In order to analyze the effect of noise signals on the estimation of harmonic and inter-harmonic parameters, a new method was proposed on the basis of the essential property of noise information and the several useful strategies including the modern cross-spectrum, total least square, ESPRIT (estimation of signal parameters via rotational invariance) and improved Prony techniques. In this method, the cross-spectrum technique is used to deal with the cross-correlation matrix of different sampled signal sequences, and signal and noise subspaces are identified validly with the help of the independent principle of white noise at different times. And then the frequencies of signal components can be extracted directly by combining the total least square technique with the ESPRIT technique. Finally, based on the estimated frequency of every signal component and the characteristic of zero average value of noise signals, the magnitude and initial phase of the signal components are obtained by using the improved Prony technique. The results show that conditioned upon low signal-to-noise ratio, the proposed method can distinguish more sub-signals and the relative errors of frequencies are less than 0.4% with merely time consumption of 0.058 s, characterizing a high estimation accuracy and efficiency.
DownLoad: