A Novel Hybrid ARQ Scheme Using Packet Coding

Size: px

Start display at page:

Download "A Novel Hybrid ARQ Scheme Using Packet Coding"

Adele Hubbard
6 years ago
Views:

1 27-28 January 26, Sophia Antipolis France A Novel Hybrid ARQ Scheme Using Pacet Coding LiGuang Li (ZTE Corperation), Jun Xu (ZTE Corperation), Can Duan (ZTE Corperation), Jin Xu (ZTE Corperation), Xiaomei Xu (ZTE Corperation), Xiao Yan (ZTE Corperation) Abstract The 5th generation (5G) mobile communication system requires high throughput for high quality user experience. In this paper, a novel Hybrid Automatic Repeat Request (HARQ) scheme based on pacet coding is proposed. The idea is that a parity bloc is transmitted when code bloc error rate (CBER) of all turbo code blocs of transmission pacet is less than or equal to the threshold of P and greater than zero. Fewer resources are occupied when parity bloc is retransmitted, which can improve the throughput of system. The parity bloc is generated by applying XOR operation on all turbo code blocs in the pacet. Besides, an iterative decoding algorithm for pacet coding is presented in detail. Simulation results show that the proposed HARQ scheme yields higher throughput than traditional ones. Keywords HARQ, Pacet Coding, Single Parity Code, Iterative Decoding, I. INTRODUCTION Hybrid Automatic Repeat Request (HARQ) is commonly used in wireless systems to improve lin and system throughput [~5]. In the process of HARQ, if a transmission or retransmission pacet is not received correctly, a request will be sent to transmitter for a retransmission. The HARQ schemes based on Chase Combining (HARQ-CC) [3] and Incremental Redundancy (HARQ-IR) [6] are widely used to enhance performance in communication systems. As an important process Medium Access Control (MAC) layer in 4G wireless system [7], HARQ is very liely to be a component of 5G wireless communication system. Since the HARQ scheme may have low efficiency for each retransmission length equal to first transmission, the multilevel ACK/NACK (or soft-ack/nack) was applied to improve the throughput of system [8~]. A few more states are fed bac to transmitter, which can bring more information of channel besides the correctness of pacet. Using information of channel and the reliability information of decoder output, adaptation of retransmission size and modulation can be conducted [8]. However, the partial data of correct code blocs may be retransmitted for multilevel ACK/NACK, which will reduce the retransmission efficiency. In this paper, a flexible soft HARQ scheme based on pacet coding is presented. The new HARQ system has two inds of woring modes. ) When the code bloc error rate (CBER) is low, a parity bloc is transmitted to assist the decoding of incorrect blocs to improve efficiency. 2) When code bloc error rate (CBER) is high, conventional soft HARQ is utilized to control the retransmission. A parity bloc is generated by applying a simple XOR operation on all code blocs in pacet. The iterative decoding algorithm of pacet coding is also presented. Moreover, simulation is conducted to chec the performance. Compared with the traditional HARQ scheme, proposed method has better performance. The rest of this paper is organized as follows. In section 2, the system models of new/traditional HARQ are presented. Section 3 gives a brief introduction to iterative decoding of pacet coding. Section 4 provides the simulation results and comparisons. Finally, the conclusion is drawn in section 5. II. SYSTEM MODEL In the paper, the HARQ scheme in point-to-point transmission is considered. There are two assumptions for analysis as follows: ). each pacet transmitted in the forward direction is matched with a feedbac sign that acnowledges correct reception (Acnowledgement, ACK) or requests a retransmission (Negative Acnowledgement, NACK). 2). Each feedbac information is decoded without error. A. Transmitter The 3GPP LTE turbo code is considered in this paper. A turbo code bloc consists of the systematic bit stream and two parity bit streams interleaved by three different sub-bloc interleavers. A virtual circular buffer is constructed by placing the interleaved systematic bits first and interlacing the two interleaved parity streams bit-by-bit. Four fixed start positions are defined by four redundancy version indexes which are distributed evenly in the virtual circular buffer. As shown in Fig. for traditional HARQ transmitter, the circular buffer consists of limited number of pacets which have been transmitted. Each pacet includes several turbo code blocs. If a pacet is mared incorrect by a feedbac, the transmitter will retransmit it with corresponding redundancy version for each turbo code blocs. In the first transmission pacet, N bits are selected beginning from the first fixed start positions for each turbo code bloc. And, in the first (or 2 nd or 3 rd ) retransmission, N bits are selected beginning from the first (or 2 nd or 3 rd ) fixed start positions. Fig.. Traditional HARQ Operation Diagram of Transmitter

