THE IEEE e system, fixed and mobile broadband

Transcription

1 Piggybacking Scheme of MAP IE for Minimizing MAC Overhead in the IEEE 80.16e OFDMA Systems JuYeop Kim, Student Member, IEEE, and Dong-Ho Cho, Senior Member, IEEE Abstract This paper analyzes Media Access Control(MAC) overhead of the IEEE 80.16e systems and shows that it causes to degrade system performance critically. MAP, a control message about resource allocation, is broadcasted with high robustness and uses a great amount of radio resource for it. This paper also proposes an advanced scheme which transmits MAP IE, a component of MAP, piggybacked on data packets, and uses fast feedback to conserve the transmission reliability of the MAP IE. Then, MAP IEs can be transmitted with high data rate, and the amount of radio resource for transmitting MAP IEs becomes extremely small. Numerical analysis and simulation results show that the proposed scheme can significantly improve the MAC overhead. Subchannel logical number 1 S-1 S Preample FCH DL-MAP Fig UL-MAP Burst Burst 1 Downlink subframe OFDM symbol number Burst K-1 K Burst 4 TTG K+1 Fast Feedback Channel Ranging Burst 1 Burst Burst Uplink subframe Burst 4 Burst 5 Burst 6 Time An example of frame in IEEE 80.16e system K+N I. INTRODUCTION THE IEEE 80.16e system, fixed and mobile broadband wireless access system, is on the basis of the IEEE REVd and 80.16e standards, published in 004 and 006, respectively, and aims to cover metropolitan area. It employs various schemes to dynamically handle radio resource in time and frequency domain. Orthogonal Frequency Division Multiple Access(OFDMA) scheme assigns frequency resource adaptively based on frequency selectivity, and Adaptive Modulation and Coding(AMC) changes Modulation and Coding Schemes(MCS) level adaptively based on instant channel quality. These schemes lead to increase the efficiency of radio resource utilization and consequently to improve the system capacity. However, the IEEE 80.16e system suffers a critical problem of a huge MAC overhead, which is the percentage of radio resource used for managing MAC protocols. Those dynamic schemes, such as OFDMA and AMC, require additional control signalling and cause to increase MAC overhead. In addition, Base Station(BS) broadcasts MAP messages, defining the access to downlink and uplink information, robustly so that all of Mobile Stations(MSs) can decode them, and assigns a great amount of radio resource for it. This causes to reduce the radio resource for data transmission and leads to limit system throughput. [1] reveals that the MAC overhead of the IEEE 80.16e system can be practically up to 60%. There are conventional researches related with MAC overhead of the IEEE 80.16e system. [] analyzes the MAC efficiency and refers the importance of MAC overhead reduction, but leaves it as a future work. Also, several schemes are proposed to reduce the MAC overhead. [] proposes a This work was supported in part by the university IT research center program of the government of Korea. The authors are with the Department of Electrical Engineering and Computer Science, Korea Advanced Institute of Science and Technology (KAIST), Yuseong-gu, Daejeon , Republic of Korea (Tel: , Fax: , jykim@comis.kaist.ac.kr, dhcho@ee.kaist.ac.kr). scheduling algorithm which reduces the size of the MAP message for relieving the MAC overhead problem. The IEEE 80.16e standard defines a control message format for Hybrid Automatic Retransmission Request(HARQ) in which the MCS level applied to the message varies based on the channel quality of a receiving MS, but this message is restricted to be applied to HARQ only [4]. [5] proposes a new format of control message for periodic resource assignment to reduce duplicate signalling, and [6] defines a compressed message format to reduce the size of the control message. However, these schemes can only be applied to specific cases and cannot bring significant improvement in view of MAC overhead. In this paper, we first estimate the MAC overhead due to MAP messages by numerical method and then propose an advanced scheme for reducing the portion of the radio resource used for MAP message transmission. The paper consists of followings; Section II describes the IEEE 80.16e system, Section III shows the analysis of MAC overhead in the IEEE 80.16e system, Section IV describes the proposed scheme for transmitting scheduling information, Section V presents the numerical and simulation results for the proposed scheme. Finally, Section VI