Widely linear complex kernel risk-sensitive loss affine projection algorithm and its variant

Recently, the affine projection algorithm (APA) has shown remarkable performance in Gaussian environments, but it will perform poorly in impulsive noise. The APA with the maximum complex correntropy criterion (MCCC) as the cost function was incorporated to address the issue of handling colored input signals in non-Gaussian noises. Nevertheless, the estimation accuracy of MCCC-APA is limited. In this paper, we introduce the complex kernel risk-sensitive loss (CKRSL) as the cost function to integrate with APA, and then propose the widely linear complex kernel risk-sensitive loss affine projection algorithm (WL-CKRSL-APA) to deal well with the colored signals in non-Gaussian noises. To strike a balance between the convergence rate and filtering accuracy, we utilize a proportional coefficient and interval updating strategy to obtain an improved filtering performance. To validate its effectiveness, we conduct simulation experiments through system identification and stereophonic acoustic echo cancellation (SAEC).