A Novel Retransmission Strategy without Additional Overhead in Relay Cooperative Network

Transcription

1 A Novel Retransmission Strategy without Additional Overhead in Relay Cooperative Network Shao Lan, Wang Wenbo, Long Hang, Peng Yuexing Wireless Signal Processing and Network Lab Key Laboratory of Universal Wireless Communication, Ministry of Education Beijing University of Posts and Telecommunications, Beijing China Abstract A new retransmission strategy in relay cooperative network is proposed in this paper. The strategy removes the additional resource consumption completely of the conventional retransmission, thus is termed Retransmission with No Additional Overhead. According to, who initiates and when to start the retransmission are both decided adaptively on the channel quality indicator. consumes no additional overhead for that the retransmitted packet can be transmitted with a new packet simultaneously by hierarchical modulation, and the strategy s adaptive processing can guarantee that the new packet is sent without delay and its received quality also is not affected. Compared with the conventional strategy termed Retransmission with Additional Overhead RAO here, both analysis and simulation results show that can effectively lower the outage probability in actual observational signal-tonoise-ratio SNR regions, and the performance of throughput is improved significantly in all SNR regions. I. INTRODUCTION Recently, cooperative relay mechanism has been proposed as an effective way to reduce the outage probability and increase the diversity gain as well as throughput TP [1], [2]. The broadcast nature of radio ensures that each communication entity shares the limited resources in a cooperative manner, which improves resource utilization significantly; the destination D can receive the same information with various forms via different paths, thus a higher spatial diversity gain is achieved. Several protocols have been proposed in relay cooperative network, the popular two are amplify-and-forward AF and decode-and-forward DF [3]. In AF protocol, the relay R simply scales and forwards the analog signal waves received from the source S. In DF protocol, R decodes the received signal, re-encodes and forwards it to D. Considering that erroneously decoded bits at R could lead to severe error propagation, a selective DF SDF protocol is proposed in which a cyclical redundancy check CRC is needed at R, and only when the CRC succeeds does R participate in the cooperation, otherwise, R is idle and D recovers the original information only by the signal directly transmitted from S. The introduction of retransmission to relay cooperative network can further enhance the diversity effect [4], because the combination can provide duplicate signals towards D not only from spatial dimension but also temporal dimension. The retransmitted signal could come from S as well as R, and who initiates the retransmission depends on the strategy design, This work was supported by China NSFC under Grant /09/$ IEEE Fig. 1. Typical 2-hop relay cooperative system [5] proposes R does, while in [6], either S or R is likely to do. Although different in mechanisms, additional resources are consumed by all these retransmission strategies, which is usually intolerable especially for resource limited system. This paper proposes a new retransmission strategy which needs no additional overhead for relay cooperative network, termed Retransmission with No Additional Overhead. There are two features of, one is that who initiates and when to start the retransmission are both decided adaptively on the channel quality indicator CQI, the other is that the retransmitted packet RP can be transmitted with a new packet NP simultaneously by hierarchical modulation HM technology [7]. The merits of come from not only removing the additional resource consumption completely, but also that with adaptive processing, NP can be sent without delay and its received quality is not affected. Compared with a conventional retransmission strategy termed Retransmission with Additional Overhead RAO here, both the performance analysis and simulation results show that can effectively lower the outage probability in actual observational signal-to-noise-ratio SNR regions, and significantly improve TP in all SNR regions. The rest of the paper is organized as follows. The system model is presented in Section II. Section III first describes the conventional strategy RAO, and then the proposed one. Section IV analyzes the performances of RAO and from a theoretical point of view. Simulation results are shown in Section V. Section VI concludes the paper. II. SYSTEM MODEL A typical 2-hop relay system is shown in Fig. 1, which is based on time division half-duplexing in a quasi-static fading channel, where the fading coefficients are constant within one frame, but change independently in each frame. S, R and D are all equipped with single antenna. The system operates on SDF pattern when there is no demand for retransmission.

