Machine Learning-Based Jamming/Anti-Jamming Strategies in IoT Networks

Abstract

The pervasiveness of wireless Internet-of-Things (IoT) networks, evident in their widespread and ubiquitous presence across various sectors like homes, businesses, and public spaces, has escalated their susceptibility to malicious interference, notably jamming attacks.
This thesis presents defense strategies against such jamming attacks in IoT networks, capitalizing on machine learning (ML) techniques. To formulate these defense strategies, an in-depth understanding of jamming attacks and the effectiveness of different jammers was essential. Consequently, a low-cost software-defined radio (SDR) jamming toolkit was developed, facilitating the implementation of various types of jammers against wireless networks based on IEEE standards. Evaluating the performance of this toolkit and analyzing the results provided insights into the effectiveness of different jamming techniques.
An anti-jamming scheme was initially explored, utilizing clear channel assessment (CCA) information in a cognitive radio environment and implemented through deep reinforcement learning. Despite its potential, this approach encountered limitations in a dynamic jamming environment, particularly the challenge of extracting the necessary information from real wireless network interfaces. These shortcomings catalyzed the development of an innovative RF-jamming detection testbed. This design leverages the spectral scan capability of wireless network interfaces and combines it with ML algorithms to detect and mitigate jamming attacks effectively. A large dataset was generated using the jamming toolkit and the jamming detection testbed, and subsequently employed to train ML algorithms, achieving a jamming detection accuracy of over 90%. Upon detecting jamming, traditional anti-jamming strategies like frequency hopping (FH) can be deployed to alter the transmit frequency.
However, FH as an anti-jamming strategy has its limitations, such as requiring more spectrum resources and being ineffective against powerful jammers capable of jamming the entire spectrum. To address this, a novel approach to combating jammers was proposed, leveraging backscattering communications (BackCom) using neural network architectures. This approach allows transmitters to "face" the jammer rather than "escape" by adapting transmit power or transmitting information directly on the jamming signals via backscattering. To harness BackCom for anti-jamming, long-range capabilities are required. Therefore, a framework for Long Range (LoRa)-enhanced BackCom was proposed, extending the range of conventional BackCom significantly. The error rate performance of this framework was studied over additive white Gaussian noise (AWGN) and Rayleigh fading channels.
Given the limited computing capabilities and power availability of IoT devices, specific considerations must be addressed in implementing the proposed anti-jamming strategies. Techniques such as edge computing, low-complexity algorithms, power management, and collaborative efforts can be employed to overcome these limitations. The use of ML techniques has shown significant improvements over traditional approaches.
With appropriate implementation strategies, these techniques can be effectively employed even in resource-constrained IoT devices. The research findings contribute to the development of more robust and secure IoT networks, ensuring reliable wireless communication in the presence of malicious interference.

Date of Award	Aug 2023
Original language	American English
Supervisor	Sami Muhaidat (Supervisor)