A time-domain model-based method for the identification of multi-frequency signal parameters

Discrete Fourier Transform (DFT) is frequently used to determine parameters like frequency, amplitude, and phase; however, the finite length of data being used often leads to errors. Modeling a signal process in time-domain and estimating model parameters from observed data will provide a prediction about the data outside the sampled window and thus may lead to a better estimation of the signal spectrum. Many models were developed in literature for this purpose. While these models can provide better spectrum estimation than DFT, they may not be the best option because the parameter estimation is only suboptimal of least squares or the methods have very narrow convergence area. In this paper, the traditional nonlinear separable estimation has been modified and a new time-domain least-squares method was developed for the identification of multi-frequency signal parameters. Algorithm convergence was analyzed and improved through segmentation of data. Simulations and an experiment were provided to validate the effectiveness of the developed method