[Paper Review] Towards an AI-Driven Universal Anti-Jamming Solution with Convolutional Interference Cancellation Network

Nguyễn Ngọc Hải, Guevara Noubir|arXiv (Cornell University)|Mar 18, 2022

Wireless Signal Modulation Classification4 citations

TL;DR

This paper proposes a universal, AI-driven anti-jamming solution using a convolutional neural network (CNN) to detect and cancel strong jammers in multi-antenna systems without requiring training sequences, channel estimation, or transmitter cooperation. The approach achieves over 99% jammer detection accuracy and a BER as low as 10⁻⁶ even when the jammer is 18 dB stronger than the legitimate signal, demonstrating robustness across modulation schemes and environments.

ABSTRACT

Wireless links are increasingly used to deliver critical services, while intentional interference (jamming) remains a very serious threat to such services. In this paper, we are concerned with the design and evaluation of a universal anti-jamming building block, that is agnostic to the specifics of the communication link and can therefore be combined with existing technologies. We believe that such a block should not require explicit probes, sounding, training sequences, channel estimation, or even the cooperation of the transmitter. To meet these requirements, we propose an approach that relies on advances in Machine Learning, and the promises of neural accelerators and software defined radios. We identify and address multiple challenges, resulting in a convolutional neural network architecture and models for a multi-antenna system to infer the existence of interference, the number of interfering emissions and their respective phases. This information is continuously fed into an algorithm that cancels the interfering signal. We develop a two-antenna prototype system and evaluate our jamming cancellation approach in various environment settings and modulation schemes using Software Defined Radio platforms. We demonstrate that the receiving node equipped with our approach can detect a jammer with over 99% of accuracy and achieve a Bit Error Rate (BER) as low as $10^{-6}$ even when the jammer power is nearly two orders of magnitude (18 dB) higher than the legitimate signal, and without requiring modifications to the link modulation. In non-adversarial settings, our approach can have other advantages such as detecting and mitigating collisions.

Motivation & Objective

• To design a universal anti-jamming building block that is agnostic to communication link specifics and does not require transmitter cooperation or explicit probing.
• To address the challenge of detecting and canceling strong jammers (up to 18 dB higher than the signal) in real-world wireless environments.
• To develop a machine learning-based solution that operates without training sequences, channel estimation, or modifications to existing modulation schemes.
• To evaluate the approach across diverse modulation schemes and propagation environments using a software-defined radio prototype.
• To demonstrate the feasibility of using deep learning for real-time, universal physical-layer anti-jamming in practical multi-antenna systems.

Proposed method

• A custom convolutional neural network (CNN) is trained to infer the existence of interference, the number of interfering signals, and their respective phases from raw RF signals.
• The CNN processes baseband I/Q samples from a two-antenna receiver to estimate interference characteristics in real time.
• The estimated interference parameters are fed into a real-time cancellation algorithm that applies phase- and amplitude-adjusted nulling to suppress jamming signals.
• The system operates without training sequences, pilot tones, or channel state information, relying solely on raw signal observations.
• A two-antenna SDR prototype is implemented using USRP software-defined radios to validate the approach in real-world environments.
• The CNN is trained on diverse signal configurations to ensure generalization across modulation types and propagation conditions.

Experimental results

Research questions

• RQ1 Can a deep learning model detect and characterize multiple jammers in real time without requiring training sequences or channel estimation?
• RQ2 How effective is a CNN-based approach in canceling jammers that are significantly stronger (up to 18 dB) than the legitimate signal?
• RQ3 To what extent can the proposed anti-jamming system maintain low BER across different modulation schemes and propagation environments?
• RQ4 Can the system achieve universal applicability across existing wireless communication links without modifying modulation or signaling protocols?
• RQ5 How does the system perform in non-adversarial scenarios such as collision detection and mitigation?

Key findings

• The proposed CNN-based anti-jamming system achieves over 99% accuracy in detecting the presence of jammers across diverse test configurations.
• The system maintains a Bit Error Rate (BER) as low as 10⁻⁶ even when the jammer power is 18 dB higher than the legitimate signal.
• The approach is effective across multiple modulation schemes, including QPSK, 16-QAM, and 64-QAM, without requiring changes to the link’s modulation or signaling.
• The system successfully detects and mitigates interference in multipath environments, demonstrating robustness beyond line-of-sight assumptions.
• The method enables universal anti-jamming without training sequences, channel estimation, or transmitter cooperation, making it deployable with existing wireless systems.
• The prototype implementation on SDR platforms confirms the feasibility and real-time performance of the approach in practical settings.