High Definition Video in IEEE c mm-wave Wireless Personal Area Networks

Transcription

1 36th Annual IEEE Conference on Local Computer Networks LCN 2011, Bonn High Definition Video in IEEE c mm-wave Wireless Personal Area Networks Sandra Scott-Hayward and Emiliano Garcia-Palacios Institute of Electronics, Communications and Information Technology (ECIT), Queen s University Belfast, Belfast, BT3 9DT, Northern Ireland sscotthayward01@qub.ac.uk, e.garcia@ee.qub.ac.uk Abstract Multimedia applications place high demands on resources in wireless networks. The 60 GHz frequency band has been adopted to support high data rate applications in wireless personal area networks (WPANs). In the IEEE c standard for millimeter-wave-based high-rate WPANs, devices are allocated a dedicated transmission time called a CTA (channel time allocation) based on their request to transmit. While the duration of a CTA may be adjusted, the standard does not specify how resource allocation should be achieved. High-Definition (HD) video with variable bit rate (VBR) is a potential application for the WPAN. In this work, an IEEE c network is simulated and the throughput losses due to static resource provisioning for a series of H.264/SVC (MPEG-4 Part 10) HD video sources are identified. It is demonstrated that the high data rates proposed for the 60 GHz frequency band will not be achieved without dynamic resource allocation. I. INTRODUCTION The rapid evolution of multimedia applications has dramatically increased the requirement for high speed wireless communications. Current wireless technologies support multi-mbps data rates over varying ranges. However, many applications require Gigabits per second, not Megabits per second. An example is HD video streaming. The data rate requirement for an uncompressed HD video stream 1920*1080 (frame size) with 24 bits per pixel (colour support) and 60 frames per second (frame rate) is approximately 3 Gbps. Current standards in frame size, colour support and frame rate give a range of HD data rates from 0.5 Gbps to 5.5 Gbps [1]. Other widely discussed applications requiring Gbps or multi-gbps data rates are rapid upload/download file transfers, wireless gaming or projection and downloading movies/photos from a camera. A complete list of the usage models for high data rate communication is included in [2]. At the core of providing high data-rate, high quality of service (QoS) applications is bandwidth. In 2001, the Federal Communications Commission (FCC) allocated 7 GHz of spectrum in the GHz band for unlicensed use. With up to four 2.16 GHz channels available, the target applications finally seem realisable. The propagation characteristics at this mm- Wave frequency are, however, quite different to those in the lower frequency bands. The 60 GHz band supports high data rate, short range, line-of-sight (LOS) directional transmissions, which falls into the category of WPANs. A mm-wave alternative PHY (physical layer) for the IEEE standard (Wireless Medium Access Control (MAC) and Physical Layer (PHY) specifications for high rate wireless personal area networks (WPANs)) was approved in September It is known as IEEE c-2009 (Amendment 2: Millimeter-wave-based Alternative Physical Layer Extension) and supports new data rates, with the highest of these greater than 5 Gbps (at PHY) [3]. While the IEEE c standard specifies PHY modes capable of achieving such data rates at the physical layer, the actual throughput at the MAC layer in a network is always less than the PHY throughput. This is due to overheads introduced at the MAC layer for data packetising and error-checking, for example. In addition, and quite significantly, when multiple devices attempt to transmit in a shared medium, they are competing for the limited bandwidth available. Collisions occur when devices try to access the medium simultaneously and this results in loss of data packets, waste of bandwidth and, as a result, reduction in MAC throughput. The IEEE c standard defines improvements to the existing MAC to support the new mm-wave PHY. For example, frame aggregation techniques and block acknowledgment are introduced as means to reduce the MAC overhead. In parallel to the IEEE standardisation work, a group of technology and consumer electronics companies known as the WirelessHD TM Consortium defined a Wireless Video Area Network (WVAN) specification [4]. The main attribute of the WirelessHD system is the transport of the HD audio/video stream with high quality of service within a room at a distance of up to 10 m. An updated specification, WirelessHD 