Experimental subdiffraction source discrimination enabled by spatial demultiplexing and single-photon detectors

arXiv:2605.15929 · quant-ph · Submitted 2026-05-15 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: We're looking at the paper: "Experimental subdiffraction source discrimination enabled by spatial demultiplexing and single-photon detectors".

Mira: Here's the gist of the paper. The gist The authors experimentally demonstrate a universal, parameter-independent test for asymmetric source discrimination using spatial mode demultiplexing (SPADE) and single-photon detectors.

Kai: Experimental Setup and Methodology The experiment is supported by a complete theory on binary decision for exoplanet detection, or more in general on asymmetric HT between 1-vs-2 incoherent sources, including crosstalk.

Mira: The test implemented is the same as Ref.

37: , that is universal as it does not depend on the physical parameters.

Title and authors: Kai: So we're looking at the paper titled "Experimental subdiffraction source discrimination enabled by spatial demultiplexing and single-photon detectors." It's about using this spatial mode demultiplexing thing and single-photon detectors to tell faint sources apart even when they are too close for traditional imaging.

Mira: Yeah, it sounds like they’re tackling that subdiffraction limit directly, which is tough because you usually need a lot of light to see those tiny separations. This paper suggests a way around that using SPADE and single-photon detectors.

Lev: From an engineering standpoint, I'm interested in how they set up this tabletop experiment. Are we talking about some complex optical path or something that’s really hard to build reliably?

Kai: It is a tabletop setup, inspired by exoplanet detection, where you have two light sources mimicking a star and an exoplanet. They use fiber-coupled LEDs for this, collimated with fiber collimators and a two-lens system to match the beam waist w0 of the demultiplexer they chose.

Mira: And the core idea they're testing is a universal test for asymmetric source discrimination that doesn't depend on any specific physical parameters of the star-exoplanet system, which is pretty ambitious.

Lev: That universality is key, because it means whatever you set up in terms of separation distance or intensity ratio, this test should still work optimally.

Kai: They model arbitrary modal crosstalk in their theory and collect data across a range of source separations and intensity ratios to verify that independence from those physical parameters.

Mira: And they show the test beats the subdiffraction limit because it achieves a type II error rate that is lower than what you get from direct imaging.

The paper's summary: Kai: To summarize, this paper presents a complete theory on binary decision for exoplanet detection, which generalizes to asymmetric hypothesis testing between one source versus two incoherent sources including crosstalk.

Mira: They use quantum detection theory formalism here to represent the absence hypothesis as density matrix rho0 and the presence hypothesis as rho1, which sets up the framework for their test.

Lev: That density matrix approach sounds heavy, but I wonder how much real overhead that adds when you try to run it on actual hardware with finite fidelity.

Kai: They show that for an asymptotically large number N of repetitions, the quantum optimal test using the quantum Stein lemma attains a rate of false negatives beta approximately equal to exp negative N D Q(rho0 rho1) where DQ is related to epsilon squared d two over four times one.

Mira: The crosstalk effect is modeled by an incoherent mixing of outcome probabilities p prime j through a stochastic matrix p prime j equals C00 C01 C10 C11 P(jHi), which then leads to the relative entropy between the two crosstalk-affected Bernoulli distributions becoming DSD(p'0p'one) approximately epsilon squared d four over three times thirty times one over C zero one <ref:2605.15929#pg2>.

Lev: That final approximation, that DSD is valid when epsilon squared d two a is much smaller than C zero one much smaller than one, which tells me the results rely on some assumptions about how small the physical separation and intensity ratio need to be for this math to hold up.

Kai: They then compare these results with the expected direct imaging based hypothesis test and with the ultimate quantum limit, showing that SPADE detection DSD in Eq. (five) can actually be smaller than DSD in Eq <ref:2605.15929#pg3>. (one) for certain crosstalk values <ref:2605.15929#pg1>.

