Improvements of a state-space iterative noise reduction algorithm for harmonic retrieval

When the model of a noisy sinusoidal process is autoregressive moving average (ARMA), then the AR spectrum is biased. However, since the AR spectrum contains all the second-order information of the process, it is possible to retrieve the noiseless predictor from the noisy one. An iterative algorithm enabling the computation of the ARMA parameters from the AR parameters and a new well-suited initialization scheme are presented. Simulations of the state-space iterative noise reduction algorithm (SINA) are performed using various AR estimators. The mean-square-error graph is plotted for all these estimators and performances of the methods are discussed