A theory on Kiwi TDoA ambiguity

[redhat]

I was thinking about this last night over dinner.  I think the reason the TDoA function on Kiwi's doesn't work all that well is pretty simply;  There are no two Kiwi's with identical antennas, and more importantly, antenna cable lengths.  A few nanoseconds here and there can cause pretty severe distance errors.

If we had a network of Kiwi's with all the same antennas and cable lengths, I bet the results would be more predictable.

Thoughts?

+-RH

[~SIGINT~]

As electromagnetic waves travel at a speed of 300 km per millisecond I believe that it is safe to say that variations in cable lengths are most likely fairly negligible in this application. Antenna types should have no bearing. A larger antenna does not pickup the electromagnetic waves any faster than a small antenna.

In a nut shell, here is how the KiwiSDR TDoA feature operates. After signals are captured (30 second I/Q samples with receiver GPS time and location embedded) they are analyzed on a server. By cross-correlating the signals, the time difference of arrival is calculated then using non-linear statistical regression, a series of hyperbolic arcs are estimated coming from each receiver. Where the three arcs intersect you have the transmitter location.

I believe that the signal type (AM, SSB, data, voice etc ...) and the quality of this signal is of greater importance. I have been told that the TDoA feature suffers dramatically when trying to correlate complex signals. It is most likely all about the algorithm and the math running on the server performing the analysis.

There is a good primer available on the UDXF web site:

Direction Finding with TDoA on KiwiSDR Net: An Introduction
http://www.udxf.nl/tdoa_firstexperiences5.pdf

Myself personally, I have never had any good consistent results with the TDoA feature. I can never get those hyperbolas to cross.

More interesting reading here:

TDOA Transmitter Localization with RTL-SDRs
https://panoradio-sdr.de/tdoa-transmitter-localization-with-rtl-sdrs/

Pay attention to the theory that receivers should be close together as mentioned in Resolution of TDOA localization and as seen in the TDOA system setup in the city of Kaiserslautern, Germany image. Perhaps that is what I am doing wrong. I am selecting receivers in a triangle which are too far apart from each other.

[Shortwave_Listener]

An interesting thing with TDoA is that using CW to decode an AM signal helps a lot. I was trying TDoA on an AM station a while ago and getting bad results, every time I ran it the results were different by a large margin. I then tried to tune it in with the CW mode on the carrier and use TDoA again. The results were greatly improved, much more stable results.

[ChrisSmolinski]

Agreed, I don't think antenna types or transmission line lengths are a factor.

From my experimentation, you get the best TDoA results if you already know the transmitter QTH 

What I mean is, judicious selection of the locations of the participating KiwiSDRs immensely improves TDoA accuracy. I find encircling the suspected transmitter site often produces the best results.

What this boils down to is an iterative approach. Start with a bunch of KiwiSDRs, observe where the target seems to be. Then add/remove KiwiSDRs and repeat. Often looking at the raw results ("TDOA combined" map setting) is more helpful to figure out where the transmitter actually is, vs the "TDOA map" (just show me your answer) default option, especially when fine tuning things.

Plus of course not using any of the really bad KiwiSDRs.

I think one of the major factors is variation is the ionosphere, especially its height. I am not sure what, if any, corrections the TDoA algorithm applies to compensate for this.


One test I just ran, after much fine tuning. Apparently the actual QTH for CFRX is 43.5N, 79.6W. Not too bad. But again, I "knew" what the correct answer was before I started.  You should have seen the first result, TDoA put CFRX somewhere in southern Pennsylvania.

I still maintain pirate ops do not have much to fear from TDoA. Many of them manage to out themselves perfectly fine without any direction finding assistance 

[~SIGINT~]

Be sure to visit the following CFRB/CFRX web site for station history and photos.

CFRB/CFRX transmitter site
https://cfrx.webs.com/

[Charlie_Dont_Surf]

I agree that in the overall scheme of all the errors that are there and can be introduced, the length of the coax and the differences in antenna types is not going to be highest on the pareto chart.

TDoA is ultimately a process that relies on probability and statistics and the map result is always expressed as a probability. If you do one TDoA run, not knowing where the transmitter is ahead of time, and at the end of that run think that you have the one "true result", then I have a bridge to sell you. Statistics is a science of estimation based upon a pool of data and determining the confidence level in that estimate; one run is not a pool, for the purposes here, it's one sample. (Admittedly 30 seconds of sampling has many samples within it but with the time scale of HF propagation, for the purposes here, I call one TDoA result = one sample.)

Some of the things that I like to do are:

a) repeated runs using the same receivers
b) repeated runs using completely different receivers, if possible, or at least swapping in one or more receivers.

If I get more or less similar results each time, then I am more confident of the answer. (This is one of the basic rules of statistical inference.) However, more often than not, especially on lower-power transmitters, a) and/or b) lead to somewhat different results than previous runs. This means your confidence level in the previous result has dropped. Some of the reasons for this are covered in the links that SIGINT provided. There can be a "pulling" action depending receiver choice, for example.

Other things that have an impact are different propagation modes, daylight/nighttime over some of the receivers, and SNR.

