Tuesday, March 26, 2024

How to Feel Bad about Carrier Suppression (and How to Get Over It)


Version 2 of the 15-10 rig continues to give me trouble.  But I guess it is "good trouble" because I am learning from it. 

You see, after you build an SSB rig, one of the critical alignment steps is the placement of the carrier oscillator/BFO frequency in relation to the filter passband.  (The IMSAI Guy has a good video on this -- see above.)  You want to place this oscillator somewhere on the downward skirt of the passband curve.  This will add to the carrier suppression already done by your balanced modulator.  There will always be some carrier remaining from the balanced modulator -- putting the oscillator frequency on the downward skirt of the filter passband adds to the suppression the carrier remnant.  

But you can't overdo it.   If you place that carrier oscillator frequency too far down on the skirt, you will start to cutoff the low audio frequencies in your transmit and received signals.  You will notice that your once beautiful sounding receiver suddenly sounds tinny and high pitched.  Yuck. 

So you go back to the books and the websites.  You look at the passband promissed by the Dishal software you used to design the filter.  See below:  

Click on the image for a better view

Surely with a passband as nice as that one, you will be able to find the sweet spot where the carrier is suppressed and your audio remains pristine.  

But I couldn't do this with the 15-10 rig.  I was forced to compromise: I had to accept less than optimal carrier suppression for less than optimal low frequency passband coverage.  And here is why:
Click on the image for a better view

The curve above is a much more realistic picture of what my filter passband actually looks like (see NanoVNA picture below).  The curve above is from the AADE software.  I found out that the Dishal software DOES NOT factor in important things like Q or ESR.  Dishal treats all crystals as "loss-less radiators."  So when you get up to 25 MHz where Q is important, AADE and other programs will show you that your passband has become curved.  And you can see how this curvature makes it difficult to get the kind of carrier suppression and audio response we want.  

What my filter looked like in a Nano VNA
Click for a better view

Look, the rigs work OK.  The receivers sound good to me.  The carrier is so far down that no one can hear it.  I have to remind myself that we are using 'SSB-SC" -- suppressed carrier, not eliminated carrier.  I've worked a lot of DX with these rigs.  But still,  I would like to improve the situation.  It kind of bothers me.  Homebrewers will understand. 

I have been experimenting with different balanced modulators.  I started with the simple two diode, single transformer, singly balanced design from Farhan's BITX20.  It works fine.  But I think I get a bit better suppresson from a doubly balanced diode ring.  I may try an NE602 Gilbert Cell.   I may also try to build a higher Q 25 MHz filter using low-ESR surface mount crystals from Mouser.  Stay tuned.  

Thanks to W7ZOI, VU2ESE, WN2A, KA4KXX, KK4DAS, N6QW, W2AEW, and G3UUR for all the good advice and encouragement.  Please put any additional ideas in the comments below.  

14 comments:

  1. I would like to know more about what is causing the very-round shoulders of your passband. I really have experience only with QER types that don't take into account ESR or any of the motional parameters at all and that use equal-value shunt capacitors.

    Mathematically, they have ripple, but with close impedance matching this is virtually absent from a nanoVNA plot. In part, I understand this "equiripple" characteristic is the result of the parallel crystals at each end of the filter. That's their theoretical purpose, at least. But I think another feature of the QER scheme is that their passbands have very-sharp-cornered shoulders.

    I have a test fixture that makes it easy to swap-out shunt capacitors on-the-fly and I can see on the VNA plot that the shoulders of the passband build up from the ends with each added cap. Sometimes, in fact, there's a slight notch at each shoulder--maybe a dB or so deep--before the skirts plunge sharply downward (see plots in the document at ). I don't know if this is an artifact of the VNA or if it's real. At only a single dB or less I don't really care about it.

    Do other filters you've built using the Dishal or AADE figures also result in round-shouldered passbands?

    --K7TFC

    ReplyDelete
    Replies
    1. Did Blogspot strip out the URL I included above? Here it is again: https://bit.ly/qer-filter

      Delete
    2. Todd: I think it is essentially a Q (loss) issue. You can see this in a QEX article by the creators of Dishal. https://www.arrl.org/files/file/QEX_Next_Issue/Nov-Dec_2009/QEX_Nov-Dec_09_Feature.pdf
      Check out Figure 4 -- that Qu 150,000 filter passband is what mine looks like. I don't think QER would fix this (I made a slap-dash attempt and noticed no improvement).

