Information theoretic generalization of stochastic resonance
A. Zador 
, A.R. Bulsara 
Salk Institute MNL/S
10010 N. Torrey Pines Rd, La Jolla, CA 92037
Naval Command, Control and Ocean Surveillance Center, RDT & E Division,
San Diego, CA 92152-5000, USA
Intuition suggests that when noise is added to a signal prior to transmission across a communication channel, the recieved signal will be more corrupted than if the uncorrupted signal had been transmitted. The amount of corruption is typically quantified by the signal-to-noise ratio (SNR): noise tends to cause a monotonic decrease in the SNR. For a surprisingly large class of channels, there is a seemingly paradoxical increase in the SNR with added noise. The observation of a maximum in the SNR vs. noise relation has been widely studied under the name of stochastic resonance (SR). In many systems, howeveer, the SNR may be undefined or uninformative. In this paper we develop an information theoretic generalization of SR by defining it in terms of the mutual information between the transmitted and recieved signal. This generalization is both well-defined and informative in a much larger class of problems.