9-decade transimpedance amplifier

[piglet]

On 27/06/2019 2:06 am, George Herold wrote:

On Wednesday, June 26, 2019 at 5:22:34 PM UTC-4, piglet wrote:
On 26/06/2019 5:48 pm, Winfield Hill wrote:
Here's a TIA circuit published in 2012, in RSI,
by Yale physicist, Stephen Eckel. “A high dynamic
range, linear response transimpedance amplifier.”

It's easy to implement, and super useful. The TIA
has multiple ranges, each with its own output, but
multiple ranges are active at once; there's no loss
of data as would happen switching range resistors.

Stephen and his co-authors found a simple, clever
trick to prevent input TIA opamp saturation, using
JFETs to successively short series-placed higher-
value range resistors for strong input currents.

They suggest a three-stage implementation, with a
300:1 ratio for each, but you can use many stages
(each one takes few extra parts), to obtain high
accuracy with a say 12-bit ADC. Also, a high
input-opamp Vos needn't degrade the dynamic range.

DropBox has a draft of our x-Chapters write-up:
https://www.dropbox.com/s/fs4edz7dqgwswoj/4x.3.7_Eckel_TIA.pdf?dl=0

I think you can download Stephen's RSI article here:
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=8&ved=2ahUKEwjk6KvVwYfjAhVlRN8KHaMfD48QFjAHegQICRAC&url=http%3A%2F%2Fwww.bmo.physik.uni-muenchen.de%2F~riedle%2FElektronik_I%2FKW103%2F2012_Eckel%2CSushkov_9-decade_RSI.pdf&usg=AOvVaw3g9i6-pWAwuuZlZLvlFArd




Marvellous - thank you!

Jfets are nice because when fully depleted off there is no parasitic s-d
diode that curses gumdrop mosfets. But gumdrop mosfets are much cheaper
than jfets these days. I wonder if one could use series connected
source-source 2N7002 and s-s BSS84 to replace those jfets? Eight mosfets
could work out cheaper than four jfets? While gate leakage should be a
non-issue I don't know how s-d channel leakage compares?

piglet

I was going to ask what about source-source fets, and found this,
https://electronics.stackexchange.com/questions/79028/understanding-two-mosfet-with-sources-connected

Is that right... just looking at pics not reading comments/ words.

George H.

Yes, here is sketch showing how might be done using mosfets to replace
jfets:

<https://www.dropbox.com/s/bidm47zj29moq6k/SteppedTIA.jpg?dl=0>

piglet

 

[Winfield Hill]

George Herold wrote...

We sell a PD with a 10 position switch, 1,3.3,10...
It's OK, when I tried to make it faster I found
the switch added ~5 pF of C. Otherwise I'm not
sure what's wrong with a gain switch.

Well, as I said, the Eckel scheme, with outputs
running simultaneously, insures fast, accurate
digitizing, especially with noisy signals. But
John's scheme looks very practical for separating
a high Rf first stage from a lower Rf second stage.
The first stage would not suffer from any extra
capacitances to slow it down. You could safely
add switches to the lower-gain stage to get lots
of ranges. I might use 10x for 1st to 2nd stage,
to insure speed, and do the 3.3x bit later.


--
Thanks,
- Win

 

[Winfield Hill]

piglet wrote...

On 27/06/2019 2:06 am, George Herold wrote:
On June 26, 2019, piglet wrote:

Jfets are nice because when fully depleted off there is no parasitic s-d
diode that curses gumdrop mosfets. But gumdrop mosfets are much cheaper
than jfets these days. I wonder if one could use series connected
source-source 2N7002 and s-s BSS84 to replace those jfets? Eight mosfets
could work out cheaper than four jfets? While gate leakage should be a
non-issue I don't know how s-d channel leakage compares?

I was going to ask what about source-source fets, and found this,
https://electronics.stackexchange.com/questions/79028/understanding-two-mosfet-with-sources-connected

Is that right... just looking at pics not reading comments/ words.

Yes, here is sketch showing how might be done using
mosfets to replace jfets:

https://www.dropbox.com/s/bidm47zj29moq6k/SteppedTIA.jpg?dl=0

I'm not sure you need the 2nd MOSFET. With MOSFET
sources to the opamp output, all MOSFETs are off
for voltages within your +/-(9V + Vgs-th) window.
A MOSFET problem is finding parts with low Coss.
E.g., BSS123 and BSS84 are about 7pF. This is
across your highest-value feedback resistor.


