Stability-controlled hybrid adaptive feedback cancellation scheme for hearing aids

Adaptive feedback cancellation (AFC) techniques are common in modern hearing aid devices (HADs) since these techniques have been successful in increasing the stable gain. Accordingly, there has been a significant effort to improve AFC technology, especially for open-fitting and in-ear HADs, for which howling is more prevalent due to the large acoustic coupling between the loudspeaker and the microphone. In this paper, the authors propose a hybrid AFC (H-AFC) scheme that is able to shorten the time it takes to recover from howling. The proposed H-AFC scheme consists of a switched combination adaptive filter, which is controlled by a soft-clipping-based stability detector to select either the standard normalized least mean squares (NLMS) algorithm or the prediction-error-method (PEM) NLMS algorithm to update the adaptive filter. The standard NLMS algorithm is used to obtain fast convergence, while the PEM-NLMS algorithm is used to provide a low bias solution. This stability-controlled adaptation is hence the means to improve performance in terms of both convergence rate as well as misalignment, while only slightly increasing computational complexity. The proposed H-AFC scheme has been evaluated for both speech and music signals, resulting in a significantly improved convergence and re-convergence rate, i.e., a shorter howling period, as well as a lower average misalignment and a larger added stable gain compared to using either the NLMS or the PEM-NLMS algorithm alone. An objective evaluation using the perceptual evaluation of speech quality and the perceptual evaluation of audio quality measures shows that the proposed H-AFC scheme provides very high-quality speech and music signals. This has also been verified through a subjective listening experiment with N = 15 normal-hearing subjects using a multi-stimulus test with hidden reference and anchor, showing that the proposed H-AFC scheme results in a better perceptual quality than the state-of-the-art PEM-NLMS algorithm.