Regularized LAD algorithms for sparse time-varying system identification with outliers

Abstract

Two regularized least mean absolute deviation (LAD) algorithms are proposed for sparse system identification, which are referred to as zero-attracting LAD (ZA-LAD) and re-weighted zero-attracting LAD (RZA-LAD), respectively. The LAD type algorithms are robust to the impulsive noises. Furthermore, ℓ1-norm penalty is imposed on the filter coefficients to exploit sparsity of the system. The performance of ZA-LAD type algorithms is evaluated for linear time varying system identification under impulsive noise environments through computer simulations.