Cascaded feedforward neural network decryption framework for chaotic optical communication

Most existing decryption methods for chaotic optical communication rely on chaos synchronization, which can be susceptible to external interference and suffer from performance limitations. This paper proposes a chaotic optical communication decryption framework based on a cascaded feedforward neural network (CFNN). The framework constructs a two-dimensional matrix corresponding to intermediate features within the neural network using BiMatch. Through the continuous inference of the cascaded neural network, it progressively extracts the features of the encrypted signal and ultimately recovers the message, without the need for chaos synchronization, alignment, and differential. Under the condition of maintaining comparable parameter size and computational complexity to traditional models, CFNN can reduce the bit error rate (BER) to below 3.8 × 10^-3 in most cases, demonstrating clear advantages in both decryption accuracy and robustness. Additionally, security analysis and experimental validation further confirm the potential for the practical application of the proposed method.