Fortifying Lifelong Security for O-RAN Ecosystem: An Incremental Learning Framework for NextG Seamless Networking

Next-generation cellular networks are shifting towards cloud-based infrastructures emphasizing programmability, virtualization, and modular architectures. The open radio access network (O-RAN) paradigm is a promising approach to overcoming traditional RAN limitations by offering an open framework for data-driven optimization at the individual user level. However, this openness also introduces vulnerabilities, where malicious attacks or unauthorized requests can compromise user privacy, disrupt resource allocation, and degrade overall service quality. In addition, new user equipment (UEs) can be introduced with new attacks and vulnerabilities over time. To address these challenges, we propose a deep incremental learning framework designed for the O-RAN environment, aimed at managing and mitigating these cyber threats while ensuring optimal user experience and resource utilization. We formulate an optimization problem to learn new attacks or vulnerabilities while retaining the knowledge of previous threat knowledge. To solve the optimization problem, we propose an exemplar-based convolutional neural network (CNN) model, implemented within the non-real-time RAN intelligent controller (non-RT RIC), to effectively monitor traffic at the radio units (RUs). Experimental results demonstrate that our proposed framework secures the O-RAN ecosystem with over $\mathbf{9 1 \%}$ accuracy, outperforming non-incremental methods by approximately 31% in distinguishing between benign and malicious traffic, thereby greatly enhancing network security and reliability for lifelong.