Multi-rate Opportunistic Spectrum Access in Multi-hop Ad Hoc Networks

Transcription

1 22nd IEEE Personal Indoor Mobile Radio Communications Multi-rate Opportunistic Spectrum Access in Multi-hop Ad Hoc Networks Ari Raptino H Graduate school of Shizuoka University Hamamatsu, Japan ari@aurum.cs.inf.shizuoka.ac.jp Masaki Bandai Faculty of Sophia University Tokyo, Japan bandai@sophia.ac.jp Takashi Watanabe Graduate school of Shizuoka University Hamamatsu, Japan watanabe@inf.shizuoka.ac.jp Abstract Opportunistic spectrum access (OSA) is a suitable approach to increase spectrum efficiency for the current spectrum management. This paper considers the temporal aspect of based OSA scenario. Firstly, we investigate the secondary users (SU) multi-rate transmission and number of hops effects to the SU performance when exploiting primary users (PU) idle duration. Numerical analysis and simulation results show that, during limited PU idle duration, higher data rate transmission is essential even though it results in more hops but as the SU source traffic is getting busier, the balance between the data rate and number of hops is required. Secondly, according to the multirate multi-hop investigation, we propose a rate adaption scheme with PU activity consideration, called Auto Rate Increase (ARI). ARI is based on the Auto Rate Fallback (ARF) scheme with the difference that ARI aims to increase the SU data rate whenever it detects PU interruption failure. ARI also adopts ARF simplicity, hence, there is no need for any prior nor additional knowledge of the networks environment. Simulation results show, under tight OSA environment, ARI scheme can increase the SU performance up to 56% compared with the original ARF scheme. I. INTRODUCTION Opportunistic spectrum access (OSA) approach is a promising solution to overcome spectrum inefficiency problem. OSA is a hierarchical spectrum access model which introduces primary user (PU) and secondary user (SU) notation. OSA allows SU (cognitive user) to exploit PU s (legacy user) spectrum opportunity as long as it is not intrusive to the PU performance. Spectrum opportunity can be discovered in spatial and temporal domains. In this paper, we focus on two aspects. The first one is the temporal spectrum opportunity, which is the PU idle duration. The second one is multi-rate multi-hop ad hoc SU networks. These two aspects mark this paper unique contributions as the following, firstly, assuming that SUs form a multi-hop ad hoc networks and that they transmit opportunistically during PU idle duration, we investigate the SU performance given the following variation of multiple data rate transmission, number of hops, and PU idle duration. Focusing on the SU performance, the numerical and simulation results give the fundamental upper boundary of SU performance under the constraints of data rate, hops, and PU idle duration. Secondly, based on the investigation, we propose a rate adaption scheme which takes PU activity into account. The proposed scheme is versatile and simple, which means that it can work well under normal or OSA model and it also does not require any prior knowledge of the environment. Multi-rate and multi-hop impact on wireless ad hoc network have been extensively studied over years. Works in [1] [3] summarize that the data rate and the number of active nodes are the conflicting factors that affect network throughput. Our work complements this observation by investigating the multirate multi-hop performance under temporal OSA. For data rate adjustment, Auto Rate Fallback (ARF) [4] scheme remains the most widely applied in wireless local area network (WLAN) based devices, mainly because its simplicity. [5] and [6] extend the ARF scheme by applying receiver based rate decision and opportunistic back to back packet transmission, respectively. Further in this paper, we will discuss why ARF is not suitable for OSA model. The proposed scheme presented in this paper extends the ARF so that it can work well under OSA model. The temporal aspect of OSA is investigated in [7] and [8] for single-hop SU. Our work focuses on the multi-hop multirate environment and furthermore, a rate adaption scheme is proposed. The remainder of this paper is organized as follows. Network model and assumptions for the work in this paper are explained in section II. Investigation and analysis of multirate multi-hop SU will be discussed in section III. Section IV explains the proposed rate adaption scheme designed for OSA environment followed by the performance evaluation of the scheme. And section V marks the conclusion and future work of this paper. II. NETWORK MODEL AND ASSUMPTIONS We consider a system with PU and SU sharing single wireless channel. As legacy user, PU has higher priority to access the channel over SU. SU s objective is to utilize PU idle duration. Therefore from the viewpoint of an SU, PU activity is often generalized as one PU with alternating busy and idle duration traffic where the idle duration is exponentially distributed [7]. We investigate with the following PU traffic, busy duration is fixed to 14 msec and various mean idle duration. PU channel access is not affected by SU activity. SU, on the other hand, can only access the channel when /11/$ IEEE 567

