optimizing overall system s welfare. On the other hand, even when plethora of system resources is available, RT users potential bad instantaneous channel condition and their physical limitations may make the goal of their short-term QoS prerequisites fulfillment unreachable. The following proposition defines an RT user s utility function parameter b RT,i upper bound. Proposition 3 (Definition of b max,i for all i S RT ). One defines as b max,i for all i S RT the maximum value of a realtime user i utility function parameter b RT,i at time slot t such that b max,i = b RT,i :min b RT,i L max, B MAX,i (A i ), brt,i=bmax,i λ max i (A.3) where L max is a large positive number and b RT,i b RT,i. Thus, B MAX,i (A i ) = 1 { ( ) c ln i 1 + awp }. max a 1+(A i /aw(p max + A i )) + c i d i R max i A i (A.4) Proof. SeeAppendix D. From the definition of B MAX,i (A i ) we can observe that the worse RT user s channel conditions are (i.e., parameter A i increases), the smaller is our ability of influencing their selection priority, since the range of their utility function parameter b RT,i decreasesaswell. Algorithms for Computing b min,i and b max,i. We conclude this section by introducing two low complexity algorithms for computing RT users parameters b RT,i lower and upper

105 14 EURASIP Journal on Wireless Communications and Networking bounds at each time slot. Initially, by using Proposition 2, we provide a divide and conquer -based algorithm for computing a real-time user s parameter b RT,i lower bound b min,i. Algorithm for Computing (b min,i for all i S RT ). Let us refer to Pi LIM (γ i, A i ) brt,i(t)=b min,i(t) (ν) > P i (ν) brt,i(t)=b min,i(t) and (P, b RT,i ) P( U R i i [U R i i as conditions A and B, respectively. (P, b RT,i )/ P)] P=P LIM i (γ i,a i) brt,i =b min,i (ν) > (i) Set ν = 1, l (ν) =, r (ν) = b RT,i and b min,i (ν) = b RT,i. (ii) If A AND B are true then l (ν+1) = (l (ν) + r (ν) )/2, r (ν+1) = r (ν) and b min,i (ν+1) = l (ν+1) else l (ν+1) = l (ν), r (ν+1) = ( l (ν) + r (ν) )/2andb min,i (ν+1) = r (ν+1). (iii) If l (ν+1) l (ν) >εor r (ν+1) r (ν) >εthen v = v +1, go to (ii). (iv) If NOT (AANDB) is true then b min,i (ν+1) = b min,i (ν+1) ε. The maximum number of algorithm s iterations ν is v log 2 (b RT,i /ε)+1,whereε is a small positive constant. Finally, with respect to Proposition 3, we propose the adoption of a Steepest Descent -based algorithm for computing at every time slot t, a real-time user i parameter b RT,i upper bound, b max,i. Algorithm for Computing (b max,i for all i S RT ). (i) Let L max be a large positive constant. (n=) (ii) n =, b max,i = brt,i. (n+1) (iii) For n > then b max,i = b (n) max,i +1 D(n) 1 where: D (n) = b RT,i / λ max i brt,i=b. (n) max,i (iv) If b RT,i / λ max i brt,i=b < L (n+1) max,i max, go to (vii) (iv) If b (n+1) max,i >B MAX,i (A i ), then go to (vii) (vi) n = n + 1 and go to (iii). (vii) Stop. The previous algorithm is a modified Steepest Descent - based algorithm, adapted to the needs of our problem (Proposition 3). Specifically, in our case, we are not interested in finding the minimum of the function b RT,i (λ max i ), but in accordance to (A.1), in finding a large value of b at which the absolute values of gradient b RT,i / λ max i brt,i=bmax,i are very large (for practical considerations, we approximate infinite with a large number L max ). Therefore, to improve convergence time, we use as a step the corresponding power of 1 (in step iii). It is easy to show, considering the complexity of the Steepest Descent [28], that the maximum number of iterations required for convergence are ν,wherev (1/2) ln(l max ). Finally, the total maximum number of iterations required to compute the minimum and the maximum values of b RT,i is v log 2 (b RT,i /ε)+1+ (1/2) ln(l max ),whereε and L max are a small and a large positive constants, respectively. B. Proof of Proposition 1 From (14), we can observe that when for two users i, j, A i = A j and R max i = R max j, then the only parameters that are affected by variations in the value of parameters b i, b j, and hence determine the properties of their utility functions U R i i (P, a, b i )andu R j j (P, a, b j )areγ i, γ j and the corresponding values of their f i and f j functions, respectively. Therefore, we first provide the following two lemmas that determine the way that a user i, parameter b i affects his f i function and his γi parameter. Lemma 4. If b i < b j then f i (γ, a, b i ) > f j (γ, a, b j ) for all γ (, ). Proof. Ifonesetsasx i = e abi in the generic definition of a sigmoidal function: ( ( f i γi Ri, P ) ) { } 1, a, b i = c i 1+e a(γi bi) d i, (B.1) where c i = (1 + e abi )/e abi and d i = 1/(1 + e abi ), one can rewrite itas f i (γ, a, b i ) = (1 e aγ )/(1 + x i e aγ ). Thus, if b i < b j then 1 + x i e aγ < 1+x j e aγ and f i (γ, a, b i ) > f j (γ, a, b j )forallγ (, ). Lemma 5. If A i γi <γ j. = A j and R max i = R max j,ifb i < b j, then Proof. From (14), one can compute a user i, parameter γi value as follows: γi = max γ 1 {(1/γ) f i (γ, a, b i )} = max γ 1 {(1/γ)(e aγ 1)/(e aγ + x i )}, wherex i = e abi.letone defines as g(x i, γ) = (1/γ)(e aγ 1)/(e aγ + x i )andash(x i, γ) = g(x i, γ)/ γ = e aγ (aγx i e aγ x i +1+aγ)+x i /(e aγ + x) 2 γ 2. When γ i = γi, then g(x i, γ)/ γ = since γi is maximum and h ( x i, γ ) > γ <γ i, h ( x i, γ ) < γ >γ i. (B.2) Thus, if x i >x j, then h(x i, γ) >h(x j, γ) since h ( x i, γ ) = eaγ aγ(e aγ 1) x i (e aγ + x i ) 3 > a γ 2, γ, b i. (B.3) If there exists x i, γi such that h(x i, γi ) =, and x j, γ j such that h(x j, γ j ) =, then if x j > x i, from (B.3), one has h(x j, γi ) >h(x i, γi )b, and hence h(x j, γi ) >. Thus, since h(x j, γ j ) = andh(x j, γi ) >, from (B.2) one has γi <γ j. Finally, since if x j >x i then γi <γ j,andwhenx j > x i then b i <b j, one concludes that if b i < b j then γi <γ j. Finally, based on Lemmas 4 and 5 one proves Proposition 1. From Remark 1 (Proposition 3,[24]) one has seen that if for any two users i, ju R i i (P, a, b i ) >U R j j (P, a, b j )for P P max, then λ max i >λ max j. Therefore one has to prove that if b i < b j, then U R i i (P, a, b i ) >U R j j (P, a, b j )for P P max.

106 EURASIP Journal on Wireless Communications and Networking 15 By the definition of a user i utility in (14), and Lemma 5, let one defines: ( ) K i = Rmax i γi (θp max + A i ) W + θr max i γi Rmax j γ j θpmax + A j W + θr max j γ = K 2. j (B.4) Furthermore, there are three possible cases for the value of parameter P,when P P max. Case 1. If P K 1 < K 2, then from (14) if b i < b,, then f i (γi, a, b i )/γi f j (γ j, a, b j )/γ j, and hence it easily follows that U R i i (P, a, b i ) U R j j (P, a, b j ). Case 2. If P K 1 and P K 2, then from Lemma 4 if b i <b j we have f i (γ i, a, b i ) f j (γ j, a, b j ), and it follows that U R i i (P, a, b i ) U R j j (P, a, b j ). Case 3. If P K 1 and P K 2, then since R max i f i (γ i (R max i, P), a, b i ) WP/(γi (θp max θp + A i )) f i (γi, a, b i )and(as in Case 1) R max i f i (γ i (R max i, P), a, b i ) WP/(γ j (θp max θp + A j )) f j (γ j, a, b j ), we can conclude that U R i i (P, a, b i ) U R j j (P, a, b j ). Finally, since when b i <b j, U R i i (P, a, b i ) >U R j j (P, a, b j ) for P P max, the proof is completed. C. ProofofProposition 2 In order to determine a lower bound of an RT user s parameter b RT,i, we first identify some of the main properties of his utility function with respect to b RT,i [, b RT,i ). The following lemma defines the relationship between the inflection point of a user s sigmoidal-like utility function and its parameter b RT,i. Lemma 6. For any two values b RT,i, b RT,i, ofuseriutility parameter b RT,i such that b RT,i <b RT,i,itholdsthatP i <Pi where Pi, Pi are the inflection points of their corresponding utilities U R i i (P i, b RT,i) and U R i i (P i, b RT,i),respectively. Proof. SeeAppendix E. Furthermore, we can also prove that when b RT,i =, then a user s utility function inflection point has always smaller value than his utility separation point, Pi LIM (γi, A i ). Lemma 7. When b RT,i =, P i brt,i= <PLIM i (γ i, A i ) brt,i=. Proof. SeeAppendix F. Based on the two previous lemmas, we can see that there always exists a value b RT,i for a user i utility function parameter b RT,i when b RT,i [, b RT,i ), such that for smaller than b RT,i values of b RT,i, the inflection point of function (P i, b RT,i ) is always smaller that its separation point (i.e., (γi, A i )). Therefore, we can provide the following proposition. U R i i Pi <Pi LIM Proposition 8. Therealwaysexistsavalueofareal-timeuseri utility function parameter b RT,i when b RT,i [, b RT,i ),denoted as b RT,i, such that Pi LIM (γi, A i ) brt,i= b = P RT,i i brt,i= b,and RT,i hence Pi LIM (γi, A i ) brt,i< b > P RT,i i brt,i< b when b RT,i RT,i [, b RT,i). Proof. SeeAppendix H. We can now proceed to prove that when b RT,i [, b RT,i), there always exists a value for a user i utility function parameter b RT,i denoted as b min,i such that when b RT,i < b min,i then their maximum willingness to pay is calculated by the second part of (17), because P i in (18) doesnotexist and hence, condition (A.1) is fulfilled. Towards that, we first provide the following proposition. Proposition 9. Therealwaysexistsavalueforareal-time user i utility function parameter b RT,i when b RT,i [, b RT,i), indicated as b RT,i such that when b RT,i b RT,i then there is no [Pi, P max ] such that P i U R i i ( P, b RT,i ) Proof. SeeAppendix I. ( ) P UR i i P, b RT,i = for Pi P P max. P (C.1) According to Proposition 8, there always exists a value for a real-time user i utility function parameter b RT,i (i.e., b RT,i) when b RT,i [, b RT,i ) such that for b RT,i < b RT,i < b RT,i then Pi LIM ( ) γ i, A i brt,i< b >P RT,i i (C.2) brt,i< b RT,i, andinaccordancetoproposition 9, there always exists a value for a real-time user i utility function parameter b RT,i (i.e., b RT,i) when b RT,i [, b RT,i ) such that for b RT,i < b RT,i b RT,i <b RT,i then U R i i (P) P UR i i (P) >, (C.3) P P=Pi LIM (γi,a i) brt,i (t)< b RT,i (t) and hence (18) has no solution. Therefore, there always exist a value for parameter b RT,i when b RT,i [, b RT,i ), denoted as b min,i, such that both (C.2)and(C.3) are satisfied. Specifically, b min,i = b RT,i. Moreover, since when b RT,i [, b RT,i = b min,i ) then, (18) has no solution, and by (17), we have λ max i = U R i i (P max )/P max = (WP max /A i )((1 e (anipmax/ai) )/(1 + e a( b RT,i (N ip max/a i)) )), where N i = W/R max i. Furthermore, since b RT,i N i P max /A i, we conclude that

107 16 EURASIP Journal on Wireless Communications and Networking when b RT,i [, b min,i ), then λ max i = WP max /A i and hence λ max i ( b RT,i )/ b RT,i brt,i<b min,i =. Moreover, the proof is completed. D. Proof of Proposition 3 In the following, we determine the sufficient conditions that will allow us to define an upper bound of an RT user s utility parameter b RT,i when b RT,i b RT,i and to justify its purpose with respect to condition (A.1). Moreover, since b RT,i b RT,i, then without loss of generality in the rest of this subsection, we assume that a user i function f i parameters c i ( b RT,i ) = (1 + e a b RT,i)/e a b RT,i c i and d i ( b RT,i ) = 1/(1 + e a b RT,i) di have fixed values independent from the value of parameter b RT,i. Initially, we determine an upper bound of a real-time user s parameter b RT,i,denotedasB MAX,i (A i )foralli S RT, without considering the satisfaction of condition (A.1). The existence of such an upper bound is based on the base station s total downlink power limitation, as explained in the following lemma. Lemma 1. Therealwaysexistsanupperboundofareal-time user s i S RT utility function parameter b RT,i, when b RT,i b RT,i,denotedasB MAX,i (A i ),where B MAX,i (A i ) = 1 { ( ) c ln i 1 a 1+(A i /aw(p max + A i )) + c i d i + awp } max, R max i A i (D.1) due to the power limitations of the base station (P max ) and the corresponding user s channel conditions per time slot (i.e., A i ). Specifically, if b RT,i >B MAX,i (A i ) then Pi >P max where Pi is the solution of (18). Proof. See Appendix I Moreover, we have already proven that λ max i ( b RT,i ) is a continuous and decreasing function of b RT,i (in Proposition 1) and vice versa. Furthermore, the following lemma states that when b RT,i b RT,i, b RT,i is also a concave up function of λ max i. Lemma 11. When b RT,i b RT,i an RT user s i S RT utility function parameter b RT,i is a concave up function of his parameter λ max i,since 2 b RT,i / (λ max i ) 2 >. Proof. SeeAppendix J. Based on the previous lemmas, we can now formally define an upper bound for a real-time user s parameter b RT,i, with respect to (A.1), as follows. Let L max be a positive constant such as when b RT,i / λ max i λ max i L max, λ max i ( b RT,i )/ b RT,i. Moreover, for a user s parameter b RT,i, which is a function of their maximum willingness brt,i to pay λ max i, we have proven the following properties: (1) b RT,i <B MAX,i (A i )(inlemma 1). (2) For b RT,i < b RT,i B MAX,i (A i ): (a) b RT,i is a continuously decreasing function of parameter λ max i (in Proposition 1). (b) b RT,i is a concave up function of parameter λ max i (in Lemma 11). Therefore, we can conclude that there always exists a b RT,i for all i S RT,denotedasb max,i, such that b max,i = b RT,i :min b RT,i L max,b MAX,i (A i ), brt,i=bmax,i λ max i which completes the proof. E. ProofofLemma 6 If P i U R i i (D.2) is the inflection point of the sigmoid like function (P i, b RT,i ) U R i i (P i, b i ) U R i i (P i ) (for simplicity in the presentation, in this proof, we denote b RT,i as b i ) the following equation must be satisfied 2 U R i i (P i ) P i 2 Pi=P i =. (E.1) Moreover, we can compute the second derivative of a users utility function in accordance to the partial derivatives chain rule as follows, 2 U R i i (P i )/ P 2 i = ( 2 U R i i (γ i )/ γ 2 i )(dγ i /dp i ) 2 +( U R i i (γ i )/ γ i )(d 2 γ i /dp 2 i ) and after mathematical manipulations, we can conclude that 2 U R i i (P i ) 2 = N i(p T + A i ) P i (P T P i + A i ) 3 ( ) ( ) 2 U R i i γi ( ) U R i γi + N γ 2 i +2 i γi, i γ i (E.2) where N i = W/R max i. Furthermore, with respect to (B.1) we can easily derive that

108 EURASIP Journal on Wireless Communications and Networking 17 2 U R i i (P i ) P i 2 = N i(p T + A i ) (P T P i + A i ) 3 ( ) 1+e ab i a i R max i e a(bi γi) e abi ea(bi γi)[ a ( ) ] [ ( ) ] ] γ i + N i +2 a γi + N i 2 ( 1+e a(b ) i γ i) 3. (E.3) Let us further define as F i ( γi (P i ), b i ) = e a(b i γ i) [ a ( γ i + N i ) +2 ] [ a ( γi + N i ) 2 ], (E.4) where F i (γ i (P i ), b i ) Pi=P i = inorder(e.1) tobesatisfied when P i = Pi. Function F i (γ i (P i ), b i ) is an increasing function of variable b i (, ) since ( ) F i γi (P i ), b i = ae a(bi γi)[ a ( ) ] γ i + N i +2 > b i (E.5) b i (, ), and a decreasing function of variable γ i (, ) since ( ) F i γi (P i ), b i = ( a) { e a(bi γi)[ a ( ) ] } γ i + N i 1 +1 < γ i γ i (, ). (E.6) Now, let us consider the case where b i = b i, γ i (Pi ) γ i and with respect to (E.1) and(e.4), F i (γ i (P i ), b i ) Pi=P = since i Pi is the inflection point of user i utility function U R i i (P i, b i ) when b i = b i (i.e., U R i i (P i, b i )). Then, if we increase the value of user i parameter b i from b i to b i,whereb i <b i, then F i ( γi (P i ), b i) Pi=P i >, (E.7) since F i is an increasing function of parameter b i and the value of parameter γ i is fixed (i.e., γ i (Pi ) γ i ). Furthermore, with respect to (E.1) and(e.4), there must also exist a value of parameter Pi regarding the inflection point of function U R i i (P i, b i ), where 2 U R i i (P i, b i )/2Pi 2 Pi=P i =, and hence F i (γ i (P i ), b i ) Pi=P =. (E.8) i According to (E.7) and(e.8), and since F i is a decreasing function of parameter γ i, we can easily conclude that, γ i (Pi ) > γ i (Pi ) when b i < b i. Finally, since γ i is an increasing function of P i we proved that if b i <b i, Pi < Pi. F. Proof of Lemma 7 It can be easily shown that when b RT,i = then γ i = 1 (F.1) Moreover,wecanseefrom(14) that when b RT,i (t) = then Pi LIM ( ) γ i, A i brt,i= = P T + A i N i +1 (F.2) Thus, if 2 U R i i (P i )/ P 2 i P=PT +A i/n i+1 <, then Pi LIM (γi, A i ) brt,i= belongs to the concave part of the sigmoid-like function U R i i (γ(p i ), P i ), and hence Pi brt,i= < Pi LIM (γi, A i ) brt,i=. Towards that and in accordance to (E.3) in Lemma 6, when b RT,i (t) = wehave 2 U R i i (P i ) P i 2 brt,i=, P=P T+A i/n i+1 = 2aRmax i e γia e 2a (N i +1) 3 N 2 i (P T + A i ) 2 (1+e γia ) 3 {e γia[ a ( γ i + N i ) +2 ] [ a ( γi + N i ) 2 ]}. (F.3) In order (F.3) to have negative values for γ 1andα i 1 the following inequality must be asserted, e γia [a(γ i + N i )+ 2] [a(γ i + N i ) 2], which is true. G. ProofofProposition 8 In accordance to Lemma 6 the inflection point Pi of a user i sigmoidal like utility function is an increasing function of his utility parameter b RT,i when b RT,i [, b RT,i ]. Therefore, since when b RT,i =, then Pi LIM (γi, A i ) brt,i= > P i brt,i= according to Lemma 7, and when b RT,i = b RT,i, then Pi LIM (γi, A i ) brt,i= < P i brt,i= according to the definition of a user s utility function, we can conclude that there always exists a value of a real-time user i utility function parameter b RT,i where b RT,i (, b RT,i ), (i.e., b RT,i) such that Pi LIM (γi, A i ) brt,i= b = P RT,i i brt,i= b, and hence when b RT,i RT,i [, b RT,i), then Pi LIM (γi, A i ) brt,i< b >P RT,i i brt,i< b. RT,i H. ProofofProposition 9 Let us define h i (P) = U R i i U Rmax i i ( P, b RT,i ) (P) P UR i i P ( ) P UR i i P, b RT,i P (P). (H.1) Moreover, as it has been proved in [24], Lemma 6, auser i net utility P i (λ) is maximized only when their power allocation value is in the concave part of their utility function

109 18 EURASIP Journal on Wireless Communications and Networking (i.e., P [Pi, P max ]). Therefore, we search for the solution of h i (P) P=P i = only within the range Pi [Pi, P max ], as it further can be observe from (18). Furthermore, since from Proposition 2 when b RT,i [, b RT,i), then P LIM i (γi, A i ) brt,i< b > P RT,i i brt,i< b, the concave part of a RT,i user s utility function is for b RT,i < b RT,i within the range of Pi LIM (γi, A i ) brt,i= b to P RT,i max and if b RT,i < b RT,i, then [ Pi Pi LIM ( ) γ i, A i brt,i< b, P RT,i max ]. (H.2) When P max P Pi LIM (γi, A i ) (+) brt,i b, then h i(p)/ RT,i P = P( 2 U R i i (P)/ P 2 ) because 2 U R i i (P)/ P 2 < forallp > P i therefore is an increasing function of P and Pi LIM (γi, A i ) brt,i b P RT,i i brt,i b. Furthermore, if RT,i we prove that there exists b RT,i [, b RT,i) such that h i (P) P=P LIM i (γi,a i) (+) > when b RT,i [, b RT,i), we will brt,i < b RT,i have conclude the proof of the proposition, since Pi will not exist. After some algebra, we have from (H.1) that h i (P) P=P LIM i (γi,a i) (+) brt,i < b RT,i =R max i Ni +N i e a(b γ i ) N i e aγ i N i e a(b 2γ i ) N i aγi e aiγ i ( N i 1+e a(b γi ) ) 2 N i aγi e a(b γ i ) a ( γi ) 2 e aγ i a ( γi ) 2 e a(b γ i ) ( N i 1+e a(b γi ) ) 2, (H.3) wherewehavedenote b RT,i b for presentation purposes. Since the denominator in (H.3) takes no negative values, we must examine the properties of the numerator. Therefore, let us define the following function Thus, H i (P, b) = N i +N i e a(b γ i ) N i e aγ i N i e a(b 2γ i ) N i aγ i e aγ i N i aγ i e a(b γ i ) a i ( γ i ) 2 e aγ i a ( γi ) 2 e a(b γ i ) (H.4) H i (P, b) b = ( 1 aγi ) Ni ae a(b γ i ) N i ae a(b 2γ i ) ( aγi ) 2 e a(b γ i ) <, H i (P, b) brt,i= γ i =1 = N i + N i ( e 2a 2ae a) 2ae a >. (H.5) By (H.5), we can prove that since the numerator of h i (P) P=P LIM i (γi,a i) (+) is an decreasing function of brt,i < b RT,i b, h i (P) P=P LIM i (γ i,a i) (+) brt,i = of h i (P) P=P LIM i (γi,a i) (+) brt,i < b RT,i > and the denominator is always positive, then if we decrease a real time user s utility function parameter b RT,i from b RT,i to, then there always exists a value for parameter b RT,i, namely, b RT,i, such that h i (P) P=P LIM i (γi,a i) (+) brt,i = b RT,i h i (P) P=P LIM i (γi,a i) (+) brt,i < b RT,i b RT,i there is no Pi [Pi LIM h i (P) P=P LIM i (γi,a i) (+) brt,i < b RT,i I. ProofofLemma 1 = andwhen b RT,i < b RT,i then >, therefore, when b RT,i < >. (γ i, A i ) brt,i< b RT,i, P max] such In accordance to (17) and(18), a real-time user s willingness to pay can be estimated when b RT,i b RT,i as λ max i = UR i i (P) P. (I.1) P=P i Moreover, we have that U R i i (Pi ) Pi λ max i = andaftersome mathematical manipulations, we conclude and finally that e a(γ(p i ) b RT,i(t)) = P i λ max i R max i c i + R max 1, (I.2) i c i d i b RT,i = 1 { ( R max ) i c i ln a Pi λ max i + R max 1 + aγ ( P ) } i. (I.3) i c i d i Moreover, we have that λ max i = UR i i (P) P P=P i = Wa(P max+ A i ) [ R max i c i (1 d i ) Pi λ max ] i ( Pmax Pi ) 2 ( + A i R max) 2 i c i ( P i λ max i +R max ) i c i d i ( Pmax Pi ) 2 ( + A i R max) 2, i c i (I.4) and hence after some algebra with respect to U R i i (Pi ) Pi λ max i =, we can see that A ( Pi ) 2 + B ( Pi ) + C = (I.5) A = k i (λ max i ) 2 + λ max i c i (R max i ) 2, [ k i (1 2d i ) 2R max B = R max i c i λ max i i (P max + A i )], C = (R max i ) 2 c i [λ max i (P max + A i ) 2 c i d i (1 d i )], where k i = Wa i (P max + A i ). In order (I.5) to have a real solution, Pi, when b RT,i b RT,i,andthusPi P max, the following two conditions must be satisfied (without loss of generality and for simplicity in the presentation since b RT,i b RT,i in the following, we set c i = 1, d i = ) :

110 EURASIP Journal on Wireless Communications and Networking 19 P i (1) Δ = (B ) 2 4A C andaftersomealgebrawe can conclude that the following inequality has to be satisfied λ max i aw +4Rmax i A i +4P max. (2) P i P max, and thus, in accordance to (I.5) = B ± Δ 2A [ aw +2R max i = R max i (P max +A i ) ] ± aw [ aw +4R max 2 [ k i λ max i i + ( R max i (I.6) 4λ max ) ] 2. i (P max +A i ) ] (I.7) Furthermore, since Pi > andaw 4R max i 4λ max i (P max + A i ), in order Pi P max the following inequality must be satisfied R max i (P max + A i ) ( aw + R max ) i k i λ max i + ( R max ) 2 P max (I.8) i and after some algebra we conclude that λ max i Rmax i A i R max i + P max awp max (P max + A i ). (I.9) Finally, from (I.3) and(i.9), we can determine the upper bound of a real-time b RT,i b RT,i user s i S RT utility function parameter b RT,i as follows: b RT,i { 1 a ( ln B MAX,i (A i ). c i 1+A i /aw(p max + A i ) + c i d i 1 J. Proof of Lemma 11 + ) + awp } max R max i A i (I.1) In accordance to (I.3) and(i.7) inlemma 1, we have that when b RT,i b RT,i then b RT,i = 1 { ( λ max i aw(p max + A i ) + ( R max ) 2 ) i ln a λ max i (P max + A i ) [ aw +2R max ] 1 i aw + R max } (J.1) i λ max i (P max + A i ) R max i Finally, we can also easily compute 2 b RT,i / (λ max i ) 2 = (P max + A i ){(P max + A i )[2aW +2R max i R max i λ max i ]+(R max i ) 2 }/ [λ max i (P max + A i ) R max i ] 3 >, since W R max i and hence λ max i >R max i /P max + A i (since from (I.9) inlemma 1, we proved that when b RT,i b RT,i then λ max i R max i /P max > /(P max + A i )) which concludes the proof. R max i. Acknowledgment This work has been partially supported by EC FP7 EFIPSANS Project (INFSO-ICT ). References [1] S. Lu, V. Bharghavan, and R. Srikant, Fair scheduling in wireless packet networks, IEEE/ACM Transactions on Networking, vol. 7, no. 4, pp , [2] T. Lee, J. Lin, and Y. T. Su, Downlink power control algorithms for cellular radio systems, IEEE Transactions on Vehicular Technology, vol. 44, no. 1, pp , [3] P. Bender, P. Black, M. Grob, R. Padovani, N. Sindhushayana, and A. Viterbi, CDMA/HDR: a bandwidth-efficient highspeed wireless data service for nomadic users, IEEE Communications Magazine, vol. 38, no. 7, pp. 7 77, 2. [4] S. Borst and P. Whiting, Dynamic rate control algorithms for HDR throughput optimization, in Proceedings of the 2th Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM 1), pp , April 21. [5] M. Andrews, L. Qian, and A. Stolyar, Optimal utility based multi-user throughput allocation subject to throughput constraints, in Proceedings of the Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM 5), pp , March 25. [6] F. Berggren and R. Jäntti, Asymptotically fair transmission scheduling over fading channels, IEEE Transactions on Wireless Communications, vol. 3, no. 1, pp , 24. [7] Y. Liu, S. Gruhl, and E. W. Knightly, WCFQ: an opportunistic wireless scheduler with statistical fairness bounds, IEEE Transactions on Wireless Communications, vol.2,no.5,pp , 23. [8] X. Liu, E. K. P. Chong, and N. B. Shroff, A framework for opportunistic scheduling in wireless networks, Computer Networks, vol. 41, no. 4, pp , 23. [9] J. B. Kim and M. L. Honig, Resource allocation for multiple classes of DS-CDMA traffic, IEEE Transactions on Vehicular Technology, vol. 49, no. 2, pp , 2. [1] F. Meshkati, A. J. Goldsmith, H. Vincent Poor, and S. C. Schwartz, A game-theoretic approach to energy-efficient modulation in CDMA networks with delay constraints, in Proceedings of the IEEE Radio and Wireless Symposium (RWS 7), pp , January 27. [11] T. Harks, Utility proportional fair bandwidth allocation: an optimization oriented approach, in Proceedings of the 3rd International Workshop on QoS in Multiservice IP Networks, vol of Lecture Notes in Computer Science, pp , Springer, Catania, Italy, 25. [12] P. Hande, S. Zhang, and M. Chiang, Distributed rate allocation for inelastic flows, IEEE/ACM Transactions on Networking, vol. 15, no. 6, pp , 27. [13] M. Dianati, X. Shen, and K. Naik, Cooperative fair scheduling for the downlink of CDMA cellular networks, IEEE Transactions on Vehicular Technology, vol. 56, no. 4 I, pp , 27. [14]X.Duan,Z.Niu,andJ.Zheng, Utilityoptimizationand fairness guarantees for multimedia traffic in the downlink of DS-CDMA systems, in Proceedings of the IEEE Global Telecommunications Conference (GLOBECOM 3), pp , December 23.

