Long-tailed JFET pair question

[Robert Latest]

Hello,

I'm designing a low-noise, low-frequency current-to-voltage
converter (TIA). Since the really nice JFET-input, low-noise
OpAmps such as the OPA627 are scarce and expensive I decided to
build the input stage with a discrete dual JFET followed with a
bipolar OpAmp. The idea is to keep the gain of the input stage
just high enough to "hide" the voltage noise of the OpAmp (if
the JFET has, say, 4nV/rtHz and the OpAmp has 16, a gain of 5
should be enough).

Out of curisoity I opened up a commercially available TIA to see
how they did it and if I could learn from their design. I found
a dual JFET running with Ids=3.5mA and Ugs=-1.7 V. It's probably
something like a 2N5912 (it's in a 7-leg TO78 can, part #
scraped off). The odd thing is: the drain resistors are only 220
ohms off a 9V rail. The parts (at least the ones I can find)
that fit this Vgs/Ids combo have a few mS transconductance which
would get the input stage's gain barely over unity. Parts with
significantly higher gm run at tens of mA in this gate voltage
range.

What I don't understand: If unity gain is enough, why not just
use the JFETs as source followers?

Disclaimer: I'm not ripping off anybody's product (and besides a
JFET diff amp ain't rocket science). It's just instructive to
look at other people's ideas, but in this case I can't make much
sense of it.

robert

 

[Fred Bartoli]

Robert Latest a écrit :

Hello,

I'm designing a low-noise, low-frequency current-to-voltage
converter (TIA). Since the really nice JFET-input, low-noise
OpAmps such as the OPA627 are scarce and expensive I decided to
build the input stage with a discrete dual JFET followed with a
bipolar OpAmp. The idea is to keep the gain of the input stage
just high enough to "hide" the voltage noise of the OpAmp (if
the JFET has, say, 4nV/rtHz and the OpAmp has 16, a gain of 5
should be enough).

Out of curisoity I opened up a commercially available TIA to see
how they did it and if I could learn from their design. I found
a dual JFET running with Ids=3.5mA and Ugs=-1.7 V. It's probably
something like a 2N5912 (it's in a 7-leg TO78 can, part #
scraped off). The odd thing is: the drain resistors are only 220
ohms off a 9V rail. The parts (at least the ones I can find)
that fit this Vgs/Ids combo have a few mS transconductance which
would get the input stage's gain barely over unity. Parts with
significantly higher gm run at tens of mA in this gate voltage
range.

What I don't understand: If unity gain is enough, why not just
use the JFETs as source followers?

Disclaimer: I'm not ripping off anybody's product (and besides a
JFET diff amp ain't rocket science). It's just instructive to
look at other people's ideas, but in this case I can't make much
sense of it.

No, you can't have unity gain.
Noise is summing quadratically, so your opamp input refered noise has to
be, as you first expected, the input stage gain has to be at most 1/3 of
the diff pair noise, so that the opamp accounts for at most 5% of total
noise voltage density.

Now, in its flat region a JFET will give you a 2/(3*gm) noise resistance
which is sqrt(8kT/3 gm) noise.

To achieve 4nV/rtHz you'll need 2.8nV/rtHz JFETs which translates to
470R noise resistance and at least 1.4mA/V transconductance.

Now that you have a JFET input stage, you can use a bipolar opamp which
are easier to get in the relatively low noise range and 4nV/rtHz is
easy. For ex. the OPA227/228 is a good and cheap one.

But at 2mA/V you'll still at least 1.5K/2K drain resistors for the JFETs
to noise to be dominant.

For the TIA you opened, 220R resistors means:
- either that the noise performances aren't optimal
- either that the JFETs have higher gm, which isn't optimal either.

 

[Eeyore]

Hello,

I'm designing a low-noise, low-frequency current-to-voltage
converter (TIA). Since the really nice JFET-input, low-noise
OpAmps such as the OPA627 are scarce and expensive I decided to
build the input stage with a discrete dual JFET followed with a
bipolar OpAmp. The idea is to keep the gain of the input stage
just high enough to "hide" the voltage noise of the OpAmp (if
the JFET has, say, 4nV/rtHz and the OpAmp has 16, a gain of 5
should be enough).

