Generalized maximum correntropy detector for non-Gaussian environments

摘要

This paper addresses the problem of multiple-hypothesis detection. In many applications, assuming the Gaussian distribution for undesirable disturbances does not yield a sufficient model. On the other hand, under the non-Gaussian noise/interference assumption, the optimal detector will be impractically complex. Therewith, inspired by the optimal maximum likelihood detector, a suboptimal detector is designed. In particular, a novel detector based on the generalized correntropy, which adopts the generalized Gaussian density function as the kernel, is proposed. Simulations demonstrate that, in non-Gaussian noise models, the generalized correntropy detector significantly outperforms other commonly used detectors. The efficient and robust performance of the proposed detection method is illustrated in both light-tailed and heavy-tailed noise distributions.