Fact-checked by Grok 2 weeks ago

Phase retrieval

Phase retrieval is the computational challenge of recovering a complex-valued signal from intensity-only measurements, typically the squared magnitudes of its or linear projections thereof, addressing the inherent loss of phase information in such data. This is fundamental in and , where direct phase detection is often infeasible due to the nature of sensors like cameras that record only flux (intensity). The origins of phase retrieval trace back to the 1950s in , with David Sayre's proposal for phase recovery from patterns, though practical algorithms emerged later in . In the 1970s and early 1980s, iterative methods such as the Gerchberg-Saxton algorithm (1972) and Fienup's hybrid input-output technique (1982) advanced the field, enabling two-dimensional reconstructions. A resurgence occurred in the and with coherent diffractive (CDI) using free-electron lasers, allowing lensless of non-crystalline specimens at nanometer resolutions, such as biological viruses and nanocrystals. Applications span diverse domains, including X-ray and optical imaging for and , astronomy for array phasing, electron microscopy, and for . In these contexts, phase retrieval enables high-resolution reconstructions without physical lenses, overcoming limits and supporting modalities like and the transport-of-intensity equation. Key algorithms fall into categories such as alternating projections (e.g., error reduction and hybrid methods), convex relaxations like PhaseLift via semidefinite programming, gradient-based optimization (e.g., Wirtinger flow), spectral initialization techniques, and Bayesian approaches like approximate message passing. Recent developments, particularly since the 2010s, integrate sparsity priors inspired by compressed sensing and machine learning enhancements, such as deep neural networks for regularization and generative models to improve robustness against noise and initialization issues. As of 2025, ongoing advances include complex-valued neural networks and feature-domain methods for improved performance in computational microscopy. These advances have theoretical underpinnings linking phase retrieval to single-layer neural network training, promising further efficiency in large-scale applications.

Fundamentals

Problem Statement

Phase retrieval is a fundamental in and that involves recovering an unknown complex-valued signal from intensity-only measurements, particularly the magnitudes of its or other linear transformations. This process aims to reconstruct the full signal, including both amplitude and phase components, when only the squared magnitudes—corresponding to intensities—are directly observable. The challenge stems from the non-uniqueness of the phase retrieval problem, as multiple signals can produce the same intensity measurements, necessitating additional constraints or for reliable recovery. The motivation for phase retrieval arises from physical constraints in optical systems and , where detectors, such as those in or microscopes, capture only the of due to their inability to resolve rapid phase oscillations at frequencies around 10^{15} Hz. In these setups, , which encodes critical structural details like relative positions and shapes, is lost during measurement, leaving researchers with patterns that provide data alone. This loss has long hindered high-resolution imaging in fields like and , where recovery is essential for forming coherent images from scattered . In the general setup, an input signal x \in \mathbb{C}^n is observed through measurements of the form |Ax|^2, where A is a matrix representing linear transformations, such as the matrix for frequency-domain data. Historically, the phase loss in diffraction patterns was first recognized in the context of in the mid-20th century, prompting early efforts to infer from distributions. Real-world examples include reconstructing nanoscale images of biological samples, such as viruses or nanocrystals, from coherent patterns generated by free-electron lasers, enabling non-invasive without lenses.

Mathematical Formulation

The phase retrieval problem seeks to recover an unknown complex-valued signal \mathbf{x} \in \mathbb{C}^n (up to a global ) from a set of measurements b_m = |\langle \mathbf{a}_m, \mathbf{x} \rangle|^2 for m = 1, \dots, M, where \mathbf{a}_m \in \mathbb{C}^n are known measurement vectors and \langle \cdot, \cdot \rangle denotes the standard inner product. Here, n represents the dimension of the signal, and M is the number of measurements, which must typically exceed n for practical recovery. This setup arises in applications such as X-ray crystallography and coherent diffraction imaging, where only magnitudes are recorded due to detector limitations. The problem is commonly formulated as a non-convex optimization task, such as the least-squares minimization \min_{\mathbf{x} \in \mathbb{C}^n} \sum_{m=1}^M \left( b_m - |\langle \mathbf{a}_m, \mathbf{x} \rangle|^2 \right)^2, which measures the discrepancy between observed intensities and those predicted by a candidate signal. Equivalent formulations include amplitude-based losses like \min_{\mathbf{x}} \sum_{m=1}^M \left( \sqrt{b_m} - |\langle \mathbf{a}_m, \mathbf{x} \rangle| \right)^2, though the intensity-based version is prevalent in non-convex methods. In matrix notation, letting A \in \mathbb{C}^{M \times n} have rows \mathbf{a}_m^\top, the measurements satisfy \mathbf{b} = |\!|A\mathbf{x}|\!|^2, where |\!|\cdot|\!| applies elementwise , emphasizing the quadratic nature of the constraints. A special case is Fourier phase retrieval, where the \mathbf{a}_m are rows of the unitary discrete Fourier transform (DFT) matrix F \in \mathbb{C}^{n \times n} with entries F_{k,j} = n^{-1/2} e^{-2\pi i (k-1)(j-1)/n}, so the measurements are b_k = |\hat{x}_k|^2 for the Fourier coefficients \hat{\mathbf{x}} = F\mathbf{x}. Oversampling is crucial here: for one-dimensional signals, sampling the Fourier magnitudes at more than twice the Nyquist rate (e.g., M > 2n) enables uniqueness when combined with a known finite support constraint, as the autocorrelation of the signal can be uniquely determined from the oversampled intensities. Without additional constraints, phase retrieval is ill-posed due to inherent non-: for generic measurement vectors, up to $2^{n-1} - 1 distinct signals (excluding the trivial global ) can yield identical intensities. Constraints such as finite , sparsity, or non-negativity are thus essential to resolve ambiguities and ensure . For generic complex Gaussian measurements, theoretical results establish that M \geq 4n - 4 intensity measurements suffice for up to global with high probability, marking a tight bound achieved in 2015.

