Asymptotically robust detection in partially contaminated noise (Q1057025)

This paper considers the problem of discrete-time detection of a known time-varying deterministic signal in independent and identically distributed noise, when knowledge of the noise density is imperfect. More exactly, it is assumed that the noise density f(.) is a member of the Huber-Tukey mixture class \(f(.)=(1-a)f_ 0(.)+ah(.)\) where \(a\in [0, 1]\) is a fixed known number, \(f_ 0(.)\) is a known nominal density, and h(.) is an unknown contaminating density. In addition it is assumed that f(x) is known exactly for \(x\in [c, d]\), where c and d may be arbitrary. A detector which is robust to the inexact knowledge of noise density is proposed. The basic idea is the introduction of limiting for observations of large magnitude.