2 There are two slots for the transmission of one packet. At slot 1, S broadcasts its message to both R and D, assuming that subscripts SR, SD and RD denote the S-R channel, S- D channel and R-D channel in the sequel. Then the received signals at D and R, denoted by y SD and y SR are y SD = h SD x + n SD 1 y SR = h SR x + n SR 2 where x is the signal transmitted, whose power is normalized to one; h and n are the fading coefficient and the additive white Gaussian noise AWGN of corresponding channel, they are both i.i.d. CN 0, 1, i.e. mutually independent, zero mean, and circularly symmetric complex Gaussian variables of unit variance. Upon receiving y SR, R detects, demodulates, and decodes it, also a CRC is done. If x is obtained correctly, R re-encodes and forwards it at slot 2, otherwise, R doesn t participate in the cooperation. The received signal at D, transmitted from R, is expressed as y RD = h RD x + n RD 3 let σ 2 represent the noise power, then the received SNR γ of some link SR, SD or RD can be calculated with corresponding h and n, that is γ = h 2 /σ 2. D then combines the signals from S and R if any to construct the original information. If the decoding fails, D broadcasts a negative acknowledgement NACK signaling to S and R, while RP can be sent by S, by R or by these two, depending on the strategy design. The soft information of current received packet is added to the one of previous saved erroneous packets at D. The system returns to the non-retransmission state when RP is decoded correctly or the maximum number of retransmission is reached. III. RETRANSMISSION STRATEGY A. Review of Conventional Strategy RAO In conventional strategy RAO, upon receiving a NACK signaling, R checks whether it has decoded the packet needing to be retransmitted correctly. If the packet has gotten correct decoding at R in a certain previous transmitting round first transmission or retransmission, then R initiates the retransmission, and S sends nothing when the retransmission is conducted; whereas if R has never decoded the packet correctly yet, S assumes the initiator, and R deals with RP according to SDF protocol. For RAO, the resource utilization is restricted by the consumption of additional overhead, and the improvement of TP is limited. B. Proposed Strategy The proposed strategy completely offsets the deficiency of RAO by adopting HM. HM is designated as an alternative modulation method to the conventional modulation in the DVB-T standards [8]. It provides protection levels of hierarchy between the data streams. An example of hierarchical 16 quadrature amplitude modulation H is shown in Fig. 2, which is a normalized union of two quadrature phase Fig. 2. Hierarchical schematic diagram shift keying QPSK modulation. The first QPSK occupies a large quadrant, whereas the second QPSK subjected to the first one occupies a small quadrant within the selected position, and the power ratio α is 4 : 1. By standard Gray mapping, the first and the second QPSK associate with the first two and the last two bits of H respectively. Utilizing the property of H, two data streams which have been encoded with Turbo code and whose priorities are different can be overlapped in time domain, the one that has higher priority HP is mapped to the first two bits of H, and the other with lower priority LP mapped to the last two bits, thus transmission time resource can be saved. In, RP can be transmitted with NP simultaneously by HM, or can be transmitted by R who would be idle in current slot if it doesn t transmit RP, thus no additional overhead is needed. In order to guarantee that NP is sent without delay and its received quality is not affected, the HP stream is assigned to NP and the LP stream is assigned to RP. S and R are assumed to have the full knowledge of channel quality, the processing at S and R are presented as follows. 