| A channel coding system can be converted to a lossy compression system by using the channel decoder as a lossy compressor and the channel encoder as a lossy decompressor. Perhaps, the most well known example of a lossy-compression technique based on this paradigm is trellis-coded quantization. In this thesis we identify conditions under which plugging in a channel decoder (encoder) as a lossy compressor (decompressor resp.) is feasible and investigate whether (asymptotic) optimality is preserved under such an arrangement. The foregoing results are also extended to the channel coding and lossy compression problems where side information is available only at the encoder and only at the decompressor respectively (popularly known as the Gelfand-Pinsker channel coding problem and Wyner-Ziv lossy compression problem respectively). We further construct lossy-compressors based on sparse-graph channel-codes that achieve the optimal rate-distortion performance for any discrete-memoryless source, and also show empirical performance very close to the optimal for moderate blocklengths. Finally, we characterize the performance vs. computational-complexity tradeoff for lossy compression by utilizing a divide-and-conquer paradigm that was first proposed by Forney to construct polynomial-complexity capacity achieving channel-coding schemes. |
