Cross-layer queuing analysis on multihop relaying networks with adaptive modulation and coding K. Zheng 1 Y. Wang 1 L. Lei 2 W.

Transcription

1 Pubished in IET Communications Received on 18th June 2009 Revised on 30th Juy 2009 ISSN Cross-ayer queuing anaysis on mutihop reaying networks with adaptive moduation and coding K. Zheng 1 Y. Wang 1 L. Lei 2 W. Wang 1 1 Wireess Signa Processing and Network Lab, Key Laboratory of Universa Wireess Communication, Ministry of Education, Beijing University of Posts & Teecommunications, Beijing , Peope s Repubic of China 2 Wireess System Lab, China Mobie Research Institute, Beijing , Peope s Repubic of China E-mai: zkan@bupt.edu.cn Abstract: Mutihop reaying is one of the promising techniques in future generation wireess networks. The adaptive moduation and coding (AMC) mechanisms can be appied in order to increase the spectra efficiency of wireess mutihop networks. However, most of these mechanisms concentrate on the physica ayer without taking the queuing effects at the data ink ayer into account, whose performances are overestimated. Therefore the cross-ayer anaytica framework is presented in anaysing the quaity-of-service (QoS) performances of the decode-and-forward (DF) reaying wireess networks, where the AMC is empoyed at the physica ayer under the conditions of unsaturated traffic and finite-ength queue at the data ink ayer. Considering the characteristics of DF reaying protoco at the physica ayer, the authors first propose modeing a two-hop DF reaying wireess channe with AMC as an equivaent Finite State Markov Chain (FSMC) in queuing anaysis. Then, the performances in terms of queuing deay, packet oss rate and average throughput are derived. The numerica resuts show that the proposed anaytica method can be efficienty appied for studying the issues incuding the reay depoyment and the cross-ayer design in the mutihop reaying networks. 1 Introduction Mutihop reaying is one of the promising techniques in future generation wireess networks [1]. It reduces the transmission distance and increases the amount of users under more favourabe channe conditions, aowing for better channe quaity and higher throughput [2]. Severa repetition-based cooperative reaying schemes such as ampify-and-forward (AF), decode-and-forward (DF) have been aready deveoped to fuy expoit the spectra diversity for reducing the outage probabiity [3]. However, these schemes usuay decrease the spectra efficiency of the system because of their repetition-based structure. On the other hand, by adjusting the transmission parameters to the instantaneous ink quaity, adaptive moduation and coding (AMC) technique aim at improving both spectra efficiency and ink reiabiity. Consequenty, the appications of AMC in mutihop reaying networks can provide high throughput [4 6]. Most of the iteratures ony concentrate on the physica ayer to maximise the data throughput of the individua ink with the constraint of packet error rate (PER), where the queuing effects at the ink ayer are not taken into account. In fact, because of the dynamic behaviour of packet arrivas, the queues may be empty even though the wireess channe can provide the transmission rate that is required by the system. Besides the transmission error, the overfow due to finite-ength buffer causes the packet oss. In addition, the service process of the queue with AMC is not deterministic as that with nonadaptive moduation, especiay in the wireess networks with cooperative reaying schemes. Recenty, Liu proposed a cross-ayer anaytica framework for singe-antenna ceuar systems, where the AMC scheme with finite-ength queue is appied to provide IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010

