Novel antenna selection and dimensionality reduction in massive MIMO communication with the exploitation of multiple antenna radio channels

Massive multiple-input-multiple-output (MIMO) has the potential to offer a high system capacity in the fifth generation of wireless communication systems, exploiting the large number of degrees of freedom gained by utilising many transmit antennas at the base station. For a signal to be transmitted from an antenna element at the base station, the element needs to be connected to a radio-frequency chain which comprises Digital to-Analogue converters, a power amplifier and mixers etc. The total number of radio frequency chains equals the number of antenna elements used for active transmission. Therefore, the radio-frequency switching matrix represents the hardware components required in antenna selection for interconnection of the radio-frequency chains with their selected antennas. However, having a large number of antennas there are numerous challenges, in terms of hardware system complexity, large matrix sizes and signal correlation due to less space between antennas. To confront these problems, novel antenna-selection algorithms and dimensionality-reduction algorithms are proposed and spatial-correlation based channel models are explored. Antenna selection algorithms based on central Principal Component Analysis (PCA) are developed, and antenna selection algorithms using noncentral Principal Component Analysis (NPCA) and Linear Dependence Avoidance System (LDAS) are proposed. These algorithms reduce the correlation between the signals received by users. The performance of the proposed schemes with different antenna-selection algorithms and sum-capacity is evaluated. Signal correlation between the antennas is modelled by using spatially correlated channel models such as the Kronecker Model and the Weichselberger Model. The correlation matrices at both ends of the link are approximated by using Power Azimuth Spectrum (PAS) models. These models give the analytical signal modelling in correlative environments used to test the proposed antenna-selection algorithms to study the system behaviour in different realistic environments. Finally, the analysis of PCA is used that can reduce the size of huge matrices significantly, which saves a large number of computations as compared with the full-dimensional system. This thesis analyses the dimensionality reduction of large matrices using Floating Point Operation (FLOP) methods.This method is also implemented in the above-mentioned spatially correlated channel models to evaluate the throughput. The results in this thesis are presented mainly in four parts: a) novel antenna-selection algorithms, b) application of novel antenna selection algorithms to spatially correlated channels, c) large matrix computations are reduced using dimensionality reduction techniques, d) analysis of matrix dimension reduction for spatially correlated channel models. This dissertation therefore presents the methodologies of lowering hardware complexity, reducing large matrices’ dimensions and modelling signal correlation by using spatially correlated channel models that affect the design in the fifth generation of MIMO broadcast wireless communication systems.