Multi-Output Quantum Pulse Gate

• Multi-Output Quantum Pulse Gate is a programmable quantum interface that uses engineered nonlinear optics to map input modes onto multiple output channels via dispersion-engineered sum-frequency generation.
• It leverages multiple quasi-phasematching peaks in a super‐poled waveguide to achieve simultaneous, high-fidelity frequency conversion while reducing alignment complexity compared to cascaded systems.
• The device supports reconfigurable pulse shaping for high-dimensional mode sorting, enabling efficient quantum state tomography and scalable integration in quantum networks.

A multi-output quantum pulse gate is a programmable quantum interface that maps selected input modes onto multiple distinct output channels in parallel. In the nonlinear-optical literature, it denotes a dispersion-engineered sum-frequency-generation device in a super-poled nonlinear waveguide, where pump shaping programs projective measurements or linear transformations on single-photon temporal and frequency modes, and the converted photons emerge in distinct spectral channels (Serino et al., 2024). The concept generalizes the original single-output quantum pulse gate proposed for spectrally engineered sum-frequency generation in a PPLN waveguide (Eckstein et al., 2010), later extended to engineered frequency conversion and quantum pulse shaping (Brecht et al., 2011). By 2022 and 2024, multi-output implementations had been demonstrated as high-dimensional decoders and programmable time-frequency mode sorters for single photons (Serino et al., 2022). In a broader control-engineering usage, closely related multi-output pulse-gating platforms have also appeared as programmable radio-frequency control systems for multi-qubit experiments (Keitch et al., 2017).

1. Genealogy and conceptual scope

The original quantum pulse gate (QPG) was introduced as a mode-selective frequency converter that extracts a single, well-defined broadband spectral mode from an ultrafast, multi-mode quantum state of light and upconverts it to a new frequency via sum-frequency generation in a $\chi^{(2)}$ medium (Eckstein et al., 2010). Its defining principle is spectral engineering of the sum-frequency-generation phasematching function together with pulse shaping of a bright classical gating pulse, so that orthogonal broadband modes can be addressed individually while residual modes remain unconverted because of orthogonality. A closely related 2011 generalization placed the QPG and the difference-frequency-generation-based quantum pulse shaper in a common framework of engineered frequency conversion in nonlinear optical waveguides, emphasizing Schmidt-mode selectivity, group-velocity matching, and the effective beamsplitter interaction between broadband temporal modes (Brecht et al., 2011).

The multi-output extension replaces the single phasematching lobe of a conventional QPG with multiple quasi-phasematching peaks, each defining a distinct output frequency channel. In the experimentally demonstrated mQPG, the waveguide is super-poled so that one nonlinear interaction yields $d$ simultaneous outputs, and pump shaping determines which temporal or frequency-mode superpositions are directed to which output (Serino et al., 2022). This distinguishes the mQPG from a cascaded single-output architecture, in which one mode is sorted per stage and the total complexity grows through repeated conversion and demultiplexing. A standard single-output QPG has one phasematching peak and one output frequency channel; the mQPG uses a super-poled waveguide with multiple quasi-phasematching peaks centered at distinct output frequencies, so that each peak serves as an independent output port encoded in wavelength space (Serino et al., 2024).

The terminology also acquired a broader meaning in quantum hardware. The rf pulse control system of 2017 is described as a practical, scalable pulse-gating platform for multi-qubit experiments: it synthesizes, gates, and shapes many coherent drive signals, supports conditional branching, and supplies tens of synchronized outputs for trapped-ion control (Keitch et al., 2017). This suggests that “multi-output quantum pulse gate” now names a family of architectures unified less by a single physical mechanism than by a common systems function: programmable, mode-selective, multi-channel quantum control.

2. Nonlinear-optical operating principle

In the photonic mQPG, the core interaction is three-wave mixing in a dispersion-engineered, periodically poled LiNbO$_3$ waveguide operated in sum-frequency generation. A strong, shaped classical pump at angular frequency $\omega_p$ converts a single-photon signal at $\omega_s$ to an up-converted mode at $\omega_c \approx \omega_p + \omega_s$, with the process engineered for temporal-mode selectivity (Serino et al., 2024). In the frequency domain, a convenient interaction-picture Hamiltonian is

$$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$

where $a_s(\omega_s)$ annihilates a signal photon, $b_c(\omega_c)$ annihilates a converted photon, $\alpha_p$ is the complex pump envelope, and $d$0 encodes phasematching and spatial overlap. Under group-velocity engineering, the transfer kernel factorizes as

$d$1

so that energy conservation is pump-determined while momentum conservation is waveguide-determined (Serino et al., 2024).

