Adaptive Channel Estimation Techniques for Massive MIMO Systems

Abstract

Wireless communication systems have been required to transmit a large volume of data more rapidly, and thus wireless communication systems require the utilization of more efficient bandwidth and larger channel capacity. Recently, extending multiple input multiple output (MIMO) system on a very large scale by using a myriad number of antennas at the base station was introduced. This method was so that the frequency efficiency can be greatly increased. Massive MIMO system can maximize frequency efficiency by applying a precoding scheme using the downlink channel state information (CSI) to the transmission data at the base station through the usage of large number of antennas. Time-division duplex (TDD) systems have been mainly studied as they can easily obtain the CSI by using the channel reciprocity between uplink and downlink. In frequency-division duplex (FDD) system, the computational complexity of the downlink channel estimation is proportional to the number of antennas at the base station as the channel reciprocity cannot be used. Therefore, effective channel estimation techniques may have to be studied. In this thesis, novel channel estimation algorithms using some of the adaptive techniques are proposed in a channel model with temporal and spatial correlations. The Kalman filter, known as the optimal channel estimation technique, is impossible to estimate the real-time channel due to matrix operations. When consecutive training signals are transmitted, we proposed time division operation of Kalman filter and normalized least mean square (nLMS) filter to enable channel estimation in real-time. Furthermore, we propose decision feedback nLMS filter which updates the CSI by using correctly decoded data as a training signal during data transmission period. With this approach, the performance can be greatly improved without much increase of the hardware complexity compared to the conventional nLMS filter. Simulation results show the performance of proposed algorithms compared to conventional algorithms in terms of mean square error (MSE) and bit error rate (BER).