Pulse-Intensity Increase in Two-Way QKD
Parent:
In this attack on two-way quantum key distribution schemes, the threat actor (Eve) manipulates the intensity of the optical pulses within the system by intercepting and replacing the pulses the receiver (Bob) sends to the transmitter (Alice) with ones of higher intensity. This strategic replacement forces Alice to encode her information on these intensified pulses, inadvertently increasing the intensity of the signals she sends back to Bob, compromising the system's security.
This step takes no key material of its own. What it leaves behind is a source emitting above the level its security proof assumes, while the watchdog that should have reported the change stays silent. A separate follow-on attack, photon-number splitting or unambiguous state discrimination, is what reads the key through the door this one opens [Sajeed2015].
[Sajeed2015] models that the same manipulation also breaks quantum coin tossing on the same commercial hardware. A cheating party who can raise the mean photon number wins more often than the classical protocol allows, so the quantum version loses the advantage that was its whole point [Sajeed2015].
| [BSI2023] | BSI. "Implementation Attacks against QKD Systems". (2023) https://www.bsi.bund.de/SharedDocs/Downloads/EN/BSI/Publications/Studies/QKD-Systems/QKD-Systems.pdf. |
| [Peng2024] | Q. Peng, B. Gao, K. Zaitsev, D. Wang, J. Ding, Y. Liu, Q. Liao, Y. Guo, A. Huang, and J. Wu. "Security boundaries of an optical power limiter for protecting quantum key distribution systems" (2024) 10.1103/PhysRevApplied.21.014026. arXiv:2303.12355. |
| [Qi2021] | R. Qi, H. Zhang, J. Gao, L. Yin, and G.-L. Long. "Loophole-free plug-and-play quantum key distribution" (2021) 10.1088/1367-2630/ac0478. arXiv:2010.05503. |
| [Sajeed2015] | S. Sajeed, I. Radchenko, S. Kaiser, J. Bourgoin, A. Pappa, L. Monat, M. Legr, and V. Makarov. "Attacks exploiting deviation of mean photon number in quantum key distribution and coin tossing" In: Phys. Rev. A 91, 032326. (2015) 10.1103/PhysRevA.91.032326. |
| [Sushchev2025] | Sushchev I. S., Bugai K. E., Molotkov S. N., Bulavkin D. S., Sidelnikova A. S., Melkonian D. M., Vakhrusheva V. M., Lokhmatov R. Yu, Dvoretskiy D. A.. "Realistic vulnerabilities of decoy-state quantum key distribution" (2025) 10.48550/arXiv.2507.15446. |
| [Zhao2008a] | Y. Zhao, B. Qi, and H. Lo. "Quantum key distribution with an unknown and un-trusted source" In: Phys. Rev. A 77, 052327. (2008) 10.1103/PhysRevA.77.052327. |
| [Zhao2010] | Y. Zhao, B. Qi, H. Lo, and L. Qian. "Security analysis of an untrusted source for quantum key distribution: passive approach" In: New Journal of Physics 12, 023024.. (2010) 10.1088/1367-2630/12/2/023024. |
1596fa55-9a58-4a29-b119-38166da00454
- Quantum Communication
Technique properties
Operational properties
| Description | Countermeasure |
|---|---|
|
Watchdog pulse-energy monitoring at Alice's input is the correct control against this technique in principle, and the only reason the attack is interesting is that the first shipped implementation of it failed. [Sajeed2015] is explicit that although the implemented strategy of the pulse-energy-monitoring module is generally correct, the technical realization should be revised dramatically, and it gives the specification: pull the OPA380 output down through a 2 kOhm resistor to the -5 V supply to prevent negative saturation, raise the front-end bandwidth to at least 9.5 MHz so the second pulse can be metered to 10 percent precision, rebuild the integrator so the capacitor charges by a current depending linearly on the input voltage instead of acting as a peak detector, and replace the edge-triggered monostable with simple level triggering. |
CM-0011 Watchdog Detector |
|
Folding the untrusted source into the security proof is the procedural answer to this attack: it does not stop the threat actor raising the mean photon number, it stops the proof depending on that number being what Alice believes. [Zhao2008a] proves security for standard BB84, weak-plus-vacuum decoy state and one-decoy state in the case where the source is unknown and untrusted, with rigorous lower bounds on the key generation rate and a simulated rate close to the trusted-source case. [Zhao2010] makes it practical: a beam splitter passively splits each input pulse and a cross-estimate technique characterises the source, so no fast optical switching and no quantum random number generator are needed. |
CM-0014 Full Device Imperfections Incorporation in Scientific Security Proof |
| From parent: Laser Seeding Attack (LSA) | |
|
A true single-photon source removes this attack surface by construction rather than reducing it, because both of the attack's mechanisms depend on a semiconductor gain medium that a single-photon emitter does not have. The intensity effect works by injected photons triggering stimulated emission, which amplifies the emitted pulse by a controllable factor [Huang2019]. The phase effect works by injected photons outnumbering the spontaneously emitted seed population that would otherwise set the pulse phase at random [Sun2015a]. Neither has an analogue in a deterministic single-photon emitter. The caveat is scope rather than physics: no cited source tests, models or recommends a single-photon source against laser seeding, so this link rests on the mechanism rather than on a stated result, and a true single-photon source is not a drop-in replacement for the weak coherent sources that real deployments use, which is precisely why [Sun2015a] scopes the attack to most DV-QKD. |
