[time-nuts] DMTD Ideas

[Magnus Danielson]

Lux, Jim (337C) wrote:
> 
> 
> On 10/1/09 7:24 PM, "Magnus Danielson" <magnus at rubidium.dyndns.org> wrote:
> 
>> Brian,
>>
>> Some quick comments...
>>
>>> The Op Amps at Point #2 would be a something like LT1000s or so and they
>>> b
>> Since getting rid of essentially all of the 20 MHz sum signal can be
>> done using trivial passive networks close to the mixer, you can
>> concentrate on the beat note or 100 Hz difference frequency. Notice that
>> the beat note output level actually depends on the loading network.
>> Essentially the trick is to let the 20 MHz see a short (cap) while the
>> beat note sees high impedance. The loss in level is essentially due to
>> the traditional 50 Ohm impedance, which doesn't match the mixers
>> effective output impedence and besides, for that frequency we don't
>> really care about reflections as lumped parameter models may be used and
>> we count highest voltage and not highest power.
>>
>> So, let's say you end up with 50 V/s and you want the counter to see say
>> 5 V/us (really 5 MV/s or 5 MVHz) to avoid jitter-trigger then the
>> slew-rate gain you need becomes 100.000.
> 
> What if you're going to run the beat note into an A/D (e.g. High performance
> audio interfac) and sample it, then do the sinewave fit to calculate where
> the zero crossing would have been?  Just a good low 1/f noise opamp to get
> the level up where the A/D performance is good?

That would do it nicely. Then it can be debated wither fitting the curve 
to find a suitable zero crossing is the best method. Doing that will 
defeat much of the point in doing DMTD since the cancelation of the 
transfer oscillator reduces as time difference between channels 
zero-crossing occurs. De-correlation may be done through multiplication 
with a complex digital transfer oscillator, complement one of them and 
then do complex multiplications sample by sample. That would keep 
correlation loss down to sample length level instabilities. Another 
method would be cross-correlation by means of FFT methods. This has the 
benefit of producing decorrelations for various time-shifts between the 
oscillators.

Cheers,
Magnus