Channel Interpolation And Semi-blind Estimation In MIMO-OFDM System

The dissertation studies MIMO technique, OFDM modulation, wireless channels'characteristics and several channel models, and mainly investigates the acquisition approaches of channel state information (CSI). Pilot-aided channel estimation and interpolation as well as semi-blind subspace algorithms are simulated, analyzed and compared. The main work includes the following.1. It establishes the simulation platform of MIMO-OFDM system, Rayleigh fading channel and 3GPP's spatial channel model using MATLAB. The whole analysis and comparison work of interpolation algorithms and semi-blind estimation algorithm are done based on the platform.2. It presents two improved algorithms after estimating the channel fading of pilot tones. In the light of the limited condition of time transform domain that the number of sub-carriers is multiple times of pilots', the improved algorithms are proposed to suit the situation that pilot symbols are at both the first and the last sub-carriers, so as to decrease edge errors. The algorithms are compared to linear interpolation, Gaussian interpolation, cubic spline interpolation, low pass filter interpolation and Wiener interpolation. Simulation results show that the performances of mean square error (MSE) and bit error rate (BER) of linear interpolation, Gaussian interpolation, cubic spline interpolation and low pass filter interpolation are similar to each other, while the performances of transform domain interpolation and Wiener interpolation are greatly improved. Wiener interpolation has the best performances. However, the channel correlation matrix suffers from heavy computation. If the matrix is bias, it may incur large estimation error. In comparison with Wiener interpolation, the transform domain method using fast Fourier transform has no need of correlation matrix, and its performances are close to Wiener interpolation. Moreover, one of our improved algorithms is comparable to Wiener algorithm.3. The pilot-aided method has to send pilots frequently, which will greatly occupy spectrum resources. In order to utilize spectrum resources more efficiently, a semi-blind channel estimation algorithm based on subspace decomposition is investigated. The method only needs the upper bound of channel order, whereas it needs a few pilots to resolve matrix ambiguity. The paper simulates the method to obtain its performance of MSE, anti-overestimation, convergence speed and computational complexity. Furthermore, its identification ability for non-co-prime channel is researched under 3GPP spatial channel parameters, which have good flexibility and wide application. Simulation results show that the method has good MSE performance. When the estimated channel order is larger than its true order, the MSE performance is close to the case of accurate channel order. Hence, it is concluded that the method is insensitivity to over-estimation of channel order. The method can also identify non-co-prime channel, i.e., channel transfer function has common zeros in Z-domain. As long as the channel is suitable for the identification condition, the MSE performance will keep good enough. On the other hand, the method converges slowly, and has the computation burden that is the cube of channel correlation matrix's rank.