111 2 EURASIP Journal on Wireless Communications and Networking [15] X. Duan, Z. Niu, and J. Zheng, A dynamic utility-based radio resource management scheme for mobile multimedia DS-CDMA systems, in Proceedings of the of IEEE Global Telecommunications Conference (GLOBECOM 2), Taipei, Taiwan, November 22. [16] J.-W.Lee,R.R.Mazumdar,andN.B.Shroff, Opportunistic power scheduling for dynamic multi-server wireless systems, IEEE Transactions on Wireless Communications, vol.5,no.6, Article ID , pp , 26. [17] J.-W.Lee,R.R.Mazumdar,andN.B.Shroff, Opportunistic power scheduling for multi-server wireless systems with minimum performance constraints, in Proceedings of the 23rd Conference of the IEEE Communications Society (INFOCOM 4), pp , March 24. [18]J.-W.Lee,R.R.Mazumdar,andN.B.Shroff, Downlink power allocation for multi-class wireless systems, IEEE/ACM Transactions on Networking, vol. 13, no. 4, pp , 25. [19] S. Shakkotai and A. Stolyar, Scheduling for multiple flows sharing a time-varying channel: the exponential rule, American Mathematical Society Translations, vol. 27, 22. [2] T. Kastrinogiannis and S. Papavassiliou, Satisfying elastic short term fairness in high throughput wireless communication systems with multimedia services, in Proceedings of the IEEE International Conference on Communications (ICC 7), pp , June 27. [21] T. Kastrinogiannis, S. Papavassiliou, K. Kastrinogiannis, and D. Soulios, A utility-based resource allocation approach for the downlink in CDMA wireless networks with multimedia services, in Proceedings of the 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 7), pp. 1 5, September 27. [22] T. Kastrinogiannis and S. Papavassiliou, Utility based shortterm throughput driven scheduling approach for efficient resource allocation in CDMA wireless networks, Wireless Personal Communications, vol. 52, no. 3, pp , 21. [23]M.Xiao,N.B.Shroff, and E. K. P. Chong, A utility-based power-control scheme in wireless cellular systems, IEEE/ACM Transactions on Networking, vol. 11, no. 2, pp , 23. [24] J.-W. Lee, R. R. Mazumdar, and N. B. Shroff, Joint resource allocation and base-station assignment for the downlink in CDMA networks, IEEE/ACM Transactions on Networking, vol. 14, no. 1, pp. 1 14, 26. [25] F. Meshkati, H. V. Poor, and S. C. Schwartz, A noncooperative power control game in delay-constrained multiple-access networks, in Proceedings of IEEE International Symp. on Information Theory (ISIT 5), Adelaide, Australia, September 25. [26] F. Meshkati, A. J. Goldsmith, H. Vincent Poor, and S. C. Schwartz, A game-theoretic approach to energy-efficient modulation in CDMA networks with delay constraints, in Proceedings of the IEEE Radio and Wireless Symposium (RWS 7), pp , January 27. [27] G. Stuber, Principles of Mobile Communication, KluwerAcademic, Norwell, Mass, USA, [28] J. R. Shewchuck, An Introduction to the Conjugate Gradient Method Without the Agonizing Pain, School of Computer Science, Carnegie Mellon University, Pittsburg, Pa, USA, 1.25 edition, 1994.

112 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 1 pages doi:1.1155/21/ Research Article Adaptive Reliable Routing Based on Cluster Hierarchy for Wireless Multimedia Sensor Networks Kai Lin, 1 Min Chen, 2 and Xiaohu Ge 3 1 School of Computer Science and Technology, Dalian University of Technology, Dalian, Liaoning 11624, China 2 School of Computer Science and Engineering, Seoul National University, Seoul , Republic of Korea 3 Department of Electronics and Information Engineering, Huazhong University of Science and Technology, Wuhan, Hubei 4374, China Correspondence should be addressed to Kai Lin, link@dlut.edu.cn Received 31 March 21; Accepted 7 May 21 Academic Editor: Liang Zhou Copyright 21 Kai Lin et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. As a multimedia information acquisition and processing method, wireless multimedia sensor network(wmsn) has great application potential in military and civilian areas. Compared with traditional wireless sensor network, the routing design of WMSN should obtain more attention on the quality of transmission. This paper proposes an adaptive reliable routing based on clustering hierarchy named ARCH, which includes energy prediction and power allocation mechanism. To obtain a better performance, the cluster structure is formed based on cellular topology. The introduced prediction mechanism makes the sensor nodes predict the remaining energy of other nodes, which dramatically reduces the overall information needed for energy balancing. ARCH can dynamically balance the energy consumption of nodes based on the predicted results provided by power allocation. The simulation results prove the efficiency of the proposed ARCH routing. 1. Introduction With he development of inexpensive multimedia hardware, wireless multimedia sensor networks (WMSN) have recently emerged as an important technology, which is a novel derivative network on the basis of wireless sensor network (WSN). In general, the sensor nodes of WMSN are equipped with CMOS camera, microphone, and other kinds of sensors for achieve the fine-grained, accurate information in a comprehensive environmental monitoring. Compared with traditional wireless sensor network, WMSN can capture the surrounding environment in a variety of media information and has outstanding performance in multimedia signal acquisition and processing. It cannot only enhance existing sensor network applications, but also enable several new applications, such as multimedia surveillance sensor networks, advanced health care delivery, industrial process control, and so on [1 4]. As an energy sensitive noninfrastructure network and the nodes of WMSN are generally distributed in unattended environments to complete the assigned task. Although WMSN is developed from WSN, its energy limitation is even more severe than that of WSN due to the high quantities of data that are included in multimedia content. Different from WSN, the energy consumption in WMSN is not mainly consumed in communication. Sometimes, sensing and processing multimedia data in WMSN may consume more energy than transmitting the same data. Hence, it is not adoptable in WMSN to simply ignore these two kinds of energy consumption like WSN does. Moreover, the multimedia sensor nodes are deployed in sparseness for their strong directives and far-field of view, which results in the big difference of network coverage model between the WMSN and WSN. This difference will also affect the network topology structure and the increased distance results in the energy consumption increased dramatically. Owing to the above reasons, the design of WMSN routing mechanism with high energy efficiency is still very important and face more challenges than WSN. The design of WMSN routing concerns energy constrains, limited computing power, and memory availability of the sensor nodes. By far, the resource consumption is not the only design

113 2 EURASIP Journal on Wireless Communications and Networking target, a certain level of Quality of Service (QoS) is also needed to guarantee delivering multimedia content, such as communication reliability, real-time, and so on. Obviously, it is a trade-off problem that the transmission of multimedia data should meet the requirements of energy efficiency and QoS assurance, which seriously increases the difficulty of routing design. According to the status of sensor nodes in the process of network operation, routing can be divided into flat routing and clustering routing. Clustering routing first appears in cable network, and it is also adaptable for WSN on account of the good flexibility and high communication efficiency. Similar to WSN, the WMSN routing is always built on a hierarchical architecture, comprising several clusters in almost same size where each cluster has several sensor nodes and a cluster head. The obvious advantages of hierarchical architecture in WMSN are as follows. First, for a real WMSN contains hundreds or thousands of multimedia sensor nodes, hierarchical architecture is efficient to divide and rule for the application of distributed computation and QoS management. Second, the sensory data are in high relativity because the sensor nodes are unavoidable to be distributed in redundancy. The unnecessary data transmission can be reduced by data fusion process of cluster head node. Third, most of sensor nodes can turn off radio model to reduce energy consumption and communication conflicts in a quite long period which can significant prolong the lifetime and improve the QoS of the whole network. For these reasons, the clustering routing is much suitable for WMSN than flat routing, specially in large scale network. The crucial guaranteed requirement of Qos is whether the sending data from source data can be effectively received by destination nodes. Since there are distortion, multipath interference, and multitone jamming in wireless channel, the package loss is unavoidable during transmission process. To improve the network performance and meets the application requirement, we make the reliable transmitting of end-toend as QoS requirement in our research. Although it is an easy method to satisfy the reliability requirement by selecting a much reliable route for data gathering, the energy of some nodes will be used up quickly if only the quality of communication is considered. Hence, the other important problem we must concern is how to balance the energy consumption of network. Aiming at the requirements of energy equilibrium and reliability, we propose an adaptive reliable routing based on cluster hierarchy(arch) for WMSN. The ARCH routing can balance the energy consumption with meeting the need of reliability between the source and destination node. Moreover, we design a power allocation mechanism to adjust the transmitting power of nodes and an energy prediction mechanism to realize energy aware among nodes. The main contributions of this paper are summarized as follows. (1) We design a self-adaptive power allocation mechanism, which can make sensor nodes meet the reliability requirement by adjust their transmitting power dynamically. It is very suitable for sensor node to meet the low cost requirement of WMSN, too. (2) We proposed prediction method of energy consumption for WMSN, which can makes the sensor nodes predict the remaining energy of other nodes. The introduced prediction mechanism dramatically reduces the overall information needed for balancing energy consumption. (3) We propose an adaptive reliable routing based on cluster hierarchy(arch) for WMSN. By using power allocation and energy prediction mechanism, the ARCH routing can balance the energy consumption with meeting the need of reliability between the source and destination. (4) We perform extensive simulation experiments to evaluate ARCH by several performance indexes. The results show that ARCH performs high efficiency in wireless multimedia sensor network. The rest of this paper is organized as follows. In Section 2, some related works on the WMSN routing protocol are presented. In Section3, the system models and problem statement are described. In Section4, an adaptive power allocation mechanism is explained. We proposed an energy prediction mechanism for WMSN in Section 5.InSection 6, the ARCH routing is presented. The simulation results are giveninsection 7. Finally, Section 8 concludes our paper. 2. Related Works In wireless multimedia sensor network, the routing design need to guarantee delivering multimedia content with a certain level of Quality of Service (QoS), such as communication reliability, real-time, and so on. Although providing QoS guarantees in WMSN during data gathering is a very challenging problem, some approaches have been proposed in the literature for QoS support. Especially, many researchers have proposed some routing protocols with the ability of energy apperceiving and QoS. These protocols can meet the requirements of real-time and reliability in WMSN. According to the different QoS, they are mainly divided into two parts Reliability Routing Protocol. The typical protocols are AFS [5] andreinform[6]. In AFS, the rank of QoS is introduced and the transmission reliability is guaranteed by self-adaptive retransmitting mechanism. In ReInforM, the transmission reliability and routing load are guaranteed by multirouting and random retransmitting mechanism. ReInforM considers the importance of the data in the packet and can adapt to channel errors. ReInforM can send multiple copies of a packet along multiple paths from the source to the sink so that the data can be delivered with the desired reliability. ReInforM uses the concept of dynamic packet state in the context of sensor networks to control the number of paths required for the desired reliability based on local knowledge of the channel error rate and topology. However, this protocol only addresses QoS in terms of reliability, disregarding energy issues. In addition, this protocol does not consider route delays when selecting

114 EURASIP Journal on Wireless Communications and Networking 3 multiple paths. In 26, Felemban et al. propose MMSPEED routing protocol [7]. MMSPEED considers the needs of real-time and reliability, adopting the design conception of MAC layer and network span. By localization algorithm and multirouting mechanism, MMSPEED has good properties of QoS and expendability. MMSPEED can well support flow media and meet the needs of graphic and video to real-time and reliability in WMSN. The drawback of MMSPEED is complex algorithm, large energy consumption, which limits its wide application in WMSN. The QoS routing approach presented by utilizing the geographic location of sensor nodesaswell.thisprotocolassignsanurgencyfactorto every packet depending on the remaining distance and the time left to deliver the packet. It determines the distance required for the packet to be sent closer to the destination in order to meet its deadline. Each node assigns a priority to all of its neighbors, according to their residual energy and delay, as well as the priority of the packets, and packets are forwarded to the highest priority nodes. Packets are sorted in two different queues, one for nonrealtime traffic, and the other one for real-time traffic. Real-time traffic is prioritized based on its urgency factor, scheduling those packets with more aggressive deadlines first for transmission. Reliability is achieved by using duplication of information at the source node. However, the protocol does not consider data aggregation and the network lacks a good decongestion scheme Real Time Routing Protocol. The typical ones are SAR [8], RAP [9], and SPEED [1]. Here, SAR is a kind of routing protocol with energy aware of QoS. Sensor nodes can send the information met the needs of tree to the sink node based on the path source, additional QoS measure, and package priority rank. RAP uses a velocity monotonic scheduler to prioritize packets and schedules them on the basis of their required transmission speed. This protocol does not consider energy issues and the number of hops executed by the packets. SPEED is the first real-time routing protocol for WMSN. It introduces a soft realtime end-toend to support all nonrealtime MAC protocols, providing the management of controlling network traffic and multirouting overload. A Weighted Fair Queuing (WFQ) approach is used in every node to provide the required share of bandwidth for both traffic classes. Path generation is performed in a centralized manner, at the base station using an extended version of Dijkstra s Algorithm. The advantage of this algorithm lies in the fact that it provides a guarantee for besteffort transmission, while simultaneously trying to maximize real-time traffic throughput. The main drawback is that the algorithm requires complete knowledge of the network topology at the base station to calculate multiple routes, thereby limiting the scalability of this approach. An energy aware QoS routing protocol for real-time traffic generated by a wireless sensor network consisting of image sensors is proposed by Akkaya and Younis in [11]. This approach finds multiple network routes by using a minimum path cost. Such kind of cost is a function of distance between nodes, node residual energy, energy transmission, and error rates which meet the requested end-to-end delay constraints. All traffic is divided into best effort and real-time classes. 3. System Models and Problem Statement 3.1. System Models Network Model. In this paper, we adopts a WMSN formed by n random deployed multimedia sensor nodes, m gateway nodes, and one sink node. All the sensor nodes are used for data collection in the monitoring area and do not move after the deployment. The network architecture is depicted in Figure 1. The sensor nodes are grouped into clusters based on many criteria such as communication range, number and type of sensors and geographical location. Each cluster has a gateway node that manages the sensor nodes in the cluster, which are significantly less energy-constrained than sensor nodes. The gateway node will take charge of sensor organization and network management based on the QoS requirement and available energy in each sensor node. All the sensor nodes are isomorphic with the same initial energy and the same capacity of sensing, computation and communication. The sink node is not limited by energy and capacity. Each multimedia sensor node can adjust the transmission power to save energy consumption and the links are symmetrical. If the receiver knows the transmission power, then the receiver can calculate the distance to the transmitter by the intensity of the received signal Transmission Error Model. In our research, we assume that path loss is close to log-normal shadow model [12], which is G(d)[dB] = G(d ) + ηlog 1 ( d d ) + X σ, (1) where G d is path loss, d represents the distance between transmitter and receiver, d is the reference range, η is the coefficient of path loss. X σ is distributed in (, σ 2 ) following Gaussian random process. The relationship between Packet reception rate and SNR(signal to noise ratio) is as follows prr = 1 Q 2γ B N R 8ρF, (2) where Q(x) = x (1/ 2π)exp( x 2 /2)dx, γ represents SNR, R is the noise bandwidth, ρ is data rate, F is code rate, and F is data frame length Problem Statement. Now we begin to formulate the problem. As shown in Figure 1, the network consists of a series of multimedia sensor nodes, one sink node, and some gateway nodes. These gateway nodes act as cluster head nodes, which need to manage and collect the data sent from the nodes in their clusters. Multimedia sensor nodes complete monitoring task and send their data to gateway nodes. In each cluster, the sensor nodes are the source nodes and the gateway node is the destination node. Due to communication capacity limitation, most of the senor nodes need to send their data by multihop method to the gateway node. There is at least one routing existed to collect data between each sensor node and gateway node in one cluster.

115 4 EURASIP Journal on Wireless Communications and Networking Sink node Gateway (class head) Multi-media sensor node Figure 1: Network architecture. R H 12 H 23 H 34 H 45 H 56 Figure 2: Reliable requirement for multihop path. Although the transmission of multimedia data does not require 1% reliability, it is still necessary to guarantee the reliability of end-to-end. As the energy of sensor node and bandwidth are both limited, it is difficult to transmit large data of streaming media and needs more investigation to design a novel routing to realize the reliable transmission in network. We assume that there is one multihop path as shown in Figure 2. The reliability between random neighbor node i and j is H ij, the reliability of the requirement from node 1 to 6 is R. Obviously, the data transmission should meet the needs as follows: H ij r ij, Hij R, 1 i 5, i +1= j. (3) As the above description, it has to consider the energy equilibrium of nodes besides of the reliability for the sake of avoiding the energy hole resulted from some nodes running out of their energy too fast. Based on the above two points, our optimization object is to design a routing protocol that can guarantee the reliability of data transmission and balance the energy consumption while delivering data from all source nodes in S to the sink node. This problem can be formulated as follows min ( ) 2 Eu E u S (4) s.t. for u S, p ru = Q, where E u represents the remaining energy of node u and E represents the average remaining energy of all nodes. p ru represents the probability of data from node u that can be correctly received. Q represents the reliable requirement of data transmission. The constraint specifies that it should guarantee the ensuring end-to-end reliability from each source node to sink node. 4. Self-Adaptive Power Allocation Mechanism Most transceiver chip supports programmable transmit power. The transmit power level can be adjusted by configure the corresponding status register. Take Cyclops based on mica2 platform for an example, the power range of communication module CC242 is in [ 25 dbm, dbm], including eight levels. The energy consumption of transmitting data is E tx( P tx) = 8 f R ( P cir + Ptx ηp tx ), (5) where P cir is the circle power and η is the conversion efficiency of power amplifier. The energy consumption of receiving data is E rx = 8 f R Prx, (6) where, P rx represents the receive power.

116 EURASIP Journal on Wireless Communications and Networking 5 Energy consumption (mj) Transmit power (mw) Transmit data Receive data Figure 3: Energy consumption with different transmit power. To have an intuitive expression on energy consumption of data transmission, let f = 128 bytes, P cir = 26.5mW, η(p tx ) =.6e.95Ptx (dbm), P rx = 28 mw, R = 256 kbps. The relationship between transmit power and energy consumption is shown in Figure 3. As mentioned above, there must be at least one route between each source node and the sink node to complete the data gathering. We use R(u, p, h) for representing the route p from node u to the sink node with h hops. For the purpose of energy conservation and reducing communication conflict, we also assume that the same data packet using only one route to send. Considering the average rate of data reception by single hop transmission, we use G i and L i for representing the power gain and transmit power level on the hop i of route p, respectively. The reliable transmission rate is p i (G i, L i ), which is related to G i and L i. G i is a Gaussian distributed random variable. When G i is independent and identically distributed, it is necessary for the reliable transmission rate to meet the transfer request from node u to the sink node by the route p is P ( u, p, h ) h = p i (G i, L i ) R, (7) i=1 wherer is the lowest reliable requirement. SNR increases with the increasing of transmission power, which results in the improvement of transmission reliability. In (7), P(u, p, h) is determined by each p i (G i, L i )ofroutep. When we use a higher L i,ahigherp i (G i, L i ) will be obtained. The increments of reliable transmission rate by increasing one level transmit power is defined in: ΔRi = P rr(g i, (L i +1)), L i [1, 2, 3,..., L]. (8) P(G i, L i ) It can be known from Figure 3 that the increasing of transmit power level will result in more energy consumption. In order to save more energy of the network, we use the critical condition of (7) and introduce the power control algorithm presented by Kwon H [13]. The reliability is guaranteed by gradually increasing the transmit power level. If all the nodes adopt the maximum transmit power level and still cannot meet the reliable transmission requirement, it has to rebuild another route. According to (5), our optimization target is to guarantee the reliability of end-to-end and balance the energy consumption of nodes in the network. To achieve the optimization target, we can allocate higher transmit power level to the multimedia sensor nodes with more remaining energy. Obviously, we must solve the problem that how to realize energy aware between nodes in priority. 5. Energy Prediction Mechanism In order to achieve the energy equilibrium, we first need to know the remaining energy of each multimedia sensor node. However, the energy aware among nodes is difficult to be achieved in WMSN, which is due to the high communication cost for constituent updating their remaining energy information. At the same time, some problems are also brought, such as network congestion and transmission delay. To solve this problem, we propose an energy prediction mechanism for WMSN, which can make sensor node know the remaining energy of other nodes without constituent updating. During the operation process of WMSN, the energy consumption of multimedia sensor nodes is not stable but depends on the their working states. Energy consumption of sensor node depends on the different working states. Nodes can turn to sleep when they have no any task. The working state conversion is necessary to save energy for WSMN, but it also increase the difficulty for predicting energy consumption. In traditional sensor node, the energy is mainly consumed on receiving and transmitting process, where the energy consumption on data sensing and processing can be neglected. However, the total energy consumptions of multimedia sensor nodes in WMSN increase greatly as the nodes need to collect the data from audio, video, and graphic. It consumes much more energy of sensing and processing than those of communication. According to the operation of multimedia sensor nodes under the clustering hierarchy in WMSN, we design a state conversion model for intracluster nodes as shown in Figure 4. There are seven working states in this model. (1) In sleep state, the sensor and communication module are close, the sensor nodes have no any task. (2) In sense state, the sensor module is close and the communication module is open, the sensor nodes collect the multimedia. (3) In idle state, the sensor module is close and the communication module is open, the sensor nodes monitor communication channel. (4) In receive state, the sensor module is close and the communication module is open, the sensor nodes receive data from other nodes.

117 6 EURASIP Journal on Wireless Communications and Networking Transmit Sense On the basis of working states statistic, the nodes can calculate the energy consumption of itself or other nodes in the next T time-step by (11) [14]. The accuracy of energy prediction is influenced by the validity of the probability diversion matrix. Process Sleep Idle Access Receive Figure 4: Model of working state conversion. (5) In transmit state, the sensor module is close and the communication module is open, the sensor nodes transmit data to other nodes. (6) In process state, the sensor and communication module are close, the sensor nodes process multimedia data. (7) In access state, the sensor and communication are close, the sensor nodes complete reading and writing memory. All the working state conversion occurs only at the end of one time-step. It is worthy mentioning that in transmit state, the energy consumption of node is different for transmitting data by various power level. Based on this state conversion model, we can realize the energy aware among nodes. Similar to [14], we use Markov chain to simulate the working states of multimedia sensor nodes. Each node has a series of random variants X, X 1, X 2,..., which describes different working states. P ij is further defined as one-step diversion probability, which can be expressed by P ij = P { X m+1 = j X m = i }. (9) The N-step diversion probability is defined in: M Pij(n) = P ik (r)p kj (n r), for <r<n. (1) k=1 If the node is currently in state i, the number of timeslot that it will stay in state s in the next T time-step can be expressed as T t=1 P is (t). We use E s for representing the energy consumption of node staying at state s in one timestep. Then, the total energy consumption in the next T timestep can be calculated by M T E T = P is (t) Es. (11) s=1 t=1 6. ARCH Routing In this section, we propose an adaptive reliable routing based on cluster hierarchy(arch) for WMSN. The above mentioned self-adaptive power allocation and energy prediction mechanism are both used in ARCH. To obtain a better performance, the cluster structure is formed based on cellular topology. The design objective of ARCH is to guarantee energy balance and meet the needs of reliability between the source and destination Establishment of Routing. With the existence of gateway node, we can easily establish the cluster structure for purpose. Here, the cluster structure is generated by cellular topology. In this structure, the monitoring area is divided into cellular virtual unit cells with same size. At the center of each unit cell, a gateway node plays as cluster head node and the rest ordinary nodes as member nodes belong to the unit cell. In the initialization phase, each gateway node sends one ADV message including the ID of this node and the multimedia sensor nodes receive these ADV messages. In general, each sensor node can receive more than one ADV message. At this time, the sensor node needs to compare the signal strength of the received messages and select the gateway node with strong signal to join in its cluster. If the gateway node is located at the right position, we can establish an ideal cluster structure based on cellular topology by this way. Considering low cost in network, the amount of gateway node should be as few as possible. We adopt a intracluster multihop communication method. To guarantee all the sensory data can be successively transmitted to the sink node, at least one route is necessary to be established for each source node to reach its cluster head node. According to the communication ability of sensor nodes, each cluster is divided into many concentric coronas with the center of gateway, denoted as C 1, C 2,..., C N. For avoiding the loss of efficiency data, we set the width of coronas equaling to the communication distance by using lowest transmit power level. From corona C 1, the nodes in corona C i 1 send ADV messages with the node ID to corona C i. By these ADV message, each node in corona C i (2iN) can find a relay node from the corona C i 1. All the nodes in the corona C 1 can communicate with the gateway node directly. Figure 5 shows the intracluster multihop routing. When each node finds its relay node, the establishment of routing in cluster is finished. Gateway node is responsible for gathering and processing all the sensory data, then send these data to the sink node. If the network scale is small, each gateway node can communicate with the sink node directly. If the network scale is large, even though gateway nodes have stronger ability than those of ordinary sensor nodes, they

118 EURASIP Journal on Wireless Communications and Networking 7 A B C D E F G H O Sink+ Figure 6: Inter-cluster multihop routing. Figure 5: Intra-cluster multihop routing. 1 still need using multihop method to transmit their data to the sink node. We can form robust inter-cluster multihop routings according to the property of cellular structure. As shown in Figure 6, there are six gateway nodes with the same distance to the center node O (sink node). They can send data to the sink node directly. The external data can be transmitted by more than one pathway to guarantee the success delivery to the sink node. For example, the gateway node A can send data to O by different pathways: A C F O, A D G O, A B D G O, or A C D G O. Whena pathway is broken, it is convenient to find another pathway as alternative. Additionally, some important data can be sent by more than one pathway to guarantee its successful transmission to the destination node Time-Slot Assignment. To avoid the conflicts during data transmission, the gateway nodes need to set up a table of time division multiple access (TDMA) to distribute the time-slot for sensor nodes belonging to its cluster. According to the TDMA table, the member sensor nodes can turn off their radio module in the nontransmission period to reduce their energy consumption. When the intracluster single-hop communication method is adopted, the TDMA table is a simple one variable linear table and each member sensor node can be distributed an isometric time-slot. But for ARCH with intracluster multihop communication method, this kind of distribution is not suitable any longer. Here, the cluster head node will not distribute time-slot for all the member sensor nodes, but only for the member sensor node with one hop. If a member node does not contain any son node, its time-slot is distributed as 1, then notify to the upper sensor node. If the member node contains only one son sensor node, they 1 A D C B Gateway Sensor node 5 1 E Figure 7: Intra-cluster time-slot distribution of multihop. will set the time-slot distribution as same as the son sensor node. If the member sensor node contains more than one son node, the time-slot of this member node is distributed as sum of all its son nodes. Figure 7 shows an example of intracluster the time-slot distribution. For node A, B,andD do not have any son nodes, they are assigned one time-slot for transmitting their data. Node C serves as a relay node for node A and B, so it is assigned two time-slots. Node E serves as a relay node of node C and D,so it is assigned three time-slots. In this example, the gateways node informs each node in its cluster of the time-slots it is going to receive packets from other nodes and the time-slots it can use to transmit the packets. G

119 8 EURASIP Journal on Wireless Communications and Networking 6.3. Realization of Optimization Target. In order to reduce energy consumption, transmit power level L i should be set as 1 for all multimedia sensor nodes. Then, the intracluster routing starts to be established. As mentioned above, all the nodes in corona C i select their relay nodes in corona C i 1. If the routing is failed to be established, the L i of nodes in this transmitting direction will be set as 2. If it is still not successful until L i of these nodes equal to L, the process of routing establishment will stop. For an established route, it needs to determine whether it is satisfy for (7). If the route does not meet the condition, the power allocation algorithm will be started. In network initialization phase, all sensor nodes have the same remaining energy. We can randomly select routing nodes and increase their transmit power level until met the requirement. To achieve energy aware, According to (12), each node calculate its own energy consumption rate as ΔE = E T /T, then notify it to other nodes on the route. Each node can predict the remaining energy of other nodes by ΔE, avoiding a large amount information exchange. To eliminate the unavoidable predicting error, each node should keep its recent ΔE and recalculate it when error occurs, then notify it with its current remaining energy to the other nodes on the route. To reduce updating number, we set an error threshold F. Only when the predicting error is higher than the value threshold, sensor nodes need to resend the message. In order to balance the network energy consumption, we dynamically adjust the transmit power level of sensor nodes based on their remaining energy. We use S p for representing the node set on route p. Foranynodei belonging to S p,ifit met the following (12), the transmit power level of all nodes in S p will be reallocated. i S p ( ) 2 Ei 1 >D, (12) E i where D is a configurable threshold. Obviously, increasing energy consumption of the node with more remaining energy can meet our optimization goal. Under the premise of meeting (7), the nodes with higher remaining energy will be allocated higher transmit power level, while the nodes with low remaining energy just need to adopt transmit power level which can satisfy the requirement of single-hop reliable transmission. 7. Simulation and Analysis In this section, we evaluate the performance of the ARCH via simulation experiments. We assume that 4 multimedia sensor nodes and 8 gateway nodes are uniformly deployed into a circle with diameter of 2 m, where a sink node is at the center of circle. The initial energy of each senor node has 5 J. All the sensor nodes are the source of information and can be the relay nodes. The original packet is 128 bytes generated by each sensor node. Referred to the index value of CC242 chip, configure eight transmit power level: 25 dbm, 15 dbm, 1 dbm, 7dBm, 5dBm, 3dBm, 1dBm,anddBm. Energy (J) Remaining energy ratio Time(s) Predicted value Actual value Figure 8: Predicted value and actual value Number of transmit power level ARCH ARCH without EP ARCH without PA Figure 9: Remaining energy ratio with different number of transmit power level. Figure 8 records the node s remaining energy and the value of predicted result in 1 s by ARCH. It can be seen that the value of predicted result is very close to the actual value. There are only three times that the error exceed when the threshold is set at 3%. When the error exceeds the threshold, the nodes will recalculate the parameter of energy prediction and make the current actual value as the initial value for the next prediction. Figure 9 shows the normalized remaining energy with different number of transmit power level during 1 s. It can be seen that the normalized remaining energy by ARCH increased obviously with increasing the number of transmit power level. If the energy prediction or power allocation mechanism are not adopted (ARCH without EP or PA), the

120 EURASIP Journal on Wireless Communications and Networking 9 Average remaining energy ratio Reliability requirement ARCH ARCH without EP ARCH without PA Figure 1: Remaining energy ratio with reliability requirement. Remaining energy ratio Number of gateway node ARCH ARCH without EP ARCH without PA Figure 11: Remaining energy with different number of gateway node. remaining energy of nodes decreased obviously. Specially, the energy consumption of network is slightly influenced by number of transmit power level for ARCH without PA. Figure 1 shows that the increasing reliability requirement will lead to declining remaining energy. Although a higher transmit power level will increase energy consumption, the reliability of end-to-end will be effectively enhanced by the power allocation. The increased energy consumption for a high reliability is far less than that of retransmission energy with a low reliability. Therefore, the remaining energy of ARCH is much more than that of ARCH without PA. Figure 11 shows that the normalized remaining energy with different number of gateway nodes during 1 s. It can be seen that the remaining energy of network increase with increasing the number of gateway nodes. The increased number of gateway nodes decreases the area of cluster, hence the intracluster transmission hop is also reduced, which is benefit for saving energy. 8. Conclusion As a resource-constrained network, wireless multimedia sensor network should try to reduce the unnecessary energy consumption. In this paper, we study the optimization of balancing energy consumption with reliable data transmission. Aiming at the needs of energy equilibrium and reliability, we propose an adaptive reliable routing based on cluster hierarchy(arch) for WMSN. The ARCH routing can balance the energy consumption with meeting the need of reliability between the source and destination. To achieve better performance, we form the cluster structure by cellular topology. Moreover, we design a power allocation mechanism to adjust the transmitting power of nodes and an energy prediction mechanism to realize energy aware among nodes.we perform extensive simulation experiments to evaluate ARCH by several performance indexes. The results show that ARCH performs high efficiency on energy equilibrium and reliability in wireless multimedia sensor network. Acknowledgments The authors acknowledge the support from the National Natural Science Foundation of China (NSFC), Contract/Grant no ; National 863 High Technology Program of China, Contract/Grant no. 29AA1Z239; The Ministry of Science and Technology (MOST), International Science and Technology Collaboration Program, Contract/Grant no. 93. References [1] I. F. Akyildiz, T. Melodia, and K. R. Chowdhury, A survey on wireless multimedia sensor networks, Computer Networks, vol. 51, no. 4, pp , 27. [2] E. Gürses and Ö. B. Akan, Multimedia communication in wireless sensor networks, Annals of Telecommunications, vol. 6, no. 7-8, pp , 25. [3] S. Misra, M. Reisslein, and G. Xue, A survey on multimedia streaming inwireless sensor networks, IEEE Communications Surveys and Tutorials, vol. 1, no. 4, pp , 28. [4] K. Lin, L. Wang, K. Li, and L. Shu, Multi-attribute data fusion for energy equilibrium routing in wireless sensor networks, KSII Transactions on Internet and Information Systems, vol. 4, no. 1, pp. 5 24, 21. [5] S. Bhatnagar, B. Deb, and B. Nath, Service differentiation in sensor networks, in Proceedings of the 4th International Symposium on Wireless Personal Multimedia Communications, pp , Aalborg, Denmark, 21. [6] B. Deb, S. Bhatnagar, and B. Nath, ReinforM: reliable information forwardingusing multiple paths in sensor networks, in Proceedings of the 28th Annual IEEE International Conference on Local Computer Networks, pp , Los Alamitos, Calif, USA, 23.

