Opportunistic Beamforming and Asymptotic Throughput Analysis of Hybrid Analog-Digital mmWave Multi-User MIMO Systems

Abstract

In this thesis, we address the problem of large training/feedback overhead and the requirement of computationally intensive algorithms to determine the phase angles of the analog precoder for downlink data transmission in hybrid analog-digital (A-D) multi-user (MU) millimeter wave (mmWave) systems. We investigate the use of opportunistic beamforming (OBF) using a dumb analog precoder as a solution to these issues. The OBF based schemes work as follows. The BS transmits a known pilot symbol over a random analog precoding vector. Using this, all the users in the system estimate their effective channels, and the best user efficiently feeds back its SNR to the BS, e.g., using a timer-based scheme. Then, the BS schedules the best user using the chosen analog precoding vector. Since the randomly chosen precoding vector is likely to be optimal to a subset of the users, and since the best user is selected in each time slot, deep fades at any given user are avoided, and the overall system throughput improves. Note, also, that this scheme requires very little feedback and no optimization of the analog precoding vector is necessary. This thesis has two parts. In the first part, we consider a single radio frequency (RF) chain at the base station (BS). We analyze two schemes of OBF, called fully random precoder (full RP) and channel structure-aware random precoder (CSA-RP). In the full RP scheme, we use random phase angles across the antenna array. In the CSA-RP scheme, we consider the use of structured random beams, where the phase angles of the analog precoder are chosen so as to have the same structure as the array response of the geometric channel model of mmWave systems. For the CSA-RP scheme, we derive the asymptotic scaling laws of the average throughput as a function of the number of users, number of antennas, and the SNR, using extreme value theory, as the number of users gets large. Our simulation results show that the second scheme achieves near-optimal throughput, i.e., close to that achieved using coherent beamforming with best user selection (called the maximum rate baseline (max-rate BL) scheme, since the user obtaining the maximum rate is scheduled to be served), via opportunistic selection among fewer users in the system compared to the first scheme. Next, we use M pilot symbols with M different random orthonormal analog precoding vectors instead of a single pilot symbol, to further reduce the number of users required to obtain near-optimal throughput. We derive the scaling law of the average throughput for OBF using M random orthonormal precoding vectors also. In the second part, we consider the case where the BS is equipped with multiple RF chains. We propose two schemes (called greedy high and greedy low) with OBF to simultaneously serve multiple users, with very low feedback overhead. The BS randomly generates an analog precoding matrix whose columns form a set of random orthonormal precoding vectors, and transmits pilot symbols using the precoding matrix. Each user measures the SINR for each of the random precoding vectors. The two schemes we consider differ in terms of how many beams are assigned to the users in each round of feedback: the greedy high scheme entails a larger number of rounds of feedback compared to the greedy low scheme, but is purportedly better in its performance. Through simulations, we find that the two schemes offer nearly identical throughputs, and thus conclude that the greedy low scheme with lower feedback overhead is more attractive for practical implementation. Furthermore, we derive the average throughput scaling laws using extreme value theory when many users exist in the system for both schemes. The results in this thesis show that as long as there are a reasonable number of users in the system, OBF is an attractive, low-complexity, and low-feedback approach for practical implementation of MU mmWave MIMO communications