A neural network classifier for cyclostationary signals

A signal classifier based on features that measure cyclostationarity has been developed and tested on simulated signals. The results demonstrate that features that represent the “cyclostationary signature” of a signal can be be used to automatically categorize a wide variety of signal types. The signature consists of a vector of “structure coefficients” that are computed on the outputs of various non-linear transformations of the target signal. This approach can considerably simplify and shorten the development time of the classifier. It is necessary to precede the computation of the signature with rejection of narrowband interference. The decision logic for the classifier is implemented using the probabilistic neural network