Fun out-of-scope-bandwidth results...

[Piotr Wyderski]

I took a 100MHz DSO with 100MHz probes in 10:1 mode and connected it to
a 400MHz LVTTL clock. It is barely able to display anything: the signal
is visible, but distorted and heavily attenuated. No wonder.

But then I attached the same to the output of my GaN pulser intended for
nanosecond-scale ferrite testing. Behold:

https://i.postimg.cc/tgL9JsrF/DS1-Z-Quick-Print25.png

This is the voltage at the drain (which means that the current through
the load resistor ramps about that fast from 0 to 3.3A). I have all
reasons to believe that this curve is accurate -- the fall time is 1.5ns
or better. So what makes this scope so incredibly good this time?

BTW, here is a tiny 1 turn:1 turn transformer saturating:

https://i.postimg.cc/X7vtMVLW/DS1-Z-Quick-Print26.png

Best regards, Piotr

 

[Bill Sloman]

On Sunday, October 25, 2020 at 4:41:43 AM UTC+11, Piotr Wyderski wrote:

I took a 100MHz DSO with 100MHz probes in 10:1 mode and connected it to
a 400MHz LVTTL clock. It is barely able to display anything: the signal
is visible, but distorted and heavily attenuated. No wonder.

But then I attached the same to the output of my GaN pulser intended for
nanosecond-scale ferrite testing. Behold:

https://i.postimg.cc/tgL9JsrF/DS1-Z-Quick-Print25.png

This is the voltage at the drain (which means that the current through
the load resistor ramps about that fast from 0 to 3.3A). I have all
reasons to believe that this curve is accurate -- the fall time is 1.5ns
or better. So what makes this scope so incredibly good this time?

Look at the Fourier transforms of the signals you are looking at. A 400Mhz clock hasn\'t got any frequency content below 400MHz.

The Fourier transform of Dirac pulse has equal amplitudes of every harmonic of it repetition rate out to infinity.

A real pulse loses the harmonics which fit within it\'s width. What you are seeing is all the harmonic content below 100MHz and progressively reducing amplitudes of the content above 100MHz, which is to say most of what you\'d see with an even faster scope.

BTW, here is a tiny 1 turn:1 turn transformer saturating:

https://i.postimg.cc/X7vtMVLW/DS1-Z-Quick-Print26.png

Cute.

--
Bill Sloman, Sydney

 

[Bill Martin]

On 10/24/20 10:41 AM, Piotr Wyderski wrote:

I took a 100MHz DSO with 100MHz probes in 10:1 mode and connected it to
a 400MHz LVTTL clock. It is barely able to display anything: the signal
is visible, but distorted and heavily attenuated. No wonder.

But then I attached the same to the output of my GaN pulser intended for
nanosecond-scale ferrite testing. Behold:

https://i.postimg.cc/tgL9JsrF/DS1-Z-Quick-Print25.png

This is the voltage at the drain (which means that the current through
the load resistor ramps about that fast from 0 to 3.3A). I have all
reasons to believe that this curve is accurate -- the fall time is 1.5ns
or better. So what makes this scope so incredibly good this time?

BTW, here is a tiny 1 turn:1 turn transformer saturating:

https://i.postimg.cc/X7vtMVLW/DS1-Z-Quick-Print26.png

    Best regards, Piotr

Heh, driving the probe with a nuke? What\'s the impedance looking back
into your GaN pulser? vs. the 400Mhz thing...

 

[Piotr Wyderski]

Bill Martin wrote:

Heh, driving the probe with a nuke? What\'s the impedance looking back
into your GaN pulser? vs. the 400Mhz thing...

Well, kind of. The 400MHz signal comes out of the FIN1028, which can
source/sink up to 16mA according to the datasheet. The GaN part is a
15mOhm R_DS_ON transistor (the bottom part of the LMG5200, actually,
as the final pulser should be bidirectional). It is capable of a full
0-10V swing in about a nanosecond.

But the impedance should not matter much IMHO, the probe is 10:1.

Best regards, Piotr

 

[Piotr Wyderski]

Bill Sloman wrote:

Look at the Fourier transforms of the signals you are looking at. A 400Mhz clock hasn\'t got any frequency content below 400MHz.

The Fourier transform of Dirac pulse has equal amplitudes of every harmonic of it repetition rate out to infinity.

The pulser is set to produce a pulse train of 1kHz repetition rate with
1% duty cycle. While certainly most of the energy resides below 1GHz, I
would expect all of the harmonics to add up to a ~5ns edge due to the
HF attenuation of the input amplifiers. Instead, I can see a razor-sharp
1ns edge with virtually no distortion. Even the tiny undershot looks
realistic.

I would not be surprised if the ADC were driven directly with my low
impedance source. It would be its peak performance point without
resorting to undersampling. But there is a ton of analogue hardware in
front of it in the scope. Pretty amazing.

Best regards, Piotr

 

[Bill Sloman]

On Sunday, October 25, 2020 at 6:17:14 PM UTC+11, Piotr Wyderski wrote:

Bill Sloman wrote:

Look at the Fourier transforms of the signals you are looking at. A 400Mhz clock hasn\'t got any frequency content below 400MHz.

The Fourier transform of Dirac pulse has equal amplitudes of every harmonic of it repetition rate out to infinity.

The pulser is set to produce a pulse train of 1kHz repetition rate with
1% duty cycle. While certainly most of the energy resides below 1GHz, I
would expect all of the harmonics to add up to a ~5ns edge due to the
HF attenuation of the input amplifiers. Instead, I can see a razor-sharp
1ns edge with virtually no distortion. Even the tiny undershot looks
realistic.

To work out what was going on you\'d best simulate the pulse shape you expect, do a Fourier transform on it, snip out the frequencies above 100MHz and convert the frequencies you have left back into a repeating pulse.

It\'s difficult to imagine that this would give you a 1nsec edge, but the asymmetrical pulse you see might deliver this as an artefact.

A 1kHz repetition rate give you one million harmonics out to 1GHz, and cutting off at 100MHz gets rid of 90% of them, but that 90% only shows up as the sharpest feature of the pulse. Intuition isn\'t a good guide to what might be going on.

I would not be surprised if the ADC were driven directly with my low
impedance source. It would be its peak performance point without
resorting to undersampling. But there is a ton of analogue hardware in
front of it in the scope. Pretty amazing.

The analogue hardware is probably designed to be phase linear up to rather above 100MHz.

--
Bill Sloman, Sydney