CM-0002 True Single Photon Source |
|
Real-time source-intensity monitoring inside the transmitter, which is the countermeasure [Zheng2019] proposes and analyses for the continuous-variable case. A beam splitter taps a fraction of the undiminished local-oscillator signal before attenuation, a photodiode measures it, and the seeding gain is recovered as the ratio of measured to preset intensity; the key rate is then evaluated at the true parameters rather than the assumed ones. The local oscillator's own loss is compensated by adjusting the variable optical attenuator in its path. In CV-MDI-QKD the loophole is closed only if both light sources are monitored simultaneously, and in a local-oscillator-at-receiver system the transmitter monitors the undiminished reference signal instead [Zheng2019]. |
CM-0016 State Preparation Monitoring |
|
Every quantitative defence claim for this technique is an isolation budget in dB, and that budget silently assumes the isolator still meets its specification. [Ponosova2025] shows it may not: out-of-band pulsed light permanently degrades an isolator's reverse isolation while forward transmission is less affected, so the link may keep running. [Huang2019] independently notes the isolator's isolation is temperature-dependent and can be reduced by heating it. Periodically re-verifying that protective components still perform to specification is therefore the control that keeps the rest of the budget honest. |
CM-0012 Continuous Device Functionality Monitoring |
|
Optical isolation is the defence of record for this technique and every quantitative countermeasure claim about it is expressed as an isolation budget, so its limits matter more here than anywhere else. Finite isolation only reduces, never eliminates, the injected photons: commercial isolators reach 50 dB and can be placed in series, but the threat actor (Eve) can always increase the power of her control laser [Sun2015a], and even behind a 25 dB isolator a 0.6 mW control laser still imposes a non-random phase [Sun2015a]. Quantified from the attack side, with 10 W injected an effective isolation above 80 dB holds the intensity gain below 2 on one tested diode, but any gain above 1 still produces key-rate overestimation unless the security analysis itself is modified [Huang2019]. Isolation is also not a fixed quantity in service. It is temperature-dependent, because the isolation comes from magneto-optic polarisation rotation and the magnets used vary with temperature, so raising the temperature reduces it [Huang2019]; and it can be permanently degraded by out-of-band light, while forward transmission largely survives [Ponosova2025]. |
CM-0008 Employment of Optical Isolation |
|
A watchdog detector at the transmitter's input can in principle see the threat actor's (Eve's) injected light, but both primary sources describe ways around it and neither endorses it as sufficient. It can be bypassed by pulses at a high repetition rate, because the monitor's bandwidth is limited [Huang2019]. It can be evaded by pulsing the control laser, which lowers the average power the monitor sees [Sun2015a]. And it can be removed outright, since a classical monitor is damageable by bright light and is reported to be the first component in the transmitter that a laser damage attack destroys [Huang2019]; [Sun2015a] gives damage limits for real monitor parts. [Lovic2023] adds that certain detectors can be blinded by bright light, and argues on that basis that passive countermeasures are preferable to active ones, because an active watchdog adds attack surface of its own. |
CM-0011 Watchdog Detector |
|
The only mitigation the primary source itself endorses. Rather than trying to keep the light out, bound how much the threat actor (Eve) can inject, derive from that an upper bound on the intensity multiplier, then minimise the key rate over the interval from to so the worst case is covered [Huang2019]. The same source explains why no optical control substitutes for this: any above 1 still produces overestimation unless the security analysis itself is modified [Huang2019]. [Sun2015a] reaches the same layer from the phase side, recommending a protocol or proof valid for unrandomised phase. [Wiesemann2024] supplies the measurement that makes the bound concrete, a source-agnostic heterodyne method characterising the phase de-randomisation on a real transmitter and yielding a lower bound on the isolation that transmitter requires. |
CM-0014 Full Device Imperfections Incorporation in Scientific Security Proof |
|
Active phase randomisation closes the phase axis and leaves the intensity axis fully open, so it is a genuine but explicitly partial control. [Sun2015a] states the position precisely: it guarantees the phase-randomisation assumption and partially reduces the risk, but does not remove the attack, because the threat actor (Eve) can still tamper with the intensity and the pulse shape, and because bright light destabilises the diode's intensity. The same passage notes it is also an incomplete defence for a continuous-wave-laser-plus-external-modulator architecture. Since the axis it leaves open is the one that breaks the decoy-state estimator [Huang2019], it must not be read as a fix. |
CM-0015 Active Phase Randomisation |
| Description | Category | Impact Type |
|---|---|---|
|
