Before we delve too far into the various applications of wireless sensor networks, it may be of some use to review the characteristics of such a system. While wireless sensor networks allow us to deploy and gather information in a variety of places that are impractical to reach – whether due to cost or inhospitableness – they also have a number of constraints that need to be understood. These constraints inform the design, configuration and deployment – and the data collected and resulting applications.
A wireless sensor network is a spatially distributed system for collecting and monitoring phenomena including temperature, pressure, vibration, motion, salinity, acidity, or virtually anything else that can be collected or measured. Sensors are typically autonomous, performing their function with little centralized command and control. Because they are distributed in the environment, they must carry their own power source, processor, memory, and communications infrastructure – in addition to the sensors themselves. Many sensors not only collect data, but also perform an action based on that data – requiring an actuator as well.
These devices have limited power, memory and processing power. Thus any deployed configuration must optimize based on these tradeoffs. How much data does the sensor collect before it communicates – since radio frequency communications are very expensive from a power perspective. But on the other hand, there is a limit to how much data the sensor can collect before it must broadcast because it is memory constrained.
As a distributed computing network, algorithms must determine how and when to access data from the constituent nodes in order for the algorithm to perform reliably. Design tradeoffs abound. Sensors are not designed to operate in singularity – their power is in the collective. The behavior of the system as a truly complex system must inform the network design as much as the constraints of individual nodes.
One important characteristic of wireless sensor networks is that they are self-organizing – determining optimal communications pathways to nearest neighbors based on algorithm requirements. This self-organization is particularly important given the high failure rate of individual nodes – else the network would require constant field trips to reconfigure based on changes to topology.
Once (or as) nodes are deployed in the landscape, they must localize their position in the overall sensor network; sometimes in two dimensions, and sometimes in three. This is one of the most essential – and most difficult – steps in deployment. Various approaches to localization can be applied, and again understanding the options will inform the tradeoffs in the data acquired – because each of the tradeoffs is a potential source of error.
If the sensor nodes are manually deployed one at a time, then a GPS collector may be used by the deployment team to explicitly encode the location in the node (typically through an uplink to avoid keying in, with associated error). This is by far the most straightforward approach, but sensor nodes can’t always be deployed one at a time or manually. In hostile environments, nodes may simply be scattered and their final location may be ad hoc. In this case, there is no opportunity to manually collect the GPS location from each node. In such cases as these, depending on factors such as cost, size and power, the sensor node itself may be equipped with a GPS receiver. In either case, once the nodes know their position, they communicate with each other to exchange relative positions – self-organizing to optimize the energy budget with respect to communications.
The other approach uses range rather than explicit distance. Range-based solutions make use of range finders, or alternatively synchronized clocks to measure signal delay, with line of sight to determine distances between nodes – but not absolute positions in space. Positions are computed based on triangulation between the nodes. Range-free solutions instead use the number of hops between nodes as a proxy – one hop is nearby neighbor, two hops is a little farther, and so on. The hierarchy of neighbors is then computed. The advantage to range-free solutions is that they don’t require additional hardware, but at the expense of an increase in messages.
A final variable in determining deployment strategies is coverage and connectivity – which is dictated by the application. Full coverage and connectivity means that every location in the area of interest is covered by one or more sensors – and that each of those sensors have a communications path to report acquired data. Full coverage and connectivity is warranted in situations where error tolerance is very low – for example, military applications where error would lead to loss of life. Partial coverage with connectivity, where sampling is adequate, can support applications such as temperature which varies little locally. There is reduced data availability, but with the desirable tradeoff of reduced cost.
In the next column, we’ll explore some applications of sensor networks to modeling, analysis and simulation. Until then, understanding tradeoffs at the planning stage means a better outcome at the results stage.