Blind Channel Identification Based On Subspace Method Of Second-order Statistics

Since blind identification just needs the outputs statistics to estimate the channel parameters, the study on blind system identification is the important technology in modern signal processing. Earlier approaches to blind identification exploited the higher order statistics (HOS) of the outputs. Because of the proposed algorithm by Tong, the second order statistics methods became the main algorithm for blind identification. Among the existing methods, the performance of subspace is outstanding. So the problem of subspace method is studied in this thesis, and the results are as follows:Firstly, the single-input multiple-output (SIMO) model is discussed, and some assumptions about model are given. The principle of the subspace method is reviewed in this section.Secondly, an improved subspace identification algorithm for SIMO system model is developed. The covariance matrix divided into blocks based on its rank. Considering the existence of noise and any other errors, the proposed method can obtain a matrix related to the noise subspace by using the total least squares (TLS). The orthonormalization of the above matrix is the noise subspace, which is important for subspace methods. Compared with the traditional subspace method, the proposed algorithm does not need eigenvalue decomposition of the covariance matrix. The analysis result shows it reduces computation labor.Finally, the major problem of subspace methods lies on that they need precise channel orders, which are not easy to obtain. In contrast, channel order upper bounds can be obtained from some priori knowledge. This method not only tolerates channel order overestimation, but also reduces the computation. Simulation indicates the performance of the proposed method.