Release : WSJT-X 2.2.1 Giving higher SNRs

Discussion in 'Ham Radio Discussions' started by ZS1SBW, Jun 12, 2020.

ad: L-HROutlet
ad: l-rl
ad: L-Geochron
ad: l-BCInc
ad: Left-2
ad: Left-3
ad: L-MFJ
ad: abrind-2
  1. ZS1SBW

    ZS1SBW Platinum Subscriber Platinum Subscriber Life Member QRZ Page

    We are noticing higher SNRs on FT8 HF for Release : WSJT-X 2.2.1 , when compaired to previous version.

    Apparently the old version gave SNRs that were too low, and that's now fixed.

    So if you're runnig < 2.2.1 don't get exited when your SNRs are vastly greater than the ones your giving !

    WSPR decoding has also been improved in 2.2.1 and it's possible that signals with lower SNRs that would not have been decoded in older versions.. are now decoded.

    73 ZS1SBW

    Last edited: Jun 12, 2020
  2. ZS1SBW

    ZS1SBW Platinum Subscriber Platinum Subscriber Life Member QRZ Page

    Rolled back to v2.1.2 as v2.2.1 is deaf.

    Hope Joe, Bill and Steve get to the bottom of the 2.2.1 "deafness" issue

    73 ZS1SBW
  3. K3XR

    K3XR Ham Member QRZ Page

    I have been using the upgrade for the past several days on 6 meters. Not noticed any problem including lots of DX from across the pond. I did not work him but did copy 4X4DK on 6 the other day. This weekend was very busy with both DX and contesters.
  4. KK5JY

    KK5JY Ham Member QRZ Page

    There's really no way to get accurate SNR calculations in a busy waterfall, because all the other strong sigs in the passband will cause an overestimate of the noise floor.

    That said, yeah, I have noticed some fairly strong sigs decode as "-24" even without wall-to-wall sigs, so there was probably some tweaking in order. :)
  5. N3KE

    N3KE Ham Member QRZ Page

    The other signals in the waterfall don’t matter. SNR should be computed from the variance in the received symbols compared to the decoded symbols. There is no need to measure any noise floor outside the 50 Hz of the actual transmission being decoded.
  6. KK5JY

    KK5JY Ham Member QRZ Page

    If you want to calculate signal-to-noise ratio, you need to measure the noise in the channel. That's how the SNR calculation works.

    If the bandwidth of the two power measurements is filled with nothing but the signal (as in your suggestion to only consider 50Hz) or if it is filled with other signals, the computation on the signal in question is rather meaningless.

    Minor edit: Please see:

    ...which says...

    Signal reports are specified as signal-to-noise ratio (S/N) in dB, using a standard reference noise bandwidth of 2500 Hz.

    On the flip side, the numbers generated by WSJT-X are obviously just estimates, and they're certainly an improvement over giving everybody "599." :)
  7. N3KE

    N3KE Ham Member QRZ Page

    Again, no. The 50 Hz is not “only signal”. If it was then the SNR would be infinite. The 50 Hz contains the signal and noise. And the only noise that matters to the demodulator is the noise in that 50 Hz. Once you demodulate the signal you can now measure the noise that was “underneath” it so to speak. This is the way SNR is measured for communications signals. You don’t need a “clear” portion of the channel to measure the noise floor separately from the signal. You can actually measure the noise floor as part of demodulation process.

    Some parts of the demodulator actually perform best with an estimate of the SNR prior to demodulation so there are methods of estimating SNR before going all the way through the FEC decoding, but again these estimators are only looking at the noise in the actual 5o Hz channel. They don’t look outside that band, noise or signals outside the band do not affect the demodulator.

    WSJT always normalizes SNR for all the different modes it processes to a noise bandwidth of 2500 Hz. It doesn’t measure the noise in 2500 Hz as that would be a meaningless thing to do for a narrowband mode. This is a convention entirely unique to Joe Taylor and no one else in the communications field has ever normalized things this way that I’ve seen. When comparing different bandwidth communications the rest of the world uses noise spectral density (essentially a 1 Hz noise bandwidth). Of course hams are used to S meters and SSB channels and not very familiar with digital communications and so I think JT chose to use normalization to 2500 Hz to perhaps make things more familiar to hams.

    But whether you normalize to a 1 Hz or a 2500 Hz noise bandwidth in reporting an SNR doesn’t matter, the bandwidth you measure the noise in is always the bandwidth of the communications signal itself. In this case 50 Hz. As you’ve already pointed out if you did anything else you’d get nonsense since adjacent channels contain other signals.
    W6RZ likes this.
  8. N3KE

    N3KE Ham Member QRZ Page

    Sorry I don’t have time to give a longer explanation but here is a link to a quick visual look at how a demodulator can compute the SNR solely within its channel:

    Those are for PSK modulations which are easily visualized. The size of the “ball” at each constellation point tells you the SNR. FT8 of course is FSK but the concept is the same if more difficult to visualize. And FT8 uses a low rate FEC which makes things even more difficult to visualize.
  9. KK5JY

    KK5JY Ham Member QRZ Page

    So what you are saying is that the "SNR" presented by the software isn't the signal-to-noise ratio at all. It's something they cooked up on their own, specifically for WSJT-X.

    Okay. :)
  10. N3KE

    N3KE Ham Member QRZ Page

    Your inability to understand how SNR is actually computed and used doesn’t mean mean it is “cooked up”. SNR requires a reference bandwidth. That bandwidth is essentially arbitrary. For some communications types there are conventions for that bandwidth. For others it is less clear what is a meaningful reference bandwidth. JT chose 2500 Hz, the familiar ham SSB channel. He could have chosen something else. All that matters is you specify it, but then that is another thing you have to specify and convert to.

    For folks that work in the comms field usually SNR is not used for that reason. Things like Eb/No, Es/No, C/No are used as they are much less ambiguous and they are more informative when evaluating and comparing modulations. But those figures of merit are all easily converted to SNR once you specify a reference noise bandwidth.

    The larger point being though you don’t measure or care about noise or interference outside your channel even if your reference bandwidth is wider than your channel.

    The misunderstanding you are having with what WSJT is reporting as an SNR is exactly the reason some object to JT’s choice of referencing all SNRs to 2500 Hz. He probably chose it to make things more “familiar” but instead it often confuses people into thinking something different is being measured.

Share This Page