Combined Source Adaptive and Channel Optimized Matrix Quantization algorithm

Matrix Quantization (MQ), a very promising source coding technique, has already been successfully applied for speech signals and noiseless channels. MQ is also shown in the literature to outperform Vector Quantization (VQ) when applied over noisy channels. Considering that most sources of practical interest are non-stationary, this paper introduces a technique which adapts MQ to varying source statistics and optimizes MQ for noisy channels, thus designs a matrix quantizer/decoder that considers both non-stationary source and noisy channel statistics. The resulting algorithm, Combined Source Adaptive and Channel Optimized Matrix Quantization (CSACOMQ) is evaluated for a source modelled as the non-stationary Wiener process and over the memoryless Binary Symmetric Channel (BSC). It is shown that CSACOMQ offers substantial Signal-to-Noise Ratio (SNR) performance improvement compared to the Channel Optimized Matrix Quantization (COMQ).