Robust Adaptive Inverse Control Based on Maximum Correntropy Criterion

Adaptive inverse control, proposed by Bernard Widrow, is mainly based on the well-known least mean square (LMS) algorithm. The LMS is a stochastic gradient algorithm under the minimum mean square error (MSE) criterion, which performs well for linear and Gaussian systems. However, its performance will become poor when signals are non-Gaussian, especially when systems are disturbed by impulsive noises. In this work, in order to improve the robustness of the adaptive inverse control against impulsive noises, we propose a new adaptive inverse control method, which is based on the recently developed maximum correntropy criterion (MCC) algorithm. The MCC algorithm aims at maximizing the correntropy between the model output and the desired response. Since correntropy is a nonlinear similarity measure that contains higher-order statistics of the signals and is insensitive to large outliers, the proposed method can achieve desirable performance in impulsive noise environments. Theoretical results on optimal solution and convergence are derived. Simulation results are also presented to demonstrate the superior performance of the new method.