Spatial Mode De-Multiplexing (SPADE)

• Spatial Mode De-Multiplexing (SPADE) is an optical technique that projects light onto an orthonormal spatial mode basis to extract detailed spatial information.
• It replaces traditional pixel-based detection with mode-resolved photon counting, overcoming Rayleigh’s curse and enhancing subdiffraction parameter estimation.
• Applications include superresolution imaging and optical communications, implemented via methods such as MPLC and integrated photonics for low-crosstalk channel separation.

to=shell 天天中彩票为什么 code python - <<'PY' import urllib.request, urllib.parse q='all:spatial mode demultiplexing' url='http://export.arxiv.org/api/query?search_query='+urllib.parse.quote(q)+'&start=0&max_results=3' try: print(urllib.request.urlopen(url, timeout=10).read()[:500]) except Exception as e: print('ERR',e) PY 【อ่านข้อความเต็มanalysis to=all code omitted 731 characters】 Spatial Mode De-Multiplexing (SPADE) denotes a class of optical measurements and devices that sort an optical field into an orthonormal spatial-mode basis before detection. In far-field imaging of incoherent sources, SPADE replaces pixel-basis intensity detection with mode-resolved photon counting, typically in Hermite–Gaussian (HG) or other orthogonal bases, and is used to estimate separations, centroids, scales, and higher moments in the subdiffraction regime. In optical communications and guided-wave photonics, the same term is used for hardware that separates multiplexed spatial channels—fiber modes, vector modes, or supermodes—by means such as q-plates, photonic lanterns, Multi-Plane Light Conversion (MPLC), matched phase masks, or integrated waveguide structures (Tsang, 2016, Richardson et al., 2013).

1. Mode-space formulation

The common mathematical structure of SPADE is a projection of the field onto orthonormal spatial modes. In the imaging literature, a paraxial field on the image plane is expanded as $E(x,y)=\sum_n a_n \psi_n(x,y)$ with $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$, and mode-resolved detection corresponds to projectors such as $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$ or, for HG modes, $\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$ (Titov, 21 Sep 2025, Santamaria et al., 2022). For two equally bright incoherent point sources separated by $\theta$, the single-photon density operator may be written as

$$\rho_1=\tfrac12\bigl(|1,u_1\rangle\langle 1,u_1|+|1,u_2\rangle\langle 1,u_2|\bigr),$$

with detection probability in mode $\psi_n$

$$P_n(\theta)=\operatorname{Tr}[\Pi_n\rho_1]=\tfrac12\bigl(|\langle\psi_n|u_1\rangle|^2+|\langle\psi_n|u_2\rangle|^2\bigr)$$

(Titov, 21 Sep 2025).

Tsang’s semiclassical treatments generalized this construction beyond Gaussian point-spread functions (PSFs) by introducing PSF-adapted bases in which higher modes resemble derivatives of the PSF. In that framework, the image-plane mutual coherence is

$$\Gamma(x,x'|\theta)=\int dX\,\psi(x-X)\psi^*(x'-X)F(X|\theta),$$

and SPADE measures diagonal and, with interferometric combinations, off-diagonal elements of the coherence matrix in the chosen mode basis (Tsang, 2017). In the weak-flux regime, the resulting mode counts are modeled as independent Poisson variables with means determined by the mode powers, which makes Fisher-information analysis and Cramér–Rao bounds analytically tractable (Tsang, 2016, Tsang, 2017).

The same mode-space viewpoint appears in communications. In few-mode and multicore fibers, the relevant states are LP modes, true vector modes such as HE and EH families, or multicore supermodes. Orthogonality and coupled-mode theory determine how perturbations transfer power among channels, and de-multiplexing is the inverse problem of extracting those channels with low crosstalk (Richardson et al., 2013).

2. Resolution, Fisher information, and the Rayleigh regime

SPADE’s central role in imaging arises from the fact that direct imaging and mode-resolved detection concentrate information differently. For direct imaging of two close incoherent sources, the Fisher information for separation vanishes as the separation approaches zero; this is the standard “Rayleigh’s curse.” In an HG basis matched to a Gaussian PSF, by contrast, the mode probabilities take the Poisson-like form

$$P(q|\theta)=e^{-Q}Q^q/q!,\qquad Q=\theta^2/(16\sigma^2),$$

which gives

$\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$0

independent of $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$1 and equal to the quantum Fisher information for the two-source model (Titov, 21 Sep 2025). This is the canonical statement that ideal SPADE evades Rayleigh’s curse.

The same conclusion appears in broader moment-estimation problems. For an arbitrary subdiffraction object, SPADE was shown to estimate second or higher moments much more precisely than direct imaging under photon shot noise, and to approach the optimal precision allowed by quantum mechanics for location and scale parameters (Tsang, 2016). In the semiclassical extension, PSF-adapted direct detection and interferometric mode measurements yield estimators whose variances scale far more favorably in the subdiffraction limit than those of direct imaging, especially for low-order nontrivial moments (Tsang, 2017).

