SDR-Captured Drone Controller LoRa Spectrogram Dataset
Description
This dataset contains spectrograms of drone controller captured during non-cooperative drone operations with the signal level attenuated to the loss-of-connection level. The dataset can be used for validating and testing detection algorithms such as YOLO. It consists of the following components: - captured: non-labeled spectrograms of LoRa signals; - labeled_BW500_2025: (500, 7-8) subset labeled in YOLO format with [1] algorithm and manually adjusted; - labeled_BW500_2026: label indexes adjusted for [2] model class names; The hardware setup: - TX12 Mark II radio controller (RC), and a Nomad Dual Gemini Xrossband ExpressLRS Module. - The 30dB attenuator was plugged in to the RC antenna to decrease the power of the control signal. - FPV drone equipped with a BAYCKRC 900/2400 Dual Band Gemini RX receiver - ADALM-Pluto SDR RevC - Panel Antenna - Apple M2 Max MacBook Pro laptop Both the transmitter and receiver are compatible with ExpressLRS 3.5.4 firmware and are equipped with dual 3rd-generation Semtech LR1121 transceivers. Software and settings: - MATLAB R2024b with Communications Toolbox Support Package - SDR sampling rate: 32 MHz - RC ELRS control bands: 2.4 GHz and 915 MHz - RC packet rates: 50 kHz and 100 kHz for both bands. Collection process: The operation was conducted indoors using a drone, positioned close to the capture setup. The operator, equipped with an attenuated radio controller, moved a considerable distance into another room with thick brick walls until the control signal was lost. The controller-reported RSSI (Raw Signal Strength Indicator) was about -104 dBm. At this point we turned off the drone telemetry and started capturing controller signals. Result: - LoRa packets with (500, 7-8) and (812, 7-8) bandwidth, spreading factor (BW, SF) parameters The visual assessment of the spectrogram closely resembles the synthetic LoRa signal mixed at a signal-to-noise ratio of –20 to –15 dB.
Files
Steps to reproduce
For detailed guidance on steps to reproduce and usage please refer to our related works: [1] G. Dudarek and S. Martyniuk, “Identifying LoRa parameters using convolutional neural networks,” Radioelectronics and Communications Systems, Oct. 2025, doi: 10.20535/S0021347025020013. [2] G. Dudarek and S. Martyniuk, “From Discrete to Continuous LoRa Parameter Estimation Us-ing Vision-Based Deep Learning,” 2026, Submission is in progress.