> If I recall correctly Hans has shelved the QSX
?
No, I've never said that. To be sure, literally speaking, it has spent some time sitting on my shelf; but in the more commonly understood meaning, it has not been shelved as in retired, given up on, etc. Just that I suspended work on it while I have generated lots of other interesting designs in the meantime, until I get back to it and finish it. But more on this below.??
?
> Are you able to give us an update.
> Your last update was very welcome.
?
Sure. Well, I had a somewhat disappointing week in which I had to spend most of the days on some very urgent SteppIR work, on which I made a lot progress and it's almost completed, but nonetheless it consumed most of the time. Which I don't regret because I love the guys at SteppIR, they're a great bunch of people and a great company with very fine products; and also because I have always found that if I follow my passion and don't question my gut feel too much, it has a strong tendency to lead somewhere interesting. In this case perhaps things I learn along the way are going to be very useful applied back into QRP Labs transceivers, so it all becomes worthwhile in the end with hindsight, even if in present day we can't foresee the future.?
?
Interleaved between that, I did make some progress on the QMX SSB firmware. I had concluded that the SSB modulation is about as good as I can make it - which is very good indeed with intermodulation performance that exceeds several?respected commercial transceivers costing 20x the price.
?
I was reminded more than once of an event at school when I was around 15, in the "Technology" class. We were paired to do a major end-of-course (2 year course) project. My friend and I had decided to make a pneumatic robot arm but that turned out to be hopelessly over-ambitious (and perhaps an indicator of? future tendencies); we made nearly no progress, certainly not enough to suggest success within the limited timescales. In the end 3 weeks before the deadline, we abandoned it, changed course and made a burglar alarm system instead with various detectors like ultrasonic movement detector, infra-red beam, magnet reed switch, pressure mat, etc. We did OK on that, even with the very little time available.?
Anyway that's a digression from the digression. The real story is the chap who, unpaired (doing his project alone) had decided he should build a radio-controlled duck out of balsa wood. Then he would plan to sail this duck on a pond where there were other (real) ducks, and thought the other ducks would follow his fake duck and not notice it was fake, then he could lure them closer and take a nice photograph of the ducks. Such was his plan.
I recall not really paying much attention to his plan for the balsa wood duck but it all went wrong one day, when cutting balsa wood with an extremely sharp industrial strength box cutter-style knife. A small slip and the blade half chopped off his thumb. It cut down at least an inch and a half between thumb and forefinger. He didn't yell or scream, just stood and stared at it, frozen with a perplexed look on his face. His predicament was quickly noticed by the teacher, who was actually the head of the physics dept and a good friend of mine (I ended up doing physics at uni); anyway he was a very practical and calm fellow and did not panic at the situation. He just grabbed the kid's arm and held it up high, to reduce blood pumping that way; then he tried to squeeze the wound together, and started making plans to attend hospital which fortuitously was only about a 5 minute drive from our school.
Which comes to the point. As he squeezed the two pieces of hand together, bits of flesh, blood, bones, tendons, muscles and whatever else is inside hands, sort of came out the other side. So he would then squeeze that other side but then bits of innards came out this side. And no matter where he squeezed, the poor kid's insides seemed to ooze out one place or another. And in the end there wasn't much point trying any further, so off they went to the hospital to get repairs done properly (which did take some time).
And this is why it reminds me so much, on many occasions, in other areas. Specifically in this situation, here I was with my? 700 / 1900 Hz two-tone test signal, trying to reduce IMD product spurs. Some things reduce IMD3 but worsen IMD5, 7, 9. Some the other way around. Sometimes it's possible to improve what happens above the twin peaks at the expense of what goes on below, and vice versa. Squeeze somewhere, and elsewhere things go wrong.?
?
So eventually like the venerable Mr Plougman, head of physics dept, it became time to shrug my shoulders, accept I made SSB as good as I can right now, which is actually very good and way better than expected, and move on to the next phase.?
?
I spent some time researching compression techniques and ALC, AGC, and then CESSB. Because all the questions now center around how to connect up the microphone samples to the SSB transmitter and get the best results.?
?
So currently I am working on CESSB (Controlled Envelope Single Sideband). The best reference to learn more about CESSB is the original ARRL QEX article published in November 2014 by?David L. Hershberger, W9GR
.?
See also:?
?
CESSB is a technique to prevent overshoot (clipping and splatter in the PA) and increase average
power by several dB relative to peak envelope power. It is implemented in a rather small number of transceivers currently limited (as far as I know) to SmartSDR software by Flex Radio Systems, Apache labs ANAN series and the Elecraft K4.
?
Now before a subset of you rapidly start angrily typing a reply along the lines of, why are you bothering with CESSB before you have even released SSB? There?is method to my madness.
?
