---
title: Selective De-multiplexing in Photonic Systems
url: https://www.emergentmind.com/topics/selective-de-multiplexing
type: topic
---

# Selective De-multiplexing in Photonic Systems

Searching arXiv for the cited works to ground the article in current metadata and ensure accurate referencing.
arxiv_search.query({"search_query":"id:1008.3424 OR id:1611.02294 OR id:2203.08682 OR id:2306.04100 OR id:1301.1564 OR id:1602.08414 OR id:1501.02125 OR id:1803.07907 OR id:1902.10408 OR id:1709.08809 OR id:1803.00206 OR id:2606.17894 OR id:2304.12956 OR id:2311.03675 OR id:2304.09622","max_results":15,"sort_by":"submittedDate","sort_order":"descending"})
Selective de-multiplexing, across the literature considered here, denotes demultiplexing in which the destination channel is determined by a discriminating variable rather than by random splitting alone. That variable may be optical wavelength, emission time bin, guided spatial mode, polarization sector, orbital-angular-momentum label, or even the informational type—classical versus quantum—carried by a quantum system. Accordingly, the topic spans plasmonic wavelength routing, active temporal-to-spatial single-photon routing, integrated mode-division receivers, free-space phase-matched decoding, and selector-controlled quantum instruments [1008.3424] [1611.02294] [1602.08414] [2606.17894].

## 1. Conceptual scope and defining features

A recurring distinction in this literature is between **passive splitting** and **selective routing**. In passive architectures, output assignment is fixed or probabilistic; in selective architectures, the output is chosen by a control variable or by a channel-dependent physical response. In the temporal-to-spatial single-photon setting, this distinction is explicit: passive beam-splitter demultiplexing incurs the super-exponential penalty \(\left(\frac{1}{n}\right)^n\), whereas active routing assigns successive photons to designated outputs by synchronized electro-optic control [1611.02294]. In wavelength-selective plasmonics and integrated photonics, selectivity arises from dispersive diffraction, inverse-designed interference, or nonlinear phase matching rather than from random power division [1008.3424] [1803.00206].

| Domain | Selective variable | Representative realization |
|---|---|---|
| Spectral | Wavelength or pump-controlled phase matching | Plasmonic Rowland grating; inverse-designed wavelength routers; SFG demultiplexer |
| Temporal | Emission time bin | Electro-optic single-photon demultiplexers |
| Spatial/modal | Mode order or LP mode | LNOI waveguide arrays; InP FMF DEMUX; MPLC |
| Polarization / angular sector | Polarization sector or matched phase mask | Conical refraction; spatial phase decoding |
| Informational | Selector-controlled instrument realization | Quantum DEMUX |

This suggests a broad but coherent definition: selective de-multiplexing is a routing operation in which a multiplexed input is separated according to a prescribed physical or informational label, and successful recovery depends on matching the receiver or device response to that label. In some cases the mapping is fixed by structure, while in others it is externally reconfigured.

## 2. Wavelength-selective and spectrally selective realizations

A clear spectral example is the concentric-groove plasmonic demultiplexer of "Plasmonic Demultiplexer and Guiding" [1008.3424]. The device is a two-dimensional structure in a \(50~\mathrm{nm}\) gold film on glass that simultaneously couples normally incident free-space light into surface plasmon polaritons and demultiplexes the resulting SPPs by wavelength. Its operating principle combines grating coupling, wavelength-dependent diffraction, and Rowland-circle focusing. The key relations are the SPP diffraction equation,
\[
d\sin\theta = m\lambda_{\mathrm{spp}},
\]
and the focusing condition,
\[
\Delta r = m\lambda_{\mathrm{SPPC}}.
\]
Because different \(\lambda_{\mathrm{spp}}\) produce different diffraction angles, different wavelengths focus to different positions on the focal circle. In the air-superstrate devices, measured resolutions were \(33~\mathrm{nm}\), \(20~\mathrm{nm}\), and \(15~\mathrm{nm}\) for \(m=1,2,3\), and operation with a water superstrate enabled \(m=4\) and an experimentally obtained resolution of \(10~\mathrm{nm}\). The same work also demonstrated routing into five SPP strip waveguides, showing that the device could act as a true WDM demultiplexer rather than only as a compact spectrometer [1008.3424].

