Functional Architecture for Sensor Networks

Compared to conventional distributed databases in which information is distributed across several sites, the number of sites in a sensor network equals the number of sensor nodes, and the information collected by each node (e.g., sensor readings) becomes an inherent attribute of that node [9]. To support energy-efficient and scalable operations, sensor nodes could be autonomously clustered. Furthermore, the data-centric nature of sensor information makes it more effectively accessible via an attribute-based naming approach instead of explicit addresses [10]. In addition, as these sensors are integrated into and extract information from physical environments, many applications also require the location information to be passed along with their sensor data. As a result, a generic functional architecture for sensor networks consists of the following components:

Hierarchical clustering. To facilitate scalable operations within sensor networks, sensor nodes could be aggregated to form clusters based on their energy levels and proximity. The aggregation process could also be recursively applied to form a hierarchy of clusters (Figure 8.2). Within a cluster, a cluster head will be elected to perform information filtering, fusion, and aggregation, such as periodic calculation of the average temperature of the cluster coverage area. In addition, the clustering process should be reinitiated in case the cluster head fails or runs low in battery energy. In situations in which a hierarchy of clusters is not applicable, the system of sensor nodes is perceived by applications as a one-level clustering structure in which each node is a cluster head by itself. The clustering algorithm introduced by Estrin and colleagues [10] allows sensor nodes automatically to form clusters, elect and re-elect cluster heads, and reorganize the clustering structure if necessary.

Location awareness. Because sensor nodes are operating in physical environments, knowledge about their physical locations becomes mandatory. Location information can be obtained via several methods. Global positioning system (GPS) is one of the mechanisms that provide absolute location information. For economical reasons, however, only a subset of sensor nodes may be equipped with GPS receivers and function as location references by periodically transmitting a beacon signal telling their own location information so that other sensor nodes without GPS receivers can roughly determine their position in the terrain. Other techniques for obtaining location information are also available. For example, optical trackers [11] give high-precision and -resolution location information but are only effective in a small region.

FIGURE 8.2 Clustering and a cluster hierarchy. (From Shen, et al., IEEE Personal Commun. Mag., 8(4), 52-59, 2001. With permission.)

Attribute-based naming. With the large population of sensor nodes, it may be impractical to pay attention to each individual node. Users would be more interested in querying which area has temperature higher than 100°F or what the average temperature in a specific area is, rather than the temperature at sensor ID#101. To facilitate the data-centric characteristics of sensor queries, attribute-based naming is the preferred scheme [10]. For instance, the name [type=temperature, location=N-E, temperature=103] addresses all the temperature sensors located at the northeast quadrant with a temperature reading of 103°F. These sensors will reply to the query, "which area has temperature higher than 100°F?" Note that not only can physical or location attributes be part of a name, but so can logical attributes such as unique IDs, temporary variables, and clustering roles (e.g., cluster head or cluster member). Therefore, the traditional addressing scheme using node IDs becomes a special case of attribute-based naming.

With the integration of these three components, the following two sample queries may be effectively and efficiently carried out.

• Which area has temperature higher than 100°F? In theory, the query is broadcast to and evaluated by every node in the network. Despite possibly the best returned result, the query would suffer from long response time. In practice, each cluster head may periodically update the temperature readings of its members, and the query can now be multicast to and evaluated by cluster heads only. This results in better response time at the expense of less accurate answers. Queries under stringent timing constraints can be evaluated by cluster heads of a higher tier.

• What is the average temperature in the southeast quadrant? Similarly, the average temperature of each cluster can be periodically updated and cached by cluster heads. Furthermore, the query should be delivered to nodes located (named) in the southeast quadrant only.

Was this article helpful?

0 0

Post a comment