History

Early Developments

The phase problem in emerged as a fundamental challenge in the early , stemming from the inability to directly measure the phases of diffracted X-rays, which are essential for reconstructing atomic structures from intensity data alone. In 1934, Arthur L. Patterson introduced the Patterson function, a synthesis of squared magnitudes that maps interatomic distance vectors without requiring phase information, thereby providing a practical for structure determination in crystals. This method highlighted the intrinsic ambiguity of phase recovery, as multiple phase sets could yield identical intensities, complicating unique reconstructions. By 1944, Patterson further elucidated these ambiguities through the concept of homometric structures—distinct atomic arrangements producing indistinguishable diffraction patterns due to phase indeterminacy—underscoring the non-uniqueness inherent in intensity-only measurements from diffraction experiments. In the 1950s, David Sayre advanced the foundational ideas by proposing that oversampled intensity data, including measurements between Bragg peaks, could theoretically enable direct phase determination for structures composed of resolvable identical atoms, laying groundwork for computational solutions. These crystallography roots emphasized the pervasive challenge of phase ambiguity in wave-based imaging. During the 1950s and 1960s, early efforts in shifted toward and analysis to estimate s indirectly, building on Dennis Gabor's 1948 invention of , which recorded via interference with a beam to produce interpretable patterns. Developments in the 1960s, such as off-axis by Emmett Leith and Juris Upatnieks, refined visibility and pattern analysis for in coherent optical systems, though these methods still relied on waves rather than magnitude-only retrieval. In 1963, Adriaan Walther formally posed the phase retrieval question in , questioning whether a complex-valued image could be uniquely recovered from its finite-aperture intensity, recognizing similar ambiguities as in . Initial proposals for direct phase retrieval in appeared in the mid-1960s, notably Walter Hoppe's 1969 work on electron microscopy, which introduced iterative techniques using overlapping patterns to resolve phases in inhomogeneous wave fields, bridging and applications. These efforts revealed persistent challenges, such as the flip and non-uniqueness from intensity data, particularly in -limited setups. By the early 1970s, the limitations of analog interferometric methods spurred a transition to computational approaches, enabling algorithmic solutions to the problem without physical references.

Key Milestones

The Gerchberg-Saxton algorithm, introduced in 1972, marked a foundational milestone in computational phase retrieval by providing an to recover the of a signal from its intensity measurements, initially applied in for reconstructing images from patterns. In 1982, James R. Fienup advanced this framework with the hybrid input-output algorithm, which significantly improved convergence rates and success in retrieving phases from intensity data by incorporating feedback mechanisms to handle stagnation issues in iterative projections. During the 1990s, key extensions emerged to address uniqueness challenges in phase retrieval, including , which uses overlapping scanned illuminations to enable stable recovery of extended objects from diffraction patterns, as demonstrated in early experimental implementations in electron microscopy, with implementations following in the . In the , the advent of X-ray free-electron lasers facilitated a major resurgence through coherent diffractive imaging (CDI). The first soft X-ray CDI experiments at FLASH in 2006 and hard X-ray at LCLS in 2009 enabled high-resolution, lensless imaging of non-crystalline specimens, such as biological samples. From 2011 to 2013, the PhaseLift method by Emmanuel J. Candès and colleagues introduced a relaxation that lifts the nonlinear phase retrieval problem into a matrix completion framework, offering polynomial-time guarantees for exact recovery under conditions. In 2015, Candès and Xiaodong Li proposed Wirtinger Flow, an efficient non- optimization algorithm that initializes with spectral methods and refines via on the amplitude-based loss, achieving global convergence to the true signal with high probability using a linear number of measurements. Theoretical milestones further solidified the field's foundations, with results establishing in phase retrieval using as few as 4n - 4 generic measurements for signals in \mathbb{C}^n, ensuring stable recovery up to a global . Similarly, for short-time Fourier transform (STFT) phase retrieval, guarantees were proven for bandlimited signals from measurements with Gaussian windows, requiring order n log n samples for reliable .

Classical Iterative Algorithms

Error Reduction Algorithm

The error reduction algorithm, also known as the Gerchberg-Saxton algorithm (introduced in ), is a foundational for phase retrieval that operates by alternating projections between the object domain and the domain to enforce known constraints. In the object domain, it applies constraints such as (zero values outside a known region) or (non-negativity), while in the Fourier domain, it replaces the magnitude with the measured data while preserving the . This projection-based approach was originally developed for reconstructing phases in electron microscopy from image and diffraction plane intensities. The algorithm begins with an initial guess for the complex-valued object or its , often a random attached to the measured . It then proceeds iteratively: perform an inverse (IFFT) to shift to the object domain and apply the object constraint (e.g., enforcing support by setting values to zero outside the known region); next, perform a forward (FFT) to return to the domain and replace the with the measured values while retaining the estimated ; repeat these steps until the change between iterations falls below a or a maximum number of iterations is reached. The projections can be formally defined as follows. Let y denote the estimate in the Fourier domain and x the estimate in the object domain. The Fourier projection P_f(y) enforces the measured magnitude: P_f(y) = |\hat{F}| \exp(i \arg(y)), where |\hat{F}| is the measured Fourier magnitude and \arg(y) is the phase of y. The object projection P_o(x) enforces the object constraint, such as support: P_o(x) = \begin{cases} x & \text{if } x \in S, \\ 0 & \text{otherwise}, \end{cases} where S is the region; for constraints, it might instead set P_o(x) = |O| \exp(i \arg(x)), with |O| the known object . Each iteration alternates x_{k+1} = \mathrm{IFFT}(P_f(\mathrm{FFT}(P_o(x_k)))). Despite its simplicity and monotonic decrease in the mean-squared error metric due to orthogonal projections, the algorithm often stagnates in local minima, producing ambiguous solutions like twin images or shifted replicas rather than the true object, particularly for complex or undersampled data. It also converges slowly with noisy measurements, as perturbations amplify inconsistencies between domains. These limitations have motivated extensions, but the method remains widely used in for its computational efficiency in preliminary reconstructions.