1.1, was released in May 2010 extending the technology to improve video support and include data transfer creating both WVAN and WPAN. While current demand for high-speed point-to-point data transfers are driven by High Definition Multimedia Interface (HDMI) cable replacement, consumer demand for high-speed communications anytime, anywhere will quickly lead to multiuser, multi-application scenarios. The multi-user scenario (Fig. 1) considered in this work could be in a living room, a conference room, a hot-spot or an aircraft cabin, for example. In this context, multiple high data rate applications will have to co-exist. Several research questions arise when characterising such coexistence. In an attempt to satisfy bandwidth demands, terminals will compete for bandwidth in a selfish manner. How will devices determine what bandwidth they require? /11/$ IEEE 93

2 On-Line Gaming Projection to TV Wireless Audio Fig. 1. Content Download HDTV Piconet Controller Video Camera mm-wave WPAN Architecture On-Line Gaming What happens when their requirement changes? How are these varying transmissions to be scheduled to use the bandwidth most efficiently? Can service provision be guaranteed and with what quality of service? Performance analyses of the IEEE c standard have to date focussed on simplistic traffic models. Multiple devices in a network transmit the same amount of data at a constant rate enabling static scheduling of a CTA for the duration of the transmission [5], [6]. However, in reality devices will have heterogeneous traffic, which will likely vary over time requiring bandwidth adjustment. In this work, such realistic traffic is modelled demonstrating the impact of variable demands on the IEEE c protocol. We consider the devices in the network to be sending HD Video but other variable bit rate (VBR) traffic combinations from the applications detailed in [2] could equally be considered. The rest of this article is organised as follows: in Sect. II the IEEE c MAC protocol is introduced. In Sect. III we highlight related work in this area. In Sect. IV VBR video traffic is discussed. Section V includes description of the simulation parameters and presentation of results. Finally, in the conclusion the results are summarised and our proposed further work on this topic is identified. II. IEEE C MAC PROTOCOL An c network, or piconet, is a wireless ad-hoc data communications system in which a number of independent devices communicate with each other in a peer-to-peer fashion [3]. In the c protocol, medium access is controlled by a piconet controller (PNC). This role can be held by any device in the network and is usually allocated to the first member of the network. Medium access is based on a frame structure, as shown in Fig. 2. The superframe is initiated by a beacon from the PNC. A contention phase known as the contention access period (CAP) follows the beacon. It is essentially a CSMA/CA (Carrier-Sense Multiple Access/Collision-Avoidance) phase for command and data exchanges using the Binary Exponential Backoff (BEB) mechanism. In c, devices compete during the CAP to transmit a request for dedicated channel time during which the application data can be transmitted. The time slot is called a CTA (channel time allocation) and occurs during the channel time-access period (CTAP), which follows the CAP. The CTAP operates under the TDMA (Time Division Multiple Access) principle. In this phase, a time slot or CTA is assigned to any devices that have requested an access period, provided the time is available. For asynchronous data e.g. a one-off file transfer request, the assignment of channel time is indicated in the subsequent beacon. The asynchronous request is assigned a single CTA for the total amount of time needed for the data transfer. The assignment is scheduled in the superframe on a best effort basis. If additional data is to be sent, the user must request another CTA. For isochronous stream requests such as a video stream transmission request, the PNC replies to the CTA request in the CAP with a CTA response. If resources are available, the isochronous request is then assigned a CTA in each subsequent superframe for the duration of the video. The isochronous stream does not expire until it is terminated either by the source device, the destination device or the PNC. In order to improve the MAC efficiency, frame aggregation and block acknowledgment are supported by the standard. Frame aggregation involves fragmenting and mapping MAC Service Data Units (MSDUs), as received from the upper