2 PU is idle. SU channel access scheme will further described in section II-B. The system is assumed to be stationary and ergodic. Also, the channel is considered to be ideal. A. SU nodes topology To emphasize multiple data rate and various number of hops, SU nodes topology is modeled to an ad hoc chain topology with disjoint source and destination node. With this configuration, each data rate has different number of hops. We use the data rate specified by IEEE b radio, but we exclude 2 Mbps from simulation because 2 Mbps will result in the same number of hops as 5.5 Mbps. Fig. 1. illustrates the SU nodes topology for each data rate where source (node 1) and destination (node 7) is separated by distance of 900 m. PU busy and idle periods are assumed to be similar for every SU nodes so there is no need to schedule synchronization between each node. Each SU is assumed to be equipped with single half-duplex transceiver. B Based OSA scheme SU channel access scheme follows the Distributed Coordination Function (DCF) scheme with few modification for OSA model. Due to the lack of space, DCF operation detail can be seen in [9]. Basically, as SU has packets to send, it will employ listen-before-talk mechanism. Firstly, SU carries out spectrum sensing for a τ duration to find a vacant channel of PU, once SU finds and decides a specific vacant channel, it will start DCF mechanism for channel access. The inter-frame spaces, i.e distributed interframe space (DIFS) and short inter-frame space (SIFS), and backoff (BO) duration are used to sense other SU transmission and also PU transmission. After SU senses idle channel for τ +DIF S +BO duration, it will proceed to transmit a packet. Let T s and I r denote the total duration of successful packet transmission and the remaining idle duration, respectively. In temporal OSA, I r is not guaranteed to accommodate T s. When I r < T s then collision with PU will occur. [10] assumes SU knows the I r and uses that knowledge to decide whether to transmit or go back to BO state. Fig. 2. describes the mechanism of based OSA. Our work assumes the following. SUs do not know I r value. Because we focus on how well SUs can salvage the temporal spectrum opportunity of PU, SUs follow an aggressive mechanism by deciding to transmit all the time after some idle duration is detected. Initial spectrum sensing is also can be ignored (τ = 0) because the channel which SU worked on is already specified. DIFS, SIFS, and BO duration act as both PU and carrier sensing to determine whether the channel is idle or not. All sensing results are assumed to be perfect. III. MULTI-RATE MULTI-HOP INVESTIGATION In this section, we analyze and investigate the impact of SU multi-rate multi-hop on ad hoc networks. Firstly, we conduct a numerical analysis on SU multi-rate multi-hop performance under OSA temporal aspect. Secondly, we further validate the numerical analysis through simulation using QualNet simulator [11]. PHY MAC Fig. 1. Fig. 2. SU nodes chain topology Based OSA illustration TABLE I SIMULATION PARAMETERS Channel Radio type T x power T x Range 1 Mbps T x Range 2 Mbps T x Range 5.5 Mbps T x Range 11 Mbps Carrier Sense Range PHY header 2.4 GHz b Radio 15 dbm 450 m 339 m 328 m 270 m 550 m 24 Bytes IEEE DCF RTS 20 Bytes CTS 14 Bytes ACK 14 Bytes MAC header 34 Bytes DISF 50 µsec SIFS 10 µsec σ 20 µsec CW min 32 Network Routing Static Application Simulation duration A. Numerical analysis SU traffic SU packet size 300 seconds CBR 1500 Bytes In our model, multi-hop SU network may experience collision under two following factors: interruption from PU transmission and intra-flow contention between SUs. During low SU traffic, PU interruption dominates the collision probability while intra-flow contention is negligible. Let P pc denotes the collision probability because of PU interruption, given the PU mean idle duration as µ and successful transmission duration as T s, P pc can be written as 1 t P pc = 1 T s µ e µ dt = 1 e Ts µ (1) We also determine that PU interruption is dominant during DATA transmission. Following the scheme, a total of 4 568

