Non-stationary signal processing using time-frequency filter banks

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