NEAR-END DISTORTION IN OVER-SAMPLED SUBBAND ADAPTIVE IMPLEMENTATION OF AFFINE PROJECTION ALGORITHM (TuePmOR5)
Author(s) :
Hamid Sheikhzadeh (Dspfactory Ltd., Canada)
Robert L. Brennan (Dspfactory Ltd., Canada)
Kevin R. L. Whyte (Dspfactory Ltd., Canada)
Abstract : The affine projection algorithm has been employed for echo cancellation using subband adaptive filters implemented with over-sampled filterbanks. We show that the near-end signal may be distorted or partially cancelled if proper regu-larization is not applied. We attribute these distortions to the nonlinear effects that appear mostly when narrow-band sig-nals are involved in adaptive filtering. Supporting this as-sumption, it is shown that as the over-sampling rate in-creases and subband signals become more bandlimited, near-end cancellation also increases. The effect of adaptive filter length and the regularization parameter on near-end distortion is analyzed, leading to the result that near-end distortion can be avoided by proper regularization. As a re-sult, a practical low-cost on-line regularization method is proposed for the affine projection algorithm, effectively eliminating near-end distortion.

Menu