ML-aided communication network design beyond communicating bits
File(s)
Author(s)
Bian, Chenghong
Type
Thesis
Abstract
The shift toward 6G represents a paradigm change in wireless network design — moving beyond the traditional paradigm of merely transmitting bits. Future communication systems will handle semantically rich data types. Moreover, the integration of radar and communication further exemplifies this transition, as wireless systems increasingly aim to jointly perceive and communicate, rather than just deliver payloads. These trends underscore the necessity of applying machine learning to wireless communication — from efficient multimedia transmission to resolving complex optimization problems that arise in the radar and communication pipeline.
Within this context, this thesis investigates robust multimedia transmission, semantic information delivery over wireless networks, and signal processing for integrated sensing and communication (ISAC) systems, all with the aid of state-of-the-art machine learning techniques.
We formulate multimedia transmission as a joint source-channel coding (JSCC) problem, and design deep learning aided encoding and decoding schemes, where images and point clouds are encoded into latent vectors. To further improve the rate-distortion performance, we represent data sources as neural networks and transmit their weights over the channel.
We then explore JSCC over wireless networks. We propose a novel process-and-forward protocol that can achieve competitive performance over cooperative relay channel. Alternatively, a neural compressor is introduced by treating received symbols as a source to be conveyed, and achieve enhanced performances for image transmission over multiple hops. We also develop a over-the-air edge inference framework for image classification over multi-hop MIMO channels.
Finally, in ISAC, we design superior waveforms through deep learning and develop corresponding signal processing algorithms where an iterative algorithm that combines channel decoding with parameter sensing is proposed for enhanced sensing accuracy and communication reliability under the context of passive sensing. All these advances proposed in this thesis pave the way for future 6G wireless networks.
Within this context, this thesis investigates robust multimedia transmission, semantic information delivery over wireless networks, and signal processing for integrated sensing and communication (ISAC) systems, all with the aid of state-of-the-art machine learning techniques.
We formulate multimedia transmission as a joint source-channel coding (JSCC) problem, and design deep learning aided encoding and decoding schemes, where images and point clouds are encoded into latent vectors. To further improve the rate-distortion performance, we represent data sources as neural networks and transmit their weights over the channel.
We then explore JSCC over wireless networks. We propose a novel process-and-forward protocol that can achieve competitive performance over cooperative relay channel. Alternatively, a neural compressor is introduced by treating received symbols as a source to be conveyed, and achieve enhanced performances for image transmission over multiple hops. We also develop a over-the-air edge inference framework for image classification over multi-hop MIMO channels.
Finally, in ISAC, we design superior waveforms through deep learning and develop corresponding signal processing algorithms where an iterative algorithm that combines channel decoding with parameter sensing is proposed for enhanced sensing accuracy and communication reliability under the context of passive sensing. All these advances proposed in this thesis pave the way for future 6G wireless networks.
Version
Open Access
Date Issued
2025-10-02
Date Awarded
2026-01-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Deniz, Gunduz
Publisher Department
Department of Electrical and Electronic Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)