DriveShield: FL-Based Framework for Privacy Preserving and Jamming Attack Detection in V2X Communication Underlying 6G

In this paper, we introduce DriveShield: a decentralized framework based on Federated Learning (FL) for real time jamming attack detection in VANETs under 6G communications. The local Feedforward Neural Network (FNN) trained on each client is aggregated by the popular Federated Averaging (FedAvg) approach and achieves a global accuracy around 83%, and reaches a peak of 87% across the training rounds. In order to preserve data privacy and integrity, the integration of blockchain and the InterPlanetary File System (IPFS) is used to make tamper resistant, secure model weight transfers. DriveShield is distinguished from such static dataset reliant approaches in the fact of their reliance on static datasets and real time data simulation of network conditions enhancing adaptability to dynamic network conditions. Performance evaluation shows that 6G-THz communication has better latency performance than 5g uRLLC, as well as better transaction scalability using IPFs. Robust detection capabilities with precision = 0.832-0.879 and F1-score = 0.822-0.880, over centralized baselines, are confirmed via metrics.