Quantum-Network Nodes with Real-Time Noise Mitigation Using Spectator Qubits
Paper details
Section titled “Paper details”| Field | Value |
|---|---|
| Journal | PRX Quantum |
| Published | 2026-01-13 |
| Authors | S.J.H. Loenen, Y. Wang, N. Demetriou, C.E. Bradley, T.H. Taminiau |
| DOI | 10.1103/zgm8-6b8l |
Two nuclear spectator qubits improved memory fidelity of an NV center during emulated remote entanglement when dephasing had accumulated. In an ideal model, one spectator restored fidelity from 1/2 to 3/4 at (g=1) and phase standard deviation (\sigma=1). The gate-based implementation also produced a higher memory Bloch vector length than measurement-based readout for most XY initial-state assignments.
Correlated noise supplies the signal
Section titled “Correlated noise supplies the signal”Spectator qubits are auxiliary nuclear spins that sample the correlated phase noise acting on a memory nuclear spin. The NV electron spin supplies the optical interface and drives this noise through always-on hyperfine coupling. In the paper’s idealized limit, small transverse coupling makes the electron-conditioned nuclear dynamics commute, so the loss is pure dephasing. Spectator outcomes update the memory’s phase distribution through a Bayesian posterior. Each later readout basis is chosen from the outcomes already collected. Nonzero transverse coupling breaks this simple relation and reduces the accuracy of spectator inference.
This method differs from protected subspaces and quantum error correction because the memory need not be encoded before the noise acts. The authors could therefore skip spectator operations when remote entanglement succeeded early and little noise accumulated. The remaining trade-off is that extracting the spectator information adds control and readout disturbance.
Emulated remote-entanglement sequence
Section titled “Emulated remote-entanglement sequence”The experiment used one NV center in diamond as a network-node emulator. The C0 carbon spin served as memory, while C1 and C2 served as spectators. The (M) spectators were initialized in the X basis.
Each attempt prepared an electron-spin superposition, waited for (t_e), and optically reset the electron stochastically. The sequence repeated (N_{\mathrm{REA}}) times and applied a nuclear-spin echo halfway through. It omitted the optical (\pi) pulse needed for spin-photon entanglement, so it did not create a remote entangled state.
After the attempts, the electron was reset and (K) of (M) spectators were read out sequentially. Real-time feedback set each readout basis from prior outcomes and applied phase corrections to the memory.
Here (N_{\mathrm{REA}}) denotes the number of emulated remote-entanglement attempts.
| Experimental quantity | Value | Condition |
|---|---|---|
| Magnetic field | 404 G | Along the NV symmetry axis |
| C0 memory coupling | (A_\parallel=24.4) kHz, (A_\perp=24.8) kHz | Carbon nuclear spin |
| C1 spectator coupling | (A_\parallel=-36.3) kHz, (A_\perp=26.6) kHz | First spectator, (g=-1.49) |
| C2 spectator coupling | (A_\parallel=20.6) kHz, (A_\perp=41.5) kHz | Second spectator, (g=0.84) |
| Waiting-time scan | (N_{\mathrm{REA}}=30) | C0 memory and C1 spectator |
Measurement-based results and overhead
Section titled “Measurement-based results and overhead”The first test showed the expected correlation between C0 and C1. Conditioning on a bright or dark C1 result projected C0 toward opposite Y directions. The two-spectator sequence produced a marked fidelity gain after substantial dephasing.
C2 gave the more marked gain at high (N_{\mathrm{REA}}). The authors linked this to (g=0.84), which was more favorable than (-1.49) for large dephasing. Adding spectators reduced fidelity when the attempt count was low because readout caused more dephasing than it corrected.
The Z-basis state lost little fidelity, confirming that dephasing dominated the process. The method does not protect Z-basis states from bit flips.
For the XY-state measurement-based fit, (A_{\mathrm{SPAM}}) rose from (0.082 \pm 0.007) without spectators to (0.279 \pm 0.008) with two. The gate-based fits used fixed (A_{\mathrm{SPAM}}) values of 0.066 and 0.125 for zero and two spectators. Initializing more spectators barely affected the memory, but reading them out reduced the Bloch vector length.
The experimental trend agreed with density-matrix simulations that included nonzero transverse coupling. Two mechanisms explain the readout cost. A faulty readout outcome causes an incorrect Bayesian update of the memory phase. Stochastic electron flips during readout add an unknown phase and further dephasing.
The readout laser used 100 pW. It stopped after 98 µs if no photon arrived and switched off immediately after detection. A 1 µs feedback clock allowed the laser to switch off after photon detection. The sequence used about 500 ns of repumping at 600 nW. The fitted electron reinitialization decay constant was 92 ± 4 ns.
Gate-based refocusing
Section titled “Gate-based refocusing”The gate-based method replaced measurement and classical feedforward with a spectator-controlled electron flip followed by a wait. That wait imposed an electron-controlled phase rotation on every nuclear spin.
