Massive MIMO and millimeter wave communications for 5G and beyond wireless systems
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Over the last decade, wireless data traffic has increased faster than ever before. Fourth-generation long-term evolution wireless cellular systems are reaching the theoretical limit of spectral efficiency with advanced signal processing, e.g., multiple-input multiple-output (MIMO), orthogonal frequency-division multiplexing, link adaptation, turbo coding, and hybrid automatic repeat request. So, to meet the expected future demands, new wireless solutions are required for fifth-generation and beyond wireless systems. Among many new proposals, massive MIMO and millimeter wave communications are two key technologies that can offer a significant improvement in system capacity. ☐ In massive MIMO, with a large number of antennas deployed at the base station, many mobile stations can be served simultaneously via multiuser beamforming over the same frequency band. With the high spatial resolution in massive MIMO, simple linear precoding (e.g., zero-forcing) can achieve almost optimal performance. In millimeter wave communications, the large available spectrum can easily support bandwidths of 1 GHz, which can enable gigabit-per-second data rates. In this dissertation, we study several challenging problems in massive MIMO and millimeter wave communications. ☐ We start with the general case for massive MIMO in frequency division duplex (FDD) systems, as it does not have the benefit of uplink/downlink channel reciprocity that time division duplex systems have. To reduce the downlink training and associated feedback in FDD systems, we first study user clustering, where mobile stations with similar channel spatial correlations are clustered together. Then, based on the clustering result, we propose an efficient eigenspace training and precoding framework, where two different prebeamforming matrices are designed to minimize the channel estimation error in the channel training and to cancel the inter-user interference in data transmissions, respectively. The new design achieves significant savings in downlink training and feedback because the overhead is associated with the number of beams in the training and no longer depends on the number of antennas at the base station. ☐ In millimeter wave systems, we study hybrid analog/digital precoding with finite-size codebooks. We first point out that it is the design of the outer analog precoder/combiner that fundamentally determines the spectral efficiency, because the inner digital precoder/combiner only generates a linear combination of the analog precoding/combining vectors. Then, we propose a new metric, which we call singular space coverage, to evaluate the outer analog precoder/combiner design, i.e., the performance limitation due to the finite-size codebooks and a small number of RF chains. Based on the new metric, we propose two types of algorithms with different performance complexity tradeoffs to select the analog precoding and combining vectors from the predefined codebooks, jointly or sequentially. Furthermore, we derive the achievable rates in hybrid precoding under constraints on the size of the codebooks at both the hybrid precoder and hybrid combiner, and propose an optimal baseband digital precoder. We show that, when the codebook sizes for the analog precoder/combiner are small, if we want to improve the spectral efficiency, it is more important to enlarge the codebook rather than increase the number of RF chains. ☐ Finally, we study the channel characteristics, beamforming training, and beam codebook designs for millimeter wave MIMO systems. In the study of channel characteristics, we analyze the spatial channel correlations, and demonstrate that, when the MIMO channels are highly correlated, the achievable spatial multiplexing order is smaller than the number of antennas at the transmitter and the receiver. Then, we propose a multipath channel model, in which the channel response is decomposed into two orthogonal subspaces. The orthogonality in the new channel decomposition helps to improve the channel estimation and beam codebook designs. In the study of beamforming training, we discuss closed-loop and open-loop training, and illustrate the main advantages of beamforming training over omnidirectional channel training. Then, we derive the conditions for reconstructing the channel information from the beamforming training when the spatial channel statistics are available and not available. In the study of beam codebook designs, we propose generalized DFT-based beam codebooks for uniform linear arrays and two-dimensional planar arrays. In the proposed design, the codebook size can be flexibly adjusted.