Comment by mannykannot | Hacker News Reader

Comment by mannykannot | Hacker News Reader

Parent Full thread

I took a look at this a few months ago, and when I left off, I was unsure exactly how they conducted the validation of the technique. In particular, there was a test on June 3 2021 where they attempted to reconstruct the path of a flight from Samoa to Australia[1]. Certain phrases in the report leave me wondering whether the test simply amounted to looking for anomalies in the WSPRnet data in close spatiotemoral proximity to where the airplane was known to be:

"There are no WSPRnet anomalies observed at departure. At 03:30 UTC there is a drift anomaly 2 minutes behind the aircraft as well as 5 WSPRnet links just behind the aircraft, which may be due to the wake." [my emphasis here and below.]

"There is no obvious way to choose whether the aircraft turned to port or starboard to execute a 180° turn back towards Australia. I tried out both options and the best fit appears to be a turn to starboard."

And from a comment by someone apparently involved:

"There was one error of note, the initial turn of the aircraft after departure was to port and not starboard. Mike commented 'I should have told you that aircraft departing NSFA on RWY 08 will always turn to port due to terrain clearance considerations. If you’d known that then no doubt the first part of the route would have been a bit more accurate.'”

These quotes seem to suggest, if not outright imply, a detailed knowledge of the airplane's track by the person(s) searching the WISPRnet data for indications of that track, and if this is the case, then the test falls short of demonstrating an ability to determine the previously-unknown track of an airplane, which is what is needed for the MH370 investigation. I would be more than happy to update this opinion once I get a clearer picture of how these tests are conducted.

[1] https://www.dropbox.com/s/1cox7y91m89sjsv/GDTAA%20V2%20Blind...

Thank you for the link and I need to spend some time to read the details. It was stated in the paper I read that tracking one of the proving flights failed but other flights had data which correlated to part of an aircraft's flight path.

I agree with your analysis that some pre-knowledge of the track is apparent here and therefore more fully blinded independent experiments would be needed to add credibility.

Given the number of airline flights at any one time and the near constant WSPR radio traffic I would have thought there is enough raw data out there to conduct further research. There may a PhD in this for someone :-)

I do worry, as implied in your emphasis, that the temporal resolution is very low. Given that the transmit time of all symbols is approx 110.6 seconds [1] plus the time for a WSPRnet station to decode a message will translate into spatial errors of various magnitudes proportional to the speed of the target aircraft, but perhaps this has been factored in.

I do remain intrigued enough to try and learn more about the robustness of the techniques.

[1] https://swharden.com/software/FSKview/wspr/

Several hours ago, someone made two posts in this thread, pointing out the lack of statistical analysis (aside from an unsupported claim of statistical significance) in a similar test involving an airplane over the Baltic Sea. I'm not seeing those posts now, and I don't know what to make of that (did their author find a reasonable analysis?)

Perhaps that commenter observed that the author of the paper seems incredibly personally invested in the success of this method and tends to personally attack any detractors without regard to the soundness of the criticism. The author seems to also have a lot of fans who do the same. Publicly engaging with a person like that requires a certain tolerance for crankiness that not everyone possesses.

I think a hypothetical WISPRnet advocate might respond by pointing out that the plane's flight isn't unknown per se, and that they could markedly narrow the search space through first applying other information about the flight, but you're probably still right anyway in that it wouldn't be enough narrowing to get useful results. That phrasing does seem to imply needing to know just about exactly where to look already to get useful results.

I think the proponents of this method can assume a last known position (with time), but for the tests to be relevant, they need to show that, given this, the data provides significant support for one and only one subsequent track. I wonder why, if the people performing the tests understand this, they would present a report that raises doubts about whether they have done so.

Reply on news.ycombinator.com