Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Pramod Varshney


detection and estimation;signal processing


Over the years, wireless communication systems have evolved into the most widely used framework for communication devices and networks. These wireless networks form the backbone of wireless sensor networks (WSNs) that have been employed in many applications such as military surveillance, autonomous driving systems and smart-homes. When designing WSNs, two crucial factors must be considered. The first factor is the security of WSNs, which is a concern due to the deployment of low-cost and potentially insecure sensors. The second critical factor is the energy efficiency of WSNs, as they often rely on battery-limited sensors. In this dissertation, we consider the design of various resilient energy-efficient WSNs for the inference task under Byzantine attacks, which are one of the most significant security threats faced by WSNs. When the system suffers from Byzantine attacks, some sensors in the network might be compromised and fully controlled by adversaries. Our goal is to design WSNs that are both energy-efficient and resilient. The first part of this dissertation (Chapters 2 and 3) focuses on enhancing the resilience of WSNs that achieve energy-efficiency through quantization, particularly in scenarios where Byzantine nodes are prevalent, and the fusion center (FC) lacks knowledge of the attack strategy. Our in-depth exploration, analysis, and enhancements center around a promising energy-efficient mechanism known as the audit bit-based mechanism. For the traditional audit bit-based mechanism, we demonstrate how a simple attack strategy can compromise the entire system. To address this concern, we introduce an enhanced audit bit based mechanism, which effectively relaxes the stringent constraints on the attack strategies that this mechanism can withstand. Building upon the enhanced audit bit framework, we propose an advanced audit bit-based scheme that not only improves system robustness but also significantly reduces redundancy related to audit bits. Furthermore, drawing inspiration from both the audit bit-based mechanism and reputation-based mechanisms, we develop some advanced schemes designed to help systems effectively address challenges in scenarios where prior knowledge of attack strategies is unavailable, and Byzantine nodes are a prevailing threat. In the next section of this dissertation (Chapters 4 and 5), we study the resilience of WSNs operating under constraints of limited power supply. Our research focuses on the security aspects of two promising energy-efficient frameworks: ordered transmission (Chapter 4) and compressed sensing (Chapter 5). In Chapter 4, we investigate the impact of Byzantine attacks on the performance of both the traditional order transmission based (OT-based) system and the communication-efficient OT-based (CEOT-based) system. We investigate the error probability and the number of saved transmissions for those OT-based systems under various Byzantine attack strategies. Furthermore, we derive upper and lower bounds on the number of transmissions saved for OT-based systems under various Byzantine attack strategies. A comparison of the resilience of CEOT-based and conventional OT-based systems is presented, offering guidance on implementing OT-based frameworks in potentially hostile environments. In Chapter 5, we investigate the distributed detection problem of sparse stochastic signals with compressed measurements in the presence of Byzantine attacks. We propose two robust detectors based on the traditional Generalized Likelihood Ratio Test (GLRT) and traditional Quantized Locally Most Powerful Test (LMPT) detectors with adaptive thresholds, given that the sparsity degree and the attack strategy are unknown. The proposed detectors can achieve detection performance close to the benchmark likelihood ratio test (LRT) detector with perfect knowledge of the attack strategy and sparsity degree. Furthermore, we explore situations where the fraction of Byzantines in the networks is assumed to be known. In this context, two enhanced detectors building on the previous proposed robust detectors are proposed to further improve the detection performance of the system by filtering out potential malicious sensors. In addition to our primary focus on traditional WSNs, our research extends to the domain of human-machine collaborative networks. These networks are particularly relevant in high-stake scenarios, such as remote sensing and emergency access systems, where automatic physical sensor-only decision-making may not be sufficient. A combination of human and machine inference networks leverages the cognitive strengths of humans and the sensing capabilities of sensors to enhance situational awareness of the systems. Chapter 6 introduces a belief-updating scheme designed to enhance the resilience of these collaborative networks against potential attacks. The proposed belief-updating scheme, which builds on a human-machine hierarchical network, can also mimic the real-world decision-making process where the sensors' local decisions are collected by humans to make a final decision. Our research reveals that our proposed scheme can improve system performance, even in scenarios where a significant fraction of physical sensors in the system are compromised, and where knowledge about the exact fraction of malicious physical sensors is lacking. Additionally, we conduct an analysis of the impact of side information from individual human sensors, and compare different operations used to incorporate the side information.


Open Access