Paper 2014/1027

Simple Lattice Trapdoor Sampling from a Broad Class of Distributions

Vadim Lyubashevsky and Daniel Wichs

Abstract

At the center of many lattice-based constructions is an algorithm that samples a short vector $$, satisfying $$ where $$ are public matrices and $$ is a trapdoor. Although the algorithm crucially relies on the knowledge of the trapdoor $$ to perform this sampling efficiently, the distribution it outputs should be independent of $$ given the public values. We present a new, simple algorithm for performing this task. The main novelty of our sampler is that the distribution of $$ does not need to be Gaussian, whereas all previous works crucially used the properties of the Gaussian distribution to produce such an $$. The advantage of using a non-Gaussian distribution is that we are able to avoid the high-precision arithmetic that is inherent in Gaussian sampling over arbitrary lattices. So while the norm of our output vector $$ is on the order of $$ to $$ - times larger (the representation length, though, is only a constant factor larger) than in the samplers of Gentry, Peikert, Vaikuntanathan (STOC 2008) and Micciancio, Peikert (EUROCRYPT 2012), the sampling itself can be done very efficiently. This provides a useful time/output trade-off for devices with constrained computing power. In addition, we believe that the conceptual simplicity and generality of our algorithm may lead to it finding other applications.

Note: Fixed some typos.