Back to search

Aristotle University Of Thessaloniki (AUTH)

Development of large-scale optimization techniques and algorithms for use in wireless communication networks

Abstract

dc:description

Wireless communication networks have evolved through successive generations, progressing from 1G to 5G and the imminent 6G. Throughout this evolution, mathematical optimization has been a fundamental tool for theoretical design and research of wireless networks. The novel characteristics and demands of each generation have led to a fundamental shif tin the structure of the underlying optimization problems, necessitating the development of novel solution techniques. Convex optimization, for instance, has been extensively utilized, and it has been proven to be adequate in addressing the majority of the challenges posed by the 3G research. Nonetheless, during the 4G research era, distributed optimization methods garnered attention due to their ability to leverage network densification and coordinate multiple base stations or network nodes, there by augmenting the available computing power. Concurrently, machine learning (ML) techniques have recently begun to complement or totally replace some traditional optimization methods which aim to address more complex problems. Finally, in the ongoin research context of 6G, quantum computing is being explored as a novel approach to address problems stemming from classical and quantum communications. In this regard, the second chapter of this thesis presents an over-the-air computing(AirComp) framework that significanlty reduces the communication latency in constrained distributed convex optimization problems. Specifically, the optimization procedure utilizesa distributed primal-dual (DPD) subgradient method. Under general assumptions, we prove that the DPD-AirComp method can asymptotically achieve zero expected constraint violation. Thus, DPD-AirComp ensures the feasibility of the original problem despite channel fading and additive noise. Furthermore, proper power control of the users’ signalscan mitigate the expected non-zero optimality gap. We present two practical applications of the proposed framework: smart grid management and wireless resource allocation. Finally, numerical results confirm the excellent performance of DPD-AirComp, which converges an order of magnitude faster than two digital orthogonal multiple access schemes: time-division multiple access (TDMA) and orthogonal frequency-division multiple access(OFDMA). In Chapter 3, we demonstrate how ML methods can serve as scalable optimization solvers, appropriate at solving the high-dimensional, non-convex design challenges that have surfaced with the 5G networks and the ongoing research for the 6G. Specifically, in the first part of this Chapter, we propose a novel peak-to-average power ratio (PAPR)-aware adaptive modulation scheme for Orthogonal Frequency Division Multiplexing (OFDM). The aim of the proposed PAPR-aware protocol is to maximize the energy efficiency ofthe OFDM transmission, where the energy efficiency is defined as the ratio of the OFDM throughput to the PA consumption, and is thus strongly affected by PAPR. For this purpose, we formulate the energy efficiency maximization problem with respect to powerand adaptive modulation constraints per subcarrier. Due to the joint optimization of power allocation, adaptive modulation, and the non-convex terms introduced by PAPR, the problem cannot be solved by using standard convex optimization methods. Therefore, a solution based on online deep learning is proposed. Numerical results showcase that the energy efficiency of the proposed PAPR-aware protocol is greater up to 3-dB compared to the conventional PAPR-unaware protocol, while the PAPR is also reduced by 3-dB. In the second part of Chapter 3, a multi-task learning approach is studied, towards a more general resource allocation paradigm. Specifically, the optimal solution to anoptimization problem depends on the problem’s objective function, constraints, and size. While deep neural networks (DNNs) have proven effective in solving optimization problems, changes in the problem’s size, objectives, or constraints often require adjustments to the DNN architecture to maintain effectiveness, or even retraining a new DNN from scratch. Given the nature of wireless networks, which involves multiple and diverse objectives that can have conflicting requirements and constraints, we propose a multi-task learning (MTL) framework to enable a single DNN to jointly solve a range of diverse optimization problems. In this framework, optimization problems with varying dimensionality values, objectives, and constraints are treated as distinct tasks. To jointly address these tasks, we propose aconditional computation-based MTL approach with routing. The multi-task DNN consists of two components, the base DNN (bDNN), which is the single DNN used to extract the solutions for all considered optimization problems, and the routing DNN (rDNN), which manages which nodes and layers of the bDNN to be used during the forward propagation of each task. The output of the rDNN is multiplied with all bDNN’s weights during the forward propagation, creating a unique computational path through the bDNN for each task. This setup allows the tasks to either share parameters or use independent ones, with the decision controlled by the rDNN. The proposed framework supports both supervised and unsupervised learning scenarios. Based on this framework, a soft modularization and a hard parameter sharing approach of lower complexity are proposed. The numerical results demonstrate the efficiency of the proposed MTL approaches compared to several multi-task benchmarks and the single-task DNN approach. Chapter 4 studies the optimal beamforming of a multiple-input multiple-output(MIMO) system, with b-bit quantized pre- and post-processing vectors. Specifically, the Chapter explores the use of the Quantum Approximate Optimization Algorithm (QAOA) and alternating optimization to address the problem of b-bit quantized phase shifters both at the transmitter and the receiver. We demonstrate that the structure of this quantized beamforming problem aligns naturally with hybrid-classical methods like QAOA, as the phase shifts used in beamforming can be directly mapped to rotation gates in a quantum circuit. Notably, this work is the first to show a theoretical connection between quantum circuits and phase shifts in pre/post-coding, which could have broader applications in other fields, such as integrated sensing and communication. The analysis for the b-bit caseis presented, along with a warm-start QAOA approach to improve computational efficiency. Numerical results highlight the effectiveness of the proposed method in achieving optimal quantized beamforming and the performance gain of the proposed warm-start approach.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Mitsiou, Nikolaos
  • Μήτσιου, Νικόλαος

Subjects

dc:subject × 14

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/60485

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Mitsiou, Nikolaos; Μήτσιου, Νικόλαος. Development of large-scale optimization techniques and algorithms for use in wireless communication networks. Aristotle University Of Thessaloniki (AUTH), 2025. http://hdl.handle.net/10442/hedi/60485