Joint Localisation of Line‐of‐Sight and Non‐Line‐of‐Sight Targets via Deep Learning‐Assisted Millimetre‐Wave Radar

ABSTRACT

Joint localisation of line‐of‐sight (LOS) and nonline‐of‐sight (NLOS) targets using millimetre‐wave radar is challenging in urban occluded environments due to complex multipath propagation and the large energy imbalance between strong LOS echoes and weak NLOS returns. This paper proposes a joint LOS/NLOS localisation framework for typical urban T‐shaped scenes by combining multidimensional parameter estimation, region‐aware sparse reconstruction and deep‐unfolded iterative optimisation. The scene is first geometrically modelled and partitioned into LOS and NLOS regions. Target localisation is then formulated as a sparse recovery problem in the range‐angle‐Doppler domain, where a region‐weighted strategy is introduced to enhance weak NLOS target responses. Based on this formulation, two network‐enhanced iterative methods are developed to learn adaptive regularisation and parameter update strategies from data. Both methods retain the interpretability of conventional iterative reconstruction whilst improving robustness under low‐SNR and mixed LOS/NLOS conditions. Simulation and real‐data experiments show that the proposed methods outperform conventional iterative algorithms in reconstruction accuracy, false‐alarm suppression and sidelobe/artefact reduction, achieving robust super‐resolution localisation in complex urban occluded scenes.