layers, into multiple subframe payloads. The sub-header for each subframe is combined to form the MAC sub-header and up to 8 subframes are aggregated into a single frame thus increasing throughput. Standard and low-latency aggregation mechanisms are defined [3]. The low latency version limits transmission delay under low traffic load by transmitting empty subframes of zero length if sufficient MSDUs are not received. A block acknowledgment (blk-ack) is sent for the aggregated frame. Each subframe is acknowledged within the blk-ack. The standard aggregation method along with the blk-ack frame is shown in Fig. 2. The benefit of these mechanisms is improved efficiency by reducing frame/ack overhead. However, the loss of a large frame can be critical to quality of service so that when selecting subframe- and frame size, the wireless channel condition should be considered [7]. III. RELATED WORK As presented in Sect. II, the IEEE c standard has a hybrid medium access protocol, where hybrid refers to the combined CSMA/CA(CAP)-TDMA(CTAP) MAC. The coordination between the CAP and the CTAP for the (2.4 GHz) protocol is studied in [8]. Network performance is improved by varying the retry limit and the length of the CAP and allowing simultaneous transmissions between non-interfering devices. 94

3 MSDU 1 FCS MSDU 2 Fragment 1 FCS MSDU 2 Fragment 2 FCS Aggregated Frame Subframe 3 Subframe 1 Subframe 2 Subframe 8... HCS Subheaders HCS MAC Hdr PHY Hdr Preamble Blk- Isochronous CTA request FCS sf 1 sf 2... sf n MAC Hdr PHY Hdr Preamble Req Resp SIFS SIFS SIFS Asynchronous CTA request Frame Frame SIFS SIFS SIFS... Req SIFS B e a c o n Contention Access Period MCTA MCTA CTA 1... CTA n Channel Time Allocation Period Superframe Fig. 2. Superframe structure, Frame aggregation and Block acknowledgment in IEEE c For the mm-wave IEEE c standard, Pyo and Harada [5] consider the overall network throughput based on contention in the CAP for transmission of CTA requests and the resulting allocation of CTAs for high-speed data transmission in the CTAP. The effect of varying the time allocated to the CAP is studied. The CAP must not be so large as to cause low data transmission time in the CTAP, nor so small as to limit successful transmission requests. Throughput improvement is proposed by providing an optimum access time for the CAP based on the number of devices in the network and through a private channel-release time, which provides a dedicated time for the device to send its CTA release request. The latter method avoids the possibility of collision/delay while contending to send the CTA release request during the CAP, which could result in an unused CTA for one or a number of superframes. However, this is achieved at the expense of bandwidth used by introduction of the dedicated private channel-release time. A number of performance analyses demonstrate the benefit of implementing frame aggregation and block acknowledgment in IEEE c [9] [12]. A 30% MAC efficiency gain by using blk-ack or NACK for the aggregated frames is presented in [9]. In [10] the results show that using the standard frame aggregation mechanism with a large subframe length improves system throughput. In [11] the authors consider the limitation that the PHY frame size places on the aggregated frame size. The proposal is to overcome this by adding a PHY frame sequence number to the MAC header such that an aggregated frame can be transmitted via multiple PHY frames. A combination of frame aggregation, Dly-ACK (similar to the blk-ack mechanism), and an Improved-Back-off Algorithm (IB) is used in [12] to increase the throughput of the WPAN MAC. The balance between time allocated to the CAP and the CTAP in the superframe has an impact on the upper layers of the protocol stack. For the specific example of TCP (Transmission Control Protocol) flows, this performance dependence is analysed in [13], [14]. The problem for TCP relates to the round-trip time (RTT). As a device needs to wait for the CAP to send a request (in this case the SYN packet), in the worst case it may need to wait for the complete duration of the CTAP. The same delay is also possible for the device to receive the ACK. As a result, the congestion window growth rate can be slow leading to underutilisation of the CTA. The problem exists for any delay-constrained application. A request for a CTA is made in one superframe and the allocation of the CTA, if possible, is provided in the subsequent superframe. Only limited solutions to the problem of contention for allocated channel time are presented in the