Segmented Autoregression Pitch Estimation Method

The paper is devoted to method of parameters estimation of modulated signals and its application to vibration analysis. Spectral diagnosing methods are the most frequently used. Spectral leakage due to amplitude and frequency variance significantly decreases reliability of results of frequency domain methods. Windowed analysis methods such as Short-time Fourier transform or wavelet transform are usually used to overcome these issues. They may be applied to track instantaneous frequency, to average or resample signal for frequency variation elimination. Then time-frequency resolution problem arises. Autoregressive methods are proposed for short time data analysis without resolution losses. Segmented Prony technique is known as non-steady signals analysis method. Its results are more accurate then Fourier methods results at short signal frames. Its disadvantages are false components if SNR is low and strong results dependence on the window width and time shift. This dependence is especially significant for nonstationary, modulated and pulse components that are essential informative features in vibrational signal. Preliminary wavelet decomposition is proposed to detect and separate single components. Time-frequency windows for segmented Prony analysis were selected using wavelet transform. Performance of the method in detection and estimation of parameters of modulated signals was examined and compared with conventional windowed Prony procedure. Efficiency of real vibration analysis was evaluated on the example of difficult for processing cases.