One-Dimensional Discrete-Time Phase Retrieval

Robert Beinert, Gerlind PlonkaTopics 2020 · DOI · OpenAlex

Abstract

The phase retrieval problem has a long and rich history with applications in physics and engineering such as crystallography, astronomy, and laser optics. Usually, the phase retrieval consists in recovering a real-valued or complex-valued signal from the intensity measurements of its Fourier transform. If the complete phase information in frequency domain is lost then the problem of signal reconstruction is severely ill-posed and possesses many non-trivial ambiguities. Therefore, it can only be solved using appropriate additional signal information. We restrict ourselves to one-dimensional discrete-time phase retrieval from Fourier intensities and particularly consider signals with finite support. We study the structure of the arising ambiguities and show how they can be characterized using the given Fourier intensity. Furthermore, we study different kinds of a priori assumptions on the signal, such as non-negativity, additional time-domain magnitudes, or interference measurements, to reduce ambiguities or ensure uniqueness.

Introduction

Phase retrieval is a fundamental challenge in signal processing and imaging science. In many physical applications, sensors can only record the intensity (magnitude) of a wave field, while the phase information—crucial for reconstructing the original object—is lost. Mathematically, this involves recovering a signal $x$ from the magnitude of its Fourier transform $|\hat{x}|$.

While phase retrieval in two or more dimensions is often uniquely solvable (up to trivial factors) because multivariate polynomials are generally irreducible, the one-dimensional (1D) case is notoriously ill-posed. In 1D, the problem is plagued by a large number of non-trivial ambiguities, meaning many physically distinct signals can produce the exact same Fourier intensity.

The Structure of Ambiguities

For a discrete-time signal of finite length $N$, the phase retrieval problem can be framed using algebraic polynomials. A signal $x = (x[0], \dots, x[N-1])$ is associated with a polynomial $P_x(z) = \sum_{n=0}^{N-1} x[n] z^n$. The squared Fourier intensity corresponds to the Fourier transform of the signal's autocorrelation.

The non-trivial ambiguities arise from the "zero-flipping" phenomenon. The zeros of the autocorrelation polynomial come in pairs: if $z_k$ is a zero, then $1/\bar{z}_k$ is also a zero. Constructing a valid signal involves choosing one zero from each pair. For a signal with $N-1$ zeros, there can be up to $2^{N-2}$ such combinations, each representing a mathematically valid solution that matches the observed Fourier intensity.

Figure 4: Illustration of non-trivial ambiguities in 1D phase retrieval. Three distinct signals ($x_1, x_2, x_3$) yield identical Fourier intensities (a) despite having different moduli (b) and phases (c).

As shown in the figure above, signals $x_1, x_2,$ and $x_3$ are indistinguishable in the frequency domain (a) but differ significantly in their time-domain characteristics (b and c).

Overcoming Ambiguity with A Priori Constraints

To achieve a unique reconstruction, researchers must impose additional constraints or "a priori" information on the signal. This paper evaluates several common strategies:

1. Non-Negativity

In many optical applications, it is assumed that the signal is real-valued and non-negative ($x[n] \ge 0$). While this constraint significantly prunes the solution space, the authors demonstrate that it is generally insufficient to guarantee uniqueness. There exist infinite families of non-negative signals that still possess non-trivial ambiguities.

2. Time-Domain Information

Providing additional information about the signal in the time domain can help. For instance, knowing the magnitude of certain signal components or the value of an endpoint ($x[N-1]$) can lead to uniqueness. The paper establishes that for "generic" signals, knowing the magnitudes $|x[n]|$ is enough to ensure a unique solution, though specific "pathological" cases remain where multiple solutions persist.

3. Interference Measurements

One of the most powerful methods for ensuring uniqueness is the use of interference. By mixing the unknown signal with a known reference signal before measuring the Fourier intensity, the symmetry of the zero-flipping is broken. The authors prove that using two such interference measurements with different modulations (e.g., rotations in the complex plane) allows for the unique recovery of any signal.

Key Mathematical Results

The paper formalizes these insights through several key theorems:

  • Theorem 24.3 provides a precise count of non-trivial ambiguities based on the geometry of the polynomial zeros relative to the unit circle.
  • Theorem 24.11 confirms that interference-based measurements provide a robust path to uniqueness, which is highly relevant for practical experimental setups in holography and coherent diffraction imaging.

Conclusion and Future Directions

While this work provides a rigorous theoretical foundation for 1D phase retrieval, practical implementation faces further hurdles. Numerical stability is a major concern; even when a solution is theoretically unique, noise in the intensity measurements can make the problem ill-conditioned. Future research is directed toward developing more stable reconstruction algorithms and extending these rigorous uniqueness proofs to more complex measurement schemes, such as those involving the Fresnel or fractional Fourier transforms.

Summary updated

Abstract

Abstract The phase retrieval problem has a long and rich history with applications in physics and engineering such as crystallography, astronomy, and laser optics. Usually, the phase retrieval consists in recovering a real-valued or complex-valued signal from the intensity measurements of its Fourier transform. If the complete phase information in frequency domain is lost then the problem of signal reconstruction is severelly ill-posed and possesses many non-trivial ambiguities. Therefore, it can only be solved using appropriate additional signal information. We restrict ourselves to one-dimensional discrete-time phase retrieval from Fourier intensities and particularly consider signals with finite support. In the first part of this section, we study the structure of the arising ambiguities of the phase retrieval problem and show how they can be characterized using the given Fourier intensity. Employing these observations, in the second part, we study different kinds of a priori assumptions on the signal, where we are especially interested in their ability to reduce the non-trivial ambiguities or even to ensure uniqueness of the solution. In particular, we consider the assumption of non-negativity of the solution signal, additional magnitudes or phases of some signal components in time domain, or additional intensities of interference measurements in frequency domain. Finally, we transfer our results to phase retrieval problems where the intensity measurements arise, for example, from the Fresnel or fractional Fourier transform.
BibTeX
@inproceedings{beinert2020one,
  author = {Robert Beinert and Gerlind Plonka},
  title = {{One-Dimensional Discrete-Time Phase Retrieval}},
  booktitle = {Topics},
  year = {2020},
  doi = {10.1007/978-3-030-34413-9\_24},
}
Figure