Canonical correlation analysis (CCA) for ARMA spectral estimation

The canonical correlation analysis (CCA) of rational system identification is investigated for autoregressive moving-average (ARMA) spectral estimation at low SNR. The method is used to compute the parameters of the state-space Markovian model and its spectrum using CCA. It is shown that this approach yields significantly better results and improved resolution for low SNR. An interesting feature of the CCA is that the system parameters are sign symmetric, which reduces the computation cost by half. The performance of this method for spectral estimation of multiple sinusoids in noise is compared with singular value decomposition and the canonical vector method.<
>