Nowadays, Wireless Sensor Networks (WSN) have been widely used in environmental monitoring applications because sensors are cheap and portable. These tiny distributed sensors provide discrete samples of environmental parameters; for example, temperature, humidity, gas pressure, decibel levels, and so on. Since sensors are always constrained by their limited battery power, how to efficiently use a large number of distributed sensors and their samples in an application with the most minimal expenditure of energy is of great importance. In this thesis we have presented a series of intelligent sampling approaches. It is worthwhile to point out that sampling in sensor networks has many interesting properties. First, sensor sampling has two dimensions. The temporal dimension decides how many...[ Read more ]

Nowadays, Wireless Sensor Networks (WSN) have been widely used in environmental monitoring applications because sensors are cheap and portable. These tiny distributed sensors provide discrete samples of environmental parameters; for example, temperature, humidity, gas pressure, decibel levels, and so on. Since sensors are always constrained by their limited battery power, how to efficiently use a large number of distributed sensors and their samples in an application with the most minimal expenditure of energy is of great importance. In this thesis we have presented a series of intelligent sampling approaches. It is worthwhile to point out that sampling in sensor networks has many interesting properties. First, sensor sampling has two dimensions. The temporal dimension decides how many samples a sensor should obtain. In our sampling approaches, the temporal sampling is used to adjust the sensor sampling rates and provide the required data quality under a noisy environment. The spatial dimension selects a subset of sensors in a distributed fashion to save sampling and transmission costs. We also find that sensor sampling depends on applications. Different applications (e.g. different queries, data cleaning, pattern search) usually require different approaches to enable the optimal use of the samples and sensors. Even different scenarios of an application (e.g. in a pattern search application, the sensory data may have or have no spatial similarity) affect the design of a sampling approach. This thesis includes three main parts: (1) intelligent sampling for summary and range queries, (2) intelligent sampling for a data cleaning application, and (3) intelligent sampling for more complex sensor queries, such as pattern query over distributed sensory streams, and max regional aggregate query. Our extensive simulation results demonstrate the effectiveness and efficiency of the proposed sampling approaches in different applications.