Re: Very simple 1kHz sinewave circuit is more than 133dB down on harmonics

On Wed, 05 Feb 2025 18:21:07 -0800, john larkin <JL@gct.com> wrote:

On Wed, 05 Feb 2025 19:17:21 -0500, Joe Gwinn <joegwinn@comcast.net>
wrote:
>

On Wed, 05 Feb 2025 14:06:47 -0800, john larkin <jl@glen--canyon.com>
wrote:
>

On Wed, 5 Feb 2025 19:08:25 +0000, John R Walliker
<jrwalliker@gmail.com> wrote:
>

On 05/02/2025 17:51, Edward Rawde wrote:

<albert@spenarnc.xs4all.nl> wrote in message news:nnd$01c0b887$041e6dfb@960b25520c3d6ac2...

In article <vnk0ul$2b1o$1@nnrp.usenet.blueworldhosting.com>,
Edward Rawde <invalid@invalid.invalid> wrote:

Elaborate DC stabilization isn't needed to get below 130dB.

>
Is this a measurement or a simulation?

 
Simulation only at present.
There has been debate about what meaningful measurments may or may not be possible below about -90dB.
In a simulation I can get down to -135dB but I make no claim to be able to achieve or measure that in practice.

>
I needed a very low distortion audio oscillator in 1997 in order to
characterise a 16-bit codec for an application report.
>
The simplest approach was to use a 20-bit sigma-delta DAC with
a combination of an active low-pass filter followed by a passive
single-pole output filter.
The output frequency was 1.1kHz.  The third harmonic, measured by
the device under test was about -82dB below the fundamental.  All
other harmonics were considerably lower.
The signal source may well have been better than that, but I had
no easy way of measuring it at the time.
>
The results are in figure 4-2 (page 37) of:
https://www.ti.com/lit/ug/slau039/slau039.pdf
>
John
>

>
TLC2272 is a 4.5 MHz opamp. So at 3 KHz, it only has a gain of 1500.
That might dominate distortion.
>
AD5791 is a 20-bit sub-1-lsb accurate fast voltage out DAC, which
should make a -120 db distortion synthesizer.  If there was any way to
measure sub-PPM distortion, one could sum it with, say, a fine-trim 16
bit DAC and tune the pair for really low distortion.
>
One might subtract two such synthesizers to measure their distortion.
>
Or three, sort of like the idea of rubbing three pieces of glass
against one another to make them flat.

>
In the Time World, this is called the three-cornered hat algorithm:
>
.<https://rubiola.org/pdf-articles/conference/2016-IFCS-Three-cornered.pdf>
>
This is actually used for more than three references, but the name is
still of three.
>
Joe Gwinn

>
When we want to measure the jitter of a delay generator vs time delay,
we use two units with slightly different delays and measure the jitter
between them.

Yep.  Was very common, but not very good at close-in PN when using a
coax cable - too lossy.  Some people use a long fiber-optic link for
the delay, which is noisier at high offset frequencies.

The RF boys sometime measure oscillator phase noise the same way: mix
two identical oscillators and look at the resulting spectrum.
>
Make sure the DUTs are uncorrelated.

Yep.

And they are bedeviled by data sheets that specify PN only above some
high offset frequency.  Sometimes, its too terrible to confess.  And
sometimes the vendor's marketing dept just doesn't know any better.

Joe Gwinn