Publication list

EBEN: Extreme bandwidth extension network applied to speech signals captured with noise-resilient microphones

In this paper, we present Extreme Bandwidth Extension Network (EBEN), a generative adversarial network (GAN) that enhances audio measured with noise-resilient microphones. This type of capture equipment suppresses ambient noise at the expense of speech bandwidth, thereby requiring signal enhancement techniques to recover the wideband speech signal. EBEN leverages a multiband decomposition of the raw captured speech to decrease the data time-domain dimensions, and give better control over the full-band signal. This multiband representation is fed to a U-Net-like model, which adopts a combination of feature and adversarial losses to recover an enhanced audio signal. We also benefit from this original representation in the proposed discriminator architecture. Our approach can achieve state-of-the-art results with a lightweight generator and real-time compatible operation.

Deeplomatics project: Multimodal UAV detection, localisation and identification for site surveillance

The Deeplomatics project aims at detecting, localizing and identifying low sound level acoustic sources generated during UAV intrusion on sensitive areas. To do that, a multimodal approach has been chosen. Acoustical sensors coupled to an AI processor running a neural network developed specifically for simultaneous detecion, localization and identification tasks allow for the analysis of the incoming acoustic waves 40 times per second. A data fusion software handles the information sent by all the sensors deployed on the surveyed area and computes the estimated position of the threat in a geographic referential. This position is then sent to an optical sensor mounted on a motorized touret that handles multiple wavelength cameras, including large and narrow field of view visible cameras, a thermal camera, and a Short wave Infrared range gated active imaging system. A second deep neural netword then focuses on detecting and identifying drones in the image in order to confirm the intrusion of a UAV on the surveyed area. A constant communication between the acoustical sensors, the data fusion software and the cameras allows for real-time detection, localization and identification of UAVs 5 times per second. The system has been deployed on Baldersheim’s proving ground in May 2022. This presentation gives an overview of this field experiment where 5 acoustic sensors and one optical sensor were connected to the data fusion software. 14 scenarios representing more than 95 minutes of UAV flights have been analyzed in real time, leading to a median radial position estimation error of 10.7 meters and a standard deviation of 15 meters. The data fusion process also lead to a significant enhancement of the identification rate of the incoming drone with 95% correct estimation in the presence of drone models that were in the training data base.

Deeplomatics: A deep-learning based multimodal approach for aerial drone detection and localization

Protection against illicit drone intrusions is a matter of great concern. The relative stealthy nature of UAVs makes their detection difficult. To address this issue, the Deeplomatics project provides a multimodal and modular approach, which combines the advantages of different systems, while adapting to various topologies of the areas to be secured. The originality lies in the fact that acoustic and optronic devices feed independent AI to simultaneously localize and identify the targets using both spatial audio and visual signatures. Several microphone arrays are deployed on the area to be protected. Within its coverage area (about 15 hectares), each microphone array simultaneously localizes and identifies flying drones using a deep learning approach based on the BeamLearning network. Each array is attached to a local AI which processes spatial audio measurements in realtime (40 estimations per second), independently to the other units of the surveillance network. A data fusion system refines the estimates provided by each of the AI-enhanced microphone arrays. This detected position is shared in real-time with an optronic system. Once this system has hooked its target, a Deep Learning tracking algorithm is used to allow an autonomous visual tracking and identification. The optronic system is composed of various cameras (visible, thermal, and active imaging) mounted on a servo-turret. The active imaging system can capture scenes up to 1 km, and only captures objects within a given distance, which naturally excludes foreground and background from the image, and enhances the capabilities of computer vision. The Deeplomatics project combines benefits from acoustics and optronics to ensure real-time localization and identification of drones, with a high precision (less than 7° of absolute 3D error, more than 90 % detection accuracy). The modular approach also allows to consider in the long term the addition of new capture systems such as electromagnetic radars.

Deeplomatics: Acoustic Localization and recognition of drones

The Deeplomatics project aims at developing an anti-drone surveillance system based on the use of Deep Learning techniques using audio and video sensors. The system consists of several microphonic antennas, each with its own embedded artificial intelligence. Each antenna is able to detect, recognize and localize in real time the intrusion of a drone. After merging this information, an optronic system is directed towards the target, to confirm the intrusion thanks to different cameras, which also have their own artificial intelligence. The deep neural network used for the acoustics is a variation of the BeamLearning architecture proposed by the authors, which allows to jointly estimate in real time the position and the nature of the source. Each antenna can monitor an area of 100 hectares, within which the angular estimation error is less than 3◦, and the recognition rate exceeds 85%. A confidence criterion is also included in the cost function, which allows during inference to estimate the relevance of each of the 40 estimates per second. This project totals several tens of hours of drone flight recordings, obtained thanks to several microphone antennas adapted to a 3rd to 5th order ambisonic encoding. The interest of this encoding is to be able to restore these recordings in the laboratory, thanks to a 3D spatialization system, controlled by the GPS positions recorded on site. The data can then be re-recorded on any microphone antenna, and to train the deep neural network for antennas not present during the initial measurement campaigns. Thanks to this approach, it is also easy to augment the data (rotation of the sound scene, addition of realistic acoustic environment…), to obtain, in fine, an accurate and robust system in real environment.

Deep Learning enhancement of speech captured with in-ear transducers

This research project was born from a collaboration between the LMSSC team (Laboratory of Mechanics of Structures and Coupled Systems) of the CNAM and the APC team (Acoustics and Protection of the Combatant) of the ISL. Its objective is to improve the intelligibility of speech captured by an in-ear microphone developed by the ISL. This non-conventional device, coupled with an active hearing protection, allows to capture the vocal signals emitted by a speaker by eliminating all external noise. However, the acoustic path between the mouth and the transducers is responsible for a total loss of information above 2 kHz. At low frequencies, a slight amplification as well as physiological noises are observed. We are therefore faced with a problem of reconstruction of a signal absent at high frequencies and denoising. Deep learning methods will be used for the reconstruction of high frequencies instead of the source-filter model which is not able to restore missing information. A first phase of analysis of the captured signals is necessary to model the degradation and observe its variability. The design of a consequent database is then made possible with a digital filtering simulating the observed deteriorations. In order to increase the richness of this database and to avoid any over-learning phenomenon, a random component will be introduced in the filtering. The design of deep neural networks is now possible for the regeneration of the emitted signal from the degraded signal. A broad exploration on the architecture of the networks, the cost functions used and the learning strategies will be undertaken. The final objective is to integrate an inference network on a programming board for real-time processing. Particular attention will be paid to the size of the network and the processing time on this type of lightweight and low power consuming architecture.

3D acoustic source localization using Deep Learning : training procedure using a higher order ambisonic spatialization process.

Previous works have proven the relevance of using the Deep Learning approach for acoustic source localization, and in particular the efficiency of the approach based on the BeamLearning deep neural network architecture that we have developed. However, the constitution of a dataset remains one of the cornerstones of the supervised learning paradigm. If the constitution of a simulated database is possible, only simple antenna geometries can be used. Indeed, the 3D diffraction phenomena by the antenna body are generally difficult to take into account by simulation. To fully capture the pressure field interactions with the antenna, it is however possible to create a database from real measurements, which can be a time consuming task and requires a priori a large number of different environments. To overcome these drawbacks, the laboratory has a 3D spatializer, which we use to create various data sets for training networks dedicated to source localization tasks. This 5th order ambisonic restitution system can be controlled in real time and offers a perfectly controlled and fully automated synthesis of a large number of source positions. After careful calibration of the spa- tialization system, a large database is built to prove that these data are relevant to train the BeamLearning architecture, and to localize a 3D acoustic source with an angular error surpassing the 3D localization performances of the SH-MUSIC algorithm, while offering a much higher angular estimation rate. We will also show that a training performed on data coming from measurements of a real antenna allows to include in the procedure an implicit calibration of the microphones composing the antenna, which greatly simplifies the problems of calibration of this type of sensors.