--
Thanks,
- Win

 

[Jan Panteltje]

On a sunny day (Thu, 27 Jun 2019 06:52:16 -0700) it happened John Larkin
<jjlarkin@highlandtechnology.com> wrote in
<usg9hehn975p0k6sm43ous31hla2g17neq@4ax.com>:

One day I want to try to visibly light an LED from our ambient RF.

I have done that, connect a LED to my CB GPA antenna,
it lit up when neighbor across the street used his set.
I called him, asked how many watts he was using.
eeehh


May also work with your cellphone if close enough.

 

[John Larkin]

On 27 Jun 2019 05:19:44 -0700, Winfield Hill <winfieldhill@yahoo.com>
wrote:

George Herold wrote...

We sell a PD with a 10 position switch, 1,3.3,10...
It's OK, when I tried to make it faster I found
the switch added ~5 pF of C. Otherwise I'm not
sure what's wrong with a gain switch.

Well, as I said, the Eckel scheme, with outputs
running simultaneously, insures fast, accurate
digitizing, especially with noisy signals. But
John's scheme looks very practical for separating
a high Rf first stage from a lower Rf second stage.
The first stage would not suffer from any extra
capacitances to slow it down. You could safely
add switches to the lower-gain stage to get lots
of ranges. I might use 10x for 1st to 2nd stage,
to insure speed, and do the 3.3x bit later.

I am currently designing a fiberoptic clock distribution system, and
want to light an LED if there is activity on an ECL differential
signal pair. The cheap way to do that is with an AC-coupled diode
detector and a comparator, with very small pickoff caps driving the
diodes. That gets me into understanding the behavior of low-barrier
schottky diodes (SMS7621, BAT15) near zero volts. My little
multi-range photodiode amp also cares about "sub-threshold" diode
behavior.

I did low-voltage measurement on an SMS7621 yesterday, but it ocurrs
to me now that I may have been detecting ambient RF, which is after
all what those diodes are made for. I should do it again, without the
leads to power supplies and DVMs acting like nice antannas.

One day I want to try to visibly light an LED from our ambient RF.

I wonder how good the diode Spice models are at, say, +-0.5 volts,
especially close to zero volts. I guess I'll find out. The data sheets
are "RF", all about silly things like sine wave dBms.

The more expensive, and less interesting, way to do my signal detector
is digitally, namely to clock an EP52 flop off my diff pair and reset
it periodically or something.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

 

[piglet]

On 27/06/2019 2:10 pm, Winfield Hill wrote:

piglet wrote...

On 27/06/2019 2:06 am, George Herold wrote:
On June 26, 2019, piglet wrote:

Jfets are nice because when fully depleted off there is no parasitic s-d
diode that curses gumdrop mosfets. But gumdrop mosfets are much cheaper
than jfets these days. I wonder if one could use series connected
source-source 2N7002 and s-s BSS84 to replace those jfets? Eight mosfets
could work out cheaper than four jfets? While gate leakage should be a
non-issue I don't know how s-d channel leakage compares?

I was going to ask what about source-source fets, and found this,
https://electronics.stackexchange.com/questions/79028/understanding-two-mosfet-with-sources-connected

Is that right... just looking at pics not reading comments/ words.

Yes, here is sketch showing how might be done using
mosfets to replace jfets:

https://www.dropbox.com/s/bidm47zj29moq6k/SteppedTIA.jpg?dl=0

I'm not sure you need the 2nd MOSFET. With MOSFET
sources to the opamp output, all MOSFETs are off
for voltages within your +/-(9V + Vgs-th) window.
A MOSFET problem is finding parts with low Coss.
E.g., BSS123 and BSS84 are about 7pF. This is
across your highest-value feedback resistor.

But without the 2nd mosfet you'd effectively have anti-parallel diodes
across the feedback resistors and be limited to 0.6V?

Good point about capacitance.

piglet

 

[Winfield Hill]

piglet wrote...

On 27/06/2019 2:10 pm, Winfield Hill wrote:
piglet wrote...

On 27/06/2019 2:06 am, George Herold wrote:
On June 26, 2019, piglet wrote:

Jfets are nice because when fully depleted off there is no parasitic s-d
diode that curses gumdrop mosfets. But gumdrop mosfets are much cheaper
than jfets these days. I wonder if one could use series connected
source-source 2N7002 and s-s BSS84 to replace those jfets? Eight mosfets
could work out cheaper than four jfets? While gate leakage should be a
non-issue I don't know how s-d channel leakage compares?

