Experimental realization of a bidirectional quantum analog-to-digital converter between photonic wavefront and qubits Authors: Lei Chen, Gong-Chu Li, Xu-Song Hong, Si-Qi Zhang, Hua-Qing Xu, Yuan-Cheng Liu, Geng Chen, Chuan-Feng Li, Guang-Can Guo Published: 2026 ABSTRACT AbstractQuantum systems can be fundamentally classified into discrete-variable (DV) and continuous-variable (CV) systems. The degree of quantum advantage achieved with a quantum system is often dictated by the scalability, robustness, and operational feasibility-factors that are closely tied to the nature of the variables employed. Notably, DV and CV systems are largely complementary in these factors. Consequently, integrating DV and CV variables holds significant promise for overcoming the inherent challenges associated with relying solely on one type of variable system. In this study, we build a bidirectional quantum analog-to-digital converter (QADC/QDAC), which facilitates deterministic and reversible quantum information transfer between CV and DV systems. This QADC/QDAC is hardware-efficient, and the error decreases exponentially with the number of qubits, thereby allowing high-fidelity information transfer with a few qubits. With this photonic QADC/QDAC, we demonstrate efficient CV mode tomography leveraging DV tomography techniques and achieve deterministic generation of non-Gaussian CV mode from qubits. Our findings address a critical barrier in constructing hybrid quantum processors and open up new avenues for scientific exploration previously inaccessible to single-variable quantum systems.Subject terms: Quantum information, Quantum optics Introduction Quantum information processing (QIP) seeks to achieve quantum advantage over classical methods by encoding quantum information into various physical variables, which can be broadly classified as continuous variables (CV) or discrete variables (DV), depending on the specific QIP tasks and the underlying physical systems. The degree of quantum advantage ultimately realized is largely determined by fundamental system characteristics, such as scalability, robustness, and feasibility-properties that are closely linked to the nature of the chosen encoding variables. While DV systems allow for universal control in noisy, many-qubit environments, their limited scalability poses a significant challenge for implementing effective error correction1–3. In contrast, CV systems are well-suited for generating large-scale quantum states through long-range interactions4–6 and are compatible with hardware-efficient single-mode error correction7–9 and high-dimensional operations10–12, owing to the infinite dimensionality. However, achieving universal quantum computation within CV systems remains a major challenge due to the difficulty of implementing controlled non-Gaussian operations10,13. Importantly, DV and CV systems demonstrate complementary strengths in terms of scalability, error correctability, and operational flexibility12,14–17. Consequently, a hybrid CV-DV architecture presents a promising strategy for integrating the respective advantages of both systems and addressing current limitations in quantum information science. In support of this line of thinking, one recent theory indicates that hybrid approaches reduce the gate complexity of quantum Fourier transforms (QFT) from O(n2) two-qubit gates to O(n) CV-DV interaction gates18. To fully harness the potential of hybrid CV-DV systems, it is essential to achieve deterministic and reversible quantum information transfer between analog and digital signals-effectively realizing a quantum analog-to-digital converter (QADC) and its inverse counterpart, a quantum digital-to-analog converter (QDAC). Unlike state transfer between disparate physical systems19–22, the implementation of QADC/QDAC presents a fundamental challenge: establishing a one-to-one mapping between finite- and infinite-dimensional Hilbert spaces. One potential approach involves performing a finite sampling of the CV system using DV systems. However, for such a protocol to be practically viable, the QADC/QDAC must not only ensure efficient utilization of qubit dimensionality but also rely on experimentally accessible interaction criteria that are not concurrently satisfied by earlier protocols cited in refs. 23, 24. Recent theoretical advances18,25 offer a promising resolution to this challenge by enabling CV-to-DV conversion through two experimentally feasible interactions, with an error rate that diminishes exponentially as the number of qubits increases. Nevertheless, the