Massive MIMO Transmission Based On Machine Learning

With the development of science and technology,the demand for wireless transmission rate in the future human society increases exponentially.With more antenna connections,the massive multiple-input multiple-output(MIMO)technology can achieve a faster transmission rate,stronger reliability,lower latency,and lower power consumption.It has become one of the key technologies of 5th generation mobile communication systems(5G)and beyond.In order to cope with the challenges of high complexity in traditional wireless physical layer transmission technologies,the application of machine learning in the field of wireless communication has been actively explored,such as channel estimation,signal detection,precoding,end-to-end communication systems,resource allocation,etc.One of its advantages is to improve online performance or reduce online computational complexity through offline training.For these considerations,this dissertation studies the massive MIMO transmission based on machine learning.Firstly,we investigate multi-cell scheduling for massive MIMO communications with only statistical channel state information(CSI).The objective of multi-cell scheduling is to activate a subset of users so as to maximize the ergodic sum rate subject to the per-cell total transmit power constraint.By adopting beam division multiple access(BDMA)based on the statistical CSI,i.e.,channel-coupling matrix(CCM),we simplify multi-cell scheduling as a power control problem in the beam domain,by which the ergodic sum rate is maximized.To reduce the computational burden on finding the ergodic sum rate,we propose a learning-to-compute strategy,which directly computes the complex ergodic rate function from CCMs via a deep neural network.Specifically,by modeling the probability density function of the ordered eigenvalues of the Hermitian CCM matrices as exponential family distributions,a properly designed hybrid neural network makes the ergodic rate computation feasible.With the learning-to-compute strategy,the online computational complexity of multi-cell scheduling is substantially reduced compared with the existing Monte Carlo or deterministic equivalent(DE)based methods while maintaining nearly the same performance.Secondly,we consider massive MIMO communication systems with a uniform planar array(UPA)at the base station(BS)and investigate the downlink precoder design with imperfect CSI.By exploiting channel estimates and statistical parameters of channel estimation error,we aim to design precoding vectors to maximize the utility function on the ergodic rates of users subject to a total transmit power constraint.By employing an upper bound of the ergodic rate,we leverage the corresponding Lagrangian formulation and identify the structural characteristics of the optimal precoder as the solution to a generalized eigenvalue problem.The Lagrange multipliers play a crucial role in determining both precoding directions and power parameters,yet are challenging to be solved directly.To figure out the Lagrange multipliers,we develop a general framework underpinned by a properly designed neural network that learns directly from CSI.To further relieve the computational burden,we obtain a low-complexity framework by decomposing the original problem into computationally efficient subproblems with instantaneous and statistical CSI handled separately.With the offline pre-trained neural network,the online computational complexity of the precoder is substantially reduced compared with the existing iterative algorithm while maintaining nearly the same performance.Thirdly,we investigate the downlink robust precoding with imperfect CSI for massive MIMO communications.With the estimated channel and channel error statistics,the general design of the robust precoder is to maximize the ergodic sum rate subject to the total transmit power constraint.To make the problem more tractable,we find a lower bound of the ergodic sum rate and propose the robust weighted minimum mean-squared-error(WMMSE)precoder to maximize the bound.We characterize the structure of the precoding vectors by low-dimensional parameters,which are learned directly from the available CSI through a neural network.As such,the precoding vectors can be immediately computed without iterations.To extend the deep learning design to multi-antennas users,we present a flexible approach that allows the various antenna configurations at the user side to be handled.Thanks to the offline-trained neural networks,the online computational complexity of precoding is significantly reduced while maintaining nearly the optimal performance.Lastly,we investigate robust precoding with fast convergence for downlink massive MIMO communication systems with CSI at the base station.To address the computational complexity issues,deep learning has been actively explored.Nevertheless,labels for supervised learning require offline iterations,while the generalization requires a large number of samples to be guaranteed.Besides,transfer learning for generalization requires inevitable online training.The fast convergent iterative design can be used for dataset generation and online adaptation in deep learning design.The general design of the robust precoder is to maximize the ergodic sum rate subject to a total transmit power constraint.We leverage an approximated formulation with high accuracy in massive MIMO,by which the complicated expectation-related computations can be addressed.Thereby,the robust iterative signal-to-interference-plus-noise-ratio(ISINR)precoder is proposed,which converges within a couple of iterations,as shown in simulation results.The robust iterative SINR precoder approaches the maximum ergodic sum rate and simultaneously reduces the computational complexity compared with the existing precoders.