Exploring a Multi-Tiered Whiteboard Infrastructure for Information Fusion in Wireless Sensor Networks

It is important for the life time of a wireless sensor network (WSN) to reduce the amount of data transferred through the network. As a typical approach, sensor data is filtered before propagating updates, to a node at the edge of a network, where it can be fused. Information Fusion inside the network can reduce the amount of data propagated, by fusing data before and in propagation, without losing the information value in it. We explore infrastructures for distributed fusion, with fusion nodes located at strategic nodes inside the network, as an approach of structured distributed fusion for WSNs. We propose an infrastructure for a white-board approach that uses a distributed real-time database with virtual full replication. With such an approach, both raw and fused data are logically available at all nodes and physically available where used, such that only used data will be propagated and use resources. The actual resource usage will be relative to the actual demand for data, rather than to the amount of data published at the white-board. We present an exploration of such an infrastructure, and points out future key research questions for such a white-board approach. I. INFORMATION FUSION IN WIRELESS SENSOR NETWORKS Wireless sensor networks (WSN) are resource constrained large-scale systems that typically collects data about its environment and sends the data to the edge of the network for analysis and exploration there. Data is often filtered, compressed or aggregated along the propagation path to the edge. A number of specific challenges need to be met for WSNs [1]: 1) Usually the sensors are battery operated, thus efficient use of the available energy supply is critical. Once the energy supply is depleted, the network must be recharged or abandoned. Transmissions require a relatively high energy cost, which means that delayed, coordinated and aggregated transmissions may save much of the energy and increase the lifetime of WSN nodes. 2) In comparison with local area networks, communication links in WSNs are several orders of magnitude slower. This gives rise to high latency, in particular for multi-hop transfers. 3) The connectivity is often unreliable, as communication paths may be lost or obstructed during operation, and the communication range of sensor nodes is irregular and often dependent on battery level. 4) Further, due to power, size and cost limitations, the memory at each node is usually very limited. WSNs are used for detection, recognition, identification, tracking change detection. They are suitable for data collection in typical information fusion systems, such as used for decision making support. Information Fusion (IF) approaches can support wireless sensor networks, in improving confidence of sensor readings by fusing complementary information. The service of a WSN may be improved by information fusion, such as correlation of readings, classification for signatures, and uncertainty management. Information fusion typically make use of multiple data sources to reduce uncertainty, for improved situation awareness. IF-approaches have been used in several WSN applications. In ALARM-NET [2], Dempster-Shafer evidential theory was used to classify sensor readings to associate readings with movements of people, and to create traces for several elderly individuals in an apartment. In the survey paper of Nakamura [3] there are several other examples of using Information Fusion for WSNs. II. DATA FILTERING AND REDUCTION For many applications, the fusion of information is often done at a central location, where all information to be fused must be collected. In case the fusion is done at a dedicated fusion node, in a distributed system with communication links between nodes, all information must be sent for fusion at that single node. WSNs have very unreliable communication links, and sensor readings may need to be sent redundantly to protect from losing sensor readings due to network loss. Processing and storing sensor data close to the sensor can avoid unreliable, and slow communication of WSN [4]. A central fusion node is a bottleneck and a single point of failure, and all nodes using the fused information need to rely on communication links to the fusion node, and the availability of the fusion node itself. Also, all information need to pass the single node, which not only needs resources enough to fuse the information, but must also be able to handle the amount of requests sent to and from it. Typically, there are only a few critical sensor events that must reach the network edge in bounded time, while large amounts of less time-critical and possibly aggregated data may reach the network edges at best effort. There exists several approaches for data aggregation and filtering in WSN for reducing the amount of communication and energy usage. Often data is filtered into application specific events, and typically sensors are explicitly programmed or use queries [5]. Queries may be declared at network edges and then distributed to sensors [6]. These are specific designs, and more generic approaches are desirable. Filtering the sensor updates at the source may reduce the information content of data, and subsequent fusion becomes less effective. However, reduction of the amount of data sent is essential for scalability and for reducing energy consumption. While reducing the amount of data to send, fusion must not suffer from information reduction. To try to overcome this, fusion processing can be located close to sensors instead. This allows data reduction by fusion, to preserve the information in the data to be propagated, and reduce resource usage while improving availability and fault tolerance of information fused. Distributed fusion need to be supported by an infrastructure for effective fusion, while reducing the resource usage, with the aim to allow fusion processing at multiple and concurrent nodes, and to increase reliability and scalability. In such exploration, the properties of a useful infrastructure for distributed fusion is searched for. To the best of our knowledge, no available data-centric scheme uses a distributed real-time database to meet several challenges in WSNs. III. A MULTI-TIERED ARCHITECTURE As an approach for fault tolerance and reliability in WSNs we propose to separate the network into multiple tiers: Groups of sensor nodes make up sub networks, where each sub network connects to a database node in a distributed (realtime) database, and where data from each sub network is published in the database through the database node connected to. Database nodes connect to each other and exchange updates. Database nodes in the networks can be organized in hierarchies for increasing levels of refinement of the information. Fusion is done at database nodes, and fused data is published in the database, in a white-board fashion. We propose to use a distributed real-time database with Virtual Full Replication (ViFuR) in such networks [7].