121 1 EURASIP Journal on Wireless Communications and Networking [7] E. Felemban, C.-G. Lee, and E. Ekici, MMSPEED: multipath multi-speed protocol for QoS guarantee of reliability and timeliness in wireless sensor networks, IEEE Transactions on Mobile Computing, vol. 5, no. 6, pp , 26. [8] K. Sohrabi, J. Gao, V. Ailawadhi, and G. J. Pottie, Protocols for self-organization of a wireless sensor network, IEEE Personal Communications, vol. 7, no. 5, pp , 2. [9]L.Chenyang,B.M.Blum,T.F.Abdelzaher,J.A.Stankovic, and H. Tian, RAP: a real-time communication architecture for large-scale wireless sensor networks, in Proceedings of the 8th IEEE Real-Time and Embedded Technologyand Applications Symposium, pp , San Jose, Calif, USA, 22. [1] T. He, J. A. Stankovic, C. Lu, and T. Abdelzaher, SPEED: a stateless protocol for real-time communication in sensor networks, in Proceedings of the 23rd IEEE International Conference on Distributed Computing Systems (ICDCS 3),pp , Providence, RI, USA, 23. [11] K. Akkaya and M. F. Younis, Energy and QoS aware routing in wireless sensor networks, Cluster Computing, vol. 8, no. 2-3, pp , 25. [12] T. S. Rappaport, Wireless Communications: Principles and Practice, Prentice-Hall, Upper Saddle River, NJ, USA, 2nd edition, 22. [13] H. Kwon, T. H. Kim, S. Choi, and B. G. Lee, Crosslayer lifetime maximization under reliability and stability constraints in wireless sensor networks, in Proceedings of the IEEE International Conference on Communications (ICC 5), vol. 5, pp , May 25. [14] R. A. F. Mini, A. A. F. Loureiro, and B. Nath, Prediction-based energy map for wireless sensor networks, in Proceedings of the Personal Wireless Communications, vol of Lecture Notes in Computer Science, pp , 23.

122 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 1 pages doi:1.1155/21/ Research Article COSR: A Reputation-Based Secure Route Protocol in MANET Fei Wang, 1 Furong Wang, 1 Benxiong Huang, 1 and Laurence T. Yang 2 1 Electronic and Information Engineering Department, Huazhong University of Science & Technology (HUST), 137 Luoyu Road, Wuhan 4374, China 2 Department of Computer Science, St. Francis Xavier University, Canada Correspondence should be addressed to Fei Wang, wangfei@hust.edu.cn Received 2 April 21; Accepted 25 June 21 Academic Editor: Liang Zhou Copyright 21 Fei Wang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Now, the route protocols defined in the Mobile Ad Hoc Network (MANET) are constructed in a common assumption which all nodes contained in such networks are trustworthy and cooperative. Once malicious or selfish nodes exist, all route paths built by these protocols must be broken immediately. According to the secure problems within MANET, this paper proposes Cooperative On-demand Secure Route (COSR), a novel secure source route protocol, against malicious and selfish behaviors. COSR measures node reputation (NR) and route reputation (RR) by contribution, Capability of Forwarding (CoF) and recommendation upon Dynamic Source Route (DSR) and uses RR to balance load to avoid hotpoint. Furthermore, COSR defines path collection algorithm by NR to enhance efficiency of protocol. At last, we verify COSR through GloMoSim. Results show that COSR is secure and stable. 1. Introduction A mobile Ad Hoc network is a collection of autonomous nodes that communicate with each other. There are no base stations, access points, and any centralized control equipment. The entering and exiting of any node occur freely and without any management. Further, as the wireless transmission range of each individual node is limited, the establishment and maintenance of all route paths in the MANET depend on all other nodes. In this situation, a trustworthy environment is important to Ad Hoc network routing protocol. Recently, a lot of researches have been done on learning multihop path in the MANET. This body of literature can be categorized into two main groups. (1) Table-Driven Routing Protocols, such as DSDV [1], and OLSR [2]. These protocols maintain a consistent and up-to-date route table to all available destinations at each node. (2) On-Demand Routing Protocols, such as DSR [3], and AODV [4]. These protocols can react well to frequently node mobility and rapid network topology changes, and they are not necessary to maintain all routes periodically. We focus on on-demand routing protocols, especially DSR. This group of routing protocol has a common assumption that all nodes in an MANET are not malicious nodes and all of them are cooperative. Due to this assumption, misbehavior could destroy route paths established by routing protocol easily. So far, there are some secure routing protocols have been proposed, such as Ariadne [5], SEAD [6], CONFIDANT [7], and CORE [8]. In this paper, we make two contributions to the area of secure routing protocols for MANET. First, we propose a reputation-based secure source routing protocol, called COSR. According to the mobility, self-organization, and secure problems of MANET, node reputation in COSR is both regarded as its trustworthiness and CoF which the node claimed or promised to others. Further, COSR uses route reputation to choose the best route path. Second, we present the simulation of several routing attacks and execute performance evaluation of COSR. Relative to previous schema in reputation-based routing protocol, COSR is more secure and more efficient. The rest of this paper is organized as follows. Section 2 introduces main secure problems of current routing protocol

123 2 EURASIP Journal on Wireless Communications and Networking in the MANET, and Section 3 introduces related work. We present reputation model and routing protocol of COSR in Section 4. Section 5 performs simulation and discusses results. Section 6 provides our conclusion and future work about COSR. 2. Problems Current MANET routing protocols face a lot of problems, such as security and performance. We will describe them in detail as follows. (i) DoS: Due to the lack of packet certification in route discovery and path collection, attacker would inject a large number of protocol messages with wrong route information. In general, DoS attack can be performed as Route Cache Overflow [9] and Sleep Deprivation Torture [1]. (ii) Blackhole: An attacker attracts route paths to pass it by promising the shortest route or service and then drops or forwards data packets to other malicious nodes for mounting more sophisticated attacks. This attackincludestwo types: Active Blackhole and Passive Blackhole. Active Blackhole attackers that would inject wrong route information refer to received RREQ or overheard data packets, so that they might attract other nodes to choose them as relay station. On the contrary, Passive Blackhole attackers pose normal nodes during route discovery, and launch attack during data transmitting. (iii) Rushing [11]: This attack targets against on-demand routing protocols. The attacker relays received RREQ without any modification as soon as possible, suppressing any later legitimate RREQ. (iv) Wormhole [12]: Attackers forward RREQ packets by tunneling between attackers to disrupt communication. If a wormhole attacker tunnels all packets through wormhole honestly and reliably, no harm is done. (v) Selfish: In the MANET, nodes own limited resource, especially battery power and bandwidth. Some nodes refuse to forward or selectively forward the packets from other nodes to save its resource. 3. Backgrounds and Related Work As the MANET is dynamic topology and self-organized, traditional security mechanisms have no effect on routing attack. Hence, a lot of secure routing protocols have been proposed to refer to those attacks. According to their main idea, they can be categorized into four groups Based on Symmetric-Key Encryption. This group uses symmetric-key encryption to enhance the security of routing protocol in the MANET. They mostly apply One-Way Hash function and Hash link algorithm through symmetric key. This primary group includes Ariadne [5], SEAD [6], and SRP [13] Based on Asymmetric-Key Encryption. This group uses asymmetric-key encryption to protect routing protocol. They need a trustworthy independent authority, such as Certification Authority (CA). The authority is responsible for creating and publishing certificate for each node. The certificate contains permanent identity and public key. This group just includes ARAN [14] Based on Hybrid Encryption. This group uses both of upper encryption technologies. According to the common idea, the fixed part of route message would be signed by privatekeyforintegrality.further,privatekeyalsoisusedto finish node authentication. Furthermore, these protocols use Hash link to protect distance parameter. The representative proposals are SAODV [15] and SLSP [16] Based on Reputation. The above protocols use cryptography, authentication, and digital signature to protect security of data contained in the routing messages, and consequently, they enhance security of routing protocols. They belong to the area of hard security [17]. However, the hard security technology can hardly prevent any nodes from malicious or selfish behavior. Further, it also can not promote cooperation among nodes. Hence, reputation is to be implemented in the routing protocol to improve security of routing protocols. The representatives of them are CONFIDENT [7], CORE [8], ASU [13], Watchdog [18], and OCEAN [19]. Table 1 compares the capability of these protocols against above attacks. 4. COSR Protocol According to such security problems of the MANET, we proposed COSR protocol, a reputation-based secure dynamic source routing protocol. COSR assumes that the link between two nodes is bidirectional, and each node can work in promiscuous mode Protocol Architecture. COSR protocol follows cross-layer design [2]. In the COSR, node s reputation depends on the information from Physical layer, Media Access Control (MAC) layer, and Network layer, and it can be computed by node s CoF, history action, and recommendation. Hence, COSR can be divided into monitor, statistics, reputation model, reputation protocol, and routing protocol. Its architecture is shown in the Figure 1. (i) MONITOR: This part includes three modules: neighbor monitor, data relay monitor, and CoF monitor. Neighbor monitor works with MAC layer. It is used to monitor neighbors in its radio range and maintain neighbor list. Data relay monitor is placed in the network layer. It requires MAC layer working in promiscuous mode, so that it could check whether the next hop had transmitted its packets. CoF would collect information about capability of forwarding from physical layer and MAC layer, and itincludes node s bandwidth, interface state, mobility status, and power.

124 EURASIP Journal on Wireless Communications and Networking 3 Attacking ARAN [14] DoS Ariadne [5] Blackhole Rushing Table 1: Comparison of Secure Routing Protocols. Secure Route Protocols ASU CONFIDENT CORE OCEAN SAODV SEAD SLSP SRP [13] [7] [8] [19] [15] [6] [16] [13] WatchDog [18] Wormhole Selfish Note: -capable of defending such attack; -cannot defend such attack; -can be solved with improvement. Reputation model Route reputation Route protocol Node reputation Direct reputation Recommendation reputation Reputation protocol Statistics Monitor Data relay monitor Neighbor monitor CoF monitor Figure 1: Architecture of COSR protocol. (ii) STATISTICS: This module is responsible for providing statistics data about neighbors history behavior. These data include the number of requested and forwarded protocol messages, and data packets. (iii) REPUTATION model: This is the core module of COSR. It is used to evaluate node s reputation and integrate route reputation relying on the data from MONITOR and STATISTICS. (iv) Reputation Protocol: This part defines reputation discovery in the MANET. Reputation Protocol clings with routing protocol and uses routing protocol to pigback reputation control message and data. (v) Routing Protocol: It is an extension of DSR by reputation model. It uses NR and RR to choose the best route path rather than path length. Further, COSR provides a secure mechanism of path collection in route discovery Node Reputation. Firstly, because the COSR uses D-S approach [21] as basic theory to evaluate node s reputation, we introduce the key concept of it simply. Definition 1. Let Θ be a frame of discernment, Θ = {G, B}, where{g} means good, {B} is contrary, and {G, B} represents unknown. If a function m :2 Θ [, 1], where (1) m(ø) =, and (2) Â Θ m(â) = 1, then m is a basic probability assignment (bpa) over Θ. In the COSR, Node Reputation (NR) is defined as a combination of direct reputation, recommendation reputation, and CoF. It is defined as follows: NR ij = R direct ( i, j ) α + Rrec ( i, j ) β +cof ( j ) γ, (1) where NR ij stands for the node reputation value by node N i on node N j R direct (i, j), and R rec (i, j) are defined under Θ.

125 4 EURASIP Journal on Wireless Communications and Networking Weights α, β, andγ are the discount factors for different elements of NR, they are defined as follows: α, β, γ [, 1], α + β + γ = 1. (2) Direct Reputation (R direct ). R direct evaluates the trustworthiness of each next hop neighbor. And, it is a metric of a serial of good, bad and unknown actions. In the COSR, a node seems good that it forwarded all requested protocol messages and data packets. On the contrary, it seems bad. If there are no forwarding requests transmitted to target node, then it is unknown. Further, COSR uses different counters to do statistics of requested and forwarded protocol message and data packet, and received forwarding request. Refer to these counters, packet as metric for control message and byte for data packet. In this method, COSR can prevent some selfish nodes from selectively forwarding short data packets rather than long data packets. The node reputation contains three parts: m({g}), m({b}), and m({g, B}). They denote trust, distrust, and unknown, respectively, and, they are computed by above counters. Hence, R direct is defined as follows: m({g}) m({b}) R direct = m({g}) + m({g, B}), m({g}) >m({b}), (3), others Recommendation Reputation (R rec ). R rec represents others subjective evaluation according to target s behavior, cooperation, and so forth. Therefore, it is a combination of a lot of recommendations from neighbors. In the COSR, a recommendation is defined as follows: Rec = m({g}), m({b}), m({g, B}). (4) A recommendation is originated from direct reputation and subjective view about target node. In the COSR, a recommendation is a basic probability assignment of Θ,andCOSR uses D-S formula to combine all received recommendation. Hence, the metric of R rec is defined as the following formula: m({g}) m({b}), m({g}) >m({b}), R rec =, others Capability of Forwarding (CoF). CoF denotes the capability of forwarding packets of a certain node. Simply, we use the remained power, bandwidth, and mobility state to evaluate it. In the CoF, remained power and bandwidth are mandatory, however, mobility state is optional. Only when the node supports Global Positioning System (GPS), it should provide mobility state and its velocity. As the information of CoF is provided by its owner, malicious node might cheat others by false data. To avoid the emergence of such malicious behavior, COSR takes the following strategies. (1) Discounting.COSRusesnode sreputation to discount those providing CoF data. (2) Punishment. Once COSR finds that any node provided a false CoF, it will punish such node through reducing its reputation level. (5) 4.3. Route Reputation. As the metric of the best route in traditional routing protocols does not concern security, data transmitting must fail when a route path contained one or more malicious nodes. Hence, to avoid such phenomenon, COSR uses two metrics, hops, and intermediate nodes reputation, and then, we define Route Reputation as the metric of the best route path. In the COSR, Route Reputation of {N i N k1 N k2 N km N km +1 N j } is defined by NR iki NR kn 1k n, NR klk m, RR ij = 1, other, where N k1, N k2,..., N km+1 are the intermediate nodes. The RR is composed of NR of all intermediate nodes in a certain route path. As NR is a real number between and 1, when such route contained more intermediate nodes, its RR would be lower even if it closes 1. Hence, shorter route path gains higher RR, if there are no malicious nodes. However, if there are malicious nodes in any short route, the RR of such route must be lower than a secure longer route. Therefore, RR gives attention to both of efficiency and security of route path. Further, according to formula (6), NR is independent of RR. In other words, any node can earn higher reputation even if it is included by a route with lower reputation. Hence, COSR does not like [13] that penalizes such nodes around a malicious node Recommendation Game (RG). As direct observation about a strange node mostly is limited even not existent, many solutions (such as CONFIDANT [7], CORE [8], and COSR) use second-hand evaluation to accelerate collecting evidence. However, a malicious node may be described as a perfect node through unfair positive recommendation, and then they can easily inject wrong route into network and launch attacks. Consequently, unfair or false recommendation can break down the constructed reputation system easily. Therefore, we provide a novel mechanism, recommendation game, against false recommendation Framework Assumption 1. All nodes in the MANET are rational. In the recommendation game, there are two kinds of players (node): reputation requester and provider. No matter what reputation requester or provider are, they both deal with reputation request and recommendation rationally. To formulate the recommendation game, we now give its definition. Definition 2. Let RG be a Recommendation Game: RG = {I, A, T, P, U},whereI is the set of mobile nodes (reputation requester and providers); A ={A i },wherea i is the action space of node i I; T ={T i },wheret i isthetypespaceof node i; P ={p i },wherep i is a belief of other players type basedonthetypeofnodei; U ={u i },whereu i is the payoff function for node i. (6)

126 EURASIP Journal on Wireless Communications and Networking 5 Reputation Requester Table 2: Payoff table of recommendation game. Reputation provider Truth No comment Lie Trust S, t, S, l Distrust S, t, S, l According to the definition, recommendation game has the following properties. Property 1. Recommendation game is an n-player game, where n 2. In the node set I, there are only one reputation requester and at least one provider. If there is no reputation provider, then this game cannot be played. Property 2. Recommendation game is an incomplete information game. Firstly, a requester contained in an RG does not know if a provider is lying. Secondary, requester does not know the relationship between a given provider and the target, while providers know whether the target is their confederate or not. At last, reputation provider does not know whether the requester would trust it or not, and it also does not know how the requester evaluates a recommendation. Definition 3. Let A r and A i p be action spaces of reputation requester and reputation provider, respectively, then the action spaces are defined by A r ={Trust, Distrust}, A i p ={Lie,Truth,No-Comment}, i I. (7) Definition 4. Let T r and T p be type spaces of reputation requester and reputation provider, respectively. Given a pair of independent constant parameters TC r and TC p,where TC r, TC p (, + ), we can define their type spaces as follows: [ ] T r = [, TC r ], T p =, TCp. (8) Definition 5. Let p r and p p be belief of reputation requester and reputation provider, respectively, they are defined as ( ) p r tp = 1, p p (t r ) = 1, (9) TC p TC r where t r and t p are private information of reputation requester and reputation provider, respectively, and t r T r, t p T p. Both of them are random variables and follow uniform distribution in their type spaces Strategy and Payoff. The strategy of reputation requester and provider can be shown in Table 2. If reputation provider gives a fair recommendation, no matter the requester believes or not, it could gain positive payoff, namelyt. However, if it cheats the requester, then it would be denoted as malicious node. And then, such provider would lose others trust in the later transactions. According to the requester s strategy, Tit for Tat, it would gain a negative reputation, namely l. t, and l, both are positive parameters and t is less than l commonly. It means that it is more difficult to build reputation and easier to destroy it. On the other hand, if the requester trusts a lie, then the related attacks maybe come true. Consequently, the requester would gain negative payoff about data transmitting. On the contrary, it might obtain a positive payoff for distrusting a lie. Theory 1. Recommendation requester does not have the absolute best strategy. The strategy of recommendation requester depends on the evaluation of private information about recommendation provider, hence, it does not have the absolute best strategy. In other words, recommendation requester has the risk of the final strategy for ever. Theory 2. The recommendation provider conditional makes TRUTH as it is the absolute best strategy, and the condition is defined as TC p < (t + l). (1) Equation (1) is the design reference about parameters of RG. We can choose appropriate t and l according to the detail application scenario, to make recommendation provider has only one strategy, TRUTH, and then, unfair positive recommendation should be reduced sharply Reputation in Route Discovery. Routing protocol in the COSR is based on DSR, hence, route discovery of COSR contains two mechanisms which are shown in Figure RREQ and RREP. RREQ and RREP are basic methods to discover route paths, however, they only can obtain limited route paths which are shown as white blocks in Figure 2(b). In this procedure, malicious nodes may launch attack in two cases: broadcasting wrong RREQ and transmitting false RREP. The possible attacks include DoS, Blackhole, and selfish. Due to this problem, COSR uses reputation against such attacks. When a node received an RREQ or RREP packet, it should check the sender s reputation, firstly. If sender s reputation does not refer enough to reputation threshold, such node would not believe route information contained in received packet and would abort it without any notification Path Collection. Path collection is an extended mechanism to discover more route paths in only one procedure of RREQ and RREP. Path collection includes two directions: forwarding and reversing. Firstly, forwarding direction only allows source node and intermediate nodes collect route paths along the direction to destination contained in RREP.

127 6 EURASIP Journal on Wireless Communications and Networking Table 3: Simulator parameters Dimensions of Space 1 m 1 m Total Number of Nodes 4 Node Placement Uniform Mobility Random Waypoint Model Move Speed 2m/s Mobile Pause Time s Max Transmission Range 25 m MAC Link Bandwidth 2Mbps Application 3 CBR Connections, 64 B/packet, 1 packet/s Simulation Time 9 s Payoff Parameters of RG α =.3, β = 1. It is shown as lightgreen blocks in Figure 2(b). Secondary, reverse direction permits all nodes related to this route discovery to collect reverse route paths to upriver nodes. It is shown as lightcyan blocks in Figure 2(a). Obviously, path collection could accelerate route discovering. However, it still has to face a serial of malicious attacks, such as DoS, Blackhole, and selfish, because relative packets and route information were not certificated. Similarly, COSR uses reputation of packets sender to verify whether to believe or not. 5. Evaluation of COSR 5.1. Environment. The evaluation of COSR is performed upon the GloMoSim, a simulator for wireless network. In the GloMoSim, we configure a mobile Ad Hoc network with a number of connections. Each experiment was repeated FIVE times with different seed which is a parameter configured in the GloMoSim and affects placement and movement of nodes. The main parameters of environment are listed in Table Scenario. The evaluation would be done at following scenarios. (i) Blackhole: This attack includes active Blackhole (ABH) and passive Blackhole (PBH). Attackers would drop all data packets which need forwarding. Blackhole attackers would never initiate a CBR connection. Active Blackhole attacker will actively sniff neighbors RREQ and inject RREP with false route path. Passive Blackhole attackers guise normal nodes in the route discovery, but they pose blackhole in the data transmission. (ii) Selfish: Selfish nodes discard data packets selectively according to preconfigured probability. (iii) DoS: Attackers would inject various protocol messages with wrong and long route, including RREQ, and RREP. Once attack is running, attacker would not participate in any application. Protocols Table 4: The comparison of ratio of packet received Attacks ABH PBH Selfish DoS COSR 79% 76% 82% 69% DSR 34% 42% 76% 61% CONFIDENT without Configured Friends 4% 52% 77% 45% CONFIDENT with Configured Friends 76% 8% 85% 77% 5.3. Performance Metrics (i) Average Path Length (APL): This is defined as the average hop number of delivered data packets between sources and destinations. It denotes end-to-end delay. (ii) Percentage of Packet Received (PPR): This is defined as the ratio of the number of data packets received by the destinations to all sent by the source nodes. PPR not only shows the throughput of routing protocol, but also reflects the security and reliability of it. (iii) Normalized Protocol Load (NPL): This is defined as the ratio of the number of originated control messages to the number of delivered data packets. NPL describes the efficiency of routing protocol Results and Discussing Varying Proportion of Blackhole and Selfish Nodes. Figure 3 shows the capability of COSR against Blackhole and selfish attack. According to the result, COSR improves the PPR largely contrasts with DSR on both active Blackhole and passive Blackhole. Due to lack of any security mechanism, DSR s PPR drops down sharply and only 2% of data packets are delivered successfully at the worst situation. On the Contrary, COSR can maintain PPR to be about 5% even 9% of nodes are malicious. The result of selfish attack is similar to Blackhole Varying Proportion of DoS Nodes. Figure 4 shows the capability of COSR against DoS attack. The result shows that DSR s PPR decreases sharply with the growing of the ratio of rushing attacker. Due to injecting a large number of wrong route information by DoS attacker, mostly nodes route cache within DSR would overflow rapidly. At the worst situation, less than 1% of data packets can be delivered successfully. However,COSRplaysbetteranditsPPRismuchsmoother than DSR s Varying Proportion of Liars. As recommendation is the main element of reputation, false recommendation produced by liars may destroy reputation system. Figure 5 verifies the capability of COSR against lies. This experiment is to be done at the situations in which the network contains 5% active Blackhole attackers. Further, all liars are selfish. The result shows how many data packets are delivered when there are liars in the MANET for two scenarios: COSR without RG

128 EURASIP Journal on Wireless Communications and Networking 7 A B D E A (A) C C (A, C) B C C, A (A,C,B) (A, C) (A, C) (A,C,D) C C, A D F E C C, A E D (A, C, E) (A, C,E) (A,C,D) G (A,C,E,F) D C D, C C, D D, C, A C, D, G F C, E F, E C, E, F F, E, C C, E, F, G F, E, C, A A (A,C,D,G) (A,C,E,F,G) C F C, D B E C, D, G E, C C, E E, C, A C, E, F C, E, F, G D D, G E E, F E, F, G C (A, C, E, F, G) (A,C,D,G) E D (A,C,E,F,G) F F, G D D, G G E E, F E, F, G (A,C,D,G) F G (A, C, E, F, G) G RREQ (a) Propagation of RREQ RREP (b) Propagation of RREP Figure 2: Route discovery and path collection in COSR Packet received (%) 6 4 Packet received (%) Malicious or selfish nodes(%) DoS attacker (%) 8 1 COSR: active blackhole COSR: passive blackhole COSR: selfish DSR: active blackhole DSR: passive blackhole DSR: selfish Figure 3: COSR against blackhole and selfish. DSR COSR Figure 4: COSR against DoS. and COSR with RG. By contrasting Figure 5 with Figures 3 and 4, itcanbefoundthatliescouldinfluencetheeffect of reputation model of COSR, but when RG begins work, its PPR is improved obviously Performance. These experiments are done at the environment, that there are 3% of nodes are malicious, and they are done at three scenarios: active Blackhole, passive Balckhole, and selfish. In the Figure 6, with the growing of pause time, NPL is decreased continually, because longer pause time makes topology of network stabler. By contrasting COSR with DSR, NPL of COSR is larger than that of DSR, especially when pause time is small. On the contrary, their NPLs are close when topology of network is changing slowly. Figure 7 shows the NPL of COSR and DSR with the growing of maximum velocity. It is the same as Figure 6, NPL of COSR is larger than that of DSR when topology of network is changing fast.

129 8 EURASIP Journal on Wireless Communications and Networking Packet received (%) Normalize protocol load Liars (%) Maximum velocity (m/s) 3.5 COSR without RG COSR with RG Figure 5: COSR against Liars. COSR: active blackhole COSR: passive blackhole COSR: selfish DSR: active blackhole DSR: passive blackhole DSR: selfish Figure 7: NPL of COSR and DSR with various maximum velocity. Normalize protocol load Pause time (s) COSR: active blackhole COSR: passive blackhole COSR: selfish DSR: active blackhole DSR: passive blackhole DSR: selfish Figure 6: NPL of COSR and DSR with various pause time. Packet received (%) Pause time (s) COSR: active blackhole COSR: passive blackhole COSR: selfish DSR: active blackhole DSR: passive blackhole DSR: selfish Figures 8 and 9 show the throughput of COSR and DSR with such scenarios. According to the results, COSR gains higher throughput than DSR, even topology of network is changing fast. As active Blackhole attacker performs more active malicious behavior so that COSR could detect it easier than passive Blackhole, though active Blackhole is more aggressive than passive Blackhole in DSR. Figures 1 and 11 give the end-to-end delay of data packets. With various pause time, APL of COSR is less than that of DSR greatly. However, their APLs are closed with various maximum velocity. This shows that pause time is more important on affecting network topology than mobile velocity. With growing pause time, the difference becomes significant because fixed network topology is more conducive to COSR detecting malicious nodes. Figure 8: Throughput of COSR and DSR with various pause time Comparison. Moreover, we do a comparison simulation among COSR, DSR, and CONFIDENT within the scenario which is described by Table 3 and contained 3% malicious nodes. According to CONFIDENT, we provide two sets of simulation result. In the first set, CONFIDENT does not contain preconfigured friends list. In the other set, we configure several default friends in each node s Trust Manager before simulation according to CONFIDENT s design. In this comparison simulation, we mainy compare the ratio of packet received when the routing protocol is against active Blackhole, passive Blackhole, Selfish, and DoS.

130 EURASIP Journal on Wireless Communications and Networking Packet received (%) 6 4 Average path length COSR:activeblackhole COSR: passive blackhole COSR: selfish Maximum velocity (m/s) DSR: active blackhole DSR: passive blackhole DSR: selfish Figure 9: Throughput of COSR and DSR with various maximum velocity Maximum velocity (m/s) COSR: active blackhole COSR: passive blackhole COSR: selfish DSR: active blackhole DSR: passive blackhole DSR: selfish Figure 11: APL of COSR and DSR with various maximum velocity. 6. Conclusion Average path length Pause time (s) COSR: active blackhole COSR: passive blackhole COSR: selfish DSR: active blackhole DSR: passive blackhole DSR: selfish Figure 1: APL of COSR and DSR with various pause time. According to the simulation result shown in Table 4,we can find that CONFIDENT gains the highest performance when it contains preconfigured friends. However, CONFI- DENT s performance decreases sharply when we canceled the preconfigured friends. At this situation, CONFIDENT is similar to DSR, on the contrary, COSR acquires satisfactory performance. The reason is that COSR designed a dynamical mechanism to construct friendship between strange nodes in the SON, but CONFIDENT does not provide such scheme. Therefore, COSR is more suitable for the dynamical SON in which there are a lot of nodes joining and leaving continuously. By using dynamic source routing protocol, communication among self-organized wireless network comes true. However, the existing malicious nodes destroyed the traditional routing protocol, such as DSR, and AODV. To mitigate the effect of malicious behavior, we present a reputation-based secure routing protocol, called COSR, for MANET. The COSR uses a novel reputation model to detect malicious and selfish nodes and make all nodes more cooperative. Further, reputation is not only used to evaluate the trustworthiness of any node, but also to describe its CoF. Due to such design, COSR can protect network against the primary routing attacks and balance load on all secure route paths to avoid hotpoint and enlarge throughput of whole network consequently. Under most simulation scenarios, COSR improves PPR, and APL largely refers to DSR, though NPL of COSR is more than that of DSR. Therefore, the ongoing research of COSR is improving is efficiency ofcosr. We hopeto decrease its NPL with current PPR and APL levels. Acknowledgment This work was supported by Program for new Century Excellent Talents in University (NCET-6-642). References [1] C. E. Perkins and P. Bhagwat, Highly dynamic destinationsequenced distance-vector routing(dsdv) for mobile computers, in Proceedings of the Conference on Communications Architectures, Protocols and Applications (SIGCOMM 94), pp , ACM Press, London, UK, August-September [2]T.Clausen,P.Jacquet,A.Laouiti,etal., Optimizedlink state routing protocol, Internet-Draft, draft-ietf-manet-olsr- 5.txt, October 21.

