Compromising Random Linear Network Coding as A Cipher

Due to its potential improvement on network throughput, network coding has attracted considerable research interests. Random Linear Network Coding (RLNC), a branch of network coding, is considered as a cipher to protect the confidentiality of Internet packets due to its features such as packet mixing in a decentralized approach. In this paper, we propose attacks to compromise RLNC as a cipher for confidentiality protection. The attacks are based on the blind source separation (BSS) technique, a statistical signal processing technique designed to recover original signals based on mixtures of original signals. The attacks include the scaling step to filter out artifacts generated by BSS algorithms and the cross-checking step to increase confidence on packets recovered by the proposed attacks. Our extensive experiments on packets collected from the Internet and a campus network show that the attacks can successfully recover about 1% of original packets.