Mira: The main point they land on is finding a quantitative threshold for crosstalk, Cth approximately zero point one below which SPADE outperforms direct imaging, which is pretty interesting because it’s so specific.

The paper's improvements: Kai: They suggest that the system can achieve subdiffraction source discrimination well beyond the diffraction limit by exploiting spatial mode demultiplexing and single-photon detectors, as SPADE "retains an advantage over direct imaging in the relevant regime of small separations and low intensity ratios."

Mira: The improvement they highlight is identifying an experimentally accessible crosstalk threshold Cth which is well below what can be achieved by diffraction-limited direct imaging, meaning we can actually get this advantage with current technology.

Lev: So, if I'm thinking about real hardware, the implication here is that we don't need near-perfect components to see the benefit; a crosstalk of zero point one is achievable in their setup.

Kai: Exactly. They quantify it by saying that for example, with a crosstalk of one minus two over two, SPADE needs up to one order of magnitude fewer photons than direct imaging to achieve the same error rate.

Mira: That's a significant photon budget reduction because if you are in a photon-starved regime, that means you can get the same result with less light input.

Lev: That makes sense for hardware constraints; reducing the required photon count is always good for experimental setups where detector efficiency might be limited.

Conclusion: Kai: So to wrap up, the paper "Experimental subdiffraction source discrimination enabled by spatial demultiplexing and single-photon detectors" shows that this test is robust to realistic imperfections and doesn't rely on prior knowledge of the source parameters.

Mira: The main implication is that for practical subdiffraction imaging scenarios, especially in photon-limited regimes, this approach becomes relevant because it works even with imperfections.

Lev: From an engineering view, we need to remember that the paper points out a limitation: they say that neglecting small quantities of orders of epsilon squared and d four over a leads to the same expression for DSD(p'v0p'v1) approximately epsilon squared d four over three times twenty times one over C zero one.

Kai: So, while the math is simple, it confirms that SPADE detection in this paper achieves the optimal exponential scaling of the false-negative error rate in the large-sample regime.

Mira: Overall, finding a crosstalk threshold Cth approximately zero point one separating SPADE and direct imaging regimes is a concrete result that helps us understand when to use which method.

Lev: For me, what we see is that this test would be really useful for guiding future experimental design toward systems where they can actually aim for that crosstalk level of zero point one <ref:2605.15929#pg2>.

Agenzia Spaziale Italiana ASI · Dipartimento Interateneo di Fisica Politecnico di Bari Department of Physics Politecnico di Bari Institute Nazionale di Fisica Nucleare INFN Sezione di Bari

quant-ph

Submitted: 2026-05-15

Updated: 2026-10-21

Comments: 10 pages, 3 figures

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: The gist The authors experimentally demonstrate a universal, parameter-independent test for asymmetric source discrimination using spatial mode demultiplexing (SPADE) and single-photon detectors.

Key concepts

Spatial Mode Demultiplexing (SPADE): SPADE is a technique used in the experiment to separate light beams based on their spatial modes. It works by combining two input beams and using a demultiplexer to spatially resolve them, allowing the system to distinguish between two sources even when they are close together, which is crucial for subdiffraction discrimination.
Asymmetric Hypothesis Testing: This refers to a statistical test designed to determine whether one source (the exoplanet) is present or absent against another (the star). The test is specifically set up to handle situations where the two sources are not identical, allowing researchers to distinguish between the 'presence' and 'absence' hypotheses.
Crosstalk Threshold ($C_{th}$): The crosstalk threshold is a critical value for SPADE performance. It represents the maximum level of interference or unwanted mixing between the two separated spatial modes that still allows SPADE to provide an advantage over direct imaging. The authors found this threshold to be around 0.1.
Photon Budget Reduction: This measures how much fewer photons are needed by a technique compared to another method (like direct imaging) to achieve the same level of statistical certainty (error rate). SPADE demonstrated a significant reduction, needing up to one order of magnitude fewer photons for the same error probability when crosstalk was low.