Out of curisoity I opened up a commercially available TIA to see
how they did it and if I could learn from their design. I found
a dual JFET running with Ids=3.5mA and Ugs=-1.7 V. It's probably
something like a 2N5912 (it's in a 7-leg TO78 can, part #
scraped off). The odd thing is: the drain resistors are only 220
ohms off a 9V rail. The parts (at least the ones I can find)
that fit this Vgs/Ids combo have a few mS transconductance which
would get the input stage's gain barely over unity. Parts with
significantly higher gm run at tens of mA in this gate voltage
range.

What I don't understand: If unity gain is enough, why not just
use the JFETs as source followers?

Disclaimer: I'm not ripping off anybody's product (and besides a
JFET diff amp ain't rocket science). It's just instructive to
look at other people's ideas, but in this case I can't make much
sense of it.

The front end may run into a current to voltage ( transimpedance ) stage. It's
been done for pro-audio mic pre-amps for example.

Graham

 

[Joerg]

Hello,

I'm designing a low-noise, low-frequency current-to-voltage
converter (TIA). Since the really nice JFET-input, low-noise
OpAmps such as the OPA627 are scarce and expensive I decided to
build the input stage with a discrete dual JFET followed with a
bipolar OpAmp. The idea is to keep the gain of the input stage
just high enough to "hide" the voltage noise of the OpAmp (if
the JFET has, say, 4nV/rtHz and the OpAmp has 16, a gain of 5
should be enough).

Look at the AD8628. Not exactly cheap but low noise.

Out of curisoity I opened up a commercially available TIA to see
how they did it and if I could learn from their design. I found
a dual JFET running with Ids=3.5mA and Ugs=-1.7 V. It's probably
something like a 2N5912 (it's in a 7-leg TO78 can, part #
scraped off). The odd thing is: the drain resistors are only 220
ohms off a 9V rail. The parts (at least the ones I can find)
that fit this Vgs/Ids combo have a few mS transconductance which
would get the input stage's gain barely over unity. Parts with
significantly higher gm run at tens of mA in this gate voltage
range.

What I don't understand: If unity gain is enough, why not just
use the JFETs as source followers?

In case this is for a photodiode: FETs are often only used to decouple
the high photodiode capacitance from the TIA that follows at the drain
node of the FET. Buys bandwidth.

 

[Phil Hobbs]

Hello,

I'm designing a low-noise, low-frequency current-to-voltage
converter (TIA). Since the really nice JFET-input, low-noise
OpAmps such as the OPA627 are scarce and expensive I decided to
build the input stage with a discrete dual JFET followed with a
bipolar OpAmp. The idea is to keep the gain of the input stage
just high enough to "hide" the voltage noise of the OpAmp (if
the JFET has, say, 4nV/rtHz and the OpAmp has 16, a gain of 5
should be enough).

Out of curisoity I opened up a commercially available TIA to see
how they did it and if I could learn from their design. I found
a dual JFET running with Ids=3.5mA and Ugs=-1.7 V. It's probably
something like a 2N5912 (it's in a 7-leg TO78 can, part #
scraped off). The odd thing is: the drain resistors are only 220
ohms off a 9V rail. The parts (at least the ones I can find)
that fit this Vgs/Ids combo have a few mS transconductance which
would get the input stage's gain barely over unity. Parts with
significantly higher gm run at tens of mA in this gate voltage
range.

What I don't understand: If unity gain is enough, why not just
use the JFETs as source followers?

Disclaimer: I'm not ripping off anybody's product (and besides a
JFET diff amp ain't rocket science). It's just instructive to
look at other people's ideas, but in this case I can't make much
sense of it.

robert

It's probably just an impedance converter, running into a much quieter
bipolar op amp--easier to stabilize and just as good. Using a diff pair
instead of two source followers would improve the negative supply
rejection a bit, I expect, as well as saving one resistor. With BJTs it
would improve the Johnson noise as well, since the small-signal emitter
resistance has a noise temperature of T_ambient_/2, but I really don't
know about JFETs.

