Adaptive Feedback Cancellation for Hearing Aids Using the Prediction-Error Method with Orthonormal Basis Functions

Abstract

Hearing aids are susceptible to acoustic feedback, which limits the achievable amplification and may severely degrade the sound quality by producing howling artifacts. A potential method of feedback cancellation employs prediction of the feedback path (FP) with an adaptive filter. However, estimation of the FP suffers a large model error, known as the bias, due to correlation between the loudspeaker and source signals. A prediction-error method (PEM) based pre-whitening filter has been utilized to reduce the bias. However, this approach requires a large number of adaptive parameters, thus reducing the convergence rate and limiting the added stable gain (ASG). We introduce a PEM-based method derived based upon the orthonormal basis functions (OBFs) to estimate the FP with a small number of adaptive parameters and the source signal using an autoregressive model. The OBF filter is defined by a set of fixed poles and adaptive tap-output weights. The poles are estimated using an inherently stable least-squares method and embedded into the filter as a priori information. Experimental results show that the proposed method enhances the convergence rate and significantly increases the ASG compared to the standard PEM, while uses far fewer adaptive parameters.