In-network Processing
With the development of sensor networking hardware, the ability to deploy large numbers of disposable sensors with in-network processing capabilities is becoming possible. These sensor nodes deployed in a field of interest observe the phenomenon at different locations and forward data to a central data sink. The observed phenomenon usually has spatial and temporal correlation data, which can be explored to minimize wireless communication and energy cost and prioritize the important data, leading to considerable improvement in the quality of data representation in the network.
Distributed data compression and correlated source coding
In view of the data reduction in sensor nodes, the efficacy of distributed data compression is a function of the degree of correlated information collected from each individual node. Extensive distributed compression and source coding are explored to achieve it. They fall into two categories. First, the approach operates at the right trade-off point between local processing and transmission costs by different clustering schemes. These methods combine the correlated source coding with the routing scheme to form a joint advantage in the data aggregation, including compression-driven routing and ruoting-driven compression. Second, they try to minimize the joint description costs of nodes by cutting off the amount the inter-node communication using correlated side information only from source nodes, such as Slepian-Wolf coding and DISCUS (Distributed Source Coding Using Syndromes) method. Distributed source codes compress multiple data sources by leveraging the use of a joint decoder, even without access to the realization of each source. Our objective is to find practical coding techniques that can approach the Slepian-Wolf achievable rate region and would be effective for different types of correlation.
Collaborative event detection based on correlated sources
Event detection is essentially important for wireless sensor network applications, including target tracking, environment monitoring and surveillance. Event detection can observe the severe change and predefined event in the field of interests, like crater activity in vocano, bridge vibration and forest fire situation. It can also serve as an system alert report for the maintenance purpose, such as energy depletion, radio failure and node dysfunction. By detecting the event, we can increase the sampling rate to improve the data quality and prioritize the event data to guarantee their delivery, facilitating the scientific analysis. However, noise in the real world make the event data hard to be recognized by the on-single-node processing. Noise may decrease the quality of data, or even overwhelm it.
Collaborative source detection utilizes the correlation information from seperate nodes located in the close range area. The goal is to detect the event signal embedded in the noise, which can hardly be detected by the single event detection algorithm. Collaborative event detection can not only "dig out" the unobservable event for individual nodes, but also facilitate sensor network placement. The sensors with collaborative event detection can be efficiently deployed in a cluster, with cluster header or one of them equipped by the good quality sensor. Take time synchronization as a comparison, each node synchronizes to a global time root to make sure they do the sampling in the same time slot. And in the same way, the event detection inside the cluster nodes can be expected as "synchronized".
Faculty
Students
Publications
Aaron Kiely, Mingsen Xu, Wen-Zhan Song, Renjie Huang, Behrooz Shirazi
Adaptive Linear Filtering Compression on Realtime Sensor Networks
Seventh Annual IEEE International Conference on Pervasive Computing and Communications (IEEE PerCom 2009 accept ratio 17%)