We develop secure threshold protocols for two important operations in lattice cryptography, namely, generating a hard lattice Λ together with a “strong ” trapdoor, and sampling from a discrete Gaussian distribution over a desired coset of Λ using the trapdoor. These are the central operations of many cryptographic schemes: for example, they are exactly the key-generation and signing operations (respectively) for the GPV signature scheme, and they are the public parameter generation and private key extraction operations (respectively) for the GPV IBE. We also provide a protocol for trapdoor delegation, which is used in lattice-based hierarchical IBE schemes. Our work therefore directly transfers all these systems to the threshold setting. Our protocols provide information-theoretic (i.e., statistical) security against adaptive corruptions in the UC framework, and they are private and robust against an optimal number of semi-honest or malicious parties. Our Gaussian sampling protocol is both noninteractive and efficient, assuming either a trusted setup phase (e.g., performed as part of key generation) or a sufficient amount of interactive but offline precomputation, which can be performed before the inputs to the sampling phase are known.
We develop secure threshold protocols for two important operations in lattice cryptography, namely, generating a hard lattice Λ together with a “strong ” trapdoor, and sampling from a discrete Gaussian distribution over a desired coset of Λ using the trapdoor. These are the central operations of many cryptographic schemes: for example, they are exactly the key-generation and signing operations (respectively) for the GPV signature scheme, and they are the public parameter generation and private key extraction operations (respectively) for the GPV IBE. We also provide a protocol for trapdoor delegation, which is used in lattice-based hierarchical IBE schemes. Our work therefore directly transfers all these systems to the threshold setting. Our protocols provide information-theoretic (i.e., statistical) security against adaptive corruptions in the UC framework, and they are private and robust against an optimal number of semi-honest or malicious parties. Our Gaussian sampling protocol is both noninteractive and efficient, assuming either a trusted setup phase (e.g., performed as part of key generation) or a sufficient amount of interactive but offline precomputation, which can be performed before the inputs to the sampling phase are known.
Threshold signature schemes enable distribution of the signature issuing capability to multiple users, to mitigate the threat of signing key compromise. Though a classic primitive, these signatures have witnessed a surge of interest in recent times due to relevance to modern applications like blockchains and cryptocurrencies. In this work, we study round-optimal threshold signatures in the post-quantum regime and improve the only known lattice-based construction by Boneh et al. [CRYPTO'18] as follows: - Efficiency. We reduce the amount of noise flooding used in the construction from 2^Ω(λ) down to √Q, where Q is the bound on the number of generated signatures and λ is the security parameter. By using lattice hardness assumptions over polynomial rings, this allows to decrease the signature bit-lengths from Õ(λ³) to Õ(λ), bringing them significantly closer to practice. Our improvement relies on a careful analysis using Rényi divergence rather than statistical distance in the security proof. - Instantiation. The construction of Boneh et al. requires a standard signature scheme to be evaluated homomorphically. To instantiate this, we provide a homomorphism-friendly variant of Lyubashevsky’s signature [EUROCRYPT '12] which achieves low circuit depth by being "rejection-free" and uses an optimal, moderate noise flooding of √Q, matching the above. - Towards Adaptive Security. The construction of Boneh et al. satisfies only selective security, where all the corrupted parties must be announced before any signing query is made. We improve this in two ways: in the Random Oracle Model, we obtain partial adaptivity where signing queries can be made before the corrupted parties are announced but the set of corrupted parties must be announced all at once. In the standard model, we obtain full adaptivity, where parties can be corrupted at any time but this construction is in a weaker pre-processing model where signers must be provided correlated randomness of length proportional to the number of signatures, in an offline preprocessing phase.
