Implementation of a 46-node quantum metropolitan area network

INTRODUCTION

• Earlier QKD networks had fewer than ten nodes, while practical deployment also requires attention to topology, scalability, key management, applications, and robustness.The introduction places these challenges within the broader goal of global QKD communication.Source: INTRODUCTION · pp. 1–2
• The 46-node Hefei network connects 40 user nodes, three trusted relays, and three optical switches across the city.It was designed to address topology, scalability, key management, applications, and node-failure challenges.Source: INTRODUCTION · p. 2

Network topology

• In the network diagram, switch-connected users have transmitters and receivers, whereas trusted-relay-connected users have transmitters only.Three sub-networks are directly connected, with users attached through optical switches or trusted relays.Source: Figure 1 · p. 2
• Full connectivity avoids relays and single points of failure but has quadratically increasing link costs; star networks use linearly many links but are vulnerable to central-switch failure.Trusted-relay networks can extend distance, but users must trust the relay.Source: Network topology · p. 3
• The network combines full connections among three sub-networks with star-like local access, using either trusted nodes or optical switches at star centers.The full connections provide robustness between important users, while star-like connections are used for more efficient local access.Source: Network topology · p. 3
• The optical switches redistribute quantum signals, while trusted nodes assign classical keys; the 16-port switch can support eight simultaneous user pairs.Switch losses were below 1.2 dB, compared with 50 dB channel isolation.Source: Network topology · p. 3

Standard QKD equipments

• Switch-connected users have both quantum transmitters and receivers, while trusted-relay-connected users have transmitters only, reducing the need for more expensive receivers at every node.The network uses polarization-encoding BB84 with a vacuum-plus-weak-decoy-state method.Source: Standard QKD equipments · p. 3
• Standard devices use four 1550 nm lasers for BB84 polarization states and produce signal, decoy, and vacuum pulses; calibration takes no more than five minutes.The calibration includes timing and polarization feedback; standardization lets device count scale linearly with user nodes.Source: Standard QKD equipments · p. 4
• The QKD setup uses signal and decoy mean photon numbers of 0.6 and 0.2 per pulse, respectively, in a 6:1:1 signal-decoy-vacuum ratio.The setup uses four-channel InGaAs single-photon detection with 10% detection efficiency and a 500 ps effective gate width.Source: Figure 2 · p. 4

Key management

• The switching strategy prioritizes users with the least stored key material, and the Roll-Call-Polling protocol determines the pairing queue.The 16-port switch supports 120 possible user-pairing schemes; switching intervals can be configured from 10 to 60 minutes.Source: Key management · p. 4
• A new device authenticates with 32 kbit of initial keys using HMAC based on SM4, then joins the key-generation queue within two minutes of power-on.The passage notes that this authentication does not provide information-theoretic security.Source: Key management · pp. 4–5

Security analysis

• The BB84 secret-key-rate expression combines the signal-state error-correction cost with a single-photon contribution estimated using decoy states.The passage defines the signal-state gain and QBER, and identifies single-photon yield and phase error as quantities estimated by the decoy-state method.Source: Security analysis · p. 5
• Finite-size statistical fluctuations primarily affect phase-error estimation, with the Z-basis single-photon phase error bounded using the X-basis single-photon bit error rate and an optimized deviation.The analysis uses the numbers of X- and Z-basis bits in the deviation calculation.Source: Security analysis · p. 5
• Authentication, error verification, and privacy amplification also have failure probabilities that depend on secure-key consumption and add under composable security.The analysis describes these alongside finite-size parameter-estimation effects.Source: Security analysis · p. 5

Application

• One-time-pad encryption was used for voice telephone, instant messaging, and digital-file transmission, with 10 Gbits of information encrypted.The encryption speed was 800 Mbps and total encryption delay was less than 50 µs.Source: Application · p. 5
• Real-time voice telephone ran at 2.4 kbps, while file transmission ran at 320 kbps.Source: Application · p. 5
• Twenty-two users simultaneously made calls in the quantum network for six minutes.The network capacity test lasted 50 minutes.Source: Application · p. 5

Network robustness

• Twenty-two users simultaneously made QKD calls; green areas mark the duration of each call.Source: Figure 3 · p. 6
• For representative links, key rates are plotted over time for relay-to-relay and relay-to-user connections, using monthly averages from measurements every 30 seconds.Eleven user nodes ran continuously during the robustness test.Source: Figure 4 · p. 6
• Average key rates ranged from 6 to 60.5 kbps during 31 months of continuous operation, with polarization-driven fluctuations followed by calibration.Table II lists average inter-subnetwork key rates and within-subnetwork ranges.Source: Network robustness · p. 5, Table II · p. 6

