Study Of Time—Frequency Analysis And Processing Of No—stationary Signals

ABSTRACTTime frequency analysis is an effect tool to analyze and process non-stationarysignals, which maPs a one-dimensional signal into a two-dimensional function of timeand frequency and reveals the time-varying spectra. This dissertation fOcus on therelated theories and aPplications of time-frequency analysis, and the main researchworks and results are as follows:l. The time-frequency resolution of short-time Fourier transfOrm (STFT) andthe properties of time-lag-Doppler function (TLDF) are analyzed. AnadaPtive STFT with linear frequency modulation window is presented andthe instriboeous frequency modulation rate is estimated by the TLDF and -Radon technique.2. Based on the comparison of different interference restraining methods, anoptimal phase kernel time-frequency distribution is put forwarded. Theproperties of optimal phase kemel time-frequency distribUtion, thedetermination of auto-terms region, and the optimal phase kemel time-frequency distribution with additional properties are discussed. Then analgorithIn of discrete-time, discrete-frequency optimal phase kemel time-frequency distribution is proposed. Analyking bat signals by the optimalphase kemel time-frequency distribution, the high time-frequency resolutionand accurate time-frequency structure is obtained.3. By introducing the localized Radon technique into time-frequency analysisarea, the localized Radon-Wigner transfOrm (LRWT) is suggested. First, thedefinition of LRWT and a fast algorithm is consimcted. Then the propertiesof LRWT are analyzed and the relationship of LRWT with Radon-Wignertransform (RWT), Wigner distribution (WD) and Ambiguity function (AF)is discussed. Finally, the relationship of LRWT and generalized marginaltime-frequency distribution is studied.4. The properties of localized time-frequency distribution (LTFD) are analyzed,then an iterative instantaneous frequency estimation algorithIn is presented.The convergence of this algorithIn is examined analytically, the estimationof initial frequency is discussed. and the perfOrmance in noise is comparedwith other algorithIns by computer simulation. Finally, the Doppler signal ofa radar velocity analyzing system is studied by means of the method in this paper and the speed variation of the observed object is clearly described.5.Based on the analysis of the time-frequency structure of Doppler signals from multi-object by traditional Radar, a mathematical modal of a nonstationary signal is proposed, whose validity is demonstrated by real Radar echoes. In allusion to this modal, an adaptive Chirp representation (ACR) is presented, by which multi-object classification is easily realized.