Frequency Side Channel Attack in Twin-Field QKD
The Frequency Side Channel Attack exploits imperfections in intensity modulation within Twin-Field Quantum Key Distribution (TF-QKD) systems, leading to frequency-based distinguishability between signal and decoy states. Each intensity setting is produced with its own drive voltage, so each pulse leaves the transmitter carrying a slightly different frequency shift, and a threat actor (Eve) who measures a pulse's spectrum can tell which setting was sent. Once the legitimate parties publicly announce which rounds were signal rounds, those measurements hand her the corresponding raw-key bits [Lu2021].
The attack is passive: the threat actor creates and alters no quantum state, because the spectral difference is put there by the transmitter's own modulator before she is involved. That is what separates it from the two earlier frequency-domain attacks in this catalog, which must shift pulse timing actively and so reach only two-way plug-and-play systems, while this one works against a one-way source. It introduces no errors, so nothing in the legitimate parties' statistics reveals it, and it yields partial rather than complete key material [Lu2021].
| [ETSI016PP] | ETSI. "Quantum Key Distribution (QKD); Common Criteria Protection Profile - Pair of Prepare and Measure Quantum Key Distribution Modules". In: ETSI GS QKD 016. (2024) https://www.etsi.org/deliver/etsi_gs/QKD/001_099/016/02.01.01_60/gs_QKD016v020101p.pdf. |
| [Huang2018] | A. Huang, S. Sun, Z. Liu, and V. Makarov. "Quantum key distribution with distinguishable decoy states" In: Phys. Rev. A 98, 01233. (2018) 10.1103/PhysRevA.98.012330. |
| [ISOIEC23837P1] | ISO/IEC JTC 1/SC 27. "ISO/IEC DIS 23837-1:2022 Information security - Security requirements, test and evaluation methods for quantum key distribution - Part 1: Requirements". (2022) Draft International Standard (DIS) 2022 of ISO/IEC 23837-1. The published standard is ISOIEC23837 (ISO/IEC 23837-1:2023); clause numbers and wording can differ.. |
| [Jiang2012] | M. Jiang, S. Sun, C. Li, and L. Liang. "Wavelength-selected photon-number-splitting attack against plug-and-play quantum key distribution systems with decoy states" In: Phys. Rev. A 86, 032310. (2012) 10.1103/PhysRevA.86.032310. |
| [Jiang2014] | M. Jiang, S. Sun, C. Li, and L. Liang. "Frequency shift attack on plug-and-play quantum key distribution systems" In: Journal of Modern Optics 61, 147. (2014) 10.1080/09500340.2013.872309. |
| [Jiang2025] | X.-L. Jiang, Y.-F. Lu, Y. Wang, C. Zhou, H.-W. Li, and W.-S. Bao. "Effect of frequency side channels on the security of practical quantum key distribution" (2025) 10.1364/OE.579447. |
| [Li2025] | Yu-Huai Li, Ting Zeng, Min-Yan Wang, Cong Jiang, Jin Lin, Hao-Bin Fu, Xin-Yang Zheng, Jiu-Peng Chen, Zeng-Sen Lin, Cheng-Lin Li, Jian-Yu Guan, Yang Li, Qi Shen, Hao Li, Lixing You, Zhen Wang, Fei Zhou, Juan Yin, Sheng-Kai Liao, Ji-Gang Ren, Xiang-Bin Wang, Yuan Cao, Qiang Zhang, Cheng-Zhi Peng, Jian-Wei Pan. "Free-Space Twin-Field Quantum Key Distribution" (2026) 10.1038/s41566-026-01910-6. arXiv:2503.17744. |
| [Lu2021] | Y.-F. Lu, M.-S. Jiang, Y. Wang, X.-X. Zhang, F. Liu, C. Zhou, H.-W. Li, S.-B. Tang, J.-Y. Wang, W.-S. Bao. "Practical analysis of sending or not-sending twin-field quantum key distribution with frequency side channels" In: Appl. Sci. 11(20), 9560. (2021) 10.3390/app11209560. |
fb18db87-e9fa-43e3-a361-518e4fa1e8e4
- Quantum Communication
Technique properties
Operational properties
| Description | Countermeasure |
|---|---|
|
Incorporating the frequency side channel into the security proof is the second of the three mitigation routes [Lu2021] names, described there as developing the mathematical models in theory to include the side channels, for example the loss-tolerant method combined with a characterisation of the real apparatus. The route fits because this control's own scope already covers the relevant flaw class: its text lists emission of multimode signals, that is signals inadvertently encoding information in undesired modes, as a device imperfection the proof must absorb. A proof that models the per-intensity spectral difference removes the threat actor's advantage rather than detecting her, since the leak is accounted for in the privacy-amplification ratio instead of being assumed absent [Lu2021]. |
CM-0014 Full Device Imperfections Incorporation in Scientific Security Proof |
|
The measurement half of the second mitigation route [Lu2021] names, which it gives as the loss-tolerant method together with the characterisation of the real apparatuses. CM-0086 is that characterisation: pre-deployment study of each security-critical component under wavelength and intensity sweeps, which its own text calls a prerequisite for any loss-tolerant or imperfection-aware proof. Sweeping the modulator across its intensity settings is what surfaces the per-setting frequency shift, so this control finds the weakness that CM-0014's proof route then prices into the key rate. It is the pre-deployment counterpart of CM-0016, which watches the same parameter at run time. |
CM-0086 Prior Hardware Component Study |
|
