Experimental Implementation of Non-Gaussian Attacks on a Continuous-Variable Quantum-Key-Distribution System

An intercept-resend attack on a continuous-variable quantum-key-distribution protocol is investigated experimentally. By varying the interception fraction, one can implement a family of attacks where the eavesdropper totally controls the channel parameters. In general, such attacks add excess noise in the channel, and may also result in non-Gaussian output distributions. We implement and characterize the measurements needed to detect these attacks, and evaluate experimentally the information rates available to the legitimate users and the eavesdropper. The results are consistent with the optimality of Gaussian attacks resulting from the security proofs.

Figure 1

Intercept-resend attack. Alice prepares a random coherent state, and Bob chooses a random quadrature measurement with a random number generator (RNG). In between, Eve makes a heterodyne measurement of each incoming quantum state, and displaces another generated coherent state according to her measurement result.

Figure 2

Experimental setup. Alice generates modulated signal pulses. Bob measures a random quadrature with a pulsed, shot-noise limited homodyne detector.

Figure 3

Variance of the noise measured by Bob that is produced by a full IR attack ($\mu =1$). We define the total added noise (referred to the input) as $\chi ={\chi }_0+ϵ$, with ${\chi }_0=1/(\eta T)-1$ denoting the loss-induced vacuum noise and $ϵ$ denoting the excess noise. The total added noise $\chi$ (stars) is measured experimentally, while ${\chi }_0$ (crosses) is deduced from the measured transmission ($T$) and from Bob’s homodyne efficiency ($\eta =0.6$). Then $ϵ$ (plus symbols) is obtained from their difference. The uncertainties margins on $ϵ$ represent the standard deviation of statistical fluctuations when computing over a finite data subset (5000 points out of 40 000).

Figure 4

Variance of the excess noise in a partial IR attack. Each point results from an average of the measured excess noise for different channel transmissions (see the $ϵ$ vs $T$ plot of Fig. 3). The solid line plots the expected excess noise due to an IR attack on a fraction $\mu$ of the pulses. Because of technical noise, the experimental data are above this line, typically less than $0.1N_0$ (dashed line).

Figure 5

Mutual information rates for a non-Gaussian partial IR attack, for $T=0.1$, 0.25, and 0.9, with $V_A=36.6N_0$, $\eta =0.6$ and a technical excess noise of $0.1N_0$. The mutual information $I_{BE}$ is plotted for a Gaussian model with equivalent excess noise (solid lines), as well as for a BS attack (dotted lines), and for a partial IR attack (dashed lines). It is compared with the Gaussian mutual information $I_{AB}$. As expected, the IR attack enables to exploit the excess noise, giving Eve extra information above the BS attack. We show statistical fluctuations ($\pm 1$ standard deviation) corresponding to a data subset of 5000 points per block, as in Fig. 3.