Diffusion Approximated Kernel Least Mean P-Power Algorithm

Considering nonlinear and non-Gaussian environments that are frequently encountered in real-world scenarios, we thus propose the approximated kernel least mean p-power (AKLM-P) algorithm to address these tough factors for nonlinear distributed systems. In order to efficiently devise this algorithm, we use the property of shift-invariant kernel function that can be approximated by random Fourier features, and consider the impulsive noise characterized by the α-stable distribution. The stability in the mean of proposed distributed algorithm is then studied. Finally, the superiorities of proposed approach compared to diffusion kernel least-mean-square (KLMS) algorithm are confirmed via the experimental results.