Multi-Timescale Online Stochastic Optimization Algorithms For Massive MIMO Systems

During the last decade,the explosive growth in the number,type and functionality of smart mobile devices has spurred the development of new mobile services,such as smart health care,smart homes and smart manufacturing.To enable efficient and reliable data transmissions in various emerging services,the current cellular systems have to be redesigned to increase both spectral efficiency and energy efficiency.The idea of employing large-scale antennas at the base station(BS),i.e.massive multiple-input multiple-output(MIMO)technology,is envisioned as one promising solution.Compared to single-antenna systems,massive MIMO achieves large spatial multiplexing and array gains to serve the users with uniformly good quality of service simultaneously,with lowcost hardware and without using extra bandwidth and energy.In order to fully harvest the gain in spectral efficiency and energy efficiency,proper resource allocation is needed.However,the literatures on resource allocation for massive MIMO systems did not consider the channel estimation error explicitly and the design criterion was based on instantaneous channel state(CSI),where the optimization variables need to be recalculated whenever the channel changes.Therefore,it is not practical as it introduces significant computational and communication overheads,which comes from the control signaling for updating the optimization variables.To improve the robustness to the CSI error/delay with the reduced power consumption and signaling overhead,this thesis focuses on devising some multi-timescale online stochastic optimization algorithms for various massive MIMO systems by exploiting the channel hardening in massive MIMO and the special structure of the corresponding problem,where novel communication techniques,such as cloud radio access network(C-RAN),simultaneous wireless information and power transfer(SWIPT),symbiotic radio(SR)and nonorthogonal multiple-access(NOMA),are involved.In the first part,we investigate a general network utility maximization problem(GUNMP)for the uplink communication of a massive MIMO aided C-RAN.We propose a two-timescale hybrid compression and forward(THCF)scheme for the uplink transmission of massive MIMO aided CRAN,to alleviate the performance bottleneck of the limited fronthaul,with reduced hardware cost and power consumption.We formulate the optimization of THCF as a GUNMP,and propose a BCSSCA algorithm to find stationary solutions of this two-stage non-convex stochastic optimization problem.of the joint optimization problem almost surely.Simulations verify that the proposed BC-SSCA algorithm achieves significant gain over the state-of-the-art baseline schemes.In the second part,we investigate a GUNMP for the downlink communication of a multicell massive MIMO cellular network.To alleviate the performance bottleneck of the massive MIMO systems caused by huge CSI overhead,severe inter-cell interference,high cost and power consumption,we propose a randomized two-timescale hybrid precoding(RTHP)scheme and develop a two-stage online stochastic optimization(TOSO)algorithm named TOSO-RTHP to find stationary solutions of the associated two-stage non-convex stochastic optimization problem for the design of RTHP.Finally,simulation results verify that the proposed TOSO-RTHP algorithm can achieve significant gain over the state-of-the-art baseline schemes.In the third part,we consider a mixed-timescale joint beamforming and power splitting(MJBP)scheme to maximize general utility functions under a power constraint in the downlink of a massive MIMO SWIPT Internet-of-Things network,to alleviate the performance bottleneck caused by imperfect channel estimation and hardware limitation.In this scheme,the transmit digital beamformer is adapted to the imperfect CSI,while the receive power splitters are adapted to the longterm channel statistics only due to the consideration of hardware limit and signaling overhead.The formulated optimization problem is solved using a mixed-timescale online stochastic successive convex approximation algorithm based on the integration of sample average approximation method with fractional programming(FP)transform.Simulation results reveal significant gain over the state-of-the-art baseline schemes.In the fourth part,we consider the stochastic transceiver design for the downlink transmission of the multi-Tags SR systems,to alleviate the performance bottleneck caused by the direct-link interference and inter-Tag interference.We formulate the optimization of transceiver design as the GNUMP under some practical constraints.Based on the online observation of some random system states,we tailor a surrogate function with some specific structure and subsequently develop a novel algorithm named batch stochastic parallel decomposition to solve the resulting problem in a mini-batch fashion.Simulation results verify that the proposed algorithm can achieve significant gain over the state-of-the-art baseline schemes.In the last part,consider the distributed design of pilot sequences for massive connectivity in cellular networks.We formulate the optimization of pilot sequences as the power-constrained average MMSE minimization problem for different channel models,specifically,the spatially white channels and spatially correlated channels.By exploiting an approximation strategy,we first convert the formulated problem into a more tractable form.Furthermore,we exploit the special structure of the spatially white channels to develop an efficient low-complexity FP alternating direction method of multipliers(FP-ADMM)algorithm to find a stationary solution of this non-convex problem.In addition,we apply some block matrix computation methods to overcome the difficulty from the Kronecker product terms,and subsequently develop a block coordinate ADMM(BC-ADMM)algorithm.We show that there is a significant advantage in assigning more orthogonal sequence to device with high access probability and better channel.Simulation results verify that the proposed distributed pilot designs achieves significant gain over the existing schemes for both spatially white channels and spatially correlated channels,respectively.