Maker Pro
Maker Pro

low noise amplifier for high impedance source

Hello everyone,
I am a student who has been working on a low noise preamplifier for a
high impedance current source. I have put the model of the detection
circuit here:
http://img146.imageshack.us/img146/9748/detectmodelid8.jpg
Its a differential ac current source with an instrinsic capacitance ~
20pF on each side. I use 1M ohm resistors to bias my input JFETs.
The bandwidth i need is only from 10 Khz to couple of MHz.
In my learning process, i have realized few things such as selecting
JFETs as my input transistors for low leekage hence low input current
noise. Maximise the value of input resistor without disturbing the bias
of the input JFETs. However i still have couple of unanswered
questions:
1. How does the Cin of the JFETs effect the signal to noise. do i have
to match the input capacitance of the JFETs to the source capacitance
for minimum noise? How about if i put multiple JFETs in parallel to
reduce input voltage noise - do i have to revise my JFETs selection in
terms of input capacitance?
2. While evaluating the performance of the designed amplifier can i
neglect the equivalent input current noise, as i am using the JFETs?
3. Are there any good review articles or texts for low noise
preamplifier design (for capacitive sensors).
Please help!
thanks
-arch
 
P

Phil Allison

Jan 1, 1970
0
** Groper alert.
I am a student who has been working on a low noise preamplifier for a
high impedance current source. I have put the model of the detection
circuit here:
http://img146.imageshack.us/img146/9748/detectmodelid8.jpg
Its a differential ac current source with an instrinsic capacitance ~
20pF on each side. I use 1M ohm resistors to bias my input JFETs.
The bandwidth i need is only from 10 Khz to couple of MHz.


** Do you really have a current source ???

Or is it more like a voltage source with small series capacitance ?

Makes a big difference.

Assuming the latter is the case, the dominant noise source in your circuit
is the bias resistors. A 1 megohm resistor has 130 nV / rt Hz of noise while
a good J-FET will have only 100th of that.

The trick is to make the bias resistors as high in value as possible and so
allow the capacitance of your source attenuate their noise contribution. For
each doubling the resistor value, the attenuation factor increases by 6dB
while noise increases by only 3 dB.

To maximise available signal voltage, make the input capacitance of the
pre-amp as low as possible, ie install the pre-amp right next to the sensor.




......... Phil
 
Thanks for your reply Phil.
** Do you really have a current source ???
Yes, to my understanding. The source consists of 2 electrodes (Penning
trap). Charged particles rotate between these electrodes and induce an
image charge. This image charge is allowed to flow in an external
circuit (like resistors) to constituting differential image currrent
out of the electrodes.
 
P

Phil Allison

Jan 1, 1970
0
** Groper alert.
** Do you really have a current source ???
Yes, to my understanding. The source consists of 2 electrodes (Penning
trap). Charged particles rotate between these electrodes and induce an
image charge. This image charge is allowed to flow in an external
circuit (like resistors) to constituting differential image current
out of the electrodes.


** No I remember - YOU posted the SAME query back on July 25.

http://groups.google.com.au/group/s...tronics.design&rnum=28&hl=en#841ce9407a2de92e


Reckon the laws of nature have changed since then ???






........ Phil
 
No I remember - YOU posted the SAME query back on July 25.
Yes i did have another post but the problem was DIFFERENT at that time.
I don't have to think about cryogenic operation at the moment.
The thing i want to understand is THIS:
HOW the input transistor capacitance and my bias resistor effect the
signal to noise ratio?
 
No I remember - YOU posted the SAME query back on July 25.
Yes i did have another post but the problem was DIFFERENT at that time.
I don't have to think about detecting a single charge and there is no
cryogenic operation.
The thing i am trying to understand is:
How the input transistor capacitance and my bias resistor effect
thesignal to noise ratio?
 
No I remember - YOU posted the SAME query back on July 25.
Yes i did have another post but the problem was DIFFERENT at that time.
I don't have to think about detecting a single charge and there is no
cryogenic operation.
The thing i am trying to understand is:
How the input transistor capacitance and my bias resistor effect
the signal to noise ratio? And i would appreciate helpful description.
 
P

Phil Allison

Jan 1, 1970
0
** DO NOT EDIT my posts - you fucking, know nothing Google Groper
prick.

** Now I remember - YOU posted the SAME query back on July 25.
http://groups.google.com.au/group/s...tronics.design&rnum=28&hl=en#841ce9407a2de92e


Reckon the laws of nature have changed since then ???

Yes i did have another post but the problem was DIFFERENT at that time.


** Like BLOODY HELL it was !!!

I don't have to think about cryogenic operation at the moment.