BeamLearning: an end-to-end Deep Learning approach for the angular localization of sound sources using raw multichannel acoustic pressure data

Sound sources localization using multichannel signal processing has been a subject of active research for decades. In recent years, the use of deep learning in audio signal processing has allowed to drastically improve performances for machine hearing. This has motivated the scientific community to also develop machine learning strategies for source localization applications. In this paper, we present BeamLearning, a multi-resolution deep learning approach that allows to encode relevant information contained in unprocessed time domain acoustic signals captured by microphone arrays. The use of raw data aims at avoiding simplifying hypothesis that most traditional model-based localization methods rely on. Benefits of its use are shown for realtime sound source 2D-localization tasks in reverberating and noisy environments. Since supervised machine learning approaches require large-sized, physically realistic, precisely labelled datasets, we also developed a fast GPU-based computation of room impulse responses using fractional delays for image source models. A thorough analysis of the network representation and extensive performance tests are carried out using the BeamLearning network with synthetic and experimental datasets. Obtained results demonstrate that the BeamLearning approach significantly outperforms the wideband MUSIC and SRP-PHAT methods in terms of localization accuracy and computational efficiency in presence of heavy measurement noise and reverberation.

Outdoor field trials for the measurement of the acoustic signals of mini UAVs

Acoustic detection and tracking of UAVs is considered by means of Unattended Ground Sensors equipped with microphonic sensors. Experimental campaigns were conducted with flying drones (DJI, Parrot…) in an anechoic chamber and in countryside. The acoustic database includes various scenario such as hovering flight, translation flight, etc. At the same time, “disturbing noises” have been recorded: ambient noises including birds, insects, people speaking, detonations and fire shot noises have been recorded to feed our database. A part of the recorded database has been used to train a classifier (learning phase). Then another part of the dataset was used to estimate the F-score to evaluate both the precision and recall of the classifier. Adding artificial noise to the data, and selecting acoustic features with evolutionary programming enabled the detection of an unknown drone in an unknown soundscape within 200 meters with the JRip classifier (Fscore of 0.88 for distances between 0 and 100 m, and 0.56 between 100 and 200 m). Main results obtained during the signature analysis and the classifier assessment will be presented and the perspectives in terms of performance improvement with the use of MEMS multi-microphones array.

3D Sound source localization in reverberant rooms using Deep Learning and microphone arrays : Simulation and experiments

Acoustic source localization is a well-studied topic in array signal processing , which could benefit from the emergence of data inference tools. We present our recent developments on the use of a Deep neural network, BeamLearning [2, 3], fed with raw multichannel audio for 3D sound source localization in reverberating environments. The data driven approaches allow to avoid the simplifying assumptions that most traditional localization methods incorporate. However, for an efficient training process, supervised machine learning algorithms rely on precisely labeled large sized datasets. There is therefore a critical need to generate a large number of 3D audio data recorded by microphone arrays in various environments. When the dataset is simulated either with numerical models or with 3D soundfield synthesis, the physical validity is also critical. Therefore, an efficient tensor GPU-based computation of synthetic room impulse responses based on fractional delays for image source models is used. We also present the use of physical 3D soundfield synthesis [1] for the learning process on microphone arrays. We discuss the advantages of this reproducible and semi-automated process, which allows to deal with arbitrary array geometries. We also analyze the localization performances of the BeamLearning approach fed with this dataset, which allows a 3D precision as high as 5 degrees in a reverberant room.

2D/3D acoustic source localization using Deep Learning techniques and arbitrary microphone array configurations

Protecting sensitive sites from drone threats requires an accurate strategy for drones detection and localization. This is why the Deeplomatics project proposes a combination of acoustic detection and localization on compact microphone arrays, and optical recognition using active imaging to monitor these threats. Rather than finding the direction of arrival of the source using an acoustic source localization algorithm based on a propagation model (such as the MUSIC method), an artificial intelligence approach, named the BeamLearnin, has been specifically designed to determine in real time the position of an acoustic source, directly from raw microphones signals. The optimization of the neural network is done using data obtained either from numerical simulations or from multi-channel microphone recordings. One of the advantages of the second method is to optimize the learning variables on real signals that integrate all the characteristics of the antenna used for the recordings, from the frequency responses of the sensors to the diffraction of the array body. In this case, the learning phase also allows an intrinsic calibration of the microphone array. These measured datasets can be recorded from any compact microphone array. Indeed, during the learning process, the arrays are positioned in the center of a sphere of loudspeakers restoring a perfectly known pressure field thanks to the formalism of spherical harmonic. This formalism also makes it possible to spatialize acoustic sources from digitally calculated signals, but also to restore a pressure field measured during UAV flights in real environments. Thus a single on-site measurement campaign can be used to build several datasets corresponding to different microphone array placed in the spatialization sphere. In addition to being able to localize the position of a source in real time, the proposed BeamLearning approach offers good results and exhibits both a better average experimental localization of the BeamLearning approach, but also a lower dispersion of localization errors than SH-MUSIC. On the GPU architecture used for source localization using the BeamLearning method, the computation time required to estimate the source DOA is also in favor of our approach, with a reduction of the computation time by a factor of 75 when compared to the SH-MUSIC method.

Timescalenet: a multiresolution approch for raw audio recognition

In recent years, the use of Deep Learning techniques in audio signal processing has led the scientific community to develop machine learning strategies that allow to build efficient representations from raw waveforms for machine hearing tasks. In the present paper, we show the benefit of a multi-resolution approach : TimeScaleNet aims at learning an efficient representation of a sound, by learning time dependencies both at the sample level and at the frame level. At the sample level, TimeScaleNet’s architecture introduces a new form of recurrent neural layer that acts as a learnable passband biquadratic digital IIR filterbank and self-adapts to the specific recognition task and dataset, with a large receptive field and very few learnable parameters. The obtained frame-level feature map is then processed using a residual network of depthwise separable atrous convolutions. This second scale of analysis allows to encode the time fluctuations at the frame timescale, in different learnt pooled frequency bands. In the present paper, TimeScaleNet is tested using the Speech Commands Dataset. We report a very high mean accuracy of 94.87±0.24% (macro averaged F1-score : 94.9 ± 0.24%) for this particular task.