Hybrid Input-Output Algorithm

The hybrid input-output (HIO) algorithm, introduced by Fienup in 1982, enhances the error-reduction algorithm for phase retrieval by incorporating a mechanism in the object domain to improve convergence speed and avoid stagnation in local minima. This method is particularly suited for Fourier-domain phase retrieval with a support constraint, where the object is known to be zero outside a specified region. Unlike pure projection methods that simply zero out invalid points, HIO adjusts those points using a combination of the previous estimate and the current output, providing directional guidance toward the solution. The algorithm proceeds iteratively as follows: Start with an initial estimate g_k(x) of the object in the spatial domain. Apply the to obtain G_k(u), then enforce the measured Fourier magnitude constraint to form G'_k(u) = |F(u)| e^{i \phi_k(u)}, where |F(u)| is the known intensity and \phi_k(u) is the estimated phase. Perform the Fourier transform to yield g'_k(x). In the object domain, identify the S; for points inside S (valid points), set g_{k+1}(x) = g'_k(x). For points outside S (invalid points), apply the HIO update: g_{k+1}(x) = g_k(x) - \beta g'_k(x), where \beta is a feedback parameter typically chosen between 0.5 and 1.0. This process repeats until , often requiring fewer iterations than error reduction for recognizable reconstructions. The feedback term -\beta g'_k(x) effectively penalizes constraint violations by subtracting a scaled version of the current estimate from the prior input, promoting exploration away from stagnant regions. This leads to faster convergence, with simulations showing image recovery in 20–30 iterations and completion under 100, compared to hundreds for error reduction. HIO also demonstrates robustness to noise in the Fourier magnitudes, such as in stellar speckle interferometry, where it reduces reconstruction errors to levels consistent with the noise floor (e.g., normalized error E_0 \approx 0.02). Tuning the \beta is crucial for performance; values near 1.0 often optimize without , though lower values may stabilize noisy cases. A variant, the output-output , modifies the update to g_{k+1}(x) = g'_k(x) + \beta [g'_k(x) - g_k(x)] for invalid points, but HIO's input-output form is more widely adopted for its balance of speed and reliability. The approach builds on the alternating projections of the Gerchberg-Saxton by introducing this non-projection feedback.

Shrinkwrap Algorithm

The Shrinkwrap algorithm (introduced in 2003), is an iterative phase retrieval method that enhances traditional projection-based techniques by dynamically refining the object's constraint through successive shrinking of a convex envelope around the current density estimate. Developed by Stefano Marchesini and colleagues, it mitigates common issues in phase retrieval, such as stagnation due to overly loose or inaccurate initial , by adapting the boundary to better enclose the object's extent while maintaining consistency with measured magnitudes. This approach is particularly valuable in scenarios where prior knowledge of the object shape is limited or unavailable. The algorithm integrates projections onto constraint sets—typically combining error reduction steps with feedback mechanisms like those in the hybrid input-output method—with periodic support updates. It begins with an initial support derived from the autocorrelation of the measured intensity pattern, which provides a coarse estimate of the object's footprint. In each iteration, the current estimate is propagated between spatial and Fourier domains: in the Fourier domain, the magnitude is enforced to match the measurements, while in the spatial domain, the estimate is constrained to lie within the current support. Every several iterations (e.g., 20), the support is refined by computing the autocorrelation of the current estimate to gauge the object's boundaries, followed by shrinking the envelope—often via computation of the convex hull of thresholded regions—to yield a tighter constraint. This process reduces extraneous artifacts and promotes convergence to a more accurate reconstruction. The support update can be formalized as
S_{k+1} = \shrink(S_k, \autocorr(x_k)),
where \shrink applies a convex envelope (e.g., convex hull) to the support informed by the autocorrelation \autocorr(x_k) of the current estimate x_k.
A key advantage of the Shrinkwrap algorithm lies in its ability to minimize artifacts arising from incorrect or static assumptions, leading to improved fidelity even with noisy or incomplete data. By iteratively tightening the , it facilitates of fine details without requiring manual intervention for support definition. This has proven especially effective in applications, such as coherent diffractive of nanostructures, where precise boundary detection is crucial for resolving object outlines at nanometer scales. For instance, it has enabled high-resolution reconstructions of isolated particles like nanocrystals from patterns alone.