Terminology

False Negatives — 2× in this paper · explained in 2 episodes

Summary

The gist The authors experimentally demonstrate a universal, parameter-independent test for asymmetric source discrimination using spatial mode demultiplexing (SPADE) and single-photon detectors.

Experimental Setup and Methodology

The experiment is supported by a complete theory on binary decision for exoplanet detection, or more in general on asymmetric HT between 1-vs-2 incoherent sources, including crosstalk. The test implemented is the same as Ref. [37], that is universal as it does not depend on the physical parameters. The experimental setup involves two fiber-coupled Light Emitting Diodes (LED) simulating two incoherent point-like sources. Both LEDs are free-space collimated using two fiber collimators (FC) and a pair of two-lens (L) system to match the beam waist w0 of the demultiplexer adopted for the collection. The LED A mimics the star and B the exoplanet, both are attenuated to respectively IA and IB intensities through fiber coupled attenuators not shown in the figure.

Theoretical Framework for Hypothesis Testing

Employing the formalism of quantum detection theory, the two hypotheses of absence H0 or presence H1 of the exoplanet are represented by two density matrices ρ0 and ρ1. In asymmetric hypothesis testing, a quantum optimal test over an asymptotically large number N ≫ 1 of repetitions attains, by the quantum Stein lemma, β ≃ exp(−NDQ(ρ0ρ1)) where DQ(ρ0ρ1) ≃ εd2 a/4 (1). The effect of crosstalk can be modelled by an incoherent mixing of the outcome probabilities p′j via a stochastic matrix p′j = C00 C01 C10 C11 P(0Hi) P(1Hi) (4). The relative entropy between the two crosstalk-affected Bernoulli distributions with probabilities p′ 0 and p′ 1 becomes DSD(p'0p'1) ≃ ϵ 2d 4a 32C10, where the last approximation is valid for εd2a ≪ C10 ≪ 1 (see Appendix C).

Performance Comparison and Crosstalk Threshold

The test beats the subdiffraction limit, as it achieves a type II error rate that is lower than the one of direct imaging. The authors find a quantitative threshold for the crosstalk Cth ≃ 0.1 below which SPADE outperforms direct imaging. For example, for crosstalk of 10−2, SPADE needs up to one order of magnitude fewer photons than direct imaging to achieve the same error rate. The crosstalk of our demultiplexer C ≃ 0.01 corresponds to one order of magnitude reduction in the photon budget required to achieve the same error rate than direct imaging, while state-of-the-art mode demultiplexers with crosstalk C ≃ 3 × 10−3 [38] achieve a photon budget reduction of ≃ 2.4 × 10−2.

Key Findings and Conclusions

The experimental results are well described by a simple noise model and confirm that the test attains the optimal exponential scaling of the false-negative error rate in the large-sample regime. The analysis identifies a crosstalk threshold Cth ≃ 0.1 separating the regimes where SPADE and direct imaging are respectively advantageous. The measured crosstalk in our implementation, of order 10−2, lies well within the SPADE-favorable regime and corresponds to an order-of-magnitude reduction in the required photon number at fixed error probability. Beyond this specific implementation, the experimental results demonstrate that the advantage of SPADE in hypothesis testing is robust to realistic imperfections and does not rely on prior knowledge of the source parameters. This makes the approach relevant for practical subdiffraction imaging scenarios, particularly in photon-limited regimes.

Experimental Results Summary

The authors counted the number N1 of photons in the HG01 and HG10 modes in a 10 ms integration time for a given intensity ratio ϵ and adimensional sources separation da≡ d/w0. From the collected counts Ni 1(ϵ = 0), that represent hypothesis H0, they calculate the 95th percentile N∗. The number of times the test fails, namely it wrongly predicts the absence of the exoplanet as N1 ≤ N∗, divided by 100, represents the type II error rates βexp(da, ϵ) for the given values of da and ϵ. These measured values are compared with the theoretical expression of β in Eq. (8), retrieving a good match between theory and experiment. The authors estimate C ≃ 10−2, that is compatible with the value of crosstalk as for the vendor specification. Remarkably, this value of crosstalk falls within the regime of C < 0.1 for which SPADE detection largely outperforms direct imaging, as shown in Eq. (6) and FIG. 3.

