Massive MIMO and millimeter-wave

Network densification and the use of extremely high frequencies (EHF), commonly known as millimeter-wave band, are the two most promising candidates for the future wireless access to fulfill the ever-increasing capacity demands. It is the small wavelength of millimeter-wave that makes it possible to increase the density of BSs significantly by reducing the footprint of BSs’ deployment sites. In addition, millimeter-wave is what made it possible to use massive MIMO in realistic scenarios. Massive MIMO might sound reasonably new as a topic of research, but as a concept, it has been a hot topic of research since the inception of multiuser information theory. At a glance massive MIMO may sound like a fancy name of multiuser MIMO but the following four key characteristics make it distinct; (i) only the BS learns the channel matrix using uplink pilots; (ii) the number of antennas on a typical BS are greater than the number of users in a cell; (iii) simple signal processing techniques; (iv) it serves multiple users (assuming mutual orthogonality in a practical sense) through spatial multiplexing in time-division duplex mode.

Massive MIMO and Millimeter-Wave: A Natural Wedlock In this section, we want to highlight the strong dependence of massive MIMO and millimeterwave on each other. Even though there is no direct relationship between these two technologies, significant amount of research has been done on massive MIMO mainly focused on conventional cellular frequency bands. While massive MIMO is an option at current cellular bands to provide array and multiplexing gain, it became an unavoidable need or prerequisite to gather enough energy at the millimeter-wave band. Since the area per antenna element is shrunk as the frequency is increased, therefore more antenna elements are required to gather the same amount of signal energy as before. Moreover, small wavelengths of frequencies in the millimeter-wave band made it easier to pack a large number of antenna elements in a compact form. For example, a carrier frequency 𝑓𝑐 = 30 GHz (i.e., 𝜆 = 1 cm), and with 𝜆/2 antenna spacing, more than 180 antennas can be placed on an area as large as a standard credit card as shown in Fig. 1.2. This number can reach up to 1300 at 80 GHz. Therefore both millimeter-wave and massive MIMO are key enablers of the next generation of high-speed wireless networks.

Ultra dense networks

The immense growth in data traffic requires a paradigm shift in all aspects of mobile networks. From the point of view of many experts and industrial consortia in this area, network densification is one of the leading ideas to tackle this challenge. A network can be considered as ultra-dense if there are more cells than active users (Ding, López-Pérez, Mao,Wang & Lin, 2015; López-Pérez, Ding, Claussen & Jafari, 2015). The basic idea is to get the access nodes as close as possible to the end-users. To yield a better idea of the dimension of this class of networks, Ding et al. provided a quantitative measure of the density at which a network can be considered ultra-dense. According to this study, a network is deemed dense if there are more than 103cells/Km2 for 600 active users/Km2 (Ding et al., 2015). The motivation behind this idea is to have access nodes as close as possible to end-users. The practical implementation of network densification can be achieved by the dense deployment of small cells in the hotspots where immense traffic is generated. In general, small cells in ultra dense network (UDN) are fully-functioning BSs that are capable of performing all the functions of a macrocell with a lower power and a smaller coverage area.

Heterogeneous Networks

The key idea behind HetNets is to densely deploy different categories of BSs to increase the network coverage and the performance of cell-edge users. Traditional macrocell BSs (MBS), characterized by high power consumption and high infrastructure costs are added to small cell BS (SBS) comprising, microcell BSs (𝜇BS), pico cell BSs (PBS) and femtocell BSs (FBS) requiring low power, little infrastructure, and low maintenance costs. The idea is that a small number of sparsely deployed MBSs provide umbrella coverage to the cell. In contrast, a high number of densely deployed SBSs, 𝜇BSs, PBSs, and FBSs, that are more closely located to the users, offload the traffic from the MBS increasing the coverage of the cell, the area spectral efficiency and improving the quality of service (QoS) of cell edge users. SBS deployment, along with MBS, is already in the specifications of long term evolution (LTE) for 4G. A key design parameter with respect to the performance of HetNets is the user association algorithm, which determines to which BS tier each user should connect.

