Recently, a type of highly mobile sensor networks known as "ad hoc wireless networks" has become increasing popular. An ad hoc sensor network is generally composed of tiny devices known as "motes" - a type of cheap programmable devices with embedded sensing and wireless communication capabilities. Different from normal wireless networks based on server-client communication, ad hoc networks are built with each node acting as both a server and a client, therefore eliminating the need of an access point. Each node is then capable of routing information and communicating with each other, as long as they're all "linked" with each other. In this type of network, mobility is highly increased, as the size of the network can be expanded indefinitely. Therefore, many researches had been done on creating efficient routing algorithms. Many of such algorithms assume the ability of motes to know their location - algorithms such as "geographic forwarding" and "Location-aware routing" (LAR). For these reasons, the purpose of my study is to create a method of localization for these ad hoc networks. While quite a few researchers have already come up with methods of doing so, most of them involve acoustic or ultrasonic time of flight, requiring the motes to have acoustic or ultrasonic sensing and emitting capabilities. This project, on the other hand, investigates the possibility of purely using the motes' radiofrequency capability to do so - a capability that is generally pre-installed in every mote.
In the beginning, it was hypothesized that there would be a stable relationship between distance and signal strength - similar distance should receive similar signal strength. It should then be quite simple to find the relative location of each mote - after converting signal to distance, the location can be found by the intersection of circles made by locus of distances. However, during the data collection phase of the project, the result is found quite different. The rotation of the receiver and sender, for example, would cause signal strength to vary as shown on the right. At that time, I realized that the RF transmitter chip and the internal antenna are both located on the side, instead of the center, therefore causing orientations to affect signal strength. Furthermore, I notice that the displayed variation only applies when the sender is on the left of the receiver, and different variation patterns are produced under different setups. To find a true pattern between all such variations would require a tremendous amount of data collection. For that reason, the project continued assuming that all the motes orient toward the same direction. By collecting data while simply changing relative location of the motes, the variation is found to be surprisingly stable, in a somewhat "8" shaped curve. Using these distance-to-signal curves, I can then convert signal to possible distance pattern - as seen in the curve on the right. By converting signal strengths into these curves, similar to the intersection of circles, the intersection of these curves can then compute the locations of the motes.

The algorithm is still under testing, and the experiments should be finished in the next few weeks. More data collection will be done in the future regarding the orientation of the motes, and hopefully the project will be expanded toward having motes in random location and orientation as well. This study was supported with funding from the Simons Foundation.