Mixture Random Features Spaces for Robust Adaptive Filtering

Kernel adaptive filters (KAFs) based on random features spaces, e.g., random Fourier features space (RFFS) and Gaussian Hermite Fourier space (GHFS) have obtained significant success in signal processing. However, RFFS and GHFS face challenges of insufficient approximation accuracy and performance sensitivity to input dimensionality, respectively. To address these issues, a novel mixture random features spaces (MRFS) method is proposed based on multikernel method. Combining the maximum correntropy criterion (MCC) and stochastic gradient method, a robust mixture random features spaces maximum correntropy criterion (MRFSMCC) algorithm is thus obtained. Simulations on nonlinear system identification and Mackey-Glass time-series prediction validate the superiorities of proposed MRFSMCC under mixed impulsive environments.