Antoine DUPRET

Dr. Antoine DUPRET

orcid

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
3 March 2016 in Sensors
Verified
Journal Article
Open access
PDF file
ABAssim Boukhayma
Assim Boukhayma
ADAntoine DUPRET
Antoine DUPRET
JRJean-Pierre Rostaing
Jean-Pierre Rostaing
CEChristian Enz
Christian Enz
This paper presents the first low noise complementary metal oxide semiconductor (CMOS) terahertz (THz) imager based on source modulation and in-pixel high-Q filtering. The 31 × 31 focal plane array has been fully integrated in a 0 . 13 μ m standard CMOS process. The sensitivity has been improved significantly by modulating the active THz source that lights the scene and performing on-chip high-Q filtering. Each pixel encompass a broadband bow tie antenna coupled to an N-type metal-oxide-semiconductor (NMOS) detector that shifts the THz radiation, a low noise adjustable gain amplifier and a high-Q filter centered at the modulation frequency. The filter is based on a passive switched-capacitor (SC) N-path filter combined with a continuous-time broad-band Gm-C filter. A simplified analysis that helps in designing and tuning the passive SC N-path filter is provided. The characterization of the readout chain shows that a Q factor of 100 has been achieved for the filter with a good matching between the analytical calculation and the measurement results. An input-referred noise of 0 . 2 μ V RMS has been measured. Characterization of the chip with different THz wavelengths confirms the broadband feature of the antenna and shows that this THz imager reaches a total noise equivalent power of 0 . 6 nW at 270 GHz and 0 . 8 nW at 600 GHz.

30 July 2019 in IEEE Instrumentation & Measurement Magazine
Verified
Journal Article
Closed access
TATzila Ajamian
Tzila Ajamian
SMSaïd Moussaoui
Saïd Moussaoui
ADAntoine DUPRET
Antoine DUPRET
Reflectometry is a wire diagnosis technique to detect and localize electrical defects in Network Under Tests efficiently. In order to achieve a very precise online diagnosis without altering the functioning of a Network Under Tests, reflectometry methods should use specific wideband Multi-Carrier signals, whose generation requires dedicated numerical tools. The basic architecture of any reflectometry system for the injection of the signal in the Network Under Tests requires appropriate Digital-to-Analog Converters. Even for the most demanding requirements in reflectometry, their specifications in terms of resolution and sampling rate can easily be found. Yet, measuring the reflected signal should be performed using Analog-to-Digital Converters capable of reaching high sampling frequencies together with sufficient resolution. Such converters are either extremely expensive or beyond the current state of the art. In that respect, this paper presents a new architectural approach for designing such reflectometry systems based on a Compressive Sampling method bypassing the Nyquist rate.

13 March 2025 in arXiv
Verified
Preprint
Open access
PDF file
YBYanis Basso-Bert
Yanis Basso-Bert
AMAnca Molnos
Anca Molnos
RLRomain Lemaire
Romain Lemaire
WGWilliam Guicquero
William Guicquero
ADAntoine DUPRET
Antoine DUPRET
In dynamic environments where new concepts continuously emerge, Deep Neural Networks (DNNs) must adapt by learning new classes while retaining previously acquired ones. This challenge is addressed by Class-Incremental Learning (CIL). This paper introduces Generative Binary Memory (GBM), a novel CIL pseudo-replay approach which generates synthetic binary pseudo-exemplars. Relying on Bernoulli Mixture Models (BMMs), GBM effectively models the multi-modal characteristics of class distributions, in a latent, binary space. With a specifically-designed feature binarizer, our approach applies to any conventional DNN. GBM also natively supports Binary Neural Networks (BNNs) for highly-constrained model sizes in embedded systems. The experimental results demonstrate that GBM achieves higher than state-of-the-art average accuracy on CIFAR100 (+2.9%) and TinyImageNet (+1.5%) for a ResNet-18 equipped with our binarizer. GBM also outperforms emerging CIL methods for BNNs, with +3.1% in final accuracy and x4.7 memory reduction, on CORE50.