Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2609.24915

Fisher-information training of optical sensing front ends from natural fluctuations

Abhinav Sinha, Kai Wang

physics.opticsquant-phSubmitted Sep 21, 2026• Announced Sep 22, 2026• 0 citations
arXivHTMLPDF

Abstract

Fisher-information-based optimization of programmable optical front ends in situ generally requires a response model or parameter-labeled measurements. We show that natural parameter fluctuations can instead provide the training signal. We construct proxies for the Fisher information (FI) of one parameter and the prior-weighted FI spectrum of multiple parameters from measured covariances after subtracting conditional detection noise, assuming locally affine responses. We use simulated photon counts from two incoherent point sources to train a mode sorter that approaches the quantum limit for separation and the Nagaoka--Hayashi single-copy bound for joint centroid--separation estimation. We demonstrate count-only updates that require no instrument response model and achieve performance comparable to model-based gradients. Our results provide a route to adaptive optical measurements without controlled parameter scans.

AI Summary

The paper asks whether programmable optical measurements can be trained to maximize Fisher information (FI) without a response model, parameter-labeled data, or controlled parameter scans. It uses naturally fluctuating parameters as the training signal: subtracting conditional detection noise from measured outcome-fraction covariances isolates parameter-induced covariance. For locally affine responses, probability whitening connects this covariance to FI. For multiple parameters, its nonzero eigenvalues approximate those of $\Sigma_{\rm pr}^{1/2}F\Sigma_{\rm pr}^{1/2}$, enabling an A-optimal objective approximating $\operatorname{Tr}(\Sigma_{\rm pr}^{-1}F^{-1})$. This recovers a weighted spectrum, not the full FI matrix. The method is tested using simulated multinomial photon counts from two equally bright incoherent point sources with Gaussian point-spread functions. Programmable mode sorters are trained using either model-based gradients or count-only simultaneous-perturbation updates. For separation estimation, two detection outcomes attain at least $99.9\%$ of the quantum FI with model-based gradients and $94.1$–$99.9\%$ with count-only updates. For joint centroid–separation estimation, nine outcomes attain $97.6$–$99.8\%$ and $98.0$–$99.4\%$, respectively, of the Nagaoka–Hayashi single-copy benchmark, assessed using the exact FI of the trained measurement. All demonstrations are numerical. Operation requires approximately fixed parameters within each shot, stable fluctuation statistics across training evaluations, and modeled or calibrated conditional noise. Narrower fluctuations reduce nonlinear bias but weaken the covariance signal; larger batches reduce sampling error, not systematic bias. The proxy becomes inaccurate near stationary responses. Count-only training requires larger photon budgets, and information in parameter-dependent total detection rates is excluded.

No sign of AI-generated textAI use disclosedStrong assumptions
Claude Opus 5 (Anthropic) was used to write, modify, and optimize the simulation code behind the numerical results and figures reported here;

Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.

Comments
11 pages main text, 20 pages supplemental material

Related papers

Quantum-Limited Time-Frequency Multiparameter Estimation through Mode-Selective Photon Measurement

Vahid Ansari, John M. Donohue, Jano Gil-López +7 more

In optics, when attempting to find the distance separations s between two incoherent emitters, it is standard practice to take a full image and extract the single parameter s from it. In such scenarios, the smallest precisely measurable separation is limited by the point-spread function on the image plane, with a vanishingly small amount of information available for smaller and smaller separations especially when photon shot noise is the dominant noise source (as in, for example, astronomical observations). This limit can be formalised by the Fisher information, which quantifies the amount of information accessible per photon detection and is directly associated with Cramér-Rao lower bound. For direct intensity imaging, the Fisher information drops to zero for object separations smaller than the spread of the optical field, sometimes referred to as the Rayleigh's curse, which limits the usefulness of photon counting for metrology.

Conference on Lasers and Electro-Optics Europe & European Quantum Electronics Conference (CLEO/Europe-EQEC)•Jun 1, 2019•0 citations