Digital signatures and key exchange protocols are the two most important public key cryptographic primitives used in the electronic transmission of data. The goal of key exchange is to preserve the secrecy of the communication, while the goal of digital signatures is to guarantee the authenticity of the exchanged messages. Constructions of digital signature schemes based on classical mathematical assumptions appeared shortly following the invention of public key cryptography in the late 1970s. And just like with key exchange, the most efficient variants are based on number-theoretic problems that are believed to be (sub)-exponentially hard for classical machines, but are solved in polynomial time by Shor’s algorithm on a powerful-enough quantum computer. Also, like for key exchange, the most efficient constructions that we believe to be quantum safe are based on the presumed hardness of lattice problems over polynomial rings. One interesting difference between key exchange and digital signatures is that key exchange appears to inherently require that some mathematical problem be computationally hard. Digital signatures, on the other hand, can be generically constructed from any one-way function [1,2]. So even though they certainly fall into the category of public key primitives based on their usage, their existence requires much weaker assumptions. Additionally, the transformation from a one-way function to a digital signature is not too inefficient. For example, the total parameter size (public key + signature) of the SPHINCS+ scheme [3] is around 40 kB. While these sizes are larger, and signing times considerably longer, than those of signatures based on factoring or discrete log, it is still a usable scheme for many applications. And being only based on symmetric assumptions (e.g. one wayness and collision resistance of cryptographic hash functions), its security is very attractive. In order to be considered an interesting alternative to the above-mentioned signature, a scheme based on a mathematical assumption would need to have significant performance advantages. Schemes based on factoring and the discrete logarithm problem were significantly shorter and faster, and so the generic approach lay dormant for over four decades. The new quantum-safe schemes will need to have similar performance advantages if they are to be used in lieu of this safe approach. Below, we describe two techniques for constructing lattice-based digital signatures with output sizes being just a few kilobytes. On a very high level, lattice-based signature constructions follow the two known approaches for constructing classical signatures. In the first approach, the signer outputs a function f and an image y = f(x) as his public key and keeps x as his secret key. To sign a message μ, he gives a non-interactive zero-knowledge proof that he knows an x satisfying y = f(x), using the message μ to create the ‘challenge’ H(μ) for the proof (where H is a public function that maps μ to something ‘random looking’). If the function f is one way then the verifier should be convinced that the proof could have only been created by the entity who knows x. A classic example of this type of scheme is the Schnorr signature scheme [4] based on the hardness of the discrete logarithm problem. The second approach is to create a function f together with a trapdoor f−1, output f as the public key and keep f−1 as the secret key. A message μ is signed by using the secret trapdoor to create a pre-image x such that f(x) = H(μ). Again, if the function f is one way then only someone in possession of a trapdoor should be able to invert it. An example of such a construction based on a ‘factoring-like’ assumption is the RSA signature scheme [5]. The high-level ideas for lattice-based signatures follow the above blueprints, but the technical details are significantly more involved. The main reason for the complications is the different algebraic structure of the hard one-way function underlying lattice cryptography. While the domain of the function in discrete log and RSA-based one-way functions are groups, the domains in lattice-based signatures are sets that are not closed under any operation—in particular, they are elements in a group that have small norms. This crucial small norm requirement precludes us from using uniformly random masking as in Schnorr signatures or having a trapdoor for a bijective one-way function as in RSA signatures. These barriers have, nevertheless, been overcome and the resulting digital signatures are quite practical. In 2017, the US National Institute of Standards and Technology (NIST) began a ‘competition’ for a quantum-safe key exchange and digital signatures standard. At the time of this writing, this process is in the third round and there are two lattice-based signatures remaining—each following one of the above high-level designs. The CRYSTALS-Dilithium [6] scheme follows the Schnorr framework, but adds a crucial rejection-sampling step to keep the size of the coefficients small. The FALCON scheme [7] utilizes a randomized trapdoor sampling technique that uses a secret trapdoor for f−1 to produce random pre-images from a particular distribution. Because there is no longer a bijection, it is crucial to also have the property that the distribution of the outputted pre-images does not leak information about the trapdoor. Both schemes are relatively fast and their parameters (public key + signature size) are the shortest of all quantum-safe signature schemes. While both schemes are based on lattices, they have rather different characteristics. FALCON has very short parameters (see Table 1), but entails a rather complicated procedure for signature generation. In particular, it uses (an optimized version of) the GPV sampler [8], which requires floating-point arithmetic with approximately 64 bits of precision. Requiring such high precision means that subtle implementation errors may not get detected even with rigorous testing. Dilithium, on the other hand, has larger parameters but a very simple implementation where all the sampling in the signing is done in a power-of-2 range; it is thus much less prone to implementation errors. It is quite possible that in the future both schemes will be used for different applications. Approximate parameter sizes (in bytes) for the CRYSTALS-Dilithium and FALCON digital signature schemes at approximately 128-bit security levels. Approximate parameter sizes (in bytes) for the CRYSTALS-Dilithium and FALCON digital signature schemes at approximately 128-bit security levels. Conflict of interest statement. None declared.