I was going to ask what about source-source fets, and found this,
https://electronics.stackexchange.com/questions/79028/understanding-two-mosfet-with-sources-connected

Is that right... just looking at pics not reading comments/ words.

Yes, here is sketch showing how might be done using
mosfets to replace jfets:

https://www.dropbox.com/s/bidm47zj29moq6k/SteppedTIA.jpg?dl=0

I'm not sure you need the 2nd MOSFET. With MOSFET
sources to the opamp output, all MOSFETs are off
for voltages within your +/-(9V + Vgs-th) window.
A MOSFET problem is finding parts with low Coss.
E.g., BSS123 and BSS84 are about 7pF. This is
across your highest-value feedback resistor.

But without the 2nd mosfet you'd effectively have
anti-parallel diodes across the feedback resistors
and be limited to 0.6V?

OK, good point, for a bipolar version. But OK for
unipolar. But I'd just add an ordinary signal diode.

> Good point about capacitance.

Carried to the extreme, it rules out JFETs too,
and one would have to use John's or Phil's scheme.


--
Thanks,
- Win

 

[George Herold]

On Thursday, June 27, 2019 at 4:16:06 AM UTC-4, piglet wrote:

On 27/06/2019 2:06 am, George Herold wrote:
On Wednesday, June 26, 2019 at 5:22:34 PM UTC-4, piglet wrote:
On 26/06/2019 5:48 pm, Winfield Hill wrote:
Here's a TIA circuit published in 2012, in RSI,
by Yale physicist, Stephen Eckel. “A high dynamic
range, linear response transimpedance amplifier.”

It's easy to implement, and super useful. The TIA
has multiple ranges, each with its own output, but
multiple ranges are active at once; there's no loss
of data as would happen switching range resistors.

Stephen and his co-authors found a simple, clever
trick to prevent input TIA opamp saturation, using
JFETs to successively short series-placed higher-
value range resistors for strong input currents.

They suggest a three-stage implementation, with a
300:1 ratio for each, but you can use many stages
(each one takes few extra parts), to obtain high
accuracy with a say 12-bit ADC. Also, a high
input-opamp Vos needn't degrade the dynamic range.

DropBox has a draft of our x-Chapters write-up:
https://www.dropbox.com/s/fs4edz7dqgwswoj/4x.3.7_Eckel_TIA.pdf?dl=0

I think you can download Stephen's RSI article here:
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=8&ved=2ahUKEwjk6KvVwYfjAhVlRN8KHaMfD48QFjAHegQICRAC&url=http%3A%2F%2Fwww.bmo.physik.uni-muenchen.de%2F~riedle%2FElektronik_I%2FKW103%2F2012_Eckel%2CSushkov_9-decade_RSI.pdf&usg=AOvVaw3g9i6-pWAwuuZlZLvlFArd




Marvellous - thank you!

Jfets are nice because when fully depleted off there is no parasitic s-d
diode that curses gumdrop mosfets. But gumdrop mosfets are much cheaper
than jfets these days. I wonder if one could use series connected
source-source 2N7002 and s-s BSS84 to replace those jfets? Eight mosfets
could work out cheaper than four jfets? While gate leakage should be a
non-issue I don't know how s-d channel leakage compares?

piglet

I was going to ask what about source-source fets, and found this,
https://electronics.stackexchange.com/questions/79028/understanding-two-mosfet-with-sources-connected

Is that right... just looking at pics not reading comments/ words.

George H.


Yes, here is sketch showing how might be done using mosfets to replace
jfets:

https://www.dropbox.com/s/bidm47zj29moq6k/SteppedTIA.jpg?dl=0

piglet

Thanks piglet. I didn't know that Jfet's have no S-D body diode.
(Why the heck is that? so much I don't know!)

George H.

 

[George Herold]

On Thursday, June 27, 2019 at 8:20:01 AM UTC-4, Winfield Hill wrote:

George Herold wrote...

We sell a PD with a 10 position switch, 1,3.3,10...
It's OK, when I tried to make it faster I found
the switch added ~5 pF of C. Otherwise I'm not
sure what's wrong with a gain switch.

Well, as I said, the Eckel scheme, with outputs
running simultaneously, insures fast, accurate
digitizing, especially with noisy signals. But
John's scheme looks very practical for separating
a high Rf first stage from a lower Rf second stage.
The first stage would not suffer from any extra
capacitances to slow it down. You could safely
add switches to the lower-gain stage to get lots
of ranges. I might use 10x for 1st to 2nd stage,
to insure speed, and do the 3.3x bit later.


