Wireless Sensor Network Abstraction – IDSQ
In a nutshell, IDSQ chooses a node as the “leader” which then takes available information and develops a “belief system.” From this it determines which node might be the next best one to investigate (say a node it believes is closer to the measurement to be made), and then passes its information to that node and declares it to be the new leader. The new node repeats the process. The first node goes back into an idle state.
The key to a robust sensor network is the right tradeoff between performance and scalability. One could take all the data from the network and centralize it, but this would not scale up. One way to handle large networks of nodes is to create groups of nodes and then deal only with those nodes. For object tracking a geographical grouping of nodes makes sense.
The algorithm is as follows:
1. The nodes sit in idle mode but wake up to sense any change in the environment.
2. If a change is detected then a leader node is elected (the one with the best sense of the change detected).
3. The leader node creates a “belief state” which contains the best known information at the time.
4. The leader node creates a group of nodes to collaborate with and disables other nodes from becoming leader.
5. The leader node propagates the belief state to the next best node and passes “leadership” status to it.
Since most nodes sit in an idle state making occasional detections, this state must be energy-efficient for the nodes.
This technique is not without its challenges. Some networks may elect multiple leader nodes as the information propagates throughout the system. Through a series of messages, a leader node can try and suppress other groups from forming.
This technique works well with object tracking because the nature of the application focuses on a subset of the nodes in a group. As the object moves through the network, the “leader” node can pass its information along to other nodes without having to rely on a centralized repository of information.
It’s not clear from the research how efficient this mechanism is or how well it performs in a real-world scenario. In this paper the authors apply statistical models to compensate for common errors in sensor networks using IDSQ techniques. By using probabilistic models, they hope to reduce the amount of data acquisition the network must perform.