FedJam: Multimodal Federated Learning Framework for Jamming Detection

TL;DR

This paper tackles jamming detection in wireless networks by introducing FedJam, a privacy-preserving, multimodal Federated Learning framework that runs entirely on resource-constrained edge devices. FedJam processes synchronized spectrograms and cross-layer KPI time-series through a lightweight dual-encoder architecture with a fusion module and a small projection head, enabling on-device training and inference without raw data transfer. Through a real-world wireless testbed and over-the-air multimodal data, FedJam achieves up to 15% higher detection accuracy than unimodal baselines and requires up to 60% fewer communication rounds to converge, while showing strong robustness under non-IID data distributions. The work highlights the practicality of multimodal FL for wireless security and provides open artifacts to support reproducibility.

Abstract

Jamming attacks pose a critical threat to wireless networks, yet existing detection methods remain largely unimodal, centralized and resource-intensive, limiting their performance, scalability, and deployment feasibility, respectively. To address these limitations, we present FedJam, a multimodal Federated Learning (FL) framework for on-device jamming detection and classification. FedJam locally fuses spectrograms and cross-layer network Key Performance Indicators (KPIs) using a lightweight dual-encoder architecture with an integrated fusion module and multimodal projection head, that enables privacy-preserving training and inference without transmitting raw data. We prototype and deploy FedJam on a wireless experimental testbed and evaluate it using the first, over-the-air multimodal dataset comprising synchronized samples across benign and three distinct jamming attack types. FedJam outperforms state-of-the-art unimodal baselines by up to 15% in accuracy, while requiring 60% fewer communication rounds to converge, and maintains low resource utilization. Its advantage is especially pronounced in realistic scenarios, where it remains extremely robust under heterogeneous data distributions across devices.

Figures (12)

• Figure 1: FedJam enables collaborative, on-device learning over multimodal data without exposing raw private measurements.
• Figure 2: FedJam system architecture.
• Figure 3: Experimental testbed components and lab layout.
• Figure 4: Collected over-the-air spectrograms illustrating benign transmissions and three distinct jamming attack types.
• Figure 5: Collected multimodal traces from an edge device, including time-aligned spectrogram images and KPI time-series for Wi-Fi (top row) and 5G (bottom row). Periods of jamming activity are highlighted with pink boxes.