Analytical and Simulation Studies for Call Admission and Resource Allocation Schemes proposed for WiMAX system

Sarabjot Singh, Sanjay K. Bose ECE Department, Indian Institute of Technology Guwahati, India sarabjot, skbose@iitg.ernet.in Maode Ma School of EEE, Nanyang Technological University Singapore emdma@ntu.edu.sg

Abstract- Quality communication over the wireless channel is
now promised because of the advent of the broadband wireless. High speed data access is expected to be provided to subscribers through broadband wireless access. We propose a joint Call Admission Control (CAC) and Bandwidth Allocation (BA) for an IEEE 802.16 based WiMAX system. The presented schemes aim to provide QoS support along with a fair resource allocation algorithm for nrtPS traffic. Two strategies for CAC namely Conservative and Non-Conservative have been proposed. Conservative CAC guarantees the QoS requirements for all classes of traffic but is more restrictive and less efficient than the Non-Conservative CAC. The performance for proposed schemes has been studied through both analytical models and simulations.

I.

INTRODUCTION

The IEEE’s 802.16 WiMAX system [1,2] aims at providing high-speed internet access and multimedia services through wireless medium. Fast deployment, the ability to reach crowded or rural areas without the use of costly wired infrastructure and the ability to support Quality of Service (QoS) required for different real-time application in wireless networks are its main advantages. QoS provides different priorities to different users or different data flows and guarantees a certain level of performance to designated data flows in accordance with requests from the application programs or the policy of the internet service providers (ISP). Four service classes are considered in this paper namely, Unsolicited Grant Service (UGS), Real-time Polling Service (rtPS), Non-real time Polling Service (nrtPS) and Best Effort (BE), with specific QoS requirements for each service class. (The ertPS service class has not been considered in this paper.) The UGS service supports constant bit rate applications through a fixed allocation. The rtPS service is meant for realtime services that generate variable rate data such as VBRvideo. The nrtPS is for non-real-time and delay-tolerant services such as FTP. The BE service supports data that do not require QoS guarantees. Packet scheduling, Resource allocation and Call Admission Control work in tandem to ensure QoS to such different traffic classes in a WiMAX

system. But the standard has left the design of these schemes to researchers. Several earlier works have considered the problem of Call Admission Control (CAC) and Bandwidth Allocation (BA) separately but works looking at the joint BA and CAC for a WiMAX system are scarce. The bandwidth allocation and scheduling approaches proposed for WiMAX can be classified as homogenous, hybrid or opportunistic algorithms [3]. The first two types use various legacy allocation approaches. For example, the scheduling algorithm of [5] assigns a fixed bandwidth for UGS, uses Earliest Deadline First (EDF) technique for rtPS, Weighted Fair Queuing (WFQ) for nrtPS and equal distribution for BE. A token bucket based conservative approach for CAC is also proposed. These are improved in [6] by a hybrid scheme that uses EDF for SSs of the rtPS class and WFQ for SSs of nrtPS and BE classes. In [7], an adaptive queue-aware algorithm is proposed for uplink bandwidth allocation and rate control mechanisms in a SS for polling services in a GPC system. The transmission rate of the connections under polling service is limited by rate control so that the overall QoS performance can be controlled. However, [7] treats real-time and non real-time services identically and does not adequately exploit QoS factors like maximum latency in its scheduling. In [4] a queue based scheduling algorithm for real-time and non real-time traffic at the Medium Access Control (MAC) layer is proposed for resource sharing between real-time and non real-time traffic depending on their queue size and latency requirements. Performance of the EDF scheduling algorithm for BWA networks is evaluated in [8]. Tsai et.al [9] propose an uplink scheduling algorithm and a token bucket based Call Admission Control (CAC) algorithm. The scheme in [10] accepts each service based on its Minimum Reserved Traffic Rate and thus cannot satisfy the bandwidth request of real-time VBR services. In [11]-[12] degradation based CAC have been proposed which aims at better handover service by minimizing the handoff connection dropping probability. The organization of the rest of the paper is as follows. Section II gives a basic description of IEEE 802.16 systems. Section III provides the description of the joint CAC and fair

