Demixing sparse signals via convex optimization

Yi Zhou, Yingbin Liang

Abstract: We consider demixing a pair of sparse signals in orthonormal basis via convex optimization. Theoretically, we characterize the condition under which the solution of the convex optimization problem correctly demixes the true signal components. In specific, we introduce the local subspace coherence to characterize how a basis vector is coherent with a signal subspace, and show that the convex optimization approach succeeds if the subspaces of the true signal components avoid high local subspace coherence. Furthermore, we illustrate via examples that our condition for exact demixing is more fundamental than existing conditions. We then verify our theoretical finding through numerical experiments.