[Paper Review] Saving phase: Injectivity and stability for phase retrieval

Afonso S. Bandeira, Jameson Cahill+2 more|arXiv (Cornell University)|Feb 19, 2013

Advanced X-ray Imaging Techniques22 references21 citations

TL;DR

This paper establishes foundational results for phase retrieval by characterizing injectivity and stability in complex and real vector spaces. It proves that $4M-4$ generic measurement vectors are necessary and sufficient for injectivity in $M$-dimensional complex spaces, and introduces a strong complement property to quantify worst-case stability, showing Gaussian measurements satisfy this with high probability.

ABSTRACT

Recent advances in convex optimization have led to new strides in the phase retrieval problem over finite-dimensional vector spaces. However, certain fundamental questions remain: What sorts of measurement vectors uniquely determine every signal up to a global phase factor, and how many are needed to do so? Furthermore, which measurement ensembles lend stability? This paper presents several results that address each of these questions. We begin by characterizing injectivity, and we identify that the complement property is indeed a necessary condition in the complex case. We then pose a conjecture that 4M-4 generic measurement vectors are both necessary and sufficient for injectivity in M dimensions, and we prove this conjecture in the special cases where M=2,3. Next, we shift our attention to stability, both in the worst and average cases. Here, we characterize worst-case stability in the real case by introducing a numerical version of the complement property. This new property bears some resemblance to the restricted isometry property of compressed sensing and can be used to derive a sharp lower Lipschitz bound on the intensity measurement mapping. Localized frames are shown to lack this property (suggesting instability), whereas Gaussian random measurements are shown to satisfy this property with high probability. We conclude by presenting results that use a stochastic noise model in both the real and complex cases, and we leverage Cramer-Rao lower bounds to identify stability with stronger versions of the injectivity characterizations.

Motivation & Objective

• Address the fundamental open problem of determining necessary and sufficient conditions for injective and stable phase retrieval in finite-dimensional spaces.
• Characterize injectivity in the complex case, identifying the complement property as necessary and proposing a conjecture for $4M-4$ measurements to be both necessary and sufficient.
• Establish worst-case stability in the real case through a new matrix condition—strong complement property—reminiscent of the restricted isometry property in compressed sensing.
• Analyze average-case stability using a stochastic noise model and Cramér-Rao lower bounds, linking stability to stronger injectivity conditions.
• Provide theoretical guarantees for practical measurement ensembles, including Gaussian and localized frames, and refine existing conjectures in quantum-inspired phase retrieval.

Proposed method

• Use algebraic geometry and frame theory to characterize injectivity in the complex case, proving the complement property is necessary for injectivity.
• Introduce the strong complement property as a numerical version of the complement property, enabling Lipschitz bounds on the intensity measurement mapping.
• Derive upper and lower Lipschitz bounds for the intensity map using singular values of submatrices of the measurement ensemble.
• Apply Cramér-Rao lower bounds to analyze average-case stability, connecting statistical estimation efficiency to injectivity strength.
• Prove the conjecture that $4M-4$ generic measurements are sufficient for injectivity in $M=2,3$ dimensions using a novel injectivity test.
• Leverage the polarization identity and expander graphs to analyze relative phase recovery, supporting stability under $\mathcal{O}(M\log M)$ measurements.

Experimental results

Research questions

• RQ1 What is the necessary and sufficient condition for a set of measurement vectors to yield injective phase retrieval in complex $M$-dimensional spaces?
• RQ2 Is $4M-4$ the minimal number of generic measurement vectors required for injectivity in $M$-dimensional complex spaces?
• RQ3 What matrix condition ensures worst-case stability in real-phase retrieval, and how does it relate to known properties like the restricted isometry property?
• RQ4 How does the choice of measurement ensemble—e.g., Gaussian vs. localized frames—affect stability in phase retrieval?
• RQ5 Can Cramér-Rao lower bounds be used to derive stronger injectivity conditions that imply stability in both real and complex settings?

Key findings

• The complement property is a necessary condition for injectivity in the complex phase retrieval problem.
• The paper proves that $4M-4$ generic measurement vectors are both necessary and sufficient for injectivity in $M=2$ and $M=3$ dimensions.
• A new matrix condition—strong complement property—is introduced and shown to yield sharp lower Lipschitz bounds on the intensity measurement map in the real case.
• Gaussian random measurements satisfy the strong complement property with high probability, implying stable phase retrieval in the worst case.
• Localized frames fail to satisfy the strong complement property, indicating inherent instability in such measurement ensembles.
• Cramér-Rao lower bounds link statistical estimation efficiency to injectivity strength, showing that stability in the average case corresponds to stronger injectivity conditions in both real and complex settings.