Codeword Decomposition: Successive Refinement Assisted Coding Scheme for Discrete Memoryless Sources

This paper proposes for memoryless discrete sources a new lossy coding scheme, codeword decomposition, which can achieve low encoding and decoding complexity and hence low latency, while maintaining high efficiency. The proposed scheme assumes assistance of successive refinement. This paper’s proposed codeword decomposition technique encodes length n sequence Xⁿ into two codewords at rate R1 and R2, respectively. A lossy version of Xⁿ can be reconstructed by independently decoding the codeword with rate R1, while Xⁿ can be losslessly recovered by jointly decoding the two codewords with R1 + R2 ≥ 1. The paper uses a hybrid majority voting (HMV) and bit flipping (BF) coding scheme, demonstrating reduced distortion in standalone transmission links versus puncturing. Moreover, the successive refinement link shows comparable performance to puncturing, even at elevated R1 values, highlighting the coding scheme’s resilience across diverse conditions.