SDR Testbed for Data-Driven Reactive Jammer Mitigation

Cooperative Mitigation against Learning-Based Reactive Jammers: SDR Validation and Experimental Analysis

Experimental implementation and evaluation of cooperative physical-layer countermeasures against learning-based reactive jammers using SDR testbeds.

Objectives of the Project

Develop cooperative physical-layer countermeasures against learning-based reactive jammers.
Experimentally implement the proposed schemes using SDR-based wireless testbeds.
Validate the correctness of the hardware setup using constellation interpretation and SNR estimation.
Investigate the covertness of the proposed countermeasures against data-driven adversaries using ML-based detection analysis.

Network Model

A crowded single-cell based wireless network with multiple user-equipments (UEs) communicating with the base station, namely Bob, as shown in Fig. 1.
Four UEs, namely Alice, Charlie, Tom, and Frank, that communicate with Bob over the uplink frequency bands fAB, fCB , fTB, and fFB, respectively.
The uplink channels are assumed to be frequency-flat and quasi-static.
Alice has critical information to communicate with Bob, and is therefore a potential victim of a Denial-of-Service (DoS) attack from an adversary.
To avoid pilot contamination attacks from adversaries, Alice communicates with Bob using non-coherent On-Off Keying (OOK) modulation.
Other UEs are not directly under adversarial attack, and communicate with Bob using coherent modulation schemes.
To support coherent demodulation over the fCB, fTB, and fFB bands, these UEs transmit pilots followed by information symbols within the coherence-time.

Threat Model

In the same network, a reactive adversary, namely Dave, is also present.
Dave is equipped with a full-duplex radio.
Dave injects jamming energy on Alice's uplink frequency band, i.e., fAB.
Dave monitors different network frequencies to detect the presence of potential countermeasures.
To detect countermeasures, Dave employs generalized energy detectors comprising:
- statistical detectors, based on instantaneous and distributional energy metrics.
- data-driven detectors that use machine learning classifiers to learn discriminitive features in the observed energy sequences.

Fig. 1: Network model depicting uplink communication between the UEs and the base station. The reactive adversary, namely, Dave, is seen injecting jamming energy on the victim Alice, while monitoring all network frequencies.

Cooperative Countermeasures

Alice tunes to the helper’s frequency band, where the helper node is Charlie.
Alice and Charlie cooperatively transmit their information symbols using fCB band using a part of their energies.
They pour their residual energy on fAB band to mimic the signaling scheme that existed prior to the implementation of the countermeasure.
The pattern of energy usage is quantified by a parameter called the energy-division factor, denoted using α, where α ∈ (0, 1).
The transmission of information symbols is divided across two time-slots.

Signaling over the fAB band

In time-slot 1, Alice and Charlie pour α and 1−α energies over the fAB band to form pseudorandom bit sequences using a shared secret-key, as shown in Fig. 2.
In time-slot 2, Alice independently handles the formation of pseudorandom bit sequences.
No information symbols are communicated over the fAB band.
The communication overhead associated with the formation of pseudo-random bit sequences is 0.5 bits/time-slot.

Fig. 2: Figure shows the signalling scheme over the fAB band during the two time-slots.

Signaling over the fCB band

In time-slot 1, Alice and Charlie pour 1−α and α energies over the fCB band to communicate with Bob, as shown in Fig. 3.
Alice and Charlie use On-Off Keying (OOK) and M-ary Phase-Shift Keying (M-PSK), respectively.
To increase reliability to Alice's bit, Charlie, who is equipped with FDR, simultaneously listens to Alice's bit while transmitting his own M-PSK symbol.
Charlie decodes Alice's bits, and incorporates it in time-slot 2 in the form of energy and phase modifications.
- If the decoded bit is 1, Charlie transmits his M-PSK symbol without modifications.
- If the decoded bit is 0, Charlie increases the transmit energy to 2−α and introduces an additional phase-shift of π/M to his M-PSK symbol.

Fig. 3: Figure shows the signalling scheme over the fCB band during the two time-slots.

Cooperative Countermeasures based on Variations in Signaling over the fCB Band

Fig. 4: Variations in signaling over the fCB band give rise to multiple cooperative countermeasures with different latency and practical FDR constraints.

Design Principles of the Cooperative Countermeasures

Owing to the fact that statistical detectors are rule-based and have low-complexity, the proposed countermeasures are primarily designed against them, such that:
- the statistical distribution of the received-symbol energies over the fAB band remains identical before and after implementing the cooperative countermeasures.
- The average received-symbol energy over the fCB band across 2n time-slots remains unity after implementing the cooperative countermeasures, which is consistent with the before-countermeasure case.