Physical Layer (Quantum & Optical) is the primary target: Eve intercepts Bob's bright pulses on the fibre and substitutes higher-intensity ones, so Alice's reflected signal carries a mean photon number above the calibrated single-photon level [Sajeed2015]. |
L1 — Physical Layer (Quantum & Optical) | Primary Target |
|
Electronics & Analog Components are a secondary target: the three exploited flaws live in the monitoring detector's analog circuit (front-end amplifier bandwidth and negative saturation, edge-triggered alarm), which Eve defeats to evade the energy alarm [Sajeed2015]. |
L2 — Electronics & Analog Components | Secondary Effect |
|
Application & Protocol Layer is a knock-on effect: the inflated, unmonitored mean photon number breaks the weak-coherent-source assumption of the BB84 and SARG04 security proofs, enabling PNS/USD information leakage [Sajeed2015]. |
L6 — Application & Protocol Layer | Secondary Effect |
| From parent: Laser Seeding Attack (LSA) | ||
|
Post-Processing and Cryptographic layer, as a secondary effect. The existing L6 mapping covers the protocol-level security guarantee, but the arithmetic that actually produces the wrong answer runs one layer lower: the decoy-state estimator relates observed gains and error rates to the single-photon yield through the photon-number probabilities of the intensity the transmitter believes it set, and that estimation is performed in post-processing on publicly announced data [Huang2019]. The continuous-variable analogue is excess-noise estimation, corrupted the same way [Zheng2019]. Nothing in the post-processing code is attacked; it is fed corrupted parameters and computes faithfully from them, which is why the impact type is secondary. |
L5 — Post-Processing & Cryptographic | Secondary Effect |
|
Physical Layer (Quantum & Optical) is the primary target: Eve injects continuous-wave light through the fibre into Alice's semiconductor laser diode, driving stimulated emission that raises output intensity and alters the optical pulse (phase, tail) [Huang2019]. |
L1 — Physical Layer (Quantum & Optical) | Primary Target |
|
Application & Protocol Layer is a secondary (knock-on) effect: the unnoticed intensity increase and phase fixing break the trusted-source assumptions of the decoy-state BB84 and MDI-QKD security proofs, undermining the protocol-level key-security guarantee [Huang2019]. |
L6 — Application & Protocol Layer | Secondary Effect |
| Relation type | Description | Technique |
|---|---|---|
| Precedes |
A-0064 opens the door and A-0044 walks through it. Raising the mean photon number leaving Alice is what makes the multi-photon components large enough to split: the primary's strong model is explicitly the combined photon-number-splitting and cloning attack already assumed by Clavis2, but with the mean photon number manipulated and increased by a factor x so that the multi-photon components can no longer be ignored [Sajeed2015]. |
A-0044 Photon-Number-Splitting (PNS) Attack |
| Precedes |
The realistic follow-on that Sajeed2015 models: intercept-and-resend with unambiguous state discrimination, which the raised mean photon number makes effective [Sajeed2015]. |
A-0045 Intercept-and-Resend Attack |
| From parent: Laser Seeding Attack (LSA) | ||
| Combines with |
[Huang2019] section V states the two can be run at the same time: the threat actor (Eve) reads back-reflected light for the transmitter's modulator settings while her injected light modifies the source. They are distinct mechanisms and the same source is careful to separate them, noting that unlike a Trojan-horse attack, which analyses light reflected back out, laser seeding manipulates the functioning of the transmitter's laser diode directly. The concurrency is what makes the pair worth recording, since one attack reads a setting and the other changes the state. |
A-0049 Quantum Trojan-Horse Attack (THA) |
| Combines with |
[Huang2019] section V names the injection-locking attack as a concurrent partner, in which the seeding is affected in real time by the transmitter's modulator state. The two sit in the same cat08 family and share physics: [Lovic2023] models laser seeding itself as optical injection locking using laser rate equations, reproducing the power increase, reduced turn-on delay, reduced chirp and jitter, and wavelength shift together. A-0066 is nonetheless a distinct technique, aimed at inferring the transmitter's polarisation from the frequency of Alice's emitted light rather than at raising intensity or fixing phase. |
A-0066 Injection-Locking Attack (ILA) |
| Precedes |
Laser seeding creates the condition that A-0055 Non-Random Phase Attack exploits. [Sun2015a] shows the emitted pulses' phase distribution collapsing from uniform to Gaussian under injected control light, and A-0055 is the catalogued technique that turns a non-random global phase into distinguishability between signal and decoy states. This is the DV half of the door-opener chain that outcome_class door_opened and produces_resource side_channel_trace encode. |
A-0055 Non-Random Phase Attack |
| Relation type | Description | Technique |
|---|---|---|
| From parent: Laser Seeding Attack (LSA) | ||
| Preceded by |
A-0053 is also the step before a laser-seeding attack. Two 70 dB isolators reduce a 40 dBm injection to -100 dBm at Alice's source, which is negligible, but the same pair degraded to 10.9 dB each leaves 21.8 dB combined, at which point seeding is feasible again at the 0.6 mW control level [Tan2022]. |
A-0053 External Magnetic Field Attack on Optical Isolators and Circulators |
|