[time-nuts] Frequency subtraction with D-flip flops

[ed breya]

I have been experimenting with three mixing devices: an RF 
double-balanced mixer (MCL SRA-1B), a DFF (74AHCT74), and an EXOR 
(74ACT86). The mixer and EXOR give similar results - the sum and 
difference frequencies, and a slew of various other products that 
need to be filtered out. What was appealing about the DFF is that I 
thought it should give only the difference, so the upper sum 
sidebands should not be possible (except for the harmonics of the 
difference frequency).

When I tried the DFF, this appeared to be the case, but it also 
contained some of the lower frequency spurs that the other mixers had 
too, but at much higher levels. I'm not sure if it's due to that 
metastability thing, or the asymmetry of the input frequency sources, 
due to their waveforms (one is nearly triangular) or the logic 
levels. In the time domain, maybe the edges (one or the other) 
correctly represent the difference, but it looks nasty in the 
frequency domain, and turned out very difficult to filter out, being 
below the desired output - I was hoping to only need low-pass 
filtering to clean it up.

The best bet so far seems to be the EXOR, since it provides some 
conversion gain, compared to the RF mixer, and a reasonable amount of 
spurious content. I'll be trying various tricks with the DC bias, 
levels, and symmetry to see if it will clean up a little easier. This 
section has to provide two outputs - one digital for the PLL, which 
can be ugly as long as the edges are right to land at the right 
frequency, and one very clean sinewave to run the PLO. It looks like 
this will need lots of filtering regardless of the mixing method.

Ed