The Reductive Technique Of Feedforward Network Stream Cipher

In information times,the security and reliability of information has become a growing concern. Feedforward network (also known as filter generator) as a kind of the important key stream generators, its design and analysis has been a hotspot of stream ciphers. On its security analysis, correlation analysis and linear analysis are the two most effective means of attacks. In this paper, the security issues of feedforward neural network stream cipher under different conditions are discussed, and an effective attack method is given.The Attack method presented in this paper is a known plaintext attack method, which is under the condition of the unknown system key, it solves a difficult problem which the above-mentioned two means of attacking feedforward network have in common, namely, the two means can not attack the system without the filter function.To solve the problem, an important conclusion of spectral theory is generalized and applied to the analysis of feedforward networks skillfully in this paper. And then, a method for reverting the filter function with the feedforward key sequences which are long enough is given.That is, for a feedforward network, if the two parameters, nonlinear filter function and tap set, are not given, we can still revert the two parameters only using feedforward sequences which are long enough.