DRONE
Software/Firmware
AutoPilot
PX4 FlightStack
iNav
LibrePilot
Paparazzi
Ardupilot
First Person View
BetaFlight
CleanFlight
dRonin
RaceFlight
MultiWii (Inactive)
BaseFlight (Inactive)
TauLabs (Inactive)
OpenPilot (Inactive)
Techniques
Acoustic
Vision
Thermal
Radar
Existing Work
https://arxiv.org/pdf/2003.01519.pdf
Amateur Drones Detection: A machine
learning approach utilizing the acoustic
signals in the presence of strong
interference
Independant Component Analysis (ICA)
- Unmix Drone, Birds, Rain, Wind..etc
Extract Audio Features
(E.g. MFCC, PSD..etc)
RF Fingerprinting
Data Comms (Encrypted)
Existing work
Deployed Systems
click to edit
Techniques
3D Lidar, Acoustic, Radio, VA
Range
250 to 300m
Features
Detect and Track up to 4.5KM
Verification, Identification & Classification up to 2.5KM
Neutralize up to 2KM
Azimuth coverage: 360°
Elevation: 20° up to 60°
Techniques
Lidar, Vision, Radar
PiNcH: an Effective, Efficient, and Robust Solution to Drone Detection via Network Traffic Analysis
https://arxiv.org/pdf/1901.03535.pdf
Works for
All sorts of drones
Binary Classification
Components required
Wifi Probe sensitive enough
to eavesdrop 2 way traffic
Features
- Packet size
- Packet inter-arrival time
- Derivatives
Multiclass Classifier
Actions
Status
Challenges
Irrelevant signals in the same freq band
Existing Works
Micro-UAV Detection and Classification from RF Fingerprints Using Machine Learning Techniques
https://arxiv.org/pdf/1901.07703.pdf
Focus on Controller
Preprocessing
- To improve low
SNR detection - Using wavelets
Detection
- Drone or Noise
- Based on observed
distinct transition signals - Use Markov models to detect
Classification
- Energy: skewness, variance,
energy spectral entropy, kurtosis - Time-Frequency
- Neighborhood Component Analysis:
Remove correlated features
Strategies
- DOA
- Detection
- Classification
- Identification
- Status
Applicable Strategies
- Status
- Detection
- Classification
Applicable Strategies
- Detection
- Classification
https://arxiv.org/pdf/1902.11130.pdf
Real-Time detection, classification
and DOA estimation
of Unmanned Aerial Vehicle
Applicable Strategies
- DOA
- Detection
- Classification
Compute 3D space
- Omni dir microphones
- Time diff
- Dir of drone
Features
- PSD
Classification
Machine Learning Framework for RF-Based Drone Detection and Identification System
https://arxiv.org/pdf/2003.02656.pdf
Datasets
Features
- off DroneRF daatset
Classification
- XGBoost outperform DNN
click to edit
RF-based Direction Finding of UAVs Using DNN
https://arxiv.org/pdf/1712.01154.pdf
https://github.com/LahiruJayasinghe/DeepDOA
Hardware
- Single channel RF Receiver
- Directional Antennas
Applicable Strategies
- DOA
Features
- PSD over antennas
Sparse Representation
- Trained with Autoencoder DNN
- Think of this as embedding
DNN Training
- Label is Direction
No Radio Left Behind: Radio FingerprintingThrough Deep Learning of Physical-LayerHardware Impairments
https://www.researchgate.net/publication/340572325_Deep_Learning_for_RF_Fingerprinting_A_Massive_Experimental_Study
Matthan: Drone Presence Detection by Identifying Physical
Signatures in the Drone’s RF Communication
https://dl.acm.org/doi/pdf/10.1145/3081333.3081354
Detection and Classification of UAVs Using
RF Fingerprints in the Presence of Interference
https://arxiv.org/pdf/1909.05429.pdf
Localisation of Drone Controllers from RF Signals using a Deep
Learning Approach.
http://eprints.lincoln.ac.uk/id/eprint/34434/1/localisation-drone-controllers%20(12).pdf
Towards RF-based Localization of a Drone and Its Controller
https://www.cs.colorado.edu/~rhan/Papers/2019_DRONET_RF-based_Localization.pdf
Deep Learning for RF-Based Drone Detection and Identification: A Multi-Channel 1-D Convolutional Neural Networks Approach
https://ieeexplore.ieee.org/document/9089657
RF-based drone detection and identification using deep learning
approaches: An initiative towards a large open source drone database
Signal Transformation
Complex to Data
Reference Datasets for Training and Evaluating
RF Signal Detection and Classification Models
https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=928336
RF Dataset of Incumbent Radar Systems in the 3.5 GHz CBRS Band
https://nvlpubs.nist.gov/nistpubs/jres/124/jres.124.038.pdf
DEEPSIG DATASET: RADIOML 2016.10A
Example ClassifierJupyter Notebook: RML2016.10a_VTCNN2_example.ipynb
Larger Version (including AM-SSB)
Over-the-Air Deep Learning Based Radio
Signal Classification
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8267032
Micro-UAV Detection and Classification
from RF Fingerprints Using
Machine Learning Techniques
Consists of post-processed segments of amplitude-time signals of drones and background
ORACLE: Optimized Radio clAssification through Convolutional neuraL nEtworks
Datasets for RF Fingerprinting of Bit-similar USRP X310 Radios
http://www.genesys-lab.org/oracle
Dataset#1: Raw IQ samples of over-the-air transmissions from 16 X310 USRP radios
Dataset#2: Demodulated IQ symbols of over-the-cable transmissions with 16 configurations of IQ imbalances
ORACLE: Optimized Radio clAssification through Convolutional neuraL nEtworks
https://arxiv.org/pdf/1812.01124.pdf
Features
IQ samples in Channel Invariant Env
Complex demodulated symbols in varying channel env
Architecture
2 CNN Layer with 2 FCN and a Softmax for classification
Pros
Cons
- Its too sensitive, able to fingerprint different devices even if identical make/model
- Performs badly in channel variant env
- A pretrained model based on data collected in Time A and Location A, will not work in Time B... or Location B... due to IQ data dependance on environment which varies in time and space.
Considerations
- HIgher quality SDR produces less impairments and make it less discrimitative for IQ data
- The same signal received over different channels actually can become 'different'. So channel variance can be a problem.
Introduced impairments to increase robustness
click to edit
Features
- Perform DFT on DroneRF dataset
- Split the 80Mhz into 8 channels of 10Mhz each
- Each channel consists of 2048 freq points
- Input: [8x2048]
Architecture
- 1D multi-channel CNN
- FC Layers
- Softmax
PiNcH: an Effective, Efficient,
and Robust Solution to
Drone Detection via Network
Traffic Analysis