Thanks for the comments.
I bought a couple of these: HW-576 mono amplifiers from Aliexpress. They cost around £5 each, and are supposed to be "100 W".... With a 24V supply, the positive speaker terminal sits at 12V DC without audio applied, and when I lashed one up to a thrown together "U-LULU" circuit (though using a PLL synthesiser rather than a crystal), I got round 24 Watts carrier and could fully modulate it without anything getting hot! The 3116 IC on the amplifier board has a small heatsink on it, and I expected it to roast, but was pleasantly surprised at how well it survived the abuse!
I have recently spent some time trying to squeeze a little more power out the U-LULU. Of course the complexity of dealing with the interaction between the oscillator circuit, the six inverters used as a gate driver and the one inverter used a buffer between them was anticipated but it's still something of a pain in the backside.
You made a good decision to use an external oscillator. When I use an external frequency source, I too can get ~25 Watts without a struggle but using the crystal oscillator as I have configured the circuit will reduce that to probably less than 20W. I had previously assumed that lower power out than 25 W or so was just simply a tuning issue. In fact I have determined the main culprit is the duty cycle of the crystal oscillator, and I am only able to get ~38% duty cycle out of it, despite a lot of tweaking and optimization work. That wouldn't be a problem but that oscillator signal then goes through two inversions in series, the first being the inverter used as a buffer and the second being the six inverters in parallel driving the FET gate. From the standpoint of duty cycle, two inversions in series is equivalent to no inversion, so this leaves the duty cycle of the gate drive only ~38%, as opposed to an external oscillator drive, which is probably ~50% (but if you use a function generator like I did, it can be anything that you want). Lower duty cycle gate drive leads to lower RF power output. This explains the discrepancy in RF output power.
I tried bypassing the buffer and that delivers a higher duty cycle to the FET gate (100-38% = 62%) which is excellent but, because of the interaction between the oscillator, the gate driver inverters and the FET, the output needs to be retuned (big sigh). My first cut without retuning gave me 26 Watts out but the efficiency was terrible, ~50% versus ~90% previously, and this has lead to a thermal issues, SOA (Safe Operation Area) violations and some destroyed FETs as I work on it.
I'm still getting this sorted out.
I'm not excited about dealing with the circuit interactions and I'm really not interested in frittering away my life on a circuit like this but I also want to make this a little better for folks if I can. I will "pull the eject cord" soon enough and work on other things that aren't so fundamentally messy when I've decided I've either had enough of this tomfoolery or I beat this issue into the ground.
I added extra filtering to the output of the amplifier, since I really didn't want spurs either side of the carrier - offset by the amplifier switching frequency! My next iteration of the design will continue to use this audio module (I don't think that I could do much better), a programmable oscillator module (one of the really cheap ones programmable with an Arduino), and a PCB that's off at JLC at the moment. The high efficiency and small number of cheap parts required suggest that it might be worth extending this design for higher power, adding some simple audio processing and possibly a little MP3 player board that works with either SD cards or USB sticks. I can see lots of these being shipped to Eastern Europe in the 48m and 41m bands!
I have already made the schematic and created the PCB layout for a follow-up that is similar to what you describe:
1) frequency source of either a) a selectable crystal oscillator (with a real crystal driver chip that is designed to work at HF) or b) an external oscillator input.*
2) selectable audio preamp circuit for those that want to use laptop computer, mobile phone or MP3 player as the audio source
3) RF filtering on the input of the preamp to keep the amount of RF injection hopefully low enough
3) continue to use a Class-E output, tuned for a lower Q than the U-LULU that should be able to deliver roughly equal power output from 6200-6300 KHz,
4) probably with an output somewhere between 30 to 40 Watts, better than U-LULU, largely due to being able to run the gate drive voltage much higher than 6 Volts.
5) PCB footprint currently equal to the Sure audio module I recommend for U-LULU so that the two PCBs can be stacked one on top of the other.
*Or I might just ditch the external oscillator input (though it doesn't take up much PCB space) in lieu of the neat little synthesizers I found (which I mention here
https://www.hfunderground.com/board/index.php/topic,115861.0.html)
