Channel Optimized Sample Adaptive Produ t Quantization

Channel optimized ve tor quantization (COVQ), as a joint sour ehannel oding s heme, has proven to perform well in ompressing a sour e and making the resulting quantizer robust to hannel noise. Unfortunately like its ounterpart in the noiseless hannel ase, the ve tor quantizer (VQ), the COVQ en oding omplexity is inherently high. Sample adaptive produ t quantization was re ently introdu ed by Kim and Shro to redu e the omplexity of the VQ while a hieving omparable distortions, even for moderate quantization dimensions. In this paper, we investigate the SAPQ for the ase of noisy hannels and employ the joint sour ehannel approa h of optimizing the quantizer design by taking into a ount both sour e and hannel statisti s. It is shown that, like its ounterpart in the noiseless ase, the hannel optimized SAPQ a hieves omparable performan e results to the COVQ (within 0.2-1.0 dB), while maintaining onsiderably lower en oding omplexity (half of that of COVQ) and storage requirements.