3 DATA transmissions are allowed before the packet is dropped. Let i-th transmission denotes the initial and 3 retransmission of DATA, where 0 i 3. Let T si denotes the successful transmission duration at i-th transmission. We define T si = T c + BO i, with BO i = 2i.CW min 1, and T c = 2 T P HY +T RT S +T CT S +T ACK +T DAT A +DIF S +3SIF S. Where T P HY, T RT S, T CT S, T ACK, and T DAT A denote the time needed to transmit PHY header, RTS, CTS, ACK, and DATA packets, respectively. T DAT A is a function of SU source data rate R, while the others are transmitted at 1 Mbps. BO i is the expected backoff value at i-th transmission and CW min is the minimum contention window. We can substitute T si to (1) where we can derive P pci for each i transmission as a function of BO i, with that, the probability of successful SU transmission during µ can be written as P succ = (1 P pc0 ) + P pc0.(1 P pc1 )+ P pc0.p pc1 (1 P pc2 ) + P pc0.p pc1.p pc2 (1 P pc3 ) (2) hence, the end-to-end throughput S e2e for low SU source traffic rate S src can be approximated with S e2e = S src.(p succ ) h (3) where h denotes the number of hops. Analysis for busy S src is more difficult because we need to consider the collision caused by the intra-flow contention. First, we strictly define that if the total time needed to transmit successfully at i-th transmission is longer than inter-arrival time then intra-flow contention will occur [12]. Take S src = 800 Kbps for example with the following parameters: SU packets inter-arrival time of 15 msec, expected busy PU duration of 14 msec, and SU packet size of 1500 Bytes. In such case, one transmission failure makes the next transmission duration longer than the inter-arrival time and only the first successful transmission can avoid intra-flow contention. Therefore, we can rewrite and approximate (3) for 800 Kbps as S e2e (800Kbps) = S src.(1 P pc0 ) h. We will leave the generalization of the busy traffic analysis for our future work. B. Simulation validation We validate the analysis in the previous section through simulation. Unless stated otherwise, general simulation parameters are being listed in Table I. All simulation results are obtained within the 95 % confidence interval. In Fig. 3, we plot the end-to-end throughput (S e2e ) versus the PU mean idle duration (µ) for two different SU source traffic rate(s src ) 200 and 800 Kbps, where they represent low and busy traffic, respectively. The result shows some noticeable gap between analytical and simulation results, especially at the 800 Kbps traffic measurement. The main reason of this inaccuracy is that the numerical analysis does not properly address the intraflow contention problem. Nevertheless, both analytical and simulation results show similar trends and agree that high R is essential under limited µ duration. In low S src, high R is preferable even though high R results in more h, while during busy S src, balance between R and h is preferable as shown in (a) SU source traffic rate 200 Kbps (b) SU source traffic rate 800 Kbps Fig. 3. Analysis and simulation result comparison for various SU source traffic rate and µ the Fig. 3(b) where 5.5 Mbps gives better performance than the other two data rates. In summary, analytical and simulation results give the following conclusions: higher data rate is essential when SU exploits temporal spectrum opportunity. Higher data rate gives more guarantee for successful packet transmission during limited idle duration. Hence, it negates the packet drop caused by the PU interruption. As the SU source traffic rate is getting busier, intra-flow contention shows its effect. The contention is worsen as multi-hop nodes increase in numbers. In this case, it is preferable to balance the data rate and the number of hops. IV. PROPOSED RATE ADAPTION SCHEME In this section, we use the observation from the section III to propose a rate adaption scheme for SU to transmit opportunistically in the temporal OSA multi-hop ad hoc networks. We choose to extend the ARF scheme, mainly for its simplicity and flexibility. ARF requires no prior nor additional information of the network which makes ARF is widely implemented in many WLAN devices. Our proposed rate adaption scheme, called Auto Rate Increase (ARI), adopts the simplicity of ARF and versatile to be used for normal or OSA environment. Network model and assumptions explained in the section II are still hold for the rest of this section. A. ARF Problem description The basic of the ARF algorithm is to decrement the data rate after a specific number of f consecutive transmission failures and increment it after s successful transmissions. Rate adjustment is decided at the source side by detecting whether 569

