This article documents several possible solutions to a problem of low sensitivity on 20m.
UPDATE 09-Jul-2023: Modification #2 is now incorporated into the official QMX assembly manual and schematics; firmware 1_00_003 is released which for a new QMX, automatically populates the Band Configuration screen on first power up, to suit this configuration. People applying the modification to an existing QMX will need to configure the bands in the Band Configuration screen according to the screenshot shown in the solution #2 section below.
It has been found that the QMX as built by the original assembly instructions has a poor receive sensitivity on 20m. Below is a sweep, this is about the best I could get using the published instructions, no matter how much I adjusted the windings. Others have also reported 20m sensitivity around 20-25dB down. The screenshot below (CLICK to enlarge it) is produced using the QMX's built-in signal generator and BPF sweep tool, available when you log into QMX with a terminal emulator:
This issue has been discussed extensively on the groups.io forum. One factor is that the 180pF and 270pF capacitors in the 20/30m Low Pass Filter (C516 and C525) are swapped, incorrectly, compared to QDX. This can also reduce the output power a little by changing the LPF response. However, swapping the capacitors to correct this mistake (which I have done, and is implied in the analysis below) does NOT materially improve the receive band pass filter response on 20m.
A theory put forward is that stray capacitance of the OFF switch MUX positions for the coil tap produce parasitic resonances, one of which occurs near to 14MHz and puts a notch in the 20m band. Some people did simulations making assumptions about stray capacitance (approx 5pF for an OFF switch input), and concluded the 80m winding plus assumed stray capacitance would cause a notch in the area of 14MHz.
It should be noted that the QMX is on a 6-layer PCB and therefore the traces are closer to the ground plane rather than 1.6mm away (the board thickness, of a 2-layer board). Thicker traces from the coil taps to the MUX switch are also used on the QMX. Despite this, my rough calculation of the capacitance of these approximately 1cm-long traces, with their given width and spacing from the internal groundplane layer, only indicates a capacitance around 1pF. Another factor is that there is ground plane under the T50-2 band pass filter inductor on the internal layers. In QDX there is no groundplane under the inductor on any layer. So this could also add some extra capacitance.
Despite the additional potential for stray capacitance on the QMX compared to QDX, I still find it surprising that the QDX does not suffer this problem but the differences are such that QMX does suffer it. This is still a mystery to me.
In my own extensive investigation I started with a single, 20m winding; then added progressively 30m then 40m. The 14MHz notch appeared as soon as I added the 40m winding. Therefore I must conclude that the parasitic resonance which is troubling the 20m performance is caused by the stray capacitance combined with the 40m winding. I did a large number of other experiments with different windings and capacitors to search for a solution to this problem.
1. One suggestion was to use a ferrite core such as FT50-61 so that far fewer turns would be required, and there would therefore be a much lower stray capacitance, which would then resonate at a higher frequency, hopefully out of the way.
2. Another idea was that BPFs are not required at all, because the Quadrature Sampling Detector has an intrinsic high-Q, narrow band pass filter characteristic already built-in by the behaviour of the sampling capacitors.
I explained in this groups.io post the reasons why I don't really buy into either of those two potential solutions.
PLL out of range
Before going any further I want to talk about something you may very likely see, in filter sweeps. See below:
Specifically I am talking about the right hand side of the sweep, where the signal drops to -80dB, in this example, at about 18MHz. It so compresses the vertical scale to accommodate this -80dB noise that it is hard to see what is actually going on at the resonance in the area if interest.
The reason for this effect is that the PLL range of the MS5351M synthesizer chip, being officially 600-900 MHz, will actually extend far beyond this on either side; but not indefinitely. The way the sweep is coded, The MS5351M MultiSynth divider ratio is set to a fixed (even integer) value for the entire sweep. As the frequency gets higher, the PLL feedback ratio therefore increases, and this means the internal PLL frequency keeps increasing; somewhere around 1200 MHz it quits and you get just a bunch of random rubbish, hence the -80dB noise shown.
Now: for historical reasons (QDX), and for the reason that nobody dares change what works, when there is already so much else to go wrong, the MultiSynth dividers for 80-20m are fixed and are hard-coded based on recognizing the band name in the QMX Band Configuration screen. For 20m for example, the divider is 64. Now if you consider 18 MHz * 64 = 1152 MHz you will see what I mean about the VCO running out of track when it gets up near 1200 MHz and it gives up.