An operator can run this control on a deployed system. CM-0016 checks that each emitted signal's time, wavelength and polarisation meet expectation, and wavelength is exactly the parameter that deviates here, since each intensity setting leaves the transmitter with its own spectrum [Lu2021]. The measurement such a monitor makes is the one the source itself performed, reading the modulator output on an optical-spectrum analyser before the fixed attenuation [Lu2021], and the drive amplitude would fall under [ETSI016PP]'s general definition of Calibration Data, a user-adjustable and verifiable physical parameter. |
CM-0016 State Preparation Monitoring |
|
The three-modulator source arrangement is the first of the three mitigation routes [Lu2021] names, and the paper describes the mechanism precisely: in practical sending-or-not-sending systems the first modulator carves the continuous light into five pulses of different intensities, the second sets the designed intensity ratio, and the last modulates the signal pulse width to a fixed width, which eliminates the side channels caused by the first two. This is the configuration the attack's own precondition excludes: [Lu2021] deliberately modelled a simplified source with a single intensity modulator, applying only one modulator to the signal pulses. |
CM-0064 Triple Intensity Modulation for Side Channel Elimination |
| Description | Category | Impact Type |
|---|---|---|
|
Physical Layer, Quantum and Optical, as the primary target. The leak is in the emitted optical field, a per-intensity frequency shift riding on the pulses, and the threat actor reads it optically with a wavelength-division multiplexer and three single-photon detectors at the transmitters' output ports [Lu2021]. |
L1 — Physical Layer (Quantum & Optical) | Primary Target |
|
Electronics and Analog Components, as a secondary effect. The frequency shift is produced by the analog drive waveform: [Lu2021] shows pure intensity modulation holds only while the drive voltage is constant, and that the amplitude difference between the signal, decoy and vacuum drive voltages is what separates their spectra. The leak originates in the drive electronics even though it is read optically [Lu2021]. |
L2 — Electronics & Analog Components | Secondary Effect |
|
Post-Processing and Cryptographic, as a secondary effect, and the layer that distinguishes this technique from both siblings. The attack does not merely leak at L1, it corrupts parameter estimation: [Lu2021] states it is impossible to pick out the decoy states that have undergone the same operation as the signal states, so the decoy-state method does not work properly, and [ETSI016PP] makes parameter estimation the input from which the privacy-amplification ratio is deduced. the draft ISO/IEC DIS 23837-1 [ISOIEC23837P1] clause 6.5 c) places parameter estimation inside post-processing. |
L5 — Post-Processing & Cryptographic | Secondary Effect |
| Relation type | Description | Technique |
|---|---|---|
| Alternative to |
A-0057 Frequency-Shift is the closest sibling and the two are alternatives rather than steps in a chain. Both turn a modulator into a frequency modulator and read the shift, but the route differs on every axis that matters: [Jiang2014] has the threat actor time-shift the pulse onto the phase modulator's rising edge, which is possible only in a two-way plug-and-play system, while [Lu2021] needs no time shift at all because the intensity modulator's drive amplitude already separates the settings, so this technique reaches one-way sources. The defences do not transfer either, and [Jiang2014] names the one that closes its own attack: if the users close the loophole by monitoring the arrival times of the signal and reference pulses at the transmitter, both the phase-remapping attack and the frequency-shift attack are invalid. That monitor has nothing to catch here, because no arrival time is altered. The two also differ in yield, A-0057 claiming full information where [Lu2021] claims partial raw-key bits, and in how the stealth is obtained. A-0057's route is intercept-and-resend, so its threat actor re-encodes the receiver's pulse from what she read, and she cannot read perfectly: [Jiang2014] concedes she can never distinguish the four signal states without error because their spectra always overlap, so a non-zero error reaches the legitimate parties, priced at 0.7 percent over 50 km with a realistic receiver and below 0.1 percent only with a near-ideal one, and bought in both cases by tuning thresholds down at the cost of her own detection probability. This technique's threat actor resends nothing. She takes a narrow spectral slice and forwards the rest, which [Lu2021] calls a loss without phase noise, so her error contribution is zero by construction rather than by tuning. |
A-0057 Frequency-Shift Attack |
| Alternative to |
A-0067 Wavelength-Selected Photon-Number-Splitting is the other alternative on this mechanism, and it is the nearer of the two because it rides the same component, the intensity modulator. The separation is the trigger and what the label is worth. [Jiang2012] depends on a time shift onto the drive edge and on intensity modulation not being performed on the signal state at all, so only the decoy states acquire a shift; [Lu2021] modulates every intensity externally, so there is no unmodulated reference frequency and the separation comes from the differing drive amplitudes instead, with no time shift and therefore one-way reach. A-0067 then needs a photon-number-splitting stage to convert its decoy label into key material, where this technique takes raw-key bits from the window click directly. |
A-0067 Wavelength-Selected Photon-Number-Splitting Attack |
|