Conference Paper

We propose a Group-Regularized Atomic Norm minimization (GRAN) framework for spectral estimation and target localization in widely separated MIMO (WS-MIMO) radar. GRAN combines the off-grid accuracy of atomic-norm minimization with a group-sparsity regularizer that enforces global consistency across spatially distributed receivers, thereby addressing the complementary weaknesses of existing atomic-norm (AN) and Group Lasso (GL) approaches. The proposed formulation jointly reconstructs the continuous-domain spectra of all receivers while enforcing physically consistent target locations, reducing spurious solutions and improving robustness in compressed settings. Numerical experiments demonstrate that GRAN retains the low-SNR advantage of GL while avoiding its high-SNR performance saturation caused by off-the-grid errors. At the same time, GRAN consistently outperforms the AN approach by a constant margin. These gains translate into improved position estimates, showing that GRAN provides a practical and computationally tractable solution for exploiting the global structure of WS-MIMO radar systems without incurring discretization-induced errors.