Radar Target Detection Based on Bayesian Inference With Horseshoe Prior

Résumé fourni par la source

A conventional radar system can exploit the Doppler sparsity of measurement data using the least absolute shrinkage and selection operator (LASSO). When the conventional scheme is applied to the separation between target signals and clutters in external environments, such as the ocean surface, it tends to estimate noise components at zero and thus cannot set an appropriate threshold to achieve a constant false alarm rate. Furthermore, the target-signal components are likely to be underestimated, which results in biased estimates and significantly reduces the estimation accuracy. Hence, this paper proposes a Bayesian inference approach that does not explicitly require any models and is thus superior to the conventional point estimation in terms of both flexibility and estimation accuracy. To ensure low complexity, a variational Bayesian method is used in the proposed scheme to estimate the parameters with ana posterioriprobability. Furthermore, the horseshoe prior is assumed to be ana prioriprobability for Bayesian inference, in contrast to the conventional variational Bayesian approach named Bayesian LASSO, and is expected to compensate for the degraded estimation accuracy. Computer simulations and real-world data experiments show that the proposed scheme can improve the signal-to-clutter plus noise ratio of single-target detection and that it outperforms conventional schemes such as non-uniform discrete Fourier transform and Bayesian LASSO.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.