| Channel optimized vector quantization (COVQ), as a joint source-channel coding scheme, has proven to perform well in compressing a source and making the resulting quantizer codebook robust to channel noise. Unfortunately like its counterpart in the noiseless channel case, the vector quantizer (VQ), the COVQ encoding complexity is inherently high. Sample adaptive product quantization was recently introduced by Kim and Shroff to reduce the complexity of the VQ while achieving comparable distortions, even for moderate quantization dimensions. In this thesis, we investigate the SAPQ for the case of noisy memoryless channels and employ the joint source-channel approach of optimizing the quantizer design by taking into account both source and channel statistics. It is shown that, like its counterpart in the noiseless case, the channel optimized SAPQ achieves comparable performance results to the COVQ (within 0.2–0.8 dB), while maintaining considerably lower encoding complexity (half of that of COVQ) and storage requirements. |