literature, as described above. It is one outstanding issue. Secondly, high data rate applications are proposed for the 60 GHz WPAN. However, the models developed to date and discussed here predominantly analyse the performance of the WPAN MAC when the techniques of frame aggregation, block acknowledgment and simultaneous transmission are exploited. While these are useful in terms of increasing the data throughput for a given bandwidth allocation, the consideration of constant frame length/frame payload at each device is limiting. The variety of multimedia applications identified in Sect. I have quite different traffic profiles, which are not realistically represented by constant bit rate traffic. In this study we implement the more realistic variable bit rate traffic. The specific case of HD video is explored in the next section. IV. VBR VIDEO TRAFFIC REQUIREMENTS Operating at a high PHY data rate, such as possible at mm-wave frequencies, the transmission of uncompressed HD video is possible in a WPAN. The advantage of uncompressed 95

4 TABLE I STATISTICS OF VIDEO TRACES Mean Frame Peak Frame Frame Size Bit Rate Bit Rate Peak/Mean (Mbps) (Mbps) Ratio Transporter Blue Planet Finding Neverland Speed The LakeHouse video is that the protocol can be designed around constant bit rate transmissions. In terms of allocating radio resource, this simplifies the scheduling task as a certain data rate is requested and used throughout the transmission. However, given the bandwidth requirement for uncompressed video ( Gbps [1]), this does not leave much resource available for other applications in the same network. In contrast, using an encoding scheme to compress the video means that good video quality can be provided at substantially lower data rates. This is based on the difference in coding of the different frame types (intra-coded (I), inter-coded (P), and bidirectional coded (B) frames). Using a coding scheme such as H.264/MPEG-4 Part 10 in which the frames are coded with forward and/or backward prediction can reduce the bandwidth requirement to between 64 kbps and 240 Mbps, depending on the application. For HD formats (Level 4), the maximum compressed bit rate is Mbps [15]. To produce nearly constant quality video, quantisation parameters are fixed (compression without rate control) but the video traffic is highly variable as a result. The difference in peak and mean frame bit rate for the set of video traces [16], [17] used in the simulation is shown in Table I. Clearly, a simple, constant CTA as used with uncompressed video will not suit the VBR of compressed video. The question is how much bandwidth is wasted in this scenario and how can the network efficiency be improved. In order to satisfy the QoS of multiple device requests with heterogeneous traffic, bandwidth must be dynamically allocated. For example, allocating bandwidth to match the peak bit rate should be avoided as this will lead to unused bandwidth during many portions of the transmission when the actual bit rate is well below the peak. This is highlighted in [18]. As a means to optimise bandwidth usage, for a TDMA system when time-slots are allocated and the full time-slot is not required for the current frame transmission, it is proposed that the remaining time be offered for use by other transmissions [19]. This re-allocation can be moderated by the access point in an infrastructure network or by a nominated controller in a distributed network. The increase in network throughput would have to outweigh the cost of managing the re-allocation. Some resource management methods have been proposed. In [6] a method to support Internet Protocol TV (IPTV) flows, which require high data rate and high QoS is presented. This method takes advantage of simultaneous transmissions using directional antennas. The successful transmission of multiple flows of a HDTV format video is demonstrated in the simulation. However, competition for resource from other applications with different parameters is not considered. A set bandwidth is reserved per flow, which does not focus on the variance in frame size within that flow. Resource allocation for this type of traffic has also been explored in [20], [21] using variable step-size least-mean square traffic prediction and a scene change detection algorithm. Results are presented for a single device transmitting a VBR video to another device and do not account for contention with other devices competing for the available bandwidth. Two issues can be identified with this proposal. The first is that the device sends a request