In 6G non‑terrestrial networks, severe Doppler shifts, long propagation delays and fast channel variations cause distribution shifts that degrade conventional physical‑layer authentication (PLA). To address this, we present a secure adaptive multi‑zone framework (SAFA‑MZ) that leverages causal meta‑learning for distributed PLA. First, a multi‑feature fingerprint is designed by fusing spatial, angular, combiner, subspace and Doppler‑delay attributes; the fingerprint is collected across several aerial nodes and adapts to changing conditions. Second, a structural causal model (SCM) is built to capture the causal links among design choices, environmental factors, extracted features and authentication outcomes. Third, we apply model‑agnostic meta‑learning (MAML) together with invariant risk minimization (IRM) and a causal‑consistency regularizer, enabling rapid adaptation to unseen NTN scenarios with only a few labeled samples. Fourth, a two‑stage authentication scheme is proposed: local nodes perform fast recognition, and only when necessary they trigger a time‑difference‑of‑arrival (TDOA) localization powered by a graph attention network (GAT), thus reducing backhaul overhead. Simulations show that SAFA‑MZ achieves 92 % accuracy and 96 % AUC across diverse environments, outperforming centralized deep‑learning and single‑feature baselines.
Review