Maximum Likelihood Based Parameter Estimation of Stationary and Non-Stationary Multi-Component Signals

摘要

In this paper the problem of extracting signal parameters of stationary and non-stationary signals by applying a parameter estimation scheme is discussed. At first, a Maximum Likelihood based approach is introduced, which combines the most classical spectral analysis tool, the Fourier transform, with a modern parameter estimation scheme, the EM algorithm. After explaining the ideas of local polynomial approximation, an extended signal model is proposed, which considers linear parameter variations in order to extract the evolutional characteristics of a non-stationary signal. By replacing the EM algorithm with its extended derivative the SAGE algorithm, an efficient method for extracting the parameters of superimposed chirp signals is derived. The performance of the novel algorithm is illustrated on the basis of a synthetic test signal. Furthermore, an application to the problem of extracting time-varying parameters of a frequency selective radio channel is presented to show the flexibility of the proposed estimation scheme.