Extract before detect, coherent extraction based on gridless Compressed Sensing

The goal of radar processing chain is to extract a few target echoes from their noisy sum made of thousands of complex numbers. Current radar chains are composed of matched filters (Pulse Compression, Doppler filters, Digital Beam Forming), noise/clutter estimation and thresholding, then extraction (hits clustering, location estimation). In this paper, extraction function is performed thanks to a Compressed Sensing approach. An important difference with current extraction function is that it performs a real “coherent extraction” (complex subtraction in each burst) of targets from the observed signals. This is a key factor to increase extraction capacity: coherent extraction can extract a higher number of target echoes than non-coherent extraction. This paper considers a multidimensional target domain and multidimensional input signals that are ambiguous in range and radial velocity, where target echoes fluctuate from burst to burst. The algorithm used to recover the sparse representation is Orthogonal Matching Pursuit (OMP) where the dictionary matrix is continuous over the target domain, therefore overcoming the grid problem: “gridless OMP”.