[time-nuts] On some pitfalls of the dual mixer time difference method of horology

[Poul-Henning Kamp]

I would find two of the best Audio ADCs I could lay my
hands on.  Some of these are incredibly good by any standard.
Sampling the mixer outputs at approximately 96kHz with
16 bit effective resolution is well within the possible.

All the repeated steps of amplification/limiting to find the zero
crossing can be almost entirely replaced by a a single FFT.

You are not limited by the 96kHz sampling frequency in the timing
of the resulting zero crossing, because you can interpolate between
the sample points.

On a Loran-C signal, I have been able to do zerocrossing
measurements with 1nsec resolution, despite the fact my sampling
clock was only 10MHz.

You can calibrate out almost all the asymetries
by simply flipping the input signals and pretty much all of them
if you can also do A/B switching on which mixer feeds which ADC.

I wouldn't at all be surprised if this is how TSC does it.

Obviously, damn good electronics is still necessary to control
cross-talk etc.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.