Cheers,

Phil Hobbs

 

[Jim Thompson]

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]

Phil Hobbs

Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson

 

[Phil Hobbs]

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]

Phil Hobbs

Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson

Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs

 

[Joerg]

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]

Phil Hobbs

Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson

Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs

Hello Phil,

Tell your mom she did good

Just reading your "Photodiode Front Ends, The Real Story" because I have
to build one. This one has to go up to 100MHz or so. Thinking about
doing it all discrete again but other newsgroupers here have suggested
modern opamps like the OPA847. The noise specs look ok. Would you trust
these new amps?

The photodiode will be AC coupled to the TIA, the DC is handled
elsewhere. Since the PD might be switched to another type some day I'd
cascode it even if going with an amp like the OPA847. That way the
feedback stability doesn't go to pots. My main concern is the usual,
long term availability of the chips.

 

[Jim Thompson]

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]

Phil Hobbs

Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson

Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs

Son-of-a-gun! I've just been reading all your papers ;-)

(Working with the other Phil ;-)

...Jim Thompson

 

[Joerg]

Jim Thompson wrote:

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]


Phil Hobbs


Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson

Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs

Son-of-a-gun! I've just been reading all your papers ;-)

(Working with the other Phil ;-)

...Jim Thompson

Same here. So you are really doing HW design now? The stuff where you
hold a hot piece of iron to some stuff that comes off a spool and melts?

 

[Jim Thompson]

Jim Thompson wrote:


On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]


Phil Hobbs


Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson


Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs

Son-of-a-gun! I've just been reading all your papers ;-)

(Working with the other Phil ;-)

...Jim Thompson

Same here. So you are really doing HW design now?
Sure.

The stuff where you
hold a hot piece of iron to some stuff that comes off a spool and melts?

Naaaah! They have a technician (Bob Toler) from West Virginia to do
that ;-)

...Jim Thompson

 

[Joerg]

Jim Thompson wrote:

On Mon, 27 Nov 2006 18:03:44 -0500, Phil Hobbs



Jim Thompson wrote:



On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]



Phil Hobbs


Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson


Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs


Son-of-a-gun! I've just been reading all your papers ;-)

(Working with the other Phil ;-)

...Jim Thompson

Same here. So you are really doing HW design now?

Sure.

That'll open a whole new world. Now you can use inductors larger than a
few picohenries

Naaaah! They have a technician (Bob Toler) from West Virginia to do
that ;-)

Great. Working on two client projects right now. One client has a nice
large group of technicians, the other has none. Either I'll have to
solder the whole prototype enchilada myself or hire a temp. But how do
you find a good temp?

 

[Spehro Pefhany]

Great. Working on two client projects right now. One client has a nice
large group of technicians, the other has none. Either I'll have to
solder the whole prototype enchilada myself or hire a temp. But how do
you find a good temp?

A temp agency might be one choice.


Best regards,
Spehro Pefhany

 

[Joerg]

A temp agency might be one choice.

I have used them in the past. Was ok for not too difficult jobs but when
in need for someone with skills to solder 0402 parts and hotrod RF
transistors the results were mixed. Hiring engineers that way did not
work at all.

 

[Phil Hobbs]

Son-of-a-gun! I've just been reading all your papers ;-)

(Working with the other Phil ;-)

...Jim Thompson

Re-reading, you mean.

Wow, they must have got some more VC money if they can afford you. I hope
he still owns some stock by the time they're over the hump. Interesting
gizmo, that. (For everyone else: Stheno Corp. is a startup that has an
interesting technology for measuring chirality in biological molecules, by
way of very small amounts of optical activity. It ought to separate the NIH
from a small boatload of money if they can get it into production.)

Cheers,

Phil Hobbs

 

[Jim Thompson]

Re-reading, you mean.

Wow, they must have got some more VC money if they can afford you. I hope
he still owns some stock by the time they're over the hump. Interesting
gizmo, that. (For everyone else: Stheno Corp. is a startup that has an
interesting technology for measuring chirality in biological molecules, by
way of very small amounts of optical activity. It ought to separate the NIH
from a small boatload of money if they can get it into production.)