in every CAP (or MCTA - Management CTA - for increased reliability) identifying its bandwidth requirement for the next superframe. For a large number of devices in the network, this will significantly reduce the available data transmission time. Secondly, what happens if the requested bandwidth is not available in the next superframe? In this work we study the transmission of compressed HDvideo using the IEEE c protocol. The impact of the VBR traffic on the network throughput is identified and the requirement for a dynamic scheduling mechanism to support this is highlighted. This contrasts with the static scheduling or single device resource allocation assumed in previous research. V. SIMULATION RESULTS AND PERFORMANCE ANALYSIS A MATLAB simulation of the IEEE c MAC protocol has been developed by the authors. A network of a PNC plus N devices is created representing a WPAN such as that pictured in Fig. 1. In the c standard a common mode signalling is defined based on the single carrier (SC) PHY, supporting coexistence and interoperability. The mandatory modulation and coding schemes are used (see Table II). An ideal channel is assumed. This means that the packets dropped in the contention access period can be attributed to collisions alone and are not a result of varying channel conditions. The traffic in the simulation is a set of H.264/SVC (MPEG-4 Part 10) HD video sources [16]. Five video traces have been selected to represent a variety of content e.g. Action/Thriller, Nature/Documentary and Drama. The peak-to-mean ratio of the frame sizes reflects the traffic variability. MPEG-4 is selected for its high compression efficiency and suitability for HD content transmission. Some statistics of the video traces are listed in Table I. The video traffic is represented by a stream of frames of varying size, as produced by the MPEG-4 codec. This traffic is then packetised into RTP (Real-Time Protocol) packets as described in RFC 3016 [22]. For the high data rate, mm- Wave network represented, one frame or fragment of a frame is contained in one RTP packet. The RTP header is 12 bytes. This traffic is then encapsulated in 4096-byte UDP packets, which introduce an additional 8-bytes header. The value of

5 Parameter TABLE II PARAMETERS FOR THROUGHPUT ANALYSIS Value Number of devices in piconet, N 2-40 Preferred Fragment Size 4096 Bytes Data Rate 1.65 Gbps Basic Rate, Beacon Rate 12.5 Mbps, 25.8 Mbps Length of Beacon (27+(7*N)) Bytes SIFS, MIFS 2.5 μs, 0.5 μs CCA Time 4 μs Channel Time Request, T req μs Channel Time Response, T resp μs Length of Imm-, T imm ack 10 Bytes Length of Blk-, T blk ack 7+(2*No. Subframes) Bytes Length of frame, T frame Preamble (1.96 μs) + Header (9.6 μs) + T payload bytes has been chosen to take advantage of the larger fragment size (maximum of 1 MB) allowed in the c protocol, which reduces overhead, but also to limit the impact of frame loss, as identified in [7]. With large frame sizes, higher SNR (signal-noise ratio) is required to maintain a certain PER (packet error rate). The IP packet is created by adding a 20-byte IPv4 header to the UDP datagram. Typically, the encoded audio is packetised independently from the video so that two separate streams are used to transport the data in an IP environment. MPEG-4 High Efficiency Advanced Audio Coding (HE-AACv2) can be used with H.264 video and can offer good audio quality at kbps Mono [23]. In the simulation, a 5-channel audio stream of 120 kbps accompanies each video. The IP packets are the MSDUs (MAC service data units). In accordance with the standard aggregation technique, as described in Sect. II and shown in Fig. 2, the MSDUs are fragmented or mapped into subframes. The subframe is based on the preferred fragment size identified by the device when associating with the piconet. In this work it is set to the UDP/IP packet size. Provided the frame deadline will not be exceeded, the frame can also be split across multiple superframes/ctas. The MPDU (MAC protocol data unit) is produced at the MAC layer ready for transmission at the PHY layer. The parameters used in the simulation are listed in Table II. To study the throughput of VBR traffic in the c WPAN, each device has a video to transmit and sends an isochronous stream CTA request. The CTA request is based on the mean bit rate of the relevant video, as per Table I, and calculated with respect to the superframe duration. A device would not know the mean bit rate of the video in advance but this value is used for comparison. The video traffic is generated from the video trace [16] and