Proportionate subband filtering technique with $l_{1}$-norm for feedback cancellation in hearing aids

摘要

Adaptive feedback cancellers (AFCs) based on normalized subband adaptive filtering (NSAF) exploiting the proportionate adaptation technique (PNSAF) have been recently reported for alleviating the feedback effect. The PNSAF algorithm exhibits faster convergence at the initial stage of adaptation whereas stagnating the convergence at later phase of the estimation process. An attempt has been made in this work to maintain the AFC convergence and misalignment in such conditions by introducing a penalty element in the cost function of PNSAF algorithm obtained from l₁-norm of the coefficients. This results in inclusion of the zero attracting term during weight update mechanism and hence the proposed algorithm is termed as zero attracting PNSAF (ZA-PNSAF). Furthermore, prediction error method is utilized to reduce the bias related issues encountered in adaptive feedback cancellation for hearing aids. The derivations and convergence analysis of the proposed algorithm have been carried out. Simulation results demonstrate the efficacy of the proposed feedback cancellation method as compared to existing techniques using speech segments as input signal. The proposed ZA-PNSAF based AFC enables 3.5 dB more lowering of misalignment value in comparison to other algorithms while maintaining faster convergence.