A new algorithm for linear system identification

This correspondence considers the on-line parameter identification of a forced linear discrete-time dynamic system from a sequence of white-noise-corrupted output measurements. In contrast to other approaches, the proposed stochastic approximation algorithm does not require knowledge of the noise statistics and converges to the true value of the parameters in the mean-square sense. If the input measurements are also corrupted with white noise, an additional term depending on the variance of the noise is required.