The "dB" column in WSJT-X

Discussion in 'Working Different Modes' started by KT5MR, Jun 28, 2020.

ad: l-rl
ad: Left-3
ad: Subscribe
ad: abrind-2
ad: L-MFJ
ad: l-BCInc
ad: Left-2
  1. AI3V

    AI3V Ham Member QRZ Page

    Uncorrelated random noise= qrm

    It really is that simple.

    Some of us understand the absolute relationship between frequency and time.

    Don't fall for the sales reps b.s.

  2. N3KE

    N3KE Ham Member QRZ Page

    Sorry, some us have years of education in RF communications and actually do spread spectrum modulators and demodulators for a living. *Believe* what you want but don’t try to convince others you have any *understanding*. Do yourself a favor and actually study before making silly claims about the relationship between frequency and time and it’s relationship to DSSS.

    I didn’t say it was anything other than QRM. It is CDMA is a multiple access technique with all the same limitations of multiple users as TDMA and FDMA. It’s not magic, it is just a different way of doing the same thing as the other MA techniques. And no one in the comms community claims anything beyond that.

    But back to the original point - the instantaneous bandwidth of a DSSS signal is its chip rate. And it’s SNR as defined over the signal’s instantaneous bandwidth is very low. Arbitrarily low as you increase spreading. What doesn’t change at all as you increase spreading though are C/No, Es/No or Eb/No. This is precisely why “SNR” is such a poorly defined figure of merit and you won’t see people in the comms community ever really use it.

    Spreading, DSSS or FHSS, does nothing to help in a AWGN channel over a non-spread signal. It is useful for reducing power spectral density (there are regulatory limits for this in some bands), a relatively low effort way to deal with common narrowband interference, another option for multiple access and if the chip or hop sequence is non-repeating and of cryptographic quality has very good anti-jam characteristics.

    But fundamentally your original post and your attempt to double down here are absolutely wrong. And as you said yourself - it’s a common misunderstanding. Take some time to read up on DSSS. Note especially a DSSS is spectrally identical to a non spread BPSK or QPSK signal with a symbol rate equal to the chip rate of the DSSS signal. In fact an encrypted BPSK signal and a DSSS signal with a non repeating code are impossible to distinguish. Hence if your claim about instantaneous bandwidth were correct then suddenly BPSK signals have a instantaneous bandwidth lower than their symbol rate.
    Last edited: Jun 29, 2020
  3. AI3V

    AI3V Ham Member QRZ Page

    Please, you are the one trying to spread (pun intended) misinformation.

    Eb/no rules.

    Full stop.

    There is no modulation method that will recover a signal weaker than noise.

    Sammy hammy programs that report numbers like -24 are at best misleading.

    This mode depends on a extremely low data rate, and a very narrow rx bandwidth to allow marginal stations to communicate, there isn't anything wrong with that of course, but thinking the eb/no is a negative number....:rolleyes:

  4. N3KE

    N3KE Ham Member QRZ Page

    Actually I'm trying to point out exactly why reporting SNR is not a useful way to compare signals. I think we agree on that.

    You made a factually false claim about the instantaneous bandwidth of spread spectrum signals. I corrected it. I corrected it because the instantaneous bandwidth of a communications signal has little to nothing to do with either its information rate, Eb/No performance or much else. It does set a constraint on how close to the Shannon limit you can get for a given information rate but that really isn't relevant to this discussion.

    Absolutely. But people frequently report signal characteristics and performance using other figures of merit. Some of them are useful for evaluating other communications performance parameters (Es/No, C/No) while others have very little utility (SNR over any particular arbitrary bandwidth).

    You haven't defined what "noise" is here so the statement on its own is somewhat meaningless. And the Shannon limit is not 0 dB in fact so again you've ill defined this.

    Absolutely! But if you understand the context of how they are reported they do provide one way of comparing modes though.

    For instance for a ham who might not actually care very much at all about data rates then Eb/No isn't in fact the best figure of merit. They've got the antenna they have and only say 100W to work with. If they don't care whether the QSO takes 0.1 sec or 15 seconds or 1 minute then Eb/No is not a useful figure of merit since it normalizes across data rates while the ham is perfectly happy to lower the data rate in order to close the link.

    Is it actually reporting the Eb/No as a negative number? If so that's b.s. But remember SNR is NOT Eb/No. SNR is an ill defined term which sometimes people will equate to Eb/No, sometimes to C/No, sometimes to Ps/Pn or any of a number of other things. So if it is reporting negative SNRs and then you are equating that to Eb/No then that's on you. SNR is whatever someone wants to define it as. Eb/No on the other hand has a very specific definition.

    So if you see something comparing "SNR" you should not assume it means Eb/No. As alluded to above if a specific data rate is not a hard requirement then a different figure of merit is often more appropriate for comparing modes than Eb/No is.

    It sounds like we actually mostly agree on things here...
  5. AI3V

    AI3V Ham Member QRZ Page

    Yes, I do believe we are in agreement on the meat. :)

    Does a WSJT-X contact require a lot less power than a ssb qso?

    Yes, absolutely.

    Are you sending the same amount of info?

    Nope, not even close.

    Could the author of the software picked a "better" number to report the quality of the channel?

    I think so. :)

    Heck, why not just report the improvement of a 50 hz bandwidth compaired to the entire hf band 3-30mhz, that gets you what, 58 db below the noise? :p

  6. W2VW

    W2VW XML Subscriber QRZ Page

    Wondering when REGE-X 1.1.0 will be ready for download.
  7. N3KE

    N3KE Ham Member QRZ Page

    Totally on the same page with you!

    Having had to present these kinds of comparisons over the years I’ve come to the conclusion while there is never a single “right” way to compare there certainly are wrong ones!

