The sliding frequency-domain adaptive filter algorithm amenable to parallel implementation

The adaptive filter has applications in noise reduction, echo cancellation, and equalization. In this paper, we present an implementation of the sliding frequency-domain LMS (SFDLMS) adaptive filter used as a noise canceller. A compelling interest in this algorithm is the capability to implement it in parallel form. Comparisons are made with the recursive least squares (RLS) and frequency-domain/block LMS methods for speed of convergence, number of computations, implementation complexity, MSE performance, and the uniformity of convergence across frequencies. The comparisons are made for the recovery of speech signals contaminated by additive loud audio backgrounds in nonstationary acoustic environments. Experimental results are presented for filter sizes in excess of 8000 taps.