experimental implementation of this protocol remains technically intractable. Specifically, the two-mode interaction must be applied sequentially and iteratively across individual qubits, requiring the ability to store qubit states and preserve coherence both within and between qubits throughout the entire process. In this work, we develop a photonic QADC/QDAC architecture that enables deterministic, bidirectional quantum information transduction between discrete-variable qubits and continuous-variable photonic wavefronts. Unlike traditional field quadratures - which are defined by canonical quantum statistical properties - the photonic wavefront constitutes a high-dimensional quantum resource enabling advanced information processing tasks, including holographic quantum imaging and high-dimensional quantum communication. The polarization degree of freedom (DOF) serves as an intermediary qubit that directly couples with the CV state, after which the sampled information is transferred to a separate which-path qubit. Repeated application of these operations facilitates deterministic conversion between CV and DV states without the need for quantum memory or continuous coherence preservation across the entire qubit ensemble. We show that with only 2 or 3 qubits, the transfer fidelity reaches approximately 0.97 for Gaussian modes and surpasses 0.80 for higher-order optical modes, including non-Gaussian CV states. This hardware-efficient QADC/QDAC provides a foundational framework for the development of hybrid CV-DV quantum processors. In our experiment, we demonstrate the versatility of our QADC/QDAC by implementing efficient CV mode tomography leveraging DV tomography techniques via QADC and by achieving deterministic generation of non-Gaussian CV modes from qubits via QDAC. Results The algorithm of QADC/QDAC The interaction between CV and DV systems constitutes a critical element required for the implementation of QADC/QDAC. The QADC/QDAC achieved in our work relies on two kinds of non-commuting Jaynes-Cummings (JC)-type26,27 interactions, which can be viewed as controlled displacement gates associated with a pair of non-commuting quadratures, namely, X and P. Given a total of N qubits, the two-mode interactions applied to the n-th qubit are expressed as: where 12Δ2n and Δ2n2 represent the coupling strength updated with the sampling iteration. With this two-mode interaction, ref. 25 introduced a quantum circuit for converting CV states into qubit-based DV representations, accomplished by sequentially interacting the CV mode with each individual qubit. However, this prototype approach requires the preservation of both individual qubit information and inter-qubit coherence throughout the entire process-challenges that remain significant obstacles in experimental implementations. To overcome these limitations, we propose a variant algorithm of the prototype one presented in ref. 25, which is enhanced by quantum state swapping and depicted in the quantum circuit diagram in Fig. 1. Specifically, an intermediary qubit is selected to interact with the CV mode, and the acquired quantum information is then transferred to additional qubits through a following swapping operation. Through the execution of this quantum circuit, a CV mode ψCV is mapped onto an N-qubit system initialized in the state 0⊗N via the application of the two-mode interaction, and the resulting state evolution can be formulated as where ΨDV is the outcome qubit state, and 0~CV=∫dxsincπx2Δx is an input-independent state on the CV mode. Due to the dimensional mismatch between CV and DV systems, the residual entangled state ΦCV-DV forms an orthogonal basis with respect to 0~CV, and the real number ϵ denotes the error probability, which depends on the input state. As expected, ϵ approaches zero in the limit as N → ∞25. The quantum circuit to implement bidirectional quantum analog-to-digital conversion (QADC/QDAC). The input signal corresponds to an arbitrary continuous-variable (CV) wavefunction ∣ψ⟩ with an unknown functional form, while the qubits are initialized in the ground state ∣0⟩. During each sampling iteration, a two-mode interaction W^nV^n is applied to entangle the CV mode with the qubits. A designated qubit (labeled with subscript 1) is selected to interact with the CV mode, and the resulting sampling information is subsequently transferred to another qubit. Upon completion of the QADC circuit, the information encoded in the CV state is mapped onto the discrete-variable (DV) qubits. The entire process operates in a deterministic and