      And there is this:
      "The other factor that limits the narrowest bandwidth that can
      be achieved is the Q of the crystals. The minimum acceptable
      Q is a function of the type of design used and the number of
      poles required as well as the center frequency and bandwidth.
      A ’fair’ rule of thumb is the minimum Q must be larger than

      2N Fo
      BW , where N is the number of poles required, Fo is the
      center frequency and BW is the width of the passband. If this
      condition cannot be met the passband will be rounded, the
      insertion loss will be high, the stopband will appear to flare
      and any attenuation peaks near the passband will round off or
      even disappear. "
      From page 3 of https://www.networksciences.com/pdfs/tutorial.pdf In this one too, check out Figure 4 on page 3. Again, that looks like the passband in my filter. Sad, but true, I'm afraid. This is probably why we see most crystal filters designed at lower frequencies. My 5.2 MHz filter looks GREAT! 73 Bill

      Delete
    3. Ah, I missed that your results are for a 25MHz filter.

      Delete
  2. It is easier to get good carrier suppression using a Gilbert cell but you can improve a 2 diode bal modulator. Back in the 1960's we used FT243 crystals with lower activity than today's HC6U. The oscillator had to oscillate fiercely to make sure it started reliably. I found it needed to be in its own screened compartment to avoid carrier leakage. In those days filters (FT243 crystals) had poor shape factor.
    Have a look at the carrier osc level as well. The rounded filter will give you rounded audio (with presence?)
    73 Brian

    ReplyDelete
  3. I've always had good results using the 1496 Gilbert cell mixer as both a balanced modulator and a heterodyne mixer in transmitters. I know ICs are a fair step down the path toward soullessness, but if you have begrudgingly made peace with the NE602, then the 1496 is only a handfull of more pins.
    In fact it also requires a handful of additional biasing resistors which are not required by the 602. But once built, I have always found I could null out carrier to a very deep null with these devices (I always use a Bourne style 10-turn miniature balancing trimpot for precise adjustment), far more so than the few 602 balanced modulators I tried, which were disappointing in my projects.
    I have a soft spot for the 1496. I first used one in my very first attempt at a solid state homebrew SSB receiver and transmitter in the late 1970s, and those stages worked really well, and far better than my attempts at VFOs or PAs. They can be found all through the homebrew solid state SSB projects in the legendary HamRadio magazine from this period.
    As for Dishal giving an idealised plot, I never knew that!
    Paul VK3HN.

    ReplyDelete
    Replies
    1. Wow, thanks Paul. That 1496 is calling me OM. And I won't feel bad about it being an IC, because I have already figured out how a Gilbert cell works: https://soldersmoke.blogspot.com/2021/11/how-to-understand-ne-602-and-gilbert.html

      Do you have handy a bal mod circuit using the 1496? Any bidirectional circuits for this part? Thanks, 73 Bill

      Delete
  4. Vasily Ivanenko (!) has a great blog post (updated just last year) on this kind of chip:
    https://qrp-popcorn.blogspot.com/2022/10/some-analog-ic-gilbert-cell-mixer-notes.html

    ReplyDelete
  5. I just noticed that Mouser currently stocks a 14-pin SOIC version of the MC1496 for about a dollar each, manufactured by ON Semiconductor.
    Circuits are shown in Solid State Design on pages 45, 72, 97, 185, 203, and 204. I am tempted...

    ReplyDelete
  6. I got 5 of the old style (with pins) for 10 bucks on Ebay. Walter's SSDRA references are interesting: page 185: " For an SSB transmitter using the filter method carrier suppression of 50 db or greater is sufficient. This is because the filter will often add another 20db of carrier suppression." Yea, that's my problem.

    ReplyDelete
  7. SSDRA page 186, regarding the simple two-diode circuit: "We have measured over 50 db of carrier suppression at 144 MHz with this circuit." Yikes. Makes me wonder if improvement using the MC1496 is possible. I mean, if you can get 50 db of suppression from a simple two-diode and a transformer circuit, it seems that you wouldn't be doing any better with the MC1496.

    ReplyDelete

  8. Good point. Referring to SSDRA, pg 185, Fig11, it seems that 'D' would be preferable as it applies the LO to the transformer with the CT grounded, not fed with AF. That helps balance. Then why not add the small trimmer of 'B' between the diodes of 'D' to null the carrier? I think Pete may have suggested something along these lines(?). Again, matched diodes, carefully balanced windings, and a tiny cermet 1-turn trimmer. Adjust for min carrier leakage on a spectrum analyzer.
    Amazing, SSDRA provides insight 47 years later!

    ReplyDelete
  9. Also from SSDRA pg 74, the single-balanced mixer using 4 diodes may have better carrier suppression. Add the trim pot to that circuit.

    ReplyDelete
  10. Back to the basics eh? This discussion reminds me again about the remarkable performance of the original BiTx design, which used hand made diode mixers.
    At the risk of pushing this thread back to modern devices those rice grain dual diodes, as used in the uBiTx can be highly matched, as the two diodes are manufactured side by side on and of the same material, as it were.
    No big deal, just grab that bag of diodes and a multimeter and off you go.
    Paul VK3HN

    ReplyDelete