978-1-4244-7006-8/10/$26.00 ©2010 IEEE

522

(Extending the proposed schemes for variable PDU lengths can be easily done. This has been shown in Figure 1. II. An active UGS source s is assumed to generate Us packets per frame. streamed video or audio. we propose using conservative CAC for highest priority rtPS sources (with zero excess delay requirements) and non-conservative CAC for the other (lower priority) rtPS sources which can tolerate higher excess delays.) We consider two strategies for CAC: Conservative and Non-Conservative. the rtPS application must be one which can tolerate some packet loss. access delays cannot be guaranteed. Bandwidth Allocation Once admitted into the system. Section V concludes the paper. Here λi. C. L rtPS and M nrtPS sources already present in the system and let C be the maximum number of packets that can be carried in a frame. For call admission in the above scenario. Each rtPS source specifies both the mean and maximum number of packets it generates per frame. The Conservative CAC ensures that all UGS and rtPS packets arriving during frame j will be transmitted in frame j+1. The packet transmission requests of UGS and rtPS are satisfied on a per frame basis while that of nrtPS are satisfied on a long term basis. Hence packets accumulating over a frame at a SS can only be transmitted from next frame onwards. Each nrtPS source specifies the mean number of packets it generates per frame.16 SYSTEM A WiMAX system contains at least one Base Station (BS) and many Subscriber Stations (SSs) each having a unique MAC address. However. The bandwidth allocation is done in the order of Pi. IEEE 802. the condition on accommodating the maximum rtPS traffic in one frame and admit a rtPS flow even if only (1) is met. We assume all PDUs (packets) to be of the same size for modeling convenience. what is left after UGS and rtPS allocation) among the M nrtPS connections. A particular queue Qi is subdivided into i+1 parts P0 to Pi. Time-Division-Multiple-Access (TDMA) is used by SSs on the uplink to transmit back to the BS in specific allocated time slots. Conservative CAC This aims to guarantee the QoS for UGS. Hence we waive (2). In the UL contention period. Non-Conservative CAC Here. We consider the centrally controlled IEEE 802. NRm for nrtPS source m. To fairly allocate the remaining resource packav (i. e. A. where the packets entering Qi are from rtPS sources which can tolerate upto i frames of excess delay. In the TDD/TDMA technique the transmissions are divided into frames (in time) with each frame further divided into uplink and downlink sub frames. i= 1… M.16 protocol (though our proposed scheme can be easily extended for Mesh mode operation). A new nrtPS source must still meet (3) to be admitted. a rtPS packet need not be carried in the immediate next frame. III.16 handles QoS of real time applications like VoIP. collisions might occur as when two or more SSs place their request PDUs in the same minislot. We label the sources from 0 to N-1. Each sub frame consists of number of time slots which are dynamically assigned to SSs in the uplink sub frame by the BS based on the requests received from them.e. Packets belonging to Pl of each queue can be transmitted over the next l frames and will be discarded thereafter. packets are the respective demands in the current frame of each We consider a single BS serving multiple connections from SSs through a TDMA/TDD access mode. with assignments of unsolicited bandwidth grants and polling.g. rtPS and nrtPS traffic. After each subsequent frame. There are two ways to contend for a transmission opportunity either to transmit in periodic intervals or to contend with the other SSs transmitting request for grants. i. Lower the i higher the priority. In case of same index i. index of corresponding queue considered. Video on demand etc. Because of this contention based access.resource allocation algorithm developed by us. The terms bandwidth and capacity are used interchangeably for the number of PDUs that can be transmitted in one uplink subframe. PROPOSED RESOURCE ALLOCATION AND CALL ADMISSION SCHEMES rtPS Source: A new rtPS source (Rnew. Rl & Rlmax for rtPS source l. Based on the above formulation. IEEE 802. A SS may establish multiple connections with BS where each connection is identified by a 16-bit Connection Identification Number (CID). the following algorithm is used. the packets not transmitted in each queue will be transferred from Pl to Pl-1 for l=2 to K. i. the packets arriving from rtPS sources which can tolerate up to K frames of excess delay will enter the queue QK and will be kept in PK from next frame. A discrete time Markov chain based analysis is also presented and the corresponding analytical and simulation results are compared in Section IV. Rnewmax) is admitted if ∑ s =1 S S Us + Us + ∑ ∑ NR Rl + l =1 L m=1 L M m + Rnew ≤ C (1) and ∑ s =1 S s =1 ∑R l =1 L l max + Rnew max ≤ C M (2) nrtPS Source: A new nrtPS source (NRnew) is admitted if ∑ Us + ∑ l =1 Rl + ∑ NR m=1 m + NRnew ≤ C (3) B.e.e. The conditions for admitting a new rtPS or nrtPS source are: 523 . Let there be S UGS sources. Either TDD or FDD method are used for transmissions. We consider a generic system which has N different types of rtPS sources differentiated on the basis of the maximum excess delay (the mean delay excluding the inherent one frame delay) that they can tolerate. This system will have J separate queues Q0 to QJ-1 (J≤N) where the new arrivals from the corresponding N sources would enter. the connections (UGS/rtPS/nrtPS) are allocated resources in the priority UGS>rtPS>nrtPS.