TimeScaleNet : a Multiresolution Approach for Raw Audio Recognition using Learnable Biquadratic IIR Filters and Residual Networks of Depthwise-Separable One-Dimensional Atrous Convolutions

In the present paper, we show the benefit of a multi-resolution approach that allows to encode the relevant information contained in unprocessed time domain acoustic signals. TimeScaleNet aims at learning an efficient representation of a sound, by learning time dependencies both at the sample level and at the frame level. The proposed approach allows to improve the interpretability of the learning scheme, by unifying advanced deep learning and signal processing techniques. In particular, TimeScaleNet’s architecture introduces a new form of recurrent neural layer, which is directly inspired from digital IIR signal processing. This layer acts as a learnable passband biquadratic digital IIR filterbank. The learnable filterbank allows to build a time-frequency-like feature map that self-adapts to the specific recognition task and dataset, with a large receptive field and very few learnable parameters. The obtained frame-level feature map is then processed using a residual network of depthwise separable atrous convolutions. This second scale of analysis aims at efficiently encoding relationships between the time fluctuations at the frame timescale, in different learnt pooled frequency bands, in the range of [20 ms ; 200 ms]. TimeScaleNet is tested both using the Speech Commands Dataset and the ESC-10 Dataset. We report a very high mean accuracy of 94.87 ± 0.24% (macro averaged F1-score : 94.9 ± 0.24%) for speech recognition, and a rather moderate accuracy of 69.71 ± 1.91% (macro averaged F1-score : 70.14 ± 1.57%) for the environmental sound classification task.

Source localization in reverberant rooms using Deep Learning and microphone arrays

Sound sources localization (SSL) is a subject of active research in the field of multi-channel signal processing since many years, and could benefit from the emergence of data-driven approaches. In the present paper, we present our recent developments on the use of a deep neural network, fed with raw multichannel audio in order to achieve sound source localization in reverberating and noisy environments. This paradigm allows to avoid the simplifying assumptions that most traditional localization methods incorporate using source models and propagating models. However, for an efficient training process, supervised machine learning algorithms rely on large-sized and precisely labelled datasets. There is therefore a critical need to generate a large number of audio data recorded by microphone arrays in various environments. When the dataset is built either with numerical simulations or with experimental 3D soundfield synthesis, the physical validity is also critical. We therefore present an efficient tensor GPU-based computation of synthetic room impulse responses using fractional delays for image source models, and analyze the localization performances of the proposed neural network fed with this dataset, which allows a significant improvement in terms of SSL accuracy over the traditional MUSIC and SRP-PHAT methods.

Méthodes Temporelles en Acoustique : Réseaux de transducteurs -Retournement temporel - Problèmes inverses -Deep Learning

Ce mémoire d’habilitation à diriger des recherches fournit une vue d’ensemble sur mes travaux, depuis mon recrutement en tant que Maître de Conférences en Acoustique au Conservatoire National des Arts et Métiers, en 2009. Les différents axes de recherches sur lesquels je me suis concentré depuis cette date ont pour fil conducteur l’utilisation de méthodes temporelles multicanales, pour le traitement des données associées aux réseaux de transducteurs. J’ai développé ces méthodes pour les appliquer à la résolution de problèmes inverses en environnement industriel, à l'élastographie du corps humain pour le diagnostic médical, à la localisation de snipers et de drones, ou encore pour proposer des stratégies d’apprentissage profond appliquées à la localisation de sources sonores, et à la reconnaissance de parole ou de sons environnementaux. Pour chacune de ces applications, je fournis dans ce document une description des méthodes proposées, ainsi qu’une synthèse des résultats les plus importants obtenus. La cohérence et les liens entre les différents projets sont mises en exergue, et les chapitres sont systématiquement illustrés de résultats numériques et expérimentaux. Le développement de dispositifs miniaturisés pour la synthèse de champs sonores et la captation est également mise en avant, puisque c’est l’une des caractéristiques de mes travaux de recherche ces dernières années. En fin de document, je propose également six propositions concrètes de recherches pour les années à venir, permettant ainsi d’exposer ma vision à court et moyen-terme des développements qui pourraient être réalisés sur la base de mes travaux.

A linear phase IIR filterbank for the radial filters of ambisonic recordings

Higher order Ambisonics decomposition of natural sound fields is often performed using spherical, rigid microphone array measurements, mainly because of its simple implementation[1-2]. All the electronic equipment can be conveniently placed inside the spherical measurement array, without affecting the scattered acoustic field. However, restitution systems for HOA sound field synthesis generally exhibit a much larger radius than measurement arrays. The well-known “bass-boost” effect is directly linked to this size discrepancy: low frequencies have to be amplified, especially for higher order components of the Ambisonics decomposition. The dynamic range for filtering purposes is limited, mainly by the signal-to-noise ratio of the microphone array. In order to overcome this problem, we developed two microphone array prototypes using analogic MEMS microphones, which have become a viable solution in a small packaging and with a reasonable price thanks to the growing use of these sensors in domotics and in the mobile phone industry. MEMS microphones from the same production batch exhibit very similar characteristics and can be used for array signal processing without any level or phase calibration. The two proposed prototypes are made of group of 4 MEMS microphones for the same sensor position to improve the signal-to-noise ratio by 6 dB. The first prototype is a 5-th order Ambisonics system (50 sensors – 200 mems – lie on a Lebedev grid) and the second prototype is a Mixed Order Ambisonics (MOA) system (42 sensors – 168 mems – 3-th order in 3D and 11-th order in 2D, 24 sensors in the equator of the sphere) [3].Nevertheless, this approach does not dispense from the need to filter higher order coefficients. A simple high-pass filtering on each order component is not sufficient, since this would not only cause losses in terms of amplitude and power but also would affect the loudness of the restitution. A filter bank is therefore needed to cut-off noise amplification at low frequencies and apply appropriate gains for loudness equalization. Baumgartner and al [4] proposed a non-linear phase filter bank based on Linkwitz-Riley IIR filters. In order to avoid group delay distortions, Zotter proposed a linear phase filter bank based on FIR filters and the use of fast block convolution [5]. This solution is although not very flexible, since the FIR strongly depend on the radius of the measurement array and on the filter bank’s cut-off frequencies. Any change in the measurement system require a new computation of each FIR filters corresponding. In the present paper, a linear phase IIR filter bank is implemented. Thanks to the use of local overlap and add time reversal blocks [6], the filter bank exhibits a linear phase delay which only depends on the time reversal blocksize. The proposed implementation of the filter bank allows to change in real-time the frequency bands and loudness equalization (diffuse or free field equalization) using of Faust programming language [7].

[1] S. Bertet, J. Daniel, E. Parizet, L. Gros, and O. Warusfel, “Investigation of the perceived spatial resolution of higher order ambisonics sound fields: a subjective evaluation involving virtual and real 3D microphones”, AES 30th International Conference, Saariselkä, Finland, 2007 March 15–17.
[2] J. Meyer and G. Elko, “A highly scalable spherical microphone array based on an orthonormal decomposition of the soundfield,” in Acoustics, Speech, and Signal Processing, 2002. Proceedings.(ICASSP’02). IEEE International Conference on, vol. 2, Orlando, FL, USA, 2002.
[3] S. Favrot, M. Marschall , J. Käsbach , J. Buchholz, T. Weller, “Mixed-order Ambisonics recording and playback for improving horizontal directionality”, presented at the AES 131st convention, New York, USA, 2011.
[4] R. Baumgartner, H. Pomberger, and M. Frank, “Practical Implementation of Radial Filters for Ambisonic Recordings”, in Proc. first International Conference on Spatial Audio, Detmold, Germany, 2011.
[5] F. Zotter, “A Linear-Phase Filter-Bank Approach to Process Rigid Spherical Microphone Array Recordings”, Proceedings of Papers – 5th International Conference on Electrical, Electronic and Computing Engineering, IcETRAN 2018, Palić, Serbia, June 11 – 14, 2018
[6] S.R Powell and P.M Chau, “A technique for realizing linear phase IIR filter”, IEEE transactions and signal processing, vol 29(11), november 1991, 2425-2435.
[7] For Faust programming language, see https://faust.grame.fr

Source localization and identification with a compact array of digital mems microphones

A compact microphone array was developed for source localization and identification. This planar array consists of an arrangement of 32 digital MEMS microphones, concentrated in an aperture of fewer than 10 centimeters, and connected to a computer by Ethernet (AVB protocol). 3D direction of arrival (DOA) localization is performed using the pressure and the particle velocity estimated at the center of the array. The pressure is estimated by averaging the signals of multiple microphones. We compare high order pressure finite differences to the Phase and Amplitude Gradient Estimation (PAGE) method for particle velocity estimation. This paper also aims at presenting a method for UAV detection using the developed sensor and supervised binary classification.

