Advancing semantic and digital communications through machine learning

Artificial Intelligence (AI) is becoming increasingly prevalent in daily life, driven by rapid advancements in Machine Learning (ML) since 2010. These breakthroughs, enabled by innovations such as specialized hardware, Deep Neural Networks (DNNs), and advances in training techniques, have allowed AI systems to match or even exceed human performance in tasks like autonomous driving and medical diagnostics, with systems such as ChatGPT and AlphaGo standing out as key examples. These achievements have raised public awareness and acceptance of AI technologies.

In the realm of wireless communication, emerging applications such as virtual reality and autonomous systems are pushing traditional digital communication systems to their limits. Conventional content-agnostic approaches struggle to meet the growing demands for bandwidth, power efficiency, and low latency.

This dissertation explores mastering these challenges by integrating ML techniques into wireless communication systems. It introduces CMDNet, a novel framework for symbol detection, designed to improve communication efficiency by combining strengths of traditional model-based designs with those of advanced ML methods. Furthermore, integrating semantic content into communications is identified as crucial for further enhancing system efficiency. Semantic communication aims at transmitting the meaning conveyed by the data rather than the exact bits, which can introduce a model deficit that challenges traditional communication designs. This challenge in design of semantic communication is addressed using advanced ML techniques, as demonstrated in the SINFONY approach.

Together, these contributions demonstrate how ML advances, such as DNNs, can overcome existing limitations in terms of model and algorithmic deficits and significantly enhance the efficiency and capabilities of future communication systems.