When doing Bluetooth direction finding, the AOA is mostly vertical.

[shuang]

Hi, I'm doing experiments with Bluetooth direction finding. I am using the example of the nRF52833 development board, and I have obtained the IQ representation of CTE and tried to solve the AOA using the MUSIC algorithm to process the IQ data. I have obtained several sets of experimental data at different incidence angles and solved the probability distribution of the angles in several experimental scenarios such as dense indoor areas and open outdoor areas. I found a strange phenomenon. Regardless of the actual angle of incidence, the AOA derived from more than 60% of the data falls in the range of 0 degrees to 20 degrees (where 0 degrees is the vertical angle). This seems a little strange, because there seem to be other factors dominating the angle of incidence. I'm curious about the possible reasons for this, or if anyone else has encountered a similar phenomenon. I sorely look forward to your help!

[JONATHAN LL]

Hi,

How did you measure this 0 -20 degree angle?

What are you expecting?

Antenna array on the central ?

If you where to walk around in one floor(flat plane) then i think 0-20 degrees seem reasonable results. 

Regards,
Jonathan

[shuang]

First of all, thank you for your timely response. 

My description does contain some confusing points, so allow me to briefly recapitulate the above questions.

As shown above, I am using θ to represent the true angle of incidence (the ground truth of AOA), which ranges from -90° to 90°, where 0° means that the incident angle is vertical (the signal from TX is perpendicular to the antenna array of RX). I use θ' to denote the angle of incidence derived from the IQ samples received from the RX. Ideally, θ' should be equal to θ. However, there is some error between θ' and θ due to interference from other obstacles or other signals, etc. If the error is α, then θ' ∈ [θ-α, θ+α]. Although some extreme cases of θ' may be beyond this range, repeating the experiment numerous times in the same scenario, θ' should generally fall within this range. Assuming an error of 10°, then at least half of the data can be derived for θ' ∈ [θ-10°, θ+10°].

However, I currently encounter a situation where whatever the true θ is (e.g. -60°, -45°, 30°, 60°, etc.), more than 60% of the data yields θ' ∈ [0°, 20°] instead of θ' ∈ [θ-10°, θ+10°]. In other words, it seems that something else (not θ) is dominating θ'. This looks a little strange. I have carefully checked my code and experimental settings and found no obvious problems, so I am seeking your help. I'm curious about what the possible causes of this phenomenon are.

[Simon]

Hi

It might be that you are measuring a strong reflection rather than the line of sight signal. This will be especially so in a small room rather than a large warehouse. We have a BluetoothLocationEngine that processes IQ. It has to use anti-reflection and noise reduction techniques to obtain reliable angles from IQ data.

If you look at our unprocessed data you can see smaller peaks that can become larger than the main peak in some circumstances.

Simon

[JONATHAN LL]

Thanks for the clarification, 

Support for a algorithm that is experimental and we do not have much insight on is not something we can nor will support as we do not currently support angle resolution algorithm work, so I cant be of much assistance, but feel free to continue discussing the topic. 

Two likely causes: the algorithm is just not working as it should or it is not robust against incorrect data for some of the channels (which is how multichannel looks) and does not work in real life due to this.

Regards,
Jonathan

[Miguel Roque]

Any news? I have the same problem, currently I'm using only 2 antennas from the 12 available. I have changed form different rooms, however the problem persists, for a static scenario, the angle is always jumping...