2 27-28 January 26, Sophia Antipolis France A transmitter of new HARQ scheme based upon pacet coding is presented in fig 2. The circular buffers in Fig. 2 for each transmission pacet concludes 2 buffers: buffer with all mother turbo code blocs and buffer 2 with parity bloc. During the new HARQ process, the first transmission has the same pacet with traditional HARQ process. According to the feedbac, the traditional retransmission pacet (in buffer ) or parity bloc (in buffer 2) or nothing (correctly received) is sent. CBER Nerr N () turbo Where N is the amount of all false turbo code blocs in a err pacet and N is the amount of all turbo code blocs in a turbo pacet. If CBER of current pacet is equal to zero, ACK is fed bac to transmitter. When CBER is less than or equal to the threshold of P and greater than zero, NACK2 is returned to transmitter. And, if CBER is greater than the threshold of P, NACK is returned. Fig. 2. New HARQ Operation Diagram of Transmitter The soft ACK/NACK is defined as more than 2 states fed bac to the transmitter, in which one of the states must mars that the pacet is received correctly. In the paper, the soft- ACK/NACK has three states as follow. ACK: the current pacet is correctly decoded, and a new initial pacet will be sent. NACK: the current pacet is decoded incorrectly, and a retransmission pacet will be sent which has the same length and redundancy version as traditional one does. NACK2: the current pacet is decoded incorrectly, and a parity bloc will be sent which has the different lengths for different retransmissions. For the transmitter, if ACK is received, a new initial pacet will be sent. If NACK is returned, a retransmission pacet will be sent which has the same length and redundancy version as traditional one does (incremental redundancy for each turbo code bloc). Else if NACK2 is received, a parity bloc will be sent. The retransmission of the parity bloc has 3 types of lengths. ) st retransmission: the length is equal to that of each turbo code bloc in first transmission. 2) 2nd retransmission: the length is equal to 2 times of that of each turbo code bloc in first transmission. 3) 3rd retransmission: the length is equal to 3 times of that of each turbo code bloc in first transmission. Why is the length of retransmission parity bloc different for each retransmission? That is because the turbo code blocs have been transmitted for copy (or 2 copies or 3 copies) at st (or 2 nd or 3 rd ) retransmission. Therefore, more bits of parity bloc are retransmitted for more times retransmission. B. Receiver Since parity bloc is not transmitted for first transmission pacet, each turbo code bloc is decoded respectively. If the parity bloc is transmitted for retransmissions, the iterative decoder will be applied at receiver, which is introduced in later chapter. The correctness of each turbo bloc can be learned from Cyclic Redundancy Chec (CRC) attached behind the information bits. And, the code bloc error rate (CBER) in a pacet is given by C. Pacet Coding In pacet coding, a single parity code (SPC) is used to interrelate all the turbo code blocs in a pacet, as shown in Fig. 3. Assuming that all the mother turbo code blocs are expressed as: c, c,, c ( K ), each has been interleaved in circular buffer. The mother parity bloc c K is generated by XOR operation on all mother turbo code blocs, which is defined as ck c c c, as shown in Fig. 3. ( K ) Fig. 3. Pacet Encoding with SPC c c c 2 c( K ) D. Normalized The normalized throughput is defined below for HARQ system. The parameter of average repeat times N average for initial pacets is expressed as Naverage NTot N, where, N is Ini Tot the amount of total transmissions, and N Ini is the amount of initial pacets. And, the normalized throughput of HARQ system is calculated as thoughput ( PER) N (2) average where, PER is the initial pacet error rate. In the new HARQ scheme, since the parity bloc may be retransmitted which occupies less resources than traditional retransmission. Therefore, during the simulations, if a parity bloc is retransmitted, the amount of total transmissions is accumulated as N Tot N Tot N Parity N. Otherwise, the amount of total transmissions is accumulated as NTot NTot. In the paper, we assume that the traditional retransmission has the same length with initial transmission pacet. The parameter of N is the length of parity bloc, and N is the length of Parity each traditional transmission pacet. c K 2

