Group sparse LMS for multiple system identification

Armed with structures, group sparsity can be exploited to extraordinarily improve the performance of adaptive estimation. In this paper, a group sparse regularized least-mean-square (LMS) algorithm is proposed to cope with the identification problems for multiple/multi-channel systems. In particular, the coefficients of impulse response function for each system are assumed to be sparse. Then, the dependencies between multiple systems are considered, where the coefficients of impulse responses of each system share the same pattern. An iterative online algorithm is proposed via proximal splitting method. At the end, simulations are carried out to verify the superiority of our proposed algorithm to the state-of-the-art algorithms.