Design of a Chaotic Communication System Based on Deep Learning With Two-Dimensional Reshaping

The chaotic communication system can be applied to enhance the security of communication between vehicles by utilizing the randomness of chaotic signals. In the traditional multi-carrier chaotic communication system, its application scenarios are greatly limited by bandwidth constraints and channel noise interference. Thus, the paper proposes a novel approach to enhance the reliability and transmission rate of multi-carrier chaos shift keying (MC-DCSK) systems. The proposed multi-carrier chaos shift keying (DL-IM-MCDCSK) system utilizes deep learning (DL) and index mapping (IM) techniques to mitigate the information leakage risk associated with conventional MC-DCSK systems. The system operates without a reference signal and utilizes a two-dimensional reshaping (TDR) index mapping structure to equalize the chaotic signals in both frequency and time domains. In addition, an auxiliary deep neural network (DNN) classifier with an improved LeNet5 architecture is designed to assist the receiver in information demodulation. The DNN classifier mainly consists of two convolutional layers (CLs) for extracting features of chaotic signals and two fully connected layers (FCLs) for classifying information symbols. The offline-trained DNN classifier can significantly improve the bit error rate (BER) performance during information recovery without requiring conventional maximum likelihood estimation (MLE). The performance of the proposed system was evaluated through Monte Carlo simulations over additive white Gaussian noise (AWGN) and multipath Rayleigh fading channels (RFC). The simulations demonstrated that the system outperforms various conventional multi-carrier systems in terms of transmission rate, spectral efficiency, and reliability.