SDR-Based Validation and ML-Based Evaluation

The practical feasibility of deployment of the proposed countermeasures is validated using an SDR-based hardware testbed. Using this setup, real energy sequences are collected for both countermeasure and no-countermeasure configurations, and are subsequently used to train the data-driven detectors at the adversary using supervised and unsupervised ML-classifiers. Finally, the trained classifiers are used to evaluate the ML-based detection capability at the adversary, and the results show that the proposed countermeasures cannot be detected with a high probability under data-driven detection mechanisms, both when countermeasure samples are unavailable and available during training.

Fig. 5: Experimental workflow for SDR-based validation and ML-based detection analysis of the proposed cooperative countermeasures.

SDR-Based Hardware Validation

The practical feasibility of the proposed cooperative countermeasures is validated using an SDR-based hardware testbed comprising Alice and Charlie as transmitter nodes, and Dave as the adversarial monitoring node, as shown in Fig. 6. The theoretical analysis of covertness is carried out assuming that Dave employs an instantaneous energy detector over the fCB band. Consequently, the SDR-based implementation focuses on the signaling scheme over the fCB band, while the signaling scheme over the fAB band is excluded from hardware implementation. Furthermore, from the perspective of covertness at Dave, all proposed schemes exhibit similar signaling structures. Therefore, only the implementation of the DTRTF scheme is discussed using the SDR-based hardware testbed. Specifically, we employ three SDRs, operated via Adalm-Pluto devices, representing Alice, Charlie, and Dave. The Adalm-Pluto SDR nodes are connected to desktop computers/laptops running Windows-11 and MATLAB-2024b. The Adalm-Plutos perform RF transmission and reception, while all the baseband signal processing is carried out in MATLAB/Simulink on the computing machines. Furthermore, the transmitter and receiver configurations follow the standard Adalm-Pluto QPSK examples, with the necessary modifications incorporated to implement the proposed cooperative countermeasures.

Detailed Implementation of the DTRTF Scheme

Fig. 6: SDR-based hardware testbed using Adalm-Pluto devices connected to desktop computers to implement Alice, Charlie, and Dave in the experiments.

This video presents an SDR-based hardware testbed for implementing and evaluating countermeasures against reactive jammers in wireless communication systems. We discuss the threat model, jammer behavior, practical mitigation techniques, and real-time implementation using software-defined radio platforms.

ML-Based Detection Analysis

To investigate the robustness of the proposed cooperative countermeasures against data-driven adversaries, real energy sequences are collected using the SDR-based hardware testbed for both before-countermeasure and after-countermeasure configurations. These energy sequences are subsequently used to train supervised and unsupervised ML-classifiers at Dave for detecting the presence of cooperative countermeasures. Specifically, the supervised ML-classifiers comprise Random Forest (RF), Multi-Layer Perceptron (MLP), and Convolutional Neural Network (CNN), while the unsupervised ML-classifier comprises Isolation Forest (IF). Furthermore, we consider both scenarios where the countermeasure samples are unavailable to Dave during training, and scenarios where Dave is adaptive, observes the post-countermeasure behavior, and retrains the supervised models using countermeasure samples. In Fig. , we plot the receiver-operating characteristics (ROC) curves for both scenarios of different ML-classifiers, along with that of the random-classifier. We observe that the ROC curvesdo not exhibit a pronounced bulge away from that of the random classifier. This observation confirms that at low values of probability of false-alarm, high probabilities of detection cannot be achieved, thereby demonstrating that the proposed countermeasures cannot be detected with a high-probability by a data-driven adversary equipped with ML-classifiers for both scenarios.

Detailed Discussion on the Dataset for the ML-Classifiers

Detailed Discussion on the Detection Performance Analysis of the Countermeasures against Data-Driven Adversary

Fig. 7: Figure plots the ROC curves of different ML-classifiers where countermeasure samples are available and unavailable during training at Dave.

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.

Related Publications:

Refeered Journals:

Soumita Hazra and J. Harshan, ``On Reliable Communication under Reactive Adversaries with Generalized Energy Detectors," in IEEE Transactions on Vehicular Technology, vol. 74, no. 4, pp. 5970-5985, April 2025.
Soumita Hazra and J. Harshan, ``Cooperative Mitigation against Learning-Based Reactive Jammers: Analysis and SDR Validation," under review at IEEE Transactions on Vehicular Technology, May 2026.

Conferences:

Soumita Hazra and J. Harshan, ``On Distribution-Preserving Mitigation Strategies for Communication Under Cognitive Adversaries," in the Proc. of IEEE International Symposium on Information Theory (ISIT) 2023, Taiwan.
Soumita Hazra and J. Harshan, ``On High-Rate, Low-Overhead Mitigation Strategies Against Cognitive Adversaries," National Conference on Communications 2024, Feb. 2024.

Acknowledgement

This work was supported by the research grant “Secure Networks and Edge-Computing Hardware for Industry 4.0.”, funded by the Ministry of Electronics and Information Technology, New Delhi, India.

Page updated

Google Sites

Report abuse