[time-nuts] Bye-Bye Crystals

jimlux jimlux at earthlink.net
Tue Mar 14 20:57:56 EDT 2017

On 3/14/17 3:17 PM, Magnus Danielson wrote:
> So well yes, you learn the hard way what those 4-leggers do when you
> have a bit of requirements. Later I dug up the patent for the process,
> which was focused more on the production of one standard product and
> late setting the frequency for customer needs. For it's purpose a great
> concept, not for all cases thought.

actually, we use a lot of them in breadboards where you want to get 
"cycle accurate" timing from FPGA logic at oddball clock rates when 
developing software.

Historically, deep space radios have had a crystal that is related to 
the assigned deep space channel number.  This would be referred to as 
the f0 (f-naught) frequency and is around 9-10 MHz.

So if your assigned channel was 14, and your S-band downlink is at 
2295.000000, you'd have a VCXO crystal cut for f0=9.5625 MHz (240xf0). 
Your uplink frequency would be 2113.3125 or 221*f0.  The receiver LO 
would be 220*f0, the IF at f0, and you'd set your VCXO PLL to lock to 
the received IF.  If you work all the multipliers and such carefully, 
all the drifts and noises mostly cancel out.

If you had a different channel assignment, you'd have a different crystal.

Same for X-band, except the ratio is 880/749

It took 2-3 years to get the crystal, but you're applying for your 
frequency allocation years before anyway.   These days, the oscillator 
runs at 4*f0 or 8*f0.

In any case, when we started using digital tracking loops, you'd run the 
ADC and DAC at the same f0, and the FPGA at that same f0 (or some 
multiple, like 76.5 MHz).

Similarly, for near earth comm, they use PN spreading codes at around 3 
MHz where the chip rate is a integer fraction of the carrier frequency 
(just like GPS), and it's convenient to have the DACs running at an 
exact multiple of the chip rate, so you have an even number of samples/chip.

The GPS folks like a frequency that's something like 48.xxx MHz, because 
sampling at that rate makes all the signals alias down to somewhere 
convenient even at max negative Doppler.

Then you have a microprocessor that might be running at some other 
convenient speed (like 66 MHz).  Since it takes years go get your 
crystal (and what if you want to re-use the breadboard for another 
mission on a different channel) it's handy to have a plug in oscillator 
at the "right" frequency- yeah, it's got a lot of jitter, but at least 
you can do things like test the logic behavior for all possible values 
of register settings and stuff like that, which would be impractical 
with the simulator which runs very, very slowly compared to real time.

The other reason we might run at something like 45 MHz is that it allows 
us to use a converter rated at 50 MHz without having to explain why 
we're running right at the rated speed with no design margin.

Woe to the poor guy or gal, though who starts to run bit error rate 
tests on the RF generated from these convenient devices.

BTW, there are two spacecraft at Mars with the same channel number.. 
because one of them used the spare radio from the previous mission - if 
you have dual redundancy, and a project with a nice budget, you buy 
three radios.  Then, when you've launched, that spare can be adopted by 
a subsequent mission, so you might wind up with a spacecraft with prime 
and redundant on different channels (a pain in test and operationally), 
and yet another channel in the testbed.

Our newer designs use DDS synthesis, so we're going to standard clock 
frequencies like 50 or 80 or 100 MHz.

For missions doing radio science that carry a USO, all this same stuff 
about getting the frequency right also counts (most of our radios have 
an "ext ref" input for this purpose). ANd for the same "what if the 
channel changes" reason, Applied Physics Lab (who make USOs) has a uso 
design with a DDS in it.

Both we and APL (and I imagine anyone else in this limited business 
area, like Thales) have spent some time working on DDS designs with 
"good" spur behavior, but overall, having a radio that tunes the entire 
band is a good thing.

Today, I'd build a radio with a reference oscillator at a "nice 
frequency" (probably 100 MHz), multiply that up for a block converter, 
then digitize the entire band (or maybe a chunk) (X-band is only 50 MHz 
wide, Ka-band is 500 MHz wide) and do the carrier recovery and tracking 
entirely in digital.

More information about the time-nuts mailing list