A variable regularization parameter widely linear complex-valued NLMS algorithm: Performance analysis and wind prediction

Abstract

To overcome the conflict that the adaptive regularized complex-valued NLMS algorithms cannot have optimal performance when the regularization parameter is large or small, a widely linear complex-valued NLMS algorithm with the variable regularization parameter (VRP-WL-CNLMS) is proposed in this paper. The proposed algorithm can adaptively change the regularized parameter by exploiting a time-varying parameter that is obtained via making the power of noise-free a posteriori error minimum. A proper estimated method is provided to compute the power of the measured noise when the noise is unknown, and the moving-average method is employed to update the regularized parameter for avoiding large fluctuations. Then we provide the analysis of the transient and steady-state (TAS) behaviors of the proposed algorithm. Simulation results with different input signals illustrate that the VRP-WL-CNLMS algorithm has better advantages than other algorithms, and verify the theoretical validity of TAS analysis of the proposed algorithm in the system identification (SI) environment. Finally, the experimental results of wind prediction show that the predicted value of the proposed algorithm has a smaller error value with the original signal than the WL-CNLMS algorithm and can predict the signal better that can support the superiority of the proposed algorithm as well.