ARMA spectral estimation based on non-linear least squares

An autoregressive moving average (ARMA) spectral estimation method based on the nonlinear least squares technique is presented. Simultaneous estimates of the AR and MA parameters are obtained by minimizing the prediction error using the Levenburg-Marqudit optimal search algorithm. If the model order is properly selected, it can be shown that this technique converges to the true model parameter values assuming that the model is stable and minimum phase. This technique is valid when applied to non-Gaussian white noise. Results for simulated data are presented to illustrate the application and accuracy of the technique