Spline-based Reconstruction of Periodic Signals with Sparse Innovations

Abstract

Optimization-based problems have become of great interest for signal approximation purposes, as they achieved good accuracy results while being extremely flexible and versatile. In this work, we put our focus on the context of periodic signals sampled with spatial measurements. The optimization problems are penalized thanks to the total variation norm, using a specific class of (pseudo-)differential operators to use well-chosen reconstruction functions. We introduce three algorithms and their adaptation to this specific use case. The first one is a discrete grid-based $\ell_1$ method, the second is called CPGD, which relies on the estimation of discrete innovations within the FRI framework, and the last one is the Frank-Wolfe algorithm. We put the emphasis on that later algorithm as we underline its greedy behavior. We consider a refined version of this algorithm, that leads to very encouraging outputs. We pursue the algorithmic treatment with a theoretical analysis of the optimization problem, based on convexity and duality theory. We draw away two conditions that, if verified by the differential operator, would inform on the shape of the solutions and even ensure uniqueness, in some cases. Moreover, we demonstrate the general expression of periodic one-dimensional exponential splines, as a first step to verify the latter conditions for exponential operators.