ED Sciences Physiques et de l'Ingénieur
Contribution to the implementation of Artificial Intelligence algorithms for radiocommunications application in a hardened embedded context
by Antoine SIEBERT (Laboratoire de l'Intégration du Matériau au Système)
The defense will take place at 10h00 - Amphithéâtre Jean-Paul DOM Laboratoire IMS, 351 Cours de la Libération, 33405 Talence Cedex
in front of the jury composed of
- Guillaume FERRE - Professeur des universités - Bordeaux INP - Directeur de these
- Matthieu ARZEL - Professeur des universités - Université de Bretagne Sud - Rapporteur
- Vahid MEGHDADI - Professeur des universités - Université de Limoges - Rapporteur
- Nathalie DELTIMPLE - Professeure des universités - Bordeaux INP - Examinateur
- Bertrand LE GAL - Maître de conférences - INRIA - CoDirecteur de these
- Aurélien FOURNY - Ingénieur - THALES SIX GTS FRANCE - Examinateur
The emergence of new operational requirements in the field of military radio communications is leading to increasing complexity in wireless transmission systems. The disturbances caused to signals by multipath impose constraints on the design of radio receivers, requiring precise and robust signal equalization. This equalization process requires high-performance, reliable propagation channel estimation. Faced with these challenges, the adoption of solutions based on artificial intelligence (AI) is pushing back the limits of traditional algorithmic methods. The integration of AI represents a strategic innovation lever for companies such as Thales. However, in an embedded context subject to severe constraints (energy consumption, computing costs), direct implementation of these AI algorithms is often unfeasible. It then becomes imperative to amend and compress the models to enable their integration into radiocommunication products. The aim of this thesis work is twofold: firstly, to improve the performance of a conventional signal processing algorithm using AI. Secondly, to optimize the model in order to adapt it to the requirements and constraints associated with execution on an embedded platform, while limiting performance degradations.