A Variable Step-Size l0-PRLS Algorithm and its Application in Sparse Channel Estimations

We recently proposed a sparsity-aware recursive least squares (RLS) adaptive filtering algorithm, named the proportionate RLS with l
0
norm regularization (l
0
-PRLS). It shows better performance than its non-sparse counterparts under sparse systems. The l
0
-PRLS algorithm, however, employs a fixed step size which trades off the convergence speed and steady-state performance. To simultaneously obtain fast convergence and low steady-state error, we resort to the variable step size (VSS) technique and propose a VSS-l
0
-PRLS adaptive filtering algorithm in this paper. The superiority of the VSS-l
0
-PRLS was verified by simulation results of sparse system identification as well as experimental results of channel estimation for underwater acoustic (UWA) communications.