The geometry of ambiguity in one-dimensional phase retrieval

TL;DR

This paper explores the geometric structure of ambiguities in one-dimensional Fourier phase retrieval, introducing a root covering space to analyze trivial and non-trivial ambiguities and providing criteria for generic phase retrieval.

Contribution

It introduces the root covering to understand ambiguities and extends the analysis to criteria for generic phase retrieval in real subvarieties.

Findings

The space of signals has a finite covering where ambiguities are trivial.

The incidence variety describes how ambiguities vary with signals.

A criterion for generic phase retrieval in real subvarieties is established.

Abstract

We consider the geometry associated to the ambiguities of the one-dimensional Fourier phase retrieval problem for vectors in ${\mathbb C}^{N+1}$. Our first result states that the space of signals has a finite covering (which we call the root covering) where any two signals in the covering space with the same Fourier intensity function differ by a trivial covering ambiguity. Next we use the root covering to study how the non-trivial ambiguities of a signal vary as the signal varies. This is done by describing of the incidence variety of pairs of signals with same fourier intensity function modulo global phase. As an application we give a criterion for a real subvariety of the space of signals to admit generic phase retrieval. The extension of this result to multi-vectors played an important role in the author's work with Bendory and Eldar on blind phaseless short-time fourier transform recovery.