Mean square cross error: performance analysis and applications in non-Gaussian signal processing

Abstract Most of the cost functions adaptive filtering algorithms include square error, which depends on current error signal. When additive noise is impulsive, we can expect that will be very large. By contrast, cross correlation signal and its delay, may small. Based this fact, propose a new function called mean for filters, provide value performance analysis in detail. Furthermore, present two-stage method to estimate closed-form solutions proposed method, generalize solution information theoretic learning methods, including least fourth, maximum correntropy criterion, generalized minimum kernel risk-sensitive loss. The simulations show effectivity method.