Resonance frequency estimation of time-series data by subspace method

This paper studies an estimation problem of a dominant resonance frequency from time-series data. We proposed an estimation method which incorporates system identification technique into time-series analysis. However, this method has a problem that the estimated resonance frequency is biased. In this paper, a new method which uses subspace method is proposed based on time-series data. The key idea of this method is to use an auto-covariance function of the time-series data instead of impulse response or ordinary input-output data. Hankel matrix of the time-series is constructed by the auto-covariance function. Then, subspace method is applied to the Hankel matrix, and the resonance frequency can be calculated. Effectiveness of the method is examined through numerical examples.