3 27-28 January 26, Sophia Antipolis France III. ITERATIVE DECODING OF PACKET CODING If a parity bloc is retransmitted, the received parity bloc and incorrect pacet can be composed into a pacet code. The parity bloc is a turbo code bloc since it belongs to turbo code space. Therefore, the iterative decoding algorithm between turbo decoding and soft SPC decoding can be applied for pacet code decoder. The log-lielihood ratios (LLRs) of pacet s bits are generated via the algorithm of IR & CC for first/second/third retransmission. Iterative decoding steps are shown as follow. ) LLRs and hard-decision bits of each turbo code bloc are obtained by turbo decoder. 2) If all the turbo code blocs are correctly decoded, iterative decoding is finished. Otherwise, every turbo code bloc is mared by true when it is correct or false when it is incorrect. 3) The LLRs of false turbo blocs are updated by soft SPC decoding. Then, we go bac to step. A maximum number of iterations can be set to avoid endless iteration. Starting from second iteration of the iterative decoding described above, only the false turbo code blocs are decoded. Let the indexes set of correct code blocs be ψ, and error code blocs be ψ 2. During the process of the iterative decoding, hard-decision bits are directly used for those correct code blocs. And, extrinsic information from other code blocs is provided to enhance the LLRs of each turbo bloc. The updated LLRs for the -th false code bloc can be calculated by the expression (soft SPC decoding) as: ( ' ) ' ( ) n n LLR LLR ( ) min( LLR m ) (3) m φ Where, LLR is the -th error code bloc s LLRs before updated, LLR ' is the -th false code bloc s LLRs after updated, n is a vector whose elements respectively denote the amount of bit in each bit locations of code blocs in set ψ, φ is a set includes all elements of ψ except the -th 2 false code bloc, n ' is a vector whose elements respectively denote the amount of positive LLRs in each bit locations of code blocs in set φ. The condition of parity bloc retransmission is that the CBER of pacet less than or equal to threshold of P. Therefore, if parity bloc is retransmitted, it is nown that just some of turbo code blocs are decoded with error. And, just the false turbo code blocs are decoded in the iterative decoder. Besides, it is found that only 2 or 3 iterations are enough for successful decoding. Furthermore, from the 2 nd iteration, the false decoded blocs have been greatly reduced. In the conventional HARQ system, it may need to decode all the turbo code blocs again if requesting a retransmission. Therefore, the proposed HARQ method has negligible extra or less complexity and latency compared with conventional ones. IV. SIMULATION RESULTS This section describes the optimization of the threshold P and performance comparison between the new HARQ scheme and the traditional one in AWGN channel. The logarithmic domain maximum a posteriori (Log-MAP) algorithm [3] is used for turbo decoder. Iterative decoding algorithm described above is applied for pacet decoder. A. Optimizing of Threshold P According to the parameter of P defined above, it may have an appropriate value with intuitively sense. If the value of P is too big, the pacet may be decoded incorrectly for retransmission of parity bloc. Or if it is too small, lower throughput is obtained since a traditional pacet is retransmitted with big probability. The simulation parameters are shown in Table. Table. The Simulation Parameters Case Case 2 Code Rate /3 3/4 Modulation QPSK 6QAM Systematic Bits Lengths 432/26 432/26 Amount of Turbo Blocs The simulation results are shown in Fig. 4. It is found that the performance is best when the value of P equals to.9 for first retransmission. As shown in fig. 4(a) for code rate of 3/4, the values of P for different retransmissions are set as [.9.7] with the best performance. Where, the value of P equals.9 (or 9 incorrect turbo blocs in blocs of a pacet) for the st retransmission, while equals (or 8 incorrect turbo blocs) and.7 (or 7 incorrect turbo blocs) for the 2 nd retransmission and the 3 rd retransmission respectively. The reason for the value of P decreasing gradually is that there is a few performance loss at 2 nd and 3 rd retransmission. From right to left in the Fig. 4, four flat areas are corresponding to four transmissions (first transmission and three retransmissions), respectively QAM,Rate=3/4, Turbo Blocs New 26[.7.6.5] New 26[.9.7] New 432 [.7.6.5] New 432 [.9.7] (a) 3

4 27-28 January 26, Sophia Antipolis France.6 QPSK,Rate=/3, Turbo Blocs.4 New 26[.7.6.5].2 New 26[.9.7] New 432 [.7.6.5] New 432 [.9.7] (b) Fig. 4. Normalized vs. SNR for Rate=3/4 and Rate=/3 We also find that the best performance can be obtained for code rate of /3 when P equal to [.9.7], as shown in Fig. 4(b). During the simulations, it is found that the pacet with most (about 9%) of the turbo code blocs incorrect can be decoded correctly with iterative decoding if a parity bloc is retransmitted. In later chapter, the values of P are set to [.9.7] for simulations. B. Simulation Results The performances of the proposed HARQ scheme are presented and compared with the traditional ones in AWGN channel. The simulation parameters are described in Table 2 for the new or traditional HARQ scheme. Three code rates are considered for comparison and each has three different turbo systematic lengths. Since low code rate is generally used with low-order modulation and high code rate with high-order modulation, the parameters of simulations are set with this rule. Table 2. The Parameters of Comparison Simulations Case Case 2 Case 3 Code Rate /3 /2 3/4 Modulation QPSK 6QAM 64QAM Systematic Bits Length 432/26/ /26/ /26/644 Amount of Turbo Blocs According to simulation results in Fig. 5, we can find that the proposed HARQ scheme has better performance than the traditional one for the first/second/third retransmission. In Fig. 5(a) for rate of /3 and modulation of QPSK, the performance gains of db,.5db and.3db are obtained for turbo code length of 432, 26 and 644 respectively at the first retransmission. And, the performance gains are.7 db,.4 db and.2 db at second retransmission. As seen in Fig. 5(b) and Fig. 5(c), the performance gains are also achieved for other code rates and modulations QPSK,Rate=/3, Turbo Blocs New HARQ 644 New HARQ 26 Trad HARQ 644 Trad HARQ (a) 6QAM,Rate=/2, Turbo Blocs New HARQ New HARQ 26 Trad HARQ 644 Trad HARQ (b) 64QAM,Rate=3/4, Turbo Blocs New HARQ 644 New HARQ 26 Trad HARQ 644 Trad HARQ (c) Fig. 5. Performances Comparisons with Different Rates and Modulations V. CONCLUSIONS In this paper, a new scheme of HARQ system based on pacet coding is presented in detail. The encoding and iterative decoding are introduced for pacet coding. Moreover, optimization simulations are also conducted for the threshold P. Simulation results show that the proposed schemes have 4

