ChipCon CC1000 RSSI data

This data was collected to answer three questions:

  1. Is Rssi useful for localization outdoors
  2. Is Rssi useful for localization indoors
  3. To what extent is calibration required to make Rssi useful

The data resides in the tinyos cvs in /tinyos-1.x/tools/matlab/contrib/kamin/chipconRssi.  To see graphs of the data, add that entire directory to your matlab path and run the "analyzeRssi.m" function.  You should see plots that look like:


The first type of graph is a density plot of the distance estimates gotten from that data set after calibration.  In general, four of the runs were used as training data for calibration and one was used for testing.  The density plot should be highest along the true distance=estimated distance line.  The second plot shows histograms of the ranging error after calibration in both percentages and absolute distance (cm).

There were three experiments conducted.  You should see 6 plots pop up corresponding to the 3 experiments.  The format of the data and the ways to calibrate/process it are all written in the analyzeRssi.m file.

Mica2Dot's were used in all experiments, which have a ChipCon CC1000 radio, which was running at 433Mhz with a wire antenna on each node.  In each experiment, N nodes were placed in a grid, rather carelessly such that each node was in a random orientation/position.   The positions themselves were not measured exactly but were confirmed to be within a foot or two of the positions that were recorded.  A base station in the center of the network was connected to a computer running matlab, which commanded each node in turn to transmit 10 messages.  All other nodes sampled each message 10 times for a signal strength readings and stored it in memory.  Matlab then asked each node for all the signal strength readings and stored them all in a big array.  All the nodes were then collected, randomly shuffled and redistributed in the same grid formation, but with different positions.  The experiment was repeated in this fashion five times for each environment.  Thus, in each environment, approximately five different distances were measured for each transmitter/receiver pair, where the transmitter/reciever were in different relative orientations for each distance.  All radios were transmitting with the PA_POW register set to 0xFF, ie. highest transmission power.

In the first experiment, nodes were approximately 2-meter spacing in a 5 x 3 grid in the east wing of the Intel Lab at berkeley.  The lab is relatively empty in the middle with desks around the outside of the room.  Obstructions such as computers and posters and chairs were not moved, but were left in place to obtain an indoor laboratory environment.  Results indicate that Rssi contains absolutely no distance information in this environment.

In the second experiment, nodes were approximately at a 3-meter spacing in a 4 x 4 grid in the field at the west gate of the UC Berkeley campus.  The field is relatively wide open, and the ground is at a slight slope.  Nodes were dropped directly onto the grass.  The experimenters and the laptop were the only obstructions on the field.  Results indicate that Rssi can give about 3-meter resolution in this environment.

In the third experiment, nodes were in the same environment as that in the second experiment, except that nodes were in a 3 x 8 grid, thereby measuring farther distances.  Results indicate that distance beyond the 9 meters measured in the second experiment are not extremely useful or are at least very noisy.  Furthermore, Rssi measurements taken at more than 9 meters are difficult to distinguish from those taken at 6-9 meters, indicating that the data at 9 meters is also not very useful unless data from farther distances can be somehow eliminated.  One important note about this data set, however, is that halfway through the fourth experiment many of the batteries had died, so I replaced about half of them before the fifth experiment.  While in theory lithium batteries don't change voltage levels, it is very likely that theory is incorrect, which could have affected these RSSI readings.