Training Sequences Based Channel Estimation For MIMO Systems And Related Study

Channel estimation based on training sequences has the advantage of low complexity, high speed and excellent performance, which plays a very important role in modern wireless communications. From the viewpoint of data and training sequence transmissions, there are three major schemes of training based multi-input multi-output (MIMO) channel estimation. One is time-division multiplexed (TDM) scheme and the other two are superimposed training (ST) and data-dependent superimposed training (DDST) schemes. In this thesis, algorithms to improve the system performance and performance comparison of the three schemes are investigated.First of all, the relationship of channel capacity and training sequence length of TDM is analyzed. Optimal training length of TDM for frequency selected MIMO channel is derived when the power of training and data sequence or peak to average power ratio (PAPR) is given. The effect of training length on channel capacity and the relationship of optimal training length between signal to noise ratio (SNR) and PAPR is analyzed by simulation.Next, the optimal power allocation of ST scheme for frequency selective MIMO channel is derived. The relationship between the SNR of the channel equalizer and the training sequences power is analyzed. The optimal power allocation of the training sequence is derived based on the criterion of maximizing SNR of the equalizer. Analysis and simulation results show that the SNR of the channel equalizer is maximized at the optimal training sequence power, and the optimal power of the training sequences is increased with increase of the signal to noise ratio at the received antennas.Then, several algorithms are presented to improve the system performance of DDST. (1). For data detector, the data dependent sequences (DDS) added on the training and data sequences act as noise and thus degrading the data detection performance. A new DDS removal algorithm, which is not only suitable for BPSK signal but also suitable for high order equi-spaced amplitude or equi-spaced square quadrature amplitude modulation (QAM), is presented in this thesis. Symbol and bit error floor of the proposed detection method is analyzed too. To remove the error floor, a data coding method is also proposed and the redundant ratio of the coding algorithm is given. Analysis and simulation results show that the proposed detection method has lower complexity and better performance than the existing methods. The data coding algorithm can remove or reduce the error floor by much lower redundant ratio. (2). The existing DDST block synchronization algorithms work well for Single-input Single-output (SISO) systems, but can hardly work for MIMO system. A new joint block synchronization, channel and dc-offset estimation algorithm based on balanced zero correlation zone (ZCZ) sequence for MIMO system is proposed. Analysis and simulation results show that the new algorithm has the same performance as the existing algorithms for SISO systems when their block and cyclic prefix lengths are the same. While for MIMO systems, the performance of the proposed algorithm is much better than that of the existing algorithms. (3). Similar to the ST scheme, for a fixed transmission power, the data detection performance will degrade with the increase of training power. Relationship between the SNR of the data detector and the training sequence power is analyzed. The optimal power allocation of the training sequences and data sequences is derived when DDS is treated as noise and DDS is known. Analysis and simulation results show that the optimal power of DDST training sequences is independent of SNR and whether the DDS removal algorithm is employed.Finally, the channel capacity lower bounds of ST and DDST schemes are derived when optimal training power is employed. And the performance of TDM, ST and DDST is compared by training sequence selection, channel estimation MSE, data detection BER and system throughput. Simulation and numerical results show that, if the length and power of training is optimal and peak-to-average power ratio (PAPR) of the TDM and DDST is the same, almost all of the above performance of DDST outperforms that of TDM except DDST data detection performance of the existing DDS removal technology.