Bayesian sparse Fourier representation of off-grid targets with application to experimental radar data

Marie Lasserre (1) , Stéphanie Bidon (1) , Olivier Besson (1) , François Le Chevalier (2)

1 DEOS - Département Electronique, Optronique et Signal
2 Department of Electrical Engineering, Mathematics and Computer Science [Delft]

The problem considered is the estimation of a finite number of cisoids embedded in white noise, using a sparse signal representation (SSR) approach, a problem which is relevant in many radar applications. Many SSR algorithms have been developed in order to solve this problem, but they usually are sensitive to grid mismatch. In this paper, two Bayesian algorithms are presented, which are robust towards grid mismatch: a first method uses a Fourier dictionary directly parametrized by the grid mismatch while the second one employs a first-order Taylor approximation to relate linearly the grid mismatch and the sparse vector. The main strength of these algorithms lies in the use of a mixed-type distribution which decorrelates sparsity level and target power. Besides, both methods are implemented through a Monte-Carlo Markov chain algorithm. They are successfully evaluated on synthetic and experimental radar data, and compared to a benchmark algorithm

Mots clés: Monte-Carlo Markov Chain; Grid mismatch; Bayesian inference; Sparse representation

Domaines: Traitement du signal et de l'image [eess.SP]

Marie Lasserre, Stéphanie Bidon, Olivier Besson, François Le Chevalier. Bayesian sparse Fourier representation of off-grid targets with application to experimental radar data. Signal Processing, 2015, vol. 111, pp. 261-273. ⟨10.1016/j.sigpro.2014.12.024⟩. ⟨hal-01111139⟩