AoA IQ Data soundness

[ReubeyGalleos]

Hi, I'm currently using the Direction Finding Connectionless examples to evaluate using the nRF5340 for numerous projects we have lined up, and I'm having difficulty ascertaining whether or not I'm actually collecting valuable data.

I've made minimal changes to the example, just set up the GPIO overlay to match the antenna array we are using and setting up a very basic switching sequence.

My overlay file is currently set to use 3 antenna, with 0x3 set as the reference antenna.

	dfe-antenna-num = <3>;
	dfe-pdu-antenna = <0x3>;

My switching sequence is then set to {0x5,0x9}.

My understanding then is that of the 45 samples I receive from the CTE, the first 8 would be using ANT 0x3, allowing me to ascertain a reference phase using arctan2(Q,I), then the remaining would alternate between ANT0x5 and ANT0x9. From the switching samples I calculate a phase offset and use the basic θ = arccos((ψλ)/(2πd)) formula to calculate an angle.

My first 2 most basic questions are so:

1. Obviously, this approach is going to struggle with noise and reflections etc, but I'm unsure as to how big those effects might be - how flakey I should expect this naive appraoch to be. 

2.  The IQ samples themselves, I'm unsure as to what range of values I should be expecting. In most of my testing so far the majority of my samples are in the range 0±50, which from looking around on this site seems to be low.

I'm reasonably confident the data I've collected so far is garbage, and will have more equipment available soon to properly interogate my hardware setup (WFH introduces some unfortunate delays at times), but I want to check it's not my analysis of the data that's the issue.

When calculating the reference phase, I use atan(Q,I) and see that odd samples are consistently ~180° out of phase with even samples. Is this expected behaviour, and if not how am supposed to ascertain the reference phase, are there supposed to be two reference phases for odd an even sample slots? My intuition is no but as I'm sure is obvious, I'm very much learning on the job here.

[swz_nordic]

If you want to use the 3 antenna, you should configure the antenna switching parttern for 3 antennas not 2 antennas.As for reflection and noise,how do you distinguish it?I think that there are error  factors,but i can't prove it.Besides that,my I/Q values are smaller than yours, they are in the range 0±30.i use the nrf52833dk,if you have the interest,we can make a talk.

[Simonr]

Hi

As mentioned already here, you need to add a switching pattern for the number of antennas you want to use which seem to be 3 antennas here. If you're new to Direction Finding we also have a whitepaper here with some more in-depth information on how Direction Finding works. 

1. Indeed, 3 antennas are not likely to give a very accurate result as the triangulation will be very rough. I don't think you can expect better than ~1 meter worth of accuracy. As "how flakey" is very much a case of definition, and will vary very on the range from transmitter to receiver and the environment it operates in, I'm not sure what I can share on this.

2. The raw IQ data that is transmitted must be converted into vector form to be reviewed properly. We don't have have any specific sample code showing this, but you can check out this page for information on the various formulas used to translate IQ data between polar and rectangular form. Nordic will not provide a Direction Finding algorithm for you I'm afraid. 

Best regards,

Simon

[ReubeyGalleos]

Apologies, I don't think I made my enquiry very clear. I'm familiar with the whitepaper, but as I'm using the NCS as that would be what we use in production, certain things are unclear. In the whitepaper, the switching sequence the first 2 writes to the SWITCHPATTERN register are used to set the antenna's used in the Guard then Reference periods. Subsequent writes are used to set the sequence that will be repeated if there are slots remaining after having being traversed. In NCS however, we set the "PDU" antenna in the .overlay file, and the README.rst says "If the number of switch-sample periods is greater than the number of stored switching patterns, then the radio loops back to the first pattern." This to me implies that the ant_patterns array shouldn't contain the Guard and Reference antenna patterns, however from your response I assume this is incorrect, and the format of the ant_patterns array should instead be [GUARD, REFERENCE, SWITCH1, SWITCH2...]. Can you confirm this is correct?

[Simonr]

Hi

From the Direction Finding technical overview from Bluetooth, you can see that the Guard and Reference period is done first, then you do switching between antennas and sampling. Here you need to switch between all (3 in your case) antennas until the sampling period is over.

Guard periods, used in many communications systems, are a technique designed to ensure there’s a gap between adjacent transmissions so that they do not interfere with each other.

During the reference period, 8 x IQ samples are acquired from the first antenna at 1µs intervals. No antenna switching occurs during the reference period. The host may be able to use the 8 reference samples to estimate the frequency of the signal and, from this, the wavelength. This allows more accurate angle calculations to be performed.

Sample and switch slots may either be 1µs or 2µs long. Support for 2µs slots is mandatory while support for 1µs is optional. HCI configuration indicates which slot length is to be used by the controller.

The full technical overview can be downloaded here.

Best regards,

Simon

[ReubeyGalleos]

Hi Simon, thank for your reply.

I understand the function of the guard and reference periods, my confusion lies with exactly how the NCS direction_finding_connectionless_rx example assigns antenna patterns for these periods. The demo is highly obfuscated as it makes heavy use of HCI RPMsg calls to do most of the interfacing with the radio registers. Because of that I can't determine what data it uses to perform the first two writes to the SWITCHPATTERN  register, whether this comes from the first two entries in the ant_patterns array or whether it uses the 'PDU' antenna pattern provided in the .overlay file. If this is the wrong place to ask for specific technical details about the NCS sdk I apologise.

My other main question in the opening post was regarding the expected range of IQ values. While the technical documention provides information on what size of information can fit in the DMA registers used to store the IQ values, I haven't been able to find information on what range I should expect "good" IQ values to fit within. Any information you could provide on this would be invaluable.