Optimal quantization in non-orthogonal subband coders
R. Gandhi, Sushmita Mitra · 2003
It is well known that the mean square error (MSE) is not preserved under non-orthogonal transformations. This poses a significant challenge to quantize the subband signals in coders that involve non-orthogonal filter banks, since the nearest neighbor (NN) encoding rule can no longer be applied to quantize the coefficients. In this paper, we develop techniques for optimal quantization of the subband coefficients in non-orthogonal subband coders. An exhaustive-search based quantization approach is proposed. The complexity of this approach is shown to increase exponentially with the length of the signal. Next, a reduced complexity solution in the form of a trellis-based search is proposed. Simulation results indicate appreciable SNR gains over standard coding techniques for non-orthogonal filter banks.