2 quaity-of-service (QoS) supports [7]. This framework was then appied in mutipe-input mutipe-output (MIMO) systems to anayse the queuing mode with AMC, especiay on the deay constrained traffic [8]. Zhou et a. [9] further proposed a cross-ayer design of diversitymutipexing switching agorithm to optimise the QoS performance of MIMO systems. Aso, Liu s framework was extended directy into two-hop wireess reay network without considering the characteristics of reaying behaviour at the physica ayer [10]. Furthermore, in order to support more than one cass of traffic in a physica frame, a physica and appication ayer cross-ayer adaptation mechanism is proposed in [11]. However, in our knowedge, there is amost no work on queuing performance anaysis of mutihop DF reaying networks so far, which sti remains open. In this paper, we anayse the queuing behaviour of DF reaying wireess networks with AMC under the conditions of unsaturated traffic and finite-ength queue. For the sake of impementation, the common moduation and coding scheme (MCS) is assumed in both hops and its seection can be based on the equivaent signa-to-noise ratio (SNR) in DF reaying networks. Then, a two-hop DF reaying wireess channe can be modeed as an equivaent Finite State Markov Chain (FSMC). In other words, the AMC with common MCS in both hops yieds an equivaent varying-rate queue server. Correspondingy, an anaytica method is presented to evauate the QoS performance incuding packet oss rate, queuing deay and average throughput in a two-hop DF reaying network. Finay, in order to demonstrate its effectiveness, the proposed anaytica method is appied in studying the issues incuding the reay depoyment and the cross-ayer design. The rest of this paper is organised as foows. Section 2 gives a brief description of the system mode. The performances of wireess networks with DF reaying are anaysed in Section 3. In Section 4, the numerica resuts are presented and discussed. Finay, Section 5 gives the concusion. Notations: A circuary symmetric compex Gaussian variance x with mean m and covariance R is denoted x CN (m, R): E[x] denotes the expectation of the random process x. 2 System mode Fig. 1 iustrates a simpe two-hop wireess cooperative communication network, where a reay (R) node cooperates with a source (S) node to transmit information to a destination (D) node. A the channes incuding source to reay (S-to-R) and reay to the destination (R-to-D) are independent to each other and are assumed to be fat Rayeigh fading for the sake of simpification. There is aso independent, zero-mean additive white Gaussian noise with unit variance at each receiver. In our anaysis, we Figure 1 Iustration of a simpe two-hop cooperative reay system mode normaise the distance between the S and the D, and assume that R is ocated between the S and the D, on the straight ine connecting S and D. We denote the S-to-R distance as d and the R-to-D distance as 1 d, respectivey, where 0, d, 1. One frame for two-hop transmission with the duration of T f consists of two successive time sots. In the first time sot, the ink ayer packets are buffered in a queue at the source node, and merged into physica ayer frames for the first hop transmission. Then, the reay node detects and decodes the received signa from the source. The packets which pass cycic redundancy check wi be re-encoded and remoduated for the second hop transmission in the second time sot. The outputs from S and from R are denoted as x 1 and bx 2, respectivey. We assume that E{jx 1 j 2 } ¼ P 1 and E{jbx 2 j 2 } ¼ P 2 are the ong-term average transmit power from S and R, respectivey. For a fixed pathoss exponent a, the average received power is dependent on the distance between the transmitter and receiver [12]. Usuay the pathoss exponent depends on the environmenta factors and its vaue is normay in the range of 2 4. The corresponding signas received by R and D are expressed as y 1 ¼ d a b 1 x 1 þ z 1 y 2 ¼ (1 d) a b 2 bx 2 þ z 2 (1) where b i denotes the independent compex fading channe gain of the ith hop transmission, modeed as b i CN (0, s 2 i ) with s 2 i ¼ E{jb i j 2 }, i [ {1, 2}. For sow fading (quasi-static) cases, the fading coefficients are constant within the transmission of the entire hop. Without oss of generaity, we assume that the noise term z i has equa variance s 2 N and is modeed as z i CN (0, s 2 N ). The instantaneous SNR at R and D can be cacuated as g (1) ¼ d a jb 1 j 2 ðp 1 =s 2 N Þ and g (2) ¼ (1 d) a jb 2 j 2 ðp 2 =s 2 N Þ, respectivey. 