Anomaly-Driven Jamming Attack Identification Using Hybrid Deep Neural Networks

Abstract

Healthy life enhances the quality of life by allowing individuals in engage able Wireless communication systems supporting Internet of Things (IoT), cyber-physical systems, and next- generation networks are increasingly vulnerable to sophisticated jamming attacks that degrade spectrum availability and service reliability. Conventional anti-jamming techniques based on fixed thresholds and handcrafted signal features fail to adapt to dynamic and intelligent interference behaviors. In this paper, an Anomaly-Driven Jamming Attack Identification Framework (AD-JAIF) is proposed using a hybrid deep neural network architecture that integrates convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) networks. Time–frequency spectrogram representations of received signals are employed to enable effective spatial–frequency feature extraction through CNNs and temporal dependency modeling through BiLSTMs. To address the challenge of limited labeled attack data and zero- day jamming scenarios, an anomaly-driven learning mechanism is incorporated to model normal communication behavior and detect deviations indicative of jamming attacks. Extensive simulations were conducted under realistic wireless channel conditions considering constant, reactive, and adaptive jamming strategies over a wide signal-to-noise ratio range. Experimental results demonstrate that the proposed framework achieves superior detection performance, exceeding 98% accuracy with reduced false alarm rates and low detection latency compared to traditional machine learning and standalone deep learning models. The proposed AD-JAIF offers a robust, scalable, and real-time solution for securing emerging wireless communication systems against modern jamming threats.