Simo Blind System Identification Algorithm

Blind system identification (BSI) is a fundamental signal processing technology aimed at estimating a system's unknown information from its outputs and some auxiliary information only. The technique of system identification without pilot sequences is very attractive due to the pursuit of large capacities and high reliabilities in mobile communications. The algorithms of blind identification about single-input multiple-output (SIMO) system are mainly investigated in the thesis.First of all, problem statements of blind system identification and the general identifiability conditions are described based on the introduction of two equivalent SIMO models.Then, batch and adaptive algorithms based on linear time-invariant channels and time-varying blind identification algorithms are discussed systematically.Batch algorithms are to estimate channels of linear time-invariant or varying slowly. Two deterministic batch algorithms, which are CR and TSML algorithms, are first introduced. And the first step of the TSML algorithm is to estimate the channel parameters initially by the CR algorithm. Due to the large computations of the TSML method, a kind of Minimum TSML method with its computation diminished is proposed, and the larger of the sub-channels' number, the larger of the computations' diminishment. It is verified by simulations that the performance of the Minimum TSML algorithm is quite approximate to that of TSML algorithm. In addition, another kind of batch algorithm based on second-order statistics, linear prediction blind identification algorithm, is introduced. This algorithm can keep good robust performance even if the channel's order is overestimated.For the sake of tracking the channels' characteristic adaptively, several adaptive blind identification algorithms based on the celebrated CR property of the sub-channels' outputs, are discussed. They are adaptive blind identification algorithms based on LMS and Newton methods, parallel multi-channel adaptive algorithm based on pair wise sub-channels' estimation and multi-channel adaptive algorithm based on RLS method. Furthermore, the multi-channel RLS adaptive blind identification algorithm proposed by this thesis is the generalization of the RLS blind identification algorithm about single-input double-output system. The algorithm performs well and converges in high speed.When the channel undergoes fast fading, the adaptive methods will yield degraded performance because they cannot track the channel's variation, so it is necessary to research on approaches based on time-varying model accounting for fast fading. The linear predicton batch algorithm and time-varying model based on complex exponential basis expansion are first introduced, and then the linear prediction blind identification algorithm and blind equalization algorithm based on this kind of time-varying model are described in detail.One of the identifiability conditions of BSI is that the entire sub-channels do not share any common zeros, i.e., the system is identifiable so long as two sub-channels of the system do not. It is verified by simulations that for blind identification algorithms estimating the whole channels jointly, the better of the identifiability performance of the most different two sub-channels, the better of that of the whole channels, and furthermore, the estimatimation performance of the most different two sub-channels can predict that of the whole channels. And the mean error of the NRMSEs of both is not bigger than 2dB.In addition, the performance evaluation formula of blind identification algorithms about complex channels' impulse responses is developed and used in the whole thesis.MATLAB simulations have also been done to verify the algorithms' performance and conclusions in the thesis.