A separable kernel filter design procedure for signal time-frequency analysis

A method for designing quadratic time-frequency distributions (TFDs) with separable kernels is proposed. It is based on a detection of autoterms and crossterms peaks in the signal ambiguity function slices along the Doppler and lag axes. The peaks separations are used to define the time and lag smoothing windows lengths. The method is tested on multicomponent signals with linear and nonlinear FM components, as well as on a multicomponent nonstationary real-life signal. In all of the presented examples, the separable kernel TFDs, obtained using the proposed method, have been shown to have a significant interference reduction, with the components energy concentration around their respective instantaneous frequency laws being well preserved.