stochastic resonance

The principle behind stochastic resonance is quite simple. Imagine a system that has two stable equilibrium points separated by an unstable one. Consider one of the stable point as being the "input not detected" state, and the other one the "input detected" state. Now consider the system to be around the "input not detected" stable point, and excite it with a periodic stimulus that makes its state oscillate around this point. Periodically, the stimulus brings the state of the system close to the unstable point, but not enough to switch to the other side towards the "input detected" stable point. The stimulus is thus not detected. Now add noise in the stimulus. The oscillation of the stimulus is in overall degraded, but sometimes, if the noise is powerful enough, the state of the system will be pushed past the unstable point and will hit the "input detected" point.

If the noise has a rather small power, it is likely that the state of the system will be stuck around the the "input detected" stable point, and the oscillatory nature of the stimulus won't be perceived. Increase the noise power a little, and transitions between the two stable points will occur more often, thus giving more information about the stimulus' period, and thus increasing the signal to noise ratio. If noise is too powerful of course, transitions will occur at random and signal to noise ration will drop.

Since most mathematical models of neurons include this kind of thresholds, and since most living things live in a noisy environment, it is not surprising that stochastic resonance has been found to enhance the sensory abilities of various insects, shrimps, fishes, and even humans.