5 27-28 January 26, Sophia Antipolis France higher throughput compared with the traditional ones. Therefore, proposed HARQ scheme is very suitable for the future wireless communication system, such as 5G. We will analyze the realistic system-level simulation for the new HARQ scheme in different channel scenarios in future. ACKNOWLEDGMENT This wor was supported partially by the National major projects of China under grant 2ZX33-3-2, networ coding technology in IMT-A cooperative relay system. REFERENCES [] J. F. Chengi, Coding performance of hybrid ARQ schemes, Communications, IEEE Transactions on, vol. 54, no. 6, pp. 7-29, 26. [2] S. Lin, and D. J. Jr. Costello, Error control coding: Fundamentals and applications, Englewood Cliffs, Prentice- Hall,983. [3] D. Chase, Code combining-a maximum-lielihood decoding approach for combining an arbitrary number of noisy pacets, IEEE Trans. Commun., vol. COM-33, pp , 985. [4] S. Sesia, I. Toufi and M. Baer, LTE: the UMTS long term evolution. Wiley Online Library, 29. [5] A. Ghosh, J. Zhang, J. G. Andrews and R. Muhamed, Fundamentals of LTE, Prentice-Hall, 2. [6] S. Lin and P. Yu, A hybrid ARQ scheme with parity retransmission for error control of satellite channels. Communications, IEEE Transactions on, vol. 3, no. 7, pp. 7-79, 982. [7] 3GPP TS 36.3 V9.2.-2, Evolved Universal Terrestrial Radio Access(E-UTRA) and Evolved Universal Terrestrial Radio Access Networ(E-UTRAN), Overall description, Protocol specification [S], 2. [8] M. Woltering, D. Wubben, A. Deorsy, V. Braun, U. Doetsch, Lin Level Performance Assessment of Reliability-Based HARQ Schemes in LTE, VTC Spring, 24 IEEE 79th, pp. - 5, 8-2, 24. [9] W. Younghoon and H. Seung-Hoon, Multilevel Control Signaling for Hybrid ARQ, Wireless Communications, Networing and Mobile Computing. WiCOM 8. 4th International Conference on, pp. -3, 28. [] M. Woltering, D. W ubben, A. Deorsy, V. Braun and U. Doetsch, Performance of HARQ with Reduced Size Retransmissions using Networ Coding Principles, VTC23- Spring, Dresden, Germany, 23. [] A. Chelli and M. Alouini, Performance of hybrid-arq with incremental redundancy over relay channels, Proc. GLOBECOM, pp. 6-2, 22. [2] B. Mai and T. Erisson, On the performance of MIMO-ARQ systems with channel state information at the receiver, IEEE Trans. Commun., vol. 62, no. 5, pp , 24. [3] H. Claussen, H. R. Karimi and B. Mulgrew, Improved Max- LogMAP turbo decoding by maximization of mutual information transfer, EURASIP Journal on Applied Signal Processing, pp , 25. 5

II. FRAME STRUCTURE In this section, we present the downlink frame structure of 3GPP LTE and WiMAX standards. Here, we consider

II. FRAME STRUCTURE In this section, we present the downlink frame structure of 3GPP LTE and WiMAX standards. Here, we consider Forward Error Correction Decoding for WiMAX and 3GPP LTE Modems Seok-Jun Lee, Manish Goel, Yuming Zhu, Jing-Fei Ren, and Yang Sun DSPS R&D Center, Texas Instruments ECE Depart., Rice University {seokjun,