Machine learning assisted authentication using chaotic diversity index modulation for data centers

We propose a data center secure communication scheme that uses chaotic diversity index modulation for identity authentication, and design a neural network (NN) that balances decryption and detection for the receiver. Chaotic diversity is obtained from a randomly partitioned standard 16QAM constellation. The watermark is subsequently embedded into the symbol index by leveraging this diversity. Finally, the constructed data frames composed of symbols across different time slots and subcarriers is disrupted, and the watermark is concealed within the massive data transmission. The receiver takes the received signal and chaotic diversity as inputs for the NN to obtain the watermark, and achieves identity authentication by comparing it with the locally stored watermark. The 56.37Gb/s transmission verification was conducted on a 10 km standard single-mode fiber (SSMF), and the results showed that the scheme has high security. The total key space can reach 10⁹⁰. The watermark is hidden in the index of the symbol, making it difficult for eavesdroppers to detect. The combination of chaotic diversity and NN can cause data pollution to eavesdroppers, making it difficult to steal the correct NN. The accuracy of NN detection exceeds that of the log-likelihood ratio (LLR) detector, and it can reduce the time complexity of detection by 4 orders of magnitude. The recognition accuracy of watermarks can reach 100% under high optical power, and the legitimate and illegitimate parties can be effectively distinguished, making identity disguise attacks difficult to achieve.