COMSEMA

Wireless networks are currently witnessing a radical shift from a purely data-oriented architecture to service and intelligent-based architectures, allowing hence the support of a diverse set of verticals. Thanks to the development of AI, future networks are expected to incorporate an even larger set of applications and services such as ReID applications, human activity recognition, interactive hologram, e-health, intelligent humanoid robot, etc. To avoid the huge amount of data transmitted by these services over the wireless networks, the fundamental semantics-blind approach to communication, so far prevalent in today's wireless systems, should be questioned. In fact, thanks to the development of generative AI and LLM-based techniques, it is nowadays possible to extract the semantic representations of an application and to transmit them instead of exchanging the whole data. However, when transmitted over wireless networks, these representations/embeddings are subject to interference, noise, and channel fading. In this project, we consider video interpretation applications and propose a fundamental semantics-approach to redesign the entire process of information generation and transmission in the network. More specifically, we will provide a comprehensive study on the impact of wireless errors on the semantic mismatch/errors and design novel semantic encoding/decoding schemes, that consider the disruptions of the wireless network, going thus beyond current methods that assume an error-free environment between the semantic encoder and decoder. Furthermore, we propose a shift from current bit error-free transmission metrics towards generic tasks related metrics and develop novel wireless transceivers that compresses further the data transmitted over the wireless system. Finally, novel AI-based interference management that focuses on the task achievement at the destination, rather than the bit rate improvement over the air interface, will be investigated.