131 1 EURASIP Journal on Wireless Communications and Networking [3]D.B.JohnsonandD.A.Maltz, Dynamicsourcerouting in ad hoc wireless networks, in Mobile Computing, chapter 5, pp , Kluwer Academic Publishers, Dodrecht, The Netherlands, [4] C. E. Perkins and E. M. Royer, Ad-hoc on-demand distance vector routing, in Proceedings of the 2nd IEEE Workshop Mobile Computing Systems and Applications (WMCSA 99),pp. 9 1, IEEE Press, [5] Y.-C. Hu, A. Perrig, and D. B. Johnson, Ariadne: a secure ondemand routing protocol for ad hoc networks, in Proceedings of the 8th Annual International Conference on Mobile Computing and Networking, pp , September 22. [6] Y.-C.Hu,D.B.Johnson,andA.Perrig, SEAD:secureefficient distance vector routing for mobile wireless ad hoc networks, in Proceedings of the 4th IEEE Workshop on Mobile Computing Systems & Applications, IEEE, June 22. [7] S. Buchegger and J.-Y. Le Boudec, Performance analysis of the CONFIDANT protocol (cooperation of nodes: fairness in dynamic ad-hoc networks), in Proceedings of the 3rd ACM International Symposium on Mobile Ad Hoc Networking and Computing (MOBIHOC 2), pp , ACM Press, June 22. [8]R.M.P.Michiardi, Core:acollaborativereputationmechanism to enforce node cooperation in mobile ad hoc networks, in Proceedings of the 6th Joint Working Conference on Communications and Multimedia Security, pp , IEEE, December 22. [9] J. Lundberg, Routing Security in Ad Hoc Networks, [1] F. Stajano and R. Anderson, The resurrecting duckling: security issues for ad hoc wireless networks, in Proceedings of the 7th International Workshop on Security Protocols, pp , Cambridge, UK, April [11]Y.-C.Hu,A.Perrig,andD.B.Johnson, Rushingattacks and defense in wireless ad hoc network routing protocols, in Proceedings of the ACM Workshop on Wireless Security, pp. 3 4, ACM, September 23. [12]Y.-C.Hu,A.Perrig,andD.B.Johnson, Packetleashes:a defense against wormhole attacks in wireless networks, in Proceedings of the 22nd Annual Joint Conference on the IEEE Computer and Communications Societies, vol. 3, pp , San Francisco, Calif, USA, March-April 23. [13] P. Dewan, P. Dasgupta, and A. Bhattacharya, On using reputations in ad hoc networks to counter malicious nodes, in Proceedings of the 1th International Conference on Parallel and Distributed Systems (ICPADS 4), pp , Newport Beach, Calif, USA, July 24. [14] B. Dahill, B. N. Levine, E. Royer, and C. Shields, A secure routing protocol for ad hoc networks, Tech. Rep. 1-37, Department of Computer Science, University of Massachusetts, August 21. [15] M. G. Zapata and N. Asokan, Securing ad hoc routing protocols, in Proceedings of ACM Workshop on Wireless Security, pp. 1 1, ACM Press, September 22. [16] P. Papadimitratos and Z. J. Haas, Secure link state routing for mobile ad hoc networks, in Proceedings of IEEE Workshop on Security and Assurance in Ad Hoc Networks, pp , IEEE, 23. [17] L. Rasmusson and S. Janssen, Simulated social control for secure internet commerce, in Proceedings of the New Security Paradigms Workshop, ACM, [18] S. Marti, T. J. Giuli, K. Lai, and M. Baker, Mitigating routing misbehavior in mobile ad hoc networks, in Proceedings of the 6th Annual International Conference on Mobile Computing and Networking (MOBICOM ), pp , Boston, Mass, USA, August 2. [19] S. Bansal and M. Baker, Observation-based cooperation enforcement in ad hoc networks, Research Report cs. NI/3712, Standford University, 23. [2] M. Conti, G. Maselli, G. Turi, and S. Giordano, Cross-layering in mobile ad hoc network design, Computer,vol.37,no.2,pp , 24. [21] G. Shafer, A Mathematical Thoery of Evidence, Princeton University Press, Princeton, NJ, USA, 1976.

132 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 13 pages doi:1.1155/21/ Research Article A Multimedia Application: Spatial Perceptual Entropy of Multichannel Audio Signals Shuixian Chen, 1 Ruimin Hu, 1, 2 and Naixue Xiong 1 1 Computer School, Wuhan University, Wuhan 4372, China 2 National Engineering Research Center for Multimedia Software, Wuhan University, Wuhan 4372, China Correspondence should be addressed to Naixue Xiong, n.xiong@whu.edu.cn Received 17 November 29; Accepted 11 February 21 Academic Editor: Liang Zhou Copyright 21 Shuixian Chen et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Usually multimedia data have to be compressed before transmitting, and higher compression rate, or equivalently lower bitrate, relieves the load of communication channels but impacts negatively the quality. We investigate the bitrate lower bound for perceptually lossless compression of a major type of multimedia multichannel audio signals. This bound equals to the perceptible information rate of the signals. Traditionally, Perceptual Entropy (PE), based primarily on monaural hearing measures the perceptual information rate of individual channels. But PE cannot measure the spatial information captured by binaural hearing, thus is not suitable for estimating Spatial Audio Coding (SAC) bitrate bound. To measure this spatial information, we build a Binaural Cue Physiological Perception Model (BCPPM) on the ground of binaural hearing, which represents spatial information in the physical and physiological layers. This model enables computing Spatial Perceptual Entropy (SPE), the lower bitrate bound for SAC. For real-world stereo audio signals of various types, our experiments indicate that SPE reliably estimates their spatial information rate. Therefore, SPE plus PE gives lower bitrate bounds for communicating multichannel audio signals with transparent quality. 1. Introduction A central goal in multimedia communications is to deliver quality contents with the lowest possible bitrate. By quality, we mean the perceived fidelity of the received contents against the original contents. And the lowest possible bitrate depends on two disparate concepts: entropy and perception. Entropy measures the quantity of information [1]. But not all information is perceptible. To pursue this goal, we want to know how many bits are sufficient to convey quality multimedia contents. Lossless compression always ensures the highest possible quality, in which the objective redundancy in the multimedia contents is the only source of compression, and there is a limit, the Shannon entropy, the lowest possible bitrate with perfect decompression. Nevertheless, this limit is very hard if not impossible to compute due to the diversity and complexity of probability models of multimedia contents. By Huffman coding, run-length coding, arithmetic coding, and other entropy coding techniques, the state-of-the-art lossless audio coders today typically achieve a compression rate of 1/3-2/3 or kbps per channel for CD music [2]. Lossless compression generally conveys higher than necessary quality in multimedia communications. Multimedia contents abound subjective irrelevancy objective information we cannot sense. Perceptually lossless compression suffices. For audio signals, this means lossless to the extent that the distortion after decompression is imperceptible to normal human ears (usually called transparent coding), the bitrate can be much lower than the true lossless coding. Perceptual audio coding [3] by removing the irrelevancy greatly reduces communication bandwidth or storage space. Psychoacoustics provides a quantitative theory on this irrelevancy [4 7]: the limits of auditory perception, such as the audible frequency range (2 2 Hz), the Absolute Threshold of Hearing (ATH), and the masking effect [8]. In state-of-the-art perceptual audio coders, such as MPEG-2/4 Advanced Audio Coding (AAC [9, 1]), 64 kbps is enough for transparent coding [11]. The Shannon entropy cannot measure the perceptible information or give the bitrate bound in this case.

133 2 EURASIP Journal on Wireless Communications and Networking In 1988, Johnston proposed Perceptual Entropy (PE [12, 13]) for audio coding based on psychoacoustics. PE gives the lower bitrate bound for perceptual audio coding: PE = 1 25 b i+1 1 ( ( ) ) log N 2 2 nint Re(ωk ) +1 i=1 6ni k=b i /k i ( ) ) +log 2 (2 nint Im(ωk ) +1, 6ni /k i (1) Other senses Psychology (cognition) Physiology (excitation/inhibition) Physics (sound wave propagation) Prior knowledge where PE is measured in bits per sample, N the length of block transform (usually DFT), nint() integer rounding, b i the index of starting bin of subband i, ω k the kth transform coefficient, n i the undetectable distortion upper bound of band, i and k i the number of bins in subband i. Table 1 lists PE for various mono audio signals. The last column gives nears transparent bitrates of current coders, slightly lower than the upper bound of PE. We can see that if n i in (1) assumes conservative values (smaller), PE will be larger. On the other hand, Adaptive Multirate (AMR [14]) and Adaptive Multirate Wide Band (AMR-WB [15]) use a priori knowledge of human voicing, also reducing bitrate. Apart from these two points, PE reliably predicts the lowest bitrate required for transparent audio coding. Since formulated, PE has found widespread use in audio coding and has become a fundamental theory in this field. Main stream perceptual audio coders, such as MP3 [16] and AAC, all employ PE as an important psychoacoustic parameter, leading to various practical methods not just theory. Nevertheless, PE has significant limitation to measure perceptual information. This limitation primarily comes from the underlying monaural hearing model. Human has two ears to receive sound waves in a 3-dimensional space: not only is the time and frequency information perceived needing just individual ears but also spatial information or localization information needing both ears for spatial sampling. Due to the unawareness of binaural hearing, PE of multichannel audio signals is simplified to the supposition of PE of individual channels, which is significantly larger than real quantity of information received because multichannel audio signals usually correlate. The purpose of this paper is to measure the perceptual information of binaural hearing. We first analyze the localization principle of binaural hearing and give a spatial hearing model on the physical and physiological layers. Then we propose a Binaural Cue Physiological Perception Model (BCPPM) based on binaural hearing. Finally using binaural frequency-domain perception property, we give a formula to compute the quantity of spatial information and numerical results of spatial information estimation of real-world stereo audio signals. With the left and right ears, human being is able to detect spatial information: sound source localization and sound source spaciousness. The former comprises of the range, azimuth, and elevation, in other words, the 3-dimensional spherical coordinate. The later can be measured by angle span of auditory images. Figure 1: 3 Layers of auditory sound source localization. Human spatial hearing is a complex procedure of physics, physiology, and psychology (Figure 1). Psychology stays on the top of this procedure. On this layer, hearing is transformed to cognition, substantially influenced by subject psychological state, other senses, especially visual perception, and knowledge, implying that the same sound does not necessarily produce the same hearing perception. In Spatial Hearing, Blauert gives examples that different subjects in the same sound environment have diverse description of the environment [17]. In1998, Hofmanet al. reported innature that subjects with modified pinnae shape lost the elevation detection ability at the beginning but gradually regained that full ability [18]. This phenomenon demonstrates that the subjects were able to learn the correspondence between frequency response characteristics with the modified pinnae and sound from different elevations and used the knowledge to guide elevation detection. Due to the above reasons, spatial hearing on the psychological layer is too complicated to be exploited in audio compression systems, which cannot assume any specific states, senses, and knowledge of listeners. On the physical layer, sound waves propagate from sources along different paths to the ears and then in the ear canals and finally to the cochlea, absorbed and reflected by walls, floors, torso, head, and other objects on the way. Those sound waves carry objective localization information. On the physiological layer, sound waves are transformed to neural cell excitation and inhibition by the auditory system. There are different types of auditory neural cell responding to different types of sound stimulus, such as intensity, frequency, and delay. Thus physical quantities become physiological data. In audio compression, irrelevancy removing is mainly on the physical and physiological layers. In the following, we discuss the representation of binaural cues on the two layers BCPPM Spatial Information on the Physical Layer. As early as 197, Rayleigh studied the physics of spatial hearing [19]: Interaural Time Difference (ITD) and Interaural Level Difference (ILD). Also Rayleigh has two seminal discoveries: the famous duplex theory, that is, below 1.5 khz, ITD is the primary localization cue and above 1.5 khz ILD instead, the head-shadow effect, that is, the blocking and reflection of sounds by head produce a maximum of 2 db intensity

134 EURASIP Journal on Wireless Communications and Networking 3 Table 1: PE and bitrate of various mono audio signals [13]. Sampling Rate (khz) Band Width (khz) PE (bits/sample) Bitrate (kbps) Near Transparent Coding Bitrate (kbps) (AMR [14]) (AMR-WB [15]) (AAC [9]) Plane wave Figure 2: The rigid ball model of human head used by Rayleigh. difference. Both discoveries are derived based on the rigid ball modeling of head (Figure 2). 2. Physiological Perception Modeling of Binaural Hearing Although a real head is far from being the rigid ball, the above results are basically correct. In 22, Macpherson and Middlebrooks demonstrated that the duplex theory is suitable for a variety of audio signals: pure tones, wide band signals, high pass signals, as well as low pass signals [2]. Exception is high frequency signals with envelope delay [17]. ITD and ILD are not all the localization cues. On the medial plane (which cuts perpendicularly through the middle of the line connecting the left and right ears), all sound sources have ITD = msandild = db. But when they have different elevations, our auditory system can detect the difference by elevation-related spectral characteristics [21 24]. Due to the asymmetric structure of pinnae [25], the interference of sound waves is both wavelength related and elevation related (Figure 3). For example, the frequency of the lowest spectral amplitude (interference annihilation) is a function of the elevation [26]. This is the root of our elevation detection ability. This spectral cue does not depend on binaural hearing, so it is also called monaural cue. Unlike ILD and ITD, the spectral cue needs prior knowledge to provide elevation information. In principle, sounds may have arbitrary spectra. A listener is not able to detect the elevation angle based solely on the spectra: any characteristics may come from sound sources themselves and may come from the filtering effect of pinnae. The listener cannot tell. Blauert reported a very interesting auditory phenomenon of narrow-band sound sources on the medial plane: the elevation angles given by subjects are independent of the real elevation angles but depended on the signal frequencies [17]. For wide-band signals of familiar types, it is easy for our auditory system to compare the pinnae filtered spectra (some frequency amplified and some decayed) to the spectra in memory, and based on the difference, reliable elevation angle estimation can be given (Figure 3). But for narrowband signals, pinnae filtered spectra do not have detectable shape difference, just level difference. Thus the elevation angle detection will be very unreliable. In fact, the elevation angles given by the subjects are the angles at which the narrow-band signals have the maximum gain due to the pinnae filtering. For example, the peak gain frequency when the sounds come from the front is 3 khz for most people [21]. So wherever a sound of 3 khz came from, most subjects pointed at the front. From the perspective of signal processing, sound wave propagation is roughly a Linear Time Invariant (LTI) system. To describe this LTI system in binaural hearing, we have Head-Related Transfer Function (HRTF [27 29]) or equivalently Head-Related Impulse Response (HRIR). In open space, HRTF/HRIR is the function of source location, that is, range, azimuth, and elevation. Figure 4 shows the HRTFs in binaural hearing. Signal S(jω) goes from the source though the left and right paths to the left and right ears, respectively. Denote by Hl θ (jω) the left path HRTF and by Hr θ (jω) the right path HRTF. Then S θ l (jω) = Hθ l (jω)s(jω) is the entrance signal of the left ear, so is S θ r (jω) = Hr θ (jω)s(jω). Since the signal may have any spectra, localization cannot be determined solely by S θ l (jω) or S θ r (jω). Suppose that there are no strict zeros in the signal and the HRTFs. To exclude the effect of S(jω), we define Binaural Difference Transfer Function (BDTF): ( ) S θ ( ) ( ) jω = r jω ( ) = Hθ jω r ( ), (2) jω jω H θ Δ S θ l which is independent of S(jω) and located related. BDTF contains the same spatial information as S θ l (jω)andsθ r (jω). In fact, we can find ILD and ITD from it: ( H θ( ) ) ILD = 2log 1 Δ jω, ITD = d dω arg( HΔ θ ( ) ) (3) jω. Obviously, ILD and ITD are not only source location dependent, but also frequency dependent. To obtain accurate relationship between sound source locations and sound wave propagation, more realistic head models or real heads are needed. In 1994, the MIT Media H θ l

135 4 EURASIP Journal on Wireless Communications and Networking Up Front (a) (b) (db) (db) Frequency (khz) (c) Frequency (khz) (d) Figure 3: Elevation angle detection (Modified from tutorial/3d psych/elev.htm). Lab collected HRTFs on 71 locations in the 3-dimensional space using the KEMAR head [3]. In 21, CIPIC of U.C. Davis examined HRTFs of 45 subjects and 2 KEMAR heads [31]. Individual difference of HRTFs is revealed in HRTFs obtained by the experiments. Nevertheless, there are common characteristics that are sufficient to derive subjectindependent spatial information. H θ l (jω) S(jω) H θ r (jω) 2.1. Spatial Information on the Physiological Layer. In human auditory system, ITD and ILD of external sound sources stimulate or inhabit specific neural cells in the full audible frequency range. This process comprises of two steps: Frequency-to-Place Transform (FPT) [32, 33] and Binaural Processing (BP). In 196, Bèkèsy reported that sounds of different frequencies generate surface waves on the basilar membrane in cochlea with peak amplitudes at different places, which are determined by the frequencies [34]. In other words, a specific frequency is mapped to a specific place on the basilar membrane, or FPT, and this specific frequency for a given place is called Characteristic Frequency (CF [35]). Hair cells on that place then transform the mechanical swing into electric signals of auditory nerves. H θ l (jω)s(jω) H θ r (jω) H θ l (jω) θ H θ r (jω)s(jω) Figure 4: Binaural hearing transfer functions. The neural signals from the left and right ears corresponding to the same frequency meet in the brain. Our auditory system then extracts the ITD and ILD information in the signals. Currently, there are two kinds of theories on

136 EURASIP Journal on Wireless Communications and Networking 5 Output 1 Output 2 Output n 1 Outputn Left ear in ΔT ΔT ΔT ΔL ΔL ΔL ΔL Left ear in ΔT ΔT ΔT Right ear in ΔT ΔT ΔT. ΔL EI EI EI ΔL ΔL ΔL. Figure 5: Jeffress model: delay line network. EI EI EI ΔL ΔL ΔL ΔL this process: Excitation-Excitation (EE [36]) and Excitation- Inhibition (EI [37]). The former proposed that there are auditory nerve cells of EE-type located between the inferior colliculus and the medial superior olive, and specific EEtype cells there have maximum excitation for signals with specific ITD and ILD; the latter proposed that there are auditory nerve cells of EI-type located between the inferior colliculus and the lateral superior olive, and specific EI-type cells there have maximum inhibition for signals with specific ITD and ILD. The common ground of the two theories is that specific nerve cells are only sensitive to specific ITD and ILD, which are called characteristic ITD and characteristic ILD. In some literatures, characteristic ITD is also called Best Delay (BD [38]) or Characteristic Delay (CD [39]). Both the EEtype and EI-type have supports from physiological research, but the latter explains better the various binaural hearing phenomena [4]. In 1948, Jeffress gave a physiological model for ITD perception [41, 42] delay line model the foundational contribution, having lasting impact in the field (Figure 5). Neural signals in the form of spike train from the left and right auditory pathways meet at some coincidence counter after traveling along the left and right delay lines and trigger the counter, which is in fact a physiological cross-correlation calculator. The specific counter having the largest counts is the counter to which the delay difference along the left and right delay lines exactly compensates the ITD. For example, sounds from the medial plane (ITD = ) generate the largest counts in the middle counter of the Jeffress network. The coincidence counters can be classified as EE-type auditory nerve cells. In 21, Breebaart et al. extended the Jeffress model by incorporating attenuators [43 45] (Figure 6). An important difference to the Jeffress model is the use of EI-type elements instead of the EE-type elements in the Breebaart model. Due to the attenuators, ILD can be extracted by the extended model. In the Breebaart model, only if the internal delay and attenuation are exactly compensated by the external ITD and ILD, the corresponding EI-type elements will have the largest inhibition. Thus, knowing the position of the EI-type element with the largest inhibition, the auditory system finds the ITD and ILD of the external audio signals. The Breebaart model also implies the calculation of Interaural Coherence (IC), which manifests as the trough ΔT ΔT ΔT Figure 6: Breebaart model: delay-attenuation network. Right ear in of the excitation surface, in accordance with the EI-type assumption. Nevertheless, there is no direct physiological quantity related to IC in this model. In 24, Faller and Merimma reported that IC relates to perceiving sound image width and stability, as well as sound field ambience [46, 47]. On the other hand, by the precedence effect [48, 49] of spatial hearing sound source localization depending primarily on the direct sounds to the ears and essentially irrespective to reflection and reverberation which contributes to lowering IC, Faller proposed that our auditory system use ITD and ILD to localize sound sources only if IC approaches 1. Since direct sounds to the ears have near 1 cross-correlation, this explains the precedence effect Binaural Cue Physiological Perception Model (BCPPM). From the viewpoint of the information theory, the channel from the physical layer to the physiological layer is lossy, and less spatial information survives during the course (Figure 7). Since the wavelength ( m) of sound in the audible range (2 2 Hz) is much longer than light, and comparable to normal objects in our surrounding leading to significant interference and diffraction spatial information from hearing is limited initially. This limited information is first compromised by noises and other interferences from other sound sources, as indicated by Δp 1 in Figure 7. Then during transformation from mechanical swing to electric impulses, part of the information is lost again due to the limited frequency range and dynamic range, the limited frequency and temporal resolution, and physiological noises of our auditory system, as is indicated by Δp 2 in Figure 7. The loss of spatial information manifests as offset and disperses, related to multisource interference, limited SNR in the physical and physiological system. For example, sometimes a single source becomes multiple sources of mirrored sound images due to reflection by, say walls and floors. These sources have the same frequency range, so

137 6 EURASIP Journal on Wireless Communications and Networking Sound field Physical layer Physical data Physiological layer Physiological data Offset Disperse Δp 1 Δp 2 Spatial information Figure 7: Spatial information loss. auditory filtering cannot separate them. And the perceived ITD and ILD are determined by the combined effects of BDTFs of those sources, typically leading to biased and vaguer location perception (Figure 8). A large sound source has similar localization effects. In the Breebaart model, the resolution of ITD and ILD is limited by the fineness of the delay elements and attenuation elements: no ITD smaller than the delay offered by one delay element can be detected and no ILD smaller than the attenuation offered by one attenuation element can be detected. This is in analogy to the ATH in monaural hearing. The limited ITD and ILD resolution turns out to limited localization resolution. In Section 1.1, we see that the physical data of sound source localization in binaural hearing are in form of ITD and ILD. In Section 2.1, we see that ITD and ILD are transformed to maximum inhibition of specific EItype auditory nerve cells in the Breebaart model, and the physiological data are in the form of coordinates of the delayattenuation network. When there are multiple sound sources, background noises, reflection, diffraction, and reverberation, IC becomes another type of physical data conveying the overall sound field information. Since spatial hearing on the physiological layer is too complex and uncertainty to be incorporated in computational model for common listeners, we restrict the calculation of perceptible spatial information to that directly related to ITD, ILD, and IC and physiological data corresponding to the three cues. In fact, spatial coding systems use the cues to represent spatial information. We first review the psychoacoustic foundation of PE, mainly the nonlinear frequency resolution (Critical Band, CB [5, 51]) of our hearing system, spreading functions in the frequency domain for noises and tones and tonality estimation. To calculate PE, Johnston used a Monaural Hearing Model (MHM, Figure 9). In this model, a 25-subband filterbank filters incoming audio signals. Each subband has a bandwidth of CB at the corresponding frequency (CB 1 -CB 25 in Figure 9), increasing from low to high frequency. Each subband also acts as a lossy subchannel, and the loss of audio information is due to the intrinsic noises of hearing system (ATH) and interchannel interference (masking effect). ATH is signal dependent, usually as a table or a fitting function of experimental data. Masking is signal dependent, usually obtained by convoluting the tonality-dependent spreading functions with the signal spectra. Combining both, we have effective channel noises (n 1 -n 12 in Figure 9). Real location Perceived location Figure 8: Two types of spatial information loss. Audio signal CB 1 CB 2. CB n n Perception Perception n 25 + Perception + Figure 9: Monaural Hearing Model (MHM) used to calculate PE. There is no place for localization in the MHM. The critical limit of the model is the lack of binaural processing only spectral-temporal information but not spatial information. The Breebaart delay-attenuation network just models the binaural processing. So we borrow the idea of lossy multichannel in MHM and combine MHM with the Breebaart model Binaural Cue Physiological Processing Model (BCPPM, Figure 1). The BCPPM consists of 3 modules. Frequency-to-Place Transform in Cochlea. This process separates sounds into a bank of subband signals, essentially the subband filtering in MHM. The subband filter can be implemented by DFT with spectral lines grouped to subbands according to CB or by the Cochlear Filter Bank (CFB [52]) proposed by Baumgarte in 22. Delay-Attenuation Network. This is the same as that in Figure 6. After the Time-to-Place Transform, external audio signals change into spike trains of auditory nerve signals, which arrive at the corresponding delay-attenuation networks. Then the networks output ITD, ILD, and IC for each critical band. From the location of the maximum inhibition (lowest excitation, the trough of the neural excitation surface in Figure 11), we can derive ITD and ILD. From the gradient of the trough, we can derive IC: faster descending or larger gradient implies larger IC ( 1); slower descending or smaller gradient implies smaller IC ( ).

138 EURASIP Journal on Wireless Communications and Networking 7 Left ear in CB 1 CB 2 CB 25 Delay-attenuation network IC ILD ITD Delay-attenuation network IC ILD ITD. Delay-attenuation network IC ILD ITD CB 1 CB 2 CB 25 Right ear in n ITD n ILD n IC n ITD n ILD n IC n ITD n ILD n IC Figure 1: Binaural Cue Physiological Perception Model (BCPPM). Effective Channel Noises. The effective channel noise for ITD, ILD, and IC (n ITD, n ILD,andn IC in Figure 1) isa simplified method to model the limited precision, intrinsic noises, and intersource interference in our hearing system. Part of the noise comes directly from grains of delay and attenuation (ΔT and ΔL in Figure 6). For example, if ΔT = 1 μs, n ILD 1 μs. Generally, ΔT and ΔL are functions of frequency. A related concept is Just Noticeable Difference (JND) in psychoacoustics, indicating the overall sensitivity of our auditory system. On the other hand, ITD, ILD, and IC are not independent, there are interactions among them. The effective channel noise should also incorporate the interactions. 3. Computing Spatial Perceptual Entropy (SPE) Based on BCPPM In this section, we will define SPE using the BCPPM and then discuss in detail the computational implementation of BCPPM, including 3 core components: the CB filterbank, binaural cues computation, and perceptible information computation (Figure 12) SPE Definition. From the information theory viewpoint, we see BCPPM as a double-in-multiple-out system (Figure 1). The double-in is the left ear entrance sound and the right ear entrance sound. The multiple-out consists of 75 effective ITDs, ILDs, and ICs (25 CBs, each with a tuple of ITD, ILD, and IC). Like in computing PE, we view each path that leads to an output as a lossy subchannel. Then there are 75 such subchannels. Unlike PE, what a subchannel conveys is not Excitation activity (MU) Delay (ms) Attenuation (db) Figure 11: An example of auditory nerve excitaton surface with ITD = ms and ILD = db,adaptedfrom[42]. a subband spectrum but one of ITD, ILD, and IC of the subband corresponding to the sub-channel. In each sub-channel, there are intrinsic channel noises (resolution of spatial hearing), and among sub-channels, there are interchannel interferences (interaction of binaural cues). Then there is an effective noise for each sub-channel. Under this setting, each sub-channel will have a channel capacity. We denote SPE(c), SPE(t), and SPE(l) for the capacity of IC, ITD, and ILD sub-channels respectively. Then SPE is defined as the overall capacity of these sub-channels, or the sum of capacities of all the sub-channels: SPE = SPE(c) +SPE(t) +SPE(l). (4) all subbands To derive SPE(c), SPE(t), and SPE(l), we need probability models for IC, ITD, and ILD. Although the binaural cues are continuous, the effective noise quantizes them into discrete values. Let [L P], [T P], and [C P] denote the discrete ILD, ITD, and IC source probability spaces: L : l 1, l 2,..., l i,..., l NL, [L P] : P(L) : P(l 1 ), P(l 2 ),..., P(l i ),..., P ( ) l NL, T : t 1, t 2,..., t i,..., t NT, [T P] : P(T) : P(t 1 ), P(t 2 ),..., P(t i ),..., P ( ) t NT, C : c 1, c 2,..., c i,..., c NC, [C P] : P(C) : P(c 1 ), P(c 2 ),..., P(c i ),..., P ( ) c NC, where l i, t i,andc i are the ith discrete values of ILD, ITD, and IC, respectively, and p(l i ), p(t i ), and p(c i ) the corresponding probabilities. Then we have N L SPE(l) = p(l i )log 2 p(l i ), (6) i=1 i= (5) N T SPE(t) = p(t i )log 2 p(t i ), (7) N c SPE(c) = p(c i )log 2 p(c i ). (8) i=1

139 8 EURASIP Journal on Wireless Communications and Networking Table 2: Critical Bands for 248-point DFT, sampling frequency 48 khz [4]. CB Index Frequency Range (Hz) Spectral Index CB Index Frequency Range (Hz) CB Index For some probability distributions, say uniform distribution, (5), (6), and (7) can be readily calculated CB Filterbank. We use the same method as that in PE to implement the CB filterbank. Audio signals are first transformed to the frequency domain by DFT of 248 points with 5% overlap between adjacent transform blocks. Then a DFTspectrumispartitionedinto25CBsaccordingtoTable 2 [41]. Then basic processing unit is the subspectra of each CB Binaural Cues Computation. ILD is the ratio of left ear entrance signal intensity to right ear entrance signal intensity. Since DFT preserves signal energy, we can use DFT subspectra energy ratio to compute ILD on each CB [53]: ILD(b) = 2log 1 kb+1 1 k=k b X l (k) 2 kb+1 1 k=k b X, (9) r (k) 2 where b is the indexes of CB, k b and k b+1 the starting DFT spectral index of CB b and CB b+1 (Table 2), X l (k) and X r (k)the kth spectral lines from left and right ear entrance signals. Time shift corresponds to linear phase shift in the frequency domain. Therefore, we can use group delay (slope of phase-frequency curve) of subband signal to derive ITD on each subband: ITD(b) = 1 k b+1 1 ( arg Xl (k +1) arg X l (k) ) w b k=k b 1 k b+1 1 ( arg Xr (k +1) arg X r (k) ), w b k=k b (1) where w b = k b+1 k b is the bandwidth of CB b,andarg represents the phase of a complex number. A more reliable but also more complex method is to use least square fitting to find the group delays and then ITD: ITD(b) = w b k arg Xl (k) k arg X l (k) w b k 2 ( k) 2 w b k arg Xr (k) k (11) arg X r (k) w b k 2 ( k) 2. The summation range, k b to k b+1 1, is left out for simplicity. Due to the property that time-domain normalized correlation is equivalent to the real part of correlation in the frequency domain, IC of each CB can be derived as the following: IC(b) = Re { X l (k)x r (k) } Xl (k) 2 Xr (k) 2, (12) where the summation range is also k b to k b+1 1, and represents conjugate Effective Spatial Perception Data. The resolutions or quantization steps of the binaural cues (Figure 12) can be determined by JND experiments. Denote by Δτ, Δλ,andΔη the resolutions of ITD, ILD, and IC, respectively. Generally, they are signal dependent and frequency dependent. For simplicity, we use constant values [44, 54]: Δτ =.2 ms, Δλ = 1dB,andΔη =.1. IC has different impacts on ITD and ILD perception. In 21, Hartmann Constan reported that the difference of JND of ILD for correlated noises and uncorrelated noises is only.5 db [55]. This can be explained by the fact that signal power is independent phase, which influences correlation, and lower IC is partly the result of increasing phase noise. This is illustrated in Figure 13: when IC decreases, the gradient along the ILD axis keeps almost unchanged, but the gradient along the ITD significantly decreases. Larger IC usually implies higher ITD perception precision or equivalently morespatial information. When IC

140 EURASIP Journal on Wireless Communications and Networking 9 Resolution of spatial hearing Left Right CB filterbank CB filterbank Binaural cues computation IC 1 IC Δη ITD ILD Δτ = Δτ IC ITD Δτ ILD Δλ Effective information SPE(c) SPE(t) SPE(l) Figure 12: SPE calculation. Attenuation (db) Delay (ms) IC Delay (ms) Figure 13: The different effects of IC on ITD and ILD perception. approaches 1, the activity surface will have a very sharp decreasing toward the point with the lowest auditory nerve activity. In this case, the uncertainty of ITD is very small and is determined precisely. When IC decreases to, the surface becomes flatter, leading to larger uncertainty or lower precision of ITD. In the extreme case, when IC =, the gradient along the IC axis will be constantly, there is no well defined trough point and ITD is completely indeterminable. By the above analysis, we ignore the effectoficon ILD and only consider the effect of IC on ITD for SPE computation. Lower IC leads to lower resolution of ITD. This is equivalent to higher JND of ITD. Then the effective JND on subband b, denotedasδτ (b), can be formulated as the following: Attenuation (db) Δτ (b) = Δτ(b) IC(b). (13) From (13) we see that when IC(b)=1, Δτ (b) assumes the minimum Δτ(b) and the auditory system has the highest resolution for ITD; when < IC(b) < 1, Δτ(b) < Δτ (b) <, the resolution of ITD is lower but there is still spatial information from ITD; when IC(b) = 1, Δτ (b) =, the resolution of ITD is and there is no spatial information in ITD. Then we have the following effective perception data q ILD (b), q ITD (b), and q IC (b) of ILD, ITD, and IC, respectively by quantization: ILD(b) q ILD (b) = 2 Δλ(b), ITD(b) q ITD (b) = 2 Δτ(b)/IC(b), (14) 1 IC(b) q IC (b) =, Δη(b) where represents the round down function. Suppose that q ILD (b), q ITD (b), and q IC (b) are uniformly distributed by (6), (7), and (8), the SPE of IC, ITD, and ILD are SPE(c) = 1 25 ( ) 1 IC(b) α log N 2 +1, Δη(b) b=1 SPE(t) = 1 25 ) ITD(b) α log N 2 (2 Δτ(b)/IC(b) +1, b=1 SPE(l) = 1 N 25 b=1 ) ILD(b) α log 2 (2 Δλ(b) +1, (15) where N is the number of spectral lines in one transform, or 124 in this case; ILD(b), ITD(b), and IC(b) can be found from (9), (1), and (11), respectively; Δλ(b), Δτ(b), and Δη(b) are the JNDs of ILD, ITD, and IC on CB b, respectively, obtained from subjective listening experiments; and α is the amplitude compression factor, assuming.6 [5]. 4. Experiments We evaluate SPE of 126 stereo sequences from 3GPP and MPEG, which are classified into speech, single instrument, simple mixture, and complex mixture, all sampled at 44.1 khz. For comparison, we also evaluate PE of these sequences.