Contrôle actif décentralisé de transparence acoustique

Conception d' antennes microphoniques à base de MEMS analogiques ou numériques : Retour d' expérience

Antennes non calibrées, suivi métrologique et problèmes inverses : Une approche par Deep Learning

Antennes microphoniques intelligentes : Localisation de sources par Deep Learning

Imagerie acoustique (stationnaire et temporelle), audio 3D, rayonnement d' instruments de musique

Axis retrieval of a supersonic source in a reverberant space using time reversal

Localizing the axis of the Mach cone created by the supersonic displacement of a bullet in a reverberant environment is a challenging task, not only because of the high velocity of the moving source, but also because of the multiple wave reflections off of the walls. Although time reversal (TR) techniques allow static acoustic source localization in a re-verberant space, they have not been explored yet on non stationary waves caused by supersonic displacements in urban canyons. The acoustic wave produced by a supersonic projectile has a conical wavefront and a N-shaped acoustic pressure signature. In this paper, this acoustic wave is reproduced using a line array of point-like sources (simula-tions) and loudspeakers (experiments). During the propagation of this conical wave in an urban canyon, the resulting pressure signals are measured using a time reversal array flush mounted into the ground. These acoustic signals allow to automatically retrieve with a high accuracy the location of the Mach cone axis using time reversal techniques. This inverse problem is solved using the maximization of a fourth-order statistical criterion of the backpropagated pressures. This criterion allows to estimate the intersections between the Mach cone axis and several vertical planes in the urban canyon. These estimations are then fitted to a 3D trajectory with a robust three dimensional interpolation technique based on the Random Sample Consensus (RANSAC) algorithm. This method allows to automatically retrieve the axis of the supersonic source with an angular accuracy of less than 0.5° and a misdistance of 0.5 cm for both numerical simulations and experimental measurements.

A distributed network of compact microphone arrays for drone detection and tracking

This work focuses on the development of a distributed network of compact microphone arrays for unmanned aerial vehicle (UAV) detection and tracking. Each compact microphone array extends in a 10 cm length aperture and consists in an arrangement of digital MEMS microphones. Several arrays are connected to a computing substation using the I2S, ADAT and MADI protocols using optical fiber. These protocols used together allow to collect the signals from hundreds of microphones spread over a distance of up to 10 km. Sound source localization is performed on each array using measured pressure and particle velocities. The pressure is estimated by averaging the signals of multiple microphones, and the particle velocity is estimated with high order finite differences of microphone signals. Multiple calibration procedures are compared experimentally. Results in sound source localization, noise reduction by spatial filtering and UAV recognition using machine learning are presented.

Source localization using a compact differential microphone array, application to drone tracking

Noise reduction on a compact microphone array, application to drone detection

Le retournement temporel en milieu réverbérant pour localiser une source supersonique

Localiser l’axe du cône de Mach causé par le déplacement supersonique d’un objet balistique dans un environnement urbain réverbérant représente un défi scientifique et technique considérable, notamment en raison des réflexions multiples sur les murs. Le retournement temporel rend possible la localisation de sources en milieu réverbérant mais n’a pas encoré eté testé lorsque les sources sont en mouvement supersonique. Le présent article expose lapremì ere approche d’un telprobì eme dans le cadre d’un modèle géométrique simple, reproduisant un espace réverbérant constitué de deux murs et d’un sol d’impédance infinie. Le principe de Huyghens permet de synthétiser un cône de Mach par superposition de fronts d’ondes monopolaires. Dans le cadre de cetté etude, le principe est utilisé pour des simulations numériques et pour une validation expérimentale en laboratoire, où une ligne de haut-parleursémetparleurs´parleursémet le cône de Mach artificiel. Une méthode faisant appel aux fonctions de Green des sources images modélise la réverbération pour la propagation directe et la rétro-propagation par renversement du temps, grâcè a un réseau de microphones déployé sur le sol. Un calcul numérique rétro-propage ensuite les données mesurées jusque dans des tranches verticales intersectant l’axe du cône de Mach. La maximisation d’un critère statistique d’ordre 4, qui supprime les forts niveaux dus aux microphones, détermine le point d’intersection en question. Une méthode de tri permet de garder les meilleures estimations servantàservant`servantà l’interpolation géométrique de l’axe. La méthode proposée présente une précision angulaire de 1 • et une distance entre les axes de 1 cm, à la fois sur les simulations et les mesures expérimentales.

Détection, classification et suivi de trajectoire de sources acoustiques par captation pression-vitesse sur capteurs MEMS numériques

L’utilisation de drones aériens est en plein essor, et la surveillance contre une utilisation inappropriée de ces appareils est un sujet de préoccupation majeure. Dans une stratégie multimodale acoustique et optronique de détection et de suivi de trajectoire par fusion de données, l’attention est ici portée au sous-système acoustique en cours de développement. Le dispositif acoustique est un ensemble d’antennes compactes (diamètre < 10 cm) et autonomes, mises en réseau afin de couvrir une zoné etendue de surveillance. Chaque unité du réseau est constituée de 10 microphones MEMS numériques permettant de mesurer demanì ere optimisée la pression et les composantes du vecteur de vitesse particulaire sur une large gamme de fréquence. Nous présentons ici les contraintes matérielles de cette approche, et les traitements réalisés pour chaque unité du réseau. Pour augmenter la robustesse de l’approche, nouscompì eterons la localisation de la source mobile par uné etape de détection et de classification de signature acoustique. Pour cela, un apprentissage sera effectuéeffectué`effectuéà partir d’une base de données de signatures acoustiques pré-enregistrées. Une fois la source détectée, l’algorithme proposé permet de réaliser un suivi de sa trajectoire, dans plusieurs sous-bandes de fréquences adaptées auxécartsaux´auxécarts inter-microphoniques et aux caractéristiques du signal. Il est fait usage d’une approche par analyse en composantes principales dans le domaine temporel. Des résultats de la localisation en présence d’une source sont présentés, ainsi que des pistes de développement pour une localisation en présence de sources concurrentes, et d’amélioration du suivi de trajectoire par filtrage particulaire et fusion de données.