The mathematical structure is governed by a Schmidt decomposition of the transfer kernel,

$d$2

with high mode selectivity corresponding to one dominant Schmidt coefficient per output channel. In the multi-output device, the waveguide and poling are engineered so that the phasematching depends primarily on output frequency and yields multiple, narrow phasematching peaks $d$3. Each peak defines a distinct output channel, while the pump shape selects the input mode or superposition associated with that channel. In the Heisenberg picture, the channel transformation is written as

$d$4

with $d$5 set by pump shaping and phasematching, and $d$6 representing loss, inconclusive events, and noise. The per-mode selectivity is

$d$7

This is the formal statement of the mQPG as a programmable, probabilistic multi-element measurement or, in the appropriate regime, as a mode-selective linear transformation (Serino et al., 2024).

A central physical condition is near group-velocity matching between pump and signal so that the joint transfer function is nearly separable. Earlier QPG theory already identified this separability as the route to single-mode conversion in PPLN waveguides and to an effective broadband-mode beamsplitter Hamiltonian (Eckstein et al., 2010). The later mQPG preserves that logic but replicates it across multiple spectrally distinct output ports.

3. Device architectures and programming

The experimentally demonstrated mQPG uses a titanium indiffused LiNbO$d$8 waveguide, periodically super-poled, with poling period $d$9 for type-II sum-frequency generation. The polarizations are input $_3$0 H, pump $_3$1 V, and output $_3$2 H. Five quasi-phasematching peaks yield five output channels, with center frequencies separated by $_3$3 for the Fancy Frequency Bin scheme in $_3$4, or $_3$5 otherwise; the phasematching bandwidth per peak is $_3$6 (Serino et al., 2024). The pump is supplied by a mode-locked Ti:sapphire laser with $_3$7 repetition rate, center $_3$8, and spectral FWHM $_3$9, then shaped by a folded 4f line with grating, cylindrical mirror, and SLM. The input is provided by an OPO seeded by Ti:sapphire, centered at $\omega_p$0, with optional state preparation through a commercial telecom waveshaper.

Programmability resides primarily in the spectral amplitude and phase shaping of the classical pump. The phasematching structure is fixed by fabrication, whereas pump shaping tailors the implemented measurement basis across channels. The device supports pulse-mode bases, frequency-bin states, time-bin states, and arbitrary superpositions. Hermite–Gauss modes are used experimentally; examples include HG0 spectral FWHM $\omega_p$1 in $\omega_p$2 and $\omega_p$3 in $\omega_p$4. Frequency bins use FWHM $\omega_p$5 with separation $\omega_p$6 in $\omega_p$7, and FWHM $\omega_p$8 with separation $\omega_p$9 in $\omega_s$0. Time bins have FWHM $\omega_s$1 and separations $\omega_s$2 in $\omega_s$3 or $\omega_s$4 in $\omega_s$5 (Serino et al., 2024).

Two programming strategies are central. In standard frequency-bin sorting, $\omega_s$6 channels and $\omega_s$7 input bins require $\omega_s$8 shaped pump sub-bands aligned to the phasematching peaks. In the Fancy Frequency Bin (FFB) scheme, $\omega_s$9 pump bins spaced by the channel separation are used, so that one global pattern of pump phases sorts $\omega_c \approx \omega_p + \omega_s$0 input modes. The FFB method reduces technical complexity and increases pump-to-phasematching bandwidth ratio (Serino et al., 2024). In the five-dimensional 2022 decoder, the pump spectrum is carved into five spectral peaks, each peak is shaped to implement one temporal mode of a five-dimensional basis, the fundamental HG bandwidths are set to FWHM $\omega_c \approx \omega_p + \omega_s$1, and the pump peaks are separated by $\omega_c \approx \omega_p + \omega_s$2 (Serino et al., 2022).

This architecture is programmable across mutually unbiased bases. In $\omega_c \approx \omega_p + \omega_s$3, the mQPG decoder was programmed to realize measurements in any of the 6 MUBs, with the additional bases generated as superpositions of five HG modes through programmable phase and amplitude control on the pump spectrum (Serino et al., 2022). In 2024, the same general platform was shown to switch between pulse modes, frequency bins, time bins, and their superpositions with no hardware changes beyond SLM-based spectral reconfiguration (Serino et al., 2024).