packetised as described previously. Repeated simulation runs of 15 s duration were made. For each of the selected videos, the expected throughput, which is based on N devices successfully receiving a CTA to transmit at the mean bit rate is plotted. This is compared with the actual throughput, which is based on N successful devices transmitting within their allocated CTA according to the VBR of the video trace. The results are shown in Fig. 3. The maximum number of devices that can be allocated a CTA under the given system settings varies but it is seen that for a video such as Transporter2, the limit is 28 devices/flows. For the lower mean bit rate videos, such as The LakeHouse, a limit is reached at around 55 devices/flows. The reduction in actual throughput compared with expected throughput varies between 10% and 30% across the five videos. The difference can be attributed to the variation in frame size leading to either allocated CTA time being underused or insufficient. The penalty for inappropriate provisioning is both waste of bandwidth, due to underused CTAs, and loss of quality, due to frames that can t be transmitted within their deadline being dropped. Buffering could, of course, be used to compensate for the delay in transmitting frames, rather than dropping them. However, we present here a generic platform, which would satisfy real-time transmission requirements for the applications identified in [2]. As such, if the frame deadline would be exceeded, the frame is dropped. This enables some quantification of the effect of the static CTA allocation on the QoS of the application. Figure 3 illustrates the case of homogeneous traffic in the network (all N devices use the same video trace). Simulations of random video traffic in the network have also been performed. Over the course of a 1 minute simulation, each of the 40 devices in a network attempts to send a video flow. The start time, end time and duration of the flow are random. Similar to the single video examples, once a device s request to transmit is successfully received and provided there is sufficient time available in the CTAP, the CTA is assigned. The CTA duration is based on the request made by the device, which is, as previously, the mean bit rate of the particular video. In Fig. 4a the difference in actual and expected throughput over the course of the simulation is shown. The average difference in throughput is 12%. Also illustrated in Fig. 4b is the effect on the video application of setting a static CTA duration based on the mean bit rate. The percentage of dropped frames varies between 1% and 60%. While some loss of B and P frames in the transmission may be tolerated, losing an average of 5% of I frames will significantly reduce the quality of the video transmission. These results show that the allocation of resources poses a significant challenge if quality of service and bandwidth efficiency are to be achieved. As described previously, if a video frame is too large to be transmitted in the CTA provided, it is transmitted across multiple CTAs so long as the frame deadline will not be exceeded. The superframe duration/size is key to meeting the frame deadline. If a video frame is to be split across multiple superframes, then the ability to meet the frame deadline will depend on the duration of the superframe. The larger the superframe, the more likely the frame deadline will be exceeded. According to the c protocol, the superframe duration 97

6 Throughput (Mbps) The LakeHouse (Expected) The LakeHouse (Actual) Transporter2 (Actual) Transporter2 (Expected) 30.4% 15% Throughput (Mbps) Speed (Expected) Speed (Actual) Finding Neverland (Actual) Finding Neverland (Expected) Blue Planet (Actual) Blue Planet (Expected) 19.2% 10.4% 12% Number of Devices, N Number of Devices, N (a) Fig. 3. (b) Actual Throughput vs. Expected Throughput for H.264/SVC HD video traces Throughput (Mbps) Actual Expected Dropped Frames/Used Frames P Frames B Frames I Frames Time (secs) Device Number Fig. 4. (a) (b) (a) Actual vs. Expected Throughput for heterogeneous video traffic, (b) Percentage dropped frames per device may range between 1 μs and μs. The benefit of increasing the superframe size is that the MAC overhead is reduced as the beacon is sent less regularly. However, this comes at the cost of quality of service for applications with strict delay requirements, such as a video play-out deadline. For the Transporter2 video, the simulation is run for a superframe duration ranging from 5 ms to 65 ms. The CTA allocation is maintained constant as determined