RF Fingerprinting for LoRa Device Authentication: Dataset Collection and Characterization
- Citation Author(s):
-
Priya Gautam (IIT (BHU) Varanasi)Merugu Jahnavi (IIT (BHU) Varanasi)Shubham Pandey (IIT (BHU) Varanasi)Hari Prabhat Gupta
(IIT (BHU) Varanasi)
Uma Maheswara (TII UAE)Kavin Thangadorai (TII UAE)Michael Baddeley (TII UAE) - Submitted by:
- HARI GUPTA
- Last updated:
- DOI:
- 10.21227/azwe-ca87
7 views
Abstract
We present a novel dataset for LoRa device authentication using Radio Frequency Fingerprinting (RFFI), addressing IoT security challenges in resource-constrained environments. Our dataset captures hardware-specific signal characteristics from 23 LoRa devices through three complementary representations: raw IQ samples, FFT spectra, and time-frequency spectrograms. Collected using a USRP B200 receiver with GPS synchronization, the data incorporates both coarse and fine Carrier Frequency Offset (CFO) estimations for enhanced feature analysis. The dataset supports deep learning approaches while maintaining an authentication-focused partition (20 training/3 testing devices) to evaluate real-world generalization. The primary contribution of this work is a multi-modal RF fingerprinting dataset specifically designed for LoRa networks. Beyond the core dataset, we further contribute an integrated CFO annotation, enabling hybrid authentication methods. Importantly, we also establish a reproducible experimental framework using software-defined radio that supports future research. Together, these contributions facilitate advancements in physical-layer security, lightweight device authentication, and robust wireless fingerprinting systems.
Instructions:
The key contributions of this work include:
1) Comprehensive LoRa Device Dataset Collection:We present a novel RF dataset comprising IQ samples collected from 23 different LoRa-based devices using a USRP receiver. The data was acquired under controlled conditions with consistent transmission intervals, enabling reproducible analysis.
2) Multi-Representation Signal Preprocessing: Each transmission instance is represented in three distinct formats — raw IQ samples, FFT-transformed frequency-domain data, and time-frequency spectrograms using STFT — to support diverse signal processing and learning approaches.
3) CFO-Aware Data Annotation: Coarse and fine Carrier Frequency Offsets (CFO) were estimated for each device during transmission, and a dedicated CFO database was maintained per device. This enables researchers to explore frequency-offset-related features without requiring additional synchronization.
4) Authentication-Focused Partitioning: A subset of three devices was reserved for device authentication testing, while the remaining 20 devices were used to generate training data, making the dataset suitable for evaluating generalization and cross-device performance.
5) Open Applicability for Future Research: The dataset is designed to support a range of research areas, including RF fingerprinting, physical-layer authentication, anomaly detection, and lightweight IoT security. It is compatible with both classical and deep learning-based workflows.
