Bachelor and Master Theses

Background

Wireless communication networks play an important role in modern industrial systems, Internet of Things (IoT) infrastructures, and connected applications. Although Wi-Fi security standards have evolved considerably, denial-of-service attacks targeting medium access mechanisms continue to present significant challenges.

One category of such attacks exploits Clear-to-Send (CTS) control frames and the Network Allocation Vector (NAV) mechanism. By manipulating CTS transmissions, an attacker can cause legitimate wireless devices to defer channel access, resulting in degraded network performance or disrupted communication.

Several detection and mitigation approaches have been proposed to address wireless denial-of-service attacks. These include traffic monitoring, statistical anomaly detection, threshold-based mechanisms, and NAV validation techniques.

However, many conventional defenses are designed to identify predefined attack patterns. Their effectiveness against intelligent attackers who continuously adapt their behavior in response to defensive mechanisms is less understood.

Reinforcement Learning (RL) provides a framework through which an agent can learn effective strategies by interacting with an environment and observing the consequences of its actions.

This creates opportunities to investigate whether an adaptive attacker can discover weaknesses in existing CTS attack defense mechanisms and how such findings can support the development of more robust wireless security solutions.

This thesis investigates the use of reinforcement learning to systematically evaluate the robustness of CTS attack detection and mitigation mechanisms in Wi-Fi networks.

Problem Description

Traditional CTS-based denial-of-service attacks generally use predefined transmission strategies, such as constant-rate flooding or fixed-duration forged CTS frames.

Such behavior may be identified by intrusion detection mechanisms that monitor CTS transmission rates, frame durations, or abnormal traffic characteristics.

However, an adaptive attacker may change its behavior based on feedback from the wireless environment and the observed response of the defense system.

This raises several important questions:

1. Can reinforcement learning identify weaknesses in existing CTS attack detection mechanisms?

2. Can an adaptive attack strategy reveal vulnerabilities that conventional attack patterns fail to expose?

3. How do different defensive strategies respond to previously unseen CTS traffic patterns?

4. Do detection mechanisms trained against fixed attack patterns remain effective against adaptive adversaries?

5. How can the identified weaknesses be used to improve wireless intrusion detection and prevention systems?

A key challenge is establishing a reproducible experimental framework for evaluating both the adaptive attacker and the targeted defensive mechanisms.

Aim

The aim of this thesis is to investigate whether reinforcement learning can be used to develop adaptive CTS-based attack strategies that expose limitations in Wi-Fi intrusion detection and prevention mechanisms.

The thesis will evaluate the robustness of existing reproducible defense approaches against adaptive adversarial behavior and investigate how the findings can support the design of more resilient wireless security systems.

Research Questions

1. How can CTS-based attack behavior and defensive responses be modeled as a reinforcement learning problem?

2. Can a reinforcement learning agent identify CTS traffic patterns that expose weaknesses in conventional detection mechanisms?

3. How does an RL-based adaptive attack compare with fixed-rate, burst-based, and randomized CTS attack strategies?

4. How do different CTS detection mechanisms perform against adaptive attack patterns?

5. Which improvements to existing defense mechanisms can increase robustness against reinforcement learning-based adversarial behavior?

Methodology

The student will:

1. Conduct a literature review of CTS-based attacks, wireless intrusion detection, adversarial machine learning, and reinforcement learning.

2. Select and reproduce representative CTS attack detection mechanisms from the research literature, including rule-based and statistical approaches.

3. Develop a controlled experimental environment containing a Wi-Fi network, a monitoring component, and the selected defense mechanisms.

4. Formulate the attacker-defense interaction as a reinforcement learning problem, defining relevant observations, actions, and reward signals.

5. Implement and train an RL agent to explore CTS attack strategies within a controlled simulation or isolated laboratory environment.

6. Compare the learned strategies against conventional, non-adaptive CTS attack patterns.

7. Evaluate how different defense mechanisms respond to adaptive attack behavior.

8. Investigate possible defensive improvements based on the weaknesses identified during evaluation.

9. Validate the findings using reproducible experiments and, where feasible, commodity Wi-Fi equipment.

The study will focus on authorized laboratory testing and defensive robustness evaluation. Experiments will be designed to prevent interference with external wireless networks.

Evaluation Metrics

Attack Effectiveness Metrics

• Network Throughput Degradation

• Packet Loss

• Communication Latency

• Channel Access Delay

• Attack Success Rate under predefined experimental conditions

Defense Evaluation Metrics

• Detection Rate

• False Positive Rate

• Detection Latency

• F1-score

• Robustness against previously unseen attack patterns

Reinforcement Learning Metrics

• Cumulative Reward

• Learning Convergence

• Training Efficiency

• Policy Generalization

• Performance across different defensive configurations

Comparative Evaluation

• RL-based versus fixed-rate attack strategies

• RL-based versus randomized attack strategies

• Performance against different detection algorithms

• Generalization to previously unseen defense configurations

Expected Outcomes

The thesis is expected to:

• Develop a reproducible framework for evaluating CTS-based attacks against wireless intrusion detection mechanisms.

• Implement an adaptive CTS attack agent using reinforcement learning.

• Investigate whether reinforcement learning reveals weaknesses missed by conventional attack strategies.

• Compare the robustness of multiple existing detection approaches.

• Identify limitations of conventional CTS attack detection mechanisms.

• Investigate improvements that increase detection robustness against adaptive adversaries.

• Provide recommendations for developing more resilient Wi-Fi intrusion detection and prevention systems.

• Contribute experimental findings and reusable research software that may support further research in adversarial wireless security.

Knowledge of Linux, computer networking, cybersecurity, and Python programming is beneficial.

Experience with machine learning or reinforcement learning is an advantage.

Familiarity with Wi-Fi protocols, wireless network security, and network traffic analysis is also beneficial.

The thesis will be conducted using open-source software, reproducible defense implementations, and a controlled experimental environment. Access to commercial wireless intrusion prevention systems is not required.

The project may be coordinated with a related thesis focused on developing CTS attack detection mechanisms for the Kismet platform.