Cheers,

Phil Hobbs

<Smirk :->

...Jim Thompson

 

[Phil Hobbs]

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]

Phil Hobbs




Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson

Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs

Hello Phil,

Tell your mom she did good

Just reading your "Photodiode Front Ends, The Real Story" because I have
to build one. This one has to go up to 100MHz or so. Thinking about
doing it all discrete again but other newsgroupers here have suggested
modern opamps like the OPA847. The noise specs look ok. Would you trust
these new amps?

The photodiode will be AC coupled to the TIA, the DC is handled
elsewhere. Since the PD might be switched to another type some day I'd
cascode it even if going with an amp like the OPA847. That way the
feedback stability doesn't go to pots. My main concern is the usual,
long term availability of the chips.

Joerg,

That sounds pretty sensible, as long as you don't let the 'elsewhere'
dominate the noise--which can easily happen, e.g. with simulated inductors or
current mirrors. (I'm perfectly confident that you wouldn't make that
mistake, but I've seen others do it pretty frequently.)

The cascode trick decouples the TIA performance pretty well from the fine
details of the amplifier, as you say, so if the chip goes away you ought to
be able to find another one without undue pain. Maybe qualify a couple or three?

I've been doing device work for the last 5 years or so, so my stock of parts
is getting embarrassingly antique--I still have some AD639 trig function
chips that I bought in 1988, as well as some Raytheon 8-bit flash converters.

The other approach is to pick a low-noise 50-ohm amplifier and connect the PD
right to the input--I've done that very successfully with 60-cent Mini
Circuits amps. Since you're AC-coupling anyway, a transformer is another
approach.

How big a photocurrent are you expecting?

Cheers,

Phil Hobbs

 

[Joerg]

Hello Phil,

Jim Thompson wrote:

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]

Phil Hobbs





Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson





Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs

Hello Phil,

Tell your mom she did good

Just reading your "Photodiode Front Ends, The Real Story" because I
have to build one. This one has to go up to 100MHz or so. Thinking
about doing it all discrete again but other newsgroupers here have
suggested modern opamps like the OPA847. The noise specs look ok.
Would you trust these new amps?

The photodiode will be AC coupled to the TIA, the DC is handled
elsewhere. Since the PD might be switched to another type some day I'd
cascode it even if going with an amp like the OPA847. That way the
feedback stability doesn't go to pots. My main concern is the usual,
long term availability of the chips.

Joerg,

That sounds pretty sensible, as long as you don't let the 'elsewhere'
dominate the noise--which can easily happen, e.g. with simulated
inductors or current mirrors. (I'm perfectly confident that you
wouldn't make that mistake, but I've seen others do it pretty frequently.)

One of the toppers I saw lately was a BIG inductor with way too much
inductance. Actually more of a choke. It happened to resonate smack dab
where it hurt the most.

I'll have a nice small UHF capable inductor first and then a lil' larger
one towards the DC circuitry.

The cascode trick decouples the TIA performance pretty well from the
fine details of the amplifier, as you say, so if the chip goes away you
ought to be able to find another one without undue pain. Maybe qualify
a couple or three?

I have read not so good things about the AD8099 in that it was hard or
next to impossible to get it stable. That leaves the OPA847 or maybe one
from the THS series. But they are all different, not drop-in
replacements, picky about the supply voltage and when someone dares to
place more than a few volts across the inputs, bzzzzt-poof. That's why I
was thinking discretes. There are nice cheap GHz transistors that boast
noise figures of around 1dB. Doesn't have to be a real rocket like the
BFP640. A few BFR505 would do fine.

I've been doing device work for the last 5 years or so, so my stock of
parts is getting embarrassingly antique--I still have some AD639 trig
function chips that I bought in 1988, as well as some Raytheon 8-bit
flash converters.

The other approach is to pick a low-noise 50-ohm amplifier and connect
the PD right to the input--I've done that very successfully with 60-cent
Mini Circuits amps. Since you're AC-coupling anyway, a transformer is
another approach.

