Information-theoretic stochastic resonance in noise-floor limited systems: the case for adding noise.

We show that in systems whose output must compete with a noise source, stochastic resonance (maximization of output signal-noise separation as a nonmonotonic function of input noise strength) exists even when measured in terms of fundamental statistical measures and optimal detector performance. This is in contrast to the commonly considered scenario where, without the competing noise, the system (e.g., a driven, overdamped particle moving in a double well potential) is essentially invertible and optimal detector performance monotonically deteriorates with increasing input noise strength.