1 Processing at S: S has two transmitting options when it receives a NACK signaling, the first one is to transmit NP with QPSK, the second one is to transmit both NP and RP with H, whose HP stream and LP stream are allocated to NP and RP respectively. S transmits H if SD s channel quality can guarantee that the HP stream can be correctly decoded at D, otherwise, S transmits QPSK. To determine whether the decoding of the HP stream would succeed at D with the current SD s channel quality, a received for correct decoding is needed as reference. λ HP is related to coding rate CR and number of bits in one packet N Bit. If γ SD, the HP stream can be decoded correctly. The threshold λ HP can be obtained by reference to SNR threshold λ HP SNR-block error ratio BLER curve under AWGN with the same CR and N Bit. The BLER performance of HP stream, when N Bit equals to 2320, is shown in Fig. 3. In general, when BLER drops to below 10 3, the receiver usually can decode correctly, thus λ HP is identified as 4.8 db. 2 Processing at R: When R receives a NACK signaling, the processing is divided into two cases by whether R has decoded RP correctly in a certain previous transmitting round. If R does, it has RP s complete information, otherwise, R accumulates the soft information of RP round by round and examines if it is able to correctly decode RP after each cumulative. In each case, R determines the forward strategy from two factors, one is the transmitting signal from S, and

3 0.1 HP stream of H 0.01 BLER 1E-3 1E-4 AWGN Turbo1/ Bits SNRdB Fig. 3. BLER of the HP stream of H another one is the CRC result of the received signal at R in the current slot. The processing of each case is described as follows. a R has the complete information of RP: When S transmits NP, if R fails the CRC of NP, R forwards RP with QPSK; otherwise R decides forwarding QPSK or H according to the current RD s channel quality. The threshold is once again resorted for reference, and compared with channel RD s received SNR γ RD this time. When S transmits NP and RP with H, R forwards RP only, because what S transmits has proven that the SD s channel quality can guarantee NP s correct decoding at D. b R has the incomplete information of RP: If S transmits NP only, R works at SDF pattern, otherwise, R conducts CRC towards NP and RP respectively, thus there are four possible results. One is that both NP and RP are decoded correctly, R then forwards RP with QPSK for that what S transmits has proven NP s correct decoding at D; another two results can be classified as one situation, that is, only one data stream NP or RP obtains correct decoding, then R forwards the very stream only with QPSK; R also possibly forwards nothing, the case comes when both NP and RP are decoded erroneously. TABLE I summarizes the strategy. If RP can t be transmitted in the current frame, the retransmission is postponed to the frame when channel quality permits. λ HP IV. PERFORMANCE ANALYSIS In this section, the relation between TP and outage probability, as well as the one between outage probability and the received SNR, are first given, and then the received SNRs of RAO and are derived respectively. Performance outage probability and TP comparisons of the two strategies are offered finally. Define R bit information bits per packet, and the whole transmission spans D t frames. TP is shown as in [9], [10] TP = E [R bit] E [D t ] where E[.] represents the mathematic expectation of random variable in brackets. It s assumed that the maximum number of retransmission is M, when the number of retransmission 4 TABLE I STRATEGY R has the complete information of RP γ SD S CRC at R γ RD R NP Wwrong RP Ccorrect NP NP, RP NP, RP - - RP R has the incomplete information of RP NP W - Idle C - NP NP, RP NPW RPW - Idle NPW RPC - RP NPC RPW - NP NPC RPC - RP m 1 m M reaches M, but D still can t decode RP correctly, an outage event occurs. E[R bit ] can be expressed as E [R bit ] = R bit [1 p M] 5 where pm is the outage probability of the m th retransmission m p m = Pr γ k < λ 6 γk denotes the received SNR of the k th 0 for the first transmission retransmission, and λ is the received SNR threshold for correct decoding at D. E[D t ] is obtained as E [D t ] = 1+ M m=1 M 1 m q m+m p m = 2+ p m 7 m=1 where qm is the probability of success on the m th retransmission after that previous m 1 retransmissions