Reversible photorefractive attack in continuous-variable quantum key distribution

Abstract

Quantum key distribution (QKD) offers information-theoretic security, but device imperfections introduce implementation vulnerabilities. Experimental evidence of the security consequences of transmitter vulnerabilities in continuous-variable QKD (CV-QKD) remains limited. Here, we experimentally investigate a reversible photorefractive attack in a local-local-oscillator CV-QKD system. Moderate blue-light illumination reduces the modulation variance of a lithium-niobate IQ modulator to a stable value without appreciably increasing estimated excess noise. The variance recovers after illumination is removed. We elucidate the underlying physical mechanisms and develop an attack that exploits recovery to compromise modulation calibration. Receiver measurements confirm that the resulting calibration mismatch leads to overestimated channel transmittance and underestimated excess noise. Key rate calculations based on measured parameters yield a positive estimated rate when the correctly calibrated rate is nonpositive. Recovery can also make the preceding manipulation harder to identify through later inspection. This work provides the first system-level experimental evidence that a reversible photorefractive response can be exploited to compromise modulation calibration in CV-QKD. These findings highlight reversible changes in quantum transmitters as a practical security concern, motivating security analyses and protection strategies that account for both optical perturbations and subsequent recovery.

This paper asks whether reversible photorefraction in a CV-QKD transmitter can invalidate modulation calibration without an obvious excess-noise signature. It characterizes a lithium-niobate IQ modulator under 405-nm illumination and measures the resulting parameter-estimation biases in an 800-MBaud local-local-oscillator CV-QKD system. At incident powers of $4$ and $5\,\mu\mathrm{W}$, the modulation variance stabilizes at $5.41$ and $4.78$ SNU, respectively, versus approximately $6.2$ SNU before illumination. Excess-noise estimates remain near $0.04$–$0.06$ SNU. After illumination stops, variance approaches its initial value within roughly six minutes, followed by slower recovery over hours. At $6\,\mu\mathrm{W}$, modulation remains unstable and excess noise rises to $0.22$–$0.31$ SNU. The proposed attack illuminates during calibration, then removes the light before key generation. Retaining the illuminated-state calibration gives $T_{\mathrm{est}}=kT$ and $\xi_{\mathrm{est}}=\xi/k$, where $k$ is the recovered-to-calibrated variance ratio. For $k\approx1.30$, receiver measurements give correctly calibrated parameters $(T,\xi)=(0.065,0.0538)$ versus apparent values $(0.083,0.0414)$. Standard asymptotic collective-attack calculations yield an apparent rate of $1.02\times10^{-3}$ bit/pulse under measured conditions while the correctly calibrated rate is nonpositive; this is not a demonstration of recovered secret-key information. Illumination experimentally bypasses the transmitter attenuator: the attenuator and circulator exhibit approximately $57$–$58$ dB and over $71.8$ dB loss at 405 nm, respectively. Full external-channel injection remains un demonstrated. The analysis compares stable states, does not model recovery-time parameter estimation, and leaves finite-key analysis and countermeasure evaluation to future work.