Every quantum key distribution (QKD) protocol must limit the information available outside the users' laboratories. In device-independent QKD this is essentially the only assumption made on the devices. In a prepare-and-measure setting it cannot be imposed strictly, since the sender's device must emit a signal that depends on her private input. The natural replacement, and a minimal one, is a leakage bound: a bound on how much information about the input leaves the sender's laboratory, through the intended signal or through any side channel. We ask how far this unavoidable requirement can replace a detailed model of the source. We show that this single bound, quantified by the probability of guessing the input from everything the source emits, suffices to prove security against collective attacks. In our analysis, nothing is assumed about the precise form of the emitted states, their dimension, energy or purity, nor about the receiver's measurement device, beyond the standard requirement that it does not leak Bob's data. The standard B92 and BB84 protocols do not certify any key under this minimal leakage bound assumption. But closely related protocols can and, in the ideal limit, achieve one secret bit per key round. For noisy implementations, we compute key rates using semidefinite programming relaxations. Our best protocols tolerate visibilities down to 0.84, corresponding to an 8% quantum bit error rate, compared with the 11% threshold of standard device-dependent BB84.
Every quantum key distribution (QKD) protocol must limit the information available outside the users' laboratories. In device-independent QKD this is essentially the only assumption made on the devices. In a prepare-and-measure setting it cannot be imposed strictly, since the sender's device must emit a signal that depends on her private input. The natural replacement, and a minimal one, is a leakage bound: a bound on how much information about the input leaves the sender's laboratory, through the intended signal or through any side channel. We ask how far this unavoidable requirement can replace a detailed model of the source. We show that this single bound, quantified by the probability of guessing the input from everything the source emits, suffices to prove security against collective attacks. In our analysis, nothing is assumed about the precise form of the emitted states, their dimension, energy or purity, nor about the receiver's measurement device, beyond the standard requirement that it does not leak Bob's data. The standard B92 and BB84 protocols do not certify any key under this minimal leakage bound assumption. But closely related protocols can and, in the ideal limit, achieve one secret bit per key round. For noisy implementations, we compute key rates using semidefinite programming relaxations. Our best protocols tolerate visibilities down to 0.84, corresponding to an 8% quantum bit error rate, compared with the 11% threshold of standard device-dependent BB84.
Semi-device-independent quantum key distribution leaves the measurement devices uncharacterized while placing a trusted assumption on Alice's source. We formulate this source assumption operationally on Alice's four-preparation ensemble as a scalar bound on one of four physically motivated source tasks: full-label guessing, parity guessing, or their normalized composites with label exclusion. For the two-bit random-access code, we derive the exact classical frontier for each of the four source assumptions. Numerically, the BB84 strategy attains the maximal quantum deviation from all four frontiers, while the preparation-depolarized BB84 family and the direct-sum label-leakage family trace complementary branches of the arbitrary-dimensional quantum boundary for the two exclusion-assisted assumptions. Because all four task values are monotone under input-independent quantum channels, the same scalar source bound constrains every Bob--Eve extension compatible with the complete observed behavior. Using a three-setting extension that separates RAC testing from key generation, we obtain two dimension-independent security certificates over this feasible set: lower bounds on the conditional min-entropy and conditional von Neumann entropy, obtained respectively by direct optimization of Eve's key-guessing probability and by prepare-and-measure semidefinite relaxations based on the Brown--Fawzi--Fawzi variational bound. The exclusion-assisted assumptions certify positive key rates down to nearly vanishing preparation visibility, far beyond full-label or parity guessing alone. Under direct-sum label leakage, all four independently optimized rate bounds remain positive at every sampled incomplete-leakage point and vanish only at complete label revelation. These results show that robust semi-device-independent security depends not only on what Eve can identify, but also on what she can exclude.
There is a large gap between theory and practice in quantum key distribution (QKD) because real devices do not satisfy the assumptions required by the security proofs. Here, we close this gap by introducing a simple and practical measurement-device-independent-QKD type of protocol, based on the transmission of coherent light, for which we prove its security against any possible imperfection and/or side channel from the quantum communication part of the QKD devices. Our approach only requires to experimentally characterize an upper bound of one single parameter for each of the pulses sent, which describes the quality of the source. Moreover, unlike device-independent (DI) QKD, it can accommodate information leakage from the users' laboratories, which is essential to guarantee the security of QKD implementations. In this sense, its security goes beyond that provided by DI QKD, yet it delivers a secret key rate that is various orders of magnitude greater than that of DI QKD.
