The Yule Walker equations produce biased estimates of the autoregressive AR process linear prediction coeffi-cients when the observations are contaminated by an additive noise. In this paper we present an alternative se-quential approach using two Kalman filters running in parallel. Alternatively, one Kalman filter produces the AR parameters to estimate the signal while the second updates the AR parameter values from the estimated signal. Besides, the noise statistics necessary to run the Kalman filters can be estimated using the optimality properties of the Kalman filter. This method has the advantage of providing an unbiased estimation of the parameters from noisy observations and an estimation of the signal in the steady state.