One has to do *something* to establish the appropriate gain for the transmit chain from microphone to antenna. I do not feel that just having a manual gain setting and relying on the operator to speak in level tones at a fixed distance from the microphone, is really going to be a sufficiently "good" system. And since 5W of SSB is not exactly gigantic, it doesn't have the relative SNR advantages of narrow bandwidth modes like CW and many digi modes; so SSB needs all the help it can get. The key aim is to increase the average power level relative to the peaks, without causing clipping and splatter. There are several things that can take place in an SSB transceiver:
- Some kind of automatic gain control which is similar to what's done on receive
- Actual compression, in which quieter speech is amplified by a larger amount than loud parts of the speech
- ALC -? Automatic Level Control in which the PA itself acts to limit the amplitude to prevent clipping; in the case a quite sophisticated delay-line look-ahead system could be used.?
CESSB replaces the need for ALC and increases average power. Inevitably ALC will REDUCE the gain to an extent that SSB peaks don't cause clipping; effectively it increases dynamic range not decreases it so arguably it's the opposite of the compression we want.?
?
Furthermore the unusual Envelope Elimination Restoration -like way in which SSB is being generated in QMX already sees the SSB signal split into separately modulated phase and amplitude components; we are therefore already at the point of messing with the RF envelope level so it actually makes CESSB several steps closers already. Because of this it may end up being the EASIEST way to achieve my SSB goals.?
?
For all these reasons it makes sense to consider CESSB at this stage, as long as it does not cause an unreasonably long delay in development.?
?
Current status: I wanted a speech sample to use for testing. Googling, it appeared surprisingly difficult to find a simple speech sample; there's lots of music, lots of things illustrating different compression levels and distortion effects; but little in the way of just some plain speech. The first plain thing I happened to come across was Winston Churchill's "Be Ye Men of Valour" speech from 1940
?
This is a 10+ minute speech which means I don't have to restart it too often. It doesn't suffer background noise or music and appears to be a clean recording. To start with, I opened it in Audacity (PC audio software). It's sampled at 11025 samples/second and isn't "full volume" in so far as it does not touch -1 /?+1 levels (full-scale). See attached "Winston Orignal.jpeg".?
?
I still haven't listened to the whole speech by the way. I just wanted some recording of someone talking and it was the first I found.?
?
Next I re-sampled it to 48ksps sample rate; I know the PC will automatically do this for me when I play the sample to the QMX (USB sound card) but I felt happier if I knew I already had it in the required format under my control. I always try to reduce the number of unknown variables wherever possible. I then used Audacity's Amplify function which automatically tells you the amount of amplification to apply to bring it up to full-scale; such that the highest and lowest peaks of the entire recording are exactly at (but do not exceed) the?+/- 1.0 level (fullscale). I then added a 2 second 1kHz full amplitude sinewave to use as a reference tone, at the start of the recording, before Churchill embarked on his waffleology. This is shown in the attached file "Resampled and 2 seconds 1kHz.jpeg".
?
Now I could play this sample into QMX. Of course I now had to connect up the USB sound card in QMX, to the SSB modulator which until now, has been busy only on synthesized 700 / 1900 Hz two-tone signals and other such things. Firstly to convince myself the audio was coming correctly into QMX I routed it to the audio output directly; at this point I could listen to Churchill in my QMX earphones (and understand the significance of the large overweight overpriced USB headphone dongle thing on my bench). Now I did some DSP for a factor of 4 decimation down to the 12ksps sample rate used for the transmit processing. Bear in mind that the fastest the '5351 synth can be updated is 16ksps. The faster you go, the more accurate the amplitude and phase modulation is (smaller steps) but the slower the filters and DSP are for the equivalent same shape factor of the filters etc. So a compromise has to be chosen.?
?
I set the gain such that my 1kHz test tone at the start produces 0.6W PEP. I'm far below the 4-5W I can get on 40m. So I could now see any overshoots beyond "full power" 0.6W PEP, without them actually driving the amplifier?to a clipped state.?
?
Interesting results... the attached "scope1.jpeg" shows the first 9 seconds, the 2kHz tone is at "full power" and I set the dotted horizontal blue line cursors of the 'scope to be able to clearly see any overshoots. You can see the RF envelope shown on the 'scope trace closely resembles that of the Audacity software PC audio clip in attached image "Resampled and 2 seconds 1kHz.jpeg". But - crucially -? you can also see the overshoots! They are more visible too on a long clip, attached "scope2.jpeg" shows nearly 3 minutes of Winston. You can see that the blue cursors are at 15.6Vpp (0.6W PEP). But the peaks extend to 19.01Vpp. So the overshoots are 22% beyond the "PEP" level. That means ALC would have to apply a gain factor of 0.82 to this clip to prevent any clipping of the RF envelope. Which would reduce the average power of the transmission, decreasing the SNR at the receiving end. Reducing the power of your watts PEP...?
?