141 1 EURASIP Journal on Wireless Communications and Networking i = i +1 b = b +1 Pre-processing unit CB filterbank unit Binaural cues computation Effective perception data computation Channel separation Left Right i>overall frame count No Spectral energy computation Spectral power Stereo audio signal Framing Windowing FFT CB partition b>25 No ITD, ILD, IC estimation Subband SPE(c), SPE(t), and SPE(l) Spectrum SPE SPE over frames SPE averaging SPE Normalised correlation Figure 14: Flowchart of SPE Computation. Figure 14 gives the computational procedure of SPE: stereo audio signals are windowed and block transformed to the frequency domain using 248-point DFT; then on the 25 CBs, binaural cues are derived before transformed into effective spatial perception data, the entropy of which is SPE. In the following experiments, Δτ(b), Δλ(b), and Δλ(b) assume constant and conservative values, and their frequency dependency is also ignored. The overall SPE is the sum of entropy of effectiveic,ild,anditdperceptiondata,shown in (4) Perceptual Spatial Information of Stereo Sequences. In this experiment, we compute perceptual spatial information by SPE for 4 classes of stereo sequences(figure 15): each class consists of 12 sequences, sampled at 44.1 khz; each data point is average of SPE over one sequence, measured by kbps. From Figure 15 we find that speech sequences generally have the lowest spatial information rate, mean 2.75 kbps, this is in accordance with the recording practice that voices usually stay in direct front of the sound field; single instrument Yes Yes SPE (kbps) Seq index Speech, mean = 2.75 Simple mixture, mean = 3.66 Complex mixture, mean = 6.9 Single instrument, mean = 3.49 Figure 15: Perceptual spatial information of stereo sequences sampled at 44.1 khz. sequences and simple mixture sequences have similar spatial information rate, mean 3.49 kbps and 3.66 kbps, respectively; complex mixture sequences generally have the highest spatial information rate, mean 6.9 kbps, this can be explained by multiple sound sources at diverse sound field locations in this type of sequences. In Parametric Stereo (PS [56]) coding, it is reported that 7.7 kbps of spatial parameter bitrate is sufficient for transparent spatial audio quality, agreeing very well with our SPE computation Temporal Variation of Spatial Information Rate in a Single Senescence. In this experiment, we choose two sequences es2 of German male speech and sc3 of contemporary pop music from MPEG and compute their SPE frame by frame (Figure 16). The test data show that for es2 with stable voice from the front, SPE stays at 1-2 kbps; for sc3 with multiple instruments and strong spatial impression, SPE stays at about 7 kbps. But within either sequence, the SPE changes little Overall Perceptual Information in Stereo Sequences. Using PE to evaluate the perceptual information, only intrachannel redundancy and irrelevancy are exploited; the overall PE is simply the sum of PE of the left and right channels. Using SPE based on BCPPM, interchannel redundancy and irrelevancy are also exploited; the overall perceptual information is about one normal audio channel plus some spatial parameters, which has significantly lower bitrate. For the above reason, PE gives much higher bitrate bound than SPE (Figure 17). PE is compatible with the traditional perceptual coding schemes, such as MP3 and AAC, in which channels are basically processed individually (except the mid/side stereo and the intensity stereo). So PE gives meaningful bitrate bound for them. But in Spatial Audio Coding (SAC [52, 54, 57 59]), multichannel audio signals are processed as one or two core channels plus spatial 4.11

142 EURASIP Journal on Wireless Communications and Networking 11 (a) (b) SPE (kbps) Time (s) (c) SPE (kbps) Time (s) (d) Figure 16: SPE of es2 (speech) and sc3(pop). (a): waveform of es2; (b): SPE curve of es2; (c): waveform of sc3; (d): SPE curve of sc3. Overall information rate (bits/sample) Simple mixture PE + PE PE + SPE 6 Complex mixture Speech Single instrument Figure 17: Perceptual Information of stereo sequences sampled at 44.1 khz, evaluated using PE and SPE. parameters. SPE is necessary in this case and generally gives much lower bitrate bound ( 1/2). This agrees to the sharp bitrate reduction of SAC. 5. Conclusion We have developed the Binaural Cues Physiological Perceptual Model (BCPPM) to measure the perceptible information, or Spatial Perceptual Entropy (SPE), in multichannel audio signals and have given a lower bitrate bound in multimedia communications for this type of contents. BCPPM models the physical and physiological processing of human spatial hearing into a parallel of lossy communication subchannels with inter-subchannel interference, and SPE is the overall channel capacity. Each of these subchannels carries ITD, ILD, or IC with addictive noises, resulted from intrinsic noises of binaural cues perception and interferences among the cues within the same CB. Experiments on stereo signals of different types have confirmed that SPE is compatible with the spatial parameter bitrate and spatial impression in SAC. Nevertheless, SPE gives only the lower bitrate bound for transparent quality. We will extend SPE to give the bound for given subjective quality in the future. Then in mobile, internet, and other communications networks conveying multichannel audio signals, we can use the estimated bound to allocate bandwidth for a particular Quality of Service (QoS), transparent or degraded and thus save bandwidth or improve the overall QoS. On the other hand, current SAC may benefit from SPE dynamically allocating bitrate to accommodate varying spatial contents thus improving quality and reducing overall bitrate. Acknowledgment This research is supported by the National Science Foundation of China Grant no

143 12 EURASIP Journal on Wireless Communications and Networking References [1] C. E. Shannon, A mathematical theory of communication, Bell System Technical Journal, vol. 27, pp , , [2] Lossless comparison, title=lossless comparison. [3] T. Painter and A. Spanias, Perceptual coding of digital audio, Proceedings of the IEEE, vol. 88, no. 4, pp , 2. [4]E.ZwickerandH.Fastl,Psychoacoustics Facts and Models, Berlin, Germany, Springer, 199. [5]B.C.J.Moore,An Introduction to the Psychology of Hearing, Elsevier Academic Press, London, UK, 5th edition, 23. [6] E. Zwicker and U. T. Zwicker, Audio engineering and psychoacoustics. Matching signals to the final receiver, the human auditory system, Journal of the Audio Engineering Society, vol. 39, no. 3, pp , [7] J. L. Hall, Auditory psychophysics for coding applications, in The Digital Signal Processing Handbook, V. Madisetti and D. Williams, Eds., pp , CRC Press, Boca Raton, Fla, USA, [8] B. C. J. Moore, Masking in the human auditory system, in Collected Papers on Digital Audio Bit-Rate Reduction, N. Gilchrist and C. Grewin, Eds., pp. 9 19, Audio Engineering Society, New York, NY, USA, [9] ISO/IEC JTC1/SC29/WG11, Information Technology Generic Coding of Moving Pictures and Associated Audio Information Part 7: Advanced Audio Coding (AAC), ISO/IEC , 25. [1] ISO/IEC JTC1/SC29/WG11, Information Technology Generic Coding of Moving Pictures and Associated Audio Information Part 3: Audio, Subpart 4: General Audio Coding, ISO/IEC , 25. [11] M. Bosi and R. E. Goldberg, Introduction to Digital Audio Coding and Standards, Kluwer Academic Publishers, Boston, Mass, USA, 23. [12] J. D. Johnston, Transform coding of audio signals using perceptual noise criteria, IEEE Journal on Selected Areas in Communications, vol. 6, no. 2, pp , [13] J. D. Johnston, Estimation of perceptual entropy using noise masking criteria, in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 88), pp , May [14] 3GPP, Mandatory speech CODEC speech processing functions; AMR speech Codec; General description, 3GPP TS 26.71, 28, htm. [15] 3GPP, Speech codec speech processing functions; Adaptive Multi-Rate Wideband (AMR-WB) speech codec; General description, 3GPP TS , 28, Specs/html-info/26171.htm. [16] ISO/IEC and JTC1/SC29/WG11 MPEG, Information technology coding of moving pictures and associated audio for digital storage media at up to about 1.5 Mbit/s part 3: audio, ISO/IEC , [17] J. Blauert, Spatial Hearing: The Psychophysics of Human Sound Localization, MIT Press, Cambridge, Mass, USA, [18] P. M. Hofman, J. G. A. Van Riswick, and A. J. Van Opstal, Relearning sound localization with new ears, Nature Neuroscience, vol. 1, no. 5, pp , [19] J. W. Strutt, On our perception of sound direction, Philosophical Magazine, vol. 13, pp , 197. [2] E. A. Macpherson and J. C. Middlebrooks, Listener weighting of cues for lateral angle: the duplex theory of sound localization revisited, Journal of the Acoustical Society of America, vol. 111, no. 5, pp , 22. [21] J. Blauert, Sound localization in the median plane, Acustica, vol. 22, no. 4, pp , [22] J. Hebrank and D. Wright, Spectral cues used in the localization of sound sources on the median plane, Journal of the Acoustical Society of America, vol. 56, no. 6, pp , [23] R. A. Butler and K. Belendiuk, Spectral cues utilized in the localization of sound in the median sagittal plane, Journal of the Acoustical Society of America, vol. 61, no. 5, pp , [24] B. Rakerd, W. M. Hartmann, and T. L. McCaskey, Identification and localization of sound sources in the median sagittal plane, Journal of the Acoustical Society of America, vol. 16, no. 5, pp , [25] A. D. Musicant and R. A. Butler, The influence of pinnaebased spectral cues on sound localization, Journal of the Acoustical Society of America, vol. 75, no. 4, pp , [26] F. Asano, Y. Suzuki, and T. Sone, Role of spectral cues in median plane localization, Journal of the Acoustical Society of America, vol. 88, no. 1, pp , 199. [27] H.Møller,M.F.Sørensen,D.Hammershøi,andC.B.Jensen, Head-related transfer functions of human subjects, Journal of the Audio Engineering Society, vol. 43, no. 5, pp , [28] H. Møller, Fundamentals of binaural technology, Applied Acoustics, vol. 36, no. 3-4, pp , [29] Spatial hearing, in Audio Signal Processing for Next- Generation Multimedia Communication Systems, Y. Huang and J. Enesty, Eds., chapter 13, section 2, pp , Kluwer Academic Publishers, Norwell, Mass, USA, 24. [3] W. G. Gardner and K. D. Martin, HRTF measurements of a KEMAR, Journal of the Acoustical Society of America, vol. 97, no. 6, pp , [31] V. R. Algazi, R. O. Duda, D. M. Thompson, and C. Avendano, The CIPIC HRTF database, in Proceedings of IEEE Workshop on Applications of Signal Processing to Audio and Electroacoustics, pp , New Paltz, NY, USA, October 21. [32] D. D. Greenwood, A cochlear frequency-position function for several species: 29 years later, Journal of the Acoustical Society of America, vol. 87, no. 6, pp , 199. [33] D. D. Greenwood, Critical bandwidth and the frequency coordinates of the basilar membrane, Journal of Acoustic Society America, vol. 33, no. 1, pp , [34] G. von Bèkèsy, Experiments in Hearing, McGraw Hill, New York, NY, USA, 196. [35] A. R. Møller, Hearing: Anatomy, Physiology, and Disorders of the Auditory System, Academic Press, Burlington, Vt, USA, 2nd edition, 26. [36] J. E. Rose, N. B. Gross, C. D. Geisler, and J. E. Hind, Some neural mechanisms in the inferior colliculus of the cat which may be relevant to localization of a sound source, Journal of Neurophysiology, vol. 29, no. 2, pp , [37] T. J. Park, IID sensitivitydiffers between two principal centers in the interaural intensity difference pathway: the LSO and the IC, Journal of Neurophysiology, vol. 79, no. 5, pp , [38] P. X. Joris, B. Van de Sande, D. H. Louage, and M. van der Heijden, Binaural and cochlear disparities, Proceedings of the National Academy of Sciences of the United States of America, vol. 13, no. 34, pp , 26.

144 EURASIP Journal on Wireless Communications and Networking 13 [39] R. M. Stern, DeL. Wang, and G. Brown, Binaural sound localization, in Computational Auditory Scene Analysis, G. Brown and DeL. Wang, Eds., Wiley/IEEE Press, New York, NY, USA, 26. [4] J. Breebaart, S. van de Par, and A. Kohlrausch, The contribution of static and dynamically varying ITDs and IIDs to binaural detection, Journal of the Acoustical Society of America, vol. 16, no. 2, pp , [41] L. A. Jeffress, A place theory of sound localization, Journal of Comparative and Physiological Psychology, vol. 41, no. 1, pp , [42] P. X. Joris, P. H. Smith, and T. C. T. Yin, Coincidence detection in the auditory system: 5 years after Jeffress, Neuron, vol. 21, no. 6, pp , [43] J. Breebaart, S. van de Par, and A. Kohlrausch, Binaural processing model based on contralateral inhibition. I. Model structure, Journal of the Acoustical Society of America, vol. 11, no. 2, pp , 21. [44] J. Breebaart, S. D. van de Par, and A. Kohlrausch, Binaural processing model based on contralateral inhibition. II. Dependence on spectral parameters, Journal of the Acoustical Society of America, vol. 11, no. 2, pp , 21. [45] J. Breebaart, S. D. van de Par, and A. Kohlrausch, Binaural processing model based on contralateral inhibition. III. Dependence on temporal parameters, Journal of the Acoustical Society of America, vol. 11, no. 2, pp , 21. [46] C. Faller and J. Merimaa, Source localization in complex listening situations: selection of binaural cues based on interaural coherence, Journal of the Acoustical Society of America, vol. 116, no. 5, pp , 24. [47] M. J. Goupell and W. M. Hartmann, Interaural fluctuations and the detection of interaural incoherence: bandwidth effects, Journal of the Acoustical Society of America, vol. 119, no. 6, pp , 26. [48] P. M. Zurek, The precedence effect, in Directional Hearing, W. A. Yost and G. Gourevitch, Eds., pp , Springer, New York, NY, USA, [49] R. Y. Litovsky, B. Rakerd, T. C. T. Yin, and W. M. Hartmann, Psychophysical and physiological evidence for a precedence effect in the median sagittal plane, Journal of Neurophysiology, vol. 77, no. 4, pp , [5] H. Fletcher, Auditory patterns, Reviews of Modern Physics, vol. 12, no. 1, pp , 194. [51] B. Scharf, Critical bands, in Foundations of Modern Auditory Theory, Academic Press, New York, NY, USA, 197. [52] C. Faller and F. Baumgarte, Binaural cue coding part II: schemes and applications, IEEE Transactions on Speech and Audio Processing, vol. 11, no. 6, pp , 23. [53] F. Baumgarte, Improved audio coding using a psychoacoustic model based on a cochlear filter bank, IEEE Transactions on Speech and Audio Processing, vol. 1, no. 7, pp , 22. [54] J. Breebaart, J. Herre, C. Faller, et al., MPEG spatial audio coding/mpeg surround: overview and current status, in AES 119th Convention, New York, NY, USA, October 25. [55] W. M. Hartmann and Z. A. Constan, Interaural coherence and the lateralization of noise by interaural level differences, Journal of the Acoustical Society of America, vol. 11, no. 5, p. 268, 21. [56] J. Breebaart, S. van de Par, A. Kohlrausch, and E. Schuijers, Parametric coding of stereo audio, EURASIP Journal on Applied Signal Processing, vol. 25, no. 9, pp , 25. [57] J. Rödén, J. Breebaart, J. Hilpert, et al., A study of the MPEG surround quality versus bit-rate curve, in AES 123rd Convention, New York, NY, USA, October 27. [58] J. Breebaart, G. Hotho, J. Koppens, E. Schuijers, W. Oomen, and S. van de Par, Background, concept, and architecture for the recent MPEG surround standard on multichannel audio compression, Journal of the Audio Engineering Society, vol. 55, no. 5, pp , 27. [59] J. Hilpert and S. Disch, The MPEG surround audio coding standard, IEEE Signal Processing Magazine, vol.26,no.1,pp , 29.

145 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 11 pages doi:1.1155/21/ Research Article Novel Approaches to Enhance Mobile WiMAX Security Taeshik Shon, 1 Bonhyun Koo, 1 Jong Hyuk Park, 2 and Hangbae Chang 3 1 Convergence S/W Laboratory, DMC R&D Center, Samsung Electronics, Dong Suwon P.O. Box 15, Maetan-3dong, Suwon-si, Gyeonggi-do, 442-6, Republic of Korea 2 Department of Computer Science and Engineering, Seoul National University of Technology, 172, Gongreung 2-dong, Nowon, Seoul , Republic of Korea 3 Department of Business Administration, Daejin University, San 11-1, Sundan-Dong, Pocheon-Si, Gyunggi-Do , Republic of Korea Correspondence should be addressed to Hangbae Chang, hbchang@daejin.ac.kr Received 26 February 21; Accepted 5 July 21 Academic Editor: Liang Zhou Copyright 21 Taeshik Shon et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The IEEE Working Group on Broadband Wireless Access Standards released IEEE which is a standardized technology for supporting broadband and wireless communication with fixed and nomadic access. After the IEEE standard, a new advanced and revised standard was released as the IEEE 82.16e-25 amendment which is foundation of Mobile WiMAX network supporting handover and roaming capabilities. In the area of security aspects, compared to IEEE , IEEE 82.16e, called Mobile WiMAX, adopts improved security architecture PKMv2 which includes EAP authentication, AESbased authenticated encryption, and CMAC or HMAC message protection. However, there is no guarantee that PKMv2-based Mobile WiMAX network will not have security flaws. In this paper, we investigate the current Mobile WiMAX security architecture focusing mainly on pointing out new security vulnerabilities such as a disclosure of security context in network entry, a lack of secure communication in network domain, and a necessity of efficient handover supporting mutual authentication. Based on the investigation results, we propose a novel Mobile WiMAX security architecture, called RObust and Secure MobilE WiMAX (ROSMEX), to prevent the new security vulnerabilities. 1. Introduction More and more, our life is closely related to a variety of networking environments for using Internet-based services and applications. The ever-changing trends of our lifestyle require faster speed, lower cost, more broadband capacity, as well as nomadic and mobility support. Due to these reasons and demands, IEEE working group has created new standards with mobility access called the IEEE 82.16e-25 amendment. It has also been developed by many working groups of the Worldwide Interoperability for Microwave Access (WiMAX) Forum, similar to Wi- Fi in IEEE standards. The WiMAX Forum tries to coordinate the interoperability and compatibility of various company products as a field standard. Specifically, Mobile WiMAX technology is considered as one of the best next-generation wireless technologies because it can support high-speed, broadband data transmission, fullysupported mobility, and wide coverage and high capacity [1 4]. From a security viewpoint, the Mobile WiMAX system based on the IEEE 82.16e-25 amendment has more enhanced security features than the existing IEEE based WiMAX network system. The improved core part of the security architecture in Mobile WiMAX, called PKM v2, is operated as a security sublayer in a MAC layer like PKMv1 in IEEE The PKMv2 provides a message authentication scheme using HMAC or CMAC, device/user authentication using EAP methods, and confidentiality using AES-CCM encryption algorithm [5, 6]. Even though Mobile WiMAX uses more enhanced security schemes supported by PKMv2, it can not guarantee the reliability of the whole Mobile WiMAX systems and network architectures. In addition, open architecture and various applications of Mobile WiMAX could cause much more risks to try to compromise Mobile WiMAX network than

146 2 EURASIP Journal on Wireless Communications and Networking existing systems. In Mobile WiMAX, the network domain consists of a link domain between Subscriber Station (SS) and the Base Station (BS), access network domain, and mobility domain. Each network domain has a possibility of potential risks. In case of a link domain, the Mobile WiMAX does not support any security features to authenticate peers and encrypt initial entry control and user data. In access network domain of Mobile WiMAX, it only provides a regular guideline for protecting inter-network data based on IP security, even it is not a scope of IEEE 82.16e. In addition, a handover which is one of the most distinguished features in Mobile WiMAX is left alone without security functionalities. Specifically, the problem is very critical when the handover is supporting fast handover optimization option. Therefore this paper focuses on three kinds of security vulnerabilities and their countermeasure according to each Mobile WiMAX network domain. Finally, we present security architecture of Mobile WiMAX network as called Robust and Secure MobilE WiMAX (ROSMEX). The rest of this paper is organized as follows. In Section 2, we study an overview of Mobile WiMAX security and analyze known security vulnerabilities and attacks. In Section 3, new security threats and the related works in Mobile WiMAX network are examined. In Section 4, we propose possible solutions in order to cope with the new threats we mention in Section 3. In Section 5, the comparison and analysis of the proposed approaches with the current approaches are presented. In Section 6, we discussed a reliable Mobile WiMAX architecture including our proposed solutions. In the last section, we conclude with a summary and discussion of future work. 2. Background The first stage of IEEE standard was released in 24; many researchers have tried to analyze the new standard s vulnerabilities and deal with possible attacks. In this section, we describe an overview of security features in Mobile WiMAX. Moreover, this paper analyzes the known existing security vulnerabilities and attacks [7 1] Overview of Mobile WiMAX Security. IEEE 82.16e-25 amendment-based Mobile WiMAX supports many good security features as compared to the fixed IEEE based WiMAX security schemes. Basically, the Privacy Key Management sublayer in the MAC layer of IEEE is a core part which comprises the WiMAX security. The PKM sublayer provides not only key related management functions but also strong protection for encrypting traffic and EAP-based flexible authentication for accessing valid users and devices. In the Mobile WiMAX system, more enhanced PKMv2 is supported, together with various cryptographic suites. In the research of [3, 4] from WiMAX Forum, the security features of the PKMv2-based Mobile WiMAX consist of Key Management Protocol, Device and User Authentication, Traffic Encryption/Decryption, Control Message Authentication, Hard Handover, and IP Mobility Support. PKMv2-based key management protocol manages and maintains various keys for EAP authentication, message authentication, traffic encryption, handover (Authentication Key transfer), and multicast/broadcast security Known Vulnerabilities and Attacks. The security architecture of Mobile WiMAX is partially originated from wireless networks based on IEEE In the case of IEEE based wireless networks, a great deal of security-related research has already been studied, and a few vulnerabilities have been known as those in [7, 8]. Among many interesting researches, John Bellardo and Stefan Savage s research [7] showed a possibility of Denial of Service attacks using identity vulnerability and Media Access Control vulnerability in MAC layer of IEEE at the USENIX conference. In this section, we investigate well-known vulnerabilities based on the IEEE network architecture from the existing researches [9, 1]. In the case of an attack using Auth Invalid vulnerability, Auth Invalid event is internally generated by the SS when there is a failure authenticating a Key Reply or Key Reject message, or externally generated by the receipt of an Auth Invalid message sent from BS to SS. If SS sends Key Request message with unauthenticated MAC code, BS responds to Key Request with Auth Invalid. Thus, when SS receives the Auth Invalid message, SS will transit from Authorized state to Reauth Wait state, and SS will wait there until SS gets something new from BS. If the Reauth Wait time is expired before SS receives something from BS, SS sends a Reauth Request in order to get into the network again. AlsowhileSSisinReauthstate,SSmayreceiveanAuth Reject message. This is a Permanent Authorization Failure. When SS receives such a message he is pushed into silent state, ceases all subscriber traffic, and will be ready to respond to any management message sent by BS. This way the attacker is able to manipulate the Authorization State Machine. Moreover, the Auth Invalid message is not safe and is easy to modify because HMAC- or CMAC- based message authentication is not provided and PKM identifier is not included. Auth Invalid contains only the error code identifying the reason and the display string describing the failure condition. Even better for attackers, this message s error code provides stateless Auth Invalid with unsolicited property. Finally, in a security vulnerability known as a Rogue BS attack, SS can be compromised by a forged BS. At this time, SS maybe believe he is connected to the real BS. Thus, the forged BS can intercept SS s whole information. In other words, the rogue BS attack is a kind of Man-In-The-Middle attack which is one of the well-known attacks in wireless networks. In IEEE using PKMv1, Auth Request message includes only the contents for SS authentication itself without correspondent BS s authentication. When SS tries to establish a connection to BS, there is no way to confirm whether the BS is authorized or not. The authorization process based on RSA authentication protocol allows only BS to authenticate SS in PKMv1. Thus, it is possible to masquerade as a Rogue BS after sniffing Auth-related message from SS. However, in the case of Mobile WiMAX

147 EURASIP Journal on Wireless Communications and Networking 3 using PKMv2, it is difficult to use the Rogue BS vulnerability because mutual authentication function between SS and BS is mandatory during authorization process. In authorization state, the mutual authentication has two modes. In one mode, RSA-based mutual authentication is used for only mutual authentication. In the other mode, mutual authentication is followed by EAP authentication during initial entry process Related Works. Recently there are a lot of Mobile WiMAX researches and related to its security. In [11], the authors focused on the EAP-based security approach when a Mobile WiMAX user wants to get a handover service. The possible solutions for secure handover in IEEE 82.16e networks are proposed, and the handover protocol guarantees a backward/forward secrecy while giving little burden over the previous researched handover protocols. However, the proposed approaches are not considered the overhead of EAP authentication procedure according to frequent handover. If the frequent handover is occurred, the preauthentication mechanism-based PKMv2 is closely coupled to system performance. In [12 14], the research reviewed the study of WiMAX and converged network and security considerations for both the technologies. They presented many security issues and vulnerability in Mobile WiMAX and then proposed possible solutions. In addition, the papers discuss all the security issues in both point-tomultipoint and mesh networks and their solutions. Some performance-related researches are studied. In [15], the authors analyzed the performance effect when RSA and ECC algorithms are used in WiMAX. However, the research is only evaluating the result, not proposing new approach using ECC-based cryptographic approach. Moreover, the initial network procedure in Mobile WiMAX is not effectively secured that makes man-in-the-middle attack possible. In [16], Diffie-Hellman (DB) key exchange protocol enhance the security level during network initialization. The modified DH key exchange protocol is fit into mobile WiMAX network to eliminate existing weakness in original DH key exchange protocol. But, it can cause additional overhead to distribute initial DH random number, and it can not present the concrete modified DH scheme. 3. New Security Threat 3.1. Initial Network Entry Vulnerabilities. According to IEEE 82.16e-25 standard, the Mobile WiMAX network performs initial Ranging process, SS Basic Capability (SBC) negotiation process, PKM authentication process, and registration process during initial network entry as illustrated in Figure 1. Initial network entry is one of the most significant processes in Mobile WiMAX network because the initial network entry process is the first gate to establish a connection to Mobile WiMAX. Thus, many physical parameters, performance factors, and security contexts between SS and BS are determined during the process. However, specifically, the SBC negotiation parameters and PKM security contexts do not have any security measures to keep their confidentiality. So, the possibility of exposure to malicious users or outer network always exists in initial network entry process. Even though Mobile WiMAX has a message authentication scheme using HMAC/CMAC codes and traffic encryption scheme using AES-CCM based on PKMv2, the security schemes are only applied to normal data traffic after initial network entry process not to control messages during initial network entry. Therefore, it is necessary to prepare a solution to protect important messages such as security negotiation parameters in SBC messages and security contexts in PKM messages during initial network entry Access Network Vulnerabilities. The WiMAX Forum defined Network Reference Model (NRM) which can accommodate the requirements of WiMAX End-to-End Network Systems Architecture [4] for Mobile WiMAX network. The NRM is a logical representation of Mobile WiMAX architecture consisting of the following entities: SS, Access Service Network (ASN), and Connectivity Service Network (CSN). SS means one of the mobile devices that would like to join Mobile WiMAX network. ASN is a complete set of network functions needed to provide radio access to Mobile WiMAX subscribers. ASN consists of at least one BS and one ASN Gateway (ASN/GW). Also, CSN is a set of network functions that provide IP connectivity services to the WiMAX subscribers. CSN consists of AAA Proxy/Server, Policy, Billing, and Roaming Entities. Basically, Mobile WiMAX architecture originated from the IEEE standards. At the view point of NRM, IEEE standards only define a set of functions between SS and BS. It means that the security architecture given by IEEE standards does not cover intra-asn and ASN-to-CSN. In Figure 2, we are able to distinguish a secure domain covered by IEEE standard and insecure domains required additional security services. In the case of communication range between SS and BS, the exchange of messages during network entry process (by the end of registration process) is belonging to insecure domain A. The security threat of insecure domain A is already mentioned in Section 3.1, and a possible solution will be described in Section 4. On the other hand, the communication range after network entry (at the beginning of normal data traffic) belongs tothe securedomain because it can be protected by TEK encryption scheme and message authentication function using HMAC/CMAC. Thus, there remain two insecure domains: insecure domain B between BS and ASN/GW and insecure domain C between ASN and CSN. The reason we called the areas insecure domains is because Network Working Group in WiMAX Forum just assumes that the insecure domain B as illustrated in Figure 2 is a trusted network without suggesting any security function [4]. Moreover, in the case of insecure domain C, the research of [4] only mention a possibility of applying an IPSec tunnel between ASN and AAA (in CSN) [4]. Therefore, in order to make a more robust Mobile WiMAX network, more concrete and efficient countermeasures are needed Handover Vulnerabilities. Mobile WiMAX supports a variety of handover methods for mobile access. There are three handover methods supported within the IEEE 82.16e- 25 amendment: Hard Handover (HHO), Fast Base Station

148 4 EURASIP Journal on Wireless Communications and Networking Initial network entry SS BS ASN/GW AAA UL-MAP (initial ranging codes) Selected ranging code RNG-RSP Insecure communication RNG-REQ RNG-RSP SBC-REQ (Security negotiation parameters) SBC-RSP PKM-RSP (PKMv2 EAP-transfer) (EAP request/identity) PKM-REQ (PKMv2 EAP-transfer) (EAP response/identity - NAI) SBC negotiation Context-request Context-report Authrelay-EAP-transfer Authrelay-EAP-transfer Authentication procedure DER PKM-RSP (PKMv2 EAP-transfer) (EAP success) PKMv2 SA-TEK-3way handshake (security contexts) Context-report Context-request-Ack DEA PKM-REQ (PKMv2 key request) PKM-RSP (PKMv2 key reply) Secure communication Figure 1: Overview of initial network entry procedure. An overview of access network security SS BS ASN/GW AAA Ranging SBC negotiation Authentication and key exchange Registration Insecure domain A Insecure domain B Insecure domain C Data communications Secure domain Figure 2: Overview of access network security.