Several later variants refine the Fisher-information picture rather than replacing it. In a bi-photon scheme based on spontaneous parametric down-conversion (SPDC), coincidence SPADE onto joint HG product modes yields

$\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$2

with enhancement by $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$3, where $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$4 is the Schmidt number of the two-photon state (Grenapin et al., 2022). In fluctuation-enhanced SPADE, temporal cumulants of mode counts provide access to higher even moments, and in the odd output of image inversion interferometry one has

$\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$5

so cumulants can replace higher-mode sorting for certain tasks (Kurdzialek, 25 Nov 2025).

3. Bases, architectures, and experimental realizations in imaging

HG modes are the most common basis because a Gaussian PSF and its derivatives map naturally into that family, but the cited work does not restrict SPADE to HG modes alone. Zernike modes, Laguerre–Gaussian modes, and PSF-adapted polynomial bases also appear, depending on whether the task is aberration sensing, polar-coordinate analysis, or general PSF adaptation (Titov, 21 Sep 2025, Tsang, 2017).

A major practical route to SPADE is MPLC. In a thesis-scale implementation for quantum-inspired superresolution, MPLC was described as alternating local phase modulations and free-space propagation to realize an arbitrary unitary in the transverse plane. A 6-mode simulation from Gaussian spots to co-located HG modes used 7 phase planes spaced 27 mm apart and obtained fidelity-matrix diagonals of 0.94–0.96 with off-diagonals below 0.1; a 3-mode experimental sorter using a reflective spatial light modulator and 4 phase masks achieved measured diagonal fidelities of approximately $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$6 and average fidelity of about 90.6% (Titov, 21 Sep 2025).

Multimode MPLC has also been used as an estimation instrument rather than only as a sorter. A 9-mode demultiplexer was employed to estimate the transverse separation of two incoherent beams in both transverse directions, with measured intensity curves agreeing with theoretical projections after calibration. The reported sensitivity showed a plateau of about $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$7 for $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$8, while cross-talk limited exact saturation of the ideal Cramér–Rao bound (Boucher et al., 2020). For bright incoherent sources, a related six-mode HG demultiplexer was used to estimate both transverse separation and relative intensity; the reported resolving power was $\int dx\,dy\,\psi_n^*\psi_m=\delta_{nm}$9 for SPADE versus about $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$0 for direct imaging, and the intensity-ratio detection limit was $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$1 versus $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$2 (Santamaria et al., 2022).

Task-specific reductions of the full HG basis are also important. In micro-oscillation frequency estimation, a two-mode “plus–minus” construction

$\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$3

was designed to retain the zeroth- and first-moment sensitivity of the displaced PSF while using only two detectors. The reported experiment used a 770 nm laser, a diffraction-limited PSF of width $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$4, and two CMOS pixels for the $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$5 diffraction orders, and showed that under increasing background light direct imaging variance rose steeply whereas PM-SPADE remained near-optimal over a wide range of $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$6 (Hu et al., 6 Apr 2025).

4. Imperfections, noise, and finite-sample limitations

The idealized claim that SPADE removes Rayleigh’s curse requires qualification once realistic noise and misalignment are introduced. A formal analysis of noisy demultiplexing showed that random-unitary noise generated by polynomials of creation and annihilation operators can reintroduce the destructive small-separation behavior. The proposed remedy was to repeat the demultiplexers and interleave them with a rotation group

$\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$7

so that first-order noise terms cancel in the effective map. For displacement noise, the protocol simplifies to two demultiplexers separated by parity, and if the noise is frozen between the two steps the combined action becomes exactly noise-free (Sakuldee et al., 2024).

Misalignment has a more structural effect in binary-SPADE reductions. Singular-learning analysis of one-versus-two source discrimination showed that aligned Gaussian direct imaging and aligned SPADE share the same real log canonical threshold $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$8 but differ in multiplicity, yielding a universal subleading advantage for aligned SPADE in the local prior-weighted regime. In the misaligned setting, however, binary-SPADE develops nontrivial local power on the scale $\Pi_n=|1,\psi_n\rangle\langle 1,\psi_n|$9, direct imaging on $\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$0, and finite-$\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$1 Neyman–Pearson comparisons on the plotted grids favored direct imaging. The same analysis identified an exact blind separation

$\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$2

at which misaligned binary-SPADE power collapses to the test size $\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$3 (Kariya, 14 May 2026).

