A least-squares based method for autoregressive signals in the presence of noise

The problem of estimating the unknown parameters of an autoregressive (AR) signal observed in white noise, including the signal power and the noise variance, is studied. A new type of least-squares method is developed which is based on a simple technique of estimating the observation noise variance by increasing the degree of the underlying AR model by one. The main feature of the presented method is that the consistent estimates of AR parameters can be directly achieved, with no need to prefilter noisy data or to make any parameter transformation.