149 EURASIP Journal on Wireless Communications and Networking 5 Switching, and Macro Diversity Handover. Of these, the HHO is the only mandatory one in Mobile WiMAX. Especially, HO process optimization flags are supported for providing seamless mobility service. The HO optimization flag consists of eight kinds of optimization options and are used as an aim to shorten a network re-entry process when occurring handover. Among the HO optimization flags, PKM authentication phase (HO optimization flag #1) and PKM TEK creation phase (HO optimization flag #2) are related to Mobile WiMAX security as illustrated in Figure 3. If these two flags are used, PKM authentication phase and TEK creation phase do not occur during the re-entry process in handover. Therefore, even though HO optimization flag #1 and #2 are necessary to fast handover to decrease HO latency for real-time services, these flags can give a chance to cause critical security holes to malicious users like a lack of valid entity authentication and man-in-the-middle attack. Supporting security-related HO optimization, flags are tradeoff between handover performance and secure communication. Thus, a possible alternative is required which can cope with the security vulnerability and does not interrupt seamless mobility service during handover. 4. Proposed Countermeasures 4.1. Approach to Initial Network Entry Vulnerabilities. Although much significant information is exchanged during initial network entry, there are not appropriate methods to protect critical information in the entry process. In order to eliminate the security vulnerability during initial network entry, this paper applies Diffie-Hellman (DH) key agreement scheme [17] to initial ranging procedure. Basically, DH key agreement is to share an encryption key with global variables known as prime numbers p and g a primitive root of p. However, the original DH scheme has a threat of Manin-the-Middle attack. Thus, in this paper, we suggest a kind of modified DH scheme using hash authentication. In a ranging process, one of the ranging codes is used as a prime number seed, and then hash authentication is applied to the exchanging process for protecting Man-in-the-Middle attack. In Figure 4, initial ranging procedure is started when SS receives UL-MAP message including ranging codes. Among the received ranging codes, SS selects one of the ranging codes (RC i )inss sstep1.ifrc i consists of A 1 and A 2, SS sends only a part of RC i (A1 or A2) and Hash value of RC i to BS in order to protect Man-in-the-Middle attack. In BS s step 1, BS receives a part of RC i (A1) and the hashed value H(RC i ). BS finds A2 from ranging code pool using A1, and then BS authenticates SS through verifying the received hash value. Thus, the selected ranging code is not only Mobile WiMAX communication but also used for generating a prime number p asoneofglobalvariablesindhprocess. In SS s step 2, 3, and 4, SS generates the other global variable g and public/private key pair and then sends them to BS. BS receives a public key of SS and global variables (prime number and its primitive root). If the received key and variables are verified, BS also sends his public key to SS in BS s step 3. Thus, BS and SS can share DH global variables and public key with each other through initial ranging process. Of these, they can generate a shared common key called pre-tek separately and establish secret communication channels in step 4 and 5 separately. Therefore, the proposed approach can protect SBC security parameters and PKM security contexts using the shared traffic encryption key (pre- TEK) during initial network entry procedure Approach to Access Network Vulnerabilities. As we already mentioned, PKM which is the main security architecture in Mobile WiMAX only covers wireless traffic between SS and BS because other communication ranges required security functions that are beyond IEEE 82.16e-25 standard. Moreover, technical documents of Network Working Group (NWG) in WiMAX Forum assume that ASN network is trusted and AAA connections between ASN and CSN may be protected with IPSec tunnel. However, there are a lot of possibilities new security holes to happen including various zero-day attacks. Moreover, IPSec requires additional s/w and h/w facilities for supporting whole Mobile WiMAX domains. Thus in this paper, we present a simple and efficient key exchange method using a device-certificate. Basically, network devices in Mobile WiMAX have a device certificate, so they can be applied to make more robust access to network domain based on PKI structure. In order to applying device certificate based approach to access network domain, we assume that Mobile WiMAX devices are certified from public authority and they can verify certificates of each other using certificate chain. In Figure 5, all devices in Mobile WiMAX have their own certificate and a certificate chain for verification. If BS would like to exchange important messages with ASN/GW, BS needs to generate a session encryption key for secure communication between BS and ASN/GW. In this case, BS first searches for an appropriate certificate (including correspondent s public key) to verify ASN/GW s identity and obtain public key. After getting public key, BS generates asn-tek as a session encryption key for secure communication with ASN/GW. Using the asn-tek, BS encrypts a message and sends the encrypted message together with the encrypted asn-tek key using ASN/GW s public key, Timestamp, and Authority s certificate to ASN/GW. When ASN/GW receives the messages from BS, ASN/GW first tries to verify the authority s certificate and checks the validation time from Timestamp. If the verification process is successful, ASN/GW decrypts the asn- TEK key and the original message. Thus, a problem of insecure communication between BS and ASN/GW can be solved by using asn-tek key as a encryption key between BS and ASN/GW. In the case of ASN-to-CSN, the proposed method generates a common encryption key called asn-csn- TEK using the same method as a way for BS-to-ASN/GW to establish secure connection. In Figure 5, we can show an example using asn-csn-tek Approach to Handover Vulnerabilities. In Mobile WiMAX, a handover process adopts a kind of fast handover method based on Handover (HO) process optimization flags to provide seamless communication by reducing the number of message exchange. However, such a handover process still

150 6 EURASIP Journal on Wireless Communications and Networking An overview of handover procedure SS MOB MSHO-REQ Serving BS HO-request Target BS HO-request Context-request Context-report ASN/GW MOB MSHO-RSP HO-response MOB HO-IND HO-acknowledge HO-confirm RNG-REQ RNG-RSP PKM authentication phase EAP authentication methods (EAP-AKA ) HO-acknowledge TEK creation phase PKM-RSP (PKMv2 SA-TEK-challenge) PKM-RSQ (PKMv2 SA-TEK-request) PKM-RSP (PKMv2 SA-TEK-response) PKM-REQ (PKMv2 key-request) PKM-RSP (PKMv2 key-reply) Figure 3: Overview of handover procedure. Proposed initial network entry approach SS s step: 1. Choose a ranging code H(RC i ), RC i = A 1 A 2 2. PNG (RC) 3. Generating p,g 4. Generating SS S pub SS UL-MAP Ranging codes = {RC 1, RC n } Initial ranging code (H(RC i ), A 1 ) Global parameters (p,g), MS s public key BS ASN/GW BS s step: 1. Verifying initial tanging code A 1 H(RC i )! = H(RC i ), 2. Verifying p 3. Generating BS S pub AAA 5. Generating pre-tek RNG-RSP 5. Generating pre-tek BS s public key Connection establishment Initial ranging with DH key agreement Secure ranging message with pre-tek Secure SBC negotiation Secure authentication and key exchange Figure 4: Proposed network initial entry approach.

151 EURASIP Journal on Wireless Communications and Networking 7 Proposed access network approach Authority Public key infrastructure Cert chain BS Cert chain ASN/GW Cert chain AAA << Authority s cert >> << RAS s cert >> << ACR s cert >> << AAA s cert >> << Authority s cert >> << RAS s cert >> << ACR s cert >> << AAA s cert >> << Authority s cert >> << RAS s cert >> << ACR s cert >> << AAA s cert >> Generating asn- TEK timestamp 1. Verifying auth s certificate 2. Decrypting message E asn-tek (context-request (auth policy)) E pub asn/gw (asn-tek) TS Auth s Cert E asn-tek [context-report (authorization policy)] Auth s Cert Generating asn-csn-tek timestamp 1. Verifying timestamp and auth s certificate 2. Decrypting asn-tek and message E asn-csn-tek [message] E pub asn/gw [asn-csn-tek] TS Auth s Cert Access service network Connectivity service network Figure 5: Proposed access network approach. has a problem as we already mentioned in Section 3.3 handover vulnerability. In this section, new handover approach with embedded mutual authentication parameters is proposed. The proposed HO approach includes a few additional fields for the embedded parameters of providing mutual authentication such as Nonce, Certificate (Cert), Authorization Key (AK), and Acknowledge (Ack). First, the challenge-response scheme with Nonce, Cert, and AK is used to provide Target BS (TBS) authentication. Moreover, HMAC/CMAC tuple is used for SS authentication as well as message authentication. Thus, the proposed approach can take an effect on mutual authentication using the embedded parameters even though HO optimization process is used. From message 1 to message 3 in Figure 6, TBS authentication is first processed during HO-Request process. When HO process is started, Serving BS (SBS) sends HO-Request message with Nonce to TBS. TBS replies HO-Response with an encrypted Nonce and Cert to SBS. If SBS verifies the included Nonce and Cert in HO-Response message, SBS sends HO-confirm message with Ack to TBS, and then TBS authentication is finished. In the case of MS authentication, CMAC/HMAC tuples are applied to authenticate MS as illustrated message 4 and 5 in Figure 6. After HO process, MS tries the Ranging process and TBS can authenticate MS using MAC code verification because RNG-REQ message includes CMAC/HMAC tuple generated by AK. Therefore, our proposed method takes an effect on getting confidentiality by including a few information fields in the existing HO message in spite of using additional HO optimization flags. This approach enhances security and performance factors during handover without full authentication process based on PKMv2. 5. Comparison of the Proposed Approaches 5.1. Mobile WiMAX-Based DH Approach in Initial Entry Procedure. In order to protect a vulnerability of initial network entry, modified DH approach was proposed in this paper. The proposed approach provides both confidentiality and countermeasure against Man-in-the-Middle attack in comparison with IEEE 82.16e and original DH scheme. Moreover, there are not communications overhead because the parameters of the proposed approach are embedded in the existing initial ranging message exchange. However, the modified DH approach performs a couple of cryptographic operations. Even though the processing delay is very small and the original DH-based scheme is one of the best known schemes for communication system security, more optimized cryptographic operations for Mobile WiMAX entry process are considered in the near future. Table 1 shows that the proposed approach provides confidentiality and countermeasure against man-in-the-middle-attack.

152 8 EURASIP Journal on Wireless Communications and Networking Proposed handover security SS MOB MSHO-REQ SBS TBS ASN GW 1.HO-request (nonce) MOB MSHO-RSP Context-request/report (AK context) 2. HO-response (E AK [nonce cert]) 3. HO-acknowledge (Ack) TBS authenticatoin SS authentication MOB HO-IND 4. RNG-REQ (CMAC) Omitted HO-confirm(TEK context) HO-acknowledge 5. RNG-RSP( CMAC) EAP authentication methods (EAP-AKA, ) PKM-RSP (PKMv2 SA-TEK-challenge) PKM-REQ(PKMv2 SA-TEK-request) PKM-RSP(PKMv2 SA-TEK-response) PKM-REQ(PKMv2 Key-request) PKM-RSP(PKMv2 Key-reply) Figure 6: Proposed handover security approach. Moreover, in the side of performance consideration, it does not have any communication overhead, but a little bit security overhead. In [16], the authors proposed a similar approach to apply the enhanced DH scheme to Mobile WiMAX network initialization process. Basically the DH approach including ours is better than the existing approach. However, the Rahman s approach is not clear to understand and they can not provide how the random is generated and distributed to others to use DH scheme. certificate-based operations are already verified in the RSAbased authentication scheme in IEEE 82.16e standard. The RSA-based certificate approach is similar to the proposed approach but it requires much processing delay than our ECC-based certificate approach. In case of RSA approach, the processing delay is about 12 ms. On the other hand, in ECC case it is only under 1 ms Mobile WiMAX s Device Certificate-Based Key Exchange in Access Network. In Mobile WiMAX network domain, IPsec has been recommended as one of solutions to provide network security. However, every device in Mobile WiMAX has to be installed and support IPSec functionality in order to use IPSec-based security services. Thus, the proposed approach concentrated on simple and efficient key exchange without any additional features. In Table 2, we can see that IPSec needs two-phase negotiation procedure (6 9 times message exchange) and a little bit complex security operation such as security association, key exchange, and key generation. On the other hand, the proposed approach provides very intuitive and systematic solution (2 times message exchange) using a device certificate. Although the proposed approach uses certificate-related functions such as signing and verifying, it does not have any side-effect in Mobile WiMAX network because the overhead and delay of 5.3. Mutual Authentication Approach in Handover Procedure. Fast mobility support is one of the most distinguished capabilities in Mobile WiMAX. However, it can not support an authentication function during handover because of using HO optimization. The proposed approach used embedded parameters for mutual authentication during handover process. Thus, at the view of security aspect, the proposed approach can provide mutual authentication in comparison with the default IEEE 82.16e handover scheme. Moreover, in performance analysis, it shows low processing overhead (1 time encryption/decryption for verifying Nonce) and no communication delay because it does not cause any additional authentication-related message exchange. In case of Hur s [11] approach, it is based on EAP-based preauthentication. However, it can not be a stable solution under frequent handover environment. In Mobile WiMAX, the dynamic mobility is one of the best advantages.

153 EURASIP Journal on Wireless Communications and Networking 9 Table 1: Security and performance analysis in network entry security. Security Performance IEEE 82.16e Applying original DH Rahman s approach [16] Proposed approach Confidentiality None O O O Man-in-the-middle attack None None O O Processing overhead Communication overhead None Random number and key generation Random number and key generation Not clear to generate and distribute DH values Random number and key generation and hash processing None Table 2: Security and performance analysis in access network. IEEE 82.16e Applying IPSec Certificate approach with RSA 124 bit Proposed approach with ECC 163 bit Security Confidentiality None O O O Medium (Depends on Low (depends on key High Processing overhead key size) size) Certification Certification Performance SA negotiation/ike/key operation/key operation/key generation generation generation Communication overhead Requiring additional features High Low Low 2 phase 6 9times 1 phase 2 times packet 1 phase 2 times packet packet exchange exchange exchange O None None Table 3: Security and performance analysis in handover security. Security Performance Mutual authentication Processing overhead Communication overhead IEEE 82.16e Handover Hur s [11] Approach None O O None None High (depends on EAP protocols) High (depends on the number of handover) Proposed handover approach Low 1 time encryption/decryption None 6. Discussion of Secure and Robust Mobile WiMAX with Proposed Approach This paper proposed novel approaches to minimize security risks in Mobile WiMAX network. We showed a reliable Mobile WiMAX architecture applying the security approaches called RObust Secure MobilE WiMAX (ROS- MEX) as illustrated in Figure 7. In ROSMEX, the enhanced network entry process has an initial ranging process with modified DH key agreement. The approach assigns a temporary Security Association (e.g., pre-tek and predefined cryptographic suites) to prevent a primary management connections between SS and BS. Thus, ROSMEX can give confidential communications to whole wireless communication range because the proposed approach generates temporary traffic encryption key and then uses the key for traffic encryption before SBC negotiation. Moreover, ROSMEX supports secure communications in all access networks. Any two entities in Mobile WiMAX can establish a secure channel using device certificate-based simple and efficient key exchange. The approach eliminates all possibilities of disguising as a valid entity in Mobile WiMAX. In Section 5.2, we already knew that it is a more efficient method than IPSec approach.. Finally, ROSMEX can support secure mobility despite omitting authentication and TEK phases by using HO optimization flags. The improved HO process has embedded mutual authentication parameters in order to provide authentication during handover. The challengeresponse scheme embedded in HO messages authenticates TBS, and SS is authenticated by MAC scheme. Specifically, our proposed approaches do not need any additional message passing and do not prevent original control flow

154 1 EURASIP Journal on Wireless Communications and Networking Internet Robust and secure mobile WiMAX AAA ASN/GW ASN/GW AAA Novel approach II: PKI-based key exchange in network domains BS BS CSN ASN ASN CSN Novel approach I : Enhanced network entry security SS SS SS SS SS SS Novel approach III: Enhanced HO with mutual authentication Figure 7: Robust and secure mobile WiMAX network. of Mobile WiMAX. Therefore, ROSMEX architecture with enhanced security countermeasures can satisfy not only the security requirements but also performance requirements during Initial Network Entry, Access Network Communication, and Handover process. 7. Conclusion Mobile WiMAX is one of the best candidate systems to accommodate demands for broadband wireless access. It can support worldwide roaming capabilities, superior performance, low latency, supporting all-ip core network, advanced QoS, and security. Moreover, Mobile WiMAX can cooperate with existing and emerging networks. However, Mobile WiMAX technology is not perfect and is not an ultimate solution for beyond 3G networks, but a kind of bridging system toward 4G networks. In the case of security aspects in Mobile WiMAX, it still has a potential possibility of a few security-related vulnerabilities. In this paper, we investigated new security vulnerabilities such as a disclosure of secret contexts during initial entry procedure, a lack of a protection mechanism in access network communication, and a possibility of rogue SS or BS attacks during HO process (in case of using HO optimization flags). Therefore, in order to eliminate the security vulnerabilities, we proposed three possible countermeasures. In the case of an initial entry process threat, modified DH key agreement is applied to initial ranging process to generate session encryption key. Using the temporal encryption key, the messages including security contexts can be protected during initial entry procedure. Secondly, a simple key exchange scheme based on device certificate was proposed as a solution to settle the vulnerability of the access network. Thus, each network component in ASN and CSN can generate session encryption keys and the correspondents also can verify them. Finally, the handover threat could be reduced using the modified HO procedure approach including a challenge-response scheme and MAC code verification in the existing HO messages. Based on the proposed approaches, we analyzed and compared our approach called ROSMEX architecture with the existing solutions. We believe that our ROSMEX architecture will contribute to make an enhanced Mobile WiMAX network. In future work, more research for IMT-advanced architecture is needed. Acknowledgment This research was supported by the MKE (The Ministry of Knowledge Economy), Korea, under the ITRC (Information Technology Research Center) support program supervised by the NIPA (National IT Industry Promotion Agency) (NIPA-21-C ). Moreover, a preliminary version of this paper appeared in NBIS 27, September 3 7, Regensburg, Germany [18]. The revised paper includes new related work in Section 2.3, the updated security analysis in

155 EURASIP Journal on Wireless Communications and Networking 11 Section 5 with the existing approaches and overall parts like abstract, introduction, and conclusion are rewritten, and the main approach was also revised with coherence. References [1] IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems, IEEE Std IEEE, 24. [2] IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems, Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum, IEEE Std 82.16e-25. IEEE, 25. [3] WiMAX Forum, Mobile WiMAX: The Best Personal Broadband Experience, 26. [4] WiMAX Forum, WiMAX End-to-End Network Systems Architecture Stage 2, 3, 26. [5] Airspan, Mobile WiMAX security, Airspan Networks Inc. 27, [6] E. Yuksel, Analysis of the PKMv2 Protocol in IEEE 82.16e- 25 Using Static Analysis Informatics and Mathematical Modeling, TUD, 27, views/publication details.php?id=5159. [7] J. Bellardo and S. Savage, denial-of-service attacks: real vulnerabilities and practical solutions, in Proceedings of the 12th USENIX Security Symposium, vol. 12, Washington, DC, USA, August 23. [8] C. Wullems, K. Tham, J. Smith, and M. Looi, A trivial denial of service attack on IEEE direct sequence spread spectrum wireless LANs, in Proceedings of the Wireless Telecommunications Symposium (WTS 4), pp , 24. [9] D. Johnston and J. Walker, Overview of IEEE security, IEEE Security and Privacy, vol. 2, no. 3, pp. 4 48, 24. [1] M. Barbeau, WiMax/82.16 threat analysis, in Proceedings of the 1st ACM International Workshop on Quality of Service and Security in Wireless and Mobile Networks (Q2SWinet 5), pp. 8 15, 25. [11] J. Hur, H. Shim, P. Kim, H. Yoon, and N.-O. Song, Security considerations for handover schemes in mobile WiMAX networks, in Proceedings of IEEE Wireless Communications and Networking Conference (WCNC 8), pp , Las Vegas, Nev, USA, March-April 28. [12] M. Habib and M. Ahmad, A review of some security aspects of WiMAX and converged network communication software and networks, in Proceedings of the 2nd International Conference on Communication Software and Networks (ICCSN 1), pp , 21. [13] P. Rengaraju, C.-H. Lung, Y. Qu, and A. Srinivasan, Analysis on mobile WiMAX security, in Proceedings of IEEE Toronto International Conference on Science and Technology for Humanity (TIC-STH 9), pp , September 29. [14] S. S. Hasan and M. A. Qadeer, Security concerns in WiMAX, in Proceedings of the 1st South Central Asian Himalayas Regional IEEE/IFIP International Conference on Internet (AH- ICI 9), November 29. [15] M. Habib, T. Mehmood, F. Ullah, and M. Ibrahim, Performance of WiMAX security algorithm: the comparative studyofrsaencryptionalgorithmwitheccencryption algorithm, in Proceedings of the International Conference on Computer Technology and Development (ICCTD 9), vol. 2, pp , November 29. [16] M. S. Rahman and M. Md. S. Kowsar, WiMAX security analysis and enhancement, in Proceedings of the 12th International Conference on Computer and Information Technology (ICCIT 9), pp , December 29. [17] W. Diffie and M. E. Hellman, New directions in cryptography, IEEE Transactions on Information Theory, vol.22,no.6, pp , [18] T. Shon and W. Choi, An analysis of mobile WiMAX security: vulnerabilities and solutions, in Proceedings of the 1st International Conference on Network-Based Information Systems (NBiS 7), vol of Lecture Notes in Computer Science, pp , September 27.

156 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 9 pages doi:1.1155/21/ Research Article A Speed-Adaptive Media Encryption Scheme for Real-Time Recording and Playback System Chen Xiao, 1 Shiguo Lian, 2 Lifeng Wang, 3 Shilong Ma, 1 Weifeng Lv, 1 and Ke Xu 1 1 State Key Laboratory of Software Development Environment, School of Computer Science & Engineering, Beihang University, Haidian District, Beijing 183, China 2 France Telecom R&D Beijing, Haidian District, Beijing 18, China 3 Department of Electronic & Information Engineering, Beijing Electronic Science and Technology Institute, Fengtai District, Beijing 17, China Correspondence should be addressed to Shiguo Lian, sglian@gmail.com Received 29 March 21; Accepted 2 August 21 Academic Editor: Liang Zhou Copyright 21 Chen Xiao et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The recording and playback system (RPS) in video conference system needs to store mass of media data real-timely. Considering the security issue, media data should be encrypted before storing. Traditional full encryption and partial encryption algorithms are not applicable to RPS because they could not adjust their speed to meet the throughput variation of media data in real-time RPS. In this paper, a novel lightweight speed-adaptive media-data encryption (SAME) scheme is proposed firstly. Secondly, the SAME is improved to a packet-based algorithm according to the implementation of data storage in RPS system. Thirdly, an RPS oriented queue theory-based autoadaptive speed control mechanism for SAME is designed. Finally, these schemes are integrated into the practical system, that is, AdmireRPS, an RPS of a heterogeneous wireless network-(hwn-) oriented video conference system. Theoretical analysis and experimental results show the SAME is effective enough to support real-time applications. In addition, the proposed schemes also can be used in video surveillance and other video recording systems. 1. Introduction With the development of multimedia and network technologies, video conferences and some other streaming media systems have attracted significant research efforts [1 8]. Recording and playback system (RPS), which stores large-volume media data and plays them back according to requirements of users, plays an important role in a video conference system. Since the media data stored in RPS contain all the crucial information of the meetings, they should be encrypted properly. However, traditional full-encryption algorithms are not applicable due to the high-volume of media data and real-time requirements of multimedia applications [1]. How to maintain the security of media data becomes a challenge task [1, 8, 9]. To solve this problem, a lot of multimedia (e.g., image, video, or audio) content encryption schemes have been proposed during the last decade. In this paper, more attention is paid on video encryption, since bandwidth of image or audio data is comparatively less than that of video. According to different viewpoints, those encryption schemes can be divided into different kinds, respectively. From the way they reduce the computational complexity, these schemes could be divided into three types, that is, partial encryption [1, 11], joint compression encryption [1],and improved full encryption [12]. Partial encryption encrypts a selected crucial subset of media data. One kind of partial encryption focuses on I- frames or intramacroblocks in video frames. The algorithm proposed in [1] encrypts only the I-frames in MPEG2. Although the other data are still clear, and might be eavesdropped by attackers, it is very hard to rebuild a clear video copy without the I-frames. However, the practical analysis given in [13] shows that without the I-frames video contents are still discernible, because the unencrypted I macroblocks in the P-frames can be fully decoded. So, it is not secure enough for some confidential applications [14]. An improved scheme called SECMpeg is proposed in [15].

157 2 EURASIP Journal on Wireless Communications and Networking It has implemented 4 levels of security. One important level is to encrypt all the I-blocks in P- and B-frames. This scheme can provide higher security but suffers a substantial increasing of complexity. Moreover, it needs the encryption process to parse the P- and B-frames, and is not suitable for RPS. Another kind of partial encryption focuses on discrete cosine transform (DCT) coefficients, motion vector, and other sensitive data. For example, Tang [16] encrypted videos by scrambling discrete cosine transform (DCT) coefficients. Shi and Bhargava [17] proposed a video encryption algorithm, where the 64 most significant sign bits of DCT coefficients and motion vectors in each macroblock are encrypted by a symmetric cipher. In some other schemes [18, 19], intraprediction mode, residue data, and motion vector are encrypted partially to keep format compliance. The third kind of algorithms use some lightweight algorithms to encrypt the sensitive data, for example, in image encryption, Lian et al. [2] used chaotic cipher to encrypt JPEG2 images. With regard to audio encryption, several partial encryption schemes are also proposed [1, 11]. Joint compression encryption is another kind of media content encryption. Wu and Kuo [21] implemented encryption operations in entropy coding. In [22], the scheme combined fixed length code (FLC) and variable length code (VLC) codeword, which is achieved by permuting the codeword and encrypting the index of code table in MPEG-4. In [23], a perceptual video encryption scheme is proposed based on exploiting the special feature of entropy coding in H.264. Improved full-encryption schemes use some lightweight algorithms to encrypt the whole bit stream. In [1, 2, 24], chaotic stream ciphers are constructed and used to encrypt video data. Besides chaotic stream ciphers, VEA scheme is another improved full-encryption scheme [12]. The main idea of VEA is as follows. Divide the plaintext into two segments: Even and Odd, encrypt the Odd with a standard encryption algorithm E(Odd), and the other half Even is exclusive-ored (Xor) with the plaintext of Odd. This mechanism can reduce the complexity to nearly one half and achieve a sufficient security level. This algorithm is extended to encrypt one fourth of the plaintext in [25]. These schemes can encrypt the compressed video data, and fit the RPS. But their encryption rate is changeless, so the encryption throughput of them could not adjust to meet the variation of media data throughput in real-time RPS. On the other hand, according to the data format or application they oriented, these encryption schemes can be classified into two types, that is, one is all-purpose scheme, and the other focuses on special codec standards or applications. As mentioned above, the schemes in [1, 15] select the I-frames or I-blocks to encrypt, and the ones in [16 19, 26] choose the DCT coefficients and sign bits of the motion vectors to encrypt. These schemes can be used in different codec standards. The other kind of scheme focuses on the special image, audio, and video standards. For example, the schemes in [2, 27] aim at encrypting the image of JPEG2, the one in [11] aims at G.729 data encryption, and the ones in [18, 19] study the AVC encryption. There are also some schemes focusing on special applications, for example, joint fingerprint embedding and en/decryption algorithms are proposed and used in digital rights management (DRM) [3, 28]. The security of multimedia content in IPTV is studied in [4, 5]. Chen et al. [2] proposed a mathematical model-based dynamic optimal selective control mechanism for optimizing the security of multidatastreams in video conference. These schemes are designed based on the special needs of applications. Among the mentioned algorithms above, joint compression encryption needs to analyze the compressed data, which is infeasible for practical RPS systems. Partial encryption and improved full-encryption algorithms have also some limitations, because they could not adjust their speed to meet the variation of media data throughput of RPS. Therefore, in this paper, we design a secure real-time recording and playback system, named AdmireRPS, based on a novel Speed-Adaptive Media-Data Encryption (SAME). As of our knowledge, this is the first media encryption scheme considering the adaptive media throughput. Firstly, by combining the block cipher and stream cipher, a lightweight speed-adaptive media-data encryption is proposed. Secondly, a packet-based SAME (PSAME) is proposed, according to the implementation of data storage in a practical RPS system. Thirdly, a queue theory based autoadaptive speed control mechanism for SAME is designed. Finally, those schemes are integrated into AdmireRPS system. Theoretical and experimental analyses show the performance of SAME is effective enough to support real-time applications. The rest of the paper is organized as follows. Section 2 presents the overview of AdmireRPS system and its security problems. The design and implement of AdmireRPS are given in Section 3. The speed-adaptive media-data encryption (SAME) and autoadaptive speed control mechanism for SAME are presented in Section 4, andsection 5 discusses the performances and presents the theoretical analysis and experimental results. Finally, Section 6 concludes the paper. 2. Admire and Encryption Bottleneck 2.1. An Overview of Admire System. As is shown in Figures 1 and 2, the Admire (ADvanced Multimedia Interactive Real-time Environment) system [2, 6, 7] is a heterogeneous wireless network-oriented large-scale multimedia collaboration system. It is compatible with multicast, unicast, wired, and wireless clients, and includes not only video conference system but also RPS, immediate message system (IMS), electric white board, collaborate edit, desktop sharing, and so forth. This system has been used in hundreds of universities and institutes. In the biggest session, there are more than 1 users joined in at one time Encryption Bottleneck in Admire System. To meet the application demands of some confidential departments, a special edition with security concern is developed, in which data encryption is operated by a specified confidential encryption algorithm which is similar to DES and implemented in a PCI card [2]. Although the card has

158 EURASIP Journal on Wireless Communications and Networking 3 Figure 1: A practical video conference in Admire system. a declaratory encryption throughput of 224 Mbps, we find its stable external throughput is only 12 Mbps in fact [2]. We also find that the software implementation of traditional encryption algorithm on universal processor is no more than 2 Mbps, which is inadequate to the data throughput of MediaGateway [7], RPServer, and other servers. Consequently, the encryption becomes a bottleneck in the system. The encryption speed can be accelerated according to the growth of the CPU speed, but in the recent years the improvement of CPU speed decelerates because of the limit of manufacturing technology of semiconductor device, so the perspective of the growth of encryption speed is not optimistic. Comparatively, the disk densities improved 1 percent per year, which is faster than Moore s Law, something like 6 percent a year [29]. Moreover, core network bandwidth and image process ability growing even faster than disk densities. Predicatively, the dissymmetry between the throughput of encryption algorithm and the data volume to be encrypted will aggravate. Furthermore, there is a remarkable variation of media data throughput in Admire system, so a speed adaptive encryption scheme is needed. 3. The Proposed AdmireRPS 3.1. AdmireRPS in Admire System. AdmireRPS is the secure recording and playback system of Admire system. It can encrypt and save the specified streaming media data into media files according to the requirements of the participants. When a user misses a meeting, RPS can playback the audio/video for her/him, and asynchronous collaboration could be achieved. Figure 2 shows the architecture of AdmireRPS. On the left of the box with broken lines are RPS clients, in the box are the servers, including RPServer which records and playbacks media data, ControlServer [2] which works as the GateKeeper (GK) and ControlUnit (CU) of H.323 system, and AdmireDB which saves the session and access control information. All these components exchange control message using AdmireMBus [2], a lightweight message bus which is fully encrypted. Media data are transmitted in media channel, which is built based on multicast or application layer multicast [7]. The details of the secure recording and playback process will be presented in the following content Media Data Recording and Encryption. The media data record process in AdmireRPS is shown in Figure 2, which includes the following 3 steps. Firstly, the client sends the record request to ControlServer via the secure control channel. Then, ControlServer inserts the media information, including session information, multicast address of the session, Synchronization SouRCe (SSRC) identifier, and so forth, into AdmireDB. Finally, ControlServer asks RPServer to record media data in the specified multicast address (the client can also require RPS to record data with specified SSRC in a multicast address). The data in one session could be saved into a single file or several files according to SSRC. However, we found stable throughput of hard disk is correlate inversely with the number of files accessed at one time, thus single-file scheme could achieve a higher performance. On the other hand, multifile scheme can playback media data with selected SSRCs in one session separately. This scheme can achieve higher flexibility. Client has the right to choose either of single or multifile in record request. The record process in RPServer saves each received RTP [3] packet from the specified multicast address. Considering the playback process should send those packets according to the received time, the file structure is designed as follow (shown in Figure 3). For each packet, there is a 96-bits header. The first 16-bits records the total block length. The 17th bit EF is an encryption flag, which specifies the encryption mode and will be explained in Section 4. The second 32-bits is time stamp, which is the time offset from the beginning packet to the current. Because the playback module could not decode the encrypted RTP packet, the last 32-bit records the synchronize source identifier of the current packet, which is got from the RTP header. As is shown in Figure 4, the RTP packet is encrypted using packet-based SAME algorithm. A speed control algorithm is also introduced, which calculates the encryption speed control parameter l according to the input packet rate. ThosetwoalgorithmsareproposedinSection The Secure Playback Process. Similar with the record process, playback process in AdmireRPS works as follows.