reversible manner, effectively implementing a unitary transformation UAD. The corresponding QDAC circuit can be realized by applying the inverse transformation UDA=UAD† to the input qubits. The resulting qubit state retains the majority of the information from the initial CV mode and can be expressed as where the summation is over 2N terms, and ∣ϕ~s⟩ is the basis of the qubit space. The function ψ(xs) represents the wavefunction of the input CV mode evaluated at equidistant positions, which are sampled by the qubit state with a uniform sampling step of 2Δ, ranging from −(2N − 1)Δ to (2N − 1)Δ. Consequently, the wavefunction of the input CV mode at position xs can be reconstructed by measuring the complex amplitude of the DV basis ∣ϕ~s⟩. Interpreting Eq. (3) in this manner clarifies the selection of appropriate transfer parameters, namely Δ and N. Specifically, a sufficiently small sampling step is required to resolve rapid variations in Ψ, while a sufficiently large N ensures coverage of the entire wavefunction support. In other words, reducing Δ and increasing N lead to a higher fidelity in the state transfer. Notably, the transfer process is non-destructive, as it follows from the unitary evolution UAD depicted in Fig. 1. As a result, the original CV state can be fully reconstructed by applying the inverse unitary transformation UAD† to couple the qubits with a sinc-distributed CV state ∣0~⟩CV. The experimental implementation of QADC/QDAC In the experiment, the CV mode is selected as the photonic transverse wavefront, of which the position and momentum operators satisfies the commutation relation [X^,P^]=iℏ, which is mathematical isomorphism to the commutator of field quadrature operators X1^ (amplitude) and X2^ (phase), and thus can be viewed as dimensionless quadratures28. Due to this mathematical isomorphism, the JC-type interaction formalized in ref. 25 can be implemented on a photonic wavefront using a high-efficiency diffractive element fabricated from liquid crystal, specifically a polarization grating (PG). The PG possesses a spatially modulated optical fast axis and imparts two opposing spatial phase gradients on the incident wavefront depending on the input circular polarization of the photons, thereby enabling controlled momentum displacement operations (see Fig. 2). Consequently, the PG functions as an ideal coupling interface between the photon wavefront and polarization, and is equivalent to V^n in Eq. (1). The other JC-type interaction W^n can be expediently implemented by positioning a PG on the conjugate focal plane of an optical 4-f system with a focal length f = 10 cm (see Supplementary Material Section I for mathematical analysis). The explicit form of the two-mode interaction used in the experiment is given by where Λ denotes the period of the PG, which determines the coupling strength analogous to Δ in Eq. (1), and λ denotes the photon wavelength, experimentally set to 800 nm. The Pauli operators σ^y=RL−LR and σ^z=HV−VH are defined with respect to the circular polarization basis R/L and the linear polarization basis H/V, respectively. Experiment setup of photonic bidirectional quantum analog-to-digital conversion (QADC/QDAC). Photons carrying an arbitrary pattern of the continuous-variable (CV) mode enter the circuit from the left and undergo several sampling iterations, each consisting of a two-mode interaction W^nΛV^nΛ. These interactions can be implemented using a pair of polarization gratings (PGs), with one PG placed on the focal plane of a 4-f system and sand-switched by quarter-wave plates (QWPs). The coupling strength hinges on the period of the applied PG, which is updated iteratively during the sampling process, as defined in Eq. (4). Between consecutive sampling iterations, a beam displacer (BD) is inserted to perform a swapping operation between the polarization and which-path qubits, followed by a half-wave plate array (HWPA) that resets the polarization qubit to ∣H⟩. Upon exiting the circuit, the CV input is sampled at discrete points indicated by the blue dots, and the corresponding information is encoded into the qubit space spanned by the polarization basis (∣H⟩ or ∣V⟩) and the which-path basis (∣0⟩ or ∣1⟩). By introducing a predefined discrete-variable (DV) mode and a CV null state ∣0~⟩CV from the right-hand side, the circuit is also capable of realizing QDAC, mapping arbitrary multiqubit states into a single CV mode. In each sampling iteration, the operator W^nΛV^nΛ is applied to the joint state of the optical wavefront and photon polarization. Subsequently, a