Application-Aware Routing Costs for Wireless Sensor Networks

The scheduling of sensor activity in wireless sensor networks so that redundancy is reduced but fidelity constraints are met has become an important issue. Because sensor networks are often deployed in random distributions, node distribution and data redundancy in some regions of the network may be lower than in others. Sensors in these regions are required to remain active more often than sensors in denser regions and consume more energy in transmitting their own locally generated traffic. Efforts must be made to balance, as well as reduce, energy consumption among the nodes within the network. For this reason, sensors in the more densely deployed regions should be considered more favorable as candidates to route the traffic of other nodes in the network. In this work, we propose several routing costs that allow traffic to be routed around the sparsely deployed regions so that the coverage of the environment can remain high for a long lifetime.

In some applications, it may be critical that the entirety of the region being monitored is covered as long as possible. In other words, the utility of the application drops significantly as the coverage falls from 100% to just below 100%. For such situations, we define a worst coverage-based cost



This cost assignment method finds the least-covered subregion (in terms of energy) of each node's coverage area and sets the node's cost equal to the inverse of the sum of the energy of the individual sensors capable of monitoring that subregion.

Consider the scenario illustrated below, where the rectangular area is the region to be monitored and each sensor is capable of monitoring the regions within the circle representing its sensing range. Sensor 1 can monitor regions A and B and since the coverage in region A is the poorest in terms of total energy, its cost is set to 1/E(A)=1/2. Similarly, sensor 2's cost is set to 1/2 and sensor 3's cost is set to 1.



More realistically, the utility of a sensor network application may degrade gracefully with the amount of area that is covered. To account for this, we propose another routing cost that considers the comprehensive coverage in the regions that a sensor can monitor instead of the single least-covered region. This comprehensive coverage-based cost is set as a weighted sum of 1/E(x), weighted by the area of each subregion. In other words, to obtain this cost, we integrate the inverse of E(x) over each sensor's coverage region.



Again, consider the scenario illustrated above. Sensor 1 will set its cost as area(A)/2+area(B)/2. Similarly, sensor 2's cost is set to area(A)/2+area(B)/3+area(C)/2 and sensor 3's cost is set to area(B)/3+area(C)/2+area(D). This comprehensive coverage-based routing cost provides a more balanced view of a node's importance to the sensing task.

In addition to these routing costs, in this work we also propose an integrated protocol for route discovery and sensor selection that further lengthens network lifetime by jointly selecting routers and active sensors. The premises for the design of DAPR are twofold --- that sensors critical to the sensing applications as data generators should be avoided as routers and that the selection of a sensor for the active sensor set affects its potential routers as well as the sensor itself. In DAPR, finite-length queries, which are triggered by the sending of Query packets, are processed by a subset of the sensors available in the network for a predetermined query length. Before the query is processed, the network undergoes a Route Discovery Phase followed by a Role Discovery Phase. Upon completion of the Role Discovery Phase, sensors process the query and provide data to the querying node for the duration of the query, as shown below.



Simulation results show the effectiveness of DAPR and the proposed application aware rouitng costs. The figure below show coverage degradation over time in a radnom uniformly deployed network. The lifetime before the first break in coverage is highest for the worst coverage-based routing cost, giving an improvement of 12% over the energy-aware routing cost. Networks using the comprehensive coverage-based cost, which was designed so that coverage degrades more gracefully, were the last to drop below 95%, improving by 15% over networks using the energy-aware routing cost.

The figure below shows coverage degradation over time in a more non-uniform deployment scenario. Because coverage is less uniform throughout the network, the gains that can be obtained from the use of the application-aware routing costs are higher than in the case of the uniform deployment scenario. The worst coverage-based routing cost gives an improvement of 28% over the energy-aware routing cost in terms of lifetime before the first break in coverage. The comprehensive coverage-based cost gives an improvement of 20% over the energy-aware routing cost in lifetime before coverage drops below 95%. energy-aware routing cost.

In DAPR, sensors are deactivated by sending a deactivation beacon to neighboring sensors after a backoff timer expires. In the figure below, we compare network lifetime when setting the backoff timer according to different criteria - randomly, based on the individual sensor node's cost, and based on the sensor node's cumulative route cost. As the activation or deactivation of a sensor affects its routers as well as itself, we expect lifetime to be highest when setting the backoff timer according to the cumulative route cost. The figure, which considers the uniform deployment scenario using the comprehensive coverage-based cost, agrees with our expectations and shows that predetermining routes increases network lifetime.