Ah, that was a concern of mine, but I thought of two ways to fix it.
1) Point the directional antenna directly upwards. In this scheme we are interested in getting a unique signal profile, not necessarily the strongest signal, so I think this should be ok. Pointing upwards (hopefully) does not depend on orientation.
2) For a robot, take the orientation into account. I plan on putting a digital compass on my robot, and aligning it to the proper direction when taking a location measurement.
Admittedly, the orientation issue is a real one, but hopefully one of these methods can get around it.
How this works (not triangulation)
on
WiFi Triangulation
·
· Score: 3, Informative
Hi all, this is my first/. post. I did a research project last semester and implemented a system like this, and got about 1 meter accuracy on average.
Rather than using signal strength for triangulation, you use it to record a "radio map", and compare your current position to the map. The basic steps are:
1) Walk around a room, recording the signal strength to each AP (so you get a file such as
"Access Point #1, Avg signal: 96
AP#2, Avg signal: 74..."
). Netstumbler or other software can help you make this file.
Create a "profile" like this for every location you wish to map (roughly, one every square foot or meter). The number of profiles determines the granularity of the system, but too many profiles can cause "collisions" in the sense that different locations have similar profiles, for some reason or another. There are ways to combat this, one of which is to make an educated guess on the new location based on the last one. (i.e., the user could not have walked over 10m in one interval)
2) When a user connects, they can compare their current signal strength info ( such as
AP#1, signal: 34
AP#2, signal: 74)
to the map: the closest point is probably their location.
I did a simple euclidean distance calculation (taking each profile as a vector in some large space [cool how the pythagorean thm. generalizes, eh?]. There are many better ways, which I am researching this semester, but euclidean distance is fine for now.
I'm pretty sure this is why they must spend an hour per 10,000 square feet to "calibrate" the system. I had to do the same, but it was a *lot* slower; I need to make a tool to do this automagically.
This semester I am also looking to get my system working with an ipaq robot running familiar. It's the combination of the palm pilot robot kit and this positioning system. Hopefully, the little robot should know (roughly) where it is, and be able to be controlled via the internet.
Check out my webpage if you are interested in more details.
Slashdot Mirror
Ah, that was a concern of mine, but I thought of two ways to fix it. 1) Point the directional antenna directly upwards. In this scheme we are interested in getting a unique signal profile, not necessarily the strongest signal, so I think this should be ok. Pointing upwards (hopefully) does not depend on orientation. 2) For a robot, take the orientation into account. I plan on putting a digital compass on my robot, and aligning it to the proper direction when taking a location measurement. Admittedly, the orientation issue is a real one, but hopefully one of these methods can get around it.
Rather than using signal strength for triangulation, you use it to record a "radio map", and compare your current position to the map. The basic steps are:
1) Walk around a room, recording the signal strength to each AP (so you get a file such as "Access Point #1, Avg signal: 96 AP#2, Avg signal: 74 ..."
). Netstumbler or other software can help you make this file.
Create a "profile" like this for every location you wish to map (roughly, one every square foot or meter). The number of profiles determines the granularity of the system, but too many profiles can cause "collisions" in the sense that different locations have similar profiles, for some reason or another. There are ways to combat this, one of which is to make an educated guess on the new location based on the last one. (i.e., the user could not have walked over 10m in one interval)
2) When a user connects, they can compare their current signal strength info ( such as AP#1, signal: 34 AP#2, signal: 74) to the map: the closest point is probably their location.
I did a simple euclidean distance calculation (taking each profile as a vector in some large space [cool how the pythagorean thm. generalizes, eh?]. There are many better ways, which I am researching this semester, but euclidean distance is fine for now.
I'm pretty sure this is why they must spend an hour per 10,000 square feet to "calibrate" the system. I had to do the same, but it was a *lot* slower; I need to make a tool to do this automagically.
This semester I am also looking to get my system working with an ipaq robot running familiar. It's the combination of the palm pilot robot kit and this positioning system. Hopefully, the little robot should know (roughly) where it is, and be able to be controlled via the internet.
Check out my webpage if you are interested in more details.