Deployment Experiences
LUSTER has been deployed for durations of hours to weeks in three different locations, including in labratory, forested abd barrier island environments. During deployment, we gain some experiences, which may also benefit other EWSNs projects.
Deployment in forests
A test deployment of several sensor nodes in a forested environment is shown in Figure 9. A SeeMote with the SeeDTV software is shown in Figure 9(b) being used to monitorWSN communications and to query each individual node’s health status. Using this technique we found that one of the nodes was not working as expected and had to replace it with a spare. The plots in Figure 10 illustrate the data provided by a single node with eight light channels attached to it. In Figure 10(a), a node has captured a whole day’s light-cycle at a time resolution of four measurements per second. It graphically shows the obvious optimization of entering an extended sleep mode during the night hours. Figure 10(b) shows detail of nightfall on a windy day.
Figure 9: LUSTER test deployment (background lightened to show sensor nodes).
Figure 10: Light data for eight light channels, from an outdoor deployment location.
Sensor Localization by Excitation
Since LUSTER supports multiple sensors per node, node-level localization is not precise enough, and the placement of the individual sensors must be mapped. We took an approach similar to Spotlight, in which we sequentially excited each individual light sensor on each node in a known pattern.
Figure 11: A method of sensor localization by sequential sensor excitation.
Figure 11 illustrates our method of sensor localization, which appears as clustered light dots against the otherwise dark sensor readings. Vertical white bars show lost packets, and are unrelated to the localization discussion.
In the deployment, we shined a flashlight on each sensor in an order determined by their location on a map. Each was exposed for about two seconds (with one second sampling rate), and the readings were recorded by the system. Actual sensor IDs were mapped to their locations by inspecting the clusters portrayed in Figure 11.
Hog Island Deployment
Hog Island is a research site in the Virginia Coast Reserve (part of LTER), located off the Eastern Shore of Virginia. A deployment to the barrier island is ongoing and data are still being collected and analyzed. Figure 12 shows the layout of the deployment area. Sensors are placed on the ground in a cross-shaped area with increasing distance from the center, on a regularly-spaced grid, and along a tree branch half-way to the canopy ceiling.
Figure 12: A layout for deployment.
A Crossbow stargate acts as the gateway between the WSN and a 2 km back-haul wifi link, only reachable viaa high-gain directional antenna. The remote access point loses power at night, necessitating the use DTN techniques in LUSTER. A solar panel charges a 12 V marine battery, which supplies power to the stargate. In total, there are 133 light sensors connected to 19 sensor nodes, with five storage nodes distributed throughout the deployment area.
Two Deployment Trips
We went to Hog island to deploy LUSTER systems twice in this summber. The first trip is at July, 22, 2007, in which we set up the solar panel and the directional antenna, and deployed sensors and storage nodes with Steven Brantley (Gradute student from VCU). At July 31, 2007, we went to Hog island again for the second deployment. This time, we set up the WiFi link and successfully uploaded data streams to the remote servers in our departments, and storage nodes also record data for a duration of 14 days. A Result data review and some sildes about depolyment experiences are available.
Lesson Learned
During deployment, we learned the following lessons about how to deploy an EWSN system