swapping operation is performed between the polarization qubit and a which-path qubit, through which the which-path qubit acquires the sampled information from the CV mode. At the conclusion of each sampling iteration, the polarization qubit is reset to H and prepared to act as the intermediary qubit that will directly interact with the CV mode in the subsequent iteration. As a result, the wavefront on each path undergoes the same evolution regardless of the number of sampling iterations executed, thereby maintaining consistent operational coherence across all spatial modes. Given that the optical components responsible for the interaction - specifically the PG and BD - possess time-reversal symmetry, the QDAC can be implemented by applying the time-reversed counterparts of the operations used in QADC. Consequently, photons propagating in the reverse direction through the circuit shown in Fig. 2 undergo an inverse sequence of operations expressed as ∏n=1N(V^N−n+1†W^N−n+1†), through which the information encoded in the qubit state is transferred to the CV mode. As illustrated in Fig. 2, QADC protocols for two-qubit and three-qubit systems were realized within the photonic setup. The two-qubit QADC employs 4 sampling points, spanning from −3Δ to 3Δ, with an interval of 2Δ = 0.8 mm (see Supplementary Material Section II for two-qubit QADC experimental details). The three-qubit QADC achieves finer sampling across 8 points, ranging from −7Δ to 7Δ, with an interval of 2Δ = 0.4 mm. Three representative CV modes are demonstrated in the experiment – specifically, the Gaussian mode, the Hermite–Gaussian mode, and the cosine-modulated mode, which are respectively mathematically equivalent to the Gaussian state, Fock states, and the cat-like state29–31 when viewed in terms of the Wigner distribution (see Supplementary Material Section III). For all these modes, the proposed QADC circuit enables high-fidelity conversion using only 2 or 3 qubits. The performance of this QADC process is evaluated using the fidelity between the ideal CV mode and the reconstructed CV mode by applying a perfect QDAC to the qubit state generated during the QADC experiment, as illustrated above each plot in Fig. 3. This figure of merit accounts for both experimental imperfections and algorithmic errors, and the theoretical fidelity accounting exclusively for algorithmic errors is also calculated. The close agreement between these two fidelity metrics indicates that our QADC/QDAC implementation faithfully realizes the algorithm. In practice, discrete sampling with phase misalignment maps the real-valued wavefunction onto qubits with a non-zero imaginary part, leading to deviations from an ideal real-valued CV state reconstruction, and such deviations are quantified and visualized by the green curves in Fig. 3. The Gaussian mode features a relatively smooth spatial profile, and a 2-qubit QADC configuration already achieves a fidelity of 0.969, which further improves to 0.984 with the addition of a third qubit. In contrast, the Hermite-Gaussian (HG01) and cosine-modulated modes exhibit greater spatial complexity and are more sensitive to errors. Nevertheless, the QADC fidelity remains above 0.80 with 2 qubits and increases to approximately 0.95 when a third qubit is employed. The evaluation of the QADC via qubit tomography (see Supplementary Material Section IV for state tomography details) demonstrates a feasible methodology for conducting efficient and accurate tomography of CV modes, which remains a significant challenge for conventional methods. QADC output for three CV modes. Panel (a, d) correspond to the Gaussian mode, (b, e) to the HG01 mode, and (c, f) to the cosine mode. The upper panels (a, b, c) present the results obtained from a 2-qubit QADC configuration, whereas the lower panels (d, e, f) illustrate those from a 3-qubit setup. In each panel, the blue and purple bars represent the real and imaginary parts of the amplitudes in each DV basis, reconstructed from the quantum state tomography result (see Supplementary Material Section IV). Applying the QDAC unitary UAD† on these amplitudes, the encoded CV information is reconstructed and visualized as red lines (real part) and green lines (imaginary part). The degree of overlap between the red line and the black dashed line-which represents the ideal wavefunction of the input CV mode-serves as a measure of QADC performance, quantified by the transfer fidelity Fexp reported above each plot, alongside the ideal theoretical transfer fidelity Fideal, which accounts exclusively for