High-resolution time-frequency analysis of EEG signals using multiscale radial basis functions

An efficient time-varying autoregressive (TVAR) modeling approach using the multiscale radial basis functions (MRBF) method is presented for nonstationary signal processing, with applications to time-frequency analysis of electroencephalogram (EEG). In this new parametric modeling framework, the time-varying coefficients in the TVAR model are approximated by using MRBF that can better identify time-varying parameters with a variety of dynamic processes in nonstationary signals. Thus, the time-varying modeling problem is simplified to optimal scale determination of MRBF and parameter estimation, which can be effectively resolved by a modified particle swarm optimization (PSO) method and an ordinary least square (OLS) algorithm, respectively. To evaluate the performance of the proposed approach, a comparison with recursive least squares (RLS) and the Legendre polynomials expansion method for a synthesized EEG signal is performed. Results demonstrated that the proposed approach could indeed provide optimal time-frequency resolution as compared to RLS and Legendre polynomials expansion. The new TVAR modeling approach was also applied to the analysis of experimental EEG signals to demonstrate the performance of the proposed method.