159 4 EURASIP Journal on Wireless Communications and Networking 1 Record/ play request 4 RPServer and session info Encrypted ctrl channel msg & ctrl module msg & ctrl module msg & ctrl module 2 Update /search Audio/ video module Audio/ video module ControlServer 3 MCast addr info Database AdmireDB Wireless client Wired client msg & ctrl module Secure record & play module Wireless network RPServer 5 Send record/play data Server set Media channel Ctrl data Media data Database Figure 2: Architecture and control process of AdmireRPS. Block length EF For extension Time stamp SSRC Header IP header UDP header NSPD header Encrypted RTP packet use PSAME Figure 5: The protocol data unit in playback system. Data: encrypted RTP packet Figure 3: Block structure of encrypted RTP packet in RPS media file. Input RTP packets FIFO buffer Speed control (l) Packet based SAME Payload Media file storage buffer Figure 4: RTP data encryption in RPS system. Firstly, Client sends media file inquiry request to ControlServer via the encrypted control channel. Secondly, ControlServer inquires the media file information from AdmireDB, and sends it to Client. Thirdly, Client selects the media file her/she wants. Fourthly, ControlServer asks RPServer to playback the specified media file in a negotiated multicast address. Fifthly, RPServer reads packets are from the media file and sends them to the negotiated multicast address, according to the time stamp. RPServer can send the encrypted packets or plaintext packets. When the packets sent in ciphertext, a session key will be sent via the encrypted control channel simultaneously. To minimize the refactoring of the other modules in Admire system, an AdmireNSPD (Network Service Provider Daemon of Admire system) module [7] is designed to substitute the Winsock, and accomplish the NAT penetrating task. The PDU in our system is depicted in Figure 5. RPServer packets the encrypted RTP data into the NSPD payload, and uses one bit in NSPD header as the encryption flag. According to this flag, client determines the decryption way (more details are in Section 4). 4. Speed-Adaptive Media-Data Encryption (SAME) Statistical characteristics of compressed audio/video data are dramatically different from the ones of text data, because the variable-length codes and other processes used in compression remove the redundant information from the

160 EURASIP Journal on Wireless Communications and Networking 5 original data. Statistical analysis in [12] shows that the coded data have high randomness at the byte level. Based on this statistical characteristic of media data, we extend the idea of VEA algorithm to a new method that uses traditional block cipher to encrypt a part of data (part I), and uses its plaintext as the stream cipher key to encrypt another part of data (part II). By changing the ratio between parts I and II, we can adjust the speed of the encryption algorithm The Basic SAME Algorithm. In the basic algorithm, firstly, the plaintext is divided into segments with a same length. Secondly, a selected traditional block cipher algorithm is used to encrypt one segment. Thirdly, for the next l-blocks, use the plaintext of the previous segment as its stream cipher key. Assuming media data are saved in a FIFO buffer, the basic algorithm consists of the following steps (also shown in Figure 6). This algorithm is designed for media file rather than real-time packets in RPS. The improved algorithm for packetsisproposedinsection 4.2. To avoid the file header being guessed by the attackers, the first n-segments are full encrypted in step (2), where n is calculated from the session key. Although the probability of a segment being got by attackers is very little, the permutation proposed in [12] could be used before the dividing in step (1). The encryption speed control parameter l in step (4) is given in Section 4.3. This important parameter can adjust the speed of the encryption algorithm, and the experimental result is shown in Section 5.2. For file encryption, this parameter should be properly saved either by saving in a separate file, or using 1 bit in the header of each segment as encryption flag EF. The decryption process can determine the decryption way according to EF. Since most standard encryption algorithms need the length of plain-block divided exactly by a very number n (e.g., 8 bytes), the rear filling method in (1) is used in step (6). Here, the length of filled bytes is also the value to be filled. Figure 7 gives two examples of n = 8, Filling Length = n ( Length mod n ) Filled Value = n ( Length mod n ) (1) 4.2. Improved Algorithm for RTP Packets. The former algorithm, which is designed for byte stream, is suitable for encrypting large-volume media file. Since both recording and playback processes in AdmireRPS work on RTP packets, a packet-based algorithm can achieve higher efficiency. Therefore, we design a packet-based improved algorithm shown as follows. TheonebitEFofblockheaderinFigure 3 can be decided by. 1, if packet is fully encrypted, EF =, if Xor with the previous packet. Because adjacent packets could have different length, the Xor operator in step (3) is implemented as (3), where p i j is (2) jth byte of packet i, c i j is its ciphertext, and PL i is the length of packet i. That is to say, if the current packet is longer than the former one, duplicate the former packet at its rear. An example is given in Figure 8, wherepl i 1 = 1 and PL i = 15, c i j = p i j p i 1 jmodpl i 1. (3) 4.3. Adaptive-Speed Control Mechanism. In this subsection a speed control mechanism is designed to determine the encryption speed control parameter l in SAME, while the input throughput and upper limit of the expected queuing delay is given. A FIFO queue is used in RPServer to buffer the input data (asisshowninfigure 4). The new packets are inserted to its rear, while the encryption process gets packets from the front. Since the volume of media data in video conference change dramatically, speed control mechanism should ensure that the queuing delay is stable and under control, while take full advantage of encryption recourse. In order to find the relationship among the input bandwidth, queuing delay and encryption throughput, we make the following assumption: (1) packets arriving follows λ-poisson distribution, (2) encryption capacity is C, (3) packets length L packet is a constraint, and (4) memory is much greater than packet length. This is a typical model of an M/M/1/K queuing system [31]. Then, the average queuing delay d queue is d queue = 1 μ λ 1 (K +1)ρk + Kρ k+1 1 ρ k+1, (4) where μ = C/L packet is the packet number algorithm can encrypt in a unit time interval, ρ = λ/μ is the load rate, and λ is the packets arrive rate. If we assume that main memory capacity of RPServer is much greater than packet length, so the parameter k is approaching to the infinite, and the following equation can be got: lim d queue = 1 k μ λ, μ = 1 + λ, d queue ( ) 1 C = L packet + λ. d queue Therefore, given an upper limit of the expected queuing delay d queue (d queue > 1/μ), the minimum encryption speed C should satisfy. C (5) (( ) ) 1 + λ L d queue packet. (6) We can use (6) to calculate the minimum throughput of SAME with a limited queuing delay. For example, using

161 6 EURASIP Journal on Wireless Communications and Networking Plaintext: stream media data Permutation Divide into SegLength -byte segments i seg i + 1 seg i +2seg i + l seg Key Full encrypt l blocks Key Full encrypt i cipher i + 1 cipher i +2cipher i + l cipher Figure 6: Speed adaptive media-data encryption (SAME) algorithm The basic SAME algorithm Begin: (1) Permute and divide the byte stream in the FIFO buffer into segments with a length of SegLength. (2) Use the traditional block cipher algorithm F to encrypt the first n-segments Do until the last segment: (3) Use algorithm F to encrypt the first segment Seg i in the buffer. (4) For the next l blocks, its ciphertext CSeg j = Seg j 1 Seg j. (5) Repeat the steps (3) and (4). End Do (6) For the last segment, fill it using the filling method shown in Figure 7, then encrypt it using F. End Algorithm 1 The decryption process If (EF = full encrypt) then PlainText = F 1 (Cipher i ); Else PlainText i = PlainText i 1 Cipher i. End 5byte 8byte Algorithm Figure 7: Rear padding in SAME. encryption algorithm in PCI card, if L packet = 8Kb, λ = 11, and d queue = 4 ms, we can find that C should be bigger than 9 Mbps. Then we look up in Figure 9 and find l should be not less than 64 (the fifth asterisk of the blue line in Figure 9(a)). In addition, in the practical system when the number of queuing packets exceeds a gate valve, the throughput of the SAME can be increased by adding the parameter l. 5. Performance Analyses 5.1. Theoretical Security Analysis Based on Shannon Theory. The following analysis is focused on the basic SAME algorithm, and the result could be deduced to the improved algorithm. Considering Seg i to Seg i+l, based on Shannon theory [32], the attack difficulty of ciphertext-only attack is H(KP C), where K is the key of encryption algorithm

162 EURASIP Journal on Wireless Communications and Networking 7 The improved encryption algorithm Begin: (1) Use the traditional block cipher algorithm F to encrypt the first n-packets Do until the end: (2) Use algorithm F to encrypt the first packet in buffer. (3) For the next l packets, let its ciphertext CPacket j = Packet j 1 Packet j. (4) Repeat steps (3) to (4). End Do (6) Full encrypted the last packet, End Algorithm 3 Xor i 1 p1 p2 i 1... p1 i 1 p1 i 1 p2 i 1 p3 i 1 p4 i 1 p5 i 1 i p1 p2 i... p1 i p11 i p12 i p13 i p14 i p15 i c1 i c2 i... c1 i c11 i c12 i c13 i c14 i c15 i Figure 8: Rear padding in packet-based SAME. Table 1: Throughputs of SAME with different CPU loads (measured in MBps). Algorithms Load 1 Load 2 Load 3 Load 4 Algorithm in PCI Software DES F, C is the ciphertext, and H is the entropy function. After segmentation, we get H(KP C) = H ( KSeg i Seg i+1 Seg i+l Cphr i Cphr i+1 Cphri + l ). (7) Since SAME uses the previous segment as the stream cipher key of the current segment, the security of the current segment depends on the previous one, finally security of Seg i+1...seg i+l depends directly or indirectly on Seg i, whose security depends on the theoretical secrecy of algorithm F. Thus, the ciphertext-only attack of SAME has an attack difficulty of H(KP C) >= H ( KSeg i Cphr i ). (8) Only when adjacent segments are related with each other, that is, H(KP C) = H ( KSeg i Cphr i ). (9) However, statistical analysis of compressed video stream shows that the data have high randomness at the byte level, moreover the permutation before data dividing can also reduce the relativity. Thus, the security is often not smaller than the first segment s encryption. The attackdifficulty of known-plaintext attack is H(K PC) =H ( K Seg i Seg i+1 Seg i+l Cphr i Cphr i+1 Cphr i +l ). (1) Because only Seg i and Cphr i relate to Key, we can obtain that the known plaintext attack of SAME has an attack difficulty of H(K PC) = H ( ) K Seg i Cphr i. (11) That is, the security depends only on the first segment s encryption Throughput Analyses on SAME. Owing to the limited scale of our Admire system, we study the throughput of SAME in simulation programs. As is shown in Figures 9(a) and 9(b), two tests, that is, Test A and Test B, are designed to evaluate the throughputs of SAME with DES similar algorithm in PCI card and DES implementation in software, respectively. These two simulation programs are implemented in C++, and each of them occupies only one processor each time. Test A runs in a Windows XP system with Intel Core Duo T GHz and 512 M RAM, and Test B runs in a Windows NT workstation with dual-processor Intel PIV 2.4 G and 512 M RAM. When the parameter l +1is{1, 4, 16, 64, 256, 124, 496}, the throughputs of SAME in Test A and Test B are {1.45, 2.774, 5.796, 23.12, 9.3, 334.1, 989.1, 1973} and {21.91, 41.4, 86.12, 314.9, 959.5, 1935, 267, 2843}, respectively. The results of SAME are compared with the original fullencryption algorithm and VEA. When l =, the SAME algorithm is equal to full encryption, and when l = 1, the SAME is equal to VEA. Experimental results also show that when l is small, the speed nearly increases linearly. When l is very large (e.g., l > 1), limited by the speed of stream cipher algorithm, the throughputs are both less than 3 GBps for the two cases. Table 1 shows the throughputs in Tests A and B when l + 1 = 124 and the computers have different loads. In column 1, only the SAME process is busy, so there is one CPU busy and the other is always idle. In column 2, beside the SAME process, there is another process with heavy load running on the other CPU, thus the two CPUs both have heavy loads. Column 3 gives the results when a process with on average 35 MBps hard disk transmission is added. The last column shows the results when the SAME process contests

163 8 EURASIP Journal on Wireless Communications and Networking 3 3 Speed of encryption algorithms (MBps) Speed of encryption algorithms (MBps) Parameter l Parameter l DES similar algorithm in PCI card VEA use algorithm in PCI SAME use algorithm in PCI (a) DES implementation in software VEA use DES implementation in software SAME use DES implementation in software (b) Figure 9: The throughput of SAME. CPU time with other two CPU intensive process. Comparing column 2 with column 1, we can find that, if only one CPU intensive process contest CPU time with SAME, the throughput of SAME reduces less than 3%. That is because both two process could use one CPU every time, the impact is not notable. But when the number of CPU intensive processes is bigger than the number of CPUs, the other processes would observably impact the throughput of SAME, as is shown in column 4. On the other hand, as in column 3, the I/O intensive process would not impact the SAME alot. In Admire RPServer, only the SAME encryption process is CPU intensive, other process like RTP parsing and file saving are either I/O intensive or not CPU intensive. Therefore, the SAME is effective enough for the high-volume real-time data in AdmireRPS. 6. Conclusions and Future Work In this paper, a security scheme for RPS system is designed and implemented, which is based on the speed-adaptive media-data encryption (SAME) algorithm. Firstly, combining the block cipher and stream cipher algorithm, the basic SAME algorithm is proposed. Secondly, a packetbased SAME is proposed according to the implementation of data storage in AdmireRPS system. Thirdly, a queue theory based autoadaptive speed control mechanism for SAME is designed. Finally, the packet-based SAME algorithm and the speed control mechanism are integrated into AdmireRPS system. Theoretical analysis and experimental results show the security and speed of SAME are suitable for real-time applications. Furthermore, the proposed schemes can also be used in video surveillance and other video recording systems. Acknowledgment This work was supported by the Major State Basic Research Development Program of China (973 Program) (Grant no. 25CB32192) and Project (no. SKLSDE-21ZX-6) of the State Key Laboratory of Software Development Environment. References [1] S. Lian, Multimedia Content Encryption: Techniques and Applications, Auerbach Publication, Taylor & Francis Group, London, UK, 28. [2] X. Chen, M. Shilong, X. Ke, W. Lifeng, and L. Weifeng, A dynamic optimal selective control mechanism for multidatastream security in video conference system, in Proceedings of the IEEE International Conference on Multimedia and Expo (ICME 7), pp , Beijing, China, July 27. [3] S. Lian, Secure video distribution scheme based on partial encryption, International Journal of Imaging Systems and Technology, vol. 19, no. 3, pp , 29. [4] S. Lian and Y. Zhang, Protecting mobile TV multimedia content in DVB/GPRS heterogeneous wireless networks, Journal of Universal Computer Science, vol. 15, no. 5, pp , 29. [5] S. Lian and Z. Liu, Secure media content distribution based on the improved set-top box in IPTV, IEEE Transactions on Consumer Electronics, vol. 54, no. 2, pp , 28. [6] T. Jin, J. Lu, and X. Sheng, Admire a prototype of large scale e-collaboration platform, in Proceedings of the 2nd International Grid and Cooperative Computing Workshop, vol. 333 of Lecture Notes in Computer Science, pp , Shanghai, China, 24. [7] T. Huang and X. Yu, An adaptive mixing audio gateway in heterogeneous networks for ADMIRE system, in Proceedings of the Grid and Cooperative Computing (GCC 3), vol. 333

164 EURASIP Journal on Wireless Communications and Networking 9 of Lecture Notes in Computer Science, pp , Shanghai, China, 24. [8]F.LiuandH.Koenig, Anovelencryptionalgorithmfor high resolution video, in Proceedings of the 15th International Workshop on Network and Operating Systems Support for Digital Audio and Video (NOSSDAV 5), pp , Washington, DC, USA, June 25. [9] H. Yin, C. Lin, F. Qiu, J. Liu, G. Min, and B. Li, CASM: a content-aware protocol for secure video multicast, IEEE Transactions on Multimedia, vol. 8, no. 2, pp , 26. [1] G. A. Spanos and T. B. Maples, Performance study of a selective encryption scheme for the security of networked, real-time video, in Proceedings of the th International Conference on Computer Communications and Networks (ICCCN 95), pp. 2 1, Las Vegas, Nev, USA, September [11] A. Servetti and J. C. De Martin, Perception-based partial encryption of compressed speech, IEEE Transactions on Speech and Audio Processing, vol. 1, no. 8, pp , 22. [12] L. Qiao and K. Nahrstedt, A new algorithm for MPEG video encryption, in Proceedings of the International Conference on Imaging Science, Systems, and Technology,LasVegas,Nev,USA, [13] L. Agi and L. Gong, An empirical study of MPEG video transmissions, in Proceedings of the Internet Society Symposium on Network and Distributed System Security, pp , San Diego, Calif, USA, [14] T. Lookabaugh, I. Vedula, and D. C. Sicker, Selective encryption and MPEG-2, in Proceedings of the ACM Multimedia, Berkeley, Calif, USA, 23. [15] J. Meyer and F. Gadegast, Security Mechanism for Multimedia Data with the Example mpeg-1 Video, Project Description of SECMPEG, Technical University of Berlin, Germany, [16] L. Tang, Methods for encrypting and decrypting MPEG video data efficiently, in Proceedings of the 4th ACM International Multimedia Conference & Exhibition (ACM Multimedia 96), pp , Boston, Mass, USA, [17] C. Shi and B. Bhargava, A fast MPEG video encryption algorithm, in Proceedings of the 6th ACM International Multimedia Conference, pp , Bristol, UK, [18] S. Lian, Z. Liu, Z. Ren, and H. Wang, Secure advanced video coding based on selective encryption algorithms, IEEE Transactions on Consumer Electronics, vol. 52, no. 2, pp , 26. [19] S. Lian, Z. Liu, Z. Ren, and Z. Wang, Selective video encryption based on advanced video coding, in Proceedings of Pacific-Rim Conference on Multimedia (PCM 5), vol of Lecture Notes in Computer Science, pp , 25. [2] S. Lian, G. Chen, A. Cheung, and Z. Wang, A chaotic-neuralnetwork-based encryption algorithm for JPEG2 encoded images, in Proceedings of the International Symposium on Neural Network (ISNN 4), vol of Lecture Notes in Computer Science, pp , Springer, Dalian, China, 24. [21] C.-P. Wu and C.-C.J. Kuo, Efficient multimedia encryption via entropy codec design, in Security and Watermarking of Multimedia Contents III, vol of Proceedings of SPIE, pp , San Jose, Calif, USA, 21. [22] J. Wen, M. Severa, W. Zeng, M. H. Luttrell, and W. Jin, A format-compliant configurable encryption framework for access control of video, IEEE Transactions on Circuits and Systems for Video Technology, vol. 12, no. 6, pp , 22. [23] L.-F. Wang, W.-D. Wang, J. Ma, C. Xiao, and K.-Q. Wang, Perceptual video encryption scheme for mobile application based on H.264, Journal of China Universities of Posts and Telecommunications, vol. 15, supplement, pp , 28. [24] S. Lian, J. Sun, J. Wang, and Z. Wang, A chaotic stream cipher and the usage in video protection, Chaos, Solitons and Fractals, vol. 34, no. 3, pp , 27. [25] A. S. Tosun and W. C. Feng, Lightweight security mechanisms for wireless video transmission, in Proceedings of the International Conference on Information Technology: Coding and Computing, 21. [26] C. Shi, S. Wang, and B. Bhargava, MPEG video encryption in real-time using secret key cryptography, in Proceedings of the International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA 99), LasVegas,Nev, USA, [27] S. Lian, J. Sun, D. Zhang, and Z. Wang, A selective image encryption scheme based on JPEG2 codec, in Proceedings of the The 24 Pacific-Rim Conference on Multimedia (PCM 4), vol of Lecture Notes in Computer Science, pp , Springer, 24. [28] S. Lian, Z. Liu, Z. Ren, and H. Wang, Secure distribution scheme for compressed data streams, in Proceedings of the IEEE Conference on Image Processing (ICIP 6), April 26. [29] J. Gray and D. Patterson, A conversation with Jim Gray, ACM Queue Storage, vol. 1, no. 4, 23. [3] H. Schulzrinne, S. Casner, R. Frederick, and V. Jacobson, RTP: A transport protocol for real-time applications, RFC 1889, [31] T. G. Robertazzi, Computer Networks and System: Queuing Theory and Performance Evaluation, Springer, Berlin, Germany, 2. [32] C. E. Shannon, Communication theory of secrecy systems, Bell System Technical Journal, vol. 28, pp , 1947.

165 Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 21, Article ID , 14 pages doi:1.1155/21/ Research Article ESVD: An Integrated Energy Scalable Framework for Low-Power Video Decoding Systems Wen Ji, 1 Min Chen, 2 Xiaohu Ge, 3 Peng Li, 1 and Yiqiang Chen 1 1 Institute of Computing Technology, CAS, Beijing 119, China 2 Department of Electrical and Computer Engineering, University of British Columbia, vancouver, BC, Canada V6T 1Z4 3 Department of Electronics and Information Engineering, Huazhong University of Science and Technology, Wuhan, Hubei , China Correspondence should be addressed to Xiaohu Ge, xhge@mail.hust.edu.cn Received 1 April 21; Accepted 6 June 21 Academic Editor: Liang Zhou Copyright 21 Wen Ji et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Video applications using mobile wireless devices are a challenging task due to the limited capacity of batteries. The higher complex functionality of video decoding needs high resource requirements. Thus, power efficient control has become more critical design with devices integrating complex video processing techniques. Previous works on power efficient control in video decoding systems often aim at the low complexity design and not explicitly consider the scalable impact of subfunctions in decoding process, and seldom consider the relationship with the features of compressed video date. This paper is dedicated to developing an energyscalable video decoding (ESVD) strategy for energy-limited mobile terminals. First, ESVE can dynamically adapt the variable energy resources due to the device aware technique. Second, ESVD combines the decoder control with decoded data, through classifying the data into different partition profiles according to its characteristics. Third, it introduces utility theoretical analysis during the resource allocation process, so as to maximize the resource utilization. Finally, it adapts the energy resource as different energy budget and generates the scalable video decoding output under energy-limited systems. Experimental results demonstrate theefficiency of the proposed approach. 1. Introduction With the growing popularity of portable video applications, such as portable video smart phones, mobile video terminals such as PDA, and vehicle DVD devices energy consumption of video decoders becomes an important design requirement. Lots of compression codecs are issued for the several major video code standards, including MPEG4/2, H.264/3, and AVS. Generally, decoders focus on the performance while rarely support dynamic decoding process to meet the variable energy resources. However, most portable video application devices operate on batteries with limited-energy supply. The capacity of battery in portable devices is limited, as well as the usable capacity of the battery declines with using time. Thus, power should be used economically to provide longer service time. Then, how to make the video decoder adapt resource in handheld devices? How to maximum video decoding quality under battery constraint when playing on portable terminals? This paper tries to answer above-mentioned questions. In this paper, we proposed simple, energy-scalable video decoding algorithms for energy constraint terminals to save power and improve video quality. Moreover, we complement these algorithms with device energy aware method to lengthen the available time of video services. This is implemented through maximizing the decoded available video frames at a given power budget. The algorithm, called ESVD, means an integrated energy-scalable video decoding framework for low-power video decoding applications. ESVD uses energy profiles as scalable management guideline. Each energy profile is equivalent to an energy constraint budget. On such ESVD, algorithms use utility theory to find the best energy levels for each of the subfunctions in decoding. In ESVD system, video decoder can dynamically adapt the variable energy resources through energy aware technique. ESVD helps the decoder combine decoded data

166 2 EURASIP Journal on Wireless Communications and Networking with the decoding process. Video decoder can work under variable energy resources constraint marked with different energy consumption budgets and provide a wide scope adjustable decoding energy output. Besides, it uses utility theory to solve the tradeoff between decoding effect and energy consumption, so as to obtain better performance in each energy levels. This paper is organized as follows. Section 2 describes related work; Section 3 gives label and parsing method so as to provide a sufficient conditions for the ESVD; Section 4 describes the energy-scalable video decoding algorithms; Section 5 evaluates them; and Section 6 concludes. 2. Related Work and Backgrounds The contributions of the paper are related to several areas of work, which we consider in turn Designing Low-Power Video Encoders Scalable Video Decoders on Terminals. De Schrijver et al. [1] study the scalable video codec. They consider the memory, processing power, and bridge these with amount of bandwidth which comes from video fragment. Thus, scalable function is from the encoded scalable video bitstreams. Yanagihara et al. [2] propose CPU load-scalable video decoder algorithm, it uses several DCT manipulations such as low-pass filtering and resolution conversion in DCT domain. The decoder aims at the application of multichannel multicast system. Their work is rudimental to ours. Landge et al. [3] propose a systematic framework to optimize the energy consumption. They are in view of wavelet-based video decoders and use generic computational complexity metrics derived from the frequency of execution of program basic blocks. Since the decoder often does not know beforehand the encoded streams, this scalable function is obtained postmanufacturing and is unique to each codec system Designing Low-Power Video Decoders. Masselos et al. [4] design a low-power decoders based on the replacement of the image block by the selected codeword in the output image. Besides, they use efficient transformations to the codewords to compensate for the quality degradation introduced by the small codebook size in the encoder side. This method reduces its memory requirements so that it gets lower power consumption. Szu- Lee and Kuo [5] integrate the encoder selected proper interprediction modes and then generate a video bit stream. This method enables the encoder to estimate the decoding complexity and choose the best inter prediction mode to meet the complexity constraint of the target decoding platform. In a word, these methods rely on the encoder to reduce resource consumption of decoder. From integrated circuit aspect, Liu et al. [6] derive rapid algorithm in IDCT, deblocking filter and prediction, which can reduce the processing cycles and reduce the memory size and access frequency. These methods are the main measures for lowering the power consumption. The work is also complementary to ours. The other low-power design techniques include skipping computation in zero components, using lower constant multipliers, reducing transitions in the data path, and self-adaptive techniques. These methods acquire good effects in IDCT and prediction compensation modules, corresponding research examples include August and Ha [7] in IDCT and prediction, Tsung- Tsai and Fang [8] in VLC, and Xu and Choy in [9] self-adaptive prediction. We combine thoughts in scalable decoder and methods in lowpower design so as to achieve integration scalability and efficiency Complexity Power Mapping in Video Decoders. From encoder aspect, researchers have developed how to measure the power consumption in video encoders. He et al. [1] analyze the rate-distortion (R-D) behavior of video encoding system under the energy constraint. Based on power-ratedistortion (P-R-D) model in [1], they prove that power is tightly coupled with rate, thus, to trade bits for joules and to perform energy minimization are rapid method to obtain minimum energy [11]. Though these models are proposed based on the encoder, they can be used for reference in low power decoding design. From decoder aspect, existing approaches use the complexity metrics as the main measure methods on the first step; these metrics include counting the number of base operations [12], and memory access frequency [13] and occupation. On the second step, use mapping relations between complexity metrics and power or energy consumption to evaluate the accurate loss value [1, 14]. We combine the complexity metrics and power mapping methods, which in turn guide the control of optimal algorithm design to optimize the energy consumption Complexity Metrics in Video Codec Complexity Evaluation in MPEG. It is largely recognized that MPEG standards play a major role in the starting and development of multimedia communications and applications [15]. From the compression ratio point of view, MPEG possesses an important role of low-bit-rate video coding. From the complexity point of view, MPEG provides three tools to evaluate video codec complexity so that it controls the resources required at the decoder. Through these models, we can set boundaries on memory and computational requirements. The MPEG-4 standard defines video buffering verifier mechanism, which includes three virtual buffer models, named the video rate buffer verifier (VBV), the video complexity verifier (VCV), and video reference memory verifier (VMV). There, the VCV model is applied to all macroblocks in an MPEG-4 video bitstream and is used to verify the computational power required at the decoder. The model is defined in terms of the VCV MB/S decoding rate and VCV buffer size and is applied to all MBs in the scene [16, 17]. It mainly aims at the processing speed, defines in terms of the number of macroblocks (MBs) per second, and determines whether the decoding resources fit within a certain profile so as to not exceed the values specified for the corresponding profile and level.

