Phase-Retrieval in Shift-Invariant Spaces with Gaussian Generator

TL;DR

This paper investigates the problem of reconstructing Gaussian-shifted functions from magnitude-only samples, establishing conditions for real and complex cases, and characterizing solutions in the complex case.

Contribution

It provides sharp density conditions for phaseless recovery of Gaussian shift-invariant functions and characterizes all solutions for complex-valued functions.

Findings

Unique recovery for real-valued functions with density > 2

Sharpness of the density condition

Complete characterization of solutions for complex-valued functions

Abstract

We study the problem of recovering a function of the form $f(x) = \sum _{k\in \mathbb{Z} } c_k e^{-(x-k)^2}$ from its phaseless samples $|f(\lambda )|$ on some arbitrary countable set $\Lambda \subseteq \mathbb{R} $. For real-valued functions this is possible up to a sign for every separated set with Beurling density $D^-(\Lambda ) >2$. This result is sharp. For complex-valued functions we find all possible solutions with the same phaseless samples.