Now what happens for a Band Configuration band name that is not 80, 60, 40, 30 or 20m (and ignoring 2200, 630, 600 and 160 for now), is that the divider is taken as the next lowest even integer below 600 MHz / operating frequency. That results, for 14MHz, in a divider of 42. The answer indeed, to life, the universe, and everything. Now you can see that if the VCO quits at 1152 MHz again, this time we would be at an operating frequency of 27 MHz. On the other hand - the lowest configurable PLL frequency is 15 * the 25 MHz reference which is 375 MHz, and 375 / 42 = 9 MHz so this means the lowest frequency of the sweep is 9 MHz.
Accordingly what I have done in the "20mwide" sweeps in the results presented below, is configured another band called "19" which sits just above the 20m band (according to its configured limits) and is set up such that it has a much wider sweep covering 9 to 27 MHz which really lets us get a good look at the whole shape of the band pass filter response for the 20m BPF.
It's a neat trick.
The best solution to get rid of the parasitic resonance that troubles 20m could be to move it further UP in frequency so it is out of harm's way. This can be done by reducing the L:C ratio of the series resonant BPFs. A smaller inductance, means a higher capacitance. The smaller inductance will resonate with the fixed stray capacitance at a higher frequency, therefore moving the parasitic notch higher up, as required. With care, it can be pushed far enough about 14 MHz. The effect of increasing the L:C ratio is to broaden the band pass filter peaks, making the BPF less effective at restricting out of band signals. Which is a sacrifice we have to make, and the QSD is anyway a high dynamic range, high IP3 mixer, so it is less critical than it would be on lesser mixers; but anyway we should not want to take that too much further than necessary.
So different capacitors can be chosen and matched with different inductance values to hopefully try and escape parasitic resonances.
However at time of writing (05-Jul-2023) there are approximately 500 QMX already out in the wild. It is not easy for many people to remove capacitors without breaking them or without damaging the PCB. And rework on the coil can also be risky. If different capacitor values are specified there will be a proportion of people who don't have the required capacitor in their junkbox and need me to send the right values, that quickly could get prohibitively expensive and my pockets aren't infinitely deep.
Therefore it makes sense to apply a constraint that the existing capacitor values (22pF, 30pF, 56pF and 220pF) must continue to be used; best if not removed from the board. At least such a constraint would be very nice to apply, if at all possible.
After a lot of different things were tried, and the resonance was VERY stubborn in its wish to notch 14MHz almost exactly, and quite resistant to attempts to moving it upward... the following very simple solution is quite elegant and performant. Put simply:
- Cut off the 40m and 80m sections of the L401 BPF inductor completely
- Add a jumper wire from the 30m coil tap to the 80m end of the coil winding
- Reconfigure the Band Configuration screen so that the 40, 60 and 80m Band Pass Filters all use the 80m tap (end)
This means that 40, 60 and 80m are actually all one single band pass filter, and they use the remaining 30m+20m sections of the coil, and resonate it with 220pF. This results in a single broad resonant peak centered on 60m band. A disadvantage is that the 80m and 40m bands aren't peaked on this resonance, they are off the side; however as the resonance is broad, they are not attenuated too badly. The sensitivity in my example for the five bands is:
- 20m: -6 dB
- 30m: -3 dB
- 40m: -5 dB
- 60m: -1 dB
- 80m: -7 dB
Remember these sensitivities are arbitrarily referenced, so they only serve as a useful comparison to each other and to other potential solutions. In the following screenshots you can see the Band Configuration I used. You do NOT need to set the PIN diode forward bias current to 5mA, this is just something I had left in from some previous experiment and I had forgotten was in there; it has no bearing on the receive sensitivity anyway as that setting only applies on transmit. Note also the "19" band which is as I described in a previous section of this article, it is used only to get a wide sweep from 9 to 27 MHz in the screenshots below.
The advantages of this solution are that it is very easy to implement for anyone who has already build a QMX; there is no need to remove or change any capacitors, and you do not need to remove coil L401, all you need to do is carefully cut off the 40 and 80m windings, and jumper the 30m to 80m taps. The performance is quite reasonable on all bands; the slight disadvantage being the 40m band is off the right-hand side of the 40-60-80m resonance curve a bit.
