A Secure Communication Method for Unmanned Undersea Systems Oriented to Federated Learning

To address the information leakage issue in unmanned undersea systems, a secure communication method oriented to federated learning was proposed in this paper. By considering the complexities and bandwidth limitations of acoustic channels, an unbiased gradient compression method using Kashin compression was introduced. This method reduced the dimensionality of transmission gradients through orthogonal projection and quantization, thereby decreasing communication costs while preserving information integrity. To mitigate information leakage, a privacy-preserving method utilizing a feedback channel was designed, and the normalized acoustic channel transmission matrix and random sequences were employed to generate secret keys, so as to ensure that eavesdroppers cannot decrypt the model parameters. By adopting multi-objective optimization techniques, the Pareto optimal solution was found to strike a balance between model accuracy and secure throughput. Simulation results show that compared with the existing methods, the proposed method in this paper can effectively improve training accuracy and secure throughput while maintaining low latency.