--
Thanks,
- Win

Thanks Win, I like the trick. I'm wondering what the application is, where
having to switch ranges (with a switch) is a problem. Our diode laser
can use a large range of measurement currents (mA to ~10 nA) But not
at the same time.

George H.

 

[George Herold]

On Friday, June 28, 2019 at 8:11:36 AM UTC-4, George Herold wrote:

On Thursday, June 27, 2019 at 4:16:06 AM UTC-4, piglet wrote:
On 27/06/2019 2:06 am, George Herold wrote:
On Wednesday, June 26, 2019 at 5:22:34 PM UTC-4, piglet wrote:
On 26/06/2019 5:48 pm, Winfield Hill wrote:
Here's a TIA circuit published in 2012, in RSI,
by Yale physicist, Stephen Eckel. “A high dynamic
range, linear response transimpedance amplifier.”

It's easy to implement, and super useful. The TIA
has multiple ranges, each with its own output, but
multiple ranges are active at once; there's no loss
of data as would happen switching range resistors.

Stephen and his co-authors found a simple, clever
trick to prevent input TIA opamp saturation, using
JFETs to successively short series-placed higher-
value range resistors for strong input currents.

They suggest a three-stage implementation, with a
300:1 ratio for each, but you can use many stages
(each one takes few extra parts), to obtain high
accuracy with a say 12-bit ADC. Also, a high
input-opamp Vos needn't degrade the dynamic range.

DropBox has a draft of our x-Chapters write-up:
https://www.dropbox.com/s/fs4edz7dqgwswoj/4x.3.7_Eckel_TIA.pdf?dl=0

I think you can download Stephen's RSI article here:
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=8&ved=2ahUKEwjk6KvVwYfjAhVlRN8KHaMfD48QFjAHegQICRAC&url=http%3A%2F%2Fwww.bmo.physik.uni-muenchen.de%2F~riedle%2FElektronik_I%2FKW103%2F2012_Eckel%2CSushkov_9-decade_RSI.pdf&usg=AOvVaw3g9i6-pWAwuuZlZLvlFArd




Marvellous - thank you!

Jfets are nice because when fully depleted off there is no parasitic s-d
diode that curses gumdrop mosfets. But gumdrop mosfets are much cheaper
than jfets these days. I wonder if one could use series connected
source-source 2N7002 and s-s BSS84 to replace those jfets? Eight mosfets
could work out cheaper than four jfets? While gate leakage should be a
non-issue I don't know how s-d channel leakage compares?

piglet

I was going to ask what about source-source fets, and found this,
https://electronics.stackexchange.com/questions/79028/understanding-two-mosfet-with-sources-connected

Is that right... just looking at pics not reading comments/ words.

George H.


Yes, here is sketch showing how might be done using mosfets to replace
jfets:

https://www.dropbox.com/s/bidm47zj29moq6k/SteppedTIA.jpg?dl=0

piglet

Thanks piglet. I didn't know that Jfet's have no S-D body diode.
(Why the heck is that? so much I don't know!)

George H.

So I was searching for a model of a jfet.
Figure 1 here,
http://www.linearsystems.com/lsdata/others/LIS_White_Paper_Consider_Discrete_JFET.pdf

Is the gate and 'back gate' connected together?

George H.

 

[Winfield Hill]

George Herold wrote...

Winfield Hill wrote:
George Herold wrote...

We sell a PD with a 10 position switch, 1,3.3,10...
It's OK, when I tried to make it faster I found
the switch added ~5 pF of C. Otherwise I'm not
sure what's wrong with a gain switch.

Well, as I said, the Eckel scheme, with outputs
running simultaneously, insures fast, accurate
digitizing, especially with noisy signals. But
John's scheme looks very practical for separating
a high Rf first stage from a lower Rf second stage.
The first stage would not suffer from any extra
capacitances to slow it down. You could safely
add switches to the lower-gain stage to get lots
of ranges. I might use 10x for 1st to 2nd stage,
to insure speed, and do the 3.3x bit later.

Thanks Win, I like the trick. I'm wondering what
the application is, where having to switch ranges
(with a switch) is a problem.