4. Tomography, fidelities, and measured operation

The principal experimental benchmark is detector tomography. In the 2024 implementation, the mQPG acts as a multi-element measurement described by POVMs $\omega_c \approx \omega_p + \omega_s$4, and for a programmed basis $\omega_c \approx \omega_p + \omega_s$5 one expects ideal projectors $\omega_c \approx \omega_p + \omega_s$6. For each channel $\omega_c \approx \omega_p + \omega_s$7, the POVM is reconstructed by minimizing a weighted least-squares functional under positivity and Hermiticity constraints, and the channel fidelity is defined as

$\omega_c \approx \omega_p + \omega_s$8

with average fidelity $\omega_c \approx \omega_p + \omega_s$9 (Serino et al., 2024). At the single-photon level with a time-of-flight spectrograph, the average fidelity reaches up to $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$0 in $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$1 and up to $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$2 in $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$3. With a CCD spectrograph of $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$4 resolution, frequency-bin sorting reaches $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$5 in $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$6 and $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$7 in $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$8, and all tested alphabets exceed $$H = \hbar \int d\omega_s \, d\omega_c \, \kappa(\omega_s,\omega_c)\, \alpha_p(\omega_c-\omega_s)\, a_s(\omega_s)\, b_c^\dagger(\omega_c) + \text{h.c.}$$9 (Serino et al., 2024).

The 2022 five-dimensional decoder reported complementary tomography figures. With the CCD spectrograph, which reflects intrinsic mQPG performance, measurement tomography gave fidelity $a_s(\omega_s)$0 and purity $a_s(\omega_s)$1, corresponding to the demultiplexing fidelity $a_s(\omega_s)$2 stated in the abstract. With the lower-resolution time-of-flight spectrograph, fidelity was $a_s(\omega_s)$3 and purity $a_s(\omega_s)$4 (Serino et al., 2022). The same platform enabled resource-efficient state tomography of 25 random pure five-dimensional input states. “Raw” tomography assuming ideal projectors gave fidelity $a_s(\omega_s)$5 and purity $a_s(\omega_s)$6, whereas calibrated tomography using reconstructed POVMs gave fidelity $a_s(\omega_s)$7 and purity $a_s(\omega_s)$8, consistent with the average fidelity $a_s(\omega_s)$9 emphasized in the abstract (Serino et al., 2022).

Selectivity and cross-talk depend strongly on output readout resolution. In the 2022 system, the internal mQPG average selectivity was $b_c(\omega_c)$0, while the complete decoder with time-of-flight readout showed average selectivity per MUB between $b_c(\omega_c)$1 and $b_c(\omega_c)$2 because of the spectrograph’s $b_c(\omega_c)$3 resolution (Serino et al., 2022). In the 2024 study, cross-talk per off-diagonal element for frequency bins with CCD readout was typically at or below the $b_c(\omega_c)$4–$b_c(\omega_c)$5 level, whereas with time-of-flight readout of effective resolution $b_c(\omega_c)$6 the finite resolution dominated the cross-talk, especially in $b_c(\omega_c)$7 (Serino et al., 2024).

A recurrent point is that the device is intrinsically probabilistic because of finite sum-frequency-generation efficiency, yielding a “no-conversion” outcome that is excluded from fidelity estimates. In the 2024 work, pump powers of $b_c(\omega_c)$8–$b_c(\omega_c)$9 were sufficient for single-photon-level tomography with per-measurement integration times of $\alpha_p$0–$\alpha_p$1 and $\alpha_p$2–$\alpha_p$3 acquisitions per setting, but the absolute efficiencies were not reported (Serino et al., 2024). This corrects a common misunderstanding: high reported fidelities characterize the quality of converted events and reconstructed measurement operators, not unconditional deterministic throughput.

