%0 Conference Proceedings
%T Sequential sensor placement using Bayesian compressive sensing for direction of arrival estimation
%+ École Nationale Supérieure de Techniques Avancées Bretagne (ENSTA Bretagne)
%+ Laboratoire Vibrations Acoustique (LVA)
%+ Laboratoire Modélisation et Sûreté des Systèmes (LM2S)
%+ SIMulation pARTiculaire de Modèles Stochastiques (SIMSMART)
%A Courcoux-Caro, Milan
%A Vanwynsberghe, Charles
%A Baussard, Alexandre
%A Herzet, Cédric
%< sans comité de lecture
%Z Microphones/sensors arrays
%B Forum Acusticum
%C Lyon, France
%P 303-304
%8 2020-12-07
%D 2020
%R 10.48465/fa.2020.0095
%K Sequential sensor placement
%K Optimal design
%K DOA estimation
%K Bayesian compressive sensing
%Z Physics [physics]/Mechanics [physics]/Acoustics [physics.class-ph]
%Z Physics [physics]/Mechanics [physics]/Vibrations [physics.class-ph]Conference papers
%X Measurement selection is a general optimization problem with the purpose of minimizing the estimation error. It aims at answering the following question: which set of measurements will give us the best estimation? In our presentation, we propose a method to design an array for direction of arrival (DOA) estimation. Since the measurements are linked to a sensor, the optimization problem becomes a sensor placement problem. To solve it, a greedy data-dependent approach is chosen answering this new question: according to what is measured by the current array, which new sensor position could improve the DOA estimation at most? Concerning the DOA problem, we took inspiration from the compressed sensing (CS) framework, and consider the DOA estimation as a sparse localization problem. With this assumption, one can solve the DOA estimation problem in its undetermined form, by the Sparse Bayesian Inference (SBI). The algorithm estimates the source angles based on a sparsity-promoting hierarchical model. The next sensor choice minimizes a cost function related to the error covariance matrix. In Bayesian experimental design, a common choice is the D-design: it minimises the determinant of the error covariance matrix. In summary, the proposed strategy is a sequential sensor placement based on Bayesian experimental design. It alternates between a step of sparse DOA estimation, and a step to choose the sensor position in the D-optimal sense. The numerical experiments concern a DOA acoustic problem. The results show that an array designed by the proposed method needs less sensors than a random array in order to localize all sources.
%G English
%2 https://hal.science/hal-03231792/document
%2 https://hal.science/hal-03231792/file/000095.pdf
%L hal-03231792
%U https://hal.science/hal-03231792
%~ ENSTA-BRETAGNE
%~ UNIV-RENNES1
%~ IRMAR
%~ UR2-HB
%~ CNRS
%~ INRIA
%~ UNIV-TROYES
%~ INSA-LYON
%~ INSA-RENNES
%~ INRIA-RENNES
%~ IRISA
%~ ENSTA-BRETAGNE-STIC
%~ LVA
%~ INRIA_TEST
%~ CELYA
%~ UNAM
%~ TESTALAIN1
%~ IRMAR-STAT
%~ CHL
%~ INRIA2
%~ UR1-HAL
%~ UR1-MATH-STIC
%~ UR1-UFR-ISTIC
%~ AGREENIUM
%~ UNIV-RENNES2
%~ TEST-UR-CSS
%~ UNIV-RENNES
%~ INRIA-RENGRE
%~ INRIA-300009
%~ INSA-GROUPE
%~ UTT
%~ FA2020
%~ UDL
%~ TEST-HALCNRS
%~ UR1-MATH-NUM
%~ UTT-LIST3N
%~ UTT-FULL-TEXT
%~ LM2S-UTT
%~ INSTITUT-AGRO