A security analysis scheme for physical layer chaos encryption based on neural networks

Abstract

Physical layer encryption schemes hold significant importance for contemporary data protection due to their flexible operations, comprehensive data protection capabilities, and low-cost, fast computation. However, there is currently no universal and effective method for security analysis of physical layer encryption systems. With various physical layer encryption methods emerging, a critical issue is whether the designed systems can effectively resist unauthorized attacks. This paper enhances the security evaluation of physical layer encryption schemes from the perspective of cryptanalysis, moving beyond simply using the size of the key space to evaluate the security of algorithms. The computational security of physical layer chaos encryption schemes is investigated by analyzing their ability to resist attacks. A neural network-based attack scheme against physical layer chaos encryption is proposed. Computer simulations verify the feasibility and effectiveness of the attack. Experimental results demonstrate that the proposed method can learn effective ciphertext-to-plaintext mappings under the tested fixed-key conditions. The method involves a moderate offline training cost and achieves sub-millisecond model-level inference after training. The results demonstrate that the investigated physical layer chaos encryption schemes are vulnerable to neural network-based attacks under the tested conditions. The proposed method provides a complementary security-evaluation approach for the investigated Arnold- and Chen-based encryption schemes.