Fast correlation attacks through reconstruction of linear polynomials

Abstract

The task of a fast correlation attack is to efficiently restore the initial content of a linear feedback shift register in a stream cipher using a detected correlation with the output sequence. We show that by modeling this problem as the problem of learning a binary linear multivariate polynomial, algorithms for polynomial reconstruction with queries can be modified through some general techniques used in fast correlation attacks. The result is a new and efficient way of performing fast correlation attacks.