Toward Analog Implementation of Recurrent Neural Networks in the RF Domain for Proactive Anomaly Detection in B5G Wireless Networks

This paper presents an analog RF-domain implementation of a Vanilla Recurrent Neural Network (RNN) for real-time anomaly detection in 5G and beyond wireless networks. Real-time analysis is crucial to distinguish benign irregularities from true anomalies that may indicate malicious activity or network issues. The proposed design leverages passive components, such as switches and capacitors, enabling low-latency processing directly at RF frequencies. Circuit simulations validate the approach, achieving up to 89% accuracy across three anomaly classes, demonstrating the potential of RF-domain neural networks for fast and efficient wireless signal analysis.