On some system identification techniques for adaptive filtering

Three different identification methods (the Steiglitz-McBride method, the output error method, and the instrumental variable method) are discussed in the context of adaptive filtering. They can be implemented by recursive algorithms with similar structures, either in gradient or Newton form as well as in various tracking variants for time-varying systems. Their properties are discussed and compared in terms of local and global convergence, behavior for multimodal error surfaces, and form of approximation for underparametrized models. The instrumental variable method is assessed to be the best alternative in most respects. >