Random numerical linear precoding and channel estimation in massive MIMO systems

The information growth we have experienced in the immediate past and which continues to increase has consequently brought about the big data era and when pooled with the vast increase in subscriber numbers has led to an ever-escalating demand for more efficient and high-capacity communication systems. The affinity for higher capacity and efficient networks has necessitated the initiation of wireless fifth generation (5G) networks. Among the key technologies underlying the wireless 5G network are massive Multiple-Input Multiple-Output (MIMO) and Cloud Radio Access Network (C-RAN) which enhances spectral efficiency, energy efficiency, security and robustness but suffers from pilot contamination and fronthaul finite capacity. There have been several attempts to minimize pilot contamination in massive MIMO system through linear precoding. But for those precoding schemes with good performance, they suffer from intricate problem of matrix inversion owing to large antenna numbers inherent in massive MIMO system, yet they do not render themselves readily to hardware parallelization. Also, channel state information estimation remains a challenge within massive MIMO networks. While the finite fronthaul capacity remains a bottleneck in C-RAN network systems. This study presents the formulation of iterative linear precoder that is efficiently parallelizable with efficient channel estimators for massive MIMO and massive MIMO partially centralised CRAN networks. The channel precoder was formulated and adapted using the iterative linear Rapid Numerical Algorithm (RNA). This model was then extended to include coordination among multicell massive MIMO system with receive combining computational complexity and efficiency evaluation. RNA model is again used to formulate improved linear and semi-blind channel estimators for massive MIMO systems in combination with the Fast Data Projection Method (FDPM). The semi-blind channel estimator is combined with compressed data channel estimator then extended based on Givens transformations and Data Projection Method (DPM) for massive MIMO partially centralised C-RAN networks. And finally, the estimation of the signal-to-interference-to-noise ratio, bit error rates, spectral efficiency, energy efficiency and normalised mean square error for the respective modelled components was realized. The models above were simulated using MATLAB for the analysis and validation. The TDD downlink massive MIMO system was considered with varying immediate channel state information qualities for the single cell and multicell systems. For single cell system, there was optimal performance with regard to the signal-to-interference-to-noise ratio and the bit error rate when rapid numerical algorithm was used to implement the matrix inversion process in comparison to existing methods. It also rendered the precoding process highly parallelizable further reducing the complexity. For instance, for base transceiver station with 128 antennas serving 32 user terminals at signal-to-interference-to-noise ratio = 20 the average per user terminal rate was: RNA = 5 bit/sec/Hz, Regularized Zero Forcing (RZF) = 5 bit/sec/Hz and Truncated polynomial Expansion (TPE at J = 2) = 2.9 bit/sec/Hz. For the case of the Bit Error Rate (BER), for base transceiver station with 128 antennas serving 32 user terminals at signalto-interference-to-noise ratio = 10 the BER was: RNA = 1, Regularized Zero Forcing (RZF) = 1 and TPE (J = 2) = 5. For the multicell massive MIMO, it was found that the performance of rapid numerical algorithm implementation gave a good spectral efficiency and energy efficiency performance in comparison to existing methods while lowering the complexity further through parallelization. The compressed data channel estimator gave comparable performance for the spectral efficiency and normalized mean square error when compared to the improved linear channel estimators. The semi-blind channel estimators for both massive MIMO and massive MIMO partially centralised C-RAN outperformed the linear channel estimators as well as the compressed data channel estimator. These results demonstrate that rapid numerical algorithm can effectively eliminate the intricate matrix inversion associated with linear precoding while rendering itself to efficient parallelization. It also shows that the compressed data channel estimator optimally estimates the channel covariance matrix while reducing the amount of channel state information transmitted in estimation process. The semi-blind channel estimators have the optimal performance with regard to the normalised mean square error. It was also illustrated that the Givens transformation based semi-blind estimator outperforms the FDPM based semi-blind channel estimator.