Abstract
Driven by the ever-increasing data demand, future wireless networks are required to achieve a 1000× increment in mobile data throughput. As the amount of suitable spectrum for wireless communication is limited, a common consensus to attain the tar-get is through spatial densification, i.e., deploy more antennas per unit area. Various re-searches have been performed and approaches have been proposed, where in general, they can be categorized under two classes: Massive multiple-input-multiple-output (MIMO) system and heterogeneous network (HetNet). The idea of massive MIMO is to aggressively expand the antenna arrays at macro base stations (MBS), if the spacing between any two antenna elements is larger than certain distance – usually termed as coherence distance – the diversity gain can be boosted up linearly with the antenna number. Under such circustom, not only the transmit power can be greatly reduced, but also the fluctuated small scale fading channel becomes flattern, which enable net-work designers to use simple transmission strategy to achieve good performance in multi-user scenario. On the other hand, HetNets are created under another philoso-phy: By successively overlaying the conventionally big MBSs with densely deployed lower power small access points, HetNet tends to offload UE traffic from macro cells to small cells. In this fashion, it reduces not only the communication distance between transmitter and receiver, but also the number of UEs that are competing resource in each cell, thus improves the data rate as well as spectrum reused. While each of these technologies has great advantage even stands alone, we wonder how the “big” MIMO and “small” cells can co-exist with each other, and whether a happy marriage between they can bring along further improvement to the wireless network. To that end, in this thesis we study several advanced methods to enhance the physical layer performance of massive MIMO system, and its combination with HetNet. To begin with, we propose a linear precoding scheme for massive MIMO cellular networks, referred to as Cell-Edge Aware Zero Forcing (CEA-ZF) precoding, that exploits certain degrees of freedom from the massive antenna array to suppress interference towards the most vulnerable UEs located at the cell edge. In order to quantify the performance of our proposed precoding scheme, we combine random matrix theory with stochastic geometry to derive tractable expressions for the coverage probability under both CEA-ZF and the conventional Cell-Edge Unaware Zero Forcing (CEU-ZF), which uses all the spatial dimensions for the multiplexing gain. Our analysis is able to capture all the key features of wireless network, including the BS topology, interference, and estimation error in obtaining the channel state information (CSI). The accuracy of our analysis is confirmed via simulation. Based on the analytical result, we compare the CEA-ZF with CEU-ZF in terms of coverage probability, sum per cell rate, and 95%-likely, or equivalently edge, rate. The numerical analysis shows positive gain from CEA-ZF, where the gain of 95%-likely rate is remarkably large, indicating CEA-ZF is a better candidate for networks with massive MIMO BSs. We then investigate the energy efficiency of a MIMO HetNet where macro BSs con-nect with small APs via wireless backhaul. We first derive achievable rates for both uplink and downlink transmissions in macro cell, small cell, and wireless backhaul, and then obtain tractable expression for the network energy efficiency. We show that under spatial multiplexing, the energy efficiency of a HetNet is sensitive to the net-work load, and it should be taken into account when controlling the number of users served by each base station. We confirm that a two-tier HetNet with wireless backhaul can be significantly more energy efficient than a one-tier cellular network. However, to achieve this goal, it also requires the bandwidth division between radio access links and wireless backhaul to be optimally set according to the load conditions. We further extend our study to analyze a two tier HetNet underlaid with energy harvesting (EH) device-to-device (D2D) communication, where UEs can harvest and store radio frequency (RF) energy radiated from nearby BSs, and use the harvested energy to help relaying information from one BS to its intended UE. In particular, we develop an analytical framework for the design of such HetNet by introducing the energy harvesting region (EHR) and modeling the status of harvested energy using Markov chain. We derive the spatial distribution of UE relays, and propose a transmis-sion mode selection scheme including the efficient relay selection method. The network outage probability is derived in close form. Based on our analysis results, we explore the effects of network parameters on the outage probability and determine the optimal offloading bias in terms of the outage probability. Particularly, we show that having a high EH efficiency enhances the performance of EH-D2D underlaid HetNet, but can also degrade the performance, especially for dense network. We finally turn our attention to the analysis of packet throughput in a small cell network that has temporal traffic dynamics. Specifically, we introduce Bernoulli pro-cess to model the traffic dynamic, and we investigate the throughput performance of static time division multiplex (S-TDD) and dynamic TDD transmissions. By leveraging stochastic geometry and queuing theory, we derive closed-form expressions for the UL and DL packet throughput, also capturing the impact of random traffic arrivals and packet retransmissions. Through our analysis, which is validated via simulations, we confirm that i) the number of scheduled UEs plays a critical role in throughput perfor-mance, and ii) D-TDD outperforms S-TDD in DL, with the vice versa occurring in UL, since asymmetric transmissions reduce DL interference at the expense of an increased UL interference. We also find that in asymmetric scenarios, where most of the traffic is in DL, D-TDD provides a DL packet throughput gain by better controlling the queuing delay, and that such gain vanishes in the light-traffic regime. Moreover, when the SIR threshold is very small, and the DL time portion of S-TDD is configured to be the same as D-TDD, then D-TDD outperforms S-TDD in both uplink and downlink.