are all failed, and qm = pm 1 pm, thus the second equality in 7 is derived. E[R bit ], E[D t ] both depend on pm, while pm has relation with γk 0 k m, γks of RAO and k = 0 and m as representives are derived respectively. A. Received SNR 1 Received SNR of RAO: Received SNR of NP at D is { γsd 0 + γ γ RAO 0 = RD 0 γ SR 0 > λ QPSK 8 γ SD 0 γ SR 0 < λ QPSK where λ QPSK is the received SNR threshold for QPSK correctly decoding. If the decoding fails, D returns a NACK signaling and the retransmission starts. NP then becomes RP. For the received SNR of m th retransmission γ RAO m, there are two possible forms by whether R has the complete information of RP, which are shown as in 9 and 10, then γ RAO m is obtained as in 11. RAO m = γ RD m 9

4 γ SD m + γ RD m γ SR m > λ QPSK γ RP RAO m = γ SD m γ SR m < λ QPSK 10 γ RAO m 1 = γrao RP m 1 or γrao RP γ RAO m = m γ RAO m 1 = γ RP RAO m 1, γ SR m > λ QPSK γ RP RAO m other cases 11 2 Received SNR of : When a NP is received at D, the received SNR has different expressions according to different cases. If there is no demand for retransmission, then γ nonack0 equals to γ RAO0 as shown in 8, otherwise, the case is further divided into two sub-cases by whether R has the complete information of RP pre RP pre denotes the packet demanding to be retransmitted in some previous frame that NP meets, and it s not the current study object, but NP. When R has it, γ 0 is shown in 12, otherwise γ 0 is shown in 13. The expression of γ 0 is chosen from γ nonack pre 0 = 0, γ RPpre γ RPpre 0 = 0 and γ RPpre 0 accordingly. γsd 0 γ SR 0 < λ QPSK γ SD 0 γ SD 0 +γ RD 0 γ SD m + αγrdm γ SD m γ SD 0 α γ SD 0 γ SR 0 > λ QPSK γ RD 0 γ SD 0 γ SR 0 > λ QPSK γ RD 0 γsd 0 γsd 0 γ SR 0 < λ QPSK γsd 0 γ SR 0 > λ QPSK 12 γ SD 0 +γ RD 0 α γ SD m γsd 0 13 When D receives NP, if the decoding fails, it returns a NACK signaling, and NP becomes RP. The received SNR of m th retransmission γ m has two possible forms by if R has the complete information of RP NP just mentioned. If R has it, γ m is as in 14; otherwise it is as in 15. m = γ RD m γ RDm γ RD m + γsdm γsd m, γ SR m < λ QP SK γ SD m, γ SR m > λ QPSK, γ RD m γsd m 14 γ SD m, γ RD m γsr conv 0 + γsdm + m β SR k > λ LP γ RP m = γ SD m, γ SDm γsr conv 0 + m β SR k < λ LP 0 other cases 15 γsr conv 0 in 15 is the equivalent value of the first transmission received SNR QPSK or H s HP, which is converted to the H s LP; λ LP is the received SNR threshold with which s LP can be correctly decoded, and β SR k is shown as in 16. Then γ m is as in 17. { γsr k γ SD k β SR k = 0 γ SD k 16 γ m = m γ m 1 = m 1 or γ m 1 = γ RP m 1 and γ SD m, γsr conv 0 + m β SR k > λ LP γ RP m other cases 17 B. Performance Comparisons For RAO and, with the γm, the outage probability per transmission is obtained by 6, and then E[R bit ], E[D t ] are obtained, thus TP also is. But the expressions of outage probability and TP haven t closed forms, so the case when channel SR is perfect which is reasonable for actual systems is considered. Under this assumption, R always has the information of RP, and γ RAO 0 equals to γ 0, whereas the expectations of γ RAO m and γ m are E [γ RAO m] = + E [γ m] = λ HP + fγ 0 SD dγ SD λ HP 0 γ RD f γ RD dγ RD 18 γ RD 1+α fγ RDdγ RD + + fγ λ HP SD + γsd 0 1+α + γ RDfγ RD dγ RD dγ SD 19 where f. represents the probability distribution function. 1 Comparison of outage probability: 18 and 19 are compared, if γ SD λ HP, +, the former is larger than the latter obviously; otherwise, 18 is smaller, but when γ RD λ LP, +, RP can be decoded correctly for both