On super additivity of Fisher information in fully Gaussian metrology

Javier Navarro, Simon Morelli, Mikel Sanz +1 more

Famously, the quantum Fisher information -- the maximum Fisher information over all physical measurements -- is additive for independent copies of a system and the optimal measurement acts locally. We are left to wonder: does the same hold when the set of accessible measurements is constrained? Such constraints are necessary to account for realistic experimental restrictions. Here, we consider a fully Gaussian scenario focusing on only Gaussian measurements. We prove that the optimal Gaussian measurement protocol remains local, if the information is encoded in either the displacement or the covariance matrix. However, when the information is imprinted on both, this no longer holds true: we construct a simple global Gaussian measurement where the Fisher information becomes super additive. These results can improve parameter estimation tasks via feasible tools. Namely, in quantum optical platforms our proposed global operation requires only passive global operations and single mode Gaussian measurements. We demonstrate this in two examples where we estimate squeezing and losses. While in the former case there is a significant gap between the Fisher information of the optimal Gaussian measurement and the quantum Fisher information for a single copy, this gap can be reduced with joint Gaussian measurements and closed in the asymptotic limit of many copies.

quant-pharXiv:2512.20534•Dec 24, 2025•6 citations

Quantum-Inspired Multi-Parameter Adaptive Bayesian Estimation for Sensing and Imaging

Kwan K. Lee, Christos N. Gagatsos, Saikat Guha +1 more

It is well known in Bayesian estimation theory that the conditional estimator attains the minimum mean squared error (MMSE) for estimating a scalar parameter of interest. In quantum, e.g., optical and atomic, imaging and sensing tasks the user has access to the quantum state that encodes the parameter. The choice of a measurement operator, i.e. a positive-operator valued measure (POVM), leads to a measurement outcome on which the aforesaid classical MMSE estimator is employed. Personick found the optimum POVM that attains the MMSE over all possible physically allowable measurements and the resulting MMSE (Personick, 1997). This result from 1971 is less-widely known than the quantum Fisher information (QFI), which lower bounds the variance of an unbiased estimator over all measurements without considering any prior probability. For multi-parameter estimation, in quantum Fisher estimation theory the inverse of the QFI matrix provides an operator lower bound on the covariance of an unbiased estimator, and this bound is understood in the positive semidefinite sense. However, there has been little work on quantifying the quantum limits and measurement designs, for multi-parameter quantum estimation in aBayesiansetting. In this work, we build upon Personick's result to construct a Bayesian adaptive (greedy) measurement scheme for multi-parameter estimation. We illustrate our proposed measurement scheme with the application of localizing a cluster of point emitters in a highly sub-Rayleigh angular field-of-view, an important problem in fluorescence microscopy and astronomy. Our algorithm translates to a multi-spatial-mode transformation prior to a photon-detection array, with electro-optic feedback to adapt the mode sorter. We show that this receiver performs superior to quantum-noise-limited focal-plane direct imaging.

arXiv:2202.09980•IEEE Journal of Selected Topics in Signal Processing•Oct 14, 2022•34 citations

Attaining Fundamental Limits of Multiparameter Incoherent Optical Imaging Using Joint-Detection Quantum Measurements

Nico Deshler, Aakash Warke, Michael R. Grace +2 more

Resolving extended incoherent objects below the diffraction limit poses an application-rich imaging challenge whose solution may enable a new generation of observational instruments and capabilities. In this work, we invoke a practical model for general imaging by approximating an arbitrary extended incoherent object as a finite grid of thermal point emitters parameterized by their brightnesses. We derive the quantum Fisher information matrix (QFIM) for simultaneous brightness estimation and show that the symmetric logarithmic derivatives weakly commute, indicating that the Helstrom bound furnishes the ultimate quantum limit on the estimation error for incoherent imaging. Furthermore, for deeply sub-diffraction scenes, we find numerical evidence of a gap between the Nagaoka-Hayashi (NH) bound and the Helstrom bound. This gap reveals that separable measurements, though more experimentally accessible, are insufficient to reach the quantum limit, and points to the prospective advantage of joint measurements acting on multiple state copies. Additionally, we show that spatial mode-demultiplexing (SPADE) often saturates the NH bound solidifying its status as a near-optimal separable measurement strategy that significantly outperforms direct imaging. Finally, we articulate two joint detection receivers implemented with bona fide quantum resources that asymptotically achieve the Helstrom bound.

quant-pharXiv:2608.19524•Aug 21, 2026•4 citations

Performance-tradeoff relation for locating two incoherent optical point sources

Jingjing Shao, Xiao-Ming Lu

The optimal quantum measurements for estimating individual parameters might be incompatible with each other, so they cannot be jointly performed. The tradeoff between the estimation precisions for different parameters can be characterized by information regret: the difference between the Fisher information and its quantum limit. We show that the information-regret-tradeoff relation can give us not only an intuitive picture of the potential for improving the joint scheme for estimating the centroid and the separation but also some clues to the optimal measurements for the sequential scheme. In particular, we show that, for two incoherent point sources with a very small separation, the optimal measurement for the separation must extract little information about the centroid and vice versa.

