MODEL-BASED FILTER BANKS AND CORRESPONDING SUBBAND CODING SYSTEMS - MULTIRATE STATE-SPACE APPROACH

A new model-based filter bank configuration is proposed in this work a nd its corresponding subband coding schemes are also discussed. The pr oposed filter bank system is developed specifically for the autoregres sive input signals with a consideration of subband noises. First, the analysis filters are designed on the basis of the conventional frequen cy splitting algorithms. Next, the synthesis filter is derived accordi ng to the composite effects of the AR input signal, the allpass analys is filters, and the white subband noises. State-space framework is use d to describe the analysis section. Hence, a new multirate Kalman synt hesis filter is derived to achieve optimal signal reconstruction. Comp uter simulations reveal that perfect reconstruction is achieved for th e clean subband case and optimal reconstruction is obtained when the c orrupting noises exist in the subband paths. In addition, the subband coding systems pertaining to the proposed model-based filter bank conf iguration have the advantage of better reconstructed speech quality an d possess a flexible subband number and down-/up-sampling factor.