Relaxation-Based Methods

General Semidefinite Relaxation

The general semidefinite relaxation addresses the phase retrieval problem by lifting the unknown signal \mathbf{x} \in \mathbb{C}^n to a rank-one \mathbf{X} = \mathbf{x} \mathbf{x}^* \in \mathbb{C}^{n \times n}, reformulating the intensity measurements b_m = |\langle \mathbf{a}_m, \mathbf{x} \rangle|^2 as linear constraints b_m = \mathrm{trace}(\mathbf{A}_m \mathbf{X}), where \mathbf{A}_m = \mathbf{a}_m \mathbf{a}_m^* . This lifting converts the non-convex quartic objective into a convex semidefinite program (SDP) over the space of matrices. The SDP formulation minimizes \mathrm{trace}(\mathbf{X}) subject to \mathrm{trace}(\mathbf{A}_m \mathbf{X}) = b_m for m = 1, \dots, M, \mathbf{X} \succeq 0, with the non-convex rank-one constraint \mathrm{rank}(\mathbf{X}) = 1 relaxed by omitting it. The trace minimization promotes low-rank solutions while ensuring feasibility, and under suitable conditions, the optimal \mathbf{X} recovers the true \mathbf{x} \mathbf{x}^* exactly despite the relaxation. Introduced in the PhaseLift algorithm by Candès, Strohmer, and Voroninski in 2011, the method solves the SDP using interior-point or other convex optimization solvers, then extracts \mathbf{x} (up to global phase) as the leading eigenvector of the recovered \mathbf{X}, scaled by its leading eigenvalue. This approach provides global optimality for the lifted problem, contrasting with local-search heuristics. Recovery guarantees establish that for M = O(n) generic measurements, PhaseLift succeeds with high probability, yielding the exact signal provided the linear map defined by the \mathbf{A}_m satisfies a restricted isometry property (RIP) of order $2n with sufficiently small constant. Specifically, if the RIP constant \delta_{2n} < 0.04, the SDP solution is unique and rank-one. Although solvable in polynomial time via SDP solvers, PhaseLift's computational demands scale with O(n^2) decision variables and O(M n^2) problem data, rendering it practical only for moderate dimensions n \lesssim 1000 on standard hardware.

Short-Time Fourier Transform Phase Retrieval

The (STFT) provides a time-frequency representation of a signal x, defined as \text{STFT}(x)(t, \omega) = \int x(\tau) w(\tau - t) e^{-i \omega \tau} \, d\tau, where w is a localized in time. In discrete settings, this is computed on a time-frequency grid, yielding measurements of the squared, or , |\text{STFT}(x)|^2. Phase retrieval in the STFT domain involves recovering the original signal x from these measurements alone, leveraging the redundancy introduced by windowing and overlapping to compensate for the loss of phase information. This setup is particularly relevant for one-dimensional signals, such as audio, where the time-localized captures both temporal and spectral structure. To adapt semidefinite programming to STFT phase retrieval, the signal x is lifted to a rank-one positive semidefinite matrix X = x x^*, where the trace constraints encode the magnitude measurements. Specifically, for each time-frequency point (r, m), the constraint is \text{trace}(D_r F D_r^* X) = |\text{STFT}(x)|^2_{r,m}, with F denoting the Fourier transform matrix and D_r the diagonal matrix representing the shifted window. The optimization problem minimizes \text{trace}(X) subject to these quadratic constraints and X \succeq 0, ensuring a convex relaxation that can be solved efficiently for moderate signal lengths. This formulation exploits the structured measurements of the STFT, differing from generic linear measurements by incorporating the window shifts and Fourier basis. Recovery is achieved by taking the principal eigenvector of the solution \hat{X} as an estimate of x. Key algorithms for STFT phase retrieval include the semidefinite solver STliFT, which directly implements the above relaxation and provides theoretical recovery guarantees for non-vanishing signals under mild conditions, such as knowledge of a few prior samples or a hop size of one. An earlier iterative approach, the Griffin-Lim algorithm, alternates between enforcing the magnitudes and consistency in the via forward and inverse STFT operations, minimizing the to a modified STFT. While Griffin-Lim lacks global optimality guarantees, it converges to local minima and remains computationally efficient for audio applications. holds for generic windows, requiring only O(n) measurements for signals of length n, far fewer than the O(n^2) needed for full magnitudes. These methods gained prominence in the , building on foundational work to enable robust recovery in , where STFT phase retrieval facilitates tasks like and enhancement from magnitude spectrograms. The convex nature of STliFT offers stability over purely iterative methods, particularly for signals with sparse or bandlimited support, while the redundancy of STFT measurements enhances compared to oversampled phase retrieval.