A threshold signature scheme distributes the ability to generate signatures through distributed key generation and signing protocols. A threshold signature scheme should be functionally interchangeable, meaning that a signature produced by a threshold scheme should be verifiable by the same algorithm used for non-threshold signatures. To resist future attacks from quantum adversaries, lattice-based threshold signatures are desirable. However, the performance of existing lattice-based threshold signing protocols is still far from practical. This paper presents the first lattice-basedt-out-of-nthreshold signature scheme with functional interchangeability that has been implemented. To build ant-out-of-naccess structure for arbitraryt≤n, we first present a novelt-out-of-nversion of the SPDZ MPC protocol. For high concrete efficiency, we avoid using the MPC protocol to evaluate hash operations. Moreover, we design an efficient distributed rejection sampling protocol. As a consequence, the online phase of our distributed signing protocol takes only 0.5 seconds in the two-party setting and 7.3 seconds in the 12-party setting according to our implementation. As a byproduct, our scheme also presents a periodic key refreshment mechanism and offers proactive security.
We show how to construct a variety of "trapdoor" cryptographic tools assuming the worst-case hardness of standard lattice problems (such as approximating the length of the shortest nonzero vector to within certain polynomial factors). Our contributions include a new notion of trapdoor function with preimage sampling, simple and efficient "hash-and-sign" digital signature schemes, and identity-based encryption. A core technical component of our constructions is an efficient algorithm that, given a basis of an arbitrary lattice, samples lattice points from a discrete Gaussian probability distribution whose standard deviation is essentially the length of the longest Gram-Schmidt vector of the basis. A crucial security property is that the output distribution of the algorithm is oblivious to the particular geometry of the given basis.
The study of digital signatures with tight reductions has attracted considerable attention over the past two decades, as such schemes inherit essentially the same quantitative hardness as the underlying computational problem. In the context of lattice-based cryptography, the GPV approach (Gentry, Peikert, and Vaikuntanathan, STOC ’08) admits a simple tight reduction from the SIS problem, but relies on preimage sampling via trapdoors, which often leads to complex and brittle implementations. By contrast, tight proofs for trapdoor-free constructions, following the Fiat–Shamir paradigm, necessarily rely on decisional assumptions, most notably the decisional LWE assumption. From the perspective of concrete security, however, reliance on a search assumption is preferable, as known search-to-decision reductions for LWE provide only weak quantitative guarantees. \ind This paper presents the first efficient lattice-based signature scheme with a tight reduction to a search assumption (namely, the hardness of the search LWE problem) that does not require any trapdoor in the scheme itself (but merely uses, instead, a trapdoor in the proof). Our construction follows the Fiat-Shamir paradigm and can be viewed as a lattice analogue of Chevallier–Mames signatures (CRYPTO ’05). Establishing security in the lattice setting, however, requires overcoming significant technical obstacles: in particular, our proof develops several new techniques to cope with the inherently weak soundness guarantees of lattice-based interactive proofs.
We present Olingo, a framework for threshold lattice signatures that is the first to offer all desired properties for real-world implementations of quantum-secure threshold signatures: small keys and signatures, low communication and round complexity, non-interactive online signing, distributed key generation (DKG), and identifiable abort. Our starting point is the framework by Gur, Katz, and Silde (PQCrypto 2024). We change the underlying signature scheme to Raccoon (Katsumata et al, Crypto 2024), remove the trapdoor commitments, extend the scheme from two to three rounds, and then apply numerous improvements and optimizations to achieve all the above properties. We provide detailed proofs of security for our new framework and present concrete parameters and benchmarks. At the 128-bit security level, for up to 1024 parties and supporting 2> 60 signatures, our scheme has 4.4 KB public keys and 11.3 KB signatures; while signing requires communication of 852 KB per party using the LaBRADOR proof system (Beullens and Seiler, Crypto 2023). An optimistic non-interactive version of our scheme requires only 76 KB communication per party.