• Keine Ergebnisse gefunden

52 CHAPTER 4. DEPLOYMENT ON SENSOR NODES ASEDT

subset or even no sensing capability. Such a lack of capabilities could be either by design or by failed sensing units. Hence, certain branches or subtrees of the EDT cannot be evaluated by the node itself. In that case, sensor nodes need to collaborate to exchange event information.

The exchange of sensed raw data, which is done by most approaches, is very inefficient from two points of view. First, permanent exchange of sensor readings leads to a huge number of transmissions and hence, consumes much energy and reduces network performance. Second, transmitting raw sensor data requires to use rather large data packages, depending on the number of readings and their accuracy, i.e., the size of each value usually varies from two to four bytes. Since a conceptual main goal is to remain very energy efficient, the focus is on minimizing the number of transmissions and the amount of data to be exchanged. Instead of exchanging raw sensor readings at each detection interval, sensor readings are locally processed first and finally one bit is submitted only, which is the Boolean value of a particular EDT-node. There already exist approaches that share information in a “yes” or “no” style, e.g., in [35], but these can only state the final complete detection result. This concept focusses on efficiently sharing information about both, complete and partial events.

In case of using EDT, the Boolean value of only one particular EDT-node has to be transferred. Missing node values may be delivered by neighbouring nodes that share the specified region of event. To prepare these data exchanges, every sensor node has to determine which EDT-node information is missing at the local EDT. Therefore the following algorithm prunes the established EDT until it contains the minimum required EDT for local event processing:

1. Mark each leave as pruned that represents an unsupported sensing capabil-ity.

2. Search all nodes that possess at least one marked child excluding the root node1.

2.1 Mark node as pruned, if

a It represents a mathematical operation or b The unmarked child represents a constant or c All child nodes are marked as pruned.

3. Repeat step 2 until no new nodes are marked. After that, all undecidable subtrees are marked.

4. Prune all marked nodes except for the root nodes of the marked subtrees.

5. Declare all left marked nodes as “undecidable”.

4.3. LOCAL ADAPTATION OFEDTs BY PRUNING 53

Figure 4.4: Pruned EDTs for two different types of sensor nodes monitoring the introduced fire event. Nodes of type A provide sensing facilities for carbon monoxide and temperature whereas nodes of type B provide sensing facilities for carbon monoxide and smoke. Consequently, each type of node prunes a certain part of theEDTthat cannot be evaluated locally. Resulting “undecidable” nodes are labelled with “?”. Hence, the Boolean values of these nodes must be obtained from other nodes in the specified region of event.

After pruning,EDTs may contain nodes, which are marked as “undecidable”.

Respective Boolean node values must be obtained by other nodes in the region of event. Let us assume to use two different types of nodes (A and B) for the introduced fire detection example. Nodes of type A provide carbon monoxide and temperature sensors whereas type B nodes provide sensing facilities for carbon monoxide and smoke. Hence, the initial EDTs generated at these nodes must be pruned with respect to available sensing capabilities.

Accordingly, type A nodes cut the branch containing the smoke readings and type B nodes respectively cut the branch containing the temperature readings.

That results in two different EDTs at the sensing devices, each containing one node marked as “undecidable”. Thus, type A nodes require information about tree node number 9 whereas type B nodes require information about tree node number 6. Both resulting EDTs are displayed in Figure 4.4. At regular evalu-ation, the EDT also checks the status of the sensing devices. In the case that sensing devices fail during application the sensor node runs the pruning again to locally self-adapt the EDT to the current situation. In addition, sensing de-vices may fail transiently only. In that case, the sensing device again is available and hence, the EDT can be reconstructed into its original form by removing the

“undecidable” marking from respective EDT-nodes.

By pruning, the EDT may degenerate to a minimal tree consisting of the root node with “undecidable” children. Such an EDT enables sensor nodes that possess no suitable sensing capability for event detection, to serve as a “bridge”.

1Since anEDTis a binary tree, every node possesses at most two child nodes. Hence, either one or both child nodes are marked as pruned in that case.

54 CHAPTER 4. DEPLOYMENT ON SENSOR NODES ASEDT

Figure 4.5: Example deployments that may require node 2 to serve as a bridge for the nodes 1 and 3. In (a) the nodes 1 and 3 do not share an event region due to their distance . In (b) these nodes share their regions indeed, but cannot communicate directly due to an obstacle between them.

These nodes are of interest if they are located between two or more nodes that possess the required sensing capabilities but cannot communicate directly or do not share the same event region. The only prerequisite is that these nodes share the event region of the bridge node. Figure 4.5displays example deployments for both cases. Here node2 shares its event region with the nodes1 and3 and hence, may perform the bridging functionality for these nodes. In Figure4.5(a) the nodes 1 and 3 do not share an event region due to their distance. In Figure 4.5(b) these nodes share their regions indeed, but cannot communicate directly due to an obstacle between them. In such a scenario all participating nodes deliver their parts of event information to the bridge node, which is finally enabled to decide about the occurrence of that event. After having identified the “undecidable”

parts for event detection on each sensor node, those have to efficiently share necessary information. A suitable collaboration scheme maintaining this data exchange is presented in the next Section.