algorithmic imperfections under perfect experimental conditions. The reverse process, referred to as QDAC, enables the generation of CV modes from a limited number of qubits. Notably, non-Gaussian CV modes can be generated without requiring complex nonlinear operations. To validate this capability, the input qubit states are prepared to be the expected states obtained from applying an ideal QADC to the target CV mode. In our experiment, by encoding information onto one polarization qubit and an additional which-path qubit, three distinct CV modes are synthesized (see Supplementary Material Section II for QDAC experimental details). Their wavefronts are reconstructed through interference with a reference beam. The amplitude and phase distributions across the reconstructed wavefronts are displayed in Fig. 4, showing good qualitative agreement with the target CV modes. The quality of the QDAC can also be characterized by the experimental fidelity, as shown above each panel with the calculated ideal fidelity. It may initially appear counterintuitive that the experimentally measured fidelity exceeds the ideal fidelity; however, this observation is physically plausible: experimental imperfections introduce stochastic modulation of the output CV mode, and such random deviations have a chance to yield a state that more closely resembles the target mode, especially in cases where the CV modes are obtained from fewer qubits. The Gaussian mode is successfully recovered from two input qubits with a fidelity of 0.97. For the HG01 and cosine-modulated modes, the fidelities remain at satisfactory levels of 0.903 and 0.834, respectively. The conversion fidelity is fundamentally limited by the discrete and finite nature of the sampling process - particularly for highly localized states. Nevertheless, extended theoretical simulations indicate that seven qubits are sufficient to achieve a conversion fidelity exceeding 0.99, even for cubic phase states and Gottesman-Kitaev-Preskill (GKP) states, both of which play a pivotal role in fault-tolerant quantum computation (see Supplementary Material Section V). Generation of CV modes via QDAC from 2 input qubits. The target outputs for the three panels correspond to Gaussian (a), Hermite-Gaussian (HG01) (b), and cosine-modulated (c) modes, with their theoretical amplitude profiles displayed as black dashed lines. The input qubits are initialized in the expected quantum state that would result from applying UAD to the target continuous-variable (CV) mode, shown as the blue bars in each panel. The output CV wavefront generated by the QDAC circuit is reconstructed using an interference technique involving a reference beam, allowing for the simultaneous measurement of both the intensity distribution and phase map, as shown in the colormaps beneath the plots. The real and imaginary components of the reconstructed wavefront are represented by red and green lines, respectively, with the associated standard error bands derived from 20 frames of colormap data. The experimentally measured fidelity and the ideal theoretical fidelity are reported above each panel. Discussion Our work achieves a deterministic and bidirectional conversion mechanism that maps the continuous spatial amplitude distribution of an optical wavefront onto a set of discrete path-encoded qubits. Experimentally, we realize high-fidelity preparation and coherent manipulation of non-Gaussian states in spatial CV modes, alongside high-efficiency quantum tomography of such modes. By compressing CV mode information into a compact qubit register, our framework allows direct application of advanced DV state estimation protocols32–34, whereby bypassing the need for exhaustive measurement basis scanning inherent in standard CV tomography (e.g., sequential position- or momentum-space sampling via mechanical slits or waveplate-based homodyne variants)35–37. Furthermore, QADC permits direct reconstruction of the Wigner function in CV systems via optimized positive operator-valued measures (POVMs) implemented on the mapped qubits, thereby eliminating reliance on phase-stable local oscillators required in interferometric methods38,39 or the scalability bottlenecks of single-qubit-assisted readout schemes12,40 involving intensive CV-DV interactions. Beyond state characterization, the deterministic and lossless mapping between photonic wavefronts and discrete qubits enables several near-term applications: single-mode-fiber-compatible holographic imaging; faithful storage of high-dimensional spatial modes in single-mode quantum