This one I like even better, and it is only slightly harder to implement:
- Cut off the 40m and 80m sections of the L401 BPF inductor completely
- I reduced the 20m section from 19 turns to 17 turns - since I later found I had to squeeze all that together, this may not even be necessary. It could stay at 19 turns and be spread out.
- I reduced the 30m section from 11 turns to 9 turns - again this may not have been necessary, the turns could have been spread out
- The 20m winding tap has to be disconnected (removed) from the 20m hole and connected instead to the "30m" hole.
- The 30m winding tap, which is now the end of the coil, has to be removed from its 30m hole and instead connected to the 40m hole
- The 80m hole gets jumpered to the 40m hole
- Reconfigure the Band Configuration screen so that 20m uses BPF #1 (not #0), and 30m uses BPF #2 (not #1), and 40, 60 and 80m all use BPF #3
This means that the 20m BPF inductance is reduced and it uses an increased capacitance of 30pF (instead of 22pF); the 30m BPF inductance is reduced and it uses an increased capacitance of 56pF (instead of 30pF); while the 40, 60 and 80m all used the reduced total inductance against the 220pF capacitor.
Now the single broad resonance of the total inductance with 220pF has one corner on the 40m band and the other corner on the 60m band, while 80m is some way down the left hand edge of the curve. Which doesn't matter much since anyway high sensitivity on 80m isn't a requirement. And on QDX 80m was also off down the edge of the curve (we do, after all, only have a maximum of 4 switch positions for BPF combinations, so cannot be perfect on 5 bands, some compromise is needed somewhere). The comparative sensitivity on my example is:
- 20m: -4 dB
- 30m: -2 dB
- 40m: 0 dB
- 60m: 0 dB
- 80m: -7 dB
This is a little better than solution #1.
Again this solution has the advantage that it doesn't need capacitor changes; it also offers a few dB improvement in performance over #1, but it is a little harder because you have to remove and reconnect L401 inductor windings to different places.
Solution by Fred WD9HNU
Fred WD9HNU has been experimenting extensively with this problem too with various combinations of inductance and capacitance. As Fred comments: "I could probably spend many more weeks winding and fiddling" - which is indeed the way it goes with these kinds of experiments. You can go on and on endlessly.
Fred says: "The new coil is 30 Turns, #30 wire (don't have #28) tapped at 15T, 18T, and 22T. I added a 39pF across the 22 for 20M, 120pF across the 30 for 30M, 270pf across the 56 for 40 and 390 across the existing 220 for 80."
So in summary:
- 20m: 15 turns and 61pF (39pF in parallel with the existing 22pF)
- 30m: 18 turns (15+3) and 150pF (120pF in parallel with the existing 30pF)
- 40m: 22 turns (15+3+4) and 326pF (270pF in parallel with the existing 56pF)
- 80m: 30 turns (15+3+4+8) and 610pF (390pF in parallel with the existing 220pF)
I also told Fred how to solve the PLL range issue, which he did and the 20m sweep below is labelled "21" for that reason (to avoid the hard-coded divider with "20"). The sensitivities are evidently:
- 20m: -5 dB
- 30m: -3 dB
- 40m: -2 dB
- 60m: -4 dB
- 80m: -4 dB
The BPFs are nicely peaked on each band and the sensitivity looks good on each of the bands. Fred did not provide me with a 60m sweep. The relative sensitivities may not be directly comparable with my own results (solutions #1 and #2 above) because it is a different QMX, with differently adjusted LPFs and slightly different everything, no doubt. You can clearly see how the significantly reduced L:C ratios have got rid of (or moved far away) the parasitic resonance notch; and also how the reduced L:C ratio significantly broadens the peaks. Since in QDX and QMX the vertical scale adjusts automatically to the minimum and maximum recorded values during the sweep, this broadening of the BPF shape is not instantly apparent until you study the dB numbers on the vertical scale and note the shallower slopes.
I like Fred's solution too but the main disadvantage is the requirement for several additional capacitors; these could be soldered on the bottom side of the board but it would be a bit of a squeeze, there isn't much space down there. Unless you used 0603 or 0805 ceramics and were uber-careful.
In conclusion my favourite is solution #2 above. It isn't hard to do to an existing built QMX. It doesn't need any new capacitors, or any new wire to wind L401 again. And the performance is good on all 5 bands. The resonant notch is moved way up out of trouble's reach.