Cost-Effective Acoustic Feedback Cancellers for Digital Hearing Aids

To mitigate the bias caused by the strong correlation between the loudspeaker and source signals in the acoustic feedback (AF) canceller (AFC), the prediction error method (PEM) based AFC (PEMAFC) is the most popular solution but comes with two inherent drawbacks. Firstly, to improve the tracking performance of the PEM filter in the processing of nonstationary signals such as speech and music, the recursive least square (RLS) algorithm is a strong candidate, but this leads to unmanageable computational complexity in the PEMAFC system. Secondly, though the decorrelated least mean square (DLMS) algorithm is often preferred as an update algorithm for estimating the AF path due to its simple structure, it still suffers from inferior steady-state performance. By drawing inspiration from the online censoring (OC) strategy that has significantly contributed to the adaptive processing of Big Data streams, we develop two PEMAFC systems in this paper, called OC-PEMAFC-1 and OC-PEMAFC-2. The OC-PEMAFC-1 employs the OC based RLS (OC-RLS) algorithm to update the weight vector of the PEM filter and shows a comparable performance by considerably reducing the overall computational complexity of the PEMAFC arising from the classical RLS. On the contrary, the OC-PEMAFC-2 incorporates the OC based DLMS (OC-DLMS) algorithm using only informative data streams, thus substantially enhancing the steady-state performance and partially reducing the overall computational complexity of the PEMAFC. This study also provides the stability analyses of the OC-RLS and OC-DLMS used in the proposed PEMAFC systems. The mentioned superiorities of the proposed systems are confirmed through simulation results on real-world AF paths.