Wavelength-Dependent Beamsplitter Attack on CV-QKD
The threat actor (Eve) exploits the wavelength dependence of the beamsplitter in the receiver's (Bob's) Continuous-Variable (CV) QKD receiver to control measurement results and mask the noise added by an intercept-and-resend attack.
Eve stands in the line between the transmitter (Alice) and Bob, measures what Alice sends, and then sends Bob her own light at wavelengths the receiver was never designed for. At those wavelengths the splitter in front of Bob's detector no longer divides light evenly, so what Bob records is Eve's reading rather than Alice's signal, while every number Bob checks for signs of an eavesdropper still reads normal. The published analyses agree the result is the whole key rather than a fragment of it: Alice and Bob finish the protocol believing the link was secure [Huang2013,Ma2013].
The attack works against both receiver designs in use. [Huang2013] treats the heterodyne arrangement and [Huang2014] the homodyne arrangement that the commercially sold CV-QKD unit of the time actually used, so neither detection scheme protects against it on its own.
| [Feng2024] | Kangyi Feng, Yijun Wang, Yin Li, Yuang Wang, Zhiyue Zuo, Ying Guo. "Underwater Wavelength Attack on Discrete Modulated Continuous-Variable Quantum Key Distribution" (2024) 10.3390/e26060515. https://www.mdpi.com/1099-4300/26/6/515. |
| [He2020a] | Zhen He, Yijun Wang, Duan Huang. "Wavelength attack recognition based on machine learning optical spectrum analysis for the practical continuous-variable quantum key distribution system" (2020) 10.1364/josab.386394. |
| [Huang2013] | J. Huang, C. Weedbrook, Z. Yin, S. Wang, H. Li, W. Chen, G. Guo, and Z. Han. "Quantum hacking of a continuous-variable quantum-key-distribution system using a wavelength attack" In: Phys. Rev. A 87, 062329. (2013) 10.1103/PhysRevA.87.062329. |
| [Huang2014] | J. Huang, S. Kunz-Jacques, P. Jouguet, C. Weedbrook, Z. Yin, S. Wang, W. Chen, G. Guo, and Z. Han. "Quantum hacking on quantum key distribution using homodyne detection" In: Phys. Rev. A 89, 032304. (2014) 10.1103/PhysRevA.89.032304. |
| [Jouguet2013] | P. Jouguet, S. Kunz-Jacques, and E. Diamanti. "Preventing calibration attacks on the local oscillator in continuous-variable quantum key distribution" In: Phys. Rev. A 87, 062313. (2013) 10.1103/PhysRevA.87.062313. |
| [Li2011] | H. Li, S. Wang, J. Huang, W. Chen, Z. Yin, F. Li, Z. Zhou, D. Liu, Y. Zhang, G. Guo, W. Bao, and Z. Han. "Attacking a practical quantum-key-distribution system with wavelength-dependent beam-splitter and multiwavelength sources" In: Phys. Rev. A 84, 062308. (2011) 10.1103/PhysRevA.84.062308. |
| [Li2022] | D. Li, Y. Tang, Y. Zhao, L. Zhou, Y. Zhao, and S. Tang. "Security of Optical Beam Splitter in Quantum Key Distribution" In: Photonics 9, 527. (2022) 10.3390/photonics9080527. |
| [Ma2013] | X. Ma, S. Sun, M. Jiang, and L. Liang. "Wavelength attack on practical continuousvariable quantum-key-distribution system with a heterodyne protocol" In: Phys. Rev. A 87, 052309. (2013) 10.1103/PhysRevA.87.052309. |
| [Qi2015] | Bing Qi, Pavel Lougovski, Raphael Pooser, Warren Grice, and Miljko Bobrek. "Generating the Local Oscillator “Locally” in Continuous-Variable Quantum Key Distribution Based on Coherent Detection" In: Phys. Rev. X 5, 041009. (2015) 10.1103/PhysRevX.5.041009. arXiv:1503.00662. |
| [Soh2015] | Daniel B. S. Soh, Constantin Brif, Patrick J. Coles, Norbert Luetkenhaus, Ryan M. Camacho, Junji Urayama, and Mohan Sarovar. "Self-referenced continuous-variable quantum key distribution protocol" In: Phys. Rev. X 5, 041010. (2015) 10.1103/PhysRevX.5.041010. arXiv:1503.04763. |
| [Tan2021] | X. Tan, Y. Guo, L. Zhang, J. Huang, J. Shi, and D. Huang. "Wavelength attack on atmospheric continuous-variable quantum key distribution" In: Phys. Rev. A 103, 012417. (2021) 10.1103/PhysRevA.103.012417. |
08cf3307-b4b4-4766-a87d-e828bbfea27a
- Quantum Communication
Technique properties
Operational properties
| Description | Countermeasure |
|---|---|
|
Real-time quadrature measurement with random optical attenuations is the one detection control that bites on the homodyne route of this attack, but only in its three-ratio form. The two-ratio scheme does not work: [Huang2014] states that the real-time shot noise measurement scheme as proposed in [Jouguet2013] cannot detect this type of attack, because that procedure assumes the measured variance is a linear function of the attenuation ratio while the attack makes it quadratic. The repair is inside the same control. With a third attenuation ratio the receiver (Bob) can fit all three coefficients of the variance against ratio curve and test that the quadratic term is negligible. [Huang2014] gives concrete settings, ratios of 1, 0.5 and 0.001 applied with probabilities of 90, 5 and 5 percent, at no net key-rate cost against the two-ratio design, and claims the repaired control defeats all then-known attacks on the detection apparatus of Gaussian CV-QKD. Feng et al. (2024), a different group, independently recommend a third attenuation ratio in the shot-noise monitoring module. |
CM-0087 Real-time Quadrature Measurements with Random Optical Attenuations |
|
