Resilient Machine Learning for NextG Wireless Systems Under Smart Jamming
Yibin Liang, Usama Saeed, Lingjia Liu · 2024
Multi-user interference and adversarial jamming are critical issues affecting the ultra-high capacity and reliability of next-generation (NextG) wireless networks. Conventional interference mitigation techniques usually rely on model-based analysis, which cannot fully address new challenges in NextG systems. Reservoir computing (RC), a new brain-inspired machine learning paradigm, can outperform conventional methods for various tasks in wireless systems due to its efficient training algorithm and minimum requirements for training data. However, its resilience in interference scenarios has yet to be thoroughly investigated. This paper explores the performance improvement for MIMO-OFDM systems under smart jamming attacks and introduces a new resilient RC architecture (ResRC) for iterative interference detection and mitigation. The performance results from Monte Carlo simulations show that ResRC can effectively mitigate interference and improve system reliability and capacity. A software-defined radio prototype receiver further verifies the ResRC architecture design in various real-world scenarios.