This is all exactly what is talked about in the CESSB article; but it's important to be able to demonstrate it because now when I put in some CESSB code I should be able to quantify the improvement. To re-iterate - the SSB modulation process, no matter whether it is done digitally or by purely analog techniques (whether phasing or superhet with crystal filter), inherently it causes overshoot (clipping) necessitating reduction in PA gain to avoid clipping that causes splatter. Even with a completely perfect input signal, such as Churchill's speech, which has no clipping at all, it still occurs at RF. Clipping at baseband audio and clipping of the SSB envelope are NOT the same thing! And this is what CESSB is aiming to improve.?
?
Note that I also measured the spectrum - transmitting Churchill at 0.6W PEP on 7030 kHz into my dummy load, 26dB of attenuation to the spectrum analyzer; the attached "Spectrum.jpeg" shows the result, 20kHz span (2kHz / horizontal division). The video bandwidth setting is 1Hz, resulting in a slow 140 second sweep time so the spectrum indicates average power not peaks.?
?
I could also zoom in on the end of the 1kHz 2 seconds tone burst at the beginning of the audio sample. You can see that even though at audio, there is no clipping (overshoot), at RF the envelope exhibits an unwanted bulge at the end which in a PA at full power, would result in clipping and splatter.?
?
Next as an experiment I amplified Churchill by 3dB in Audacity; which is a very crude form of compression because now his peaks which were close to full scale, get clipped. Still, nothing at the audio input exceeds full scale (+/- 1.0) though now the higher frequency components will be emphasized because the clipping instances are getting like a squarewave at their edges. Now the attached "ScopeCompressed.jpeg" shows the overshoot situation got a lot worse, it is now 58%. The spectrum - attached file "SpectrumCompressed.jpeg" is still as expected, OK, because the "overshoot" is not really overdriving the amplifier because the gain was set so low, for only 0.6W PEP so that I could observe the overshoots. The average level is seen rk be higher, as expected.?
?
All of this is exactly what the QEX CESSB article talks about; so there's nothing unexpected here; but it's very nice to be able to see it and "feel it" and have a baseline to compare against when I put in improvements.?
?
> Yes, I would also be interested in an update.
> Will there actually be a difference between a?
> QMX with SSB and a QSX?
?
That's a very interesting question. I had previously expected that the transmit performance on SSB for QMX would not be as good as SSB generated conventionally by a quadrature sampling exciter and linear driver/PA chain. This would mean QSX still had value though undoubtedly the gap would be narrow and perhaps indicate an upgrade in QSX spec would be desirable.?
?
However now it seems that QMX SSB performance will exceed what was possible with QSX. Which is a very interesting concept to put it mildly (actually I find it mind-blowing). So what then IS the remaining point of QSX as it was originally planned? Almost all originally planned QSX features will have been delivered by a 160-6m QMX+. A remaining difference is that QSX has 10W output compared to 3-6W on a QMX+.?
?
So it's an existential question I spend a lot of my thinking about. A lot of thinking on long drives, lying in bed hot summer nights insomniac-style, and other such occasions when it is not wasteful to spend thought: not saying I stare at the lab ceiling for hours when I should be hammering out a thousand emails, help desk tickets or lined of C code for QMX SSB.?
?
I suspect in the end the answer will probably be, that QMX+ does indeed cover almost all of what QSX intended. So as often mentioned here and everywhere, though QSX was a lot slower than originally expected, and many other projects and products occurred in the interim, seemingly to some commentators as unwelcome interruptions, most of these actually became stepping stones towards QSX. And now we may have accidentally taken one stepping stone too far and achieved QSX before its time and name, in the form of QMX+. Apart from that 3dB output power, it's almost all in there, and more.
?
All of that still doesn't mean QSX has been or should be abandoned. It's not necessary to step down with any degree or sorrow, regret or remorse. On the contrary, QSX having been substantially delivered via QMX+, relieves the demand and pressure for launch of QSX. Which allows the imagination to run wild and free with a much more ambitious product. Which could be quietly developed to a much higher specification, to provide a more substantial differentiation to QMX+.?
?
Some may have heard me talk of 7 radios. 7 plans. Which isn't because the Harry Potter novels are 7 in number. And doesn't limit me to developing no more after that 7. Just that in my own head having gradually formulated a multi year technical and business plan, I had conceived 7 transceivers. 5 you have. QCX+, QCX-mini, QDX, QMX and QMX+. The shape and form of the remaining two, which may or may not materialize unexpectedly in the coming year or two, is imperfectly defined and in a constant state of change. One was to be QSX; but now given the success with QMX+, I started to dare dream that the QSX plan can evolve more into a super QSX.?
?
Executive summary: exciting times and I ain't done yet!
?
73 Hans G0UPL
WV0H
Donald S Brant Jr
Bill, N4QA/QRP