Imagerie acoustique instationnaire par retournement temporel en environnement complexe

Le processus d’imagerie par retournement temporel est une technique largement utilisée pour réaliser de la localisation et caractérisation de source acoustique. Dans ce papier, différentes améliorations sont proposées afin que l’efficacité et la précision de cette technique d’imagerie instationnaire soient indépendantes des conditions de mesure. Les différentes améliorations proposées ont été développées en tirant avan-tage de l’utilisation d’une antenne de mesure hémisphérique double-couche, permettant l’enregistrement simultanée du champ de pression et de sa dérivée normale. Le pro-cessus d’imagerie par retournement temporel a été optimisé dans le but de reconstruire un champ de pression avec une grande precision temps-espace et une qualité de haute résolution. Une étude expérimentale menée en chambre réverbérante et bruité (faible rapport signal-à-bruit) met en évidence la capacité de la méthode proposée pour rétro-propager, avec une grande précision temps-espace, un champ de pression rayonné par une source acoustique instationnaire présente dans un volume délimité par une surface de mesure.

Hemispherical double-layer time reversal imaging in reverberant and noisy environments at audible frequencies

Time reversal is a widely used technique in wave physics, for both imaging purposes and experimental focusing. In this paper, a complete double-layer time reversal imaging process is proposed for in situ acoustic characterization of non-stationary sources, with perturbative noise sources and reverberation. The proposed method involves the use of a hemispherical array composed of pressure-pressure probes. The complete set of underlying optimizations to sonic time reversal imaging is detailed, with regard to space and time reconstruction accuracy, imaging resolution and sensitivity to reverberation, and perturbative noise. The proposed technique is tested and compared to more conventional time reversal techniques through numerical simulations and experiments. Results demonstrate the ability of the proposed method to back-propagate acoustic waves radiated from non-stationary sources in the volume delimited by the measurement array with a high precision both in time and space domains. Analysis of the results also shows that the process can successfully be applied in strongly reverberant environments, even with poor signal-to-noise ratio.

Synthesis of a Mach cone using a speaker array

The interest of the authors concerns sniper detection using time-reversal techniques on the Mach cone in a reverberant urban environment. In order to setup a safe and reproducible experimental framework at a reduced scale, it is possible to synthesize a N-wave with a conical geometry by means of loudspeakers disposed along a hypothetical ring axis. The supersonic nature of the simulated displacement leads to a set of constraints, both on spatial and temporal samplings, correlated to the structure of the medium and to the digital sampling of the N-shaped signal. Those constraints are theoretically studied to ensure reconstruction of the conical wavefront. A rst experiment has been realized, that allowed the synthesis of a Mach wave using 15 speakers spaced by 4.36 cm. Taking into account the directivity of each speaker and the diraction eects due to the line array, the symmetry of revolution of the cone is studied. Since the loudspeakers are in their linear regime, nonlinear behaviors of the wave are no longer present. However, inverse ltering methods are possible for improving the quality of the signal. We show that it is possible to visualize the spatio-temporal evolution of the pressure eld in planes containing the ring axis using a linear microphone array mounted on a translation robot. Comparisons between experiments and simulations show encouraging results for the following. PACS no. 43.28.We, 43.28.Mw

Electroacoustique sur le web : un retour d'expérience

Attack transient exploration on sopranino recorder with time-domain Near-Field Acoustic Holography method

Directivity of musical instruments have been studied for a long time. In the last decades, there has been a growing interest in imaging methods for the characterization and localization of sound sources. These developments are of great help to study the stationary and transient radiation behaviour of woodwind instruments. The time-domain Near-Field Acoustic Holography is one of these powerful imaging methods. NAH allows to separate the sources contributions from the different parts of an instrument. One of the advantage of the time-domain holography is to observe acoustic phenomena during transitory states. Most studies on recorder acoustic radiation were conducted during steady state, yet none of them focus on the attack transients. In order to investigate the acoustic radiation of the sopranino recorder, we use a semi-cylindrical microphone array, thus taking advantage of the geometrical symmetry of the recorder. In this study, the imaging method is used for the localization and the characterization of radiating sources on a sopranino recorder played by a performer. The array consists in a 4 angular and 11 longitudinal microphone arrangement. Taking into account the symmetry, the number of measurements points is 2 × 11 × 4 = 88. The recorder is located at the center of the cylindrical array. This study aims at highlighting an acoustic coupling between finger holes, labium and bell during the attack transient. This experiment allows us to validate the experimental protocol: semi-cylindrical array and time domain acoustic holography method for woodwind instruments radiation investigations.

Acoustic imaging in confined and noisy environments using double layer Time Reversal and Field Separation Methods

Many imaging methods cannot localize precisely unstationary sources in confined and noisy environments. In this paper, the use of a Time Reversal acoustic sink (TRS) method is proposed, in conjunction with a Field Separation Method (FSM). The proposed time reversal (TR) process is based on the measurement of the sound pressure field and its normal derivative on a double layer hemispherical antenna, which bounds the region of interest (ROI). These data are time-reversed and numerically back-propagated to a surface, 0.5 cm away from the source plane. As most imaging methods, the efficiency of this process relies on the use of the most suitable Green functions, which depend on the propagating environment. A way to improve the TR process is to transform numerically the confined space problem into a free field case, for which the Green functions are well-known. The proposed FSM consists in expanding the measured fields on the spherical harmonics functions, thus allowing to compute the outgoing waves. This process allows a precise localization and characterization of the source placed under the antenna, using free-field Green functions. Thanks to this method, the influence of reverberation and acoustic fields radiated by sources outside the ROI can be suppressed. The measurements presented in this paper are performed in an anechoic room, using two acoustic sources. The first one to image in the ROI emits a filtered pulse and the second one, placed outside the ROI, is driven by a Gaussian white noise. In order to assess the reconstruction quality of the proposed imaging process, a reference field is measured in an anechoic room on the back-propagation surface, corresponding to the pressure values when the source laying in the ROI is radiating alone. Comparisons with back-propagated pressures using TRS in conjunction with FSM show a good accuracy both in space and time domains.

Evaluation of a separation method for source identification in small spaces

This paper investigates the efficiency of a field separation method for the identification of sound sources in small and non-anechoic spaces. When performing measurements in such environments, the acquired data contain information from the direct field radiated by the source of interest and reflections from walls. To get rid of the unwanted contributions and assess the field radiated by the source of interest, a field separation method is used. Acoustic data (pressure or velocity) are then measured on a hemispheric array whose base is laying on the surface of interest. Then, by using spherical harmonic expansions, contributions from outgoing and incoming waves can be separated if the impedance of the tested surface is high enough. Depending on the probe type, different implementations of the separation method are numerically compared. In addition, the influence of the walls’ reflection coefficient is studied. Finally, measurements are performed using an array made-up of 36 p-p probes. Results obtained in a car trunk mock-up with controlled sources are first presented before reporting results measured in a real car running on a roller bench.

Time-Reversal Imaging and Field-Separation-Method applied to the study of the Steelpan radiation

Time reversal (TR) is a powerful method for the imaging and localization of sound sources. We propose the use of TR imaging in conjunction with a field separation method (FSM) in order to study the radiating sources of a Steelpan. The TR-FSM method consists in measuring the acoustic pressure on a double-layer hemispherical time reversal mirror (TRM). Outgoing waves are separated from ingoing waves by using spherical-harmonic expansions. The outgoing contribution is then time-reversed and numerically backpropagated, allowing to achieve accurate imaging of acoustically radiating sources. This FSM also allows to separate contributions from noise source outside the region of interest, thus separating the contributions from zones on the instrument. In this study, this imaging method is used for the imaging of radiating sources on a Steelpan played by a performer. The results show that in some situations, the main contribution to the radiation comes from zones that were not mechanically excited by the steelpanist.

