Image Processing Reference
In-Depth Information
7.3.4 Constrained Anisotropic Diffusion Routing Protocol
In [Chu] two techniques for querying and routing in WSNs are introduced, i.e., information-driven
sensor querying (IDSQ) and CADR. The problem addressed here is how to perform queries and
route data maximizing the information gain while minimizing latency and bandwidth utilization.
The main idea of IDSQ/CADR is to introduce an information utility measure based on the estima-
tion theory that models the information content and the spatial configuration of a network. In this
way,anodecanspreadaquerybasedontheevaluationofanobjectivefunctionthatconsidersboth
information and cost, and it can forward data based on the local gradient of the objective function.
IDSQ/CADR can be viewed as a generalization of the diffusion approach, in which both information
gain and communication cost direct data diffusion. In particular, IDSQ is a sensor selection algo-
rithm that aims to find the optimal order of sensor queries that provide maximum information gain
while balancing energy cost. he CADR protocol, on the other hand, determines the optimal routing
path from the querying to the queried sensor (through the gradient of the objective function). his
approach allows sensors to send packets only when there is interesting data to report. Moreover, only
the part of the network with a better information/cost trade-off may be active. So this approach is
more energy-efficient than Directed Difusion, where queries are diffused in an isotropic fashion with
(controlled) flooding over the entire network. Also, compared with approaches that only minimize
the energy consumption for a single path, CADR achieves better performance, as both energy cost
and information gain are taken into account through an appropriate utility function.
7.3.5 Cougar Protocol
In [Yao] a data-centric protocol that models WSNs as a distributed database is presented. The
Cougar protocol aims to define sensor tasks through declarative queries.
In order to achieve declarative queries in sensor networks, nodes have to implement a query layer
between the network and application layers, which basically consists of a query proxy that performs
in-network processing. In fact, unlike typical WSN applications, where data is forwarded and then
analyzed, the Cougar approach moves part of the data analysis inside the WSN. In this way it is
possibletoreducetheamountofdatatobetransmitted.SinceforWSNnodeslocalcomputations
generally require less energy than data transmissions, this approach achieves better energy efficiency
than centralized data extraction.
A query optimizer is located in the sink node, which produces an efficient query plan that reduces
resource usage and so extends network lifetime. In order to generate a good plan, network conditions
have to be known. For this purpose, a catalog can be created at the server, which maintains the useful
information (that needs to be updated as the network parameters change). In addition to the network
condition, the optimizer also considers existing query workload and tries to merge similar queries.
A special node, the leader, is elected. his is the node where the computations of the aggregate values
will take place. hen two query plans are generated, one for the leader and one for the other nodes,
i.e., nodes send the leader local data from sensor or partially aggregated data, and the leader waits
untiltherequireddataisreceivedandthensendstheoverallaggregatedvalues.hisapproachcanbe
used for long-term queries, for example the average temperature value can be monitored for a long
timeandonlyupdatesforsigniicantchangescouldbetransmitted.hankstosuchadataaggrega-
tion, the Cougar approach is efficient for bandwidth and energy management, although it also has
some drawbacks. For example, as a sensor has to wait to receive results to be aggregated in order
to perform efficient data aggregation, it requires synchronization between sensor nodes along the
communication path. his could be a problem because with high loss rates broken links may be hard
to distinguish from long delays. In addition, this protocol requires a catalog that has to be updated
at every change in the network (i.e., sensor position, connectivity, workload, etc.), and in a network
with several thousands of nodes this is not a simple task.