    It would have been nice to have just reported C/No in WSJT-X. That’s really the figure of merit they are trying to use and is sensible for a typical ham scenario in which the station is power constrained but pretty flexible in information rate. The way it reports SNR using the SSB bandwidth is really just an offset in dB from C/N0 but why the obfuscation?
  8. W6RZ

    W6RZ Premium Subscriber QRZ Page

  9. N3KE

    N3KE Ham Member QRZ Page

    Actually counter intuitively the units in the numerator and denominator are the same so the ratio is unitless! Eb is energy (for example Joules) while No is a power spectral density (for example Watts/Hertz which of course equals Watts*Seconds which equals Joules).

    Eb, the energy per information bit, is akin to signal power (but of course is energy instead) and No, the noise spectral density, is akin to noise power (but of course normalized to 1 Hz of bandwidth). So yes it is very much like a signal to noise ratio.

    What's useful about Eb/No as a figure of merit related to SNR performance? It is independent of data rate. For any modulation and coding I construct I can always transmit the data successfully ten times faster as long as I use ten times as much power and ten times as much bandwidth.

    To see how close Eb/No is to an SNR pretend we send a BPSK signal at 1 Mbaud and want to convert Eb/No to SNR. Here we will define SNR as the signal power divided by the noise power over the bandwidth of the signal. This is a common definition of SNR but by no means the only one. Assume the Eb/No measured for this signal in whatever conditions is 10 dB.

    It's 1MBaud BPSK so the power of the signal is Eb * R where R is the baud (bit) rate.

    The noise power is going to be No times the bandwidth of the signal. For BPSK the noise bandwidth is about R typically. So the noise power is No * R.

    Look what happened: SNR = Eb/No * R/R = Eb/No

    The crux here is that as long as we scale the bandwidth in which we measure noise power to match the bandwidth of the signal then Eb/No equals the typical definition of SNR to within a scale factor.

    The rub some people complain about in WSJT-X is that it *does not* scale the bandwidth in which noise power is measured to match the signal. So FT8 which only uses 50 Hz of bandwidth gets its noise measured in 2.5 kHz of bandwidth so the ratio of S/N goes way down.

    Eb/No fundamentally tells me how efficient a given modulation or coding scheme is without having to arbitrarily pick a data rate or a noise measurement bandwidth. It is a reasonable way to make "apples to apples" comparisons of modulation and coding schemes normalizing away any differences in data rates between them.

    Now of course the absolutely most effective way to get a link to close if you have a fixed transmitter power is to send the data slower! If I send data at 1/10th the rate the Eb goes up by a factor of ten. Modulation and coding schemes fight over just a few dB of Eb/No performance assuming similar use cases (spectral efficiency and payload size being the major driving factors limiting Eb/No performance) and so we can see that even a really poor modulation and coding scheme can still close the link as long as we just slow down the data (again assuming fixed power).

    The "magic" of FT8 is mostly just in sending the data very slowly. That's where it gets a lot of its advantage from. Now it also happens to be a very, very well structured modulation and coding scheme that uses nearly optimal source coding and error correction coding for the task it was designed for. So it has optimized its Eb/No quite well, but fundamentally most of the lifting is done by sending a tiny smidge of data over 15 seconds and thus keeping Eb high even at low receive powers.

    And for hams who really have little urgency in which to make a QSO sending data really slowly is actually a really effective way to work at low receive powers. JT has made sure that while playing that simple trick of sending data slowly we might as well also design the modulation and coding to be about as efficient as it can and eek out more dB of performance.

    Yes Es/No is a different SNR figure of merit related to Eb/No.

    Again it is normalized for data rates making it more applicable. However it doesn't normalize for the effects of modulation and coding choices and so otherwise equally power efficient schemes that have nearly identical Eb/No can have wildly different Es/No.

    In Eb/No the Eb is the energy per information bit. That is the information (e.g. bits to describe my callsign) that I'm trying to send through the channel.

    Es is the energy per symbol and a symbol can be composed of far more or less than a single information bit. So it really tells you nothing about modulation and coding performance at sending information efficiently through a channel.

    For example an uncoded BPSK symbol contains one information bit, an uncoded 1024QAM symbol contains 10 information bits and a rate 1/10 coded BPSK symbol contains 1/10th of an information bit.

    So if the Eb/No were say 10 dB in the above three cases the Es/No would be 10dB, 20dB and 0dB respectively.

    So why care about Es/No at all? For one it tells you how hard a job a demodulator's acquisition and tracking loops are going to have. Lower Es/No signals are harder. Also it roughly correlates to how far above the noise floor the signal is going to appear on something like a spectrum analyzer.

    Yep so in the above spec here while Es/No tells you nothing about how optimal the coding schemes are (answer for DVB-S2 and DVB-S2X extremely close to optimal for all modes) it is a very useful table for deciding what mode you can use over a given link at a given receive signal power and noise floor. For DVB-S2 users typically they are allocated a given bandwidth (often on a satellite transponder). This Es/No table tells them essentially how much additional data rate they can pass with additional power under the constraint they can't increase their transmit bandwidth. So it serves a very useful purpose for finding the operating point of the link but it doesn't tell you anything about how wonderful the DVB-S2 modulation and coding are.

    Clear as mud?
    Last edited: Jun 30, 2020
  10. W6RZ

    W6RZ Premium Subscriber QRZ Page

    This source claims otherwise. See page 36.

    "The quantity Eb/N0 is sometimes called the “signal-to-noise ratio per information bit,” but it is not really a signal-to-noise ratio, because its numerator and denominator do not have the same units."

    I believe he's saying that the dimensions of the numerator and denominator are not the same.

    The question still remains, can you have a real-life system with a negative signal to noise ratio?

Share This Page