Combined-Statistical-Distribution Generalized Maximum Correntropy Algorithm With Affine Projection for Robust Adaptive Filtering

This brief proposes a novel Combined-Statistical-Distribution generalized maximum correntropy algorithm with affine projection to improve the convergence rate and tracking ability in system identification. We design a new generalized Gaussian kernel to measure the system error and utilize the sigmoid activation function as a transition strategy of the Combined-Statistical-Distribution (CSD) factor. Meanwhile, we use the inequality scaling method to improve the constraints to avoid reducing the convergence rate. Under the
$\alpha $
-stable noise scenario and impulsive interference, the superiority of the proposed algorithm is demonstrated in system identification experiments.