Application of neural networks to unmasking chaotic secure communication (Q1809429)

The paper deals with chaotic shift keying (or chaotic switching) which is the most robust secure communication scheme. In the shift keying a binary information signal is hidden into chaotic time series generated by two different chaotic attractors. To make both indistinguishable from the transmitted signal is the basic idea to achieve security of chaotic shift keying. The security considerations done previously only from the time domain turn out to be not enough because an information leak in the spectral domain could cause fatal failures. The chaotic shift keying is unmasked by using two simple single-layer neural networks with known text strategy which reveal information leaks and thus break this chaotic secure communication scheme from the spectral space. In the first step of the proposed algorithm the time series of the transmitted signal are transferred into the spectral-temporal space by using a spectrogram (the moving window fast Fourier transformation of the time series). So the chaotic cryptanalysis problem becomes a two-dimensional pattern classification problem which is solved in the second step by the single-layer neural networks. One network reduces the two-dimensional spectrogram pattern to a one-dimensional characteristic signal which is fed into the other network. This network refines and optimizes the signal to give the second intermediate decision from which the final decision is very easily obtained by low-pass filtering and thresholding. The authors use Chua's circuit and the Lorenz system as transmitters and give various computer experimental results.