shows conclusions. II. THE IEEE 80.16e SYSTEM The IEEE 80.16e system has an OFDMA frame structure, as shown in Fig. 1. The frame is composed of K downlink OFDM symbols, N uplink OFDM symbols and S subchannels. Each subchannel is composed of n s subcarriers. Using a terminology resource unit, composed of a subchannel and n t symbols, there are R d and R u resource units for downlink and uplink, respectively. After preamble, Frame Control Header(FCH) presents the length and the MCS level of MAP messages. The next following MAP messages are composed of DL MAP and UL MAP messages which describe the allocation of downlink and uplink packet transmission and help Mobile Stations(MSs) receive their own data correctly /07/$ IEEE 84

2 do Type PHY Synchronization Field DCD Count Base Station ID Type Uplink Channel ID UCD Count Allocation Start Time uo message for j th downlink burst in frame i, can be evaluated as following, b die (i, j) =(b di + n dc (i, j)b C ) (1) C di Fig.. DL_MAP_IE DL_MAP_IE UL_MAP_IE UL_MAP_IE Structure of DL MAP and UL MAP messages Then after MAP messages, downlink and uplink bursts, which are the concatenations of data packets and control messages, follow. Usually, a downlink burst is composed of downlink packets which is transmitted with the same MCS level, and a uplink burst is composed of uplink packets transmitted from an MS. Fig. shows the structures of DL MAP and UL MAP message. DL MAP message is composed of basic information fields, whose size is b do, and several DL MAP IEs. Each DL MAP IE has information fields, whose size is b di,forresource allocation of the bursts and several fields. The size of a field is b C. MSs can notice that the corresponding burst includes the data packets of s listed in the fields. ADLMAP IE has multiple fields since the downlink burst is composed of data packets which are transmitted with the same MCS level and may belong to different s. UL MAP message has similar structure to DL MAP message. It is composed of basic information fields, whose size is b uo, and several UL MAP IEs, and each UL MAP IE has information fields, whose size is b ui, for resource allocation of the uplink burst and a field. Unlike the downlink burst, the uplink burst is composed of uplink data packets belonging toa. Note that the large part of MAC overhead comes from MAP message transmission, because they take a great portion of radio resource. Since MAP messages should be delivered to all MSs even in bad channel condition, BS transmits them with lowest data rate to guarantee robustness. Thus, BS assigns large amount of radio resource for it compared to the other MAC management messages transmitted in the bursts. In addition, BS sends the MAP messages for every frame, which is fairly frequent compared to the other MAC management messages. Thus, our analysis focuses on the MAC overhead due to MAP messages. III. MAC OVERHEAD DUE TO MAP MESSAGES A. DL MAP Let j {1,,...N } be the index of MCS level where N is the number of MCS levels. Since downlink burst is generally a concatenation of packets whose MCS levels are the same, the number of DL MAP IEs in a DL MAP is the number of kind of MCS levels used for transmitting the scheduled packets. b die (i, j), the size of DL MAP IE ui where n dc (i, j) is the number of connections whose packets areinthej th downlink burst in frame i, and is equal to the number of connections which is scheduled in frame i and whose MCS level is j. TheDLMAP IE message for j th downlink burst is not transmitted when n dc (i, j) is zero. Then, b dmap (i), the size of DL MAP message in frame i, can be evaluated as following, b dmap (i) =b do + x(i, j)(b di + n dc (i, j)b C ) = b do + b di x(i, j)+n ds (i)b C () where x(i, j) is a binary variable which is set as following. { 1, if ndc (i, j) > 0 x(i, j) = () 0, otherwise In addition, n ds (i) is the number of downlink connections scheduled in frame i and is equal to the number of fields in the DL MAP message. It can be evaluated as following. B. UL MAP n ds (i) = n dc (i, j) (4) b umap (i), the size of UL MAP message in frame i, can be calculated as the sum of the sizes of UL MAP IEs and b uo.it is expressed as following, b umap (i) =b uo +n us (i)b ui (5) where n ura (i) and n us (i) are the number of uplink random access and scheduled connections, respectively, served in frame i. Like the downlink case, the size of UL MAP is also proportional to the number of scheduled users. C. MAC overhead due to MAP messages Based on () and (5), R MAP, the average number of the resource units used for MAP message transmission, can be evaluated as following, R MAP = E[D br (i)(b dmap (i)+b umap (i))] (6) where D br (i) is the number of resource units per bit used for transmitting broadcasting messages in frame i. Since D br (i) is independent of b dmap (i) and b umap (i), () is rewritten as following. R MAP = E[D br (i)]e[b dmap (i)+b umap (i)] = D br (b o +n b di +n ds b C +n us b ui ) (7) 85