for the previous experiment. Figure 5a shows the variation in the actual throughput and expected throughput with respect to the superframe size for a range of network sizes. Actual and expected throughput are determined as previously. The effect of keeping the CTA allocation constant is that the additional capacity of the larger superframe is not exploited. This is indicated by the decreasing throughput above a certain superframe duration. However, the importance of adapting the superframe duration to the network load is demonstrated. As the number of devices in the network increases, the superframe duration at which the maximum throughput is achieved increases. In this example, it ranges from 5msto20ms. For the same experiment, Fig. 5b shows the increase in percentage of frames dropped as the superframe size increases. This is due to the increased superframe size causing the frame deadline (based on the frame rate of 24 frames per second) to be exceeded and these frames being dropped as a result. A compromise must be reached when setting the superframe duration in order to attain the optimum throughput for the network size/traffic load and to achieve an acceptable quality of service for the applications in that network. It can be seen from the results that with fewer devices in the network, the competition for resource is reduced and the difference between actual and expected throughput is less 98

7 Throughput (Mbps) Superframe Duration (ms) Expected Actual Number of Devices, N Dropped Frames/Used Frames Superframe Duration (ms) Fig. 5. (a) (b) (a) Actual and Expected Throughput, and (b) Percentage dropped frames per device with respect to superframe duration significant. It might then be satisfactory in a network with 10/15 devices to satisfy all requirements with the available PHY bandwidth. However, it should be considered that a single device may have more than one application running. In addition, even for a small number of devices, in the situation where all applications require high data rate simultaneously, the network capacity could be exceeded. The conclusions we draw from this analysis are that in order to maximise network throughput, the superframe duration must reflect application latency requirements and CTAs must be adapted to suit VBR applications and to meet the quality of service requirements of individual applications. While methods have been proposed in the literature (as presented in Sect. IV) for estimating the bandwidth required by VBR traffic, the ability of the network to support these varying requirements has not been considered. For example, if a number of devices are admitted to the network based on an initial estimation of required bandwidth and subsequently the required bandwidth is increased beyond the capacity of the network, how can the allocation be handled to best meet the QoS requirements of each application/device? We propose an approach in which all application requirements are considered simultaneously and an appropriate allocation made, which will be satisfactory for every application. This could be achieved by the device using the characteristics of the traffic (e.g. frame sizes and deadlines) combined with feedback regarding previous transmissions (e.g. no. of frames lost) to determine if a CTA update request should be sent immediately or if a delay of a number of superframes can be tolerated by the application. The PNC could then consider the frame information along with existing allocation bandwidth requirements to schedule CTAs to meet the QoS requirements of all active devices. VI. CONCLUSION In this paper, we have studied the suitability of the IEEE c mm-wave WPAN standard for compressed HD video distribution. A MATLAB simulation modelling the hybrid CSMA/CA-TDMA MAC was developed. The analysis shows that a significant loss in throughput and quality of service is experienced when a static channel time allocation is made in the network based on the mean bit rate of a variable traffic stream. It is also shown that the superframe duration must reflect application latency requirements. The conclusion is that a dynamic system is required, which will adapt superframe duration to the applications in progress and link CTAs to the varying traffic stream. Private knowledge of a device regarding the contents and traffic type in its buffer could be used to assist the PNC in optimising resource allocation. Simultaneously, knowledge of the network and network traffic could be exploited by the PNC to shape the MAC structure by periodically modifying the superframe duration and CAP length. Our future work will explore resource allocation methods to support heterogeneous high data-rate traffic in a mm-wave network with minimum computational overhead. REFERENCES [1] S. K. Yong and C.