Quantizer design for non-orthogonal subband coders

Our prior results have shown that the optimal quantizer scheme in a non-orthogonal subband coder involves a trellis-based search of the code vectors at the encoder and filtering the quantized subband coefficients with the synthesis filter bank at the decoder. The performance of the optimal quantization scheme can be improved by using code vectors that are designed by taking into account the optimal trellis-based encoding rule. In this paper, we describe an iterative technique for designing the code vectors when the trellis-based encoding rule is used to quantize the subband signals in a non-orthogonal subband coder. For a given training sequence, the design procedure updates the code vectors iteratively, until a local minima is reached. Simulation results indicate that subband coders that use the optimized code vectors yield a higher signal-to-noise (SNR) than those that use the sub-optimal code vectors.