We present a new approach to the design of time-frequency (TF)filter banks for non-stationary noisy signals. The input multicomponentsignal is represented by the minimum cross-entropy TF distribution andthe system is based on an array of time-varying filters. Each filterprocesses one component of the signal according to its specific TFsupport. The output of the proposed TF filtering algorithm is a set ofthe signal components. To demonstrate the algorithm performance,simulation results are given. Since the suggested method is equivalentto two-dimensional matched filtering, an algorithm for non-stationarysignal classification is also presented
展开▼