Modern Methods

Non-Convex Optimization Approaches

Non-convex optimization approaches to phase retrieval address the limitations of convex relaxations, such as (SDP), by directly optimizing non-convex formulations of the amplitude-based loss function, enabling scalability to high-dimensional signals. These methods typically involve an initialization step followed by iterative , achieving global convergence guarantees under mild conditions on the number of measurements. Unlike SDP, which scales poorly with dimension due to its quadratic lifting, non-convex methods operate in the original signal space and require only O(n^2) time per iteration, making them suitable for large-scale problems. A seminal approach is the Wirtinger Flow algorithm, introduced in 2015, which formulates phase retrieval as minimizing the least-squares loss over measurements. The loss function is defined as L(\mathbf{x}) = \frac{1}{2M} \sum_{m=1}^M \left( b_m - |\langle \mathbf{a}_m, \mathbf{x} \rangle|^2 \right)^2, where \mathbf{x} \in \mathbb{C}^n is the signal to recover, \mathbf{a}_m \in \mathbb{C}^n are the measurement vectors, b_m = |\langle \mathbf{a}_m, \mathbf{x}_0 \rangle|^2 are the observed intensities for the true signal \mathbf{x}_0, and M is the number of measurements. The gradient of this loss is \nabla L(\mathbf{x}) = \frac{1}{M} \sum_{m=1}^M \left( \langle \mathbf{a}_m, \mathbf{x} \rangle \mathbf{a}_m \left( |\langle \mathbf{a}_m, \mathbf{x} \rangle|^2 - b_m \right) \right), and the update rule applies : \mathbf{x}^{t+1} = \mathbf{x}^t - \mu_t \nabla L(\mathbf{x}^t), with a decreasing step size \mu_t to ensure convergence. Initialization is performed via methods, computing the top eigenvector of the Y = \frac{1}{M} \sum_{m=1}^M b_m \mathbf{a}_m \mathbf{a}_m^H, which provides a crude estimate close to the true signal. This initialization draws from ideas in relaxations but avoids their computational overhead. Under Gaussian random measurements, Wirtinger Flow converges linearly to the global minimum with high probability when M = O(n \log n), recovering the signal up to a global . This sample complexity is near-optimal and matches information-theoretic limits, while the algorithm's iterations stabilize the recovery error exponentially fast after initialization. Empirical results demonstrate its effectiveness on and real images, with relative error dropping below $10^{-10} in hundreds of iterations for n \approx 10^4. Variants of Wirtinger Flow have extended its robustness and efficiency. Amplitude Flow, proposed in 2016, modifies the initialization and truncation to handle oversampled measurements more stably, using a truncated to mitigate outliers in the eigenvector estimate and achieving similar linear guarantees for M \geq O(n \log n). Another variant, Reshaped Wirtinger Flow from 2016, reshapes the loss landscape by incorporating a non-smooth penalty on measurement amplitudes, enabling faster rates and better performance under fewer measurements, inspired by network architectures for deeper optimization. These methods maintain the core framework while improving practical scalability for dimensions up to n = 10^6, outperforming convex alternatives in speed by orders of magnitude.

Deep Learning Techniques

Deep learning techniques for phase retrieval emerged prominently after 2018, employing convolutional neural networks (CNNs), generative adversarial networks (GANs), and unfolded iterative networks to directly map intensity measurements to . These methods leverage on simulated sets of intensity-phase pairs or real experimental , incorporating data-driven priors to handle ill-posed inversions more effectively than traditional iterative algorithms. By learning nonlinear mappings, such networks achieve faster convergence and robustness to measurement imperfections, often integrating physical models like transforms to ensure consistency with wave propagation laws. A foundational approach is prDeep, introduced in 2018, which combines a flexible CNN denoiser—based on a variational denoising framework similar to U-Net architectures—with iterative refinement for Fourier phase retrieval, demonstrating superior robustness to noise and partial Fourier coverage compared to classical methods. In 2020, physics-informed networks advanced the field through untrained deep networks that enforce measurement consistency without paired training data, enabling phase recovery from a single hologram by optimizing network parameters directly on the forward model, as exemplified in holographic imaging applications. More recent developments include unfolded Wirtinger flow networks from 2022, which unroll the Wirtinger gradient descent iterations into learnable layers with deep decoding priors, providing interpretable deep learning by blending optimization theory with neural components for phaseless imaging tasks. Training typically occurs in a supervised manner using pairs of measured intensities and ground-truth phases generated via simulations, though self-supervised strategies employ consistency losses between predicted phases and observed intensities to mitigate the need for labeled data. GAN-based variants, such as PhaseGAN (2021), further enable training on unpaired datasets by adversarially learning phase distributions that match physical constraints, enhancing generalization across diverse signal types like biological samples. These techniques offer key advantages, including resilience to noise and in partial measurements, real-time inference speeds suitable for dynamic imaging, and improved handling of complex nonlinearities in real-world data. For instance, advances from 2020 to 2025, such as the edge view enhanced phase retrieval (EVEPR) method (2025), integrate edge-detection priors into CNNs for sharper holographic reconstructions in phase-contrast microcomputed , reducing artifacts in biomedical applications. Additionally, generative priors in deep networks have reduced to linear order O(n) for n-dimensional signals, enabling recovery from fewer measurements than nonconvex optimization alone. Despite these benefits, challenges persist in generalization across unseen datasets, necessitating diverse training data to avoid to specific geometries, and in balancing model complexity with computational efficiency for deployment on edge devices.