i i ⎟ (9) +1 ⎜ i ⎜ ⎟ n =1 ⎝ ⎠⎠ ⎝ ∑ Figure 1: Priority Structure The above algorithm allocates resources fairly to the nrtPS connections by distributing the available resource in proportion to their mean requirements as different nrtPS connections have different arrival rates of packets with different means. if any. N i ) be the system state at the end of frame i where N inr is the number of where P is the total number of rtPS and nrtPS packets that can be carried in a frame (i. 2. We first study the general case where rtPS packets can be buffered for one or more frames. max ⎜ 0. i =1…M.. 7.i .i .i are those MPDUs in the rtPS queue which will be discarded if not transmitted within the next l+1 frames. (Due to lack of space. Discrete-time Markov chain is used to analyze the system and to calculate the average delay of MPDUs and mean losses.m are computed as follows. 4. P − ⎜ ⎜ ⎝ ⎝ ∑ ⎞ ⎞ r ⎟ ⎟ Nn . 2 considers the general case where a rtPS packet can be buffered for a maximum of K r r nr frames before it is discarded. the state descriptor will be ( N1. are their mean requirements (which are to be met on a long term basis).i +1 = max ⎜ N r l +1.nm ∑∑ π n = 0 m =0 ∞ ∞ n.m =1 (10) We compute LossR as the average number of packets discarded per frame for rtPS packets and the mean delay DN for nrtPS packets as follows. Algorithm for Bandwidth Allocation in a frame 1.i . j Pij . K > 1 ⎜ ⎜ ⎟ n = 1 ⎝ ⎠ ⎠ ⎝ λi = λi − λav − i . i . j (12) 524 . the state at the end of the i+1 frame can be found as l ⎛ ⎛ ⎞ ⎞ r ⎟ ⎟ Nlr. The K=1 Non-Conservative CAC is studied further through subsequent analysis and simulations.. Let ( N K . 5. LossR = i = P +1 j = 0 ∑ ∑ (i − P)π ∑∑ jπ i =0 j =0 ∞ ∞ i. D. the individual rtPS and nrtPS queues merge into a global rtPS queue and a global nrtPS queue respectively at the BS. Steps 47 iteratively allocate resources unused by connections which demand less to connections which demand more. capacity excluding the UGS packets). ∑λ i =1 prop −i λav − i = ( packav × λ prop −i ) / λtot for i = 1… M λinit − i = λi for i = 1… M Check if packav > 0 and λtot > 0. if i=M GOTO 2 END λalloc −i = λinit − i − λi for i = 1… M ∑ ⎛ ⎛ = max ⎜ rai − max ⎜ 0.P − ⎟⎟ Ninr N nra min N ra = + − − i i +1 n. 3.i . 9. π n.m = ∑∑ i =0 j =0 ∞ ∞ ∞ ∞ π i . packav = packav + (λav −i − λi ) and λi = 0 else 8. 6.. j (11) NN = i. N i ) as per our earlier notation.nrtPS connection and NRi. Analytical Model We study the analytical model of the system with only single priority rtPS sources.) The equilibrium state probabilities πn.j)→(k.. Given the arrival distribution in a frame for each rtPS and nrtPS queue. N1. P − Nn . It can then be used to calculate the transition probabilities of the Markov Chain assuming that service is provided in a FCFS manner for K-frame buffering. if λ prop −i > 0 and λi > λav −i then set Following the conventions defined earlier.i − max ⎜ 0. Fig. K − 1. The transition probability Pij. these details are not given here. A PMP network where the generated traffic stays within the network is considered. .. in the system..0 ⎟ ⎟ n =1 ⎠ ⎠ l l=K (8) K ⎛ ⎛ ⎞⎞ nr r ⎜ N nr .kl for the (i. in a proportionally fair manner until either there is no available resource to allocate or there are no unmet demands left.l) transition may be found by considering each transition event.. For allocation by the BS.e. the arrival distribution to the global rtPS and nrtPS queues can be found. otherwise GOTO 8 SET packav = 0 and λtot = 0 For each i = 1… M if λ prop −i > 0 and Figure 2: Transitions with K frame buffering for rtPS packets λi ≤ λav −i then set λ prop −i = 0 . SET λ prop −i = NRi for i = 1… M SET λtot = M MPDUs in the nrtPS queue and Nlr. Note that for this K=1 r nr case..0 ⎟ l = 1.

