Double-underdetermined target localization with multistatic MIMO radar via Vandermonde-structured double coupled canonical polyadic decomposition

This study investigates target localization in multistatic MIMO radar with coprime planar arrays (CPPAs), with a particular focus on challenging double underdetermined scenarios, where both the transmit and receive arrays contain fewer sensors than the number of targets, resulting in not full column rank (i.e., rank-deficient) factor matrices on both sides. A tensor-based model is constructed from received signals, which admits a double coupled canonical polyadic decomposition (DC-CPD). The sparse uniform linear subarrays in the CPPAs yield local Vandermonde structures in the first two factor matrices of the DC-CPD model. As a result, the model is referred to as the Vandermonde structured DC-CPD (VS-DC-CPD) model. A (semi-)algebraic VS-DC-CPD algorithm is developed by exploiting the rotational invariance of the Vandermonde structure, which allows the VS-DC-CPD problem to be converted into a sequence of joint eigenvalue decompositions, thereby reducing the computational complexity. The corresponding working conditions are derived, demonstrating the effectiveness in double underdetermined configurations. Simulation results demonstrate superior localization accuracy and average CPU time compared with existing tensor-based approaches, especially in double underdetermined scenarios.