Quantum Multiplexing: Channels & Applications

• Quantum multiplexing is a strategy that encodes multiple quantum channels using orthogonal spectral, temporal, or spatial modes, thereby enhancing photon efficiency and aggregate rate.
• It is applied across optical communication, quantum repeaters, and high-dimensional error-corrected networks to maximize hardware utilization while preserving quantum coherence.
• Optimized resource-matching and dynamic multiplexing schemes have demonstrated significant improvements in signal fidelity and rate scaling in diverse experimental platforms.

Quantum multiplexing denotes a family of strategies for carrying multiple quantum channels, correlations, or logical degrees of freedom through a shared physical resource. Across the literature, the shared resource may be an optical spectrum, a time-bin frame, a spatial mode set, a memory array inside a repeater or router, a high-dimensional single-photon Hilbert space, or even a measurement-conditioned sector of Hilbert space in a monitored many-body system. The common objective is to increase aggregate rate, photon efficiency, or hardware utilization while preserving the quantum features required for communication, networking, sensing, or computation (Heurs et al., 2010, Abruzzo et al., 2013, Schmolke et al., 2023).

1. Conceptual scope

The term has several technically distinct meanings. In continuous-variable optical communication, quantum multiplexing commonly means assigning independent information-bearing channels to orthogonal spectral modes of a squeezed or entangled optical field. In repeater theory, it means allowing non-parallel matching between occupied memory positions so that more stored elementary links can be swapped per round. In photonic encoding, it means packing multiple qubits into one photon by exploiting orthogonal degrees of freedom such as time bins, frequency bins, polarization, orbital angular momentum, or spatial modes. In monitored many-body dynamics, it can mean the coexistence of trajectory-resolved synchronization channels associated with different decoherence-free subspaces (Heurs et al., 2010, Abruzzo et al., 2013, Nishio et al., 2024, Schmolke et al., 2023).

This breadth is not merely terminological. It reflects a shared structural idea: multiplexing turns latent orthogonality into operational parallelism. Orthogonality may come from cavity resonances, wavelength channels, memory indices, graph-theoretic matchings, or distinct dynamical sectors. The practical gain then depends on whether those orthogonal sectors remain separable under loss, decoherence, detector limitations, and control imperfections.

A second common feature is that multiplexing is usually advantageous only when the physical resource is matched to the signal structure. In squeezed-light communication, the squeezing spectrum should overlap the occupied sub-bands. In repeater multiplexing, the matching strategy should minimize storage-time differences. In high-dimensional photonic transport, qubits should be assigned to photons in a decoder-aware way to avoid correlated erasures. Several of the most favorable results arise precisely from this resource-matching principle (Heurs et al., 2010, Abruzzo et al., 2013, Nishio et al., 2024).

2. Principal modalities

The main modalities described in the literature are summarized below.

Modality	Physical mechanism	Representative work
Frequency-division multiplexing	Orthogonal spectral teeth or bins	(Heurs et al., 2010, Eldan et al., 2023)
Wavelength/state multiplexing	Shared wavelengths participate in several entangled states	(Fan et al., 27 Feb 2025)
Time-bin multiplexing	Distinct temporal slots or orthogonal temporal envelopes	(Zitelli, 5 Sep 2025, Peñas et al., 2024)
Spatial/modal multiplexing	LP, HG, or few-mode-fiber channels	(Zia et al., 2024, Beraza et al., 18 Feb 2025)
Memory-array multiplexing	Non-parallel matching across stored links	(Abruzzo et al., 2013, Kunzelmann et al., 2024)
High-dimensional per-photon multiplexing	Multiple qubits encoded in one photon	(Nishio et al., 2024, Piparo et al., 2019)
Space-time hybrid multiplexing	Reuse of hardware across spatial sites and temporal modes	(Byun et al., 3 Oct 2025)

A canonical optical realization is the squeezed-vacuum frequency comb generated by a sub-threshold optical parametric oscillator. In that system, each cavity resonance is a “tooth” separated by the free spectral range, here $199\ \mathrm{MHz}$, and each tooth is an orthogonal squeezed spectral mode. Time-resolved balanced homodyne detection with a $2.5\ \mathrm{GHz}$ analogue front end and $8\ \mathrm{GS/s}$ sampling resolved the first dozen teeth of a comb spanning more than $2.4\ \mathrm{GHz}$, with digital demultiplexing performed by DFTs over $1.024\ \mu\mathrm{s}$ windows and guard bands used to suppress cross-talk (Heurs et al., 2010).

