Guo Zhiruo, Liu Bo, Ren Jianxin, Zhong Qing, Mao Yaya, Chen Shuaidong, Yang Xiantao, Song Xiumin, Wang Feng, Zhao Lilong, Tu Bing
Opt Express. 2025 Jul 14;33(14):29409-29426. doi: 10.1364/OE.564874.
In response to the issues of peak-to-average power ratio (PAPR) and dynamic degradation in multi-carrier chaotic encryption, this paper proposes the tone reservation (TR) encrypted scheme based on transformer-long short-term (LSTM), to achieve physical-layer encrypted transmission in the fiber wireless seamless integrated system. The proposed scheme integrates 23 large-scale 3D chaotic models with a hybrid transformer-LSTM neural network architecture. This combined framework generates 3D chaotic sequences via nonlinear predictive modeling, which are then applied to encrypt the original data at the physical-layer. The 3D chaotic sequences generated by the neural network are applied to TR selection, effective subcarrier scrambling, and whole subcarrier scrambling respectively. This scheme achieves physical-layer encryption while reducing the PAPR of the communication system, with an average reduction of 1.47 dB. The R (R-squared) of the chaotic sequence predicted by the neural network reaches over 0.99, and the root mean square error (RMSE) reaches 0.066. Compared with traditional single-characteristic chaotic sequence prediction, the proposed prediction scheme possesses multiple chaotic source characteristics, enabling it to resist nonlinear dynamical degradation issues. In this paper, the transmission of 12Gbaud quadrature phase shift keying (QPSK) signal at 20.6 Gb/s in 4-meter fiber wireless seamless integrated transmission system is experimentally demonstrated by experiments, and the performance of the system is analyzed. The encrypted signal meets hard decision forward error correction (HD-FEC) threshold @3.8×10 at the legal receiver, and the minimum bit error rate (BER) can reach 2.1×10. At the HD-FEC@3.8×10, the sensitivity of the input optical power (IOP) is improved by 0.45 dB after reserved subcarrier optimization. At the illegal receiver, without the correct key, the BER exceeds 0.4, and the original data cannot be decrypted, effectively preventing illegal attacks. The experimental results demonstrate that this scheme can secure data transmission from the very foundation of the communication system, indicating significant application potential in future fiber wireless seamless integrated secure communications.
针对多载波混沌加密中的峰均功率比（PAPR）和动态退化问题，本文提出了基于变压器 - 长短期记忆（LSTM）的音调预留（TR）加密方案，以实现光纤无线无缝集成系统中的物理层加密传输。所提方案将23个大规模3D混沌模型与混合变压器 - LSTM神经网络架构相结合。该组合框架通过非线性预测建模生成3D混沌序列，然后将其应用于物理层对原始数据进行加密。神经网络生成的3D混沌序列分别应用于TR选择、有效子载波加扰和全子载波加扰。该方案在实现物理层加密的同时降低了通信系统的PAPR，平均降低了1.47dB。神经网络预测的混沌序列的决定系数R超过0.99，均方根误差（RMSE）达到0.066。与传统的单特征混沌序列预测相比，所提预测方案具有多混沌源特征，能够抵抗非线性动态退化问题。本文通过实验在4米光纤无线无缝集成传输系统中演示了20.6Gb/s的12Gbaud正交相移键控（QPSK）信号的传输，并分析了系统性能。加密信号在合法接收端满足3.8×10时的硬判决前向纠错（HD - FEC）阈值，最小误码率（BER）可达2.1×10。在HD - FEC@3.8×10时，预留子载波优化后输入光功率（IOP）的灵敏度提高了0.45dB。在非法接收端，没有正确密钥时，BER超过0.4，原始数据无法解密，有效防止了非法攻击。实验结果表明，该方案能够从通信系统的根本层面保障数据传输安全，在未来光纤无线无缝集成安全通信中具有显著的应用潜力。
