Multiparameter estimation for the superresolution of two incoherent sources

Technical Summary: Multiparameter Estimation for the Superresolution of Two Incoherent Sources

Problem Statement
Conventional direct imaging (DI) of two incoherent point sources is fundamentally limited by diffraction when the separation between sources falls below the Rayleigh limit. In this regime, the information per photon regarding the separation vanishes in DI, rendering the estimation of the separation, centroid position, and relative brightness (intensity imbalance) highly inaccurate. While quantum metrology theory (specifically the Quantum Cramér–Rao Bound, QCRB) indicates that this precision limit is not fundamental and that spatial-mode demultiplexing (SPADE) in the Hermite–Gauss (HG) basis can achieve finite precision even at zero separation, previous experimental implementations have largely been restricted to single-parameter estimation or assumed equal source intensities. Real-world imaging scenarios are inherently multiparametric; the separation ($d$), centroid ($c$), and intensity imbalance ($p$) are simultaneously encoded and non-independent. The challenge lies in designing a measurement strategy that provides robust sensitivity across a wide range of scene configurations without prior knowledge of the parameters, particularly when moving from idealized indistinguishable sources to realistic, slightly non-identical sources.

Methodology
The authors experimentally demonstrate a simultaneous estimation of $d$, $c$, and $p$ using a dual-Spatial Mode Demultiplexing (SPADE) architecture.

• Experimental Setup: The system utilizes two independent continuous-wave lasers at 1550 nm to generate incoherent sources. The light is coupled into two Multi-Plane Light Converters (MPLCs). The first MPLC decomposes the field into a standard Hermite–Gauss mode basis. The second MPLC is identical but deliberately shifted laterally by $x_s = 0.3w_0$ (where $w_0$ is the beam waist), creating a complementary mode basis.
• Detection Scheme: The experiment measures the intensities of the first four horizontal modes ($HG_{00}, HG_{10}, HG_{20}, HG_{30}$) from the first MPLC and the $HG_{10}$ mode from the shifted second MPLC. A total power detector monitors the scene.
• Estimation Strategy: The authors employ a maximum likelihood estimator (MLE) assuming multivariate Gaussian noise. Calibration curves are generated by scanning individual sources to map the system's response, accounting for experimental imperfections such as misalignment and crosstalk. The estimator minimizes a loss function comparing measured mode intensities against theoretical expectations derived from the calibration data.
• Scenarios: Two distinct scenarios are investigated:
  1. Distinguishable Sources: Realistic sources with 99% visibility (slightly non-identical).
  2. Indistinguishable Sources: An idealized case where data is reconstructed from a single source to simulate perfectly identical emitters, removing any prior asymmetry.

Key Contributions

1. Dual-Basis Approach: The paper introduces a dual-MPLC configuration to overcome the intrinsic symmetries of a single HG basis, which previously prevented the simultaneous estimation of all three parameters ($d, c, p$). The shifted second MPLC lifts degeneracies (specifically between $-d$ and $d$) and enhances robustness.
2. Robust Multiparameter Estimation: The study achieves simultaneous estimation of separation, centroid, and intensity imbalance over a broad range of parameter values in a single experimental setting, without adapting the measurement to the unknown scene.
3. Benchmarking Against Bounds: The performance is rigorously benchmarked against classical Fisher-information-based Cramér–Rao bounds (CRB) and, for the indistinguishable case, the Quantum Cramér–Rao bound (QCRB).

Results

• Distinguishable Sources: The dual-MPLC setup achieved separation uncertainties reaching $10^{-3}w_0$, which is 2 to 4 orders of magnitude below the Rayleigh limit. The bias for separation and centroid was on the order of $10^{-2}w_0$, and intensity imbalance was estimated with $\sim1\%$ precision. The statistical errors closely approached the experimental CRBs.
• Indistinguishable Sources: Even in the absence of source asymmetry, the dual-MPLC setup maintained accuracy, with biases for $d$ and $c$ remaining in the $10^{-2}w_0$ range. While sensitivity degraded slightly compared to the distinguishable case, the setup successfully estimated parameters across a wide range, demonstrating that the method does not rely on source imperfections.
• Comparison to Single-MPLC: Omitting the second MPLC resulted in significant performance deterioration, including a loss of nearly an order of magnitude in sensitivity and a failure to resolve the sign of the separation (degeneracy between $-d$ and $d$).
• Quantum Limits: For indistinguishable sources, the classical sensitivity of the 2-MPLC configuration was found to be within a factor of $\sim4$ of the QCRB.

Significance and Claims
The paper claims to have experimentally demonstrated the simultaneous estimation of three parameters characterizing two incoherent sources in the sub-Rayleigh regime, enabling super-resolved scene characterization without prior knowledge of the sources. The authors highlight that their approach bridges the gap between single-parameter estimation and full image reconstruction. They emphasize that the dual-MPLC scheme is more robust and versatile than implementing a full set of optimal modes, which are scene-dependent and difficult to realize. The work clarifies the practical scope of multiparameter SPADE measurements, showing that high-precision superresolution is achievable even with indistinguishable sources, paving the way for practical applications such as astronomical imaging. The authors note that while the experiment was conducted in a high-photon-flux regime, the linear nature of the device suggests the approach could be extended to lower-flux conditions.