A later frequency-domain generalization treats an ultra-broadband squeezed source as a bank of orthogonal spectral bins processed in parallel. In that architecture, broadband SPDC around $1560\ \mathrm{nm}$ spanning roughly $150\ \mathrm{nm}$, or about $18\ \mathrm{THz}$, was partitioned into $23$ uncorrelated spectral channels of about $100\ \mathrm{GHz}$ each, with spectral shaping and parametric homodyne used for simultaneous protocol execution across the entire comb (Eldan et al., 2023).

State-multiplexing modifies wavelength-division multiplexing rather than merely stacking more bins. A dual-pump SiN microring resonator supports two degenerate SFWM processes and one non-degenerate process, so one wavelength can be entangled with several partners simultaneously. In the demonstrated four-user fully connected network, six DWDM channels sufficed instead of the twelve channels required by a conventional WDM-only implementation, while preserving simultaneous connectivity and secure communication functionality (Fan et al., 27 Feb 2025).

Spatial multiplexing appears in two related forms. One uses mode-division multiplexing in few-mode fibers with HG or LP modes selected by MPLC mode sorters or photonic lanterns. Another uses multi-core or multi-focus collection from many independent emitters into distinct fiber channels. These approaches add parallel channels without consuming additional spectral bandwidth, but they replace spectral-isolation problems with polarization dependence, mode alignment, and random mode coupling (Zia et al., 2024, Beraza et al., 18 Feb 2025, Koong et al., 2020).

3. Capacity, photon efficiency, and rate laws

In homodyne-limited Gaussian communication, the central benefit of multiplexing is often photon efficiency rather than raw channel count. For a homodyne Gaussian channel, the per-mode capacity is

$2.5\ \mathrm{GHz}$0

with $2.5\ \mathrm{GHz}$1. In the squeezed-comb implementation, placing two sinusoidal carriers on resonance at $2.5\ \mathrm{GHz}$2 and $2.5\ \mathrm{GHz}$3 increased $2.5\ \mathrm{GHz}$4 from approximately $2.5\ \mathrm{GHz}$5 to approximately $2.5\ \mathrm{GHz}$6, and increased the per-mode capacity from approximately $2.5\ \mathrm{GHz}$7 to approximately $2.5\ \mathrm{GHz}$8. The key theoretical point was that, for fixed photon flux $2.5\ \mathrm{GHz}$9, comb-matched squeezing outperforms white squeezing over the same analogue bandwidth: $8\ \mathrm{GS/s}$0 because photons are not spent squeezing out-of-band frequencies (Heurs et al., 2010).

A related but broader energy-efficiency result appears in continuous-variable teleportation. For a single mode with squeezing parameter $8\ \mathrm{GS/s}$1, the mean photon number is $8\ \mathrm{GS/s}$2, and the teleportation-induced additive noise is $8\ \mathrm{GS/s}$3 in the lossless case or $8\ \mathrm{GS/s}$4 with symmetric loss $8\ \mathrm{GS/s}$5. The resulting one-way capacity bounds are

$8\ \mathrm{GS/s}$6

Under a fixed total energy budget $8\ \mathrm{GS/s}$7, optimized multi-mode coding yields aggregate rates scaling linearly with $8\ \mathrm{GS/s}$8, whereas single-mode coding scales logarithmically with $8\ \mathrm{GS/s}$9. The paper’s optimization further gives

$2.4\ \mathrm{GHz}$0

which formalizes the claim of exponentially enhanced rate at equal energy when multi-mode and single-mode schemes are compared (Christ et al., 2012).

In network protocols, rate improvement is often expressed relative to a non-multiplexed baseline. The generalized multiplexing metric

$2.4\ \mathrm{GHz}$1

captures the gain at fixed target fidelity $2.4\ \mathrm{GHz}$2 for $2.4\ \mathrm{GHz}$3-qubit tasks. Semiclassical batching obeys

$2.4\ \mathrm{GHz}$4

and, for $2.4\ \mathrm{GHz}$5, $2.4\ \mathrm{GHz}$6. The quantum-network constructions in that work are noteworthy because they explicitly aim to exceed those semiclassical limits by using coherence across multiplexed modes and by exploiting decoherence regimes that differ between active and idle memories (Propp et al., 7 Oct 2025).

The ultra-broadband spectral architecture provides an operational demonstration of rate scaling by parallelization. There, a proof-of-principle CV-QKD experiment executed the protocol simultaneously over $2.4\ \mathrm{GHz}$7 uncorrelated spectral channels, with the same infrastructure also supporting multiplexed continuous-variable teleportation. This suggests a bridge between the frequency-comb communication picture and the multi-mode capacity results of continuous-variable teleportation (Eldan et al., 2023).

