Wireless sensor networks (WSNs) offer the promise of revolutionizing the way scientists observe the physical environment. Unfortunately, experience from the networks deployed to date has shown that planning and managing a sensor deployment is a great challenge requiring expertise in multiple areas of Computer Science. However, future sensor networks must be implemented by researchers in other disciplines. For this to happen, the process of deploying and managing sensor networks should be radically simpler.
In this project we develop a set of network design tools. These tools use site-specific signal propagation models coupled with detailed physical layer radio models to calculate packet loss rates. Based on these estimates, the tools determine the location of additional relay points and gateways necessary to create a reliable network around the required measurement locations. Furthermore, we develop a network self-monitoring tool that correlates measurements taken by individual sensor nodes to construct a global view of the operational network. Information from the deployment is fed back to the network design tool. Through this coupling, incremental adjustments to the network layout are made until the network reaches the desired level of performance.
Using the tools developed by this project, sensor networks will become predictable and robust instruments, empowering scientists to observe phenomena that were previously out of reach. We are working with scientists at Johns Hopkins University as well as high school teachers to bring the results of this project to the broader academic and educational community.
Wireless sensor networks comprise small, embedded sensing devices, called motes, deployed in the natural and built environment to measure phenomena of interest. Marquee applications of wireless sensor network technologies include environmental, structural, and process monitoring. The motes' small size and wireless communication capabilities simplify deployment and reduce impact to the environment. At the same time, to reduce power consumption, these nodes use low-power radios. For example, current motes, use radios that follow the IEEE 802.15.4 standard that consume, when active, 100 times less power than common WiFi radios. The drawback of using these radios is that their effective range is shorter and they are more susceptible to interference from external radio sources such as co-located networks and ambient noise. The limited range of low-power radios means that in many cases data need to travel multiple hops through so-called relay points that forward them as they travel from their sources to their ultimate destinations. Given the unreliable nature of wireless links, one needs to select and maintain end-to-end paths that provide packet delivery rations as close to one as possible. Furthermore, considering that low-power radios consume three orders of magnitude less energy when turned off, there is great incentive to put the nodes to sleep (or as we say 'duty-cycle' them) when not active. However, this means that nodes may need to wake up their neighbors or node wakeups should be otherwise synchronized. In summary, it is necessary to rethink the network architecture for sensor networks from the bottom up. This task is further complicated by the requirement to have these networks deployed and managed by domain scientists with little expertise in Computer Science, let alone Wireless Sensor Networks. Our work in this project has produced elements of such a wireless sensor network architecture. At the physical level, we have investigated the spatial and temporal properties of low-power wireless links and leveraged their characteristics to develop techniques for placing the relay points of a wireless sensor network deployment. Specifically, we have developed techniques that allow practitioners to create links that span distances that are two to three times longer than the naive approach while maintaing high quality. The same techniques can be used to place mobile gateways that collect data from a WSN deployment. We have also identified non-linearities in the received signal strength measurements (commonly called RSSI) of popular low-power radios. The implications of these non-linearities are widespread, since protocols ranging from link estimation to localization use RSSI. We developed techniques to resolve these non-linearities and showed that using them improves the performance of the aforementioned algorithms.