2.1 FSMC channe mode per hop Assume that there are L MCS modes avaiabe for AMC transmission in the system. Then, an FSMC S¼ {s (i) 1, s(i) 2,..., s(i) L }, i [ {1, 2}, with L states is used to refect the fading channe in the ith hop transmission [13]. We denote the state transition probabiity from s to s m and steady-state probabiity of state s as,m and p(i), respectivey. Then, ¼ [,..., p(i) L ] is the 1, p(i) IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010

3 stationary state distribution vector, and P (i) represents the transition probabiity matrix with,m,, m [ {1, 2,..., L} in the ith hop transmission. Furthermore, the transitions are assumed to ony happen between adjacent states, that is,m ¼ 0, 8j mj. 1,, m [ {1, 2,..., L} (2) Let G ¼ [G 1, G 2,..., G Lþ1 ] be the region boundary of the received SNR, by which the entire SNR region is partitioned into L þ 1 parts. For simpicity, the SNR threshod vaues for MCS seection in AMC are set as the region boundary in this paper. These threshod vaues are in the ascending order with G 1 ¼ 0 and G Lþ1 ¼ 1. Since the radio channes of two hops are assumed to be independent and statisticay identica, the same set of boundary can be used for both hops so that the index i of the hop is omitted. The channe is in state s if the instantaneous SNR g (i) is between G and G þ1. The adjacent-state transition probabiity over one frame period T f can be cacuated as,þ1 ¼ x(g þ1)t f p, ¼ 1,..., L 1 (3), 1 ¼ x(g )T f p, ¼ 2,..., L (4) Here, x(g ) denotes the eve cross rate at an instantaneous SNR in the ith hop transmission. Since the wireess channe is assumed to be Rayeigh fading, the received instantaneous SNR g (i) is distributed exponentiay with probabiity density function! f (g (i) ) ¼ 1 g(i) exp (i) g g (i), i [ {1, 2} (5) where g (i) ¼ E[g (i) ] is the average received SNR. Then, x(g ) can be expressed as x(g ) ¼ sffiffiffiffiffiffiffiffiffiffi 2pG g (i) f (i) D exp G g (i), i [ {1, 2} (6) where f (i) D denotes the mobiity-induced Dopper spread in the channe of the ith hop transmission. Finay,, can be derived from the normaising condition P L m¼1 p(i),m ¼ 1as, ¼ 8 >< >: 1,þ1 p(i), 1, ( ¼ 2,..., L 1) 1,þ1, ( ¼ 1) 1, 1, ( ¼ L) Meanwhie, the stationary probabiity (7) that the Markov chain is in state s can be given by [13] ¼ ð Gþ1 G f (g (i) )dg (i) ¼ exp(g =g (i) ) exp(g þ1 =g (i) ) 3 Performance anaysis 3.1 Arriva process Let A t denote the number of packets arriving from the upper ayer at the source in frame t. This arriva process A t is stationary with the mean of E[A t ] ¼ T f and independent of the queue state as we as the service process. Usuay A t is assumed to be Poisson distributed with parameter of T f for simpicity P(A t ¼ a) ¼ (T f ) a exp( T f ), a [ {0, 1,..., 1} a! The vaue of the random variabe A t is the non-negative integer, that is A t [ A¼{0, 1,..., 1}. 3.2 Queue service process Considering the characteristics of DF reaying scheme, the transmission over a two-hop DF reaying channe can be regarded as a dynamic server. The mutirate transmission is achieved by AMC schemes, where the transmitted data rate can be adjusted depending on the channe conditions of both hops Adaptive moduation and coding: For the reay with simpe functions, the common MCS scheme is adopted in different hops with the knowedge of the channe quaities of both hops. In this way, the packets received at the reay can be forwarded immediatey with no need of buffering. In DF reaying networks, the equivaent instantaneous SNR of two-hop transmission can be given by [3] (8) (9) g DF ¼ min {g (1), g (2) } (10) Then, the MCS seection can be made according to this equivaent SNR. The objective of AMC is to maximise the system throughput by adjusting the transmission parameters ike moduation order and coding rate according to the avaiabe channe state information (CSI), whie maintaining the required PER of P 0. The th MCS mode is chosen when the equivaent SNR of DF reaying channe enters the range as [14] Mode is chosen when g DF [ [G, G þ1 ) (11) with the corresponding rate of R, [ {1,..., L}. When the deep channe fading occurs, that is G 1 g DF, G 2,nodata IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010