Comparison to Direct Imaging

The error rate scaling for the optimal direct imaging test is found to be βDI = exp−Nϵ 2d 4a/4, confirming the asymptotic optimality of this direct imaging test, in the sense that is saturates the error rate given by the relative entropy. The SPADE test achieves a scaling DSD(p'v0p'v1) = ϵ 2d 4a 32C10, which in general is smaller than DSD(p'0p'1) in Eq. (5). The authors show that the test is approximately 1/8C10 = 12.5 times more efficient than direct imaging in the asymptotic regime.

Appendix Analysis

In Appendix A, they discuss more in depth this discrepancy between two binary tests regarding the formulation of hypotheses concerning the position of celestial bodies. The analysis shows that for small ϵ and da, both versions of ρ1 can be diagonalised in the HermitGauss (HG) basis centred in centre of brightness. In Appendix C, they show that the two binary tests are identical for SPADE within the first relevant order of the expansions in ϵ and da. The analysis confirms that neglecting small quantities of orders of ϵ squared and d 4a leads to the same expression for DSD(p'v0p'v1) = ϵ 2d 4a 32C10. The scaling in da is not the “Rayleigh cursed” scaling because the two hypothesis can be told apart by looking at the centre of brightness, although the relative entropy has suboptimal scaling in ϵ.

Experimental Details

FIG. 2 shows the experimental setup where two fiber-coupled Light Emitting Diodes (LED) with a central wavelength at about 1550 nm and a bandwidth of about 80 nm (FWHM) simulate two incoherent point-like sources. The A beam, after reflection on two steering mirrors (M), impinges on a cube beam splitter (BS) to be combined with the B beam and, after crossing a fixed polarizer (P2), is coupled with the free-space input port of a HG demultiplexer.

Improvements for AI systems

Subdiffraction source discrimination for faint signals: The improved system can discriminate faint sources well beyond diffraction limit by exploiting spatial mode demultiplexing (SPADE) and single-photon detectors, as SPADE retains an advantage over direct imaging in the relevant regime of small separations and low intensity ratios.
Photon-starved imaging tasks: The system can achieve subdiffraction asymmetric hypothesis testing under realistic imperfections, paving the way to photon-starved imaging tasks by identifying an experimentally accessible crosstalk threshold Cth which is well below what can be achieved by diffraction-limited direct imaging.
Crosstalk optimized photon budget: The system can operate with a significantly reduced photon count for the same error rate; for example, with a crosstalk of 10 minus 2, SPADE needs up to one order of magnitude fewer photons than direct imaging to achieve the same error rate.

Abstract

We experimentally demonstrate a universal, parameter-independent test for asymmetric source discrimination below the diffraction limit. The test exploits spatial-mode demultiplexing (SPADE) and single-photon detection to discriminate the presence of a faint source in the vicinity of a brighter one. We develop a complete theory of the test in the presence of arbitrary modal crosstalk and show that the crosstalk-affected test retains a favorable error exponent over diffraction-limited direct imaging, provided that the crosstalk remains below a threshold C max 1/9. This establishes a quantitative condition under which the subdiffraction advantage of SPADE survives realistic imperfections. We experimentally test the scheme for a range of source separations and intensity ratios, without requiring prior knowledge of these parameters. The measured false-negative error rates are in good agreement with the theoretical model and systematically outperform direct imaging. These results demonstrate that SPADE provides a robust and experimentally accessible advantage for asymmetric subdiffraction hypothesis testing in the photon-starved regime, even in the presence of realistic modal crosstalk.