Channel optimized vector quantization with soft input decoding

In this paper, we present a new soft input decoding algorithm based on the Channel Optimized Vector Quantization (COVQ) system. In a COVQ communication system, hard decision decoding (HD) is used on the channel outputs to recover the reproduction indices of the code vectors from a designed codebook. We refer to this decoding method as COVQ_HD. Because hard decision decoding causes an irreversible loss of information, a soft decoding (SD) algorithm, which uses the soft channel outputs to reconstruct the source signals directly, can achieve better performance. We refer to this method as COVQ_SD. A first-order Gauss-Markov source and speech linear spectral pairs (LSPs) coding at 24-bits/frame are used to examine a novel COVQ_SD. The simulation results show that the performance of COVQ_SD is better than COVQ_HD under both the channel-matched and channel-mismatched conditions. Decoding complexity of COVQ_SD is also discussed and compared to other known soft input decoding algorithms.