4 (a) Vulnerable periods of normal and PU failure (b) ARI flowchart Fig. 4. ARI scheme it receives any response from the destination. In DCF, if source does not receive CTS/ACK for longer than a certain timeout then the transmission will be deemed as failure. The failure is caused by the high bit error rate (BER) in the signal received by the destination. By utilizing slower data rate, the BER can be decreased to the point that the destination is able to decode the transmission correctly. This is the main logic for ARF scheme to decrement the data rate whenever a transmission failure is detected. Section III points out the importance of utilizing high data rate in limited PU idle duration. If the original ARF scheme were to work in temporal OSA model, ARF scheme treats all kind of failures by decrementing the data rate, this mechanism leads to significant throughput degradation in the temporal OSA model. ARI scheme addresses this issue and makes sure that the rate is adjusted correctly according to the environment. B. ARI scheme In order to solve the aforementioned problem, ARI scheme needs to distinguish between normal and PU failure. PU failure is defined as the transmission failure which is caused by the PU interruption, upon encountering such failure, SU tries to increment the data rate in the next retransmission in order to provide faster transmission during limited PU idle duration. Next, normal and PU failure differentiation process is explained as follows. Considering the hardware constraint, SU can not performs simultaneous sensing and transmitting mechanism which makes it difficult to know whether the failure is caused by PU or normal failure. Let us refer to Fig. 4(a). From the source node point of view, the vulnerable periods of transmission failure are at WAIT_FOR_CTS and WAIT_FOR_ACK duration. WAIT_FOR_CTS and WAIT_FOR_ACK duration are the timers set by the source node after RTS and DATA transmission, respectively. When both of these timers expire and no CTS nor ACK is received then the transmission is marked as failure. Extensive simulation from section III proves that PU interruption is dominant during WAIT_FOR_ACK duration. By combining MAC and PHY layer knowledge, SU marks the PU failure if the following are true: WAIT_FOR_ACK timer expires (ACK timeout), PHY state is on sensing, and for more reliability, PHY sensing detects PU signal. If one of these three conditions is not satisfied then SU treats the failure as normal failure. With these prerequisites, SU can distinguish normal and PU failure to adjust the data rate accordingly. Summary of ARI scheme is as follows, let R = {R min, R 1, R 2,..., R max } denotes the set of data rate defined by the PHY layer. R, R min, R max denote the selected data rate, lowest data rate, and highest data rate, respectively. R, R min, R max R. For the initial transmission, SU sets R = R min and as SU detects s numbers of successful transmissions and k numbers of PU failures, SU increments R until R = R max and for f numbers of transmission failure R is decremented until R = R min. Fig. 4(b) describes the flowchart of ARI scheme. C. Performance evaluation In this section, we evaluate ARI scheme by comparing it with the original ARF through simulation. General simulation parameters are similar to the parameters in the Table I with one exception, because the rate will be dynamically adjusted over times, we use AODV dynamic routing instead of static routing. Parameter f and s follow the ARF scheme which are 10 and 2, respectively. Intuitively, the best k value is not larger than the number of maximum long retransmission limit (number of ACK timeout encountered before the packet is dropped) specified by DCF which is 4. Although not shown in this paper, simulation has been done to determine the best k value. The results show that, although slightly, k = 2 seems to give the best performance overall. Hence, for the rest of this section, we use k = 2 for ARI scheme. Set of data rate R = {1, 2, 5.5, 11} follows b specification where 1 and 11 Mbps correspond to R min and R max, respectively. Fig. 5(a) shows the result of ARF and ARI performance comparison in 7 SU nodes topology as depicted in 11 Mbps nodes topology from Fig. 1. In this topology, all data rates are guaranteed to have the network connected. The result shows that ARI scheme outperforms ARF scheme in both low (200 Kbps) and busy (800 Kbps) SU source traffic during limited PU mean idle duration (µ). ARI scheme performance increase is due to the more frequent selection of high data rate as opposed to ARF scheme where it always decrements the data rate for transmission failure. Fig. 5(b) shows the result for 4 nodes, please refer to 5.5 Mbps SU topology in Fig. 1 for the illustration. The 4 nodes topology restricts the use of 11 Mbps as data rate selection. This topology will further proves 570