** What totally irrelevant CRAPOLOGY !!


The thing i want to understand is THIS:
HOW the input transistor capacitance and my bias resistor effect the
signal to noise ratio?


** YOU do not understand ANYTHING.

Bugger off.



....... Phil
 
M

Marte Schwarz

Jan 1, 1970
0
trap). Charged particles rotate between these electrodes and induce an
image charge.

But this isn't really a current source. Charce delivered into a known
capacity means a voltage, isn't it? You also can make a Current measurement
but not by enabling the current flowing through resisitve path. Then you
should use transimpedance amplifiers. TI have niche appnotes discussing
noise in transimpedande amps.

Marte
 
Hi Marte,
Thanks for replying. Here i have the fundamental paper talking about
the electrical model of my detector by Comisarow, which is a penning
trap. I
http://scitation.aip.org/getabs/ser...00069000009004097000001&idtype=cvips&gifs=Yes
ts based on rotating monopole and in Fig. 3 and 4 the schematic is
presented. I base almost all of my design thoughts on this paper.
Also, thanks for the TI reference. I found one of the AN disucssing
noise comparison of TIA and switched integrator.
Please let me know of your thoughts regarding the source modelling.
 
W

Winfield Hill

Jan 1, 1970
0
archiees wrote
I am a student who has been working on a low noise preamplifier
for a high impedance current source. I have put the model of the
detection circuit here:
http://img146.imageshack.us/img146/9748/detectmodelid8.jpg
Its a differential ac current source with an instrinsic capacitance ~
20pF on each side. I use 1M ohm resistors to bias my input JFETs.
The bandwidth I need is only from 10 Khz to couple of MHz.

archiees wrote...
Thanks for your reply Phil.
** Do you really have a current source ???
Yes, to my understanding. The source consists of 2 electrodes
(Penning trap). Charged particles rotate between these
electrodes and induce an image charge. This image charge is
allowed to flow in an external circuit (like resistors) to
constituting differential image currrent out of the electrodes.

The schematic you give in your .jpg drawing is

.. |--
.. ,---------+-------------->|-- JFET
.. | +--/\/\-- gnd
.. | ICT +---||--- gnd
.. | cs |--
.. '-(->)----+-------------->|--
.. +--/\/\-- gnd
.. +---||--- gnd

where the resistors are 1M, and the capacitors 20pF. The
signal voltage developed across a low-value resistor like
1M will likely be much less than you like, as Phil implied.
Resistor noise density is i_n = (4kT/R)^1/2, so you'll want
a high R, like 100M or higher. The load capacitance will
start reducing the signal (including the resistor noise)
voltage above a frequency fc = 1 / 2pi R C, which is only
8kHz for 1M, and 80Hz for 100M. So clearly you want to
lower value of the load C if you can.

One point, a portion of the JFET's capacitance, Crss, will
be added to C, unless you employ tricks to prevent this.

Once you know the signal voltage vs. current, as a function
of frequency, you can compare it to e_n, the JFET's voltage
noise at that frequency. You will find low e_n JFETs have
high capacitance. Just a little gotcha from Mother Nature.

For example, a classic 2n4392 has en = 3nV and Crss = 3pF.
Compare this to a bf862 with en = 0.9nV and Crss = 1.9pF.
Ooops!! Actually, that's a bad example, contrary to Mother
Nature, as well as Murphy's Law, because the Philips bf862
JFET is far, far better than it should be! :)
 
P

Phil Allison

Jan 1, 1970
0
** BEWARE: Groper Idiot on Loose

Thanks for replying. Here i have the fundamental paper talking about
the electrical model of my detector by Comisarow, which is a penning
trap. I
http://scitation.aip.org/getabs/ser...00069000009004097000001&idtype=cvips&gifs=Yes

ts based on rotating monopole and in Fig. 3 and 4 the schematic is
presented. I base almost all of my design thoughts on this paper.



** My Gawd !!!

Now *WE* have to FUCKING pay money ( US$23) to read this arrogant
wanker's reference !!





......... Phil
 
Thanks Win,
I appreciate the explaination. It makes complete sense to me now.
Earlier, i was misled about matching the capacitance stuff.
I have been working a design using JFETs from Interfet having en =
5nV/rtHz @10kHz, input capacitance 25pF and CRss of 5pF. Clearly i was
not doing the right thing and i can certainly improve by selecting the
proper device.
So besides the en in my frequency range and min. capacitance - sh'ldn't
i look at the leekage current numbers as well. As i understand it would
determine how large input resistor i can put at the input (instead of
1M ohm) without disturbing the gate bias.
I cannot do anything about the 20pF capacitance at the input, its
intrinsic to the detector and i am already keeping my preamplifier as
close as possible.
I am sorry about the reference to the paper earlier, i thought its free
online access - as i'm in school never encountered subscription
problems. I have uploaded it at the following url, incase people are
interested.
http://www.yourfilehost.com/media.php?cat=other&file=Singnal_modelling_Comisarow.pdf
Thanks again for helping.
-Arch
 
