Abstract: By applying basis-functions expansion that uses wavelet decomposition and reconstruction filters and is rich in local time-frequency features, we introduce a novel algorithm identifying time varying parameters in TVAR model. A periodical extension method for processing two terminals of signals is implemented. Some high frequency coefficients in the wavelet decomposition are neglected to overcome the restriction in solving linear equations by layer number of wavelet reconstruction and the relationship between the model order and the minimal layer number of wavelet reconstruction. Research shows that the algorithm is effective in identifying the trend and frequency features of time-varying parameters. Increasing the sampling rate can reduce the effect of the neglected high frequency component, which is helpful to the identification the high frequency features of fast and instantaneous change signal.

Key words: basis-function, time-varying parameters, wavelet reconstruction filter, minimal reconstruction layer number, model order