We explore three applications of decentralized networks in increasingly complex detection frameworks: unconstrained binary detection in computer networks, constrained binary detection in wireless sensor networks (WSNs), and M-ary detection in wireless body area networks (WBANs). In each of these applications, the goal is to investigate the interactions between the estimation and detection components, as well as any constrained resources.;In the first application, parametric models for anomaly detection in computer networks are developed, wherein only aggregate traffic statistics are analyzed in order to detect network anomalies. Both synthetic and real attacks are used to validate our methods, and we find that simulated attacks as low as 12% can be detected in live traffic in just a few seconds. The effect of the smart adversary, wherein attack packet-sizes are chosen to match a percentage of background traffic packet-sizes, is investigated; the proposed method is able to detect attacks with up to 71% matched traffic.;The optimal allocation of transmission power, for the estimation of a scalar parameter in a wireless sensor network, given an overall power budget constraint, is then considered. The simple star topology is first considered, after which the results are extended to the branch, tree and linear topologies, and finally these topologies' applicability to the generalized topology case is presented. The optimal allocation policies are a function of measurement and channel noises, and evolve from sensor selection, to waterfilling, and finally to channel equalization.;The third application is activity-detection via heterogeneous sensors in a wireless body-area network. In particular, the number of samples allocated to each sensor is optimized to minimize the probability of misclassification. Using experimental data from overweight adolescent subjects, it is found that allocating a greater proportion of samples to sensors which better discriminate between certain activity-levels can result in either a lower probability of error or energy-savings ranging from 18% to 22%, in comparison to equal allocation of samples. The current activity of the subject and the performance requirements do not significantly affect the optimal allocation, but employing personalized models results in improved energy-efficiency. Furthermore, an alternate vector optimization is derived which significantly reduces the computational complexity of the original combinatorial optimization.;Finally, future directions for the work described in this dissertation are presented.
展开▼