Inverse-designed silicon devices realize the same selective principle in a different regime. "High-performance 2D 1xN T-junction Wavelength (De)Multiplexer Systems by Inverse Design" [1902.10408] reports \(1\times2\), \(1\times4\), and \(1\times6\) wavelength demultiplexers with footprints \(2.80~\mu\mathrm{m}\times2.80~\mu\mathrm{m}\), \(2.80~\mu\mathrm{m}\times4.60~\mu\mathrm{m}\), and \(2.80~\mu\mathrm{m}\times6.95~\mu\mathrm{m}\), respectively. The selectivity is encoded in an objective-first inverse-designed dielectric pattern that routes the shortest wavelength to \(P_1\) and the longest to \(P_N\). Relatedly, "Inverse design and demonstration of a compact on-chip narrowband three-channel wavelength demultiplexer" [1709.08809] experimentally demonstrated a three-channel SOI demultiplexer for \(1500\), \(1540\), and \(1580~\mathrm{nm}\) with \(40~\mathrm{nm}\) spacing and a footprint of \(24.75~\mu\mathrm{m}^2\). It reported a simulated peak insertion loss of \(-1.55~\mathrm{dB}\) with under \(-15~\mathrm{dB}\) crosstalk, and a measured peak insertion loss of \(-2.29~\mathrm{dB}\) with under \(-10.7~\mathrm{dB}\) crosstalk [1709.08809].

A more explicitly active spectral architecture is the all-optical quantum signal demultiplexer based on sum-frequency generation [1803.00206]. There, the selected wavelength channel is determined by the SFG pump wavelength: only the signal satisfying the quasi-phase-matching condition is efficiently upconverted and therefore extracted from the common telecom line. The experiment used a type-0 PPLN crystal with \(795~\mathrm{nm}+1560~\mathrm{nm}\rightarrow525~\mathrm{nm}\), achieved a maximum quantum conversion efficiency of \(38\%\), and preserved energy-time entanglement after demultiplexing, with raw visibilities \((89.69 \pm 5.07)\%\), \((91.98 \pm 3.07)\%\), and \((89.02 \pm 3.23)\%\) across three channels [1803.00206]. This is selective spectral extraction by nonlinear phase matching rather than by fixed passive filtering.

## 3. Active temporal-to-spatial de-multiplexing of single photons

In quantum photonics, selective de-multiplexing most often means converting one temporal stream of single photons into multiple simultaneous spatial outputs. "Active demultiplexing of single-photons from a solid-state source" [1611.02294] established the canonical integrated form of this problem: successive photons from a quantum-dot micropillar source are routed into four outputs of a lithium-niobate chip using synchronized electro-optically tunable directional couplers. The routing policy is explicit—first photon to output 1, second to output 2, third to output 3, fourth to output 4—and therefore selective rather than probabilistic. The paper characterizes the scaling advantage by contrasting the passive penalty \(\left(\frac{1}{n}\right)^n\) with an active scheme that becomes polynomial in the deterministic limit \(\eta_{\mathrm{DM}}\to1\). Experimentally it obtained \(\eta_{\mathrm{DM}}=0.78\pm0.06\), in agreement with the independent estimate \(0.80\pm0.09\), on a four-output integrated device [1611.02294].

"Fast and efficient demultiplexing of single photons from a quantum dot with resonantly enhanced electro-optic modulators" [2203.08682] pushed this line toward higher switching speed and higher end-to-end efficiency using a free-space binary tree of resonantly enhanced EOMs and PBSs. The routing rate was \(38.1~\mathrm{MHz}\), the end-to-end demultiplexer efficiency was \(79(3)\%\), and the measured four-photon coincidence rate was \(0.17(1)~\mathrm{Hz}\). The switching fidelity was quantified by \(\eta_{\mathrm{sw}}=0.946(8)\), and the model of the detected \(n\)-fold rate made the active advantage explicit through the factor \(\eta_{\mathrm{sw}}^n\) relative to passive \((1/m)^m\)-type scaling [2203.08682].

A distinct architectural direction replaces switching trees by repeated use of one active element. "Single-active-element demultiplexed multi-photon source" [2304.12956] uses a single EOM together with a recurrent free-space delay geometry to load successive photons and then release them into multiple outputs. The demonstrated system produced up to eight demultiplexed highly indistinguishable single photons, with approximately \(1.4~\mathrm{kHz}\) four-photon and \(20~\mathrm{mHz}\) eight-photon coincidence rates, \(g^{(2)}(0)=1.57(2)\%\), and \(\mathcal{I}=95.35(3)\%\) [2304.12956]. The related four-channel free-space storage-loop design of "Resource-efficient low-loss four-channel active demultiplexer for single photons" [2304.09622] achieved \(39.7\%\) efficiency per channel using one Pockels cell, a \(3.6~\mathrm{m}\) loop with round-trip time matched to the \(12.1~\mathrm{ns}\) source period, and controlled polarization rotation for release [2304.09622].

