Performance Metric

http://rx.linkfanel.net/snr.html is used by some to determine how well their system is working. FWICT, it is not very good. Some of the top stations are terrible.

Comments

  • Ask him if he can improve it?
  • I think the issue is that there is no easy way to differentiate a discrete spur from a legit signal. I sent him mail a long time ago, but still seems to have the same limitation.
  • edited November 2019
    A kiwi installation with an LF/MF line fundamental and a lot of big carriers scores high even though it's a terrible system.
    I've chosen to ignore the whole metric. As Jim says "it's not very good". I'm not sure how to make it better either. A 'smart' analysis of a kiwi can be done but it seems that it almost takes AI to do it. At least my personal non-AI analysis takes a bunch of study to come up with an assessment of how well a given site is operating, and there's no guarantee that the result is all that good!
    Tough problem.
  • Previously when Marco, IS0KYB still had his "dynamic SNR" website running he had another metric called SibaQ. Although not disclosed it seemed to also factor in the 24 hour variation in SNR's for the 4 bands he measured hourly.

    Sorting Kiwi nodes this way made the better performing nodes stand out easier when comparing IMHO.

    Though problem indeed trying to capture the receiver live performance in a single metric and would need a good look as to how this performance is perceived and can be quantified. AI applied to snapshots of the spectra might be an interesting project.
  • One inspection of some of the 'Top' ones shows how skewed it is!
  • It's quite difficult to come up with an automated way to measure SDR performance. Been there, got the t-shirt :)

    I looked at the characteristics of known "good" KiwiSDRs vs "bad" ones. And wow, there are a lot of bad ones. Good SDRs have a few strong bins, and lots of low noise bins. Bad SDRs have a lot of high noise bins. This is one of those cases where you can look at the histogram plot and immediately determine if an SDR is good or bad, but it's tricker to do it in code. It boils down to distinguishing signal from noise.

    It was a few months ago that I wrote my code, so I would have to look over it again to remember all the details. Basically I built a histogram of signal levels over the spectrum, and find the amplitude of the 98th percentile bin. Then I find which bin is equal to 0.4 of that value. I call that the quality factor. Larger is better. These numbers were all empirically derived of course, not based on any theory. Just what seems to work. As with all things software, it could certainly be improved, I got it to the "good enough" stage. I mostly use it to test my own setup and various antennas, and a few others privately use it to evaluate their own KiwiSDR.

    This is what a "good" one looks like:




    And a bad one (some are even worse than this, this one is still bad enough to be effectively unusable IMHO):



    A smarter improvement might be to look at the frequency when examining the signal level to decide if it is real or noise. High signal levels are OK in the MW and SWBC bands. Maybe the ham bands Not good if all over the spectrum. Of course the time of day is going to affect what signal levels you expect on a given band. And of course the antenna type which may favor certain bands. And.... See, it's complicated :)
    WA2ZKDHB9TMC
Sign In or Register to comment.