ED Mathématiques et Informatique
Designing passive hybrid Brain-Computer Interfaces to estimate User eXperience in virtual galleries
by Marc WELTER (LaBRI - Laboratoire Bordelais de Recherche en Informatique)
The defense will take place at 9h30 - Ada Lovelace Centre Inria de l'université de Bordeaux 200, avenue de la Vieille Tour 33405 Talence CEDEX
in front of the jury composed of
- Fabien LOTTE - Directrice de recherche - Inria - Directeur de these
- Frédéric DEHAIS - Professeur des universités - ISAE SUPAERO - Rapporteur
- Maryam ALIMARDANI - Assistant professor - Vrije Universiteit Amsterdam - Rapporteur
- Nicolas ROUGIER - Directeur de recherche - Inria / Institut des Maladies Neurodégénératives Centre Broca Nouvelle-Aquitaine - Examinateur
- Tomasz RUTKOWSKI - Research Scientist - RIKEN AIP - Examinateur
Humans in the 21th century are exposed to an unprecedented amount of aesthetic stimuli, especially in digital spaces such as social media. In such spaces, presentation of aesthetic stimuli, e.g. visual art or music, relies on recommendation systems that require explicit user feedback. However, giving explicit feedback requires cognitive effort that might interrupt and lessen aesthetic experience (AE). Passive Brain-Computer-Interfaces (BCIs), on the other hand, allow implicit and real-time monitoring of such cognitive, affective and conative mental and embodied states in human users. This thesis is part of the European BITSCOPE project that aims to improve user experience by personalizing visual art presentation in virtual art galleries. In particular, the goal of this thesis is to try and design passive hybrid BCIs to estimate user experience in virtual art galleries. User experience in art spaces relates to the experience of art, i.e. the AE. AEs are complex experiences that are composed of various mental states, notably attentional, affective and reward-related states. Because AEs are rewarding, we postulate that markers of good AE also indicate good user experience and that improving AE will consequently improve user experience. Therefore, passive aesthetic appreciation decoding BCIs could allow personalization of art presentation in digital spaces without interruption, and thus, improve user experience. To this end, we first identified established neural correlates of AE and appreciation from the literature. Furthermore, we established an open database with more than 200 highly diverse visual art stimuli to be used in empirical aesthetics research. These stimuli were then used in a classical stimulus presentation and rating experiment while participants wore neuro-physiological recording devices. The data was then analyzed to find neuro-physiological correlates of aesthetic appreciation. Based on these analyses, we implemented classical and State-of-the-Art decoding pipelines to estimate AE from single-trial neuro-physiological data. We started by exploring Electroencephalography (EEG) and physiological data, i.e. Electrodermal Activity (EDA) and Photoplethysmography, independently before testing a multi-modal approach where features from these different modalities are fused together. Although, all decoding models performed worse than random chance level on average, a few models yielded relatively good results for some subjects. EDA features were particularly informative for one participant resulting in the best individual classification with ca. 88% accuracy. In contrast, the highest EEG-based performance was ca. 71% accuracy for the best participant. We observed that AE and its correlates show large variability which makes it difficult to determine optimal decoding pipelines. It seems plausible that more complex decoding methods, e.g. ensemble learning, could improve classification accuracies further. However, we discerned fundamental flaws in the common stimulus presentation and rating paradigm used by us and in the literature. Notably, this kind of experimental design imposes unecological constraints on participants that result in qualitative differences in attention compared to art experiences outside the lab. These flaws challenge the generalization of the results obtained from these protocols to real life applications. Furthermore, these protocols are challenging to adapt to online BCI. Instead, we propose a novel and ecologically valid experimental protocol that can easily be used for online or offline decoding. Moreover, we discuss the challenges of art recommender systems. In particular, we consider the risk of recommendation bias that could result in lessened AEs. Finally, we elaborate possible health and well-being application for AE decoding BCIs, as well as some of the ethical issues arising from this technology.
ED Sciences Physiques et de l'Ingénieur
Design of autonomous IoT systems: contribution of RF technologies.