167 EURASIP Journal on Wireless Communications and Networking 3 In VCV model, the computational complexity of the decoder is defined by bridging the data rate, and the number of MBs per second that the decoder has to decode. Indeed, the computational power consumption required by each MB decoding may largely vary with the MB types. According to careful analysis in [16], the ways to measure the decoding complexity of the encoded video data can be associated to the rate of the following parameters, including the number of MBs, the number of MBs per shape type such as boundary or transparent, the number of MBs per combination of texture and shape coding types, and the number of arithmetic instructions and memory Read/Write operations. Therefore, the number of MBs per combined coding type is a better method to represent the major factors determining the actual decoding complexity from the compressed data. Based on this, an alternative VCV model is proposed in [18], which allows a more efficient use of the available decoding resources. The model indicates that the decoding complexity can be measured by a combination of the MB complexity types and the number of MBs in corresponding different types. Thus, the decoding complexity can be evaluated and characterized by a combination of scenes, shape, and texture coding tools. This model enhances the VCV model because of complementing some determining factors. Furthermore, simplified control method in [18] can be adopted to distinguish the various types of MBs in terms of decoding complexity, in which the complexity weights can be defined relatively to the most complex MB type in the context of each profile. This means MPEG-4 decoders in most critical cases can be a compliant decoder, making a better supplement of the video complexity verifier model Complexity Evaluation in H.264. H.264/AVC represents many advanced techniques in standard video coding technology, and promises some significant advances of the state-of-the-art video coding techniques in a broad variety of applications [19, 2]. Compared to previous standards, H.264/AVC is given with respect to the coding efficiency and hardware complexity [21]. Indeed, assessing the complexity of a video coding standard is not a straightforward task; the same is true of H.264/AVC. Though the complexity heavily depends on the characteristics of the platform on which it is implemented, there are still mapping metrics to evaluate implementation complexity. Reference [21] analyse the complexity of H.264/AVC based on the new versions of the executable H.264/AVC specification, which includes updated tool definitions and can achieve a reduced complexity [22]. This analysis divided the H.264/AVC decoder into six parts, these are CABAC, RD-Lagrangian optimization, B-frames, Hadamard transform, deblocking filter, and displacement vector resolution. And it analyzes these parts in detail from the access frequency aspect and decoding time aspect Complexity Metrics in Video Codec. Generally speaking, the VCV model and the alternative VCV model are both based on measuring the decoding complexity in terms of the number of MB. The relative complexity weight for each MB complexity type is thus obtained as the ratio between the maximum decoding time for each MB type and the highest maximum decoding time from all the MB types relevant in the decoder profile. This method is widely adopted in the videocodec,suchas[23]. The measurement flow of video complexity evaluation systems such as video codec can be typically divided in several main steps. (1) Algorithmic development phase. This first step focuses on algorithmic performance. The algorithmic specification is typically released as a standard description plus a software verification model [24]. In this phase, complexity cost function in C-level analysis is needed. Efficient implementation based on each algorithm is adopted while it guarantees performance [25]. This phase focuses on deducing complexity, leading to high performance and enabling lowpower realizations in algorithm-specific complexity level. (2) Evaluation flow phase which deals with the actual system realization is based on a specific platform. The true implementation complexity of the algorithm based on universal platform can be acquired. Can this stage determine the cost of each module or each algorithm in some series terminals and, hence, its success and widespread diffusion or not? On the other hand, memory access consumption is another key factor in power consumption. In video decoding, the primary design goal is to reduce memory transfers between large frame memories and data paths. Many researches summarize the cost of a data transfer into a function of the memory size, memory type, and the access frequency, such as [5, 13, 26]. The measure method is the number of accesses per second instead of the clock frequency [26]. To accurately calculate the dynamic cost in each frame during decoding is a difficult job. Thus, in [12], they provide the upper limit of memory consumption. 3. Parsing and Labeling Video Decoding The main low-power techniques targeted at achieving lower consumed processing cycles and memory requirements are both described and discussed in Section 2. In this part, we address in analysis how to partition the decoder so as to providescalableoutput. In most cases, there is not enough residual capacity of battery to enable portable devices users to watch any video programs at any time as they wish, because of the exhausting battery. At the same time, in general video decoding systems, each module consumes a different amount of power and can affect a different rating of video quality. That is, the modules have different contributions in an environment with energy/battery constraint. Therefore, there is a tradeoff between maximum available lifetime of battery and minimum distortion caused by as possible as balanced decoding control. Given that the residual capacity levels of battery can be substantial, it makes sense to schedule modules and perform power management as if the scalable affected was a heterogeneous system. On the other hand, most video decoders nowadays, especially in real time mobile video

168 4 EURASIP Journal on Wireless Communications and Networking applications, are paid more efforts in improving robustness. For example, data partition techniques in H.264, decoder with little redundancy information or with little support from the encoder side. In this case, useful information can be introduced to help decoder. In this environment, there are three high-level control issues. The first is the MB types in coded data; the second is the detailed MB partition information; the third is the effect of human visual properties on single image. Based on these configurations, we present a set of energy-scalable algorithms for video decoding scheduling and energy management, aimed at minimizing power and maximizing video quality. The scheduling algorithms are intended to complement the scheduling criteria produced by the parsing and labeling control, such as priority, and fairness. In the following, we give the detailed parsing and labeling processing MB Type Information. In the first place, MB type information is considered as the primary criterion in decision since an intra MB is decoded without referencing any MB in another picture [27], but may be referred to by other inter MBs. Usually, intra MB is taken for more importance than inter MB. Thus, the intra MB block is marked as L 1 (), the inter MB is marked as L 1 (1), and inter MB in B frame is marked as L 1 (2), which are denoted in (1). It means, from block type aspect, intrablocks and intra frames are assigned and processed in high energy profile comparing with interblocks. In fact, VCV also introduced MB type information as main decoding control term, which had been discussed in Section 2.2, L 1 (), case: Intra MB, ( ) w 1 i, j = L 1 (1), case:inter MB, (1) L 1 (2), case: Inter MB in B frame, where w 1 ( ) is the results of paring MB type information. (i, j) denotes the position index of an MB MB Partition Information. In the second place, the MB partition information is considered as secondary criterion in decision. Each intramacroblock could be classified into several modes including intra 4 4, intra 8 8, and intra Each intermacroblock in P frames could be partitioned into inter 4 4, inter 8 8, inter 8 16, inter 16 8, and inter In a word, there are following partition modes in macroblock, these are 16 16, 16 8, 8 16, 8 8, 8 4, 4 8, and 4 4. Among these, if a block is partitioned into 4 4 mode, then it is the finest block and may be assigned in top level profile; while if a block is in mode, it belongs to coarse block and is in bottom level profile. The MB partition information can be easily extracted after entropy decoding. Thus, the partition information becomes a criterion in assigning the macroblock into different energy profile. Here, for simplicity, we use the energy controlling parameters to mark the blocks or macroblocks so that we can obtain a reasonable distribution in the energy profile. L 2 (x) denotes the controlling level while L 2 (x) [L 2 (), L 2 (4)], and the values corresponding to MB partition information are in the following: L 2 (), 4 4, L 2 (1), 8 4or4 8, ( ) w 2 i, j = L 2 (2), 8 8, L 2 (3), 8 16 or 16 8, L 2 (4), Generally,amacroblockcanberegardedasacombination of basic blocks which belong to different partitions. The basic block is defined in 4 4 block in H.264 [19] and is defined in 8 8inMPEG2[28] andavs[29]. Hence, the marked coefficient for a macroblock is deduced through the partition results of basic blocks. Weighted sum method is adopted in this paper. For instance, a macroblock consists of four 4 4 blocks in top left corner, two 4 8 blocks in top right corner, two 8 4blocksinbottomleftcorner,and an 8 8 block in bottom right corner. Figure 1 shows the partition results. Then, the final effected coefficient which decides the macroblock into appropriate energy profile is 4 L 2 () L 2 (1) L 2 (1) L 2 (2) Effect of Human Visual Properties. In the third place, the effect of human visual properties is considered as third criterion in decision. In many video applications, clients would pay more attention to the regions of their interest. For example, if the shoulder and head video is always existed in video applications, the region of interest (ROI) of clients is usually the human face instead of the background. Thus, for the decoder, more resources including bits and computational power are desired to be allocated reasonably according to the human subjective effects to improve the overall visual quality [3]. From the objective aspect, [31] gave a detailed segmentation strategies for an image. The paper analyses main segmentation approaches for multimedia services from the viewpoint of their features. The first one consists in estimating segmentation scope through the position of the transitions and marks the separation between neighboring regions. This approach has been mainly successful for the temporal case and being applied to both spatial and temporal segmentation problems. The second approach consists in estimating the region through homogeneous elements according to the feature space. This approach has been mostly applied to spatial and spatial-temporal segmentation. Here, we applied the segmentation thoughts and ROI technology to the image region decision. We mark the region in image based on human s attention degree. The technology of ROI is adopted as an efficient tool for the reasonable classification of image; it could be used to divide an image into several parts into different level. When the available battery energy is not enough, the ROI information is used to optimally allocate the available energy to different parts of the image according to their relative level. Since the central region in an image will be concerned firstly according to the habit of human being, (2)

169 EURASIP Journal on Wireless Communications and Networking 5 MB L(4) L(4) L(3) L(3) L(4) L(4) L(3) L(3) L(3) L(3) L(2) L(2) L(3) L(3) L(2) L(2) Figure 1: A example of MB partition information computing. the blocks in central region is allocated to higher energy profile than the surrounding region. As shown in Figure 2, the marks of the human s attention degree are dispersal from central to surrounding regions, then the energy controlling parameters can be marked as (3), where (i, j) denotes the position index of an MB, ( ) L 3 (), i, j interest region, ( ) ( ) w 3 i, j = L 3 (1), i, j sub-interest region, (3) ( ) L 3 (2), i, j normal region. These parsing and labeling configurations provide the sufficient conditions for the following energy-scalable algorithms. Then the energy profile scheduling and energy scalable management rely on the criteria produced by the parsing and labeling control, including priority, and fairness. In the next section, we develop a model of energy-scalable video decoding (ESVD).The overall energy consumption could be optimized after these methods, at the same time the ESVD can guarantee the best video decoding quality in energy constraint circumstance. 4. Energy-Scalable Video Decoding Model In this model, different energy profiles are equivalent to different energy consumption level, and video decoder runs at these profiles. In this scalable energy profiles, the most obvious optimization goal is to maximize performance at a given power or energy budget. Given the complexity or the power budget of this environment, to reasonably design the algorithm for scheduling and for energy or power management, a global optimization solution is required. Section 2 shows possible algorithms to maximize performance at the target power. To simplify the problem, we construct parsing and labeling processing in video decoder in first step, which is given in details in Section 2. These provide the foundation of ESVD. On the other hand, in most video decoding systems, especially for mobile applications, there is a limited system energy supply. Most of the services or functions in mobile devices have estimable power consumption. It means that the upper bound of the consumption can be acquired. Generally speaking, the total consumption is measured by the available battery capacity, that is, the energy consumption is inverse proportion to the available battery lifetime. Strictly speaking, the energy consumption in general video processing applications results from a number of factors, including the number of functions in using regulations, operation systems, hardware, and battery life. Most researches distinguish between two types of power constraints, namely peak constraint and average constraint. Here, we propose another type of power constraints, which is a bound constraint H(F ). We use F {f n } to represent a function, in which f n represents the nth subfunction in function F. H(F ) represents the minimum energy requirement required to implement a function F. For video decoder, a bound of energy constraints also exists. It implies that the optimal energy control method can be obtained when the total energy consumption is deduced by the method tends to the energy bound as closely as possible. Of course, the video decoding function contains many subfunctions such as interpolation (INTP), deblocking filter (DF), entropy decoding (END), and inverse transform (IDCT) [32]. According to bound constraint definition, designing an optimal energy/power consumption video decoding system can be transferred to find the best control among these subfunctions to achieve lower power/energy consumption, so that we can prolong the available battery duration. The above discussion shows the possibility to maximize performance at different target power level. To resolve the problem, we decompose it into two steps. First, we use parsing and labeling processing to map the subfunctionsinunitofmbinvideodecoder,soastogenerate scalable video decoding output. Second, we use power management algorithm to find the best configuration in subfunctions for each power profile and at the same time, maximizes overall performance while keeping lower power consumption Energy Scalable Management in Video Decoder. To compute the integrated weight of MB in order to assign it into appropriate energy profile, the proposed three decision phases in Section 3 are in combined calculation. This needs a mapping bridge between the levels L k (i) in each phase. This problem is solved as follows. Given a set of N subfunctions in video decoding function F in unit of MB, each subfunction canrun at M levels, there is N M power consumption level, correspondingly. Then this problem can be summarized as finding the best selection of power consumption levels for

170 6 EURASIP Journal on Wireless Communications and Networking the subfunctions, at the same time it can maximize the decoding quality subject to the constraints: each scalable power consumptions in whole video decoding is less than E profile(k) in each energy/power profile. Our approach is to reduce the problem to a linear optimization problem. Overall, from parsing and labeling procedure, we map the labeling results on energy/power profiles orderly. To be specific, from the subfunctions, we select the subfunctions in order for MBs and in round robin manner for the whole video sequences decoding. Here, to be simplified, linear weight control Utility Fixed power f (MB) = a 1 L1 (k) + a 2 L2 (k) + a 3 L3 (k), (4) (a) Decoding effect where k 1, k 2 and k 3 represent the effects on the total performance for each phase, separately. We give a simple example firstly. Then the final value can assign the macroblock into appropriate energy profile. For instance, if the video is encoded in AVS, assuming that the initial (a 1, a 2, a 3 ) is (c 1, c 2, c 3 ) in empirical way and L j (k) = k, thereby the maximum marked coefficient for a macroblock is f (MB) = c 1 i+c 2 i+c3 i and the minimum marked one is zero. We can get the marked bound of a macroblock as [, max( f (MB))]. Suitable levels can be classified either in theoretical way or in empirical method, then there are different intervals corresponding to the levels. (, a 1 max( f (MB))) represents coarse level, (a 1 max( f (MB)), a 2 max( f (MB))) denote half accurate level, and (a 2 max( f (MB)), max( f (MB))) is in accurate level, for the sake of clarity, equal configuration is used, that is a 1 = a 2 = 1/3. For example, given a coded frame, after entropy decoding, the macroblock information is extracted as follows, the type is intra, the partition belongs to 8 8, the position lies in central adjacency region, and L j (i) = i still comes into existence. Then the finial marked coefficient is calculated through (4). It means that the labeling energy index of this macroblock belongs to the corresponding energy profile Utility Function in Power Control Scheme. As a frame decoding is composed by subdecoding in unit of MB, MB encoding is also under common resource constrained. Each MB s decoding is a competitor of battery energy for others. On the other hand, PSNR and bit rates are the measurement of decoding quality of all MBs. Ideally, an MB unit would like to achieve normal quality of decoding effect while expending a small amount of energy. In some cases, better decoding effect or long duration decoding and playing are in anticipantion even if the available battery capacity is not enough. For example, most mobile terminals can work in different battery states including Maximum battery life mode, Battery optimized mode, Maximum performance mode, and Enhanced quality mode. Each battery state corresponds to a battery working mode of the device. These states are widely used in mobile devices and terminals. It is desired that video decoder should provide corresponding decoding output to match these working states. Thus, it Utility Fixeddecoding effect (b) Power consumption Figure 2: Utilities as a function of decoding effect and power consumption. is necessary to optimize the video decoding process under battery resource constraint. Obviously, it can be transformed into a kind of tradeoff between obtaining better decoding effect and obtaining lower energy consumption in corresponding working state. Finding a good balance between the two conflicting objectives is the primary focus of the power control component of resource management. This tradeoff is illustrated through the conceptual line in Figure 2. If the decoder power is fixed, the terminal would experience high decoding effect which leads to increased reasonable allocation of the system resources. If the decoding effect and quality is fixed, increasing the power consumption expedites the battery drain, which reduces the effective use of the mobile terminal. The optimal power control algorithm for video decoding systems should maximize the decoding quality. Traditionally, the object is to achieve acceptable PSNR as the measurement of decoding quality. However, this single target is not enough for efficient video decoding. This is because the object on power consumption is another important factor in applications. It is clear that a high PSNR level at the decoding output will result in better decoding effect. However, achieving a high PSNR level often requires the terminal to work in high power consumption state, which, in turn, results in low battery life. These issues can be quantified by defining the

171 EURASIP Journal on Wireless Communications and Networking 7 utility function of an MB decoding unit, which is defined as u i,j = Δe i,j ΔPSNR i,j mwh db, (5) where Δe i,j = E normal (i, j) E profile (i, j) andδpsnr i,j = PSNR normal (i, j) PSNR profile (i, j). For MB(i, j), E normal (i, j) represents the battery power consumption of the decoder in normal state, while E profile (i, j) means the battery power consumption in corresponding energy profile. Accordingly, PSNR normal (i, j) is the quality in full decoding state, while PSNR profile (i, j) represents the decoding quality in corresponding energy profile. Utility as defined above combines the decoding quality and power consumption. The efficiency function yields the desirable properties. Assuming perfect case ΔPSNR i,j and Δe i,j means the decoder is under the full-state decoding. The mobile terminal can work in Maximum performance mode or Enhanced quality mode. In this case, the decoding quality will obtain maximum value. On the other hand, u i,j is a monotonically increasing function of the Δe i,j. That is, in case of fixed target power consumption e i,j = E target, for decoding schemes, the best strategy for MB encoding is to make a decision for each subfunction, so as to acquire maximum utility u i. This suggests that, in order to maximize utility, all MBs in the video decoding system should try to improve the decoding effect while as possible as less consume the energy. So that the utility function is suitable for measuring power efficiency of video decoding systems Energy Allocation Scheme Based on Macroblock Tracking. As mentioned above, most mobile terminals provide many working states such as Maximum battery life mode, Battery optimized mode, Maximum performance mode and so on. Accordingly, supposing that video decoder can provide corresponding decoding output to match these working states. Each energy profile E profile(k), k N corresponds to a decoding level. Then the goal is to adjust the decoder state in unit of MB to obtain best decoding quality under energy consumption budget E budget(k). Following the arguments in (5), there is max U ( i, j ) s.t. i j ( ) E MB i, j Ebudget, i j where i = width frame /width mb and j = height frame /height mb. For example, if there is video decoding data in CIF format, then width frame = 352, height frame = 288, width mb = height mb = 16, and so forth. From the discussion above, all MBs in a frame are parsed and labeled into different scalar quantity, here we use ŵ(i, j) to represent the final labeling result of each MB. Then the MBs in a frame can be allocated into different energy profile levels according to their labeling results. As the decoder is divided into several levels in unit of MBs, we relate these MBs with different decoding state (6) to realize fine allocation. Define that the number of decoder states is γ, and then it is obviously that the number of MBs is usually unequal to γ. This leads to an optimal problem. That is, we should configure these MBs into suitable decoding states to obtain better decoding quality. From (6), we have max U ( i, j ) s.t. i j ( ( )) w level(γ) m level(γ) E MB i, j level γ Ebudget. γ i j (7) As mentioned above, we classify these MBs into three levels for the sake of simplicity and define that the MBs in the same level has the same energy budget. For each MB in level 1, let the energy budget is e 1, accordingly, for each MB in level 2 and each MB in level 3, the energy budget be e 2 and e 3, separately. Then (7)canberewrittenas max U ( i, j ) s.t. i j 3 w level(γ) m level(γ) e γ E budget. γ= Decisions Using Learning Method. As we known, it is difficult to obtain the accurate correlation between PSNR and energy consumption level. Thus we use machine learning tools [33] to exploit the correlation and derive decision table to classify the MBs into corresponding decoding levels. Machine learning method refers to the study of decoding states to acquire knowledge from experiences. It deduces new knowledge from existing rules and uses the analysis of a set of experiments or examples, for creating a set of rules to take decisions. Thus, the correlation problem is posed into two sub-problems: one is to collection the variation of PSNR and energy consumption in each decoding state; the other is to classify these data into suitable modes according to their utilities. In next section, we give the detail of subfunctions design in unit of MB. And carry out a performance evaluation of each subfunction in terms of its variation of PSNR and the variation of PSNR and energy consumption results. 5. Implementation During Video Decoding The overall energy consumption could be optimized after these methods; at the same time the ESVD can guarantee the best video decoding quality in energy constraint circumstance. For the sake of clarity, the whole energy constraints are summarized as the total summation of each function which the applications support. In practice, the functions cover a variety of applications. In contrast, the average power constraint can be imposed on the overall consuming power in universal user application circumstance. Motivated by the previous discussion that all macroblocks are classified into several energy/power profiles, we (8)

172 8 EURASIP Journal on Wireless Communications and Networking Resource and available capacity-aware Parsing and labeling Coded data Entropy decoding Scalable IDCT + Scalable deblock filtering Scalable error concealment Output Intraprediction Frame buffer Switch Scalable inter prediction Figure 3: Illustration of the power scalable control video decoding system. designadeviceresourceperceptualmodule.thismodule implements a mapping bridge between the energy profile and the device available resource. This module includes two part functions. Part 1, user can specify the working state of video service. These states include Maximum battery life mode, Battery optimized mode, Maximum performance mode, and Enhanced quality mode. As mentioned above, each state corresponds to a battery working mode of the device. Part 2, to automatic adapt the working state of video service according to remaining battery capacity perception. For instance, Maximum battery life mode can be configured automatically when the residual capacity is under 3%, while Enhanced quality mode adopted automatically in the case of available battery capacity is above 8%. To be specific, when the result of part 1 and result of part 2 are not matched, that is, user configures the device as Enhanced quality mode but the residual capacity is under 3% at that time, the final available profile is based on perceptual remaining battery capacity results. That is to say, part 2 has higher priority than part 1, and user can manually specify the working state only when the device resource is sufficient. It is widely accepted that END, IDCT, INTP, and DF are the four main subfunctions in universal video decoder. Consequently, the following discussion is based on these four subfunctions. The implementation of each energy profile is described in Figure 3, and the modules are listed as follows IDCT SubFunction. The complexity of IDCT subfunction in decoder has closed relation with the inner non-zero parameters. Researches provide many scalable methods, for example, [34] using different proportion subrectangles in blocks to output scalable computation IDCT. In general, the energy of the DCT coefficients is dissipated among the zigzag scan of the block. The low-frequency component in left-upper corner has higher energy, while the highfrequency components in right-lower corner contain lower energy. Thus, we progressively omit the data along the inverse zigzag scan, from right-lower corner to the left-upper corner, so that obtain minimal output quality degradation and at the same time achieve scalable energy consumption. Here, we classify the energy profile in IDCT subfunction into four degrees, including accurate-level, saving-level, coarselevel and non-idct. When accurate-level is selected, the whole parameters computation is implemented as shown in Figure 4(a). Many simplified methods can be used such as 1D IDCT optimization so as to minimize the energy consumption possible as. Figures 4(b) and 4(c) show the cases of optimal-level and matching-level, separately. The main difference between the two levels is the number of computing parameters. The number implies corresponding processing levels Motion Compensation and Interpolation SubFunction. Motion and residual information is generated from compressed bits after entropy decoding. Interpolation of reference samples to generate a motion-compensated prediction is generally performed for each macroblock that is intercoded [12] and occupies most complexity in motioncompensated prediction. Thus, the average time required by the interpolation subfunction is approximate to a function of the number of intercoded macroblocks. The most straightforward approach to classify this subfunction is to fully interpolate and fully compensation operations. In this level, quarter-pixel motion compensation is replaced by halfpixel operations, it forms a saving mode with little quality decline while computation is saved. Accordingly, substituted interpolation modes in unit of half-pixel and integral-pixel compensation by integer interpolation results are adopted in the other energy profile, separately Deblocking Filter SubFunction. Deblocking filter which is often referred to as a loop filter is the final stage of the decoding process. DF subfunction reduces the blocking effect that is introduced by encoding the process at block boundaries. Comparatively high complexity of the subfunction is in consensus. Even after a tremendous effort in speed optimization of the filtering algorithms, the filter can easily account for one-third of the computational complexity of

173 EURASIP Journal on Wireless Communications and Networking 9 (a) (b) (c) Figure 4: Data pruning patterns in IDCT. 5 9 Influence on energy consumption (%) Influence on PSNR (%) Decoding states Decoding states Mother Waterfall Tennis Ship Bus Paris (a) Influence on energy consumption Mother Waterfall Tennis (b) Influence on PSNR Ship Bus Paris Figure 5: Influence on energy consumption and PSNR under different decoding rules. a decoder [35]. The complexity is mainly based on the high adaptivity of the filter, which requires conditional and decisional processing on the block edge and sample levels, thus, there are many conditional branches in the filter which leads to excessive power consumption. At the same time, for a macroblock, the vertical filter begins from left-most edge and is followed from left to right by the three vertical edges; besides, the horizontal filter begins from top edge, and is followed by the three internal horizontal edges from top to bottom. Amount of relevant and candidate pixels should be loaded into the memory, this leads to additional power consumption either. Scalable energy can be achieved by classifying the filtering process into three levels, including full, half, and rough filtering. Among these, full filtering operation means that overall branch filtering is implemented for the macroblock. And, half filtering represents the operation reduced in computational complexity, which can be achieved by taking into account the fact that the image area in past frames is already filtered, and thereby optimizing or omitting the filtering process accordingly. For the rough filtering, skip operation is used with low quality degradation, while the lowest power consumption of the DF subfunction is required in this mode. Besides, learning tools are used to analyze the data sets of decoder. The decision table will be used to determine the decoding modes of an MB. Inductive learning uses the analysis of data sets for creating a set of rules to take decisions. Then a decision table is built as the decoding rules. This table is from a set of experiments or examples, collected as the training data set. We build information database to gather the decoding states. This set of data including the following properties: () full decoding mode; (1) decoding without deblocking filter mode, which corresponds to deblocking filter subfunction adjusting, (2) quarter pixel interpolation is compensated by half-pixel interpolation, (3) quarter pixel interpolation and half-pixel interpolation are both compensated by integer-pixel interpolation; these two cases are corresponding to motion compensation and interpolation subfunction adjusting, (4) data pruning pattern in IDCT complies with saving-level, (5) data pruning pattern in IDCT follows coarse-level; (6) data pruning pattern in

174 1 EURASIP Journal on Wireless Communications and Networking Energy consumption (mwh) Energy consumption (mwh) Energy consumption (mwh) (a) Sequence mother (b) Sequence waterfall (c) Sequence tennis PSNR (db) PSNR (db) PSNR (db) (d) Sequence mother (e) Sequence waterfall (f) Sequence tennis Energy consumption (mwh) Energy consumption (mwh) Energy consumption (mwh) (g) Sequence ship (h) Sequence bus (i) Sequence paris PSNR (db) PSNR (db) PSNR (db) (j) Sequence ship (k) Sequence bus (l) Sequence paris Figure 6: Stat. on energy consumption and PSNR for different video sequences in each decoding rule. IDCT follows low-level; these three cases are brought into correspondence with IDCT subfunction adjusting. Figure 5 gives the influence on energy consumption and PSNR under different decoding rules, separately. Affiliated subfunction: discussion on error concealment subfunction. Error concealment technique aims at obtaining a close approximation of the original signal or making the output of decoder closely accepted by human eyes [36]. Most error concealment techniques are based on block matching algorithms [37] or adaptive techniques in unit of block such as [38]. It can improve the decoding quality while it leads to less computational complexity. Due to the energy consumption which lies in computation, memory occupation and memory access, the effect of error concealment on additional power consumption is more than that on complexity. Here, we classify the error concealment operation into three levels to adapt the scalable energy profiles. This classification is based on scene and region change and on the unit of block. Thus, the macroblock can belong to three energy profiles, including accurate concealment in case of scene change, half concealment in case of regional variability, and coarse concealment when few and no movements take place. Reference [39] gives an analysis of H.264/AVC decoder in computational complexity, and [12] presents detailed analysis in both computational complexity and memory occupation complexity. For the aspect of the complexity in AVS video decoding, [32] is provided an approximate estimation. Generally speaking, for most video decoders including H.264, MPEG4, AVS, and so forth, the computational power allocation with emphasis on power-distortion (P-D) [1] canbeexpressedinformofcostfunctions.we take power consumption in video decoding into account by modifying the power-distortion-complexity (P-D-C) cost functions in processing unit of macroblock and subfunctions

175 EURASIP Journal on Wireless Communications and Networking Energy consumption PSNRY (db) Scalable ratio (%) Scalable ratio (%) (a) Energy Consumption in each decoding modes (b) PSNR in each decoding modes Figure 7: Influence on energy consumption and PSNR under different decoding modes. in decoder. Through the objective function in (8), dynamic scalable assignments provide a local quality optimum in each energy profile. Consequently energy scalable video decoding (ESVD) is achieved. An undeniable fact is that scalable video decoding leads to the quality degradation. Thus minimizing this degradation is another purpose in ESVD. 6. Experimental Results 6.1. Building Energy Consumption Information Database. In this subsection, we use Application Energy Graphing Tool [4], which can measure the battery power consumption of an application over time, log and graph the resulting data. We use it to profile the energy distribution of the decoding modes. To calculate the energy consumption in the case of subfunctions modes, we assume that all other possible operations among the subfunctions are running, expect the testing mode. It means it will occur in power control schemes in practices that decoding data will be ergodic to all basic subfunction units in despite of some skipped or simplified operations. The reason is that compressed video data includes multifeatures, thus the decoding process varies with these features. For instance, for the same decoding program, the decoding time is different among the typical sequences such as mother, waterfall, tennis, ship, bus, and paris. Thus we use the typical video sequences as the test video set. The format is CIF and coded in AVS standard. We recycle the decoding process until the number of decoding frames obtains 15 frames in each sequence. Figure 6 shows the total energy consumption and corresponding PSNR in each decoding rule. The results are based on statistical experimental average. The decision table will be used to determine the decoding mode of MBs, based on the information gathered during the preanalysis of the decoder. This process can be more accurate by the information update during the decoding stage. Figure 5 depicts the process for building the decision tables from the results in Figure 6. For example, when the decoder works on mode (1), decoding without deblocking filter mode, little PSNR is lost but about 15% energy consumption saving can be obtained; when the decoder works on mode (5), data pruning pattern in IDCT follows to coarse-level, only around 1% energy consumption saving can be obtained but 85% PSNR losing occurs that is, when the energy budget is not full enough to support fullmode decoding, mode (1) is a better choice than mode (5) The Performance of the ESVD Model. To evaluate the performance of the ESVD model and the energy scalable video decoding system, we implement the proposed ESVD model and energy scalability scheme in the AVS decoder software. The ESVD model is not limited to the video coding standards, and thus similar performance can be expected for other coding systems, such as H.264 and MPEG-4. We select stochastically waterfall CIF sequence at 128 kb/s and 25 fps as the testing sequence. We performed two sets of evaluations one is for evaluating decoding scalability and the other for evaluating scalability quality. We let the decoder work under four modes. The energy consumption budgets are descending. The scalable results including PSNR and energy consumption shown in Figures 7 and 8 show the subjective quality in different decoding modes, separately. Each mode corresponds to energy consumption budget ratio compared to the full decoding mode. These experiments show the scalability and efficiency of ESVD. 7. Conclusion and Future Work This paper proposed ESVD framework in power control video decoding systems. It aims at providing the scalable decoding output which is adaptive to energy resource.

176 12 EURASIP Journal on Wireless Communications and Networking (a) In 9% budget mode (b) In 8% budget mode (c) In 7% budget mode (d) In 6% budget mode (e) In 5% budget mode (f) In 4% budget mode (g) In full mode mode Figure 8: Subjective qualityin thedifferent scalable modes. It proposed a method to make the video decoder adapt resource under battery constraint, which can be widely used in handheld devices. At the same time, it gives a method to maximum video decoding quality when playing on portable terminals, through building a decoding information database. The experiments demonstrate the efficiency of ESVD. In future research, we will try to study fine-grained energy scalable control in energy consumption through improving the scalability of each decoding module.