5 Outage Probability E-3 SR RAO 25dB 15dB 5dB db SD ThroughPut Bit/symbol SR RAO dB 15dB dB SD db Fig. 4. Outage probability performance of and RAO Fig. 5. Throughput performance of and RAO RAO and, so only when γ RD 0, λ] LP and γ SD 0, λ] HP, may be worse than RAO of the outage probability performance. In the middle to high SNR regions, the probability that both γ RD 0, λ] LP and γ SD 0, λ] HP appear is small enough to be ignored, so may be worse than RAO of the outage probability performance only in the low SNR region, and this region is usually low enough to be not in the conventional study scope. 2 Comparison of TP: For the performance of TP, E[D t ]s of RAO and are first compared, E [D t ] of RAO obeys 7, but for, E [D t ] = 1, for that no additional time slots are assumed for retransmission. In addition, is superior to RAO of the outage probability performance in actual observational SNR regions, which also illustrates that E[R bit ] of is usually larger. With advantage in E[D t ] and E[R bit ], the superiority of in TP is obvious. V. SIMULATION RESULTS In this section, the simulation results of outage probability and TP are given. The Monte-Carlo simulation adopts single path Rayleigh fading channel and Turbo code whose CR is 1/2. The number of symbols per packet is 1160, and the is 4.8 db. S and R are assumed to know respective CQI, and M = 2. An average SNRρ region of channel SD is under observation. The ρ SD -outage probability performances of RAO and under different reliability of channel SR are given in Fig. 4. ρ SR is assigned 25 db, 15 db and 5 db respectively, while ρ RD ρ SD = 5dB. The results show that always displays better outage probability performance than RAO in actual observational average SNR regions, despite SR s channel reliability. It is noted that when outage probability is 10 2, there are 2 db and 1.5 db gain of over RAO threshold λ HP for ρ SR is valued 15 db and 25 db respectively. Besides, outperforms RAO more obviously with the increase of ρ SR, i.e. the advantage of is better reflected when SR s channel quality is good. The ρ SD -TP performances are given in Fig. 5. The TP of RANO is always higher than that of RAO, no matter how SR s channel quality is, which is demonstrated in Section III. B. In addition, the advantage of is even more evident in low to middle ρ SD regions, which illustrates the applicability of even for poor channel quality. The superiority in TP comes from the lower outage probability and the zero time resource consumption of. VI. CONCLUSION A retransmission strategy used in relay cooperative system consuming no additional resource is proposed. Adopting adaptive design and HM technology, RP can be retransmitted needing no additional time slot, NP can be transmitted without delay and not influenced on its received quality. Both analysis and simulation results show that the proposed strategy has a better performance of outage probability in actual observational SNR regions, and its throughput performance significantly outperforms the conventional retransmission strategy in all SNR regions. REFERENCES [1] A. Sendonaris, E. Erkip and B. Aazhang, User cooperation diversity, part I and part II, IEEE Trans. Commun, vol. 51, no. 11, pp , Nov [2] A. Nosratinia, T. E. Hunter, and A. Hedayat, Cooperative communiation in wireless networks, IEEE Commun. Mag, vol. 42, pp , Oct [3] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, Cooperative diversity in wireless networks: Effcient protocols and outage behavior, IEEE Trans. Inform. Theory, vol. 50, no. 12, pp , Dec [4] B. Zhao and M. C. Valenti, Practical relay networks: a generalization of Hybrid-ARQ, IEEE J. Select. Areas Commun, vol. 23, no. 1, pp. 7-18, Jan [5] I. Stanojev, O. Simeone, Y. Bar-Ness and C. You, Performance of multirelay collaborative Hybrid-ARQ protocols over fading channels, IEEE Commun. Lett, vol. 10, no. 7, pp , Jul [6] E. Zimmermann, P. Herhold and G. Fettweis, The impact of cooperation on diversity-exploiting protocols, IEEE VTC 04, vol. 1, pp , May [7] H. Jiang and P. A. Wilford, A hierarchical modulation for upgrading digital broadcast systems, IEEE Trans. Broadcasting, vol. 51, no. 2, pp , June [8] ETSI, EN , v1.1.1, Digital video broadcasting DVB : Interaction channel for digital terrestrial television RCT incorporating multiple access OFDM, March [9] G. Caire and D. Tuninetti, The throughput of hybrid-arq protocols for the Gaussian collision channel, IEEE Trans. Inform. Theory, vol. 47, no. 5, pp , July [10] A. Steiner and S. Shamai, Multi-layer broadcasting hybrid-arq strategies for block fading channels, IEEE Transactions on wireless communications, vol. 7, no. 7, pp , July 2008.