Abstract :
[en] Multiuser precoding techniques are critical to handle the co-channel interference, also known as multiuser interference (MUI), in the downlink of multiuser multi-antenna wireless systems. The convention in designing multiuser precoding schemes has been to treat the MUI as an undesired received signal component. Consequently, the design attempts to suppress the MUI by exploiting the channel state information (CSI), regardless of the instantaneous users’ data symbols. In contrast, it has been shown that the MUI may not always be undesired or destructive as it is possible to exploit the constructive part of the interference or even converting the interfering components into constructive interference (CI) by instantaneously exploiting the users’ intended data symbols. As a result, the MUI can be transformed into a useful source of power that constructively contributes to the users’ received signals. This observation has turned the viewpoint on multiuser precoding from conventional approaches towards more sophisticated designs that further exploit the data information (DI) in addition to the CSI, referred to as symbol-level precoding (SLP). The SLP schemes can improve the multiuser system’s overall performance in terms of various metrics, such as power efficiency, symbol error rate, and received signal power. However, such improvement comes with several practical challenges, for example, the need for setting the modulation scheme in advance, increased computational complexity at the transmitter, and sensitivity to CSI and other system uncertainties. The main goal of this thesis is to address these challenges in the design of an SLP scheme.
The existing design formulations for the CI-based SLP problem consider a specific signal constellation; therefore, the design needs to set the modulation scheme in advance. In this thesis, we first elaborate on optimal and relaxed approaches to exploit the CI in a novel systematic way. This study enables us to develop a generic framework for the SLP design problem, which can be used for modulation schemes with constellations of any given shape and order. Depending on the design criterion, the proposed framework can offer significant gains in the power consumption at the transmitter side or the received signal power and the symbol error rate at the receiver side without increasing the complexity, compared to the state-of-the-art schemes. Next, to address the high computational complexity issue, we simplify the design process and propose approximate yet computationally-efficient solutions performing relatively close to the optimal design. We further propose an optimized accelerated FPGA design that allows the real-time implementation of our SLP technique in high-throughput communications systems. Remarkably, the accelerated design enjoys the same per-symbol complexity order as that of the zero-forcing (ZF) precoding scheme. Next, we address the problem of robust SLP design under system uncertainties. In particular, we focus on two sources of uncertainty, namely, the channel and the design process. The related problems are tackled by adopting worst-case and stochastic design approaches and appropriately redefining the precoding optimization problem. The resulting robust schemes can effectively deal with system uncertainties while preserving reliability and power efficiency in the multiuser communications system, at the cost of a slightly increased complexity. Finally, we broaden our scope to new technologies such as millimeter wave (mmWave) communications and massive multiple-input multiple-output (MIMO) systems and revisit the SLP problem for low-cost energy-efficient transmitter architectures. The precoding design problem is more challenging particularly in such scenarios as the related hardware restrictions impose additional (often intractable) constraints on the problem. The restrictions are typically due to the use of finite-resolution analog-to-digital converters (DAC) or analog components such as switches and/or phase shifters. Two well-known design strategies are considered in this thesis, namely, quantized (finite-alphabet) precoding and hybrid analog-digital precoding. We tackle the related problems through adopting efficient design mechanisms and optimization algorithms, which are novel for the SLP schemes. The proposed techniques are shown to improve the system’s energy efficiency compared to the state-of-the-art.
