Physics-informed deep learning systems for secure information transmission with multimode fibers

• Applicant: Professor Dr.-Ing. Jürgen W. Czarske
• Subject Area: Measurement Systems ; Communication Technology and Networks, High-Frequency Technology and Photonic Systems, Signal Processing and Machine Learning for Information Technology
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 560574412

The aim of the Reinhart Koselleck project is to pursue exceptionally innovative and risky investigations on novel measurement systems to exploit scattering processes in multimode fibers (MMF) using physics-informed deep learning towards paradigm-shifts for secure information transmission. Fiberoptic communication technology forms the backbone of the Internet. Advances are not only important for further exponential growth of data rates, but especially for data security. Using the mode space of MMF compared to single-mode fibers enables an increase in data rates through spatial multiplexing. However, there are obstacles to information transmission in MMF due to mode crosstalk, which leads to mixing of the information channels in particular at large fiber lengths. This challenge is addressed in the Koselleck project by novel AI-based measurement systems. Artificial neural networks mimic the function of biological neural networks and are intended to be used to measure the transmission matrix (TM) of the MMF, but the learnable parameters increase exponentially with the mode number. A paradigm shift on physics-informed deep learning is pursued by both, “Physics Prior”, merging data driven algorithms with physical models, and “Physics in the Network”, an optical diffractive deep neural network (ODNN), that combines deep learning with diffractive optics. Optical neural networks are trained by artificial intelligence, but operate without energy-hungry GPUs. ODNNs have negligible energy consumption and therefore represent a paradigm shift for sustainable data processing. Real-time measurement of the transformation matrix (TM) of MMF by physics-based neural networks provides channel information between participants, which is exploited with the inverse TM through Physical Layer Security (PLS). It provides quantum-safe encryption as opposed to classical cryptography. Instead of post-quantum cryptography, physical laws are exploited not only with PLS but especially with quantum key distribution (QKD) using non-classical light, which guarantees data security due to the no-cloning theorem of single photons. The vision of the project is to explore and exploit terra incognita in information transmission through MMF using novel measurement systems and classical and non-classical light to improve the capacity and security of communications.

• DFG Programme: Reinhart Koselleck Projects