Power optimized estimation in wireless sensor networks

(eng) Spurred by ease of deployment provided by the wireless communication paradigm, wireless sensor networking is an emerging technology which finds application in many fields including military, environment and habitat monitoring, health care, and industrial automation, among others. A wireless sensor network (WSN) consists of spatially distributed sensors that cooperatively monitor physical conditions. The sensor nodes are usually powered by limited capacity batteries which limit sensing, communication, and computational functionalities of the individual sensors and thereby of the overall network. These intrinsic limitations of the sensors and the reason that sensors produce measurements which are correlated due to their spatial proximity, energy-efficient cooperative signal processing algorithms and communication protocols are required for information processing and dissemination in WSNs. The focus of this thesis is to develop and analyze power-aware and energyefficient data processing algorithms and communication strategies in the context of realizing WSNs with acceptable sensing capabilities and operational life. Specifically, the thesis aims to develop optimization techniques and algorithms to realize energy-efficient parameter estimation in WSNs with spatially correlated data, analog and digital modulation schemes, perfect and imperfect knowledge of the communication channels, and under different network topologies. Firstly, the thesis considers an estimation problem in which a WSN is deployed to observe a source where sensors send their observations to a remote fusion center (FC). Due to the spatial correlation of the sensor observations, the power allocation problem needs to be solved numerically. We show that using successive approximation approach, the solution to the power allocation problem can be given by an iterative water-filling type solution. For this network setting, the impact of channel estimation errors on the performance of the power allocation scheme is also analyzed. Secondly, the contribution of the thesis lies in the area of joint quantization and power allocation for owerconstrained estimation in WSNs with spatially correlated data. To this end, two quantization and transmission schemes are presented: one based on an optimal uniform quantization scheme and the other, an approximation, based on pseudo-quantization noise model. It has been shown that, compared to the former scheme, the latter scheme is computationally simpler, whereas the distortion performance of the two schemes matches quite well. Thirdly, noting that the centralized WSNtopology (all sensors directly send their observations to the FC) may not be optimal in the realm of energy-efficient estimation, we study the performance of power-constrained estimation in hierarchical WSNs. Finally, we study nonlinearities in discharging behavior of the sensor batteries and their effect on the network lifetime; and we also investigate the impact of degree of knowledge about the channel gains on the network lifetime.