Feedback cancellation in digital hearing aids using convex combination of proportionate adaptive algorithms

Abstract

The least mean square (LMS) based linear adaptive algorithms are commonly used for acoustic feedback cancellation (AFC) in digital hearing aids, given the simplicity, tracking capability, and robustness. Simultaneously, the LMS adaptive filter’s major limitation is slow convergence due to a compromised step-size selection. In recent times, the convex combination of LMS adaptive filter has depicted an improved trade-off between the convergence rate and steady-state while employed for adaptive feedback cancellation. However, there exists a scope for improving the convergence further for a time-varying feedback path under different noise conditions and input signals. The convex combination investigated with the mixing parameter δn is confined to the (0,1) interval. Here we observed that the convex combination of two adaptive filters estimates the undesired feedback path dependently. Therefore, the optimal convex combination coefficients occur as a sequence that mitigates the mean square error (MSE). The proposed adaptive feedback cancellation framework has been analysed statistically for convex optimization cost-function. Simulation and results present the improvements of the proposed framework in comparison to the existing state-of-the-art.