5. From mode sorter to programmable quantum network

The mQPG can be interpreted not only as a detector but also as a programmable frequency-bin interferometer. A 2024 theoretical framework describes linear optical quantum networks based on an mQPG and a type-0 PDC source, with the transfer function programmed directly by the pump spectrum (Folge et al., 2024). For an mQPG with

$\alpha_p$4

the Schmidt modes are the programmed superpositions $\alpha_p$5 and the fixed outputs $\alpha_p$6, and each pair behaves as an independent tunable beamsplitter. At unity conversion,

$\alpha_p$7

which is operator-identical to routing the input vector of frequency bins through a linear network with matrix $\alpha_p$8 (Folge et al., 2024). The practical scaling estimate is

$\alpha_p$9

and with state-of-the-art parameters the paper projects a few hundred input bins per device. Because unconverted superposition modes pass through, cascading multiple mQPGs increases the number of accessible outputs (Folge et al., 2024).

A later cavity-assisted proposal pushes this network view further. The cavity-assisted sum-frequency-generation gate of 2025 is described as deterministic, universal, and fully programmable, implementing an $d$00-by-$d$01 truncated unitary transformation, or a full unitary when $d$02, on frequency-bin modes (Chen, 5 Dec 2025). In that framework, the multi-output functionality is realized by addressing multiple cavity idler resonances simultaneously with orthogonal pump tones, with transfer matrix elements

$d$03

Under optimal coupling, $d$04, the device reaches near-unity conversion in the lossless limit, while internal loss bounds the peak conversion efficiency to $d$05 (Chen, 5 Dec 2025). The reported attainable dimensionality is $d$06 on the order of $d$07, with $d$08 up to about one thousand using current components. A plausible implication is that the mQPG lineage is evolving from high-dimensional mode sorting toward fully programmable high-dimensional unitary processing.

A distinct cavity route had already been proposed in 2017 through the dichroic-finesse cavity QPG, where a bad cavity for the signal and a good cavity for the converted field make the Green’s function separable in time and enable near-perfect temporal-mode selectivity (Reddy et al., 2017). That architecture has two inherent outputs, $d$09 and $d$10, and can realize multi-output behavior in time bins through read-out control pulses.

6. Limitations, trade-offs, and broader control-hardware usage

The nonlinear-optical mQPG is limited by finite conversion efficiency, finite spectrograph resolution, pump–signal timing synchronization, and fixed phasematching bandwidth. The phasematching peaks are narrow, $d$11, while output-channel spacing is $d$12–$d$13; as the effective spectral bandwidth grows, especially in the FFB scheme, delay stabilization and compensation of thermal drifts become more stringent (Serino et al., 2024). Pump phase noise and residual dispersion in the 4f shaper introduce small systematic errors, and calibration and pre-compensation are required. Scaling to larger dimension requires additional super-poled phasematching peaks, sufficient pump bandwidth and resolution, and better shot-by-shot spectrographs. The FFB method reduces technical demands because its pump-resource scaling is linear, $d$14 pump bins, rather than $d$15 in the standard approach (Serino et al., 2024). Relative to cascaded single-output QPGs, the multi-output device reduces insertion loss, alignment overhead, and cumulative phase instability because sorting is performed in one nonlinear interaction.

In a broader control-engineering usage, a multi-output quantum pulse gate is a control platform that can simultaneously synthesize, gate, and precisely shape many coherent drive signals used to implement quantum logic operations across multiple qubits (Keitch et al., 2017). The rf implementation described in 2017 combines direct digital synthesis per channel, a DAC-driven variable-gain amplifier for envelope shaping, FPGA event sequencing, and a synchronized backplane architecture. Each channel card provides four independent rf outputs, up to eight cards can be inserted in one backplane instrument for 32 synchronized outputs, and multiple instruments can be synchronized via a shared reference clock and Ethernet control. The native frequency range is $d$16–$d$17, extension to the gigahertz regime is achieved via mixers, pulse timing resolution is $d$18, and real-time feedback supports quantum error syndrome measurement and correction within $d$19 (Keitch et al., 2017). In trapped-ion experiments using this system, single-qubit gate fidelities of $d$20 and two-qubit gate fidelities of $d$21 were reported elsewhere, together with transform-limited shaped pulses and conditional mixed-species sequences (Keitch et al., 2017).

The photonic mQPG and the rf multi-output pulse-gating platform are physically different devices, but they share a systems-level structure: both distribute programmable amplitude, phase, and timing control across many synchronized channels, both support conditional or basis-dependent operation, and both are designed for scalability. This suggests that the term “multi-output quantum pulse gate” has become a cross-domain label for hardware that converts programmable pulse structure into parallel, high-fidelity quantum operations.