Discrete time robust detection of stochastic signals in non-Gaussian contaminated noise (Q579209)

For many applications in signal detection, imprecise knowledge of the underlying noise process often makes desirable the employment of a robust detector. In this paper we consider the discrete-time detection of stochastic signals in white noise, where the univariate noise density is known perfectly only on an interval about the origin. We present a method to enhance the asymptotic performance of the detector by exploiting this knowledge, and at the same time preserve robustness properties of the detector to the remaining inexact knowledge of the univariate noise density via a saddlepoint condition. We then provide examples to show that improved performance is indeed obtained.