Here’s some quick and dirty results.  You have to imagine all the furniture.  I placed the anchor on my desk, and then walked the tag around to various places.  I let the estimates average out over a dozen or so measurements.  
The green boxes show successful estimates.  The numbers inside the box are <Actual> / <Decawave Est>.  The pink boxes are where it failed to get an estimate.
Click on image to enlarge


  • This is clearly better than Nanotron.
  • The one true LOS estimate I made is really is really good.
  • It seems to take a long time to get an estimate – and meanwhile it is burning power.
  • The two failed estimates that are up near Brenda’s desk (top of page) have to pass through at least one layer of glass, or outside metal/concrete wall.
  • The failure in the bedroom is quite a long way.
  • The failure in the bathroom, has to pass through a large plate-glass mirror.
  • Many estimates are around +/- 500cms, but one is 30% wrong.
The next thing to do is to see how repeatable these estimates are, and what factors cause the estimates to fluctuate.

Excerpts from David’s response

It’s a little disappointing, isn’t it? What’s interesting is the inconsistency. The only commonality to the bad measurements seems to be glass. Are the closet doors in the master bedroom mirrored? Going through regular walls doesn’t seem to be a problem. Having to go through/close to metal appears to be a problem. That isn’t surprising.

It would be a PITA but measuring every square 6″-1′ in the main living area from the wall in the kitchen extending over to where Brenda’s computer is and over to about 1′ past the kitchen/nook wall would be interesting. Actually, mapping the whole living area would be interesting, but what PITA, and how would you mark all of those locations to be able to go back to them?

Since we were specifically asked to provide feedback to Decawave, picking a problem location and taking a bunch of measurements, demonstrating accuracy and repeatability, and then asking DW why it’s not working well there, might be worth the effort, as long as we have their attention.

I wish we had an instrument that could show what the RF looked like in a particular location. I don’t even know what that would be. A spectrum analyzer? But how do you look at a UWB signal on a spectrum analyzer since the bursts are so short?

The open space measurements are great. If there’s an anchor in every room it might work very well.

More thoughts

I was wondering about doing a careful survey of the whole place to see what kind of results I get. I’m trying to think of the best way to do the test.

Simply put, we need to see the ranging error from source to destination for all possibilities. If I had all the time in the world then I’d divide the whole house into small squares and check the Actual versus DecawaveEstimates ranges between every two squares. So a hundred squares would require 100*99/2 = 4500 measurements which I don’t think I have the fortitude, or time to do before I leave.

The other problem is how to represent the results if I had them to hand. There are more dimensions than we have eyes/brains to perceive. I need some kind of 5-D representation that plots src:<x1, y1>, dest:<x2, y2> against E<error>. I suppose I could simplify it by fixing src:<x1, y1> and simply plotting dest:<x2, y2> against E.

Imagine a 3D surface where height represents E at dst from a single src, perhaps even superimposed on a map of the apartment.  

Then I could somehow animate iterating over every possible src, so I can see how the topology changes.

That seems like a lot of work, and what might I see?

Leave a Reply