This occurs because the one frame buffering case allows more rtPS traffic to be carried by the system – part of which may otherwise be lost when K=0. Multiple priority rtPS sources For our performance studies. Each SS is associated with a single type of connection (UGS/rtPS/nrtPS) on the UL. The total number of requests from all UGS connections is fixed at 15 packets per frame so that a maximum of 45 rtPS and nrtPS packets can be carried in a frame. This is shown in Fig. the infinite Markov Chain of Section III is truncated to a finite one which is large enough so that increasing it further does not change the numerical results significantly. This makes this CAC a practical choice in a real system where the rtPS channels are being used for multimedia applications.e. For simulations. With just one frame buffering. The number of nrtPS sources is fixed at 4 while the number of rtPS sources is varied as per what is allowed by non conservative CAC. 4 which plots the mean excess rtPS delays. this will be at the most one frame period more and may be tackled easily by adjusting the playback delays at the respective destinations. 3 of type 1 and 4 of type 2 rtPS sources. a discrete-event simulator was run until statistically significant results have been obtained. 1 (type 1) or 2 (type 2) frames of excess delay.4] to ~U[3. the NonConservative CAC has a low enough packet loss to be acceptable for typical multimedia applications but will allow the system to accommodate more rtPS sources than the Conservative CAC.e. there is evident differentiation between the excess delays of type 1 and type 2 rtPS sources in accordance with the priority structure assigned in bandwidth allocation.13] and that of type 2 rtPS sources from ~U[0. Single priority rtPS sources We simulate a WiMAX PMP scenario with single BS and variable number of SSs.10] to ~U[3. a nrtPS source generates packets with distribution ~U[0. the rtPS packets that cannot be carried in the next frame are lost) and (b) one frame buffering (i. It may also be noted that the analytically obtained results are very close to the simulation results with some deviations only at very high traffic loads. The mean arrival rate of type 1 and type 2 rtPS sources are varied according to what is allowed by the CAC. B. ANALYTICAL AND SIMULATION RESULTS The proposed CACs and resource allocation algorithm have been studied through analysis and simulations. The one frame buffering does increase the mean rtPS delay. rtPS packets can be buffered and tried for one more frame). K=1. Figure 3 shows the average nrtPS delays for K=0 and 1 with varying normalized maximum rtPS traffic. Here in each frame. There are 2 type 0 rtPS sources. IV. ones which can tolerate either 0 (type 0). As expected. 525 . For analytical computations. We consider two cases (a) no frame buffering (i. We first present the results for a system containing only single priority rtPS sources and later analyze the effectiveness of the proposed priority structure for rtPS sources with differential QoS requirements. increasing the rtPS traffic increases the nrtPS delay since system load increases. 4] whereas a rtPS source generates packets with ~U[0. Each type 0 rtPS source has a mean arrival rate of 15 packets per frame with distribution ~U [0.e. The above generic analytical model can be modified for the case where no buffering is allowed for rtPS sources. The state transition is similar to equation (9) with K being 0. Figure 5 shows the mean excess delay of rtPS sources of type 1 and type 2 with varying traffic intensities where the ordered pair on the plots shows the respective mean arrival rates of respective sources at particular traffic intensity. we consider a system with three types of rtPS sources. Assume that the maximum number of packets of rtPS and nrtPS sources (excluding the UGS) that can be carried in a frame is fixed at 75. from K=0 to K=1).7]. Our studies focus on the rtPS and nrtPS flows as UGS traffic gets a fixed number of packets in the UL of each frame and the BE traffic gets essentially what remains unused in UL.6. In Figure 5 the varying traffic intensities (normalized traffic load) are obtained by varying the distribution of type 1 rtPS sources from ~U[0.1 to 1. The state descriptor of Markov chain at the end of the frame i is Ninr.DN = NN λnr (13) UGS traffic) ranges from 1. 10]. Figure 3: Average nrtPS delays with varying maximum rtPS traffic A. Our results also show that as the ratio of maximum rtPS traffic to maximum available capacity (after taking care of the fixed It is clear that the packet loss is drastically reduced by merely incorporating one-frame buffering in Non-Conservative CAC. The nrtPS and rtPS packets are assumed to be of equal length of 100 bytes at UL data rate of 10 Mbps. K=0. As seen in this figure.e. Note that the maximum rtPS traffic may be higher than the capacity in this case as the CAC only takes the mean traffic into account. i. all being lesser than 1 frame. The total number of packets that can be carried in a UL subframe is set at 60 with the 10 ms frame divided equally between up-link and down-link. 30]. the number of packets in the global nrtPS queue. 4. The nrtPS delays are higher when the frame buffering of rtPS is increased (i. However. the rtPS packet loss ranges from approximately 10-2 to nearly 1 for K=0 but is merely 10-14 to 10-6 for K=1 as shown in Fig.

