Improving Channel Capacity in the LTE Downlink through Channel Prediction

Abstract

Link adaptation, multiuser resource scheduling and adaptive MIMO precoding are implemented in the Long Term Evolution (LTE) downlink in order to improve spectral efficiency and enhance effective utilization of the available radio resources. These processes require the transmitter to have an accurate knowledge of the channel state information (CSI). This is typically provided via feedback from the receiver. Due to processing and feedback delays, the CSI used at the transmitter is outdated leading to performance degradation causing a decrease in the overall system capacity. Channel prediction is an important technique that can be used to mitigate the system degradation that arises as a result of the inevitable feedback delay. The minimum mean square error (MMSE) based algorithms have been proven to have high performance in channel estimation and prediction. However this superior performance is accompanied by a high computational complexity due to the matrix inversion required as well as the large size of the channel matrix. In this thesis, the problem of channel aging on the LTE downlink is discussed. After a review of the LTE architecture along with its MIMO-OFDM radio interface, an overview of transmissions through a wireless fading channel is presented. A system model for block fading channels is then presented and a MMSE channel prediction method is derived. A reduced complexity approximate MMSE channel (AMMSE) prediction algorithm is then proposed to reduce the high computational complexity inherent in the MMSE method. The complexity reduction is achieved through reduction of the size of the channel matrix as well as approximating the matrix inversion through iteration. Evaluation of the proposed approximate MMSE algorithm

indicates that the mean square error can be reduced by up to 13 dB for feedback delays of 0 – 15 ms which translates to an improvement of 22% in the average throughput for slow fading channels. Fast fading channels are characterized by high Doppler spread and are dispersive in both time and frequency. At high Doppler spread, the effects of channel aging become more pronounced and the channel capacity decreases rapidly without prediction. The block fading model which is flat fading and only utilizes the temporal correlations is inadequate for fast fading channels. A novel three dimensional minimum mean square error (3D-MMSE) channel prediction algorithm which utilizes the time, frequency and spatial correlations is developed for the fast fading scenario. The performance of the proposed algorithm in doubly dispersive channels is studied and analyzed. The results indicate that the proposed 3D-MMSE prediction technique provides 17.4 % enhancement in average throughput for feedback delays of 0 – 10 ms compared with the scenario without prediction in fast fading channels. A reduced complexity approximate 3D-MMSE (A3D-MMSE) algorithm is then proposed to reduce the high computational complexity inherent in the 3D-MMSE method. The approximate method operates through a three-step algorithm which first exploits temporal correlations, followed by separate smoothing filters to exploit the frequency and spatial correlations. A comparative analysis indicates that the proposed approximate 3D-MMSE algorithm provides 89 % reduction in complexity compared to the 3D-MMSE algorithm. The results show that an increase in the Doppler spread above 200 Hz leads to decrease in the temporal correlations and this makes the channel prediction to become increasingly inaccurate.