Applications

Optical and Electron Imaging

Phase retrieval has played a pivotal role in optical imaging, notably in the diagnosis of the Hubble Space Telescope's primary mirror flaw during the 1990s. Launched in 1990, the telescope suffered from due to a error in the mirror's , resulting in a 0.4-wave wavefront error at 632.8 nm. Iterative phase retrieval algorithms were employed to analyze on-orbit point-spread functions from star images captured by the Faint Object Camera, enabling precise characterization of the aberration without physical access to the . This non-invasive technique confirmed the mirror's flaw and informed the design of corrective for the 1993 servicing mission, restoring the telescope's diffraction-limited performance to approximately 0.1 arcseconds resolution in the . In coherent diffractive imaging (CDI), phase retrieval reconstructs high-resolution images of non-crystalline specimens from far-field diffraction patterns, bypassing the need for lenses and enabling nanoscale 3D structural analysis. This lensless method is particularly valuable in X-ray free-electron laser (XFEL) facilities, where ultrashort pulses illuminate isolated particles or biological samples, producing diffraction data before sample destruction. Seminal implementations at facilities like the Linac Coherent Light Source have achieved resolutions down to 10-20 nm for biomolecules, such as viruses and proteins, by iteratively solving the phase problem using algorithms like the error reduction or hybrid input-output methods. CDI overcomes traditional resolution limits imposed by lens imperfections, providing quantitative phase contrast that reveals distributions with minimal . Ptychography extends phase retrieval to scanning electron microscopy, achieving atomic-resolution imaging through overlapping coherent illuminations and of the exit wave. By raster-scanning a focused beam across the sample and recording patterns at each position, recovers both and with redundancies that stabilize the reconstruction against and aberrations. A landmark 2021 demonstration using a scanning transmission reached 0.39 resolution for materials like MoS₂, limited primarily by lattice vibrations rather than instrumental factors, enabling visualization of atomic columns and defects in real and reciprocal space. This approach handles effectively by incorporating partial models and dose-efficient , reducing beam damage in sensitive samples while surpassing the information limit of conventional phase plates. Lensless on-chip leverages retrieval for compact, portable systems, particularly in resource-limited settings like point-of-care diagnostics. In these setups, a sample is illuminated and its shadow or pattern is captured directly on a , with computational recovery yielding quantitative images. A 2022 advancement introduced super-resolution techniques, such as accelerated Wirtinger optimization, to break the Nyquist-Shannon of the detector, achieving at least an order-of-magnitude faster convergence for reconstruction while maintaining image quality over a full for biological samples like sections. This method mitigates artifacts through multi-frame registration and complex-valued constraints, enabling high-throughput, aberration-free without bulky .

Signal Processing and Astronomy

In , phase retrieval plays a crucial role in recovering temporal signals from magnitude-only measurements, such as (STFT) spectrograms, enabling applications in audio synthesis and reconstruction. This is particularly vital for one-dimensional signals where information is lost during intensity detection, allowing reconstruction of the underlying for further processing or analysis. In astronomy, phase retrieval addresses similar challenges in imaging sparse celestial objects from interferometric data or aberrations in telescopes, improving and image quality without direct phase measurements. A seminal application in speech and audio processing is the Griffin-Lim algorithm, which iteratively reconstructs a time-domain signal from its modified STFT magnitude spectrogram by alternating between time and frequency domains to enforce consistency constraints. Introduced in 1984, this method has become the de facto standard for spectrogram inversion, particularly in neural vocoders for generating waveforms from mel-spectrograms in text-to-speech systems, where it minimizes reconstruction error through projections onto and consistency sets. For instance, in , Griffin-Lim enables high-fidelity audio output by recovering phases that align with perceptual features, though it can introduce artifacts in highly modified spectrograms, prompting extensions like accelerated variants for faster convergence. STFT phase retrieval extends to radar and communications, where it facilitates time-frequency analysis of modulated signals from phaseless measurements, aiding in design and target detection. In systems, algorithms recover ambiguous phases from STFT magnitudes to synthesize ambiguity functions, enabling precise range-Doppler estimation even under sparse sampling, with guarantees established for signals exceeding certain thresholds in the transform. Similarly, in communications, it supports blind recovery of transmitted symbols from received intensity spectra, enhancing in phase-insensitive channels like optical links. Blind deconvolution integrates phase retrieval to jointly estimate an unknown signal and its (PSF) from convolved intensity measurements, a bilinear prevalent in pipelines. Convex relaxations, such as lifted semidefinite programs, provide guarantees for unique recovery under structured priors like sparsity, by formulating the problem as trace minimization over low-rank matrices representing the outer product of signal and PSF. This approach has been applied to recover blurred audio or radar echoes, where phase retrieval resolves ambiguities in the that linear cannot. Iterative methods further accelerate convergence, treating phase retrieval as a special case where the PSF is the signal's . In astronomy, phase retrieval enhances imaging in radio telescopes by reconstructing phases from interferometric visibilities, compensating for atmospheric turbulence or baseline errors to form high-resolution maps of sparse sources like quasars. For the , phase-retrieval wavefront sensing characterized and corrected the primary mirror's pre- and post-repair, using focus-diverse point-spread functions to estimate phases via iterative algorithms, achieving sub-wavelength accuracy for diffraction-limited performance. These techniques rely on oversampled data from fields, enabling on-orbit alignment without specialized hardware. Recent advances as of 2025 incorporate generative models to regularize phase retrieval for sparse signal in astronomy, particularly in interferometry where phases are inaccessible. Diffusion-based priors, trained on simulated data, iteratively denoise phaseless measurements to reconstruct of extended sources, outperforming traditional methods in low-signal regimes by enforcing astrophysical sparsity. For example, transformer-generative frameworks have been applied to Bayesian inverse problems in high-dimensional radio , enabling scalable of transient events from sparse arrays.

