Power-Efficient Distributed Computing and Data Processing in Wireless Sensor Networks
thesisposted on 21.10.2015 by Xi Xu
In order to distinguish essays and pre-prints from academic theses, we have a separate category. These are often much longer text based documents than a paper.
Wireless Sensor Networks are broadly applicable in many areas including: measurements of physical phenomena (temperature, pressure, humidity, hazardous material); transportation (traffic congestion, bridge structure);industrial control, assembly line control and diagnostics; and military applications (task tracking, event detection).The sensors used in WSNs are usually severely constrained in available energy while they may participate in energy-consuming tasks of data communication, aggregation, and interpretation. Power efficiency is one of the key objective that all data gathering and distributed computing tasks aim to achieve. On the other hand, advances in semiconductor technology enables ever more computing power for sensors.Clever utilization of the computational capability of sensors for in-network processing is essential for reducing the communication cost caused by data transmission redundancy in most WSN applications. It should be noted that compared with communication cost, the cost of the computation is almost negligible. The objective in most applications hence is to balance tradeoffs between computation and communication efficiency and the attendant power dissipation. Recognizing this, our research targets strategies to achieve the valuable information or data of interest over WSNs in a distributed fashion by factoring in the power efficiency. We explore power efficient solutions for well-known signal processing tasks such as data gathering and Fourier analysis. We further investigate the adaptive use of compressive sensing to achieve better computation and communication performance in such tasks.We first investigate efficient Fourier analysis of data field based on the randomly distributed sensors in a network and propose to implement Nonuniform DFT (NDFT) algorithm for the data measured from the field.We present a novel structure to realize NDFT implemented on WSNs and also propose an original algorithm with judicious design of computation and communication to reduce the required energy consumption along the data routing path. In our next study, we investigate a power-aware data collection scheme --- Hierarchical Data Aggregation using Compressive Sensing (HDACS) for large scale dense wireless sensor networks. We incorporate compressive sensing (CS) in a multi-level data aggregation hierarchy to shrink data volume for transmission and demonstrate how it works more efficiently and effectively when it is implemented in the hierarchical data gathering structure. We also explore the use of CS and HDACS for efficient and distributed/collaborative computing of NDFT in WSNs.In contrast to the existing state of the art, the methods investigated in this dissertation show significant improvement in terms of execution time, transmission power efficiency, SNR and packet collision phenomenon. Most of the existing CS-based data aggregation schemes for WSNs rely on the ideal assumption that data field are globally smooth, thus they fail to work if the data field is non-smooth. We present an adaptive data aggregation scheme referred to as Adaptive Hierarchical Data Aggregation using Compressive Sensing (A-HDACS) to perform data aggregation in non-smooth multimodal data fields. Finally, for the spatio-temporal data fields, we observe that the existing power-efficient CS-based data aggregation schemes for WSNs either remove data communication redundancy in the routing path or remove the temporal data redundancy by lowering the sampling rate at each sensor.We introduce Spatio-Temporal Hierarchical Data Aggregation scheme using Compressive Sensing (ST-HDACS) to overcome these shortcomings. ST-HDACS simultaneously incorporates the spatial and temporal data redundancies, and formulates the solution by taking advantage of HDACS-based scheme as well as MC.