DDS

The optics loop is so slow and wimpy in a rubidium, it seems to me that the mid frequency (ballpark 1 Hz and up) phase noise will be dominated by the XO inside. The low end rubidiums are probably optimized for frequency stability but not mid-frequency phase noise. I think.

Here's a rubidium manual, with schematics. It's all old fashioned parts, no FPGAs or anything, and it's impressive how simple it is.

The Allen variance gets good at 100 seconds, where the physics starts to dominate.

DDS must be more interesting mathematically, since the spurs will squirm all over the place as a function of the frequency settings.

The technique dates back to the 1960s or something, so the fundamental concept can't be protected. Tons of people do DDS in FPGAs, and there are Xilinx and such appnotes, and ADI isn't suing them.

Peter Alfke, a great guy now departed, has some DDS patents, for Xilinx. There is still a goodly amount of DDS patent activity by non-ADI inventors.

Plain vanilla DDS in an FPGA or ASIC is pretty simple and apparently unprotected. Phase accumulator, sine table, DAC. The ADI patents alone don't look like a barrier to entry.

I did a 64-bit version recently, a brute-force attack on the problem that a binary phase accumulator can't hit exact decimal frequencies. Nobody has a right to whine about a few nHz error. There are 8 of them in here:

Hmm, maybe. Most of the eBay ones are from retired cell phone towers, I believe, so the phase noise has to be good enough for CDMA at least. Telecom applications can be surprisingly demanding.

Fun. I've done a lot of measurements like that--using dither to differentiate a peak and servoing where the fundamental component goes through zero.

I've never been much of a discrete-math buff. All that Diophantine equation stuff leaves me pretty well cold. Complex analysis and Fourier and calculus and tensors and stuff are way more useful and interesting, unless you're designing DDSes. ;)

Cheers

Phil Hobbs

Simulation spares a boy a lot of hard thinking.

It's much less useful in mixed-technology systems though, which is what I'm usually doing.

Cheers

Phil Hobbs

(heroically omitting his usual rant about people who don't do their own math)

We've just been simulating a really ghastly DDS-tangled TLL (time-locked loop) to see if various goofy ideas might work. So far, most don't. We are past the precision of single floats here, borderline with doubles. DDS is hazardous because it's hard to know if a sim might fail if the frequency were set to some obscure value. Brute-force testing that might run for years.

SRS makes a clock generator with really good phase noise and jitter. They use a narrow range DDS, and a good crystal filter, to generate the reference for a fractional-N RF synthesizer, to get the best of both worlds.

Hey, I'm an engineer. I do what works with least effort. And I've employed lots of smart people who covered pages with math that produced absurd conclusions.

There's a saying that computer code has an error every 10 or 100 lines or something. Ditto for spreadsheets. Math proofs probably have a similar error rate.

Grin, especially when ~95% of the math I do is just algebra. (Well including trig and complex numbers.)

I get the feeling that there is a real disconnect (for most) between math and the real world.

"Here's a hunk of copper tubing." "Now go calculate it's resistance, and then measure it."

George H.

Well you have to be clever too.. back of the envelope type calculations to know if you are in the ballpark. Of course I've filled my share of pages with mindless calc's also.

George H.

You've gotta measure either way; length and mass, if it's copper, supports a calculation best. Trying to get inside diameter of soft bent copper tubing is a fool's errand.

I actually did some calculus, took a symbolic integral to calculate the power dissipation as a fet switched. That was about 12 years ago. It came out close to a triangle-approximation guess. LT Spice would do that now.

You're my peeps, George. ;)

One short formula for telling science and engineering apart: "In science, the experiment tests the theory, and in engineering, the theory tests the experiment."

Cheers

Phil Hobbs

I would measure it first. How else would you know the dimensions for the calculations?

:)

The AD9956 has a mode that can do that too, except you have to supply your own filter. It's actually a pretty remarkable part, it looks like. I have the eval board, but haven't played with it yet.

Cheers

Phil Hobbs

I hardly ever do integrals.. just guesstimates of areas, volumes. Derivatives are more common, but that's basically algebra in my book anyway. I bump into Bessel functions.. Cavity modes, FM modulation index.

George H.

Yeah, you are right... I was going to say a piece of copper wire but that's even easier calculation-wise.

We do this experiment with a lockin to measure the low resistance of a copper bar. You have to be real careful of the inductance in the cables... what frequency you pick...fun, (assuming you like that wort of thing.)

George H.

Peeps? Those marshmallow things you get at Easter?

Hmm, I don't know what that makes me. I do the measurement to check my math. :^)

George H.

The indispensable John Miles, KE5FX, has some good plots of various Rb standards:

Compared with his plots for the 10811 OCXO at

it looks like the LPRO-101 is very comparable above about 100 Hz, but noisier from there to ~0.05 Hz (20 second allan variance) where the OCXO starts to fall out of bed. So their VCXO isn't as good as Rick Karlquist's. (No big surprise there.)

Cheers

Phil Hobbs

Millennial-speak for "people." (Hence the smiley.)

Pay no attention to the man behind the curtain. (I do that sometimes too, as well as goofing off on Usenet--right now I'm trying to avoid aligning a gruesomely complicated heterodyne interferometer.)

Cheers

Phil Hobbs

Engineers were designing aqueducts and cathedrals and bridges and ships before there was much math or science, just acquired practice, extend the designs of the stuff that doesn't fall down.