Correlation Attacks on Block Ciphers

This report presents a new statistical attack on iterative block ciphers called the correlation attack which is a natural generalization of linear cryptanalysis. The attack is based on nding complex-valued functions on the input and the output of a cipher which have a high correlation. Their mutual relation is then exploited to yield information about the nal round key. Introducing the notions of imbalance, I/O product, and correlation matrix, it is shown how to measure a cipher's security against the attack, and the mini-cipher IDEA(8) is found to be provably secure (assuming independency of subkeys). Links to other kinds of statistical attacks are explored. In particular, it is shown that the correlation matrix of a cipher and the matrix of di erential transition probabilities used with di erential cryptanalysis are connected by the 2-dimensional Fourier transform. This implies that correlation cryptanalysis and di erential cryptanalysis are essentially of the same strength.