4 transmission is aowed, which corresponds to the mode 0 with rate R 1 ¼ 0 bits/symbo. Assuming that b radio resource units are aocated to one user by the system, the number of packets transmitted per frame is denoted as c ¼ br. Then, et C t denote the number of packets transmitted using AMC scheme in frame t, which is a random variabe with the non-negative integer vaues taken from the set C¼{c 1, c 2,..., c L } Equivaent FSMC of DF reaying: In DF reaying networks, the AMC with common MCS in both hop yieds an equivaent varying-rate queue server. Owing to the Poisson arrivas, this queue server is the one with vacations. Correspondingy, we define a new equivaent FSMC H¼{h 1, h 2,..., h L } with L states for DF reaying transmission, which considers the effects of both channe varying character and DF reaying scheme. In this newy defined FSMC, a user is regarded to be in the state h [ H ony when the worse ink of two hops is in channe state s [ S, 1 L. The corresponding data rate in state h is the same as s in the FSMC per hop. Next, et us discuss the stationary state distribution vector ~p ¼ [ ~p 1, ~p 2,..., ~p L ] and the transition probabiity matrix ~P of the equivaent FSMC, where ~P is formed by ~p,m with, m [ {1, 2,..., L}. Under the assumption that the channes in both hops are independenty, the steady-state probabiity of a user in DF reaying channe can be first computed by ~p ¼ m¼þ1 m þ m¼þ1 m þ (12) Simiar to the singe-hop channe, the transitions in the DF reaying channe are assumed to happen ony between adjacent states, that is ~p,m ¼ 0, 8j mj. 1,, m [ {1, 2,..., L} (13) There are severa possibiities that the state of a user may change from h to h þ1 in the DF reaying channe, that is Case 1: The first hop is in state s whie the second hop is in state s m, m [ { þ 1,..., L} at the current time. Then, at the next time, the former is in state s þ1 whie the atter is in state s n, n [ { þ 1,..., L}. Case 2: The second hop is in state s whie the first hop is in state s m, m [ { þ 1,..., L} at the current time. Then, at the next time, the former is in state s þ1 whie the atter is in state s n, n [ { þ 1,..., L}. Case 3: Both the first hop and the second hop are in state s at the current time. Then, at the next time, both of them are in state s þ1. Simiar cases wi happen when the state of a user switches from h to h 1. Therefore the adjacent-state transition probabiity in the DF reaying channe is cacuated by ~p,þ1 ¼ 1 ~p þ,þ1 m¼þ1,þ1 m¼þ1 ~p, 1 ¼ 1 ~p þ m, 1 m¼þ1, 1 m¼þ1 m m n¼þ1 m n¼þ1 n¼ 1 n¼ 1 m,n m,n þ X2 m,n i¼1 m,n þ X2 i¼1,þ1, 1!! (14) Finay, ~p, can be derived from the normaising condition P L m¼1 ~p,m ¼ 1as 8 >< 1 ~p,þ1 ~p, 1, ( ¼ 2,..., L 1) ~p, ¼ 1 ~p,þ1, ( ¼ 1) >: 1 ~p, 1, ( ¼ L) 3.3 Queue state transition (15) According to the anaytica framework in [7], not ony the queue service process but aso the queue state have to be considered to anayse the system performances. The queue state Q t is described as the number of packets in the queue at the end of frame t, or equivaenty, at the beginning of frame t þ 1. Its vaue region is denoted as Q t [ Q¼ {0, 1,..., K } with the queue buffer ength of K. Depending on the current channe state, the transmitter at the source moves at most C t packets out of the queue to the physica ayer at the beginning of frame t. Thus, the number of free sots in the queue at the beginning of frame t is F t ¼ K max{0, Q t 1 C t } (16) Then, the A t arriving packets are tried to be paced in the queue throughout frame t. IfA t. F t, ony F t packets can be put into the queue and A t F t packets are dropped due to finite-ength queuing. Otherwise, a A t packets are accepted by the queue. Therefore the queue transition can be given by Q t ¼ min{k, max{0, Q t 1 C t } þ A t } (17) Next, an joint queuing and serve-rate FSMC with a state pair (Q t 1, C t ) is constructed in order to anayse the system behaviour. Its state transition matrix ^P can be obtained with the entry ^P (u,c),(v,m) defined as the transition probabiity from (Q t 1 ¼ u, C t ¼ c) to(q t ¼ v, C tþ1 ¼ m). 298 IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010

