Application of neural networks to unmasking chaotic secure communication

Abstract

The chaotic shift keying, which is the most robust chaotic secure communication scheme, is unmasked by using two simple single-layer neural networks (NNs) with known text attack strategy. First, spectrograms are used to transfer the time series of the transmitted signal into spectral-temporal space. Then the first single-layer NN is trained by samples consisting of spectrogram and message signal pairs for refining characteristics from the two-dimensional signal (spectrogram) into a one-dimensional output. Finally, the second single-layer NN is trained by the output of the first one to optimize the intermediate decision. The final decision is then given by low-pass filtering and thresholding. The trained NNs are used to unmask the chaotic shift keying from examples, which are not in the training set. Chua's circuit and the Lorenz system are used as the transmitters. Computer experimental results are given.