Machine learning for circuit parameters optimisation and RF signals analysis

Job description: The mentioned circuit is the result of a thesis work, with the report available here: "Intelligent RF System for Ultra Low Power Spectrum Sensing with Machine Learning." Chapter 2 motivates its design, while Chapters 3 and 5 are of direct interest for the internship. Currently, the circuit parameters are set by human expertise, and the algorithm is chosen a posteriori through optimization on data (Chapter 5). This process must be repeated for each new task the circuit can be used for. The goal is to develop a learning algorithm that jointly optimizes the circuit parameters with the mathematical model required to perform the task (detection, classification, etc.) to achieve a more optimal solution (better performance at equal complexity or vice versa). This more generic learning approach should also facilitate the reuse of the circuit for other applications. We have a relatively large database of RF signals for various use cases (WiFi, Bluetooth, drones, etc.) and a relatively simple Python simulator of the circuit for quick familiarization. There is no clearly identified solution in the literature for this problem. As a first step, a simple solution to serve as a benchmark will be to select parameter value samples according to an experimental design and then train the learning model for each sample to determine the most performant one. For the next steps, a considered approach would be to implement reinforcement learning to enable automatic selection of the best circuit parameters and the coefficients of the learning model. Successful results could be valorized through a scientific publication.

Institute description: CEA‑Leti works every day to link micro and nanotechnology research with industrial and consumer applications, all with the aim of improving people's quality of life. Located in Grenoble, Leti employs more than 1,800 top researchers and has offices in the United States and Japan. Within the institute, the Service of Sensor Systems and Electronics for Energy (SSCE) runs a Laboratory of Sensor Signals and Systems (LSSC). The lab focuses on fusing sensor signals, exploiting multimodality through studies in signal processing, information processing, and embedded algorithms. These efforts are especially aimed at context‑capture functions and interaction with the environment from mobile sensor systems.

Your profile: Etudiant.e en dernière année d’école d’ingénieur/master avec une spécialisation en apprentissage machine, optimisation et traitement du signal; des connaissances en radio-fréquences et/ou apprentissage par renforcement sont des atouts. Le(la) candidat(e) devra faire preuve de curiosité, d’autonomie et de persévérance dans son travail car ce sujet est une question de recherche innovante qui n’est a priori pas clairement adressée dans la littérature.

Level of qualification studied: Bac+5 - Master 2

Languages: French Fluent,English Fluent