Sun. Packet level performances like average delays and loss rate are also shown for different mixes of traffic combinations. L. REFERENCES C. pp. “A Performance Study of Uplink Scheduling Algorithms in Point to Multipoint WiMAX Networks.Oct. CONCLUSION The proposed Admission Control (CAC) and Bandwidth Allocation (BA) mechanism have been analysed using a discrete Markov chain approach and simulations. K. Ji." Proc. Mag.Tsai. C. and D. 2007. Yusheng Ji and Nararat Ruangchaijatupon. vol. School of Computing. Wang. ”A study on the performance of scheduling schemes for broadband wireless access networks.[5] [6] [7] [8] Figure 4: Mean rtPS excess delays and packet loss percentage with varying maximum rtPS traffic [9] V. Liping Wang. We have also proposed a resource allocation scheme for real time sources with differential QoS requirements. Dhrona.243–249. pp.” Proceedings of Fourth European Conference on Universal Multiservice Networks. In Proceedings of the IEEE 63rd Vehicular Technology Conference. 10-13 Dec. pp. 2007 Figure 5 : Mean Excess Delays with varying Traffic Intensity in the System 526 . Niyato and E. "Queue based scheduling for IEEE 802.S.” Proceedings of IEEE 64th Vehicular Technology Conference. N. Sept. pp. The proposed prioirty structure is shown to be effective in handling such rtPS sources. Journal of Communications. thesis. no. Hossain. 2006. June 2006. on Mobile Computing. 1-857. "Queue-aware uplink bandwidth allocation and rate control for polling service in IEEE 802. vol. Vinay. 2006 D. ON. 1–5. “Admission Control for Non-pre-provisioned Service Flow in Wireless Metropolitan Area Networks. T. 2006. 98–107 [2] IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems.1-5.16e Networks.30-37.16 broadband wireless networks... no2. The analysis results are very close to the simulation results. [1] [10] [11] [12] Communications and Signal Processing. of IFIP Intl. Y. pp. Issue 1." Information. Issue 6.Jiang and C. J. D. better network resource utilization can be achieved. Feb. 16.. 12211225. 2004 [3] P. Ruangchaijatupon... “CAC and Packet Scheduling Using Token Bucket for IEEE 802. “An Efficient Admission Control Scheme for Adaptive Multimedia Services in IEEE 802. Jun." IEEE Std 802.” in Proc. Eklund et al.16 broadband wireless access systems. The Non-Conservative Connection Admission Control proves to be better in terms of overall network resource utilisation with only minor tradeoff on the delay and loss performance.." IEEE Trans.1.. Zhu. Maode Ma. The resource allocation scheme proposed for non real time traffic allocates resources with fairness in proportion to their respective QoS requirements. Sreenivasulu.. J. Queens Univ. “IEEE Standard 802. International Symposium on Communications and Information Technology (ISCIT’06). Communication Systems. Wongthavarawat and A. vol. “Packet Scheduling for QoS support in IEEE 802. pp. May 2006.” M.”Intl. vol. 2002.” IEEE Commun.162001). pp. S. "Performance evaluation of end-to-end delay by hybrid scheduling algorithm for QoS in IEEE 802. ECUMN ’07. Yau. K. 2007 6th International Conference on . The results showed that if the delay and loss tolerances for real-time traffic streams are relaxed. 8196 K. Wireless and Optical Communications Networks.16 Broadband Wireless Access Systems. 5. Bose. Fuqiang Liu. pp.16 Networks”.16: A Technical Overview of the WirelessMAN™ Air Interface for Broadband Wireless Access. Jayaram. VTC-2006 Fall.16 wireless broadband. Ganz. Feb 2003. pp.K. 668-679 N. Das.R. Quality of Service Scheduling for 802. Besides veryfying our analysis they also validate the simulation model. Kingston.16-2004 (Revision of IEEE Std 802.16 network. 2007 [4] Raghu. and H. Conf. and Y.Wang. Yin Ge and Geng-Sheng Kuo.