Abstract:

There are two ways to implement frequency domain filtered x LMS (FXLMS) algorithms. The first carries out both control signal generation and control filter updating in the frequency domain, and the second involves implementation of the control filtering in the time domain and updating of the coefficients of the control filters in the frequency domain. Most active noise control literature is focused on the second approach as the first one introduces a delay of at least one FFT block size for the control filter generation, which is usually not acceptable for active noise control. However, the second implementation has a limitation on its maximum computational complexity reduction due to its delayless requirement and it also needs quite a large on-chip memory for its FFT. The multidelay adaptive filter to be discussed in this paper is intended to solve some of the above problems. The multidelay adaptive filter has a flexible structure, which partitions a long filter into many shorter sub-filters so that a much smaller FFT size can be used to reduce the delay and memory requirement while maintaining the low computational complexity and faster convergence properties of the frequency domain algorithm.