3 schemes are employed, since the number of scheduled users in a frame increases due to improved capacity and additional MAP IEs for indicating the usage of MIMO and Band AMC scheme should be transmitted. Fig.. MAC overhead due to MAP messages vs. number of users where b o = b do + b uo, x denotes E[x(i)] for any random process x(i), and n is the number of MCS levels used in frame i, and can be approximated as following. { n = E x(i, j) nds, if n ds is small (8) N m, otherwise When n ds is small, then it is likely that at most one scheduled user utilizes MCS level j, which means that n dc (i, j) 1, j {1,,...N }, and n approaches to n ds. In addition, when n ds is big, then it is likely that at least one of the scheduled users utilizes MCS level j, which means that n dc (i, j) 1, j {1,,...N }, and n approaches to N m. Finally, V MAP, MAC overhead due to MAP messages, can be expressed as following. V MAP = R MAP = D br(b o +n b di +n ds b C +n us b ui ) (9) R d + R u R d + R u Fig. shows the result of the numerical analysis and simulation. In the simulation, it is assumed that the ratio of downlink and uplink subframe is 8:9 and the size of MAC PDU is 00 and 1800 bit. Other simulation parameters are based on [4] and [9]. As the number of users increases, the amount of MAP IE increases and MAC overhead consequently increases. Since the amount of resource is limited, the supportable number of scheduled users has limits and thus MAC overhead gets saturated when the number of users becomes 40 in case that MAC PDU size is 00 bits and 15 in case that MAC PDU size is 1800 bits, respectively. In addition, MAC overhead increases as the MAC PDU size gets small, because the number of scheduled users in a frame increases and consequently the number of MAP IEs in a MAP message increases. The results show that MAC overhead due to MAP message in the IEEE 80.16e system is up to 5% and causes to limit system capacity. The main factor of the problem is that the IEEE 80.16e system broadcasts the MAP message with the MCS level providing robustness for every user and some MAP IEs for the MSs having good channel quality are transmitted with very low data rate. MAC overhead becomes larger when multiple antenna and frequency selective transmission IV. PIGGYBACKING MAP IE ON DATA TRANSMISSION The proposed scheme transmits MAP IEs piggybacked on downlink data instead of broadcasting. Using AMC scheme, data packets are transmitted with maximal data rate based on the channel quality of the receiving MS. By piggybacking MAP IEs on the data, the MCS level used for data transmission is applied to the MAP IE and the data rate of MAP IE increases. Let N = {1,,...} and S = {1,,...S} be the set of frame indice and, respectively. Let S i be the set of indices scheduled in frame i and IE(i, j) be the MAP IE describing the data burst for j in frame i. LetD u (i, j) be the number of downlink subcarriers per bit needed for data transmission to user j in frame i. The operation of the proposed scheme is given as following. 1) Initialization. t(i, j) =0, i N and j S i =1(initial frame index is 1) ) Decide whether to include the IE(i, j) in MAP. For all j S i. If t(i, j) =0, include IE(i, j) in MAP. Otherwise, skip to transmit IE(i, j). ) Decide whether to piggyback IE(i + δ p,j) on the data packet For all j S i, piggyback IE(i + δ p,j) if D u (i, j) >D TH 4) MS with j gets the data packet and sends feedback. 5) BS ensures whether the IE(i + δ p,j) is received correctly. If ACK for the data packet including IE(i + δ p,j) is received, t(i + δ p,j)=1 6) Go to next frame. i = i +1 and go to step ) In this operation, IE(i + δ p,j) is transmitted in frame i, which means that MAP IE for the burst which will be transmitted after δ p frames is made in current frame. In fact, the piggybacked MAP IEs should describe the resource allocation in future frame, because it is meaningless that piggybacked MAP IEs indicate data bursts in the current frame. After decoding the piggybacked MAP IE with data packet, MS notices the data transmission which occurs after δ p frames. Thus, the proposed scheme is suitable for the traffic whose pattern is predictable, such as VoIP and real-time streaming traffics. Note that the proposed scheme is compatible with conventional protocols in the IEEE 80.16e system, because the proposed scheme assumes to use broadcasting MAP message. Like conventional way, BS can broadcast the MAP IE in MAP message when it was not sent before in piggybacked way. Also, the 86

