Poor accuracy using MDEK1001 kits

I have been attempting to track down measurement issues in our system and so have started simplifying everything to a minimum case and am still seeing significant measurement errors when the distances are being reported as closer than they actually are.

For the minimum test case setup I have taken two MDEK1001 kits and set them as anchors and placed them 1m apart. I then took another MDEK1001, set it as a tag at 1m from on of the anchors and used the les command to record data. Here is what it looks like:

C084[0.00,0.20,0.66]=0.88 5433[-0.75,0.00,0.66]=1.92 
C084[0.00,0.20,0.66]=0.89 5433[-0.75,0.00,0.66]=1.92 
C084[0.00,0.20,0.66]=0.92 5433[-0.75,0.00,0.66]=1.85 
C084[0.00,0.20,0.66]=0.92 5433[-0.75,0.00,0.66]=1.89 
C084[0.00,0.20,0.66]=0.88 5433[-0.75,0.00,0.66]=1.87 
C084[0.00,0.20,0.66]=0.90 5433[-0.75,0.00,0.66]=1.88 
C084[0.00,0.20,0.66]=0.90 5433[-0.75,0.00,0.66]=1.88 
C084[0.00,0.20,0.66]=0.83 5433[-0.75,0.00,0.66]=1.82 
C084[0.00,0.20,0.66]=0.74 5433[-0.75,0.00,0.66]=1.82 
C084[0.00,0.20,0.66]=0.85 5433[-0.75,0.00,0.66]=1.85 
C084[0.00,0.20,0.66]=0.77 5433[-0.75,0.00,0.66]=1.89 
C084[0.00,0.20,0.66]=0.79 5433[-0.75,0.00,0.66]=1.85 
C084[0.00,0.20,0.66]=0.76 5433[-0.75,0.00,0.66]=1.81 

And here is a photo of the test setup on a wooden table top:

Before getting to this point, I have tested the setup with a larger network, mounted on a variety of objects (plastic and metal) and a range of distances. It appears to always have an absolute error that is possibly unique to each tag/anchor pair.

I’ve done some more testing at a variety of distances, with the antennas both aligned and at 90deg.
I took several dozen measurements at each distance to get an good average and stddev at each point. Here’s the results:

ALIGNED
Actual	Reported  std dev   error
527	    298	      40	    229
1011	895	      15	    116
1492	1336	  13	    156
2136	2032	  24	    104
2548	2472	  20	     76
3048	2964	  15	     84
4369	4229	  20	    140
5624	5617	  19	      7
90 DEG
Actual	Reported  std dev   error
603	     344	  45	    259
1187	1051	  32	    136
1572	1362	  18	    210
2035	1840	  16	    195
2487	2323	  17	    164
2936	2781	  19	    155
3962	3796	  19	    166
5103	4898	  17	    205

This ends up giving an average offset error of 150mm, with a std dev on 64mm for the offset.

What can we do to improve this?

Hi @Milner
I will not make you happy. But your measurements does not look bad. You are facing following issues:

  1. Antenna measurement affection. The antenna does not behave the same in all directions - you can do a simply test - keep the TN in one place and start rotating the TN and observe measured distance. You will see that the average measured distance changes with the antenna angle. This natural for antenna and there is nothing you can do with his except change the antenna design.

  2. It is not recommended to put nodes on one surface. From our experience when you place nodes on the same surface (like table) the nodes might be facing multipath problems (increase the measurement std dev) - there is direct path (over the air) and second path (trough the table surface). I would recommend you to put nodes on tripods and increase the distance between them.

  3. Be aware with powering the nodes via power banks - our experience is pretty bad. The TN/AN has a low power consumption and the power banks do not know how to handle it - they can turn off the nodes randomly or they have a hi noise at their output as they are designed for Hi current application. We have recently solved mysterious system behavior by replacing powerbanks with power adaptors.

Additional note: The 1m Anchor separation is pretty small - unless this is your use case that you need to work with. I would recommend you to increase the distances between anchors to 5m or more.

Cheers
JK

I have similar experiences with the DWM1001-DEV as @Milner has and the solutions that @leapslabs suggest are indeed helping.

In my application it really helped to place the anchors far enough from each other. Not only in X but also in Y direction. I am using 4 anchors to determine the position of 1 tag. It also works with 3 anchors but it gives less reliable results because when moving the tag inside a metal structure it gets a block line of sight the accuracy decreased dramatically.

The 4 anchors are placed around an area where I need to maximum accuracy. When the tag is outside this square the accuracy is already less. The anchors are placed in plastic housings hanging at around 2 meters in the area.

Hope this helps!