[time-nuts] Question about precise frequency / phase measurement

[ed breya]

You may want to look at how that was done many years ago with 
frequency difference multiplication as in the old Tracor meters - I 
think the 528 was the main one. They synthesized a 9 MHz reference 
from one input, and then subtracted it from the other to get a 1 MHz 
result, which was used as a reference for another 10 MHz PL 
oscillator. The 9 MHz was subtracted from this, and so on to a next 
identical stage. The net result is that each stage of this process 
multiplies the frequency difference by ten times. These can be 
cascaded until you reach the limit of the noise performance of the 
stage designs - they managed around 10,000 times fairly readily. This 
method can be duplicated fairly easily with modern logic parts. With 
a setup like this you can produce a 1 or 10 MHz carrier that can be 
counted to very high resolution at one second gating - you just 
ignore the carrier digits and look at the multiplied difference 
frequency digits. Or, you can subtract the carrier and get just the 
multiplied difference - but you have to keep track of the phase info 
to know if it's plus or minus.

It's of course possible to use whatever frequencies and stage 
multiplication factors you want, but the tradeoffs are in making the 
numbers come out rationally (especially if you want a number of 
decade multiplier ranges), and the precision and quality of the 
intermediate frequency filtering and processing. About ten to one 
hundred times per stage is within reason. For example, I have an 
experimental (way unfinished) setup started with three stages of 1 
GHz PLOs for multiplication, and a 990 MHz reference. This will give 
100X per stage, reaching 10E6 difference frequency multiplication at 
one second gating, presuming I can manage the phase noise 
sufficiently. A quick two-stage setup indicated no problem reaching 
10E4, but that last 100X will be tricky - I have to build it up for 
real first, with extremely clean power supplies, shielded signal 
processing modules, and solid signal routing, just to see if it's 
possible. I would not recommend this approach - I'm only doing it 
because I happened to have all the main parts on hand. It would be 
better to keep everything down to 100 MHz or less for processing in 
ECL or ACMOS, and using crystal VCOs and filters.

Ed