Fast Sparse Adaptive Filtering Algorithms for Acoustic Echo Cancellation

In this communication, fast adaptive algorithms are suggested to enhance the performance of the Fast-Normalized Least Mean Square (FNLMS) algorithm in Acoustic Echo Cancellation (AEC) applications with a sparse system. We propose two new algorithms, the first one is the Zero-Attracting (ZA) FNLMS which gives a better performance when the unknown system is extremely sparse. However, by decreasing the sparsity of the system, the Mean Square Error (MSE) got significantly worse than that of the FNLMS algorithm. To overcome this issue, another algorithm named Reweighted Zero-Attracting FNLMS (RZA-FNLMS) algorithm is proposed in this paper. Simulation results with stationary and non-stationary inputs under different Signal to Noise Ratio (SNR) values of additive noise and change in the impulse response lengths show an improvement in the convergence speed.