Bit-level error detection of turbo codes and its application to stopping of iterative decoding and hybrid ARQ

Abstract

In this dissertation, we consider a bit-level error detection method of turbo codes and its applications to stopping of iterative decoding and a hybrid automatic repeat request (HARQ) system. In turbo codes, most transceiver techniques such as decoding, early stopping, and retransmission schemes are performed frame by frame. However, the decoding reliability of each bit in a frame is not the same with the iterative decoding. Based on the decoding reliability, we propose adaptive correction or retransmission methods. First, we consider a bit-level stopping scheme of iterative decoding which prevents the unnecessary decoding delay and computational complexity of frame-level stopping scheme. We investigate the average number of iterations and bit error rate performance. The proposed bit-level stopping scheme requires smaller number of iterations than the frame-level scheme while achieving the same BER performance. Therefore, the decoding delay and computational complexity can be reduced. Second, we consider a parallel iterative decoding of turbo codes and its bit-level stopping scheme. We derive an optimum combining method of parallel decoding and propose a measure of decoding reliability based on the combined LLR. We find that the parallel decoding has improved error performance over the conventional serial decoding at the expense of increased computational complexity. The proposed stopping scheme based on the combined LLR can be a solution to a drawback of the parallel decoding while maintaining the improved coding gain over serial decoding. Third, we consider bit-level hybrid ARQ systems where the unreliable bits are detected and retransmitted. Since the proposed bit-level HARQ systems adaptively retransmit only the unreliable bits, throughput efficiency improves. Furthermore, proposed schemes prevent the reliably decoded bits from being replaced by erroneous symbols after the retransmission. Therefore the error performance also improves, especially in a low SN...