by Yéro DIA (Laboratoire de l'Intégration du Matériau au Système)
The defense will take place at 14h00 - Amphithéâtre Jean-Paul DOM 351 Cours de la Libération, 33405 Talence, France
in front of the jury composed of
- Valérie VIGNERAS LEFEBVRE - Professeure - Université de Bordeaux - Directeur de these
- Daniela DRAGOMIRESCU - Professeure - Université de Toulouse - Rapporteur
- Serge VERDEYME - Professeur - Université de Limoges - Rapporteur
- Jean-Marie PAILLOT - Professeur - IUT d'Angoulême - Examinateur
- Simon HEMOUR - Maître de conférences - Université de Bordeaux - Examinateur
In a context where the miniaturization of embedded systems and the massive development of the Internet of Things (IoT) are creating increasing energy constraints, this thesis focuses on the study and development of RF energy harvesting systems for the autonomous powering of sensor nodes. Currently, these nodes are still mostly powered by button batteries, which entail both logistical challenges and ecological impact. Ambient energy harvesting thus appears as a promising and relevant solution, especially wireless power transfer using radiofrequency waves, due to their ubiquity in urban environments. An initial study of available energy sources supports the choice of far-field electromagnetic energy (1 μW/cm² in the GSM 900/1800 MHz bands). A literature review is then conducted to present existing technical solutions based on rectennas. These improvements may be implemented upstream (circuit and antenna design) or downstream (communication optimization) of the system. Chapter 2 focuses on the design and evaluation of rectifier circuits based on Schottky diodes. The SMS7630 and HSMS2850 diodes were selected and compared in four-stage rectifiers: the SMS7630 showed a conversion efficiency of 37%, while the HSMS2850 offered better sensitivity (-15.9 dBm). An original contribution lies in the introduction of a resistance compression network (RCR), which linearizes the rectifier's non-linear behavior in response to power variations. Chapter 3 presents the design of two dual-band PDMS flexible antennas aimed at maximizing energy harvesting in the GSM bands. The first, an omnidirectional antenna, powers a low-voltage clock at 17 cm from a smartphone. The second integrates an artificial magnetic conductor (AMC) and achieves higher efficiency (up to 5 dBi), while maintaining stable performance when in contact with the human body. These results demonstrate the feasibility of semi-flexible antennas with strong integration potential on curved or wearable surfaces. The final chapter introduces an original application: powering a very low-consumption RFID tag, composed of a tunnel diode-based oscillator and a four-stage rectifier. This system achieves an activation power of -28.1 dBm and a range of 90 m, surpassing the performance of industrial RFID tags. This thesis brings several original contributions: the implementation of resilient rectifiers using RCR networks, the development of biocompatible flexible antennas adapted for wearable applications, and a functional proof of concept of an ultra-low-power RFID system. These advancements pave the way for battery-free, energy-autonomous sensors, more seamlessly integrated into their environment and aligned with the energy challenges of the IoT.
dynamic safe operating area of GaN HEMT operating under RF conditions :methodology and RF test bench development
by Thomas PALLARO (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
in front of the jury composed of
- Nathalie MALBERT SAYSSET - Professeure des universités - Université de Bordeaux - Directeur de these
- Tristan DUBOIS - Maître de conférences - Université de Bordeaux - CoDirecteur de these
- Jean-Christophe NALLATAMBY - Professeur des universités - Université de Limoges - Examinateur
- Benoit LAMBERT - Ingénieur de recherche - UMS - United Monotithic Semiconductors - Examinateur
- Nathalie LABAT - Professeure des universités - Université de Bordeaux - Examinateur
- Bertrand BOUDARD - Professeur des universités - Université Caen Normandie - Rapporteur
- Olivier LATRY - Professeur des universités - Université Rouen Normandie - Rapporteur
The thesis work falls within the general framework of the evaluation of the performance and reliability of GaN HEMTs and associated integrated circuits in RF operation. The objective of the thesis work is to define a methodology for establishing a dynamic safe operating area under RF operational conditions of GaN HEMTs. In other words, we want to answer the following questions: What would be the maximum charge cycle to ensure optimal and reliable operation of a based HEMT amplifier ? Do the maximum excursions achieved in RF operation address the same degradation mechanisms as in DC operation? Another objective of this work is to establish the link between the waveforms of the signals (voltages and currents) at the terminals of the transistor during RF operation under gain compression, and the amplitude of the RF induced degradations. It will achieved thanks to the background knowledge on the physical degradation mechanisms of AlGaN/GaN HEMTs. The thesis work to be carried out to obtain the dynamic safe operating area will consist of three phases. The first one will focus on the simulation of the waveforms in different classes of operation of the amplifier using the non-linear large signal model of the transistor. The second step will consist in developing, in the environment of the NANOCOM platform of the IMS laboratory, the test bench for waveform measurements in RF operation in order to compare them with simulation results. Indeed, a conventional RF aging bench does not make it possible to record the temporal waveforms of the input and output voltages but only the evolution of the output power as a function of the input power for different impedances presented. The tests are, most of the time, only carried out around an operating point representative of the application. The third stage of the work will be dedicated to accelerated aging tests. The test protocol will include performing RF aging tests under scaled stresses by gradually increasing gain compression. The recovery measurements will include, on the one hand, an I-V characterization in static mode and in pulsed mode of the transistor, on the other hand, a measurement of the output power and an acquisition of the time waveforms. The changes in the static parameters can then be correlated with that of the output power to locate the origin of the physical degradations and assess the limits of the dynamic operating safety area. The deformations of the temporal forms will also be compared to simulations in order to identify the changes in the parameters of the non-linear model of the transistor. This will result in the possibility of determining the impact of degradations of the electrical parameters of the transistor on the performance of the circuits. Ultimately, these results should contribute to improving the prediction of the lifetime of RF circuits.