4. Repeaters, routers, and entanglement networks

In repeater settings, multiplexing does not require multiple photonic carriers per se. It can instead be implemented as flexible matching across memory arrays. In finite-range multiplexing, the repeater station contains two arrays of $2.4\ \mathrm{GHz}$8 memories, and after heralded storage a Bell-state measurement can be performed between non-parallel memory positions provided the connection length is at most $2.4\ \mathrm{GHz}$9. The limits are $1.024\ \mu\mathrm{s}$0 for strictly parallel matching and $1.024\ \mu\mathrm{s}$1 for full-range matching. For the two-segment repeater studied there, $1.024\ \mu\mathrm{s}$2 already achieved most of the benefit of full-range multiplexing in both repeater rate and memory-coherence requirements. With $1.024\ \mu\mathrm{s}$3, $1.024\ \mu\mathrm{s}$4, and $1.024\ \mu\mathrm{s}$5, the rate plateau occurred around $1.024\ \mu\mathrm{s}$6 time-bins, and Strategy 1—minimum total storage-time difference $1.024\ \mu\mathrm{s}$7 across a maximum-cardinality matching—reduced the minimal coherence time needed for positive key by roughly $1.024\ \mu\mathrm{s}$8 when moving from $1.024\ \mu\mathrm{s}$9 to $1560\ \mathrm{nm}$0 at $1560\ \mathrm{nm}$1 (Abruzzo et al., 2013).

The multipartite generalization replaces bipartite matching by GHZ-measurement scheduling at a central router. Memories are grouped by party, valid hyperedges contain one occupied memory per party, and the router solves a maximum-cardinality hypergraph matching subject to the same finite-range constraint $1560\ \mathrm{nm}$2. Because decoherence depends on the storage-round index $1560\ \mathrm{nm}$3, the best-performing policy is not merely to maximize the number of GHZ measurements, but to choose among maximum-cardinality matchings using the smallest total storage rounds. In the tripartite analysis, strategy S2—always combining the newest qubits first—and a finite cutoff $1560\ \mathrm{nm}$4 on memory age maximize the secret key rate (Kunzelmann et al., 2024).

State-multiplexed entanglement networks address a different bottleneck: spectral resource scaling in fully connected wavelength-division architectures. Conventional WDM requires

$1560\ \mathrm{nm}$5

DWDM channels for $1560\ \mathrm{nm}$6 users, because each edge consumes a signal-idler pair. The state-multiplexed dual-pump microring reduces that count by letting one wavelength participate in several distinct entangled states. The demonstrated four-user graph used six wavelength channels rather than twelve, and the ten-user design used $1560\ \mathrm{nm}$7 channels rather than $1560\ \mathrm{nm}$8. In the four-user experiment, the distributed BBM92 states yielded a total asymptotic secure key rate of $1560\ \mathrm{nm}$9 across the six edges (Fan et al., 27 Feb 2025).

These works show that, in network architectures, multiplexing is often less about a specific physical degree of freedom than about a matching or assignment freedom. The same logic appears in repeaters, routers, and wavelength-reuse schemes: rates improve when one can reassign successful but otherwise mismatched resources before decoherence erases their value.

5. Error-corrected transport and high-dimensional encodings

A distinct line of work studies multiplexing as a coding primitive. In high-dimensional photonic transport, one photon can carry a qudit of dimension $150\ \mathrm{nm}$0, so the number of qubits per photon is

$150\ \mathrm{nm}$1

This compresses resource counts but converts photon loss into a correlated erasure of all $150\ \mathrm{nm}$2 embedded qubits. For surface codes and hypergraph-product codes, the main question is therefore assignment: which code qubits should share a photon? For surface codes, the best heuristic among those tested was random plus a Manhattan-distance threshold; for HGP codes, sudoku and diagonal assignments, which avoid putting a photon’s qubits in the same rows and columns of the Tanner graph, performed at or above the no-multiplexing baseline in low-erasure regimes. Representative numerical examples include the $150\ \mathrm{nm}$3 surface code and HGP codes such as $150\ \mathrm{nm}$4 and $150\ \mathrm{nm}$5 (Nishio et al., 2024).

Multiplexing also reduces logical-gate overhead in high-dimensional quantum Reed–Solomon encoding circuits. There, a multi-controlled gate $150\ \mathrm{nm}$6 acting on $150\ \mathrm{nm}$7 time-bin controls can be implemented with a single CX plus optical switches, while $150\ \mathrm{nm}$8 reduces to one Toffoli, i.e.

