The optimal detection of a signal of known form hidden in additive white noise is examined in the framework of stochastic resonance and noise-aided information processing. Conditions are exhibited where the performance in the optimal detection increases when the level of the additive (non-Gaussian bimodal) noise is raised. On the additive signal–noise mixture, when a threshold quantization is performed prior to the optimal detection, another form of improvement by noise can be obtained, with subthreshold signals and Gaussian noise. Optimization of the quantization threshold shows that even in symmetric detection settings, the optimal threshold can be away from the center of symmetry and in subthreshold configuration of the signals. These properties concerning non-Gaussian noise and nonlinear preprocessing in optimal detection, are meaningful to the current exploration of the various modalities and potentialities of stochastic resonance.
