Optimization and reinforcement learning for emerging multi-antenna wireless networks
File(s)
Author(s)
Feng, Zhenyuan
Type
Thesis
Abstract
Multi-antenna wireless networks such as wireless power networks ( WPN) and wireless
communication networks (WCN) have attracted great attention in academia and
industry. This thesis is concerned with the optimization and reinforcement learning
for emerging multi-antenna wireless networks. Various scenarios and algorithms are
elaborated and evaluated.
First, we design a novel WPN based on the joint optimization of beamforming
and waveform with intelligent reflecting surfaces (IRS). Generalized multi-user and
low-complexity single-user algorithms are demonstrated based on alternating opti-
mization (AO) framework, successive convex approximation ( SCA), and semidefinite
relaxation ( SDR ) to maximize the weighted sum output direct current ( DC). The
energy harvester nonlinearity is explored and two IRS deployment schemes, namely
frequency selective IRS (FS-IRS) and frequency flat IRS (FF-IRS), are modeled and
analyzed. IRS can provide a promising passive beamforming gain on output DC power
over conventional WPT designs and significantly influence the waveform design.
Second, we consider a massive multi-input multi-output (MIMO) WCN under im-
perfect channel state information at the transmitter (CSIT). In order to tackle the
multiuser interference and CSIT imperfections, we develop a multi-agent codebook-
based deep reinforcement learning (DRL) framework where the transmit and receive
beamforming are jointly designed to maximize the average information rate of all users.
Leveraging this deep Q-network ( DQN)-based framework, interference management is
explored and three multi-agent DQN-based schemes, namely the distributed-learning-
distributed-processing scheme, partial-distributed-learning-distributed-processing,
and central-learning-distributed-processing scheme, are proposed and analyzed. This
work demonstrates the inherent effectiveness and robustness of the proposed designs
in the presence of imperfect CSIT...
communication networks (WCN) have attracted great attention in academia and
industry. This thesis is concerned with the optimization and reinforcement learning
for emerging multi-antenna wireless networks. Various scenarios and algorithms are
elaborated and evaluated.
First, we design a novel WPN based on the joint optimization of beamforming
and waveform with intelligent reflecting surfaces (IRS). Generalized multi-user and
low-complexity single-user algorithms are demonstrated based on alternating opti-
mization (AO) framework, successive convex approximation ( SCA), and semidefinite
relaxation ( SDR ) to maximize the weighted sum output direct current ( DC). The
energy harvester nonlinearity is explored and two IRS deployment schemes, namely
frequency selective IRS (FS-IRS) and frequency flat IRS (FF-IRS), are modeled and
analyzed. IRS can provide a promising passive beamforming gain on output DC power
over conventional WPT designs and significantly influence the waveform design.
Second, we consider a massive multi-input multi-output (MIMO) WCN under im-
perfect channel state information at the transmitter (CSIT). In order to tackle the
multiuser interference and CSIT imperfections, we develop a multi-agent codebook-
based deep reinforcement learning (DRL) framework where the transmit and receive
beamforming are jointly designed to maximize the average information rate of all users.
Leveraging this deep Q-network ( DQN)-based framework, interference management is
explored and three multi-agent DQN-based schemes, namely the distributed-learning-
distributed-processing scheme, partial-distributed-learning-distributed-processing,
and central-learning-distributed-processing scheme, are proposed and analyzed. This
work demonstrates the inherent effectiveness and robustness of the proposed designs
in the presence of imperfect CSIT...
Version
Open Access
Date Issued
2024-03-18
Date Awarded
01/08/2024
License URL
Advisor
Clerckx, Bruno
Publisher Department
Electrical and Electronic Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)