$150\ \mathrm{nm}$9

and

$18\ \mathrm{THz}$0

For the $18\ \mathrm{THz}$1 encoder, the total CX count drops from $18\ \mathrm{THz}$2 to $18\ \mathrm{THz}$3 under the multiplexed decomposition (Nishio et al., 2022).

Distributed photonic quantum error correction gives another resource perspective. Quantum multiplexed photons can carry $18\ \mathrm{THz}$4 qubits each, reducing the number of transmitted photons while preserving the code’s logical function. In the Reed–Solomon analysis at $18\ \mathrm{THz}$5 and $18\ \mathrm{THz}$6, the minimum resource point achieving $18\ \mathrm{THz}$7 was approximately $18\ \mathrm{THz}$8 qubits and $18\ \mathrm{THz}$9 photons, whereas without multiplexing both exceeded $23$0. The same paper also emphasizes that the gain is code- and block-structure dependent, because multiplexing can turn independent erasures into multi-qubit loss events (Piparo et al., 2019).

A more physically explicit entanglement-distribution variant is QMUXING, where a single multi-DOF photonic pulse sequentially entangles many remote memories and can be combined with purification. Its normalized rate metric,

$23$1

makes the resource trade-off explicit by dividing purified-pair rate by weighted matter-qubit and photon costs. In that framework, QMUXING outperforms Deutsch-style pair-by-pair generation both in normalized rate and in raw resource consumption, and its built-in purification reduces matter-qubit requirements from $23$2 to $23$3 (Piparo et al., 2018).

6. Beyond channel multiplexing

Quantum multiplexing also appears in settings where “channel” no longer means a physical subcarrier. In waveguide QED, orthogonal temporal envelopes $23$4 and frequency-separated carriers play analogous roles. A concrete orthogonal temporal family is built from sech-based pulses satisfying

$23$5

Single-photon simulations show that matched temporal controls give transfer efficiencies above $23$6, while an orthogonal mode can be rejected with probability approximately $23$7. However, simultaneous temporal-mode multiplexing with resonant nodes is largely spoiled by scattering-induced distortions and waveguide-mediated coherent exchange. The work therefore identifies frequency multiplexing as the scalable solution, with a practical spacing rule of $23$8, under which multi-channel fidelity approaches the product of single-channel fidelities and dozens of multiplexed photons become feasible in state-of-the-art experiments (Peñas et al., 2024).

In continuously monitored many-body systems, multiplexing can arise from stochastic projection into distinct decoherence-free subspaces. For a monitored Hamiltonian $23$9 and Hermitian measurement operator $100\ \mathrm{GHz}$0, a trajectory trapped in a DFS evolves noiselessly under

$100\ \mathrm{GHz}$1

If different DFSs support different eigenfrequencies, then distinct single trajectories synchronize at different frequencies. The number of possible channels is bounded by

$100\ \mathrm{GHz}$2

and the channel probabilities are set by the initial overlap weights, not by the measurement strength. In the $100\ \mathrm{GHz}$3 XY-chain example, two $100\ \mathrm{GHz}$4-dimensional DFSs support the four frequencies $100\ \mathrm{GHz}$5, $100\ \mathrm{GHz}$6, $100\ \mathrm{GHz}$7, and $100\ \mathrm{GHz}$8, so different trajectories of the same monitored system can realize different stable synchronization channels (Schmolke et al., 2023).

Space-time hybrid multiplexing extends the idea again, this time to a modular neutral-atom computer. In MAQCY, Q-Pairs—dual-species Rydberg-blockaded atom/superatom pairs—are reused across spatial sites and temporal modes, with temporal translation $100\ \mathrm{GHz}$9 moving logical information forward while physical atoms are replaced. The resource claim is structural: the architecture uses $2.5\ \mathrm{GHz}$00 physical qubits for $2.5\ \mathrm{GHz}$01 logical qubits, independent of circuit depth, in contrast to static globally controlled layouts requiring $2.5\ \mathrm{GHz}$02 atoms for full connectivity. The reported composite translation fidelity is $2.5\ \mathrm{GHz}$03, or $2.5\ \mathrm{GHz}$04 with improved transport, and a three-qubit quantum Fourier transform is implemented using only global operations and atom transport (Byun et al., 3 Oct 2025).

These nonstandard usages do not erase the communication meaning of multiplexing; they generalize it. What is being shared is still a constrained quantum resource—frequency support, controllable Hilbert-space sectors, or physical qubit hardware—but the orthogonal sectors are defined dynamically or logically rather than by a fixed external channelization.

