We implemented our work on the MICA motes developed by X-Bow. Under our architecture we implemented a simple Geographic Forwarding [16] mechanism as well as a simple buffering MAC protocol that follows a send when ready strategy with no contention resolution. See [11] for more information on the MICA platform and TinyOS Operating System.
Our MICA test bed consists of 24 motes laid out in a 12 X 2 grid with motes placed one foot apart. Nodes communicate to neighboring nodes within a specified neighborhood grid size (NGS). The awareness horizon is set to 1 hop in accordance with the small size of the network. For our experiments we tested NGS sizes of 2 and 4.
Nodes are employed with light sensors capable of detecting a shadow cast by the tracked event. Our goal was to track a rectangular object 1 square grid in size, moving at a speed of 1 grid per 10 seconds (GrPS). Nodes are programmed with our tracking architecture and an application which performs target location estimation. Additionally, nodes are pre-configured to report a detected event to a fixed location in the sensor network (node 0) every time the aggregate event location changes.
Based on the limited bandwidth, energy restrictions, expected target speed, and radio radius of the MICA motes, we experimented with and chose our heartbeat, failed leader, and entity timeout timers to be 2, 5, and 8 seconds respectively.
|
Initial experimental data show that for speeds of 1/10, and 1/15 GrPS and a NGS of 4 grids, our architecture as deployed was capable of correctly forming, maintaining, and reporting the location of a group around the event of interest. The reported path of the event for one such trial is shown in Figure 9. In this experiment, the tracked vehicle was traveling in a horizontal line from left to right as shown by the horizontal axis in the figure. The jagged quality of the reported path is a result of the limited number of motes detecting and reporting a position for the target at any specific point in time. It is possible that when the communication radius and the speed of events vary, multiple entities can be formed. Table 1 shows the number of entities formed over different speeds in the MICA testbed. The creation of spurious entities at faster speeds is due largely to the use of an unreliable MAC layer in the experiments. Message loss, as mentioned in the paper, allows spurious entities to be created. The noisy nature of the measured track is due to the fact that the light sensors do not provide a measure of distance from the target. Hence, the best one can do is simply average the positions of all sensors detecting the event. In contrast, other sensors (such as magnetic senors) can provide a better estimate of how far a measured target is. Triangulation will therefore result in a much more accurately estimated position.
Observe that the purpose of our experimental measurements is to qualitatively illustrate the success of our scheme in practice. It is not our purpose to compare experimental measurements to simulation. We have purposely decided to use a standard wireless MAC layer in the simulator instead of the simplified unreliable custom MAC layer implemented on the actual motes. Consequently, different performance results are expected. We believe that the implementation of the motes MAC layer will mature significantly in the future, making it less interesting to seek quantitative performance statements on the current testbed at this time.