Looks very enticing, have to check their noise figures. I have never
used one because they are single-sourced but Mini-Circuits is quite a
reliable company.

How big a photocurrent are you expecting?

So far they told me 1mA max. It'll always be above 300uA or so. The
challenge is to detect very minute changes in that, hence the low noise
requirement.

 

[Joerg]

<Smirk :->

...Jim Thompson

I like their abbreviation for Magneto Optical Enantiomeric Detection.
They call it MOPED

Have to confess that I don't even know what enantiomeric is. So are we
now in a race about who can create the lowest noise source/receiver?

 

[Phil Hobbs]

Hello Phil,

Phil Hobbs wrote:

Jim Thompson wrote:

On Mon, 27 Nov 2006 17:53:49 -0500, Phil Hobbs

[snip]

Phil Hobbs






Phil, Would you be Philip C. D. Hobbs?

Stheno ?

...Jim Thompson






Yeah, but it's all my mum's fault.

Cheers,

Phil(ip C. D.) Hobbs





Hello Phil,

Tell your mom she did good

Just reading your "Photodiode Front Ends, The Real Story" because I
have to build one. This one has to go up to 100MHz or so. Thinking
about doing it all discrete again but other newsgroupers here have
suggested modern opamps like the OPA847. The noise specs look ok.
Would you trust these new amps?

The photodiode will be AC coupled to the TIA, the DC is handled
elsewhere. Since the PD might be switched to another type some day
I'd cascode it even if going with an amp like the OPA847. That way
the feedback stability doesn't go to pots. My main concern is the
usual, long term availability of the chips.

Joerg,

That sounds pretty sensible, as long as you don't let the 'elsewhere'
dominate the noise--which can easily happen, e.g. with simulated
inductors or current mirrors. (I'm perfectly confident that you
wouldn't make that mistake, but I've seen others do it pretty
frequently.)

One of the toppers I saw lately was a BIG inductor with way too much
inductance. Actually more of a choke. It happened to resonate smack dab
where it hurt the most.

I'll have a nice small UHF capable inductor first and then a lil' larger
one towards the DC circuitry.

The cascode trick decouples the TIA performance pretty well from the
fine details of the amplifier, as you say, so if the chip goes away
you ought to be able to find another one without undue pain. Maybe
qualify a couple or three?

I have read not so good things about the AD8099 in that it was hard or
next to impossible to get it stable. That leaves the OPA847 or maybe one
from the THS series. But they are all different, not drop-in
replacements, picky about the supply voltage and when someone dares to
place more than a few volts across the inputs, bzzzzt-poof. That's why I
was thinking discretes. There are nice cheap GHz transistors that boast
noise figures of around 1dB. Doesn't have to be a real rocket like the
BFP640. A few BFR505 would do fine.

I've been doing device work for the last 5 years or so, so my stock of
parts is getting embarrassingly antique--I still have some AD639 trig
function chips that I bought in 1988, as well as some Raytheon 8-bit
flash converters.

The other approach is to pick a low-noise 50-ohm amplifier and connect
the PD right to the input--I've done that very successfully with
60-cent Mini Circuits amps. Since you're AC-coupling anyway, a
transformer is another approach.

Looks very enticing, have to check their noise figures. I have never
used one because they are single-sourced but Mini-Circuits is quite a
reliable company.

How big a photocurrent are you expecting?

So far they told me 1mA max. It'll always be above 300uA or so. The
challenge is to detect very minute changes in that, hence the low noise
requirement.

If you have 300 uA to work with, life is easy. A 200-ohm load will get you
to the shot noise limit--so something like a Mini Circuits T4-1 transformer
running into a garden variety 50-ohm amplifier (with no load resistor) will
be the bee's knees from a SNR point of view. Gain stability will probably be
the limitation then.

A cascode stage running into the summing junction of a current-feedback amp
would be another approach. That AD847 doesn't look as though it has the GBW
for a 100-MHz application.

Cheers,

Phil Hobbs