Edge AI-based Radio Frequency Fingerprinting for IoT Networks

TL;DR

This work tackles secure IoT authentication by deploying radio frequency fingerprinting directly at the network edge. It presents two lightweight DL models—a CNN and a Transformer Encoder—trained on IQ samples to produce device-specific RF fingerprints, with TensorFlow Lite implementations enabling real-time edge inference on devices like the Raspberry Pi. The Transformer-based RFF outperforms the CNN in accuracy and robustness, achieving high performance (accuracy >0.95, ROC-AUC >0.90) while maintaining a small footprint (≈73 KB for quantized Transformer). Edge deployment is demonstrated via GPU-accelerated training and Raspberry Pi inference, with significant gains from quantization. The results suggest practical, low-latency PHY-layer authentication for large-scale, resource-constrained IoT deployments in smart cities and beyond.

Abstract

The deployment of the Internet of Things (IoT) in smart cities and critical infrastructure has enhanced connectivity and real-time data exchange but introduced significant security challenges. While effective, cryptography can often be resource-intensive for small-footprint resource-constrained (i.e., IoT) devices. Radio Frequency Fingerprinting (RFF) offers a promising authentication alternative by using unique RF signal characteristics for device identification at the Physical (PHY)-layer, without resorting to cryptographic solutions. The challenge is two-fold: how to deploy such RFF in a large scale and for resource-constrained environments. Edge computing, processing data closer to its source, i.e., the wireless device, enables faster decision-making, reducing reliance on centralized cloud servers. Considering a modest edge device, we introduce two truly lightweight Edge AI-based RFF schemes tailored for resource-constrained devices. We implement two Deep Learning models, namely a Convolution Neural Network and a Transformer-Encoder, to extract complex features from the IQ samples, forming device-specific RF fingerprints. We convert the models to TensorFlow Lite and evaluate them on a Raspberry Pi, demonstrating the practicality of Edge deployment. Evaluations demonstrate the Transformer-Encoder outperforms the CNN in identifying unique transmitter features, achieving high accuracy (> 0.95) and ROC-AUC scores (> 0.90) while maintaining a compact model size of 73KB, appropriate for resource-constrained devices.

Figures (8)

• Figure 1: System model assumed in this paper. A set of IoT devices with transmitters Tx$_{1, ...,\text{n}}$ are deployed in an attempt to connect to an edge device, i.e., an AP, with receiver Rx. The AP uses PHY-layer information to identify/authenticate each RF Fingerprint against a pre-trained model, allowing only authorized devices to connect to it.
• Figure 2: Structure and details of the implemented CNN.
• Figure 3: Structure and details of the implemented Transformer encoder.
• Figure 4: Batch sizes impact on model validation accuracy during model training for (a) CNN, and (b) Transformer
• Figure 5: Model training and validation accuracy/loss as a function of number of epochs, for (a) CNN and (b) Transformer encoder