5 Then, this entry can be simpified as ^P (u,c),(v,m) ¼ P(Q t ¼ v, C tþ1 ¼ mjq t 1 ¼ u, C t ¼ c) ¼ P(C tþ1 ¼ mjc t ¼ c)p(q t ¼ vjq t 1 ¼ u, C t ¼ c) (18) where P(C tþ1 ¼ mjc t ¼ c) can be found from the entries of ~P, and P(Q t ¼ vjq t 1 ¼ u, C t ¼ c) is cacuated as P(Q t ¼ vjq t 1 ¼ u, C t ¼ c) 8 >< P(A t ¼ v max{0, u c}), if 0 v, K ¼ 1 P 0v,K P(Q t ¼ vjq t 1 ¼ u, C t ¼ c), >: if v ¼ K (19) The stationary distribution vector ^p of this FSMC (Q t 1, C t ) can be computed by ^p ¼ ^p ^P, X 8u,8c ^p (u,c) ¼ 1 (20) where ^p (u,c) ¼ im t!1 P(Q t 1 ¼ u, C t ¼ c) is the entry of ^p, which is proved to exist and be unique in [7]. 3.4 System performance Corresponding to the number of free sots in frame t in (16), the number of packets dropped at frame t can be easiy given by D t ¼ max{0, A t K þ max{0, Q t 1 C t }} (21) Since the imit distribution of the stationary A t exists when t! 1, we define A ¼ im t!1 A t and obtain P(A ¼ a) ¼ P(A t ¼ a) (22) Then, with the simiar method in [7], the average number of packets dropped per frame can be cacuated by D ¼ E[ im D t ] t!1 X ¼ a[a,u[q,c[c max{0, a K þ max{0, u c}} P(A ¼ a)p(q ¼ u, C ¼ c) Next, the average packet drop rate P d can be computed by P d ¼ im T!1 (23) P T t¼1 D t D P T t¼1 A ¼ (24) T t p So, the packet oss rate and average throughput of DF reaying systems can be obtained as foows, respectivey j ¼ 1 (1 P d )(1 P 0 ) (25) h ¼ (T f )(1 j) (26) Aso, the average queuing deay is computed according to Litte s aw [15] t ¼ where the average queue ength given by Q ¼ X u[q,c[c Q (1 P d )T f (27) Q ¼ E[ im t!1 Q t ] is up(q ¼ u, C ¼ c) (28) 4 Numerica resuts and anaysis In this section, the performances of two-hop DF reaying wireess networks are evauated by the numerica anaysis. We aso study the performances of the direct transmission (DT) between the source and destination as the reference. Most of the common parameters are summarised in Tabe 1. Define the ong-term transmit power difference between from R and from S as DP ¼ 10 og P 1 =P 2. The reated appications of the proposed anaytica method are given to demonstrate its effectiveness in DF reaying wireess networks. A of the channes are modeed as five-state independent and identicay distributed (i.i.d) FSMC with the same Dopper frequency. Correspondingy, the MCS modes for transmission are pre-defined in Tabe 2, referred as Mode 1 Mode 5. Athough there is no exact cosed-form PER for coded moduations, we can use the equivaent cosedform PER expression for MCS seection. Therefore for the sake of simpification, the PER performance of each MCS can be approximativey expressed as [16] PER (g) 1, if 0 g, h (29) a exp ( g g), if g h where g is the received SNR and the coefficients of a, g, h are mode dependent as shown in Tabe 2. These coefficients are generated by fitting and comparing curves to the simuated PER according to the Monte-Caro simuations with parameters given by 3G ong-term evoution (LTE) specification [17]. For various MCS modes on the assumption of target PER ¼ 10 2, the SNR boundaries can be found in Tabe Reay depoyment In Fig. 2, we study the variation of the average throughput with respect to reay ocation characterised by d, where different DP is assumed. For simpicity, the average received SNR with DT is set to be 8 db. When the reay is IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010

