Embedded Files
Objective
Objective
Evaluate the covertness of the proposed countermeasure against ML-based detection
Energy of received IQ samples obtained after the carrier-synchronizer block is used for training and testing the ML classifiers
Fig. 1: Figure illustrates the overall SDR-based dataset generation and validation workflow adopted in this work.
Hardware Testbed Description
Hardware Testbed Description
Adalm-Pluto devices perform RF transmission and reception
Baseband signal processing is carried out in MATLAB/Simulink
Dave collect the IQ samples of the received symbols
In this video, we present an SDR-based SDR testbed for evaluating covertness under data-centric reactive jammers, where the adversary employs machine learning–based detectors to monitor and adapt to countermeasures. We discuss the threat model, implementation, and experimental evaluation of covert communication under adaptive and learning-enabled detection.
Class 0: Before-Countermeasure Setup
Class 0: Before-Countermeasure Setup
Communication is established between Charlie and Dave
Charlie transmits unit-energy QPSK symbols
Energies of received IQ samples of unit-energy QPSK symbols form the dataset for Class 0
Well-defined constellation clusters are observed for unit-energy QPSK transmission
Fig. 1: Figure shows the constellation diagram of the training dataset for Class 0, obtained using the SDR-based hardware testbed.
Class 1: Surrogate Training Dataset
Class 1: Surrogate Training Dataset
The true countermeasure symbols are unavailable during training in a realistic sce
Therefore, a surrogate Class 1 is constructed assuming that the countermeasure may modify:
transmit energy
modulation format
Surrogate Class 1 Symbols
Surrogate Class 1 Symbols
16-QAM symbols
64-QAM symbols
QPSK symbols with average energy 0.1
QPSK symbols with average energy 0.01
Fig. 2: Figure shows the constellation diagram of the training dataset for Class 1, obtained using the SDR-based hardware testbed.
Key Observation
Key Observation
Reduced-energy QPSK symbols exhibit increased constellation dispersion.
Lower transmit energy amplifies the effect of noise after AGC operation..
Higher-order QAM constellations appear more scattered due to denser constellation structures
Different modulation formats and transmit energy levels produce distinguishable energy patterns.
SNR Validation
SNR Validation
The first 13 received symbols correspond to the known Barker-sequence preamble modulated using QPSK.
The received IQ samples are correlated with the known preamble symbols to obtain the channel estimate at the receiver.
Using the estimated channel coefficient, the average signal and noise powers are computed to estimate the SNR.
For QPSK with transmit energies:
1 → 30.44 dB
0.1 → 20.37 dB
0.01 → 8.17 dB
Approximately 10 dB reduction in SNR is observed for every tenfold decrease in transmit energy, thereby validating the correctness of the hardware testbed.
Testing Dataset
Testing Dataset
Countermeasure-generated IQ samples are used during testing.
Testing is performed on unseen signal characteristics.
The testing dataset contains symbols collected from both time-slots.
Experiments are conducted for:
α = 0.9877
α = 0.9901
α = 0.9964
Page updated
Google Sites
Report abuse