4 MAP IE Data MAP IE for retransmission MAP region Data region 1 BS time slot Packet loss HARQ MS time slot Feedback ACK NACK ACK region Fig. 4. Example of proposed scheme applied to the IEEE 80.16e system Percentage ofresource saved by the proposed schem e sim,du = 1 anal,du= 1 sim,du = 0.5 anal,du = 0.5 proposed scheme can utilize the conventional feedback channel for the feedback of MAP IE. Thus, the proposed scheme does not require the change of the conventional standard and protocols. In addition, the proposed scheme has novelty unlike the conventional piggybacking schemes which piggybacks simple control information such as bandwidth request or Automatic Retransmission Request(ARQ) feedback. The proposed scheme piggybacks the resource allocation information and uses feedback mechanism to assure the transmission reliability of piggybacked information. Fig. 4 shows the operation of the proposed scheme applied to the IEEE 80.16e system. A BS transmits a data packet at a periodic interval of 5 frames and uses HARQ scheme so that a fast feedback channel is available. The BS initially broadcasts MAP IE for data packet 1 and piggybacks the MAP IE for data packet which will be transmitted after 5 frames. The MS decodes data packet 1 and the piggybacked MAP IE, and sends ACK through the HARQ feedback channel. BS then transmits data packet and piggybacked MAP IE for data packet. The MS detects error from the data burst and sends NACK. Then BS broadcasts the MAP IE for the retransmission, and retransmits the data packet and MAP IE. After the MS gets the retransmission properly, it gets ready to decode data packet and the piggybacked MAP IE for the latter data packet. The proposed scheme can effectively minimize MAC overhead, since it increases the average data rate of MAP IEs and consequently saves the amount of radio resource for the MAP IEs. The saved radio resource can be easily utilized for data transmission. However, if D u (i, j) is somehow low, the piggybacked MAP IE is transmitted with low data rate, which leads to bring little performance gain. Thus, when D u (i, j) is lower than a threshold, D TH, it is better to transmit MAP IE based on conventional way. In addition, the proposed scheme may cause additional channel estimation error due to resource allocation of future transmission. Then, packet error rate will increases and the system throughput may be degraded. V. ANALYSIS AND SIMULATION FOR PROPOSED SCHEME A. Analysis R p (i), the amount of the radio resource used for piggybacked MAP IEs in frame i, is evaluated as following. R p (i) = n ds(i) n us(i) D u (i, j)b di + D u (i, j)(b ui b C ) (10) E[R p (i)] = D u (n ds b di + n u s(b ui b C )) (11) Fig QPSK 1/1 QPSK 1/8 MCS levelused forbroadcasting Radio resource gain of the proposed scheme for various cases In (10), the first term is the summation of the resource for piggybacked DL MAP IEs and the second term is the one for UL MAP IEs. Note that piggybacked MAP IEs don t contain fields. Using (11), R pmap, the average amount of the resource that the proposed scheme uses for MAP transmission, becomes as following, R pmap = D br (b o +E[b ret (i)])+e[r p (i)] (1) E[b ret (i)]= α(n b di + n ds b C + n us b us ) (1) where b ret (i) is the total size of the MAP IEs which are not delivered properly due to packet error and retransmitted in frame i, and α is the average packet error rate. Since the piggybacked MAP IEs is lost with the rate of α, those MAP IEs are broadcasted in MAP message. Using (7) and (1), R s, the average amount of the resource saved by the proposed scheme, is evaluated as following. R s = R MAP R pmap = D br (1 α)(n b di +n ds b C +n us b us ) E[R p (i)] = D br (1 α)r MAP D u (n ds b di + n us b ui ) (D br (1 α) D u )R MAP (14) (14) is evaluated with an assumption that R MAP (n ds b di + n us b ui ). (14) reveals that the resource gain of the proposed scheme increases as D u decreases. This means that the proposed scheme can obtain sufficient gain when the channel quality of the MS is good and high data rate could be supportable for the MS. B. Simulation We simulate for the performance of proposed scheme in the IEEE 80.16e TDD system [4] under the same simulation environment mentioned in section III. Fig. 5 shows the simulation result and numerical result for the percentage of the resource saved by the proposed scheme. The results show that the resource gain can be up to 0-70%, and the percentage increases as the average data rate of the MSs D u increases. Fig. 6 shows the simulation result with respect to the maximal MAC throughput of the proposed scheme compared to that of the conventional scheme. It shows 87

