Développement de nouvelles méthodes de classification/localisation de signaux acoustiques appliquées aux véhicules aériens

Abstract : This thesis deals with the development of a compact microphone array and a dedicated signal processing chain for aerialtarget recognition and direction of arrival (DOA) estimation. The suggested global approach consists in an initial detection ofa potential target, followed by a DOA estimation and tracking process, along with a refined detection, facilitated by adaptivespatial filtering. An original DOA estimation algorithm is proposed. It uses the RANSAC algorithm on real-time time-domainbroadband [100 Hz - 10 kHz] pressure and particle velocity data which are estimated using finite differences and sums ofsignals of microphone pairs with frequency-dependent inter-microphone spacings. The use of higher order finite differences, or variants of the Phase and Amplitude Gradient Estimation (PAGE) method adapted to the designed antenna, can extend its bandwidth at high frequencies. The designed compact microphone array uses 32 digital MEMS microphones, horizontally disposed over an area of 7.5 centimeters. This array geometry is suitable to the implemented algorithms for DOA estimation and spatial filtering. DOA estimation and tracking of a trajectory controlled by a spatialization sphere in the Ambisonic domain have shown an average DOA estimation error of 4 degrees. A database of flying drones acoustic signatures has been set up, with the knowledge of the drone’s position in relation to the microphone array set out by GPS measurements. Adding artificial noise to the data, and selecting acoustic features with evolutionary programming have enabled the detection of an unknown drone in an unknown soundscape within 200 meters with the JRip classifier. In order to facilitate the detection and extend its range, the initial detection stage is preceded by differential beamforming in four main directions (north, south, east, west), and the refined detection stage is preceded by MVDR beamforming informed by the target’s DOA.
Complete list of metadatas

Cited literature [259 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-02180882
Contributor : Abes Star <>
Submitted on : Thursday, July 11, 2019 - 4:38:12 PM
Last modification on : Thursday, September 12, 2019 - 3:45:45 AM

File

TheseRAMAMONJY.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02180882, version 1

Collections

Citation

Aro Ramamonjy. Développement de nouvelles méthodes de classification/localisation de signaux acoustiques appliquées aux véhicules aériens. Acoustique [physics.class-ph]. Conservatoire national des arts et metiers - CNAM, 2019. Français. ⟨NNT : 2019CNAM1234⟩. ⟨tel-02180882⟩

Share

Metrics

Record views

238

Files downloads

48