Long-time coherent integration for low SNR target via particle filter in track-before-detect
Résumé
This article addresses the problem of detecting and tracking a target at very low SNR via track-before-detect. Since in this framework targets are undetectable on one single time step, coherent integration must be performed over time to gather sufficient energy and obtain good detection performance. As this integration must be done along a given target trajectory, we propose to apply a particle filter to perform the estimation over time. This particle filter samples the target delay, velocity and amplitude to perform a coherent processing. Two detection tests based on the particles are then proposed to solve the detection problem. Finally we propose also a new particle filter inspired by a GLRT strategy that does not need to sample the amplitude. Both filters converge on the true target state at very low SNR, the latter converging faster than the former.