The Research Of The Methods Of Time-frequency For Electroencephalography Based On The DIVA Model

In recent years, as a rapid development of electronic information, communication andcomputer science, it becomes more and more perfect for the electroencephalography(EEG) signalprocessing theory and techniques. Therefore, the research area of brain-computer interface(BCI)attracts more interests as a currently hot topic, of which the way to obtain EEG and the subsequentpattern classification for the unique character extraction from EEG plays vital importance in BCIresearch system. EEG is a non-stationary time-varying signal, carrying all kinds of information. Anaccurate analysis to EEG can improve the efficiency of the processing of the EEG. Thetime-frequent analytical theory and method, an effective tool for non-stationary signal analysis, cananalyze both time domain and frequency domain simultaneously, which becomes a new researcharea of signal processing in recent years. As a new theory and method, it has provided theoreticalsignificance and application value to discuss the methods of time-frequency. In this paper, themethods of time-frequency for EEG are investigated as the following sections:First of all, the basic theory of the time-frequency analysis and several commonly usedtime-frequency analysis method are introduced, such as classic Fourier transform, Short TimeFourier Transform (STFT) and wavelet transform. Classic Fourier transform only applies tostationary signals, which can not analyze non-stationary signals. STFT can analyze brain electricalsignals from either time domain or frequency domain, but the accuracy and resolution are poor.Wavelet transform can only analyze the approximation of the signal component, while ignoresfurther decomposition, which will affect the divided band of non-stationary random signalcharacteristics. Thus, the accurate extraction of the signal characteristics will be affected.Secondly, a High Time-Frequency Resolution Analysis(HTFRA) for electroencephalographybased on DIVA model is proposed. Although Wigner-Ville distribution (WVD) can analyzetime-frequency for non-stationary signals with high resolution, it will introduce cross terms duringfeature extraction because of its quadratic time-frequency analysis methods, which will affect theunderstanding of the information of the signal characteristics. The HTFRA is based on WVD andeffectively eliminates the cross terms of WVD without affecting the signal resolution by using theMedian Affined Filter (MAF) method for nonlinear filtering. The simulated signals are analyzedwith Short-Time Fourier Transform, Wigner-Ville distribution, and HTFRA, respectively. Theresults indicate that HTFRA gives a better energy distribution in the time-frequency field comparedwith the traditional methods.Finally, time-frequency analysis of EEG based on improved S transform is discussed. S transform is a combination of short-time fourier transform and wavelet transform, which selectsappropriate mother wavelets for different EEG, and analyzes time frequency of the simulatedsignals based on the improved S transform. Simulation experiments show that the time-frequencystructure of the improved S transform is more precise compared with STFT and wavelet transform.