Physical Layer Authentication Scheme Based on Improved SVM in Static Scenarios

This thesis proposes a physical layer authentication scheme based on multiple physical layer features and machine learning for static scenarios in Vehicular Ad-hoc Network (VANET). The scheme first trains and constructs the authentication model using multiple physical layer channel feature data and introduces data augmentation methods to enhance data diversity. This data augmentation strategy, combined with the multi-feature approach, significantly improves the effectiveness and richness of the feature data, thereby enhancing the stability and accuracy of the authentication model. Moreover, the scheme employs One-Class Support Vector Machine (OC-SVM) for the construction of the authentication model, which is trained solely on data from legitimate vehicles. This not only effectively reduces the training overhead of the model but also aligns more closely with the needs of real-world application scenarios. Experimental results show that the proposed scheme performs excellently in terms of authentication effectiveness, stability, and robustness, while also having low overhead, thus verifying its feasibility and superiority in static VANETs scenarios.