References

  1. [1]
    [PDF] Phase Retrieval: From Computational Imaging to Machine Learning
    Abstract—Phase retrieval consists in the recovery of a complex-valued signal from intensity-only measurements. As it pervades a broad variety of applications, ...
  2. [2]
    [PDF] Phase Retrieval with Application to Optical Imaging
    This review article provides a contemporary overview of phase retrieval in optical imaging, linking the relevant optical physics to the signal processing ...
  3. [3]
    (PDF) Phase Retrieval: An Overview of Recent Developments
    May 16, 2016 · The phase retrieval problem is a fundamental inverse problem in applied mathematics, physics, and engineering. The objective is to reconstruct a ...<|control11|><|separator|>
  4. [4]
    [PDF] Phase Retrieval via Matrix Completion
    The phase retrieval problem consists in finding x from the magnitude coefficients |x[ω]|, ω ∈ Ω. When Ω is the usual frequency grid as above and without further ...
  5. [5]
    [PDF] Phase retrieval algorithms: a comparison - University Lab Sites
    Iterative algorithms for phase retrieval from intensity data are compared to gradient search methods. Both the problem of phase retrieval from two intensity ...
  6. [6]
    Phase Retrieval: Uniqueness and Stability | SIAM Review
    We aim to summarize our current understanding of uniqueness and stability properties of phase retrieval problems.<|control11|><|separator|>
  7. [7]
    Phase Retrieval via Wirtinger Flow: Theory and Algorithms - arXiv
    Jul 3, 2014 · This paper develops a non-convex formulation of the phase retrieval problem as well as a concrete solution algorithm.
  8. [8]
    [PDF] Phase retrieval from the magnitude of the Fourier transforms of ...
    The idea of oversampling is to sample the magnitude of a. Fourier transform finely enough to get a finite support for the object in which the pixel value ...
  9. [9]
    phase retrieval from 4n–4 measurements - ResearchGate
    Introduction. Phase retrieval is the recovery (up to a global phase factor) of signals from intensity measurements, i.e. from the absolute values of ...
  10. [10]
    Ambiguities in the X-Ray Analysis of Crystal Structures | Phys. Rev.
    In this paper a large number of cases is presented in which two, and sometimes three or four, non-congruent sets of points are homometric, ie, have the same ...Missing: crystallography | Show results with:crystallography
  11. [11]
  12. [12]
    The Question of Phase Retrieval in Optics - Taylor & Francis Online
    Nov 11, 2010 · In this paper the phase reconstruction is studied under the constraint that the aperture of the lens be finite.
  13. [13]
    A practical algorithm for the determination of phase from image and ...
    An algorithm is presented for the rapid solution of the phase of the complete wave function whose intensity in the diffraction and imaging planes of an ...Missing: original | Show results with:original
  14. [14]
    Ptychography: A solution to the phase problem - Physics Today
    Sep 1, 2021 · Ptychography was introduced in 1969 in a three-part paper by Walter Hoppe, an influential figure in the field of electron microscopy. (For more ...
  15. [15]
    Phase Retrieval from Coded Diffraction Patterns - Semantic Scholar
    Oct 11, 2013 · Coded aperture design for solving the phase retrieval problem in X-ray crystallography ... 1990. TLDR. The principles of phase retrieval in ...
  16. [16]
    PhaseLift: Exact and Stable Signal Recovery from Magnitude ... - arXiv
    Sep 21, 2011 · This novel result demonstrates that in some instances, the combinatorial phase retrieval problem can be solved by convex programming techniques.
  17. [17]
    [PDF] Fourier Phase Retrieval: Uniqueness and Algorithms - arXiv
    Nov 6, 2017 · The aim of this section is to survey several approaches to ensure uniqueness of the discrete phase retrieval problem.
  18. [18]
    STFT Phase Retrieval: Uniqueness Guarantees and Recovery ...
    Aug 12, 2015 · We first develop conditions under which the STFT magnitude is an almost surely unique signal representation. We then consider a semidefinite ...Missing: theorem | Show results with:theorem
  19. [19]
    Phase-retrieval stagnation problems and solutions
    Stagnation of the algorithm means that the output image changes very little after many further iterations while not at a solution. A solution is any Fourier- ...
  20. [20]
    Legacies of the Gerchberg–Saxton algorithm - ScienceDirect.com
    The first solution of the phase problem was obtained in 1972 by Gerchberg and Saxton and has given rise to a huge literature [33,46,47]. In the 1970s Walter ...
  21. [21]
    X-ray image reconstruction from a diffraction pattern alone
    Oct 28, 2003 · X-ray image reconstruction from a diffraction pattern alone. S ... Marchesini, M. Howells, U. Weierstall, G. Hembree, and J.C.H. Spence ...
  22. [22]
    None
    ### Summary of the Shrinkwrap Algorithm
  23. [23]
    [PDF] Iterative phase retrieval in coherent diffractive imaging: practical issues
    2.3.2 Hybrid input-output algorithm (HIO). The hybrid input-output algorithm (HIO) [2] is obtained from ER algorithm by modifying the constraint in the ...
  24. [24]
    Phase Retrieval via Matrix Completion - SIAM Publications Library
    Walther, The question of phase retrieval in optics, Optical Acta, 10 (1963), pp. ... Phase Transitions in Phase Retrieval. Excursions in Harmonic Analysis ...<|control11|><|separator|>
  25. [25]
    Phase retrieval with PhaseLift algorithm | Applied Mathematics-A ...
    Dec 28, 2020 · This paper provides a contemporary overview of phase retrieval problem with PhaseLift algorithm and summarizes theoretical results which have been derived ...
  26. [26]
  27. [27]
    [1605.07719] Reshaped Wirtinger Flow and Incremental Algorithm ...
    May 25, 2016 · We study the phase retrieval problem, which solves quadratic system of equations, i.e., recovers a vector \boldsymbol{x}\in \mathbb{R}^n from ...
  28. [28]