Since the same parallel couplings created the dephasing, a correctly timed wait partly refocused the memory and other spins. The optimum was the first maximum in the wait-time response. Later maxima depended on individual hyperfine couplings. A wait that was too long overcompensated the phase and added dephasing.
At (N_{\mathrm{REA}}=30), the C0 memory and C1 spectator scan produced a maximum in memory Bloch vector length at the optimal delay. The authors compared gate-based and measurement-based implementations through the ratio (R) of their Bloch vector lengths. They tested permutations of C0, C1, and C2 as memory and spectator qubits.
The ratio (R_Z) stayed close to 1 for a Z-basis memory, as expected because the method addresses dephasing. The ratio (R_{XY}) exceeded 1 for most, but not all, permutations with X- or Y-basis memories. The implementation omitted electron readout, the main measurement-induced noise source. This confirmed that removing mid-circuit readout improved the protocol for coherence-carrying states.
Choosing spectators by success probability
Section titled “Choosing spectators by success probability”The authors also modeled a probabilistic entanglement process with per-attempt success probability (p). Success after (N_{\mathrm{REA}}) attempts had probability (P(N_{\mathrm{REA}})=(1-p)^{N_{\mathrm{REA}}-1}p). Their expected fidelity weighted (F(N_{\mathrm{REA}})) by this distribution and compared zero, one, and two initialized spectators. The single-spectator analytical fit included state-preparation and measurement errors matched to (N_{\mathrm{REA}}=0).
Low (p) gives more weight to late successes, where accumulated dephasing makes spectator information useful. Thus one or two spectators outperform no spectators in that regime. At (p \approx 1), early successes dominate and spectator overhead lowers expected fidelity.
The paper did not implement a per-attempt real-time choice after a successful heralded event. The proposed strategy would initialize several spectators, then select measurement or gate use after success. That choice would require longer initialization and one-time calibration of readout angles or electron wait times. Spectators with limited control fidelity may still be useful because their information can be ignored when it offers no net gain.
Limits for quantum-network engineering
Section titled “Limits for quantum-network engineering”Several limits define the current result. The experiment emulated remote entanglement rather than linking separate nodes. It also omitted transfer of the electron state to an idle nuclear spin after the attempt sequence. The method targets correlated dephasing, not arbitrary errors or bit flips.
Echo sequences, extra electron pulses, and other defect systems can change the noise correlations. The authors state that effectiveness can rise or fall with those changes. The supplementary analysis found typical (A_\perp \approx 20) kHz at about 403 G, producing a 2.7-degree precession-axis tilt. Simulations repeated with (A_\perp=0) changed little in this system.
Earlier network demonstrations still had memory decoherence faster than new entangled-state generation, giving active link efficiency below 1. Spectator sensing therefore complements changes to entanglement rate, process optimization, and memory coupling.
For diamond hardware, the host must support an NV electron spin and nearby carbon spins with known hyperfine couplings. System calibration should include (A_\perp), parallel coupling ratios, electron reset errors, and spectator readout faults. Spectators become useful only when phase information exceeds the disturbance added by its extraction.
Materials and support
Section titled “Materials and support”The quantum-network node requires a diamond host containing an NV center and a coupled carbon nuclear-spin register. 6CCVD’s diamond materials catalog includes single-crystal diamond grown by MPCVD, although the paper does not specify the sample’s growth method or defect preparation. The engineering support page can be used to define the host and fabrication requirements for this application.
View Original Abstract
Quantum networks might enable quantum communication and distributed quantum computation. Solid-state defects are promising platforms for such networks, because they provide an optical interface for remote entanglement distribution and a nuclear-spin register to store and process quantum information. A key challenge toward larger networks is to improve the storage of previously generated entangled states during new entanglement generation. Here, we introduce a method that uses “spectator” qubits combined with real-time decision making and feedforward to mitigate dephasing of stored quantum states during remote entanglement sequences. We implement the protocol using a single nitrogen-vacancy (NV) center in diamond and demonstrate improved memory fidelity. Our results show that spectator qubits can improve quantum network memory using minimal overhead and naturally present resources, making them a promising addition for near-term testbeds for quantum networks.
High Concentration Quantum Grade Single Crystal Diamond
View Specs →Products
View Specs →
Single Crystal Cvd Diamond Plate
View Specs →References we cover
Section titled “References we cover”- Entanglement of Spin-Pair Qubits with Intrinsic Dephasing Times Exceeding a Minute
- Realization of a multinode quantum network of remote solid-state qubits
- Resonant Excitation and Purcell Enhancement of Coherent Nitrogen-Vacancy Centers Coupled to a Fabry-Perot Microcavity
- Quantum Network Nodes Based on Diamond Qubits with an Efficient Nanophotonic Interface
- Robust Quantum-Network Memory Using Decoherence-Protected Subspaces of Nuclear Spins