Adaptive combination proportionate filtering algorithm based on decorrelation for sparse system identification

The slow convergence rate of adaptive filters leads to the degradation of performance when input signals are heavily correlated. To solve this problem, improved proportionate normalized least-mean-square based on decorrelation (DIPNLMS) algorithm is proposed in this paper. Due to the principle of decorrelation, the proposed algorithm achieves a fast convergence rate. However, the fixed step-size DIPNLMS has a confliction between convergence rate and steady-state error. Thus, we apply an adaptive combination scheme to address this tradeoff, namely, adaptive combination of improved proportionate normalized least-mean-square based on decorrelation (CDIPNLMS) algorithm. Simulation results in the context of sparse system identification demonstrate that the proposed algorithms outperform the existing algorithms.