DDS wisdom

But wouldn't any low-pass filter also do the same thing (ie: convert vertical parts of the waveform to less vertical parts)? Perhaps a low-pass filter could be chosen which gives superior performance over a straight integrator.

arch

Fun.)

a bit

he

wo

ility

ration

ch is

et by

I can

do

erate

al.

of

folks

ularly

and

the

bits

it in

ifics.

st

ou

e out

arcs.

te) > > has gone away.

ss

Interesting theoretical question. Part of engineering is explaining what yo u are doing, and why you are doing it, in terms that even a manager can und erstand.

My contention would be that the integrator gets rid of the high-slew rate c omponent of the error signal in the first stage of filtering, which makes a difference to any active filter in a way that filter construction software doesn't usually capture. LTSpice does, but making a stair-case approximati on to a sine wave as a test signal might be tedious.

Thanks for saying more eloquently what i wish i would have said.

?-)

Almost. It helps a lot, but there are various non-linearities and jitter effects. Much of this is chip design specific, and vendor app notes are the best source on the oddities of their designs.

It's a complex area, and there are many approaches to reducing the spurs. Some are built into the chips, and some can be applied in the surrounding circuitry and/or controlling firmware and/or software. Which methods are applicable will as always depend of application details.

As a rule, chips are designed to a specific spur level, and all the various sources of spurs are balanced by design to yield roughly the same peak spur levels. In other words, no specific spur cause dominates.

Yes, and it causes much trouble in radar applications.

But is it a problem for your application? And in what way?

Given this, approaches can be suggested.

Hmm. I must be misremembering where I saw that. I think they mention the wheel, but they may not go into what happens at rollover. I discovered the effect when analyzing why a prototype DMTD (Dual Mixer Time Difference) instrument was suffering periodic phase bumps. When one is measuring the performance of Rubidium clocks (10^-11 fractional frequency error), it doesn't take much.

Anyway, the relevant data trove is at work, so it'l be a few days before I can reconstruct where I saw the business about bumps.

Joe Gwinn

Got it, and by 2-D you mean fixing the phase and measuring the amplitude and phase in both I/Q channels.

I'll check out your paper on Monday. We have the whole series of RSI that a library was pitching. Makes great fodder for the "reading room."

George H.

Youch, that smarts. ;) (But not too inaccurate for the most part.)

Cheers

Phil Hobbs

And if my low-pass filter was a biquad, how would that compare?

At 10s of MHz, passive LC filters work better. Inductors are not slew rate limited.

An integrator makes no sense here. Attenuation is a pitiful 6 dB/octave at the cost of roughly 90 degrees of phase shift.

average_phase =sum_over_f( W(f) * atan2(I(f), Q(f)) ) / sum_over_f (W(f) )

Calibration consists of switching the input from 'received signal' to 'reference signal' to get a phase zero point.

Back in the early days of DSP, when it was more of a branch of numerical analysis, there were lots of arguments about slopes and derivative discontinuities. (Hamming and Lanczos are good reads on that.)

When oscilloscopes became popular, people took a look at the actual waveforms they were arguing over, said "Never mind", and went back to work. ;)

Cheers

Phil Hobbs

If you ever need a serious LC filter design, we have the NuHertz dongled software. It makes all sorts of amazing exotic filters, using standard part values.

Cool. Got an example you could post?

Cheers

Phil Hobbs

Over here it's correctly pronounced "ech", with an e, but its still irrelevant.

That's why you dudes make the mistake, but its still wrong.

Sure, American English has bona-fide differences from UK English, but this is not one of them.

Its actually how the following word sounds. Now..,

"It's an Hewlett Packard 9816 computer"

Does that actually sound ok to you?

The Hewlett sound starts as "Hugh". That is not a vowel sound

Kevin Aylward

Yes.

Kevin Aylward

Here's one:

It looks really dumb, but it isn't. That's the beauty of the NuHertz thing.

Used here:

"HP" isn't prononuced "Hewlett Packard", it's pronounced "aitch pea". Hence the 'an'.

Cheers

Phil Hobbs

As you say, you are not an RF guy and you don't do sine waves. These issues are many orders of magnitude below your level of interest or ability to measure.

But obviously, they are real, measurable, and severely limit the use of DDS in precision applications.

But that's not your field.

Don't give up so soon. I think you may be referring to Section 4, "The Effect of Truncating the Phase Accumulator on Spurious Performance," starting on Page 19 of

Fred Harris may have ways to help reduce it, discussed in "Ultra Low Phase Noise DSP Oscillator", at

What did you use to solve the problem?

So roughly a 50 MHz LPF with a real zero at 100 MHz-ish?

Cheers

Phil Hobbs