Intelligent intrinsic physical layer security: joint authentication and encryption for data center communications

Navigating the era of artificial intelligence and massive cloud computing presents a critical challenge: securing the explosive growth of data traffic across optical data center interconnects (DCIs). However, conventional physical-layer security (PLS) methods often treat encryption and authentication as disjointed modules, lacking the intelligent adaptability for dynamic channels. Here, we propose an intelligent intrinsic PLS framework to jointly execute identity authentication and data encryption in short-reach DCIs. Enabled by end-to-end deep learning, this scheme integrates a mixture density network (MDN) for dynamic channel condition with a constellation autoencoder. Utilizing chaotic index modulation, the system nonlinearly learns highly distinguishable geometric constellations and embeds watermarks into subcarrier indices for redundancy-free transmission. Through this intelligent mapping, security attributes are natively fused into the physical signals. We achieve a secure transmission using discrete multi-tone (DMT) signaling over a 10-km standard single-mode fiber (SSMF). The results demonstrate that our scheme dynamically optimizes watermark embedding, strongly thwarting unauthorized eavesdropping and robustly resisting impersonation attacks. This avant-garde paradigm offers a highly scalable and intelligent solution for secure next-generation optical DCI networks.