Blind Multi-user Detection In MIMO-OFDM System

Blind multiuser detection is the detection technique for multi-user developed in the field of communication signal processing in recent years. The characteristic is to receive the data of users by receiver in the communication system, without using the information of other users, to get the interference cancellation among multiusers. With the rapid development of mobile communication, though there are a lot of advantages in the traditional adaptive multiuser detection, it requires to improve systematic capacity and transmission rate in the communication system. And a large amount of waste of spectrum resources will be caused by the constant re-sending of the training sequence, when the channel changes or the new user arrives. Therefore, blind multi-user detection without using the training sequence manifests the enormous application prospect in the multi-user detection of mobile communication.Since MIMO technology has been used to greatly improve the systematic capacity, OFDM system has been used to convert frequency selective fading channel to flat fading channel and get the interference cancellation among signals, the combination of MIMO and OFDM can increase the transmission rate of data in the mobile communication system. Therefore, MIMO-OFDM becomes the key technology of 4G Mobile Communication. In MIMO-OFDM system, if it tries to recover the original user signals of sender accurately in the terminal, it requires the channel state information. However, due to the large number of antennas, the low efficiency of the channel estimation comes out and it also occupies a large amount of bandwidth by using the method of sending training sequence. Basic theories and methods on the blind multi-user detection were proposed in 1995 and stochastic gradient algorithm was adopted in the related researches. Since the speed of the algorithm depends on the characteristic distribution for the relevant matrix of the data, it also bears upon the number of users and the energy with the interference users. When it meets with the large number of users and large amount of energy of the interference users, the result of slow convergence of stochastic gradient algorithm and error come out. Considering computation burden, complicated structure and the difficulties of realizing the hardware, blind multiuser detection based on ICA is proposed in the thesis. Including the following aspects:1. Transforming the data structures of received signal. Considering the easier hardware and software processing, the received data frames of OFDM signals are divided into several segments at FB. Then longitudinal permutation and combination of each segment is employed, with collecting signals at each FB viewpoint. Therefore, each user forms an instantaneous mixture at each FB. The entire delay spread signals are covered by all the signals at each FB. As the signals of user are re-arranged, it causes some difficulties for the recovery of the original signal. The terminal and the sender as well as the structure of channel should be considered, when the output of the instantaneous mixtures which are suitable for NGA algorithm are transformed by received signals of the terminal in MIMO-OFDM system. Therefore, the deduction of mathematical model is required. The input of NGA algorithm and the software processing are determined by mathematical model. Due to the Cyclic Prefix (CP), Inverse Discrete Fourier Transform (IDFT) and Discrete Fourier Transform (DFT) modulation nature of OFDM signal, the output signals of DFT modulation are assigned according to the subcarriers and FB viewpoints in order to form the instantaneous mixtures at FB of MIMO channel. Thus, blind multiuser detection in MIMO-OFDM can be treated as N sets of problems of blind source separation. Through the research of the systems for two users, the mathematical model in general situation is deduced in this thesis.2. The design for pre-filter of the sender. The recomposition of user signals at each FB after the separation by NGA algorithm is the key problem processed in the research. The representative mixture model of blind source is X = AS+N, where X is the received sequence, A is mixture matrix, S is sending sequence, N is noise sequence. The purpose of blind source separation is to find out separation matrix W , to obtain Y = WX, then to attain the complete separation of signal, without using the information of S and A . In fact, the essence of the separation is Y = GS+WN.G is the mixture/ separation matrix in the whole. The recovery source in free of noise is Y = GS. Assuming the estimation of WN is zero; it has several processing methods such as subspace method and so on. In the ideal condition, G = I(unit matrix) is required. However, in most cases, it can be obtained as G = PD. P is permutation matrix and it will disturb the sequence of source. D is diagonal matrix and it will cause the problem of undesired scaling. It leads to find out the correlation of signals after separation. Due to the independence of each user and the signal sequence of the user, pre-processing tends to be applied in the sender. If the pre-filter is designed in the sender, the signal of each user after separation has the relationship with the adjacent FBs. Therefore, fuzziness brought by the NGA algorithm can be removed completely. 3. Proposing the correction algorithm for the amplitude variation and the phase distortion of different user signals at the adjacent FBs by using the characteristic of second order statistics. Since the independence and autocorrelation of the user exist at each FB, different user signal at different FB can be discriminated by the related computations for the based user signal and the adjacent user signal. Then, the other user signal can be discriminated by the related computations of the next signal at adjacent FB, with the newly discriminated signal. According to the algorithm, the signal at each FB can be recovered correctly by iteration computation.