Source identification in small spaces using field separation method: application to a car trunk

Acoustic holography is a powerful tool for the localization and ranking of sound sources. However when dealing with non-anechoic spaces, classical methods have to be modified to take into account reflections on the walls of the testing room and, if necessary, the field radiated by secondary sources. In this paper, the field separation method is used to overcome these problems. The method used here consists in measuring the acoustic pressure on a double layer half-sphere array which base is lying on the surface of interest. Then, by using spherical harmonic expansions, contributions from outgoing and incoming waves can be separated if the impedance of the tested surface is high enough. Simulations on simple configurations and measurements on a car trunk mock-up are first presented. For the measured cases, the double layer array used is made-up of 2*36 carefully calibrated microphones. Comparison with results obtained with double layer SONAH are also shown. Finally, results obtained with a real car on a roller bench are reported.

Sonic Time Reversal Imaging optimization in reverberating, confined or noisy environments

Time reversal (TR) is a powerful method for the imaging and localization of sound sources. Classical TR is based on the recording of the pressure field on a time reversal mirror (TRM), followed by a numerical backpropagation of the time-reversed signals in a simulated propagation environment. To achieve accurate imaging, Green functions (GF) describing the environment must be well-known. When dealing with reverberating environments, precise numerical backpropagation is a rather complicated problem to solve. In order to avoid this situation, we propose a field separation method (FSM) in order to recover data that would be measured on the TRM in free-space, corresponding to the well-known free-field GF. This method consists in measuring the acoustic pressure on a double-layer hemispherical TRM. Outgoing waves are separated from ingoing waves by using spherical-harmonic expansions. The outgoing contribution is then time-reversed and numerically backpropagated using the free-field GF, allowing to achieve accurate imaging. This FSM also allows to separate contributions from sources outside the region of interest. This new method is illustrated by simulations and measurements in a car-trunk mockup and in a reverberating room. Imaging resolution will be discussed using several TR schemes, taking advantage of the double layer p-p measurements.

Quantitative elastography of renal transplants using supersonic shear imaging: a pilot study

Purpose To evaluate the reliability of quantitative ultrasonic measurement of renal allograft elasticity using supersonic shear imaging (SSI) and its relationship with parenchymal pathological changes. Materials and methods Forty-three kidney transplant recipients (22 women, 21 men) (mean age, 51 years; age range, 18-70 years) underwent SSI elastography, followed by biopsy. The quantitative measurements of cortical elasticity were performed by two radiologists and expressed in terms of Young’s modulus (kPa). Intra-and inter-observer reproduc-ibility was assessed (Kruskal-Wallis test and Bland-Altman analysis), as well as the correlation between elasticity values and clinical, biological and pathological data (semi-quantitative Banff scoring). Interstitial fibrosis was evaluated semi-quantitatively by the Banff score and measured by quantitative image analysis. Results Intra-and inter-observer variation coefficients of cortical elasticity were 20 % and 12 %, respectively. Renal cortical stiffness did not correlate with any clinical parameters, any single semi-quantitative Banff score or the level of interstitial fibrosis; however, a significant correlation was observed between cortical stiffness and the total Banff scores of chronic lesions and of all elementary lesions (R00.34, P00.05 and R00.41, P00.03, respectively).

Characterization of non-stationary sources using three imaging techniques

Over the last decade, different imaging techniques have been developed to characterize and localize non-stationary acoustic sources. This study focuses on three of them: Time Domain Holography, Real-Time Near-Field Acoustic Holography and Time Reversal Acoustic Imaging. In the first part of the paper, the principles of these three methods are reminded and the validation technique is presented. Then, the second part deals with the comparison of the results obtained with these three methods for two different kinds of sources: point-like sources (small loudspeakers) and a plate-like acoustic source emission. Advantages and draw-backs of the methods are discussed. Finally, an industrial case is studied: set-ups, measurements and analysis of non-stationary sound sources imaging are presented.

Noninvasive In Vivo Liver Fibrosis Evaluation Using Supersonic Shear Imaging: A Clinical Study on 113 Hepatitis C Virus Patients

Supersonic shear imaging (SSI) has recently been demonstrated to be a repeatable and reproducible transient bidimensional elastography technique. We report a prospective clinical evaluation of the performances of SSI for liver fibrosis evaluation in 113 patients with hepatitis C virus (HCV) and a comparison with FibroScan (FS). Liver elasticity values using SSI and FS ranged from 4.50 kPa to 33.96 kPa and from 2.60 kPa to 46.50 kPa, respectively. Analysis of variance (ANOVA) shows a good agreement between fibrosis staging and elasticity assessment using SSI and FS (p , 10 25). The areas under receiver operating characteristic (ROC) curves for elasticity values assessed from SSI were 0.948, 0.962 and 0.968 for patients with predicted fibrosis levels F $ 2, F $ 3 and F 5 4, respectively. These values are compared with FS area under the receiver operating characteristic curve (AUROC) of 0.846, 0.857 and 0.940, respectively. This comparison between ROC curves is particularly significant for mild and intermediate fibrosis levels. SSI appears to be a fast, simple and reliable method for noninvasive liver fibrosis evaluation.

Imagerie de sources acoustiques par renversement du temps - Optimisations et stratégies de dépassement de limites, du théorique à l' expérimental

Techniques d'imagerie à haute résolution de sources actives par retournement temporel dans le domaine audible

La problématique de l’imagerie et de la localisation de sources acoustiques actives dans le domaine audible trouve un ancrage dans des applications aussi variées que l’acoustique industrielle, l’acoustique des instruments de musique, et l’acoustique sous-marine. Dans le domaine des fréquences audibles, l’utilisation de techniques à haute résolution est cruciale afin de détecter et caractériser des sources acoustiques, dont les tailles caractéristiques sont bien souvent plus petites que la demi-longueur d’onde. De nombreuses techniques de formations de voies, de traitement d’antennes, et de traitement de signal ont été développées pour répondre à cette problématique. Les techniques de retournement temporel (RT), initialement développées dans le domaine ultrasonore, ouvrent un large champ d’investigation pour l’imagerie de sources acoustiques audibles. Ces méthodes doivent être raffinées afin d’atteindre un régime d’imagerie à haute résolution dans le domaine audible. Nous présenterons une vue d’ensemble des optimisations nécessaires dans le domaine audible. En particulier, nous développerons en détail les principes et l’implémentation de l’imagerie de sources actives par puits à retournement temporel numérique. Cette technique d’imagerie est dérivée des travaux de Rosny et Fink sur la focalisation à haute résolution par puits à RT . Nous avons récemment démontré que cette technique nouvelle d’imagerie de sources par puits à RT numérique permet d’obtenir des résultats extrêmement satisfaisants pour l’imagerie de sources acoustiques en milieu anéchoïque. EN milieu réverbérant, le RT fonctionne de manière optimale et permet de minimiser le nombre de capteurs pour la focalisation En revanche, dans le cas de milieux confinés et/ou réverbérants, les techniques d’imagerie par puits à RT numérique échouent, puisqu’ils sont basés sur une rétropropagation simulée numériquement, impossible à réaliser précisément dans le cas du régime stochastique en environnement réverbérant. Nous discuterons de solutions d’imagerie dérivées du RT en milieu confiné permettant de dépasser cette limite d’utilisation.

