Stable simplified gradient algorithms for total least squares filtering

An algorithm for total least squares filtering is developed based on a gradient descent approach. The total least squares approach can mitigate parameter estimation bias resulting from noise effects in the adaptive filter. The stability properties of the algorithm are analyzed and stable and convergent behavior is established The proposed algorithm is implemented without divisions or square roots, and does not require normalization. Despite recent interest in iterative total least squares methods, there are few algorithms that both avoid the computational complexities described above and have been shown to possess stable and convergent behavior.