Federated digital twin framework for spectrum handover management in vehicular THz ORAN

Abstract

Vehicular Terahertz (THz) Open Radio Access Networks (ORAN) face severe challenges due to ultra-high mobility, rapid channel dynamics, spectrum scarcity, and frequent handovers, which limit the effectiveness of existing reactive and centralized management strategies. This paper proposes a Federated Blockchain-Assisted Digital Twin framework for proactive spectrum sharing and seamless handover in vehicular THz ORAN systems. The framework employs a Contrastive Denoising Spatio-Temporal Autoencoder (CD-STAE) to refine multi-source telemetry while preserving transient anomalies, followed by a Federated Multi-View Neural State-Space Twin (FMNSS) that constructs a privacy-preserving digital replica of the network. A Spatio-Temporal Transformer with Heteroscedastic Uncertainty (ST-TrH) enables confidence-aware forecasting of load, interference, and mobility-driven congestion, while On-Chain Verifiable Reward Shaping (OVRS) integrates blockchain-derived trust scores into federated aggregation and proactive spectrum leasing. Building on these predictions, a Reinforced Graph Trajectory Encoded DiffNet (R-GTEDiffNet), comprising EM-ST-GAT, Causal Trajectory Encoder (CTE), and Differentiable Handover Cost Layer (DHCL), is introduced to perform mobility-aware spectrum slicing and predictive handover optimization. Simulation results demonstrate superior performance over state-of-the-art models, achieving up to 94.1% spectrum utilization, 6.2 ms leasing delay, 10.5 ms handover latency, 1786 Mbps throughput, and an average classification accuracy of 98.68%, showing the framework’s strength in high-speed vehicular THz ORAN scenarios.