A novel time-frequency analysis in nonstationary signals based multiscale radial basis functions and forward orthogonal regression

For time-frequency analysis of nonstationary signals, an adaptive and efficient time-varying autoregressive (TVAR) modeling method based on the multiscale radial basis function (MRBF) network and forward orthogonal regression (FOR) algorithm is investigated in this paper. Specifically, time-varying coefficients in the TVAR model is firstly approximated by the MRBF which has a better performance of tracking the time-varying parameters in nonstationary signals. Thus, the time-varying modeling problem is simplified to the selection of optimal centers and scales of MRBF, which a modified particle swarm optimization (MPSO) method aided by a FOR algorithm are resolved. Secondly, recursive least squares (RLS), Legendre polynomials expansion method and single scale radial basis function approach (SSRBF) are used to compare with the proposed method to evaluate the performance. Finally, the experimental results indicate that the proposed approach outperforms competing techniques in terms of mean absolute error and root mean squared error, and show the effectiveness of the proposed method for extracting the nonstationary signals. abstract environment.