Sparse-aware least sum of exponentials algorithms for sparse system identification

Abstract

Zero attraction techniques are introduced into the least sum of exponentials algorithm to estimate the sparse systems. The l1-norm and its reweighting form are employed and integrated into the least sum of exponentials algorithm to create desired zero attraction terms. The proposed algorithms are mathematically presented in the context of the adaptive filtering frame. The proposed algorithm is investigated for estimating a sparse system. The obtained results illustrate that the proposed sparse least sum of exponentials algorithms give better estimation performance than the classical least sum of exponentials algorithms when the system is exact sparse.