[ A ]
Sensing without sight
RF detection is passive — it uses only what the drone already radiates. Night, fog and obstacles matter little; the control link is still on the air.
SDR ONLINE — HANGZHOU, CN·--:--:--·UTC+8
Every drone talks.
I listen.
I'm Jinhao Wan, a PhD student at Zhejiang University of Technology. My research applies deep learning to RF signal identification, with a focus on detecting low-altitude UAVs from their radio emissions — passive sensing that needs no camera and no line of sight.
RF-BASED UAV DETECTION//AUTOMATIC MODULATION CLASSIFICATION//RF FINGERPRINTING//SIGNAL IDENTIFICATION//DEEP LEARNING//SOFTWARE-DEFINED RADIO//
////////////
[ 01 ] RESEARCH DIRECTION
The opposite way
to find a drone.
not visioncameras · radar · acoustics
Most counter-drone systems watch the sky. Cameras need light and a clear line of sight, radar is expensive to deploy at scale, and acoustic sensing breaks down in noisy environments.
My approach is passive RF sensing. A flying drone is also a radio — its control link, telemetry and video downlink are transmitting the whole time. I train deep networks to read these emissions from raw I/Q samples, so detection works at night, in fog and behind obstacles, and says which emitter is flying rather than just that something is.
- no line of sight
- 24/7 passive
- per-device identity
- low-cost SDR
[ 02 ] WHY THE SPECTRUM
Three reasons
to listen.
[ B ]
Identity, not just presence
RF fingerprints separate models and even individual devices. The answer is not just "a drone is nearby" but which controller, on which link.
[ C ]
Learning from raw samples
CNN, ResNet and Transformer models trained end-to-end on raw I/Q and spectrograms, with a focus on low-SNR robustness.
[ 03 ] LIVE INSTRUMENT
One signal,
four views.
Four views of the same signal — move the pointer, switch modes, click to scan.
Open the full explorer ↗
Signal feature spaceMove to reveal classes · click to probe the spectrumPSK OFDM Noise UAV
[ 04 ] RESEARCH
Four connected
directions.
From identifying the modulation on the wire to finding the drone behind it.
Open the research atlas ↗
01RF Signal IdentificationModulation type and signal class, straight from raw I/Q samples.categorize the spectrum
02UAV DetectionFinding and recognizing drones from their RF emissions, down to per-device fingerprints.find the emitter
03Deep Learning for RFCNN, ResNet and Transformer architectures that turn waveforms into decisions.learn from waveforms
04Robust RecognitionAccuracy that survives low SNR, fading and lab-to-field distribution shift.hold up under noise
[ 05 ] PUBLICATIONS
Selected work.
Two manuscripts in preparation, more as the work matures.
Search the full library ↗
Deep learning for automatic modulation classification of RF signals.
Convolutional and transformer models that classify the modulation type of received signals from raw I/Q samples, with an emphasis on robustness at low SNR.
RF-based detection and recognition of unmanned aerial vehicles.
Detecting and identifying small UAVs from their RF emissions using SDR captures and deep classifiers, including fingerprinting of individual devices.
[ 06 ] WHAT'S NEXT
The road ahead.
Three phases, plus a running log.
NOW
Foundations
Modulation classification and RF identification on SDR captures. First manuscript in preparation.
NEXT
Detection in the wild
UAV detection and per-device fingerprinting under real channels; field captures around campus.
THEN
Robust systems
Lab-to-field generalization, low-SNR robustness, deployable low-cost SDR pipelines.
LAB LOG
- Portfolio relaunch — site rebuilt as an RF research lab notebook.
- Research focus — deep learning for RF signal identification and low-altitude UAV detection.
- In preparation — AMC and UAV RF-fingerprinting manuscripts underway.
[ 07 ] CONTACT
Get in touch.
Questions, ideas, or collaboration on RF sensing and UAV detection are all welcome.
wanjinhao1@gmail.com