These works collectively show that temporal selectivity can be implemented by synchronized electro-optic state changes, by binary-tree routing, or by recurrent storage-and-release. A common misconception is that all active temporal demultiplexers are equivalent to arbitrary packet switches. The single-active-element architecture explicitly is not: it is deterministic sequential demultiplexing of regularly spaced time bins, not fully arbitrary random-access routing [2304.12956].

## 4. Spatial-mode, polarization, and phase-selective de-multiplexing

Integrated mode-division demultiplexing provides a spatial analogue of wavelength routing. "Integrated Broadband Mode Division Demultiplexer in Waveguide Arrays" [2306.04100] uses a uniform multimode waveguide array on LNOI in which different input modes acquire different transverse group velocities. The coupled-mode evolution is
\[
i\frac{da_{n,m}(z)}{dz} + K_m[a_{n-1,m}(z) + a_{n+1,m}(z)] = 0,
\]
and the mode division angle is
\[
\theta_m = \arctan(-2DK_m).
\]
Because higher-order modes have larger \(K_m\), they propagate at larger transverse angles and land at different lateral positions after \(80~\mu\mathrm{m}\). The device experimentally separated TE\(_0\), TE\(_1\), and TE\(_2\), and the measured mode division angles agreed closely with theory across \(430\), \(445\), \(470\), and \(480~\mathrm{nm}\) [2306.04100]. This is mode-sensitive but only weakly wavelength-sensitive transport.

In few-mode-fiber systems, selectivity can also be programmable. "Reconfigurable photonic integrated mode (de)multiplexer for SDM fiber transmission" [1602.08414] used an InP balanced MZI amplitude controller and a phase shifter to route LP\(_{01}\) and LP\(_{11a}\) between output ports. At the balanced point \(\phi_A=\pi/2\), the phase \(\phi_P\) determined whether the fields combined in phase or out of phase,
\[
E_{00} \propto \frac{E_1+E_2}{\sqrt{2}}, \qquad E_{10} \propto \frac{E_1-E_2}{\sqrt{2}},
\]
so changing \(\phi_P\) dynamically switched channel routing. The system demonstrated post-propagation channel crosstalk around \(-15~\mathrm{dB}\), mode excitation crosstalk down to \(-20~\mathrm{dB}\), and at least \(10~\mathrm{nm}\) operational bandwidth [1602.08414]. By contrast, "High Speed Data Transmission over GI-MMF Using Mode Group Division Multiplexing" [1501.02125] emphasized that, under direct detection with OOK, optical mode-group selective multiplexing and de-multiplexing is essential because square-law detection makes MIMO ineffective. There the receiver used LCOS-based mode conversion followed by SMF modal filtering to select the strongest mode of each mode group after \(5~\mathrm{km}\) OM4 GI-MMF transmission [1501.02125].

High-dimensional spatial selectivity was pushed further by "Fabrication and Characterization of a Mode-selective 45-Mode Spatial Multiplexer based on Multi-Plane Light Conversion" [1803.07907]. The reciprocal MPLC pair addressed all 45 guided modes of a standard \(50~\mu\mathrm{m}\) OM2 graded-index multimode fiber using only 11 phase profiles, with average \(4~\mathrm{dB}\) insertion loss and \(-28~\mathrm{dB}\) crosstalk across the C band. The work also made clear that, after propagation in graded-index fiber, the most robust separation is often at the **mode-group** level rather than the individual mode level because modes with constant \(n+m\) are degenerate and couple strongly [1803.07907]. A related practical caveat appears in "Ultra-compact and efficient integrated multichannel mode multiplexer in silicon for few-mode fibers" [2311.03675]: although the device is reciprocal and can collect all degenerate LP content into eight single-mode on-chip channels, realistic FMF demultiplexing generally requires digital MIMO DSP or an MZI mesh because LP-mode evolution and polarization rotation scramble the received basis [2311.03675].

Free-space implementations use matched phase or polarization structure instead of integrated mode transforms. "Free Space Optical Polarization De-multiplexing and Multiplexing by means of Conical Refraction" [1301.1564] generates a ring in which every point is linearly polarized and diametrically opposite points are orthogonally polarized. Angular masks then pass selected sectors, and a second biaxial crystal remultiplexes the surviving sectors into one beam; a third crystal at the receiver reconstructs them. The experiment demonstrated up to 12 channels over \(4~\mathrm{m}\), with average adjacent-channel crosstalk below \(3\%\) for the 12-channel case [1301.1564]. "A novel space division multiplexing system for free space optical communications" [1308.6405] realized a complementary matched-phase picture: a desired channel is recovered after the third lens if and only if
\[
m_{\mathrm{in},l}(-x,-y)\,m_2(x,y)=\text{constant},
\]
while unmatched channels remain off-axis or annular. The paper treated planar linear, radial linear, and hybrid radial-plus-azimuthal phase encoding in this way [1308.6405]. At the level of learned diffractive optics, "Polarized deep diffractive neural network for classification, generation, multiplexing and de-multiplexing of orbital angular momentum modes" [2203.16087] showed that a polarized D2NN can classify 14 orthogonally polarized vortex beams and de-multiplex hybrid polarized OAM inputs into Gaussian beams at two, three, and four spatial positions [2203.16087].

