Nonlocal robustness of the sign detector with nonstationary data

A popular nonlinear detection system is the sign detector, which is classically considered to be nonparametric when testing for the presence of a constant signal in independent and identically distributed noise whose univariate distribution possesses a fixed known value at the origin. It is shown via a nonlocal robustness analysis that in practice a severe loss in the nonparametric nature of the detector can be experienced. Compensation at the design stage is illustrated