This has been an interesting discussion and some good ideas have arisen. I
thought I might add my 2c into the pot.

I have been working in a related area, PLL of precision references using a
micro. As well as references for 5 and 10MHz, I have a simple transmitter
for 80/40/30m that uses a VCXO and phase locks it to an external precision
reference, such as a GPS or TV sync. I use 50 or 100Hz as the loop
frequency, and achieve stability of 2 parts in 10^7 with no real trouble,
even with a room temp cheap crystal, and incredibly low phase noise and low
sidebands (the 50Hz sidebands about 50dB down and the carrier phase noise
is -30dB only 0.1Hz out from the carrier).

Sure, lockup time is very slow, which isn't important in this application,
but you guys might like to consider some of my ideas on the subject:

* My VCXO is actually the micro's crystal. I use an ATMEL 2313, and use the
16 bit timer as the main divider. This is where you plug in the frequency
you want - for example, for 3.8MHz you divide by 3800, as the counter
reloads at 1kHz (actually, any multiple of 50Hz or 100Hz would do).

* The PLL operates by undersampling the counter at 50 or 100Hz. This adds
gain to the loop, since 10 or 20 times the phase shift occurs, but it does
mean there's an ambiguity in frequency every 50 or 100Hz. You might have to
rethink that. You could count right down to 100Hz with a VCXO as high as
6.5MHz (limited by the size of the counter, not the micro speed). To go
higher, use a divider ahead of the micro. The 2313 micro will run at 16MHz
easily, so 20m is no problem.

* I use a simple 8 bit software driven PWM D-A converter. It can easily be
extended to more than 8 bit, but the more bits, the slower it gets, so
becomes harder to get rid of the PWM sample clock. I run it at 1kHz, 8 bit,
which you would expect to give a strong 1000/256 = 3.9 Hz component, but I
get around that through "PWM randomization" - instead of incrementing the
PWM counter by one, I increment it by 13, which randomizes the output and
increases the effective PWM clock to about 40Hz. BTW, this technique also
works with hardware (R-2R) D-A converters, so can be used to extend 8 bit to
10 or 12 bit easily. 12 bit would need only 4 bit PWM.

* I keep the hardware gain (i.e. the voltage to frequency conversion of the
varicap) as LOW AS POSSIBLE and control the system gain in the firmware. I
also have filters in the firmware and techniques to remove dud phase
measurements which can happen when using a TV frame frequency reference.

* I use the UART to control the device, and also to get telemetry back, for
example the PWM loop voltage. The whole system operates in a 1kHz timer
interrupt, so the main processor can do other things (in my case it runs a
beacon, a real time clock, and sends telemetry).

* My transmitter circuit has only 3 chips. The micro, a 74HC00 that acts as
the 40mW transmitter, and a sync separator for the TV timing reference.

I think some of these things would have application in a 20m PSK receiver.
The tuning range required is small, would suit a VCXO, and there's no reason
why you couldn't step the frequency in 100Hz or 50Hz steps. Use a 1MHz
reference crystal divided to 100Hz to lock the VCXO. You'd just need to up
the settling speed by perhaps using a hardware D-A. You could change the
divider by commands from the UART.
Keep the discussion going! Very interesting!

73,
Murray ZL1BPU
www.qsl.net/zl1bpu/micro