6 Tabe 1 Parameters assumption in a DF reaying networks Parameters Vaues frame ength (T f ) 2 ms dopper frequency ( f D ) 20 Hz resource unit per frame (b) 1 queue buffer ength (K ) 50 packet arriva rate () 2 pathoss exponent (a) 4 cose to the midde between the source and destination, the throughput is increased because of the improvement of the equivaent SNR with DF reaying. We can find that the optima ocation of the reay is the point where the average SNR of the first hop equas that of the second hop. For exampe, in the case of DP ¼ 0 db, the reay is optimay depoyed at d ¼ 0:5 with g (1) ¼ g (2) ¼ 20 db. Moreover, ony when the ocation of reay is within the given range, for exampe, d [ [0:2, 0:8] in case of DP ¼ 0dB, the throughput performance of two-hop transmission with DF reaying is better than that of direction transmission. Furthermore, with ess transmit power from R, that is DP. 0, the throughput becomes smaer because of the ower received SNR at the destination. Correspondingy, the queuing deay performances with the different ocation of reay and DP are aso given in Fig. 3, where the simiar concusion can be drawn. Figure 2 Throughput performances with the different reay ocation From the view of impementation, ess transmit power from R means ess cost. So, Fig. 5 compares the average throughput performances between two-hop DF reaying and direction transmission, where the average transmit power from R is changed. It is cear that the throughput gain of two-hop DF reaying transmission become ess with the increase of DP. When the power difference DP is more than 13 db, the throughput with two-hop DF reaying transmission is even worse than that with direction transmission. Therefore to determine the transmit power from R is the trade-off between the performance and the impementation cost. Then, we compare the anaytica and simuation resuts of the throughput performances with the different packet arriva rate in Fig. 4, where the reay is depoyed at the optima ocation. With the increase of the packet arriva rate, the throughput gain with DF reaying transmission becomes arger compared with the DT. However, it is foor in the region of the arge packet arriva rate because the data rate that provided by the networks is imited by the wireess channe. It is noted that the anaytica resuts are in ine with those simuation ones, which demonstrates the effectiveness of our anaytica method. 4.2 Cross-ayer design The proposed framework can aso be used in various crossayer optimisation schemes in DF reaying cooperative networks, exampe to find the optima P 0 on the physica ayer to minimise the packet oss rate on the data ink ayer. For simpicity, the average transmit power from S is assumed to be equa to that from R, for exampe P 1 ¼ P 2.Sothe corresponding optima ocation of the reay is in the midde of the source and destination. Then, Figs. 6 and 7 show the packet oss rate performances with the different queue ength Tabe 2 Approximate coefficients of MCS modes and SNR boundary Mode moduation QPSK 16QAM 16QAM 64QAM coding Rate 1/2 1/2 3/4 2/3 R a g h (db) G (db) IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010

7 Figure 3 Deay performances with the different reay ocation Figure 6 Packet oss rate performances with the different queuing ength Figure 4 Throughput performances with the different Figure 7 Packet oss rate performances with the different maximum Dopper frequency increase of P 0, the mode switching eves of AMC scheme is changed and consequenty affects the probabiity of rate seection. As a resut, P d is affected by P 0 indirecty and becomes ess, which may decrease the PER. Therefore we observe that the packet oss rate first decreases and then becomes arger with the increase of P 0.Then,theoptima P 0 can be obtained, that is, the one with the minimum packet oss rate. Furthermore, the performances of packet oss rate can be improved with arger queue ength K and Dopper frequency f D. Figure 5 Throughput performances with the different DP K and Dopper frequency f D,whereP 0 varies from 10 4 to As shown in (25), the packet oss rate is dependent on the vaue of P 0 directy, that is, the arger P 0 eads to the arger packet oss rate if P d is unchanged. However, with the 5 Concusion In this paper, we presented a cross-ayer anaytica framework in studying the queuing behaviour of DF reaying wireess networks with AMC and finite-ength buffer. The service process of two-hop DF reaying schemes can be modeed as an equivaent FSMC when taking into account the characters of DF reaying protoco. The QoS performances of DF reaying wireess networks are evauated in terms of queuing deay, packet oss rate and average throughput. IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010

