Low drift, low noise, fast opamp

[Joerg]

[...]

But AD is 'sposed to be the Mercedes-Benz of semiconductor companies...

Same here. I 'sposed t'same until I sampled their AD9833 DDS.
"Sine, Triangle, Square" they say.
Great stuff, methinks I'll drive that prototype mixer straight from that
square output. No way mate!. Phase jitter so bad that a 'scope has
difficulty displaying the waveform. Zilch mention of this in the data sheet.

[...]

That is indeed occurring more frequently. In the first ten years of my
career I had only one instance where a chip wouldn't perform and the
vendor was a quite arrogant about that (was supposed to be operated
between 4.75V and 5.25V but wouldn't perform unless fed 6V). In the last
10 years I had half a dozen cases. The last one (a TPS regulator)
remains unexplained. Instead of throwing it on PSPICE they kept asking
for more and more data from us. So I dumped it.

Yes. I read the thread. Admit I no longer have your patience in chasing down
questionable components. Any component that dares to display 'oddities' is
immediately out on it's ear. More and more though, I'm falling back to using
those transistor things. They can be a bugger to design with and generate
big drawings but aren't 'arf cheap and reliable.
john

That might be because the guys from fresh out of school are picking up
the hot chip designer jobs but all they learned was VHDL and SPICE, no
solder irons or real scopes involved. They don't make the Robert Widlars
and Bob Peases anymore these days. I am seeing the same trend in circuit
level design which is my kind of turf. Won't complain because when the
nicely simulated "solution" goes kablouei they need consulting help ))

 

[Reinhard Zwirner]

In a 100 MHz system, there is no "higher frequency portion"!

Everything < 1 GHz = DC? Grmmpf!

Embarrassed ;-)

Reinhard

 

[colin]

Hello Colin,

What we are measuring isn't the absolute signal which will be huge in
this case. We are after subtle differences in that signal, and a whole
lot more than 60dB. Basically we need to squeeze out all we can get. Of
course, I don't yet know what amount of noise will originate at the DFB
module. It wouldn't make sense to build a super-quiet front end and then
the laser swamps it with its own noise.

how much more do you need, another 10-20db ?
or a lot more still ? that might be difficult ..

Im not sure what the shot noise from the detector would be,
but although your signal is dc-100mhz what is the bandwitdh of the change?

I expect you could do some noise reduction on the difference signal to get
the extra margin as 100mhz bandwidth means considerable noise, thats if you
could demodulate it somehow otherwise youl just have to process the heck out
of it with a DSP I gues.

Colin =^.^=

 

[Lostgallifreyan]

how much more do you need, another 10-20db ?
or a lot more still ? that might be difficult ..

Im not sure what the shot noise from the detector would be,
but although your signal is dc-100mhz what is the bandwitdh of the
change?

I expect you could do some noise reduction on the difference signal to
get the extra margin as 100mhz bandwidth means considerable noise,
thats if you could demodulate it somehow otherwise youl just have to
process the heck out of it with a DSP I gues.

Colin =^.^=

Or you could just cool the photodiode with a TEC. Joerg said that the laser
might make more noise than the detector though, so that might not help.

 

[John Larkin]

Or you could just cool the photodiode with a TEC. Joerg said that the laser
might make more noise than the detector though, so that might not help.

Photodiodes aren't very noisy; usually the amp's noise wins big-time.
In Joerg's case, laser noise might be the real winner.

The light coming into the pd is photonized, so has shot noise, but
improving the detector can't help that.

John

 

[Joerg]

Hello Colin,

how much more do you need, another 10-20db ?
or a lot more still ? that might be difficult ..

In the amp I see no problem. Never done it with a TIA chip but I
designed many discrete current amps and you can hear the grass grow with
those. But if the laser source is noisy, well, then it is all water
under the bridge unless I can sufficiently "de-noise" that source.

Im not sure what the shot noise from the detector would be,
but although your signal is dc-100mhz what is the bandwitdh of the change?

