[time-nuts] Bye-Bye Crystals
magnus at rubidium.dyndns.org
Wed Mar 15 13:38:28 EDT 2017
On 03/15/2017 05:30 PM, Richard (Rick) Karlquist wrote:
> On 3/15/2017 4:45 AM, Bob Camp wrote:
>> Where do you plan on getting an OCXO grade crystal at an odd frequency
>> that? Much of the performance of a good OCXO is in the crystal. Doing
>> a proper
>> design on one is a lot of work. You *might* think that having a design
>> for 5.000000
>> MHz would give you a good design for 5.000050 MHz. I have empirical
>> evidence that
>> this isn’t the case. Many years later, I’m still utterly amazed that
>> this is the way things
>> work in the crystal business ….( = it’s not just a design issue, it’s
>> also a business decision)
> Some 30 years ago when GPS was in its infancy, the hardware utilized
> OCXO's at 10.23 MHz. Somehow, HP got suckered into trying to make
> modified 10811's that ran at that frequency vs 10 MHz. Jack Kusters
> tried to explain to anyone who would listen that this was a major
> redesign of the crystal, because he would have to deal with a new
> set of anharmonic spurious modes. Although only a 2.3% frequency
> change, everything is different. In terms of business decisions,
> only something with as much "juice" as GPS could have gotten Jack
> to make a custom frequency. As it turned out, a few dozen crystals
> were made, and that was the end of it. I managed to snag them
> before they were thrown out, in case they might be useful for
> The original poster wanted not only an odd frequency (which I don't
> recommend for the reasons above) but also wanted to varactor tune
> the oscillator. I also don't recommend doing that because of the
> difficulty of generating a clean enough DC voltage. Against my
> advice, the HP smart clocks were tuned with DAC's driving varactors.
> They never really got that to work up to their expectations.
> Synthesis is so advanced now, 20 years later, that there is
> no reason IMHO to voltage tune an OXCO.
Agree. Synthesis chips is cheap now and should be sufficiently good.
Also, it may not be a good strategy to zero-beat to the carrier, so such
a frequency may actually be a bad choice. Modern DDS chips allows the
tuning of the LO1 such that different IF frequencies can be tried.
For instance, the RTL-SDR is typically operated in 0 Hz beat, such that
the front-end mixes down to 0 Hz for the carrier and lets the RTL chip
sample the signal. This causes a distinct "blimp" from the 1/f noise.
Shifting the IF from 0 Hz and then let the RTL downconvert after
sampling removes this blimp since the 1/f noise can be brought out of
spectrum. When I dug around, this was available as an option even in GNU
Radio if you only knew it.
This just to illustrate that best result is not always achieved by
bringing it straight on carrier. Sure, it makes the design very very
simple, but has its drawbacks.
Now, the 164 kHz carrier, as sampled by a 48 kHz clock produces a 20 kHz
beat frequency as it wrapps down in the undersampling. To digitally
convert it to DC or lock straight to 20 kHz digitally is fairly trivial.
Similarly, a typical GPS receiver often has the digitized signal offset.
With a more modern view on receiver design and considering the synthesis
tools available, you can play around quite easily and move things around
in interesting ways. For instance, consider that you have an IF filter,
you can with some care sweep the filter and "tune" the LOs to the IF
frequency working best for your needs. We do that in ham receivers to
shift filtering to where it helps from nearby strong signals.
More information about the time-nuts