What I do to maximize chances of success*:

1) a) and b) above
2) I always use IQ mode.  I am not sure if the kiwi TDoA algorithm performs separate I and Q correlations but I was told many years ago that IQ mode was essentially mandatory. I think that the TDOA algorithm can still generate phase information from AM, CW, SSB detection but I'm going to suggest that IQ mode is probably better for this to give it the most available information to work with.
3) Be wary of some receivers in daylight and some in nighttime. Seek to have all the TDoA receivers you use either all in the day or all in the night.
4) I try to use 3 receivers most of the time and only emply 4, 5 or 6 to solidify a result that I already determined using 3 to increase confidence.
5) The receivers in use can not be too close to the transmitter for HF TDoA, otherwise there will be strong errors. What's "too close"? It depends, but a minimum of ~300 miles/~500 km is a good rule of thumb.
6) Because of 5) above, I always start out wide and then decrease my focus area encircled by the receivers.
7) In general, I like equidistances between all the receivers and the target. This is why it helps to know where the transmitter is already. (Yes, I realize the irony here and so should you.)
8 ) Always favor receivers with better SNR on the signal you want to TDoA over worse. (Note that has nothing to do with a receiver's general SNR score.) To check out a receiver before you choose it for TDoA, double click on the receiver in the TDoA receiver selection map to open up a browser tab for that receiver on the frequency of choice. Make a choice of receiver(s) based upon reception quality. Don't just assume it will be OK without checking. (Garbage in --> garbage out.)
9) I generally use 15-second sample periods for initial screening and then move to 30 or 60 seconds to solidify confidence, or if things appear to be noisy.

*success to me looks like a result one high probability result in a diameter of maybe 50 km. This occurs very infrequently on non-SWBC transimitters!

[Charlie_Dont_Surf]

Often looking at the raw results ("TDOA combined" map setting) is more helpful to figure out where the transmitter actually is, vs the "TDOA map" (just show me your answer) default option, especially when fine tuning things.

Agreed. Especially when you can not seem to get to a result on the "show me the answer" map after multiple tries, I often look into the raw "receiver x vs. receiver y" results and see which one(s) are most problematic. Eliminate/refine receiver choices and then try again.

I still maintain pirate ops do not have much to fear from TDoA. Many of them manage to out themselves perfectly fine without any direction finding assistance 

Obtaining a truly "robust solution" (as they say in the statistics world, i.e., a believable result that you can be confident in) on a low-powered station can be very, very difficult in my experience. In my opinion, for a low-powered pirate transmitter, kiwi TDoA is more something to generate innuendo, suspicion and false results than to generate an answer you can believe in.

By the way, the people you really have to worry about are the FCC in the US or the equivalent in other countries (example: Ofcom in the UK). They have much better equipment than just kiwi TDoA and the FCC aren't using theirs to chase HF pirates very much these days. They are too busy chasing FM pirates in big cities.

Side note: Here's some of the unclassified capability that the US Military has. You can bet that the FCC has stuff that is on par or one step above this: https://youtu.be/MncDlhdBOlY?t=1772

[MDK2]

Is anyone accounting for internet lag? Or is the audio data timestamped somehow? I believe you need GPS to use a kiwi SDR, correct?

[~SIGINT~]

My understanding is that GPS is required to ensure that all of the receivers are locked to the same reference clock and frequency standard. I do believe that the data captured is time stamped and processed by the Kiwi TDoA server. A receiver is not capable of performing the TDoA function internally. It is just a "data collector".

[Ray Lalleu]

The length of the coax line can just move the receiving location a few meters. Nothing that counts. This wrong idea comes from the usual combination of identical antennas close to each other for standard goniometry. Once, I set two FM antennas 1 wavelength apart on top of a rotor. All bearings were 15 degrees false, and that was because I did not care to make coax lines of exactly equal length for both antennas to the combiner.

In TDOA, the phase of the RF signal is anything and unknown. What is important is the phase (or delay) of the recovered audio after detection. A difference of 1 millisecond means a difference of distance of 300 kilometers between the TX and the two receivers, thus giving an hyperbola for possible TX locations. With at least a third KiwiSDR (or more), several hyperbolas can be drawn, and where they encounter is the probable place of the TX. Of course, with signals reflected by the ionosphere layers, the hyperbolas should be blurred, and so is the 'point' where they encounter. The error can be much larger when the propagation is quickly changing between daytime and nightime conditions.

BTW, note that only modulated signals can be treated by TDOA. If there is no modulation (or even if the S/N is too bad), TDOA can't do anything. Only the traditional goniometers can give some bearings.

On the other hand, TDOA can be accurate with direct propagation along/just above the ground. And setting a TDOA receiving station is easier than a goniometer station, so FCC (and alike elsewhere) could easily relie on many stations, wherever there is a powerline and an internet connection. Less easy where solar PV panels (and batteries) and radio or satellite links are needed.

As you know, shortwave signals have a short direct range, particularly when the polarisation is really only horizontal. No, an horizontal dipole is not enough to achieve such a goal.