Adaptive frequency domain identification for ANC systems using non-stationary signals

The problem of identification of secondary path in active noise control applications is dealt with fundamentally using time-domain adaptive filters. The use of adaptive frequency domain subband identification as an alternative has some significant advantages which are overlooked in such applications. In this paper two different delayless subband adaptive algorithms for identification of an unknown secondary path in an ANC framework are utilized and compared. Despite of reduced computational complexity and increase convergence rate this approach allows us to use non-stationary audio signals as the excitation input to avoid injection of annoying white noise. For this purpose two non-stationary music and speech signals are used for identification. The performances of the algorithms are measured in terms of minimum mean square error and convergence speed. The results are also compared to a fullband algorithm for the same scenario. The proposed delayless algorithms have a closed loop structure with DFT filterbanks as the analysis filter. To eliminate the delay in the signal path two different weights transformation schemes are compared.