Study On Massive MIMO With Low-resolution ADCs/DACs

Massive MIMO is a key technology to the 5G communications,which shows large gains over traditional multiple antenna techniques in terms of capacity,spectral efficiency and etc.However,the enormous radio frequency units required in Massive MIMO will cause tremendous power consumption and system cost.To address this problem,using low-resolution ADCs and DACs at the base station can efficiently reduce the power consumption.In this thesis,we study the Massive MIMO system with low-resolution ADCs/DACs.For uplink systems with LR-ADCs,we study the multiuser detection of Massive MIMO with OFDM-IM modulation.An efficient frequency-domain component detector called pattern-aware parallel approximate message passing(PA-P-AMP)is proposed.In the proposed PA-P-AMP,first,the large-size frequency domain detection problem is addressed by multiple parallel small-size AMP detectors;second,the activation pattern constraint in index modulation is integrated efficiently.The frequency-domain detector is then utilized in the G-Turbo detection framework which achieves a prominent BER performance.For downlink systems with LR-DACs,we study the non-linear 1-bit precoding scheme.First,we introduce l₀-norm constraint on quantized transmit signals to further reduce the number of radio-frequency chains and thus the power consumption.The alternative direction method of multipliers(ADMM)is employed to solve this precoding problem with l₀-norm constraint.Specifically,an efficient two-step implementation of the non-convex projection operator in ADMM is presented.Simulation results show that the proposed precoding with l₀-norm constraint can achieve good BER performance,besides reduce the energy consumption.Moreover,we study precoding schemes to optimize the performance of the worst user based on the criterion of min-max MSE for block and slow fading channels.Specifically,two different precoding methods were proposed,SDR-Gaussian Randomization(GR)and ADMM-SF.In the first SDR-GR method,the min-max MSE problem is first formulated as a semi-definite programming problem with non-convex rank-1 constraint,and then SDR is employed to relax this constraint.Finally,the output of SDR is further processed by Gaussian randomization.In the low-complexity two-stage ADMM-SF method,the initial solution is first acquired by the efficient ADMM algorithm for min-sum MSE problem,and then refined to achieve better min-max MSE performance by the symbol-flipping.Simulation results show that the proposed method can improve the BER performance of the worst user efficiently.