Spectral monitoring at the receiver (Bob) is the half of the filtering defence that actually detects the attack. The existing procedure stands: Bob should randomly perform monitoring with and without a wavelength filter and compare the readings [Huang2013]. The two controls only work as a pair, and [Huang2014] says why. A narrow filter in front of detection forces the threat actor (Eve) to use high power in order to push enough off-band light through, and monitoring of the local-oscillator intensity is what sees those signals. Neither element is sufficient alone. Two later and independent sources agree: [Li2022] reports that filters make it necessary for an eavesdropper to increase the light intensity to conduct the attack, thus also increasing the probability of detecting the attack, and Feng et al. (2024) recommend the same pairing. |
CM-0011 Watchdog Detector |
|
CM-0086 is pre-deployment characterisation, and that is the half of the defence it carries here: sweep the receiver's splitter and its 10:90 monitoring tap across the whole band an attacker can reach and compare the measured response against what the security argument assumes. The sweep is what surfaces the imbalance. [Huang2014] shows the measurement being done, on commercial double-wavelength-type 10:90 and 50:50 parts at eighteen wavelengths from 1270 nm to 1610 nm. Acting on the result, by selecting a splitter whose ratio is flat across that band, is the preventive step the characterisation enables. Both later sources recommend it, and one of them quantifies the target: [Li2022] states that beam splitters made of mirror coatings are recommended and that the variation of splitting ratio should be restricted to lower than 1 dB across 1260 nm to 1700 nm, which is a figure a defender can check a bill of materials against; Feng et al. (2024) recommend a wavelength-independent beam splitter for the continuous-variable case. |
CM-0086 Prior Hardware Component Study |
|
Machine-learning classification of the receiver's optical spectrum is a detection control aimed at this attack, proposed by He et al. (2020) using a linear-discriminant-analysis support-vector-machine classifier over spectral data. It has something real to classify, because the attack necessarily puts light outside the system band: [Huang2014] uses wavelengths away from the 1550 nm system band. Effectiveness is moderate because the signal is faint, much weaker than the local oscillator. |
CM-0088 Pattern Recognition Using Machine Learning |
| Description | Category | Impact Type |
|---|---|---|
|
The physical quantum and optical layer is the primary target. The attack acts on a passive optical component, the receiver's fused biconical taper beam splitter, whose transmission varies periodically with wavelength changes [Huang2013], and it works by putting light into that component at a wavelength the design never anticipated. |
L1 — Physical Layer (Quantum & Optical) | Primary Target |
|
The electronics and analog layer is a secondary effect. The attack's second part works by arranging the balanced detector's differential photocurrents so that they cancel, which makes the shot-noise reading come out normal at both of the receiver's attenuation settings [Huang2014]. The deception is executed optically but it lands on the analog detection electronics. |
L2 — Electronics & Analog Components | Secondary Effect |
|
The post-processing and cryptographic layer is where the damage is realised. A falsified shot-noise unit propagates into the excess-noise estimate and from there into the privacy-amplification ratio, so the parties compress by the wrong amount and accept a key the attacker holds. [Huang2014] locates the failure in exactly that estimator and reports that she can tap all of the secure key information without being discovered. |
L5 — Post-Processing & Cryptographic | Secondary Effect |
| Relation type | Description | Technique |
|---|---|---|
| Combines with |
Scoped to the homodyne Strategy B branch only. When the receiver monitors local-oscillator intensity and also checks its linear relation to the shot noise, the wavelength manipulation is not sufficient on its own and [Huang2014] composes it with the local-oscillator calibration attack of [Jouguet2013], which manipulates the detector clock trigger timing. A-0099 Manipulation of Vacuum-Noise Estimation is that attack in the catalog: its own literature holds both [Jouguet2013] and [Huang2014], and its description is local-oscillator parameter manipulation to mask excess noise. Strategy A, which applies when the receiver does not monitor local-oscillator intensity, needs no such composition. |
A-0099 Manipulation of Vacuum-Noise Estimation |
| Relation type | Description | Technique |
|---|---|---|
| Alternative to |
A-0097 is the continuous-variable instance of this same mechanism and says so itself, describing the attack as similar to its counterpart in discrete-variable QKD systems. The two rows share the class vulnerability V-0029 and the beamsplitter carrier HW-0019. Li2011's own class claim covers the shared ground, proposing an attack "which can be applied to all practical QKD systems with the passive state modulation". Neither derives from the other: they are peers selected by which receiver the attacker faces, and neither has a parent [Li2011]. |
A-0096 Wavelength-Dependent Beamsplitter Attack |
|