Blind channel identification: subspace tracking method without rank estimation

Subspace (SS) methods are an effective approach for blind channel identification. However, these methods also have two major disadvantages: 1) They require accurate channel length estimation and/or rank estimation of the correlation matrix, which is difficult with noisy channels, and 2) they require a large amount of computation for the singular value decomposition (SVD), which makes it inconvenient for adaptive implementation. Although many adaptive subspace tracking algorithms can be applied, the computational complexity is still O(m/sup 3/), where m is the data vector length. In this paper, we introduce new recursive subspace algorithms using ULV updating and successive cancellation techniques. The new algorithms do not need to estimate the rank of the correlation matrix. Furthermore, the channel length can be overestimated initially and be recovered at the end by a successive cancellation procedure, which leads to more convenient implementations. The adaptive algorithm has computations of O(m/sup 2/) in each recursion. The new methods can be applied to either the single user or the multiuser cases. Simulations demonstrate their good performance.