Аннотация
This paper investigates an efficient method for acoustic feedback cancellation (AFC) in digital hearing aids. The conventional approach based on the normalized least mean square (NLMS) algorithm suffers from biased and slow convergence. The prediction error method (PEM)-based AFC (PEM-AFC) significantly improves upon the biased convergence issue, however, the convergence speed may still be slow. The main structure of the proposed method is based on a pervious work and comprises two adaptive filters. The hearing aid receiver signal is used as an input to the first adaptive AFC filter, whereas the second AFC filter is excited by a gain-controlled (internally generated) probe signal. The key proposal is to use a lattice filtering-based pre-processing in the update equation of the first AFC filter. This results in a Newton-like adaptive algorithm with an improved convergence speed. Furthermore, the lattice filtering is performed on sample-by-sample basis in contrast to the frame-based processing in the PEM-AFC. The level of the probe signal exciting the second AFC filter is reduced as AFC system converges. This makes the second adaptive filter being sluggish particulary at the steady-state. In order to improve the convergence speed, a maximum Versoria-criterion (MVC)-based gradient information is incorporated into the update algorithm of the second AFC filter. As in the previous method, the coefficients between the two adaptive filters are mutually transferred to ensure that both adaptive filters estimate the acoustic feedback path. Simulations demonstrate the effective performance of the proposed approach for a variety of speech signals with different gain settings of hearing aids.