Common metrics to asses the performance of a HetNet are: the outage/coverage probability (Dhillon, Ganti, Baccelli & Andrews, 2012; Singh, Dhillon & Andrews, 2013), the spectral efficiency (Hu & Qian, 2014), the energy efficiency (Su, Yang, Xu & Molisch, 2013; Liu, Chen, Chai & Zhang, 2014a; Zhu, Wang & Chen, 2012; Chavarria-Reyes, Akyildiz & Fadel, 2015), QoS (Liu, Chen, Chai & Zhang, 2014b; Liu, Chen, Chai, Zhang & Elkashlan, 2014c), and the fairness (Liao, Hong & Luo, 2014; Bethanabhotla, Bursalioglu, Papadopoulos & Caire, 2016). Typical user association algorithms associate the user to the BS with maximum received power; however, this approach is not the optimal solution in a HetNet scenario, due to disparity between the transmitted power of different network tiers, which results in most of the users associating with the MBS, hence, making ineffective use of the other tiers. A big challenge while designing user association algorithms is that, due to the combinatorial nature of the association problem, the resulting optimization problem is generally NP-hard (Liu, Wang, Chen, Elkashlan, Wong, Schober & Hanzo, 2016), rendering it computationally prohibitive to solve. Furthermore, some of the objectives are conflicting, i.e., energy efficiency and QoS, resulting in performance tradeoffs. HetNets also pose a great resource management challenge. Specially, due to the dense deployment of SBSs, 𝜇BSs, PBSs, and FBSs, the interference between different tiers and among the same tier can be a limiting factor to the overall network performance. In addition to the user association challenge discussed earlier, spectrum sharing strategies are of utmost importance to minimize the interferences.

MIMO Architectures Typically, in conventional MIMO systems, signal processing happens in baseband, and therefore such systems are referred to as fully digital, as shown in Fig.1.7 (Rusek et al., 2013). On the other hand, fully digital massive MIMO systems at least for now, do not sound like a practical idea since fully digital massive MIMO architecture requires a massive number of energy-intensive RF chains, resulting in a significant increase in the implementation cost. Therefore, different alternative massive MIMO architectures have been proposed recently. For example, a fully analog architecture employing only one RF chain, i.e., only one data stream, connected with a massive number of antenna elements using phase shifters to achieve an array gain was proposed (Kim & Lee, 2015) as shown in Fig.1.8. An extension to the fully analog architecture has been proposed recently to support multi-streams (Zhang, Molisch & Kung, 2005; Sudarshan, Mehta, Molisch & Zhang, 2006; Venkateswaran & van der Veen, 2010). This hybrid architecture divides the signal processing of very high dimension matrices into a dimension reduced digital part (i.e., requiring a small number of RF chains) and a large size analog part similar to the fully analog case, as shown in Fig.1.9.

One of the key advantages of the fully analog architecture is its low hardware cost and energy consumption. However, since its analog circuitry can not perfectly adjust the signals, it is not possible to adjust the beams precisely according to channel conditions. Hence, a significant performance loss could occur (Heath et al., 2016), specifically for mobile users. Therefore, the hybrid architecture is a good balance between accuracy, hardware cost, and energy consumption. Moreover, since effective scatterers at millimeter-wave frequencies are always small in number, this leads to low-rank MIMO channel matrix (Rappaport et al., 2013). Hence, typically the number of data streams defines the minimum number of RF chains required for parallel data transmission, this makes hybrid architecture a much better fit. The analog part of the hybrid architecture can be implemented in many different ways, each having its own constraints and limitations (Heath et al., 2016; Méndez-Rial, Rusu, Alkhateeb, González-Prelcic & Heath, 2015; Méndez-Rial, Rusu, González-Prelcic, Alkhateeb & Heath, 2016). This difference in analog architectures affects not only the signal processing techniques but also the maximum theoretically achievable performance of massive MIMO systems. Following is a brief description of the two typical implementations of the analog part of the hybrid architecture.