Fast and efficient acoustic feedback cancellation based on low rank approximation

Abstract

In an adaptive feedback cancellation (AFC) scenario, it is essential for an algorithm to track and cancel the feedback signal as quickly as possible. We analyze typical feedback paths in hearing aids and show that they exhibit a low-rank nature. Further, to exploit this knowledge and improve the convergence and tracking performance for AFC, we propose the nearest Kronecker product decomposition based adaptive feedback canceller with prediction error method based signal pre-whitening. Detailed simulation study and comparison of computational complexity show that the proposed algorithm can provide improved convergence and tracking along with improved output speech quality over traditional AFC algorithms, at a moderate computational load.