DISCUSSION

• The Hefei network uses standard commercial QKD products and systematic key management, and the authors describe scaling it with added users and relays and connecting it to backbone and satellite-linked networks.Source: DISCUSSION · p. 6

• FPGA-based Winnow reconciliation has efficiency 1.3 ∼1.5 and takes less than 50 ms for a 256 kbits sifted key when the bit error rate is below 4%.A CRC determines whether the reconciled key string is retained or abandoned.Source: §1 · p. 7
• Privacy amplification estimates the secure factor with the decoy method and uses blockwise FPGA Toeplitz multiplication, with finite-key effects included in parameter estimation.Source: §1 · p. 7
• At 20 km, QBER is generally below 1% and bidirectional key rate reaches 30 kbps; the robustness test uses a 40 MHz clock.Total loss includes fiber loss proportional to one fourth of fiber length in kilometers, detector loss of about 3 dB, and detector efficiency of about 10%.Source: §1 · p. 7

• Detector-gate calibration scans detector positions to maximize visibility, with a 15% minimum and abortion after three consecutive below-threshold readings for any detector.Alice sends 40 MHz optical pulses during this step.Source: §2 · p. 7
• Polarization feedback requires at least 20 dB visibility and checks H, V, +, and − pulses, aborting after three failed trials.After H passes, H is switched off and V is checked; + and − are then calibrated.Source: §2 · p. 7
• Synchronization adjusts Alice’s lasers to 100 kHz to match the synchronization pulses and aborts if any laser fails three consecutive times.Source: §2 · p. 7
• Total calibration time stays within 5 minutes; stronger pulses improve calibration, so laser power rises with transmission distance.Source: §2 · p. 7
• Fiber polarization drift is the network’s main noise source; TR2–TR3 calibration occurred 15 times in 24 hours, with daytime calibration every 56 minutes.Overnight, polarization was stable and the system ran steadily for almost 11 hours.Source: §2 · p. 8

• The control center authenticates devices when they log in, then controls authenticated devices and provides neighboring-device connection configurations for QKD.Source: §3 · p. 8
• Authentication keys are automatically replenished after use, but lost keys require manual reset; these keys are stored separately and reserved for identity authentication.Source: §3 · p. 8
• New secure keys are encrypted with previously generated keys before storage in secure modules, which use FPGA-based isolation and anti-detection measures.The system also restricts electromagnetic radiation.Source: §3 · p. 8

4. Details of the key rates

• Detailed key rates are reported in Supplementary Tables IV–VIII, with a typical link profiled over a week in Supplementary Figure 5.The section points to the supplementary tables for network-wide link rates and to the figure for the typical-link time series.Source: §4 · p. 8

Supplementary Note 2: Key Rate Calculation

• Vacuum and weak-decoy states bound the single-photon yield Y1 from below and the single-photon error rate e1 from above.Signal and weak-decoy intensities are μ and ν; their gains are Qμ and Qν, while the dark-count yield is Y0 and its error rate is e0 = 0.5.Source: Supplementary Note 2: Key Rate… · p. 8
• Finite-size phase-error bounds use random sampling to account for deviations between phase- and bit-error rates across bases.The calculation uses the single-photon bit-error bound and applies statistical-fluctuation substitutions to the key-rate quantities.Source: Supplementary Note 2: Key Rate… · pp. 8–9
• The finite-key calculation uses sifted-key counts nx and nz from the X and Z bases, respectively, to set qx = nx/(nx + nz).The BB84 expression is recalled as the starting key-rate formula.Source: Supplementary Note 2: Key Rate… · pp. 8–9
• Finite-size estimation uses Gaussian fluctuations set to 10 standard deviations, with bounds applied to observed gains, error rates, and detection counts.The bounds also cover successful single-photon detections, which cannot be directly observed; signal and weak-decoy probabilities are qs = 0.75 and qd = 0.125.Source: Supplementary Note 2: Key Rate… · p. 9
• The total finite-key length is Ktot = Kx + Kz, and the key rate is r0Ktot/N, where r0 is the source repetition rate.The post-processing key consumption Δ is of order 10^2 bits when each step has failure probability 10^-10.Source: Supplementary Note 2: Key Rate… · p. 9
• The example calculates a finite-size key rate for link UB-23 to TR-3 using experimental parameters listed in Supplementary Table IX.The resulting rate is reported in Supplementary Table IV; the authors plan to update the calculation module with the Chernoff-Hoeffding method.Source: Supplementary Note 2: Key Rate… · pp. 8–9