Supersonic shear imaging is a new potent morphological non-invasive technique to assess of liver fibrosis. Part 2 : comparison with fibroscan

Real time quantitative elastography using Supersonic Shear wave Imaging

Supersonic Shear Imaging (SSI) is a quantitative stiffness imaging technique based on the combination of a radiation force induced in tissue by an ultrasonic beam and ultrafast ultrasound imaging sequence (up to more than 10000 frames per second) catching in real time the propagation of the resulting shear waves. Local shear wave speed is estimated and enables the two dimensional mapping of shear elasticity. This imaging modality is implemented on conventional probes driven by dedicated ultrafast echographic devices and can be performed during a standard ultrasound exam. The clinical potential of SSI is today extensively investigated for many potential applications such as breast cancer diagnosis, liver fibrosis staging, cardiovascular applications, ophthalmology. This invited lecture presents a short overview of the current investigated applications of SSI.

Quantification non-invasive du degré de Fibrose Hépatique par Élastographie Dynamique (Supersonic Shear Imaging)

Il a été récemment démontré que lélastographie par pression de radiation (Supersonic Shear Imaging - SSI) est capable de quantifier les propriétés viscoélastiques du foie chez lhomme. Cette technique développée à linstitut Langevin a démontré sa bonne reproductibilité, répétabilité et son indépendance vis à vis de la respiration du sujet. La technique SSI est basée sur lutilisation in vivo de la pression de radiation ultrasonore, qui permet d’induire une onde de cisaillement profondément dans les tissus. La propagation de cette onde est alors imagée grâce à un échographe ultrarapide (5000 images/secondes). Lestimation de la vitesse de londe en fonction de la fréquence donne alors accès aux paramètres viscoélastiques des tissus. Cette étude a permis de construire de solides bases pour élaborer un protocole clinique de qualité. Dans le travail qui suit, la technique SSI, adaptée à une sonde courbe (C42 2.5 MHz), est utilisée pour cartographier entre les côtes les propriétés viscoélastiques de 142 patients atteints de maladies hépatiques chroniques (VHC, VHB, cirrhose connue). Sur ces mêmes patients un examen Fibroscan (élastographie transitoire monodimensionelle) ainsi quune prise de sang sont réalisés. Cette dernière permettant de déterminer le niveau de fibrose des patients et leur score METAVIR (Fib4, FORNs, APRI). Les paramètres viscoélastiques du foie sont déterminés sur une large bande fréquentielle et cartographiés sur une surface de 120x75 mm². Cette étude montre que la technique SSI permet de discriminer de manière quantitative et avec une bonne différentiation les niveaux de fibrose des patients (p-index ~=.10-16 ). La technique permet de s’affranchir des problèmes d’hétérogénéités affectant les techniques d'élastographie 1D. Cette étude suggère ainsi que la technique SSI pourrait devenir un examen de routine complémentaire précis et fiable pour la détermination du niveau de fibrose hépatique et permettrait de diagnostiquer plus rapidement les pathologies hépatiques chroniques.

Evaluation de deux méthodes d'imagerie acoustique en milieu bruité

L’amélioration du confort acoustique dans les habitacles des véhicules de transport individuels ou collectifs constitue une préoccupation de plus en plus importante pour les industriels des domaines ferroviaire et routier. La méthode de l’holographie acoustique permet d’explorer le champ proche des sources acoustiques ; l’objectif principal est de localiser et quantifier les sources de bruit. Dans ce cadre, deux méthodes sont étudiées ici : SONAH (Statistically Optimal Nearfield Acoustical Holography) et FSM (Field Separation Method). La méthode SONAH présente l’avantage de réaliser la propagation du champ non pas dans le domaine des nombres d’ondes mais dans le domaine spatial directement. Ceci permet d'éviter les problèmes d’effet de troncatures liés à l’utilisation des transformées de Fourier bidimensionnelle spatiale. Ainsi, la propagation s’effectue par le biais d’une matrice de transfert définie à l’aide de méthodes statistiques de telle façon que toutes les ondes propagatives et qu’une partie des ondes évanescentes soient projetées avec une exactitude optimale. Des simulations ont été réalisées avec un monopôle afin de calculer les erreurs relatives de rétropropagation de la pression et de la vitesse acoustique. Un deuxième monopôle est introduit par la suite dans la simulation pour représenter une source perturbatrice et étudier la robustesse de la méthode SONAH dans le cas d’un milieu bruité. La méthode FSM présente, dans le cas d’un milieu confiné, l’avantage de séparer les champs convergents et divergents sur une surface entourant la source. Elle est basée sur une formulation intégrale couplée à la méthode de décomposition en harmoniques sphériques des champs rayonnés, ainsi les réflexions des parois sont soustraites des mesures et les conditions de mesures en champ libre sont rétablies. Pour chacune des deux méthodes, les erreurs relatives par rapport au champ théorique sont calculées. Les temps de calculs et les domaines d’application seront également discutés.

A new potent morphological non-invasive technique to assess liver Fibrosis : Part I, Technical feasibility

A new potent morphological non-invasive predictor of liver fibrosis staging by supersonic shear imaging : Clinical study

NON-INVASIVE LIVER FIBROSIS STAGING USING SUPERSONIC SHEAR IMAGING : A CLINICAL STUDY ON 150 PATIENTS

High-Resolution Imaging of Sound Sources in Free Field Using a Numerical Time-Reversal Sink

Numerous practical applications – such as non destructive evaluation of industrial structures, acoustic characterization of musical instruments, and acoustic mapping of sound sources in a known propagation medium – involve source detection and characterization. In the past, this problem has been investigated using different beamforming and backpropagation methods. In this work, a new technique, based on the time reversal sink concept, is used to detect active sound sources with a limited number of measurement points. The theory and application of super-resolution focusing of sound and vibration using a time-reversal sink (TRS) have been studied, both in ultrasonic regime and in audible range. A high-resolution imaging technique based on a numerical time reversal sink has recently been developed by the authors for vibrational imaging of active sources in a dispersive medium. In this paper, the numerical time reversal sink imaging technique is adapted to the case of high-resolution acoustic imaging of active sound sources in a three-dimensional free field. This technique allows high-resolution imaging and provides a new method of characterization and detection of sound sources. All results show the high resolution imaging capabilities of this new technique when compared with classical time-reversal (TR) backpropagation. More than simply detecting the position of the acoustic source, this technique allows to detect the size of the active sources. This technique provides an alternative to other imaging and source detection techniques, such as three-dimensional acoustic holography and beamforming.