memories; programmable, diffraction-limited wavefront shaping without photon loss; and a bidirectional interface for free-space and integrated photonic platforms. Additionally, when QADC/QDAC is integrated into quantum computing platforms, it facilitates CV-to-CV mode transduction-i.e., transferring quantum information from one CV mode to another via a shared qubit bus-thus enabling distributed quantum processing across spatially separated CV nodes. Finally, our mapping supports the synthesis of Gottesman-Kitaev-Preskill (GKP)-like states: a single logical qubit can be encoded as a superposition of squeezed Gaussian peaks in phase space7,41,42. This establishes a natural bridge for porting bosonic quantum error-correcting codes, which is originally designed for CV hardware, to fault-tolerant operation in DV architectures. This exponential scaling inherently imposes a practical constraint on the maximum number of qubits feasible within the present setup. However, this limitation is significantly alleviated by the intrinsic hardware efficient property of our scheme, and high-fidelity analog-to-digital conversion is achieved even with a small number of qubits. To further enhance scalability, the sampling information encoded in the CV mode can be transferred to discrete-variable photonic encodings, such as frequency-bin or time-bin qubits, that are compatible with scalable, integrated photonic platforms. Methods CV state preparation in QADC Due to the finite sampling step of 2Δ, the CV state must be confined within a frequency range not exceeding 14Δ, corresponding to a maximum diffraction angle of 3.42′ of the wavefront. As a result, the propagation direction of light must be precisely controlled and maintained throughout the experimental setup. Under these constraints, conventional reflective spatial light modulators (SLMs) are unsuitable for preparing CV states, as they introduce non-negligible deviations in the propagation direction of light in practice. Instead, a transmissive optical device with ultra-low surface irregularity is utilized to ensure directional stability while enabling precise wavefront modulation. In our experiments, three distinct types of CV states are prepared across separate experimental configurations. First, the Gaussian state is directly acquired from the output of a collimator with a beam waist diameter of 2 mm. Second, the HG01 mode state, which corresponds to the superposition state 12+1+−1 in the orbital angular momentum (OAM) basis, can be generated using a q-plate in combination with a polarizing beam splitter (PBS). The q-plate enables the conversion of polarization states into corresponding OAM modes, while the PBS performs polarization projection to complete the state preparation. Third, a cosine-mode state is synthesized using a 3200 μm diffraction grating together with a PBS. This configuration introduces a symmetric frequency shift in opposite directions within the spectrum range, thereby generating the desired cosine-shaped spatial mode. Reconstruction of the CV mode exiting from QDAC Conventional self-interference techniques for characterizing the wavefunction of CV modes, such as differential phase detection43,44 and Hartmann-Shack wavefront sensing45,46, become unsuitable in the presence of singular points, such as the one located at the center of the HG01 mode. To address the challenge of reconstructing the CV mode generated by QDAC, a vertically polarized reference beam, split by a polarizing beam splitter (PBS) at the input of the circuit, is made to interfere with the horizontally polarized signal beam at an additional PBS located at the output. However, the extended interferometric paths in the setup are prone to phase fluctuations, which significantly degrade the performance of four-step phase-shifting algorithms47. To mitigate this issue, the combined reference and signal beams are split by using a non-polarizing beam splitter (NPBS). The two output ports of the NPBS correspond to projection measurements onto the polarization bases 12(H+V) and 12(H+iV), which induce relative phases as 0 and π/2 radians between the two beams, respectively. This dual-projection configuration enables the simultaneous acquisition of two interference patterns on a charge-coupled device (CCD) camera. By extracting phase information from these co-registered interference patterns and incorporating concurrent intensity distribution measurements, the phase profile of the CV mode can be accurately reconstructed. Supplementary information Supplementary Information Transparent Peer Review file