Practical instruments reflect these limitations. The first on-sky binary-SPADE hypothesis-testing experiment used a double-clad fiber coupler to separate the fundamental Gaussian mode from the higher-order complement and detected a binary star system below the diffraction limit in a photon-starved regime where no direct image could be formed. The measured type II error was always lower than that of a perfect direct-imaging measurement, but the scaling was heavily limited by unbalanced loss in the coupler, and atmospheric turbulence was identified as a likely source of further degradation on larger apertures (Wallis et al., 16 Jun 2026). MPLC experiments likewise reported tight tolerances: an input lateral shift of one waist reduced fidelity from about 0.94 to about 0.69, and plane-spacing errors of $\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$4 mm around 27 mm degraded fidelity by at least 5% (Titov, 21 Sep 2025).

5. Communications, multiplexing, and integrated photonics

In optical communications, SPADE is closely tied to space-division multiplexing and mode-division multiplexing. The modal degrees of freedom of a circular dielectric waveguide are described, in the weakly guiding approximation, by LP modes satisfying a scalar Helmholtz equation, while full-vector analyses yield HE and EH modes. Orthogonality, perturbative coupling coefficients, and differential mode-group delay determine how channels are launched, mixed, and recovered. The principal separation technologies summarized in this literature include photonic lanterns, spot-type and free-space couplers, integrated silicon photonic mode multiplexers, thin-film wavelength filtering for spatial superchannels, and digital $\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$5 MIMO equalizers (Richardson et al., 2013).

Free-space demonstrations illustrate one branch of this usage. A liquid-crystal q-plate mode de-multiplexer converted Gaussian beams into four orthogonal vector modes,

$\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$6

and back into single-mode fibers for coherent detection. In a proof-of-principle link, four vector modes each carried a 20 Gbit/s QPSK signal on a single wavelength channel near 1550 nm, giving an aggregate 80 Gbit/s over about 1 m with off-diagonal crosstalk below $\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$7 dB and power penalties below 3.41 dB at the forward-error-correction threshold (Milione et al., 2014). A different free-space architecture used matched spatial phase masks in a three-lens Fourier “sandwich” to multiplex Gaussian beams with planar, radial, or hybrid radial–azimuthal phase slopes and re-focus only the selected channel on axis, reporting crosstalk below 10%, even below 1% with optimized masks, and separation efficiency above 90% (Hai-long et al., 2013).

Integrated and fiber-compatible implementations provide another branch. A supersymmetric three-waveguide de-multiplexer combined a central superpartner waveguide with spatial adiabatic passage so that the TE$\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$8 mode transferred from left to right while TE$\Pi_{hk}=|f_{hk}\rangle\langle f_{hk}|$9 remained in the left guide, with output fidelities above 0.90 over broad parameter ranges and about 0.99 for optimized values (Queraltó et al., 2018). MPLC has also been scaled to high-dimensional guided-wave mode sorting: a pair of 45-mode spatial multiplexer/demultiplexers for a standard 50 $\theta$0m graded-index multimode fiber used 11 phase reflections, showed an average insertion loss of 4 dB and average inter-mode crosstalk of $\theta$1 dB across the C band, and was then used to probe mode-group crosstalk and bend sensitivity in the fiber itself (Bade et al., 2018).

6. Scope, applications, and research directions

The application space of SPADE is correspondingly broad. In imaging, the cited work identifies fluorescence microscopy, astronomy, binary-source hypothesis testing, single-particle tracking, and remote sensing of thermal sources as major targets (Tsang, 2016, Titov, 21 Sep 2025, Wallis et al., 16 Jun 2026). In communications, SPADE enters free-space optical links, few-mode and multicore fibers, ring-core fibers, and integrated photonic circuits (Richardson et al., 2013, Milione et al., 2014).

The literature also shows that SPADE is not a single measurement architecture. Full HG sorting, binary-SPADE, interferometric HG combinations, image inversion interferometry, q-plates, photonic lanterns, double-clad fiber couplers, holographic sorters, SLM-based projectors, MPLC, and supersymmetric adiabatic waveguides all instantiate the same underlying principle of orthogonal mode separation, but they target different observables and tolerate imperfections differently (Titov, 21 Sep 2025, Richardson et al., 2013, Wallis et al., 16 Jun 2026).

Current directions emphasized in the cited work include scaling MPLC to tens or hundreds of modes, hybrid wavefront-matching and adjoint-based optimization, adaptive superresolution for dynamic scenes, photon-number-resolving detection, adaptive-polarization compensation for vector-mode links, and the use of temporal cumulants or entanglement to simplify measurements or enhance Fisher information (Titov, 21 Sep 2025, Milione et al., 2014, Kurdzialek, 25 Nov 2025, Grenapin et al., 2022). A consistent implication across these strands is that SPADE’s value lies not only in surpassing diffraction-limited direct imaging under idealized models, but also in recasting spatial information into a mode basis where information can be concentrated, routed, and analyzed with hardware tailored to the estimation or communication task at hand.