Super-resolution imaging of active sound and vibrational sources using a time-reversal sink

Theory and experiments of super-resolution focusing using a time-reversal sink have been investigated in high-frequency regime [Rosny and Fink, Phys. Rev. Lett. 89] and in audible range [Bavu, Besnainou, Gibiat, Rosny and Fink, Act. Acoust., 93]. This technique, generalized to the case acoustic and vibrational imaging of active sources, allows super-resolution imaging and provides a new method of characterization of active sources in a known background medium. This imaging technique involves a measurement in the background medium using an array, and the simulation of the backpropagating-field in a fictive medium. An ideal numerical time-reversal sink (NumTRAS) is then used to refine results and obtain high-contrast, high-resolution imaging of initial sources. The algorithm has been validated in parallel supercomputer simulations, in both vibrational and acoustics fields and has been used to detect active vibrational sources in a clamped Mindlin plate and active sound sources in an anechoic room. All results show high-resolution imaging capabilities when compared with classical time-reversal backpropagation. NumTRAS provides an alternative to other imaging and source detection techniques, such as acoustic holography and beamforming. Beyond the applications of acoustic and vibrational non-destructive evaluation of industrial structures, NumTRAS has applications in evaluation of musical structures and is being tested to detect and characterize moving sources.

LE PUITS À RETOURNEMENT TEMPOREL DANS LE DOMAINE AUDIBLE : UN OUTIL DE FOCALISATION ET D'IMAGERIE À HAUTE RÉSOLUTION DE SOURCES SONORES ET VIBRATOIRES

Le développement de techniques de focalisation et d’imagerie à haute résolution pour les sources acoustiques et vibratoires à basse fréquence est l’un des enjeux de la recherche actuelle en acoustique, notamment pour exciter localement et analyser des structures vibroacoustiques complexes tout en conservant des propriétés de haute résolution. Ces propriétés sont nécessaires lorsque la taille des objets étudiés est plus petite que la longueur d’onde mise en jeu. Nous désirons une méthode flexible, rapide, précise, non invasive, et unifiée d’excitation et d’analyse. Celle-ci doit être applicable tant dans le domaine des vibrations dans les structures que dans le domaine des ondes acoustiques tridimensionnelles. Pour cela, nous nous basons sur la technique du puits à retournement temporel, qui n’a, à ce jour, été mise en oeuvre que pour la focalisation d’ondes de Lamb dans une cavité ergodique ou avec des ondes électromagnétiques. Aucune technique d’imagerie n’a, avant cette thèse, été dérivée du puits à retournement temporel. La méthode du puits à retournement temporel est adaptée pour la focalisation à basse fréquence. Elle permet d’exciter localement une structure avec une grande intensité, et possède des capacités de super-résolution. Malgré tout, nous démontrons que cette méthode est difficilement applicable en situation pratique, puisqu’elle fait perdre le caractère non invasif nécessaire à la plupart des applications. En revanche, nous présentons dans ce manuscrit une technique nouvelle d’imagerie de sources vibratoires et acoustiques, basée sur le puits à retournement temporel. Cette technique non invasive d’imagerie, utilisant des dispositifs de mesure similaires aux techniques de formations de voies ou d’holographie en champ proche, permet d’obtenir une image des sources vibratoires ou acoustiques à très haute résolution de manière rapide. L’approche de cette nouvelle méthode d’imagerie est décrite. Des applications à l’imagerie de sources d’impact sur une plaque encastrée, ainsi qu'à l’imagerie de sources acoustiques en champ libre et en milieu sous-marin profond sont proposées. Une application à l’imagerie de sources acoustiques à basse fréquence sur une guitare est développée. Ces résultats représentent les premières applications de l’imagerie par puits à retournement temporel numérique. Les limites, la théorie, et la mise en oeuvre de cette technique d’imagerie à haute résolution sont étudiées et détaillées. Il est démontré que cet outil possède des performances et des limites similaires à l’holographie en champ proche, tout en dépassant les capacités à basse fréquence des techniques classiques de localisation limitées en résolution couramment utilisées, comme le beamforming ou le retournement temporel.

Subwavelength Sound Focusing Using a Time-Reversal Acoustic Sink

Time-reversal mirrors (TRM) have been developed since 1986 in order to focus ultrasonic transient waves in complex media. In the last few years, the properties of TR of acoustic fields have been studied in many different areas. Nevertheless, few applications of TR have been developed in audible range acoustics. The aim of this paper is to demonstrate the concept of time-reversal acoustic sink (TRAS) in audible frequency regime, in order to overcome the diffraction limit imposed by the TRM focusing. The major difference between the TRAS and TRM experiments in ultrasonics and audible range is the ratio between the wavelength and the size of the transducers and objects on which the focusing is achieved. The audible range experiment are lead in Fresnel field (near field), whereas the ultrasonic experiments are lead in Fraunhoffer field (far field). We present the first experimental results with a TRAS in this frequency range. The focusing behaviour in a reverberation room using different transient sounds and frequency domains are investigated and discussed, showing that one can take advantage of reverberation in order to achieve subwavelength sound focusing using a single-element TRM. We report that a focal spot of a seventh of a wavelength has been recorded using the TRAS techniques in audible range, compared to the half wavelength obtained with normal TRM processing. A promising application of a numerical TRAS-method in acoustic imaging and localization of acoustic and vibrational sources is presented.

Techniques de Focalisation par Retournement Temporel dans le Domaine Audible

Cetté etude présente les propriétés de focalisation acous-tique spatio-temporelles par retournement temporel (RT) dans le domaine audible en fonction de l’environnement dans lequel l’expérience est menée. La possibilité de réaliser une focalisation par RT grâcè a un miroir com-posé d’un transducteur sera démontrée dans un milieu réverbérant. Malgré leuréléganceleurélégance et leur adaptivité, ces méthodes de focalisation ne permettent pas de focaliser en super-résolution. Le puits acoustique, pour la première fois mis en place dans le domaine audible permet, lui, d’obtenir une résolution accrue par rapport à la focalisation par RT classique et de vaincre la limite physique de diffraction imposée par cettedernì ere. Nos résultats expérimentaux montrent la possibilité de créer une tache focale à λ/7 .

Torsional waves in a bowed string

Bowing a string with a non-zero radius exerts a torque, which excites torsional waves. In general, torsional standing waves have higher fundamental frequencies than do transverse standing waves, and there is generally no harmonic relationship between them. Although torsional waves have little direct acoustic effect, the motion of the bow-string contact depends on the sum of the transverse speed v of the string plus the radius times the angular velocity (rw) . Consequently, in some bowing regimes, torsional waves could introduce non-periodicity or jitter to the transverse wave. The ear is sensitive to jitter so, while quite small amounts of jitter are important in the sounds of (real) bowed strings, modest amounts of jitter can be perceived as unpleasant or unmusical. It follows that, for a well bowed string, aperiodicities produced in the transverse motion by torsional waves (and other effects) must be small. Is this because the torsional waves are of small amplitude or because of strong coupling between the torsional and transverse waves? We measure the torsional and transverse motion for a string bowed by an experienced player over a range of tunings. The peaks in (rw), which occur near the start and end of the stick phase in which the bow and string move together, are only several times smaller than v during this phase.

Rotationnal and translational waves in a bowed string

We measure and compare the rotational and transverse velocity of a bowed string. When bowed by an experienced player, the torsional motion is phase-locked to the transverse waves, producing highly periodic motion. The spectrum of the torsional motion includes the fundamental and harmonics of the transverse wave, with strong formants at the natural frequencies of the torsional standing waves in the whole string. Volunteers with no experience on bowed string instruments, however, often produced non-periodic motion. We present sound files of both the transverse and torsional velocity signals of well-bowed strings. The torsional signal has not only the pitch of the transverse signal, but it sounds recognisably like a bowed string, probably because of its rich harmonic structure and the transients and amplitude envelope produced by bowing.