GenCoder++: A Protocol-Aware and Adversarially Robust Adaptive Intrusion Detection Framework for Hybrid CAN-Ethernet Vehicular Networks

This paper presents GenCoder++, a robust and adaptive Intrusion Detection System (IDS) framework designed for hybrid vehicular networks that incorporates the Controller Area Network (CAN) and Ethernet protocols. Unlike traditional static IDS approaches, GenCoder++ integrates a compression-aware deep neural classifier, generative variational autoencoder (VAE), and Adversarial Machine Learning (AML)-Shield defense module to proactively handle evolving adversarial threats and concept drift. The proposed system dynamically retrains itself using realistic synthetic samples, generated by an entropy-regularized VAE, enhancing resilience against unseen attack patterns. A novel communication layer coordinates the decision logic across modules and triggers retraining based on entropy and anomaly density metrics. Experimental evaluations on a hybrid CAN-Ethernet dataset revealed that GenCoder++ outperformed state-of-the-art models in terms of accuracy, generalization, and adversarial robustness, achieving 92.5% accuracy and recovering up to 18.8% F1-score under Projected Gradient Descent (PGD) attacks. Furthermore, the compressed GenCoder-Light variant demonstrates real-time efficiency with 4.2× faster inference and 50% less memory usage, making it suitable for deployment in embedded Electronic Control Units (ECU). Comprehensive ablation and energy profiling analyses confirmed the modular contribution and practical viability of GenCoder++ for the next-generation intelligent automotive systems.