    On the use of deep learning for phase recovery | Light - Nature
    Jan 1, 2024 · In this review, we first briefly introduce conventional methods for PR. Then, we review how DL provides support for PR from the following three stages.
  29. [29]
    prDeep: Robust Phase Retrieval with a Flexible Deep Network - arXiv
    Mar 1, 2018 · A new phase retrieval algorithm that is both robust and broadly applicable. We test and validate prDeep in simulation to demonstrate that it is robust to noise.
  30. [30]
    Development of a deep learning method for phase retrieval image ...
    May 13, 2025 · This study focuses on the development of a deep learning method, called edge view enhanced phase retrieval (EVEPR), that incorporates the ...Abstract · INTRODUCTION · RESULTS · DISCUSSION
  31. [31]
    Optimal Sample Complexity with Deep Generative Priors - arXiv
    Aug 24, 2020 · In this work, we resolve the open problem in compressive phase retrieval and demonstrate that generative priors can lead to a fundamental ...Missing: Efficient | Show results with:Efficient
  32. [32]
    Hubble's Mirror Flaw - NASA Science
    When the Wide Field and Planetary Camera 2 (WFPC2) team learned of the problem with Hubble's primary mirror, they started researching ways to correct the flaw ...
  33. [33]
    HST phase retrieval: a parameter estimation - NASA/ADS
    It was found, on orbit, that the Hubble Space Telescope had a conic constant error in the primary mirror. The result of the error is a substantial amount of ...Missing: 1990s | Show results with:1990s
  34. [34]
    In situ coherent diffractive imaging | Nature Communications
    May 8, 2018 · Coherent diffractive imaging (CDI) has been widely applied in the physical and biological sciences using synchrotron radiation, X-ray ...
  35. [35]
    Coherent diffraction imaging of cells at advanced X-ray light sources
    The single-pulse imaging at X-ray free electron laser has for the first time obtained nanoscale structural images of living cells in experiment, providing ...
  36. [36]
    [PDF] Coherent X-ray Diffraction Imaging - UCLA Physics & Astronomy
    In this paper, we will briefly dis- cuss the principle of coherent diffraction imaging, present various implementation schemes of this imaging modality, and ...
  37. [37]
    Electron ptychography achieves atomic-resolution limits set by ...
    May 21, 2021 · Ptychography is another phase-retrieval approach that dates back to Hoppe's work in the 1960s (9). Modern, robust setups use multiple ...
  38. [38]
    Low-dose phase retrieval of biological specimens using cryo ...
    Jun 2, 2020 · Here we demonstrate that electron ptychography can recover the phase of the specimen with continuous information transfer across a wide range of the spatial ...
  39. [39]
    Factors limiting quantitative phase retrieval in atomic-resolution ...
    Factors limiting quantitative phase retrieval include coherent and incoherent lens aberrations, detector positioning, uniformity, scan-distortion, noise, and ...Missing: plates | Show results with:plates
  40. [40]
    Pixel Super-Resolution Phase Retrieval for Lensless On-Chip ... - NIH
    In this work, we introduce accelerated Wirtinger flow (AWF) as a unified framework for pixel super-resolution phase retrieval. Based on the proximal gradient ...
  41. [41]
    High-fidelity pixel-super-resolved lensless on-chip microscopy via ...
    The proposed method achieves imaging resolution beyond the Nyquist–Shannon sampling limit, providing a 1.26-fold improvement in resolution across a full FOV of ...
  42. [42]
    [PDF] arXiv:2205.05496v1 [eess.AS] 11 May 2022
    May 11, 2022 · The de-facto standard for STFT phase retrieval in the context of audio signals is the Griffin-Lim algorithm (GLA) [3], which is itself inspired ...<|separator|>
  43. [43]
    Phase retrieval algorithms: a personal tour [Invited]
    Dec 21, 2012 · At the time (early 1972) I was simply photographing pictures of holograms out of the book Optical Holography (see [3], Figs. 19.8 and 19.11), ...
  44. [44]
    [PDF] Accelerated Griffin-Lim algorithm: A fast and provably converging ...
    We compared AGLA to two other algorithms, which were observed to perform well in the STFT phase retrieval in [30]. The first algorithm is the Relaxed. Averaged ...
  45. [45]
    Multi-source phase retrieval from multi-channel phaseless STFT ...
    In this paper, we extend this work and formulate a multi-source phase retrieval problem where multi-channel phaseless STFT measurements are given as input. We ...
  46. [46]
    Simultaneous Phase Retrieval and Blind Deconvolution via Convex ...
    Our method is a novel convex relaxation that is based on a lifted matrix recovery formulation that allows a non-trivial convex relaxation of the bilinear ...
  47. [47]
    [PDF] Iterative blind deconvolution algorithm - University Lab Sites
    phase retrieval can be considered a special case of blind deconvolution, in which we deconvolve f(x) and f* (-x) from r(x). Because the AD algorithm ...
  48. [48]
    [PDF] Phase-Retrieval Wave-Front Sensing for_the Hubble and Future ...
    Jan 17, 2011 · Phase retrieval is used for wavefront sensing and control, allowing for design correction of optics, and to reconstruct the field in the pupil.
  49. [49]
    Hubble Space Telescope characterized by using phase-retrieval ...
    We describe several results characterizing the Hubble Space Telescope from measured point spread functions by using phase-retrieval algorithms.
  50. [50]
    using generative models for regularization of the phase retrieval ...
    Sep 2, 2025 · In this paper we consider the numerical solution of a phase retrieval problem for a compactly supported, linear spline with the Fourier ...
  51. [51]
    High-dimensional Bayesian Astronomical Imaging and Inverse ...
    Recent progress in machine learning and generative modeling has opened new avenues to tackle previously insoluble high-dimensional inverse problems in astronomy ...<|control11|><|separator|>