A robust quasi-Newton adaptive filtering algorithm for impulse noise suppression

This paper studies the problem of robust adaptive filtering in impulse noise environment using the Quasi-Newton (QN) adaptive filtering algorithm. An M-estimate based cost function is minimized instead of the commonly used mean square error (MSE) to suppress the adverse effect of the impulse noise on the filter coefficients. In particular, a new robust quasi-Newton (R-QN) algorithm using the self-scaling variable metric (SSV) method for unconstrained optimization is studied in detail. Simulation results show that the R-QN algorithm is more robust to impulse noise in the desired signal than the RLS algorithm and the other QN algorithm considered. Its initial convergence speed and tracking ability to sudden system change are also superior to those of the quasi-Newton algorithm proposed by De Campos and Antoniou (1997).