A fast convergence normalized least‐mean‐square type algorithm for adaptive filtering

Abstract

A new adaptive algorithm with fast convergence and low complexity is presented. By using the calculation structure of the dual Kalman variables of the fast transversal filter algorithm and a simple decorrelating technique for the input signal, we obtain an algorithm that exhibits faster convergence speed and enhanced tracking ability compared with the normalized least-mean-square algorithm with similar computational complexity.A. BENALLAL AND M. AREZKI CONCLUSIONIn this paper, a new FNLMS-type algorithm for adaptive filters is presented. The adaptation gain for the new algorithm is obtained by simplification of the FTF adaptation gain calculation. The amount of computation, 2N multiplications per sample, is the same for the proposed algorithm as for the NMLS algorithm. In all presented simulations, the proposed algorithm performs better than NLMS. It exhibits faster convergence and enhanced tracking ability compared with the NLMS algorithm.

Signed Convex Combination of Fast Convergence Algorithm to Generalized Sidelobe Canceller Beamformer for Multi-Channel Speech Enhancement

“…The convex combination of y (a) (n) and y (b) (n) is given by where, ̃( ) is dual Kalman gain [14], γN(n) is the Likelihood variable [14].…”

“…The computational complexity of the LMS [11], NLMS [13], FCNLMS [14], CLMS [16], and the proposed SCCFC algorithms are compared in this section. Here the length of the adaptive algorithm is given by N. For a regular LMS algorithm takes 2N+1 multiplications to update the filter.…”

“…It gives faster convergence when compared to LMS whereas computational cost is high and is too expensive for real-time noisy environments [12,13]. Fast convergence [14] algorithm has less computational complexity but gives less performance under various noisy conditions. And also when the positions of the source signal changes, the weight coefficient information used to update the adaptive filter will be lost, due to this, poor performance in the non-stationary environment combined adaptive filter [15][16][17] are designed, which give good convergence transition compared to the single adaptive filter.…”