## 5. Quantum generalization: de-multiplexing classical and quantum information

"Demultiplexing Generalized Information via Quantum Transmission Lines" [2606.17894] extends the concept beyond physical signal labels to the informational content of a quantum system. The proposed Q-DEMUX has one input quantum port \(A\), one output quantum port \(Q\), one output classical port \(C\), and a selector \(s\in\{0,1\}\). If the data is classical, the goal is recovery from the classical output; if the data is quantum, the goal is recovery from the quantum output, with the classical output allowed as side information for correction. A crucial restriction is that the induced quantum channel on \(Q\) is independent of the selector, so the selector chooses between different **instrument realizations** of the same channel rather than between different channels [2606.17894].

The paper defines the total selective-demultiplexing strength \(\mathcal{T}_s(\mathrm D_s)\) and proves the universal bound
\[
\mathcal{T}_s(\mathrm D_s)\le 2\log_2 d_{in}.
\]
Perfect selective demultiplexing corresponds to saturation of this bound, and the characterization theorem states that this is possible if and only if the induced channel is both **classical-quantum** and a **random isometry**. The same work studies a stronger selector-less variant and proves
\[
\mathcal T(\mathrm D)\le \log_2 d_{in},
\]
showing that selector access doubles the maximal generalized-information strength. It also proves that if
\[
\log_2 d_{in} < \mathcal T_s(\mathrm D) \le 2\log_2 d_{in},
\]
the two instrument realizations must be traditionally incompatible [2606.17894]. In this formulation, selective de-multiplexing becomes a problem of quantum instrument theory rather than only of optics or switching networks.

## 6. Recurrent trade-offs, limitations, and interpretive boundaries

Several recurring boundaries appear across these otherwise disparate implementations. First, selective de-multiplexing is not synonymous with arbitrary or lossless separation. In plasmonic and waveguide-array demonstrations, the mechanism may be dispersive focusing or modal transport rather than exact one-port-per-channel filtering, and several proof-of-principle papers do not report full insertion-loss or crosstalk budgets in standard telecom terms [1008.3424] [2306.04100]. In FMF systems, selective demultiplexing can mean efficient collection of the relevant modal subspace followed by MIMO or mesh-based descrambling, not necessarily direct passive recovery of each original lane [2311.03675].

Second, selectivity is realized through different physical resources: wavelength-dependent diffraction in Rowland-type gratings, electro-optic state changes synchronized to photon emission, matched spatial phase cancellation, nonlinear phase matching in \(\chi^{(2)}\) media, reciprocal unitary spatial transforms, or selector-controlled instrument decompositions [1008.3424] [2203.08682] [1308.6405] [1803.00206] [2606.17894]. This suggests that the unifying issue is not a specific hardware class but the existence of a reliable many-to-one mapping from label space to output channel.

Third, the principal trade-offs are domain-specific but structurally similar. Plasmonic devices gain compactness and free-space-to-SPP interfacing but face higher propagation loss and moderate channel count [1008.3424]. Active single-photon demultiplexers remove passive scaling penalties, yet overall multiphoton rates remain extremely sensitive to source brightness, switching fidelity, and duty-cycle constraints [1611.02294] [2304.09622]. Mode-division systems can achieve low crosstalk and high mode counts, but degeneracy, perturbation sensitivity, and fabrication tolerance complicate strict mode-by-mode recovery after propagation [1803.07907] [2311.03675]. Active SFG demultiplexing adds tunability and visible-band detection, but at the cost of pump power, conversion efficiency, and cavity or phase-matching complexity [1803.00206].

Taken together, these results show that selective de-multiplexing is best understood as a family of routing problems unified by intentional, label-dependent output assignment. The label may be spectral, temporal, modal, polarization-resolved, or informational; the implementation may be passive, active, or hybrid; and the notion of “perfect” separation ranges from spatially distinct output spots to exact recoverability conditions on quantum instruments.

Source: https://www.emergentmind.com/topics/selective-de-multiplexing