W

Winfield Hill

Jan 1, 1970
0
Phil Allison wrote...
archiees wrote


** My Gawd !!!
Now *WE* have to FUCKING pay money ( US$23) to read this
arrogant wanker's reference !!

You don't have free access to a decent library that has it,
or can get it for you? It's an old, basic reference in JCP,
which is an longtime basic journal available in millions of
libraries, and is readily available online through tens of
thousands, if not hundreds of thousands, of libraries, etc.
Specialized journals, like the Journal of Mass Spectroscopy,
or Ultramicroscopy, etc., may be another matter, even though
for many these are now among today's basic journals.
 
C

Carl Ijames

Jan 1, 1970
0
where the resistors are 1M, and the capacitors 20pF. The
signal voltage developed across a low-value resistor like
1M will likely be much less than you like, as Phil implied.
Resistor noise density is i_n = (4kT/R)^1/2, so you'll want
a high R, like 100M or higher. The load capacitance will
start reducing the signal (including the resistor noise)
voltage above a frequency fc = 1 / 2pi R C, which is only
8kHz for 1M, and 80Hz for 100M. So clearly you want to
lower value of the load C if you can.

As Comisarow discusses in that paper, above the fc you calculate the RC
looks like a low pass filter and so attenuates the signal voltage as the
cyclotron frequency rises. However, the current is proportional to
cyclotron frequency and so these offset, giving constant signal voltage
as the frequency (and thus the mass/charge ratio) varies. Not saying
rising signal with decreasing m/z wouldn't be nice, it's just how it
falls out. It is convenient that the proportionality between number of
ions of a given m/z and the signal voltage is constant, and not a
function of m/z (assuming constant cyclotron radius).

What kind of Penning trap system are you working on, Arch?
 
R

Robert Latest

Jan 1, 1970
0
On 16 Oct 2006 15:53:38 -0700,
in Msg. said:
For example, a classic 2n4392 has en = 3nV and Crss = 3pF.
Compare this to a bf862 with en = 0.9nV and Crss = 1.9pF.
Ooops!! Actually, that's a bad example, contrary to Mother
Nature, as well as Murphy's Law, because the Philips bf862
JFET is far, far better than it should be! :)

I don't have the 4329 datasheet, but note that the BF862's noise is
spec'd at 100kHz. But the BF862 does seem to have a rather large gm for
its low capacitance. When you say "contrary to Mother nature" it sounds
as if you meant to imply "contrary to the laws of physics" which would
imply "faked datasheet". Did you mean to say that? Personally, I
wouldn't put anything past Philips and especially NXP.

robert
 
W

Winfield Hill

Jan 1, 1970
0
Robert Latest wrote...
On 16 Oct 2006, Winfield Hill wrote


I don't have the 4329 datasheet, but note that the BF862's
noise is spec'd at 100kHz.

Its noise isn't appreciable higher at 1 to 10kHz. I'll look
for my data and post it.
But the BF862 does seem to have a rather large gm for its
low capacitance. When you say "contrary to Mother nature"
it sounds as if you meant to imply "contrary to the laws of
physics" which would imply "faked datasheet". Did you mean
to say that? Personally, I wouldn't put anything past Philips
and especially NXP.

No, I was just being funny. I suppose it's really the older
part that's at fault in the comparison. The bf862 does what
Philips claims, and better. IIRC, the bf862 has a higher
gm at a given current than the '4392. JFET noise goes as
1/gm, so that must give the bf862 a leg up. NXP or not, it's
a nice JFET, easy to get, and cheap. What's not to like?

Philips states an e_n = 0.8nV spec at Idss, which could be
as high as 25mA, and even tho this is the industry-standard
way to publish noise specs, it's not the best to my way of
thinking. But the bf862 JFETs I measured are no more than
1.1nV or so at 3mA, which makes them very useful parts.
 
R

Robert Latest

Jan 1, 1970
0
On 17 Oct 2006 04:55:28 -0700,
in Msg. said:
JFET noise goes as 1/gm,

1/gm or 1/sqrt(gm)? From my understanding of FETs, gm goes up with area
-- if you parallel N FETs you get N times the transconductance, but you
also get N times Ciss and Crss and 1/sqrt(N) the noise.

I'm probably at least half wrong here, so please correct me.
so that must give the bf862 a leg up. NXP or not, it's
a nice JFET, easy to get, and cheap. What's not to like?

Nothing. I like it fine.

robert
 
Top