It's almost the whole range :-(

The dark current noise typically wins AFAICT. The last calcs had the
shot noise almost a order of magnitude above everything else. That's for
the photo diode. Thing is, the RIN for the laser diode doesn't mean much
for us because this app is also sensitive to phase noise. All they give
you in the datasheet is spectral line width and who knows what's
underneath that number. Mode jumps etc.

I expect you could do some noise reduction on the difference signal to get
the extra margin as 100mhz bandwidth means considerable noise, thats if you
could demodulate it somehow otherwise youl just have to process the heck out
of it with a DSP I gues.

There will be plenty of math horsepower in that machine.

 

[Joerg]

Hello John,

Photodiodes aren't very noisy; usually the amp's noise wins big-time.
In Joerg's case, laser noise might be the real winner.

Looks like it. When they state 10MHz line width that doesn't sound
encouraging. Who knows what's hiding behind that number. Guess we'll
find out soon. If only they had some graphs in them datasheets.

The light coming into the pd is photonized, so has shot noise, but
improving the detector can't help that.

I always wondered why nobody seems to offer the common base stages I do
in discretes as a chip but that dominant shot noise may just be the
reason. Maybe nobody really needs the stellar noise figures of a lone RF
transistor. Although, considering that most fast TIA chips are above $2
I still think there'd be a market (my amps were always much less).

BTW, an engineer from Europe mentioned a small chip company that I
almost forgot about:

http://www.ichaus.de/productgroup.php?grp=Laser_Drivers

They have a US office. But AFAICT no offerings for the receive side.

 

[colin]

Hello Colin,

It's almost the whole range :-(

The dark current noise typically wins AFAICT. The last calcs had the
shot noise almost a order of magnitude above everything else. That's for
the photo diode. Thing is, the RIN for the laser diode doesn't mean much
for us because this app is also sensitive to phase noise. All they give
you in the datasheet is spectral line width and who knows what's
underneath that number. Mode jumps etc.

can you compensate by monitoring the laser diode output noise ?
or is the laser diode miles away ?
or maybe cancel it out with feedback ?

Colin =^.^=

 

[Joerg]

can you compensate by monitoring the laser diode output noise ?
or is the laser diode miles away ?
or maybe cancel it out with feedback ?

Yes, we could, but only in the final design. I am already seeing it
coming that I have to regulate out at least some of the noise. Not
looking forward to that because it has to be done up to 100MHz plus that
still doesn't take care of the phase noise part. The latter would
require some nifty optical detector (getting stomach cramps here...).

 

[colin]

Yes, we could, but only in the final design. I am already seeing it
coming that I have to regulate out at least some of the noise. Not
looking forward to that because it has to be done up to 100MHz plus that
still doesn't take care of the phase noise part. The latter would
require some nifty optical detector (getting stomach cramps here...).

Good luck ! and dont forget to breath.

Would you digitise the full 100Mhz ?
Wonder if the sampling units in such scopes are easily available.

Is there any structure to the signal and difference ?
Ie is the full bandwidth used all the time so it just looks like white noise
?
or are there periodic repetitions etc ?
Is the difference in light signal due to the changes in
reflection/refraction/absorbtion ?

Ive been looking for an elusive ~1ps variation in delay for some time now on
and off with little success so far.

Colin =^.^=

 

[Joerg]

Good luck ! and dont forget to breath.

Would you digitise the full 100Mhz ?
Wonder if the sampling units in such scopes are easily available.

Yes, but it won't be a scope. It's going to be our own design. Scope
don't show a lot of ENOBs up there, maybe 6-7. We need a lot more.

Is there any structure to the signal and difference ?
Ie is the full bandwidth used all the time so it just looks like white noise
?
or are there periodic repetitions etc ?

All the time.

Is the difference in light signal due to the changes in
reflection/refraction/absorbtion ?

Basically all reflection.

Ive been looking for an elusive ~1ps variation in delay for some time now on
and off with little success so far.

Well, we'll be looking at femtometers ;-)