A switch, with wiring to the panel, will add 1 to
5pF across the feedback resistor. We often make
fast TIAs, with high f_T opamps, that make use of
the intrinsic 0.1pF capacitance of many resistors.
We go further, and trick the resistor into having
even less capacitance, see Figure 8.80-C. These
applications could use John's scheme to separate
the highest-gain stage from the rest.


--
Thanks,
- Win

 

[John Larkin]

On 28 Jun 2019 08:02:08 -0700, Winfield Hill <winfieldhill@yahoo.com>
wrote:

George Herold wrote...

Winfield Hill wrote:
George Herold wrote...

We sell a PD with a 10 position switch, 1,3.3,10...
It's OK, when I tried to make it faster I found
the switch added ~5 pF of C. Otherwise I'm not
sure what's wrong with a gain switch.

Well, as I said, the Eckel scheme, with outputs
running simultaneously, insures fast, accurate
digitizing, especially with noisy signals. But
John's scheme looks very practical for separating
a high Rf first stage from a lower Rf second stage.
The first stage would not suffer from any extra
capacitances to slow it down. You could safely
add switches to the lower-gain stage to get lots
of ranges. I might use 10x for 1st to 2nd stage,
to insure speed, and do the 3.3x bit later.

Thanks Win, I like the trick. I'm wondering what
the application is, where having to switch ranges
(with a switch) is a problem.

A switch, with wiring to the panel, will add 1 to
5pF across the feedback resistor. We often make
fast TIAs, with high f_T opamps, that make use of
the intrinsic 0.1pF capacitance of many resistors.
We go further, and trick the resistor into having
even less capacitance, see Figure 8.80-C. These
applications could use John's scheme to separate
the highest-gain stage from the rest.

You can sometimes switch things around with low-capacitance diodes or
phemts or even multiplexers, so rotary switch capacitance and wiring
aren't in the circuit. The switch is cold, just DC.

These Fujitsu relays are dynamite.

https://www.dropbox.com/s/14mt8y78cc5ng79/Relays.jpg?raw=1

https://www.dropbox.com/s/se162xpw86hpmzs/DSC06884.JPG?raw=1

DPDT, fraction of a pF, fraction of an ohm, good to a few GHz. But
don't water wash them.



--

John Larkin Highland Technology, Inc

lunatic fringe electronics

 

[Dave Platt]

In article <b549465b-526d-4394-bf23-84157436d2cd@googlegroups.com>,
George Herold <gherold@teachspin.com> wrote:

So I was searching for a model of a jfet.
Figure 1 here,
http://www.linearsystems.com/lsdata/others/LIS_White_Paper_Consider_Discrete_JFET.pdf

Is the gate and 'back gate' connected together?

From https://www.nxp.com/docs/en/application-note/AN211A.pdf I see

"The substrate, which functions as Gate 2 of Figure 1, is
of relatively low resistivity material to maximize gain. For the
same purpose, Gate 1 is of very low resistivity material,
allowing the depletion region to spread mostly into the n-type
channel. In most cases the gates are internally connected
together. A tetrode device can be realized by not making
this internal connection."

I have a few surplus JFETs from Linear Integrated Systems which do
have an active fourth lead - I believe it's the substrate (gate 2).
If I recall correctly one can either tie the substrate to the gate, or
to the source, or to a constant voltage which is more negative than
the source (in the case of an N-JFET).

 

[Piglet]

On 28/06/2019 18:51, Dave Platt wrote:

In article <b549465b-526d-4394-bf23-84157436d2cd@googlegroups.com>,
George Herold <gherold@teachspin.com> wrote:
So I was searching for a model of a jfet.
Figure 1 here,
http://www.linearsystems.com/lsdata/others/LIS_White_Paper_Consider_Discrete_JFET.pdf

Is the gate and 'back gate' connected together?

From https://www.nxp.com/docs/en/application-note/AN211A.pdf I see

"The substrate, which functions as Gate 2 of Figure 1, is
of relatively low resistivity material to maximize gain. For the
same purpose, Gate 1 is of very low resistivity material,
allowing the depletion region to spread mostly into the n-type
channel. In most cases the gates are internally connected
together. A tetrode device can be realized by not making
this internal connection."

I have a few surplus JFETs from Linear Integrated Systems which do
have an active fourth lead - I believe it's the substrate (gate 2).
If I recall correctly one can either tie the substrate to the gate, or
to the source, or to a constant voltage which is more negative than
the source (in the case of an N-JFET).

Having substrate access sounds like it could be useful - shame the jfets
I use (J177, J113, J107, BF256 etc) only have three pins

piglet