Description
Date depot: 6 janvier 2022
Titre: Design of an embedded intelligent system for the automatic classification of physiological signals
Directeur de thèse:
Xun ZHANG (LISITE)
Encadrant :
Frédéric AMIEL (LISITE)
Domaine scientifique: Sciences pour l'ingénieur
Thématique CNRS : Systèmes et architectures intégrés matériel-logiciel
Resumé: In the future Brain Human interface concept, one of the important and common problems in EEG analysis is the presence of muscular, ocular, electronic and other types of artefacts. The importance of artefact detection for removal or rejection purposes has been emphasized in the literature.
In classical EEG applications, the signals are either averaged or processed so that artefacts portions are rejected or cancelled out. However, this is impossible for real time applications such as Brain-Computer Interface (BCI) or clinical Neuro-monitoring wherein the signals must be processed at the same time when they are recorded. Blind Source Separation (BSS) is an effective and powerful tool for the separation of EEG signals which are a linear mixture of the original source signals and are corrupted by artefact signals and noise. BSS is almost always the first step involved in the processing of EEG signals for source separation, artefact removal and other purposes. Thus, for the aforementioned real-time applications, it is of prime importance that Blind Source separation is effectively performed in real-time. The motivation to implement BSS in FPGA comes from the hypothesis that the performance of the system could be significantly improved in terms of speed considering the optimal parallelism environment that hardware provides. Though there is no worldwide consensus on the best BSS algorithm, Second Order Blind Identification (SOBI) was chosen as it offers some unique advantages over the other BSS algorithms and also due to the fact that no prior work has been done in the domain of its implementation in FPGA. Few of the blocks of the SOBI algorithm were synthesized and it was concluded that implementation in FPGA indeed reduces the computation time and thus has great potential for real time processing and analysis of EEG signals.
In this context, the subject of the PhD is the design of an embedded intelligent system for the automatic classification of physiological signals. This subject address three main objectives:
• The definition of a generic algorithmic chain for physiological signal processing and automatic classification
• The definition of a generic embedded architecture to implement these algorithmic chains in an FPGA
• The validation of this algorithmic chain and embedded architecture on different applications (to be defined): affective computing, depression or sleep apnea syndrome.
Doctorant.e: Han Xia