-C. Chong, An overview of multigigabit wireless through millimeter wave technology: Potentials and technical challenges, Eurasip Journal on Wireless Communications and Networking, vol. 2007, [2] S. Nandagopalan, MAC Channel Access in 60 GHz, 2009, IEEE /0572r0. [3] IEEE c, IEEE Standard for Information technology - Telecommunications and information exchange between systems - Local and metropolitan area networks - Specific requirements. Part 15.3: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs) Amendment 2: Millimeter-wave-based Alternative Physical Layer Extension, IEEE Std c-2009 (Amendment to IEEE Std ), pp. c1 187,

8 [4] WirelessHD Consortium, May [Online]. Available: http: // [5] C.-W. Pyo and H. Harada, Throughput analysis and improvement of hybrid multiple access in IEEE c mm-wave WPAN, IEEE Journal on Selected Areas in Communications, vol. 27, no. 8, pp , Oct [6] L. X. Cai, L. Cai, X. S. Shen, and J. W. Mark, Resource Management and QoS Provisioning for IPTV over mmwave-based WPANs with Directional Antenna, Mobile Networks and Applications, vol. 14, no. 2, pp , Apr [7] X. Zhu, A. Doufexi, and T. Kocak, On the performance of IEEE c millimeter-wave WPANs: PHY and MAC, in th Conference on Wireless Advanced (WiAD). IEEE, 2010, pp [8] L. X. Cai, K.-H. Liu, H. Rutagemwa, X. Shen, and J. W. Mark, Performance analysis of hybrid medium access protocol in IEEE WPAN, in 4th Annual IEEE Consumer Communications and Networking Conference, CCNC IEEE, Jan. 2007, pp [9] C. W. Pyo, Z. Lan, F. Kojima, H. Harada, and S. Kato, MAC enhancement for Gbps throughput achievement in millimeter-wave wireless PAN systems, in International Wireless Communications and Mobile Computing Conference, IWCMC 2008, August 6-8, Inst. of Elec. and Elec. Eng. Computer Society, 2008, pp [10] X. An, Z. Lan, R. V. Prasad, R. Hekmat, H. Harada, and I. Niemegeers, Performance analysis of the frame aggregation mechanisms in IEEE c, in 2009 IEEE 20th International Symposium on Personal, Indoor and Mobile Radio Communications. IEEE, Sep. 2009, pp [11] J. Luo, A. Kortke, and W. Keusgen, Throughput maximization of a Hybrid Dynamic Frame Aggregation and Block-ACK scheme for 60 GHz high speed communications, in 2010 IEEE 21st International Symposium on Personal Indoor and Mobile Radio Communications (PIMRC). IEEE, pp [12] S. Mehta and K. S. Kwak, Performance Analysis and Improvement of WPAN MAC for Home Networks, Sensors, vol. 10, no. 4, pp , Mar [13] M. Veyseh, A. Anpalagan, and B. Ma, Adaptive channel and superframe allocation (ACSA) for 60 GHz wireless networks, Mathematical and Computer Modelling, vol. 53, no. 3-4, pp , Feb [14] K. W. Chin and D. Lowe, A Simulation Study of TCP over the IEEE MAC, in The IEEE Conference on Local Computer Networks, th Anniversary. IEEE, 2005, pp [15] ISO/IEC :2010, Coding of Audiovisual Objects - Part 10: Advanced Video Coding, Tech. Rep., [16] P. Seeling, M. Reisslein, and B. Kulapala, Network performance evaluation using frame size and quality traces of single-layer and two-layer video: A tutorial, IEEE Communications Surveys and Tutorials, vol. 6, no. 3, pp , [17] G. V. der Auwera, P. T. David, and M. Reisslein, Traffic and Quality Characterization of Single-Layer Video Streams Encoded with the H.264/MPEG-4 Advanced Video Coding Standard and Scalable Video Coding Extension, IEEE Transactions on Broadcasting, vol. 54, no. 3, pp , Sep [18] M. Park, C. Cordeiro, E. Perahia, and L. L. Yang, QoS Considerations for 60 GHz Wireless Networks, in 2009 IEEE GLOBECOM Workshops, 2009, pp [19] K.-W. Chin and D. Lowe, A novel IEEE CTA sharing protocol for supporting VBR streams, in 14th International Conference on Computer Communications and Networks, ICCCN IEEE, 2005, pp [20] Y.-H. Tseng, E. H. K. Wu, and G.-H. Chen, Scene-Change Aware Dynamic Bandwidth Allocation for Real-Time VBR Video Transmission Over IEEE Wireless Home Networks, IEEE Transactions on Multimedia, vol. 9, no. 3, pp , Apr [21] W. K. Kuo and S. Y. Lien, Dynamic resource allocation for supporting real-time multimedia applications in IEEE WPANs, IET Communications, vol. 3, no. 1, pp. 1 9, [22] Y. Kikuchi and T. Nomura, IETF RFC 3016, RTP payload format for MPEG-4 audio/visual streams, [23] Fraunhofer Institut for Integrated Circuits - Audio Coding HE-AAC, February [Online]. Available: produkte/audiocodec/audiocodecs/heaac/index.jsp 100