Uncertainty Relations for Shift-Invariant Analog Signals

TL;DR

This paper extends the uncertainty principle to infinite-dimensional shift-invariant spaces, enabling sparse signal representations and decompositions in analog signals using convex optimization.

Contribution

It introduces an uncertainty principle for shift-invariant spaces and demonstrates how to find sparse decompositions via finite-dimensional convex optimization.

Findings

Established an uncertainty principle for analog shift-invariant signals.

Proved tightness of the bound with a bandlimited train example.

Provided a method to find sparse representations in infinite domains.

Abstract

The past several years have witnessed a surge of research investigating various aspects of sparse representations and compressed sensing. Most of this work has focused on the finite-dimensional setting in which the goal is to decompose a finite-length vector into a given finite dictionary. Underlying many of these results is the conceptual notion of an uncertainty principle: a signal cannot be sparsely represented in two different bases. Here, we extend these ideas and results to the analog, infinite-dimensional setting by considering signals that lie in a finitely-generated shift-invariant (SI) space. This class of signals is rich enough to include many interesting special cases such as multiband signals and splines. By adapting the notion of coherence defined for finite dictionaries to infinite SI representations, we develop an uncertainty principle similar in spirit to its finite counterpart. We demonstrate tightness of our bound by considering a bandlimited lowpass train that achieves the uncertainty principle. Building upon these results and similar work in the finite setting, we show how to find a sparse decomposition in an overcomplete dictionary by solving a convex optimization problem. The distinguishing feature of our approach is the fact that even though the problem is defined over an infinite domain with infinitely many variables and constraints, under certain conditions on the dictionary spectrum our algorithm can find the sparsest representation by solving a finite-dimensional problem.