Minimum MSE Filter Banks Design Using Correlated Signal Dependent Quantization Noise Model

Abstract

A theory is proposed for how to design minimum mean square error (MSE) filter banks for lossy source coding when uniform quantizers and ideal entropy coding are used to compress the subband signals. A signal-dependent correlation model is used to find the auto and cross correlation between the additive quantization noise within the same subband and across different subbands. This theory also shows how to find the cross correlation between the original time-series and the additive quantization noise. The simulation results show that the proposed way of designing the filter banks outperforms the minimum MSE performance results found by using the classical signal independent high-rate quantization noise model.