Date of Award

8-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Engineering

Committee Chair/Advisor

Linke Guo

Committee Member

Richard Brooks

Committee Member

Fatemeh Afghah

Committee Member

Long Cheng

Abstract

The rapid development of hardware has made wireless devices smaller and easier to embed in everyday objects. The actions performed by devices have shifted from manual control by humans to automatic determination by built-in algorithms. To achieve an efficient and accurate decision-making process, devices have grown in their ability to actively sense their surroundings and communicate with other devices without human intervention. However, devices using different protocols create a heterogeneous environment. Future wireless networks are increasingly challenged by dense heterogeneous deployments, dynamic spectrum contention, and growing security threats. Traditional communication architectures are built on a passive, reactive approach to the environment, lacking the environmental awareness and adaptive intelligence required to operate efficiently in a complex Radio Frequency (RF) environment. As wireless systems increasingly integrate machine learning and autonomous decision-making, new frameworks are needed to enable real-time sensing, intelligent coordination, and resilient operation. This dissertation advances the design of resilient, awareness-driven wireless systems capable of adaptive sensing, intelligent coordination, and robust real-world applications.

Due to the dynamic IoT environment, wireless transmission among heterogeneous IoT devices is unpredictable and notoriously difficult to manage, resulting in network performance that is always compromised. To address this issue, we leverage RF sensing data to develop a deep learning-based framework for mitigating cross-technology interference (CTI) in a Wi-Fi-based IoT system. Each IoT device can provide detailed Channel State Information (CSI) of the transmission link, by which the gateway uses Generative Adversarial Networks (GAN) for building a detailed environmental RF map. Then, we propose a CSI-inspired gateway topology management strategy to determine the optimal gateway location, aiming to achieve the highest network throughput. Since the CSI is highly location-dependent, the optimal gateway location with the highest PDR can be further derived. We collected more than 2x10^8 CSI samples from 128 locations in a 1,000 m^2 area for validation. Extensive experimental results have shown that our design successfully predicts 8.99% area with 88% PDR and 1.94% area with 95% PDR, matching the ground truth with an accuracy of 96.63%.

When deploying heterogeneous IoT devices with different wireless protocols in a limited geographic area, e.g., manufacturing warehouses and clinic rooms, inevitable packet collisions will occur due to spectrum overlap. Those unpredictable collisions will ultimately degrade network performance, mainly due to the lack of coordination across coexisting protocols. To solve this problem, we revisit the classic resource orchestration problem in a practical wireless coexistence scenario with a dense indoor IoT deployment. We propose leveraging multi-protocol gateways, e.g., Amazon Echo, Google Nest Hub, and Samsung SmartThings Station, to develop a Multi-Agent Reinforcement Learning (MARL) framework that jointly considers channel status and contextual information to orchestrate limited resources. Given protocol heterogeneity and diverse transmission requests, we design a novel resource pool to enable fine-grained management of available resources, enabling gateways to collaboratively decide the system-level optimal strategy. The proposed design will also feature a cascaded RL model to determine a sequential decision for best utilizing available resources. Based on extensive real-world experiments conducted on a Software-Defined Radio (SDR) platform with up to 33 IoT devices, our proposed framework achieves more than 2.19x in throughput. It reduces the delay by 69.07% compared with current random-access mechanisms.

Advanced wireless communication systems employ deep learning (DL) techniques to enable automatic modulation recognition (AMR) for spectrum monitoring and management, particularly in bands supporting diverse coexisting wireless protocols. In practical wireless environments, wireless signals can be easily compromised by malicious noise, intentional interference, and adversarial attacks, thereby reducing the effectiveness of AMR. By exploiting DL model vulnerabilities, an undetectable perturbation added to the wireless signal can cause misclassification, resulting in serious consequences, including decoding errors, throughput degradation, and communication disruption. To address the limitations of existing approaches to defending against wireless adversarial attacks, we propose a Transformer-based AMR that leverages temporal correlation in time-series wireless signals. Instead of directly applying the Vision Transformer (ViT), we first innovate a feature extraction module specifically for RF signals from both the time and frequency domains, together with an adaptive positional embedding to the Transformer encoder for enhancing AMR accuracy. To mitigate noise effects in practical wireless communication, we propose a noise-adaptive adversarial training scheme for the developed Transformer-based model using adversarial examples crafted by white-box attackers. To demonstrate the scheme’s efficiency, effectiveness, and robustness, our proposed design has been thoroughly evaluated using a self-collected real-world dataset comprising over 30 million wireless signal samples across 21 modulation schemes in both indoor and outdoor scenarios. Our results reach a maximum accuracy of 94.17% in AMR classification and 71.2% under adversarial attacks.

Collectively, this dissertation advances the development of resilient and awareness-driven wireless systems through integrated RF sensing, intelligent orchestration, AI-driven signal understanding, and practical adversarial robustness evaluation in real-world wireless environments.

Author ORCID Identifier

0009-0007-5559-6621

Download

Available for download on Tuesday, August 31, 2027

Share

COinS