Machine Learning‑Enhanced Physical Layer Key Generation in OFDM Sub‑THz Systems

This document presents the first steps and results of a physical layer key generation (PKG) system for sub-THz wireless communications, addressing both Time Division Duplex (TDD) and Frequency Division Duplex (FDD) scenarios. Channel characteristics are extracted using OFDM-based channel sounding, and the FDD operation mode overcomes non-reciprocity challenges by incorporating machine learning techniques to predict downlink characteristics from the uplink measurements. The extracted channel features are processed through quantification methods to minimize the key disagreement rate (KDR).