5 Fig. 6. MAC througput vs. number of MSs for conventional and proposed schemes that the maximal MAC throughput can be improved up to 0%. This assures that the proposed scheme can effectively reduce the MAC overhead and contribute to improve system capacity. In addition, the maximal MAC throughput slightly decreases as the packet error rate increases, because MAP IE retransmission occurs and the more resource is used for it. Furthermore, we evaluate the effect of the additional channel estimation error caused by the piggybacked MAP IE using the channel model from ITU-R model [8]. RR and PF stand for round robin and proportional fair scheduler, respectively. MAP MCS denotes for the MCS level used for broadcasting MAP message. Fig. 7 and Fig. 8 show the simulation results of the goodput gain versus δ p in case of Pedestrian B 1 km/h channel model and 10 km/h channel model, respectively. In case of 1 km/h channel model, the goodput gain is obtained for any channel estimation delay due to prescheduling, because the channel variation is relatively slow. In addition, PF scheduler brings more goodput gain than RR scheduler. This is because PF scheduler can improve the average channel quality of the scheduled MSs. However, in case of 10 km/h channel model, the proposed scheme improves goodput in case of small δ p but goodput loss occurs when δ p is big, because of the rapid variation of channel quality. In addition, RR scheduler may bring more goodput gain than PF scheduler when δ est is sufficiently large. This is because RR scheduler generates piggybacked MAP IEs based on long term channel quality, which varies slowly. VI. CONCLUSIONS The IEEE 80.16e system suffers critical MAC overhead problem and the main factor of the problem is that the MAP message is broadcasted with the MCS level providing low data rate. The throughput degradation due to MAP overhead reveals serious problem. To solve it, the piggybacked scheme is proposed to increase the data rate of MAP IE messages. The numerical analysis and simulation results show that the proposed scheme leads to reduce MAC overhead and improves goodput up to 0% in ideal case, but it should be carefully applied since the proposed scheme may sometimes bring goodput degradation due to channel estimation error. Goodput improvement ratio MAP:QPSK1/1 MAP : QPSK 1/8 MAP : QPSK 1/4 RR PF Channel estimation delay due to prescheduling (frame) Fig. 7. Goodput gain vs. channel estimation delay δ p for Ped B 1km/h channel model Goodput improvement ratio MAP:QPSK1/1 MAP : QPSK 1/8 MAP : QPSK 1/4 RR PF Channel estimation delay due to prescheduling (frame) Fig. 8. Goodput gain vs. channel estimation delay δ p for Ped B 10km/h channel model REFERENCES [1] Taesoo Kwon, et al, Design and Implementation of a Simulator Based on a Cross-Layer Protocol between MAC and PHY Layers in a WiBro Compatible IEEE 80.16e OFDMA System, IEEE Communication Magazine, Vol. 4, No. 1, pp , December 005. [] Ariton E. Xhafa, et. al, MAC Performance of IEEE 80.16e, IEEE Vehicular Technology Conference 005 Fall, September 005 [] Juhee Kim, et al, A New Efficient BS Scheduler and Scheduling Algorithm in WiBro Systems, International Conference on Advanced Communication Technology 006, Vol., pp , February 006. [4] IEEE Std 80.16e, Part16: Air interface for Fixed Broadband Wireless Access Systems, February 006. [5] Yongjoo Tcha, et al, A Compact MAP Message to Provide a Virtual Multi-frame Structure for a Periodic Fixed Bandwidth Assignment Scheme, IEEE C80.16e-04, 68r, November 004. [6] Itzik Kitroser, et al, Compressed DL-MAP Format for OFDMA PHY,, IEEE C80.16d-0/6, September 00. [7] Alcatel, OFDM Downlink Scheduling for Control Channels, GPP TSG-RAN WG #49, R , November 005. [8] ITU-R, Guidelines for Evaluation of Radio Transmission Technologies for IMT-000, ITU-R M.15, February [9] Doug Gray, et al, Mobile WiMAX - Part I: A Technical Overview and Performance Evaluation, WiMAX Forum, pp.1-9, August