arXiv:2202.05647•Physical Review A•Jun 9, 2022•8 citations

Fisher-Optimized Multiplane Light Conversion for Quantum Parameter Estimation

Abhinav Sinha, Kai Wang

We propose a self-optimizing photonic neural network based on multiplane light conversion (MPLC) for quantum parameter estimation. Unlike static mode sorters, our system adaptively learns measurement strategies by maximizing Classical Fisher Information (CFI) employing a minimal number of detectors using a simultaneous perturbation stochastic approximation algorithm. We demonstrate this on the two-point source separation problem. Our approach achieves near-optimal precision and maintains robustness against centroid misalignment, outperforming standard direct imaging and spatial mode demultiplexing methods.

Photonics North (PN)•Jun 1, 2026•0 citations
Fisher-information-based optimization of programmable optical front ends in situ generally requires a response model or parameter-labeled measurements. We show that natural parameter fluctuations can instead provide the training signal. We construct proxies for the Fisher information (FI) of one parameter and the prior-weighted FI spectrum of multiple parameters from measured covariances after subtracting conditional detection noise, assuming locally affine responses. We use simulated photon counts from two incoherent point sources to train a mode sorter that approaches the quantum limit for separation and the Nagaoka--Hayashi single-copy bound for joint centroid--separation estimation. We demonstrate count-only updates that require no instrument response model and achieve performance comparable to model-based gradients. Our results provide a route to adaptive optical measurements without controlled parameter scans.
The paper asks whether programmable optical measurements can be trained to maximize Fisher information (FI) without a response model, parameter-labeled data, or controlled parameter scans. It uses naturally fluctuating parameters as the training signal: subtracting conditional detection noise from measured outcome-fraction covariances isolates parameter-induced covariance. For locally affine responses, probability whitening connects this covariance to FI. For multiple parameters, its nonzero eigenvalues approximate those of Σpr1/2FΣpr1/2\Sigma_{\rm pr}^{1/2}F\Sigma_{\rm pr}^{1/2}Σpr1/2​FΣpr1/2​, enabling an A-optimal objective approximating Tr⁡(Σpr−1F−1)\operatorname{Tr}(\Sigma_{\rm pr}^{-1}F^{-1})Tr(Σpr−1​F−1). This recovers a weighted spectrum, not the full FI matrix.
The method is tested using simulated multinomial photon counts from two equally bright incoherent point sources with Gaussian point-spread functions. Programmable mode sorters are trained using either model-based gradients or count-only simultaneous-perturbation updates. For separation estimation, two detection outcomes attain at least 99.9%99.9\%99.9% of the quantum FI with model-based gradients and 94.194.194.1–99.9%99.9\%99.9% with count-only updates. For joint centroid–separation estimation, nine outcomes attain 97.697.697.6–99.8%99.8\%99.8% and 98.098.098.0–99.4%99.4\%99.4%, respectively, of the Nagaoka–Hayashi single-copy benchmark, assessed using the exact FI of the trained measurement.
All demonstrations are numerical. Operation requires approximately fixed parameters within each shot, stable fluctuation statistics across training evaluations, and modeled or calibrated conditional noise. Narrower fluctuations reduce nonlinear bias but weaken the covariance signal; larger batches reduce sampling error, not systematic bias. The proxy becomes inaccurate near stationary responses. Count-only training requires larger photon budgets, and information in parameter-dependent total detection rates is excluded.
In optics, when attempting to find the distance separations s between two incoherent emitters, it is standard practice to take a full image and extract the single parameter s from it. In such scenarios, the smallest precisely measurable separation is limited by the point-spread function on the image plane, with a vanishingly small amount of information available for smaller and smaller separations especially when photon shot noise is the dominant noise source (as in, for example, astronomical observations). This limit can be formalised by the Fisher information, which quantifies the amount of information accessible per photon detection and is directly associated with Cramér-Rao lower bound. For direct intensity imaging, the Fisher information drops to zero for object separations smaller than the spread of the optical field, sometimes referred to as the Rayleigh's curse, which limits the usefulness of photon counting for metrology.
