Frequency estimation using nonlinear least squares and S-Transform

In this paper, an S-Transform (ST) based frequency estimation method is proposed. Firstly, ST gets a redundant time-frequency character from the original signal. Then, feature extraction algorithm can find multiple frequency peaks from the spectrum. Finally, we use a nonlinear least squares fitting to overcome the influence of distortion and noise, and identify the function of a time varying frequency with respect to time. The method can estimate varying frequency accurately with high computation efficiency. Simulation suggests that, the proposed method has many benefits: first, it can accurately estimate the short time frequency; second, the method is robust to noise; third, multiple simultaneous varying frequencies could be tracked; last but not least, the computation complexity is acceptable for real-time estimation.