I. Introduction
As the deployment of 5G is the first step where people are approaching a new world in terms of Internet of Everything (IoE), the data traffic in future communications will explosively grow up to 1000-fold in the next decade. To meet this demand, there are several emerging technologies attracting most attention, including massive multiple-input multiple-output (MIMO) [1], millimeter wave (mmWave) communications [2] [3], Non-orthogonal Multiple Access (NOMA) [4], etc. Among them, massive MIMO and millimeter wave communication are viewed as the most promising solutions and have been frequently combined together to complement each other in the past few years [5]. The unique feature of millimeter wave communication is that it is able to provide tens of GHz bandwidth by exploring higher unoccupied frequency resources (30 – 300 GHz) rather than improving the spectrum efficiency with the current bandwidth. However, millimeter wave communication suffers from the severe signal attenuation induced by such high carrier frequency [2]. On the other side, massive MIMO is able to provide great power gain by the large number of antennas (e.g., hundreds to thousands of antennas) at the Base Station (BS) compared to the number of users in the same cell [1]. Even with low-complexity algorithms (e.g., Minimum Mean Square Error (MMSE) for channel estimation and Matched Filter (MF) for data detection), massive MIMO can still improve energy and spectral efficiencies to a great extent through beamforming and spatial multiplexing [6] [7]. To this end, it turns out to be a good choice to employ massive MIMO to provide enough power gain to compensate for the orders-of-magnitude more path loss in millimeter wave communication systems.