Re: Photodiode amplifier noise


"John Larkin" <jjlarkin@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:lthuv4pidieva43eb3i2sde8n1bqutp42s@xxxxxxxxxx
On Mon, 04 May 2009 12:56:19 -0400, Phil Hobbs
<pcdhSpamMeSenseless@xxxxxxxxxxxxxxxxxx> wrote:

Joerg wrote:
John Larkin wrote:
On Sun, 03 May 2009 15:22:14 -0400, Spehro Pefhany
<speffSNIP@xxxxxxxxxxxxxxxxxxxxxxx> wrote:

On Sun, 03 May 2009 11:21:42 -0700, the renowned John Larkin
<jjlarkin@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote:

On Sun, 03 May 2009 09:53:37 -0700, Joerg
<notthisjoergsch@xxxxxxxxxxxxxxxxxxxxx> wrote:

John Larkin wrote:
On Sat, 02 May 2009 21:29:47 GMT, James Arthur
<bogusabdsqy@xxxxxxxxxxx> wrote:

John Larkin wrote:
On Sat, 02 May 2009 13:14:40 -0700, Joerg
<notthisjoergsch@xxxxxxxxxxxxxxxxxxxxx> wrote:

John Larkin wrote:

[...]

This weekend I'm going to buy a tin of Danish sugar cookies,
eat all
the cookies, and build a pA range diode-curve measuring setup
into it.
I have a bunch of LMC6001s (Ib is 10 fA typ) to use. The
data***
hints somewhere in one of the app circuits that Cin is big.
But it's
not specified anywhere.

It's "Danish Butter cookies" :-)

Make sure to _throughly_ clean it before you build anything
into it, until wiping shows no grease residue. Else you might
experience an impressive meeting of ants in there. BTDT.


The rule seems to be that if Cin is large, don't mention it on
the
data***.

Cin is probably largish and usually not used directly:

http://www.postech.ac.kr/ee/paust/paper_link/2007%20Development%20of%20a%20Beam-Profile%20Monitor%20for%20Cyclotron%20MC50.pdf

Interesting; that's vaguely like what I'm tying to do. But I
can't
find a data*** on the JU421, even on the Vishay site, so I
assume
it's another FormerFet.
www.vishay.com/docs/70248/70248.pdf

Cheers,
James Arthur
Cool. Vishay didn't recognize "JU421".

Nice part for leakage, but with En=30 nv/rthz and Gm=200 uS, it's
sort
of the Marching Band of Noise. This guy adds a second one in
differential mode to add bonus noise and lower the gain to boot.

Well, good luck finding one :-)

If you do it'll probably have "DoD pricing".


Why do these scientific types so love differential jfet front-ends?

In electronics they tend to hang on to stuff much longer than we do.

[...]
I hadn't paid much attention to discrete jfets in, well, some
decades.
They have low transconductance, high capacitances, miserable
matching,
and astounding data*** parameter spreads, like 10:1 Idss limits.

But for really low noise (like, under 1 nV/rthz) amplification of
high-impedance signals, nothing can touch them. The best fet opamps
are, like, 5x as noisy as the best jfet.
True.
Pity that most people, at
least the scientists, trend to use them wrong.

John
What wrongness have you seen? The commercial stuff for research
applications looks pretty decent to me.


I see mostly botched circuit protection. Last one: A freaking expensive
laser diode driver with TEC in there was running. Janitor plugged vacuum
into same circuit, turned it on ... tzzzt ... phut ... *POP*

That isn't supposed to happen.


I've seen a lot of diffamps where it wasn't necessary, really rotten
second stages (the fets have low gain, so 2nd stage noise matters) and
circuits where the Johnson noise of gate bias resistors (and
occasionally drain load resistors!) overwhelmed the fet's noise. One
paper got around that by testing the amp with a grounded input!

Could it be that there just aren't a lot of good disctete circuit
designers around? I would hate to think that.


That is most definitely the case :-(

OTOH, this puts bread on the table around here ...


Big JFETs are great, but only for slow measurements...once you get past
BF862 territory you get to choose between horrible 1/f noise (in the RF
parts) or horrible input capacitance (in the AF parts). The AC input
resistance drops quadratically with frequency once you get above the
corner frequency.

Or are there sterling counterexamples that I'm missing?

BF862 is a pretty good part for a class of signals, like particle beam
detectors in accelerators, or small piezos, where the source is
capacitive. I have one paper that calculates that the best s/n happens
when the capacitance of the fet is equal to the capacitance of the
source, sort of a capacitive maximum-power-transfer thing.

I've been meaning to look into varactors as low-noise LF amps; they
were popular for a while before fets got good. I have no idea what the
noise mechanisms are.

Distributed amps have interesting noise behavior, when you can use
them.

John



the capacitance matching is almost a fluke, a quick rule of thumb, and it
happens to be because IIRC :
the FET gain (gm) is function among other things of the device geometry ,
the FET input capacitance is also function of the device size, while the
output noise is function of the detector and fet input capacitances.

Then , the S/N just "happens" to be maximised close to the capacitance
matching point.
Actually , the optimal detector to FET input capacitance ratio may be 1:0.6
, or something not equal , but close to 1:1.

Jure Z.


.

Quantcast