Abstract Most security proofs of quantum key distribution (QKD) assume that there is no unwanted information leakage about the state preparation process. However, this assumption is impossible to guarantee in practice, as QKD systems can leak information to the channel due to device imperfections or the active action of an eavesdropper. Here, we solve this pressing issue by introducing a security proof in the presence of information leakage from all state preparation settings for arguably the most popular QKD scheme, namely the decoy-state BB84 protocol. The proof requires minimal experimental characterization, as only a single parameter related to the isolation of the source needs to be determined, thus providing a clear path for bridging the gap between theory and practice. Moreover, if information about the state of the side channels is available, this can be readily incorporated into the analysis to further improve the resulting performance.
Quantum key distribution protocols have been studied for almost four decades, and have already seen real world applications. Such implementations often operate in naturally noisy environments (e.g., sensor thermal noise, sun radiation, etc.) with known or measurable noise levels. However, existing security proofs implicitly attribute any channel noise to a possible eavesdropper, an overly pessimistic assumption that can adversely effect the guaranteed protocol key-rate. In this paper, we analyze general prepare-and-measure QKD schemes, and show how to improve the bounds on the key-rate under the assumption of an independent state replacement noise at the receiver. Taking BB84 as a concrete example, our results theoretically guarantee a higher key-rate, as a function of the minimal possible replacement probability. In practice, for some parameter-regimes, this guarantees a higher tolerable QBER.
We provide an analysis of a family of device-independent quantum key distribution (QKD) protocols that has the following features. (a) The bits used for the secret key do not come from the results of the measurements on an entangled state but from the choices of settings. (b) Instead of a single security parameter (a violation of some Bell inequality) a set of them is used to estimate the level of trust in the secrecy of the key. The main advantage of these protocols is a smaller vulnerability to imperfect random number generators made possible by feature (a). We prove the security and the robustness of such protocols. We show that using our method it is possible to construct a QKD protocol which retains its security even if the source of randomness used by communicating parties is strongly biased. As a proof of principle, an explicit example of a protocol based on the Hardy's paradox is presented. Moreover, in the noiseless case, the protocol is secure in a natural way against any type of memory attack, and thus allows one to reuse the device in subsequent rounds. We also analyze the robustness of the protocol using semidefinite programming methods. Finally, we present a postprocessing method, and observe a paradoxical property that rejecting some random part of the private data can increase the key rate of the protocol.
Quantum key distribution (QKD) enables two legitimate parties to share a secret key even in the presence of an eavesdropper. The unconditional security of QKD is based on the fundamental laws of quantum physics. Original security proofs of QKD are based on a few assumptions, e.g., perfect single photon sources and perfect single-photon detectors. However, practical implementations of QKD systems do not fully comply with such assumptions due to technical limitations. The gap between theory and implementations leads to security loopholes in most QKD systems, and several attacks have been launched on sophisticated QKD systems. Particularly, the detectors have been found to be the most vulnerable part of QKD. Much effort has been put to build side-channel-free QKD systems. Solutions such as security patches and device-independent QKD have been proposed. However, the former are normally ad-hoc, and cannot close unidentified loopholes. The latter, while having the advantages of removing all assumptions on devices, is impractical to implement today. Measurement-device-independent QKD (MDI-QKD) turns out to be a promising solution to the security problem of QKD. In MDI-QKD, all security loopholes, including those yet-to-be discovered, have been removed from the detectors, the most critical part in QKD. In this thesis, we investigate issues related to the practical implementation and security of MDI-QKD. We first present a demonstration of polarization-encoding MDI-QKD. Taking finite key effect into account, we achieve a secret key rate of 0.005 bit per second (bps) over 10 km spooled telecom fiber, and a 1600-bit key is distributed. This work, together with other demonstrations, shows the practicality of MDI-QKD. Next we investigate a critical assumption of MDI-QKD: perfect state preparation. We apply the loss-tolerant QKD protocol and adapt it to MDI-QKD to quantify information leakage due to imperfect state preparation. We then present an experimental demonstration of MDI-QKD over 10 km and 40 km of spooled fiber, which for the first time considers the impact of inaccurate polarization state preparation on the secret key rate. This would not have been possible under previous security proofs, given the same amount of state preparation flaws.