On the Use of Delayless Adaptive Filter Architecture for Feedback Control in Hearing Aids

Traditionally, time domain adaptive filters have been used in many different configurations and systems. However, for systems requiring long adaptive filters, the computational complexity becomes a problem in practical applications. Hence, frequency domain adaptive filters have been introduced to significantly reduce the computational complexity, in addition to also allow frequency dependent adaptation control. However, the traditional frequency domain adaptive filter approach introduces an additional delay in the signal path due to the analysis and synthesis processing, which is unacceptable in some applications. Thus, a new class of so-called delayless adaptive filters have been proposed and improved over the years. In this work, we reveal that the traditional delayless subband adaptive filter architecture has an undesired convergence behavior which depends on the unknown system to be identified. We show that for a closed-loop hearing aid feedback cancellation system, this undesired behavior significantly slows down the convergence of the system, and a redesigned delayless architecture without that undesired dependency improves the convergence rate by approximately 30%.