Famously, the quantum Fisher information -- the maximum Fisher information over all physical measurements -- is additive for independent copies of a system and the optimal measurement acts locally. We are left to wonder: does the same hold when the set of accessible measurements is constrained? Such constraints are necessary to account for realistic experimental restrictions. Here, we consider a fully Gaussian scenario focusing on only Gaussian measurements. We prove that the optimal Gaussian measurement protocol remains local, if the information is encoded in either the displacement or the covariance matrix. However, when the information is imprinted on both, this no longer holds true: we construct a simple global Gaussian measurement where the Fisher information becomes super additive. These results can improve parameter estimation tasks via feasible tools. Namely, in quantum optical platforms our proposed global operation requires only passive global operations and single mode Gaussian measurements. We demonstrate this in two examples where we estimate squeezing and losses. While in the former case there is a significant gap between the Fisher information of the optimal Gaussian measurement and the quantum Fisher information for a single copy, this gap can be reduced with joint Gaussian measurements and closed in the asymptotic limit of many copies.
It is well known in Bayesian estimation theory that the conditional estimator attains the minimum mean squared error (MMSE) for estimating a scalar parameter of interest. In quantum, e.g., optical and atomic, imaging and sensing tasks the user has access to the quantum state that encodes the parameter. The choice of a measurement operator, i.e. a positive-operator valued measure (POVM), leads to a measurement outcome on which the aforesaid classical MMSE estimator is employed. Personick found the optimum POVM that attains the MMSE over all possible physically allowable measurements and the resulting MMSE (Personick, 1997). This result from 1971 is less-widely known than the quantum Fisher information (QFI), which lower bounds the variance of an unbiased estimator over all measurements without considering any prior probability. For multi-parameter estimation, in quantum Fisher estimation theory the inverse of the QFI matrix provides an operator lower bound on the covariance of an unbiased estimator, and this bound is understood in the positive semidefinite sense. However, there has been little work on quantifying the quantum limits and measurement designs, for multi-parameter quantum estimation in aBayesiansetting. In this work, we build upon Personick's result to construct a Bayesian adaptive (greedy) measurement scheme for multi-parameter estimation. We illustrate our proposed measurement scheme with the application of localizing a cluster of point emitters in a highly sub-Rayleigh angular field-of-view, an important problem in fluorescence microscopy and astronomy. Our algorithm translates to a multi-spatial-mode transformation prior to a photon-detection array, with electro-optic feedback to adapt the mode sorter. We show that this receiver performs superior to quantum-noise-limited focal-plane direct imaging.
Resolving extended incoherent objects below the diffraction limit poses an application-rich imaging challenge whose solution may enable a new generation of observational instruments and capabilities. In this work, we invoke a practical model for general imaging by approximating an arbitrary extended incoherent object as a finite grid of thermal point emitters parameterized by their brightnesses. We derive the quantum Fisher information matrix (QFIM) for simultaneous brightness estimation and show that the symmetric logarithmic derivatives weakly commute, indicating that the Helstrom bound furnishes the ultimate quantum limit on the estimation error for incoherent imaging. Furthermore, for deeply sub-diffraction scenes, we find numerical evidence of a gap between the Nagaoka-Hayashi (NH) bound and the Helstrom bound. This gap reveals that separable measurements, though more experimentally accessible, are insufficient to reach the quantum limit, and points to the prospective advantage of joint measurements acting on multiple state copies. Additionally, we show that spatial mode-demultiplexing (SPADE) often saturates the NH bound solidifying its status as a near-optimal separable measurement strategy that significantly outperforms direct imaging. Finally, we articulate two joint detection receivers implemented with bona fide quantum resources that asymptotically achieve the Helstrom bound.
The optimal quantum measurements for estimating individual parameters might be incompatible with each other, so they cannot be jointly performed. The tradeoff between the estimation precisions for different parameters can be characterized by information regret: the difference between the Fisher information and its quantum limit. We show that the information-regret-tradeoff relation can give us not only an intuitive picture of the potential for improving the joint scheme for estimating the centroid and the separation but also some clues to the optimal measurements for the sequential scheme. In particular, we show that, for two incoherent point sources with a very small separation, the optimal measurement for the separation must extract little information about the centroid and vice versa.
We propose a self-optimizing photonic neural network based on multiplane light conversion (MPLC) for quantum parameter estimation. Unlike static mode sorters, our system adaptively learns measurement strategies by maximizing Classical Fisher Information (CFI) employing a minimal number of detectors using a simultaneous perturbation stochastic approximation algorithm. We demonstrate this on the two-point source separation problem. Our approach achieves near-optimal precision and maintains robustness against centroid misalignment, outperforming standard direct imaging and spatial mode demultiplexing methods.