Joint Encoding and Decoding Optimization of LT Codes over Noise Channels

The low-degree variable nodes and high residual errors of Belief Propagation (BP) decoding are key factors of high error floor of LT Codes over noise channels. A joint encoding and decoding scheme are proposed to lower this error floor. The Regularized Variable-Node (RVN) encoding method is adopted to maximize the minimum degree and avoid the lowdegree variable node. By investigating the characteristic of error patterns in error frames and the relationship between error bits and their Log-likelihood Ratio (LLR) values, the most unreliable bits are identified after BP decoding. Thanks to the Cyclic Redundancy Check (CRC) detector, the residual error bits of BP decoder are identified and corrected by bit flip. The simulation results demonstrate that the proposed method can dramatically reduce the frame error rate in the error floor region at the cost of almost negligible extra computation.