A voting median base algorithm for approximate performance monitoring of wireless sensor networks

Wireless Sensor Networks (WSNs) are expected to be a new, revolutionary technology in the same manner as the Internet. This is due to their special characteristics such as low power consumption, ad hoc operation, self-maintenance and many other features. These special characteristics help in reducing the costs of network manufacture and implementation which extends their applications in a number of areas such as health and military services. Unfortunately, network resources such as memory, power and processing capacity constitute a serious constraint. In addition, they reduce the immunity of the network against external and internal impacts (such as electromagnetic interference) which make sensor node operations frequently deviate from the norm, degrading the WSN's functionality. In some cases the data collected by the network becomes unreliable; the monitoring of the phenomenon may even fail. To ensure the reliability of the network, several tools have been proposed to detect and isolate these deviations but most use relatively high levels of resources. In certain circumstances these state-of-the-art tools are unable to avoid the instant impact of data deviations on the accuracy of the collected data and on the network's functionality. This thesis overcomes these drawbacks by proposing a new, real-time, low resources usage, distributed performance algorithm that will monitor the accuracy of collected data and network functionality in large scale dense deployed WSNs. In order to achieve this, we have used the spatio-temporal correlation between the measurements of the neighbour nodes in large scale dense deployed WSNs. This correlation arises due to near proximity (of the nodes) and/or the slow characteristics' change of monitored phenomenon. The proposed algorithm has been tested via simulation experiments using different simulated and real world application data sets. Moreover, it has been tested on a real network testbed with Mote sensors using continuous reporting and event-driven applications. The results from these experiments showed a high rate of detection of changes in the reliability levels of data and in network performance. They also showed a high level of accuracy in terms of the detection of sensor faults. This, however, comes alongside certain limitations because of the use of simple passive analysis with the proposed algorithm