5 TABLE II ARI PERFORMANCE INCREASE AND DECREASE (+/-) COMPARED TO ARF FOR VARIOUS SU SOURCE TRAFFIC, µ, AND TOPOLOGIES µ 200 Kbps, 800 Kbps, 200 Kbps, 800 Kbps, (msec) 7 nodes 7 nodes 4 nodes 4 nodes (a) 7 nodes topology % % % % % +34.6% % % % % % % % % -0.81% % % % -3.45% % % % -1.03% % % % -0.56% % % % -0.52% % % +7.05% -0.68% % % % -0.58% % Fig. 5. and µ (b) 4 nodes topology ARF and ARI performance comparison for various SU source traffic network. It also able to work in both normal and OSA model. Performance evaluation shows that ARI performs better than ARF scheme under temporal OSA model. ARI is able to select the highest available data rate to mitigate both PU and normal failure. Finally, for the future works. Firstly, we will provide the generalization of the busy traffic analysis. Secondly, we will add PU protection consideration to our model. whether ARI scheme is able to distinguish between normal and PU failure, failure to do so causes ARI scheme to choose the highest data rate available (11 Mbps) and leads to significant performance degradation. The result shows the otherwise, ARI scheme can adapts well with the change of topology and still outperforms ARF under OSA model. Both results from Fig. 5 show that ARI scheme performs better than the original ARF scheme under OSA model by choosing the highest possible data rate available depending on the various situation, e.g., busy SU source traffic rate or topology. Table II summarizes ARI performance increases and decreases compared to the ARF, represented by the (+) and (-) signs, respectively. We can clearly see that, in general, ARI outperforms ARF scheme under relatively short PU mean idle duration. Slight performance decreases of ARI can be spotted in low SU source traffic and relatively long PU mean idle duration. The decreases are around 0-3% and insignificant compared to the increases in performance which are up to 56%. V. CONCLUSIONS In this paper, we investigate SU multi-rate multi-hop impact under temporal OSA model and propose rate adaption scheme which takes PU activity into account. From the multirate multi-hop investigation, we found that high data rate is essential and balancing the data rate and number of hops is preferable as the SU traffic getting busier. The investigation leads us to propose a rate adaption scheme under temporal OSA model called ARI. ARI is simple and versatile where it does not require any prior nor additional information of the REFERENCES [1] R. Draves, J. Padhye, and B. Zill, Routing in multi-radio, multi-hop wireless mesh networks, in Proceedings ACM MobiCom 04, 2004, pp [2] X. Wu, G. Ding, and W. Zhu, Load-based route discovery through searching range adaption for manet throughput improvement, IEEE Transaction on Vehicular Technology, vol. 58, no. 4, pp , May [3] F. Y. Li, A. Hafslund, M. Hauge, P. Engelstad, O. Kure, and P. Spilling, Does higher datarate perform better in IEEE based multihop ad hoc networks? Journal of Communications and Network, vol. 9, no. 3, pp , Sept [4] A. Kamerman and L. Monteban, Wavelan-II: A high-performance wireless LAN for the unlicensed band, Bell Labs Technical Journal, vol. 2, no. 3, pp , [5] G. Holland, N. Vaidya, and P. Bahl, A rate-adaptive mac protocol for multi-hop wireless networks, in Proceedings ACM MobiCom 01, 2001, pp [6] B. Sadeghi, V. Kanodia, A. Sabharwal, and E. Knightly, Opportunistic media access for multirate ad hoc networks, in Proceedings ACM MobiCom 02, 2002, pp [7] S. Huang, X. Liu, and Z. Ding, Opportunistic spectrum access in cognitive radio networks, in INFOCOM The 27th Conference on Computer Communications. IEEE, april 2008, pp [8] K. W. Sung, S.-L. Kim, and J. Zander, Temporal spectrum sharing based on primary user activity prediction, Wireless Communications, IEEE Transactions on, vol. 9, no. 12, pp , december [9] IEEE standard - part 11: Wireless lan medium access control (mac) and physical layer (phy) specifications, IEEE Std , pp. C1 1184, [10] A. T. Hoang, D. Wong, and Y.-C. Liang, Design and analysis for an based cognitive radio network, in Proceedings WCNC IEEE, april 2009, pp [11] Qualnet simulator version 4.5. [Online]. Available: [12] J.-Y. Yoo and J. Kim, Maximum end-to-end throughput of chaintopology wireless multi-hop networks, in Proceedings WCNC IEEE, march 2007, pp [13] Q. Zhao and B. M. Sadler, A survey of dynamic spectrum access, in IEEE Signal Processing Magazine, May 2007, pp