7. Experimental platforms and practical constraints

Experimental realizations show that multiplexing gains are invariably mediated by engineering constraints. In single-mode telecom fiber, wavelength-division multiplexing can combine a quantum O-band channel with a classical C-band timing pulse. In the demonstrated system, the O–C differential group delay was $2.5\ \mathrm{GHz}$05, C-band pulse broadening was $2.5\ \mathrm{GHz}$06, and the converted O-band path provided a $2.5\ \mathrm{GHz}$07 transmission advantage over direct $2.5\ \mathrm{GHz}$08 transmission at $2.5\ \mathrm{GHz}$09, despite only approximately $2.5\ \mathrm{GHz}$10 overall quantum-frequency-conversion efficiency. The principal structured impairment was leakage through WDMs with minimal $2.5\ \mathrm{GHz}$11 isolation, which produced a separate coincidence peak that could nevertheless be removed by timing gates (Lenhard et al., 2015).

In few-mode fibers, modal multiplexing replaces spectral leakage with random mode coupling. Over an $2.5\ \mathrm{GHz}$12 low-DMGD few-mode link, the measured average modal insertion loss of the MUX–FMF–DeMUX chain was $2.5\ \mathrm{GHz}$13, average inter-group cross-talk was $2.5\ \mathrm{GHz}$14, and the quantum fractional power analysis showed that $2.5\ \mathrm{GHz}$15 of photons migrated to group $2.5\ \mathrm{GHz}$16 and $2.5\ \mathrm{GHz}$17 to group $2.5\ \mathrm{GHz}$18, with about $2.5\ \mathrm{GHz}$19 remaining in the originally excited groups. Under co-propagation with two classical channels, the quantum SNR stayed above $2.5\ \mathrm{GHz}$20 for classical output powers up to $2.5\ \mathrm{GHz}$21 per channel, implying a shot-noise-limited classical baud-rate bound of $2.5\ \mathrm{GHz}$22 per classical channel (Zia et al., 2024).

A closely related SDM experiment exploited detector dead time as a resource rather than a limitation. Three quantum spatial channels in an $2.5\ \mathrm{GHz}$23 FMF carried $2.5\ \mathrm{GHz}$24 time-bin frames, i.e. $2.5\ \mathrm{GHz}$25 qubits per frame, and frame scheduling was chosen so that detector recovery windows suppressed inter-channel modal cross-talk at the detection stage. The resulting aggregate rate was $2.5\ \mathrm{GHz}$26, corresponding to

$2.5\ \mathrm{GHz}$27

with measured interference visibility $2.5\ \mathrm{GHz}$28, symbol error $2.5\ \mathrm{GHz}$29, and $2.5\ \mathrm{GHz}$30 for $2.5\ \mathrm{GHz}$31 (Zitelli, 5 Sep 2025).

Spatial multiplexing of emitters provides another hardware route. A multi-core-fiber confocal microscope generated seven excitation and collection spots matched to a deterministically positioned nanowire quantum-dot array, with measured focus spacing $2.5\ \mathrm{GHz}$32. Two quantum dots were tuned into resonance near $2.5\ \mathrm{GHz}$33, and multiplexed background-free single photons were collected from both, with antibunching fits giving $2.5\ \mathrm{GHz}$34 for each source. The same platform demonstrated coherent control of a biexciton–exciton cascade, with a $2.5\ \mathrm{GHz}$35-pulse at $2.5\ \mathrm{GHz}$36 and resonant-excitation values $2.5\ \mathrm{GHz}$37 for $2.5\ \mathrm{GHz}$38 and $2.5\ \mathrm{GHz}$39 for $2.5\ \mathrm{GHz}$40 under pulsed TPE (Koong et al., 2020).

Demultiplexing hardware is itself an essential part of multiplexed quantum systems. An all-optical SFG demultiplexer converted selected telecom DWDM channels to $2.5\ \mathrm{GHz}$41 by tuning the pump wavelength, reached a maximum quantum conversion efficiency of $2.5\ \mathrm{GHz}$42 at $2.5\ \mathrm{GHz}$43 pump power, and preserved energy-time entanglement with post-conversion raw visibilities of $2.5\ \mathrm{GHz}$44, $2.5\ \mathrm{GHz}$45, and $2.5\ \mathrm{GHz}$46 across three channels (Li et al., 2018).

Across platforms, the recurrent limiting factors are the same: cross-talk between nominally orthogonal sectors, mode- or polarization-dependent loss, detector bandwidth or dead-time constraints, phase stabilization, and the cost of demultiplexing or matching. A plausible synthesis is that the decisive question is not whether orthogonal sectors exist, but whether the full chain—generation, routing, storage, measurement, and post-processing—preserves that orthogonality at the operating error budget. Quantum multiplexing succeeds when that systems-level condition is met; when it fails, the additional channels merely convert one bottleneck into another.