TDoA extension operating notes
edited September 2018 in TDoA topics
This discussion was created from comments split from: Direction Finding and linking existing KiwiSDRs.
It looks like you're new here. If you want to get involved, click one of these buttons!
There are still quite a few bugs to be sorted out, but I have been able to get very accurate locations, certainly close enough to be able to find the TX site on Google Earth & Street View
Brief 'How To' based on my experience so far.......
I find that in order to get consistent results, if I make any major changes to the setup of the signal being detected, I have to refresh the browser and reopen the TDoA extension
Set the mode to I/Q and reduce the bandwidth to just cover the target signal.
Then set the SDR map to cover the target area, be careful not to 'nudge' the frequency on the waterfall whilst doing this.
Check each KiWi you are intending to use can hear a clean signal by double clicking on the KiWi 'flag' on the map, before adding it to the list
Choose KiWi's carefully. You can use KiWis much closer together for signals on the lower frequencies typically <5MHz. But for higher frequencies you need to use KiWi's much further apart in order to ensure that each KiWi is the correct 'skip distance' away from the likely signal source
TDoA works the best if you can choose SDR's that are approximately the same distance away from the likely source of the transmission
Ideally the SDR's should be chosen as opposing pairs, either side of the likely source of the transmission, for example one North and another to the South, a further pair to the East and West would then give the best results.
If all of the SDR's are on one side of the likely source of the transmission, the results can be ambiguous, so it is always best to experiment with different SDR's in order to find the ones that give the clearest indication.
Northern Europe is fairly well covered but Southern Europe is not. More TDoA KiWi's in places like Portugal, Spain, Tenerife, Ascension Island, Azores and Africa are required
Once running the sampling should only take 40 seconds at most.
I suspect that the process can fail if one of the previously chosen KiWi's becomes too busy, or looses GPS lock in the time between initial selection and actual sampling.
If the sampling phase takes longer than 1 minute, refresh the browser and start again.
Once the sampling has finished it may take several minutes to get a result back from the server.
If the server running Octave can't get a good correlation it will also fail, sometimes without an error message and it just keeps trying.
If the TDoA running phase hasn't finished after 5 mins, refresh the browser and start again.
I usually start with just 3 KiWi's to get a rough location and then zoom in as required. Take a look at the maps from pairs of KiWi's and delete the KiWi's that that don't provide good clear contours from the list, then add somenew ones and try again.
Sometimes it may take two or three attempts to get a good map back.
If you don't get good clear results, try running it a few more times, as sometimes the propagation is not favourable during a sampling run, but may improve on subsequent ones.
I have tried it with AM, RTTY, STANAG, SSB and other digital modes, including some short duration burst type signals, and with perseverance have managed to get good results. The only type of signal I have had difficulty with has been Morse.
Martin - G8JNJ
There can be problems when some of the sampling Kiwis don't have good reception of the target signal. Be sure to double-click on the Kiwi map marker to open a connection to that Kiwi in a new browser tab. If you're getting individual map plots that are completely filled with red or green, or "dt" plots with unusual patterns, this probably indicates a problem.
Martin - G8JNJ
The new algorithm displays an different set of error messages. Maybe Christoph will elaborate more about the algorithm changes.
I'm not sure if the new algorithm is fussier about the GPS update rate (each receiver was reporting, at the time >=20 updates/min) but that would seem to be the more likely as all of the receivers involved had good quality signals from WWV at the time(s) that it was run.
If I understand it correctly, he's seems to be testing pairs of plots to see if they are providing good results and omitting the ones that are not (which is typically what I have been doing manually), plus some other stuff I don't quite follow so easily.
Martin - G8JNJ
With the new algorithm a station gets excluded if either the recorded WAV file is not good (too short/has gaps) or if the times of the cross-correlations with other stations are not consistent, i.e., if the number of the station is 1 and equations such as (13)-(14)-(34) == 0 do not hold.
I will have a look and see which of those conditions can be relaxed.
Thanks for the latest update.
Generally speaking I find that the new algorithm is usable most of the time providing the signals are of good enough quality. If they aren't, I can still run the original version to get a feel fro a rough location, but I wouldn't have as much confidence in the end result.
Personally I don't mind the stricter testing procedure if it means that the end result is more likely to be accurate, especially if it filters out some of the erroneous multipath plots that are problematic at certain times of the day on certain frequencies.
As a bonus it also means that I don't have to be quite so 'fussy' about the initial selection of KiWi's, as I can now add a few 'extras' and let the new algorithm sort out the best to use rather than having to do it myself.
Maybe the testing applied could be further 'tiered' with a selection box or 'slider' control so that you could have a range of 'confidence' values that could be selected depending upon the required degree of accuracy ?
Martin - G8JNJ