8 By using the proposed anaytica method, we observe that the two-hop DF reaying transmission do not aways outperform the direct transmission, which depends on the ocation of the reay (R) node. Moreover, the performance gain is reated to the transmission power from R. Onthe other hand, the cross-ayer design is aso studied by using the proposed anaytica framework. Numerica resuts show that the packet oss rate first decreases and then becomes arger with the increase of PER, where the optima PER can be obtained, that is, the one with the owest packet oss rate. In the next steps, we wi extend this work by considering the foowing issues: the impacts of mutiuser scheduing, the reay seection and retransmission schemes in DF reaying wireess networks. 6 Acknowedgments This research was funded in part by the Natura Science Foundation of China ( ) and the Research Fund for the Doctora Program of Higher Education under Grant References [1] AHMED M.H., SYED I., YANIKOMEROGLU H.: On the performance of time division mutipe access-based mutihop fixed ceuar networks with respect to avaiabe frequency carriers, IET Commun., 2008, 2, (9), pp [2] SENDONARIS A., ERKIP E., AAZHANG B.: User cooperation diversity-parts I & II, IEEE Trans. Commun., 2003, 51, (11), pp [3] LANEMAN J.N., TSE D.N.C., WORNELL G.W.: Cooperative diversity in wireess networks: efficient protocos and outage behavior, IEEE Trans. Inf. Theory, 2004, 50, (12), pp [4] LIN Z., ERKIP E., GHOSH M.: Adaptive moduation for coded cooperative systems. Proc. IEEE Sixth Workshop on Signa Processing Advances in Wireess Communications, June 2005, pp [5] YAZDIAN E., PAKRAVAN M.R.: Adaptive moduation technique for cooperative diversity in wireess fading channes. Proc. IEEE Persona, Indoor and Mobie Radio Communications, 2006, pp. 1 5 [6] STRINATI E.C., SHENG Y., BELFIORE J.C.: Adaptive moduation and coding for hybrid cooperative networks. Proc. IEEE Int. Conf. Communications, June 2007, pp [7] LIU Q., ZHOU S., GIANNAKIS G.B.: Queuing with adaptive moduation and coding over wireess inks: cross-ayer anaysis and design, IEEE Trans. Wire. Commun., 2005, 4, (3), pp [8] HARSINI J.S., LAHOUTI F.: Queuing with adaptive moduation over MIMO wireess inks for deadine contrained traffic: cross-ayer anaysis and design. Proc. IEEE Int. Conf. Communications, June 2007, pp [9] ZHOU S., ZHANG K., NIU Z., YANG Y.: Queuing anaysis on MIMO systems with adaptive moduation and coding. Proc. IEEE Int. Conf. Communications, May 2008, pp [10] CHENG P., YU G., ZHANG Z., JIA H., CHEN H., QIU P.: Cross-ayer performance anaysis of two-hop wireess inks with adaptive moduation. Proc. IEEE Persona, Indoor and Mobie Radio Communications, September 2006, pp. 1 5 [11] QUAZI T., XU H.J., TAKAWIRA F.: Quaity of service for mutimedia traffic using cross-ayer design, IET Commun., 2009, 3, (1), pp [12] GOODMAN D.J.: Wireess persona communications (Addison Wesey, 1997) [13] ZHANG Q., KASSAM S.A.: Finite-state Markov mode for Rayeigh fading channes, IEEE Trans. Commun., 1999, 47, (11), pp [14] ALOUINI M.S., GOLDSMITH A.J.: Adaptive moduation over Nakagami fading channes, J. Wire. Commun., 2000, 13, (1 2), pp [15] KLEINROCK L.: Queueing systems. Vo. 1: theory (Wiey, New York, 1975) [16] LIU Q., ZHOU S., GIANNAKIS G.B.: Cross-ayer combining of adaptive moduation and coding with truncated ARQ over wireess inks, IEEE Trans. Wire. Commun., 2004, 2, (5), pp [17] 3GPP TS Evoved Universa Terrestria Radio Access (E-UTRA); Physica channes and moduation, IET Commun., 2010, Vo. 4, Iss. 3, pp & The Institution of Engineering and Technoogy 2010