On the Geometry of Radar Waveform Design with Constraints
Many optimization problems that arise in signal processing applications are based around optimizing vectors/matrices of numeric sequences. These problems are often inherently non-convex due to non-linear constraints. Such problems are frequently solved through convex relaxations of the original problem to allow for computationally simple algorithms to be leveraged, albeit with the potential for appreciable performance costs in the solution space. When the constraints of a problem can be implicitly satisfied through the structure of a Riemannian manifold, optimizing directly on said manifold affords the opportunity to solve the problem in an unconstrained fashion. Therefore, when feasible, recasting the constrained optimization problem in Euclidean space as an unconstrained optimization over a Riemannian manifold offers greater flexibility and improved efficiency over convex relaxation based approaches. This paper introduces the minimal smooth optimization theory required and adapts the notation to a radar signal design context. In addition, we apply our approach to two different problems in direct parallel to contemporary convex relaxation-based signal design literature.