Adaptive estimation of power spectrum by using genetic algorithm

A new method for adaptive estimation of a nonstationary power spectrum is proposed. The method uses a new model based on a time-varying coefficient autoregressive (AR) model in which the order of autoregression also varies with time. The nonstationary nature of the power spectrum can be obtained by estimating the time-varying coefficients, and an abrupt change of the structure of the spectrum can be estimated by the time-varying order. The model is written in a state space representation with a system model that defines the smoothness of time-varying parameters and an observation model consisting of the time-varying parameter AR model. A Monte Carlo filter and genetic algorithm, which are very similar except for crossover, are used for the estimation of AR coefficients and the order, respectively. A simulation experiment shows the estimation result by the proposed method.