A neural network for spectrum estimation of quasi-stationary signal

The authors propose a system for adaptive processing of quasi-stationary signals, which accomplishes pattern clustering and recognition with spectral characteristics of stationary segments. With input being AR coefficients, Kohonen's self-organizing neural network plays a significant role in determining model transition time, pattern clustering and recognition, adaptive spectral estimation, and discovery of tendency or regularity in quasi-stationary signals. Finally, simulation results are given to prove its performance.<>