OT: milestone, the x-Chapters

[Winfield Hill]

pcdhobbs@gmail.com wrote...

You aren't getting 0.1% over 8 decades
with 12 bits. Your step size is 0.45%.

The technique can do 1% of a 10^-8 signal.

Wrong. Keep thinking, you can, if you use
multiple channels. OK, let me think and
calculate more carefully. The starlight-
to-sunlight version, with 15 parts, is a
simplified version, it's good to about 1%,
degrading somewhat at the bottom of a range.
OK, another hint, there are simultaneously-
overlapping ranges. But no range changing.


--
Thanks,
- Win

 

You aren't getting 0.1% over 8 decades with 12 bits. Your step size is 0.45%.

You can make a feedback square-rooter using the two FETs as a PWM multiplier.

Cheers

Phil Hobbs

 

[Winfield Hill]

John Larkin wrote...

On 24 Jun 2019, Winfield Hill wrote:

pcdhobbs@gmail.com wrote...

You aren't getting 0.1% over 8 decades
with 12 bits. Your step size is 0.45%.

The technique can do 1% of a 10^-8 signal.

Wrong. Keep thinking, you can, if you use
multiple channels. OK, let me think and
calculate more carefully. The starlight-
to-sunlight version, with 15 parts, is a
simplified version, it's good to about 1%,
degrading somewhat at the bottom of a range.
OK, another hint, there are simultaneously-
overlapping ranges. But no range changing.

Is there a single electrical output that
spans the PD current range?

Yes, the output from the uP and ADC that's
interfaced to the circuit. As I said in a
hint, the ADC converts multiple channels,
coming from the circuit. If you get enough
hints, you may be able to figure it out.


--
Thanks,
- Win

 

[Dave Platt]

In article <qeq97g02isj@drn.newsguy.com>,
Winfield Hill <winfieldhill@yahoo.com> wrote:

Last night we handed over finished latex and
eps files, for AoE The x-Chapters, to our
long-suffering editor at CUP.

There was much rejoicing!

Looking forward to the day when I can order a copy!

 

[John Larkin]

On 24 Jun 2019 12:20:37 -0700, Winfield Hill <winfieldhill@yahoo.com>
wrote:

pcdhobbs@gmail.com wrote...

You aren't getting 0.1% over 8 decades
with 12 bits. Your step size is 0.45%.

The technique can do 1% of a 10^-8 signal.

Wrong. Keep thinking, you can, if you use
multiple channels. OK, let me think and
calculate more carefully. The starlight-
to-sunlight version, with 15 parts, is a
simplified version, it's good to about 1%,
degrading somewhat at the bottom of a range.
OK, another hint, there are simultaneously-
overlapping ranges. But no range changing.

Is there a single electrical output that spans the PD current range?


--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[George Herold]

On Monday, June 24, 2019 at 4:08:50 PM UTC-4, Winfield Hill wrote:

John Larkin wrote...

On 24 Jun 2019, Winfield Hill wrote:

pcdhobbs@gmail.com wrote...

You aren't getting 0.1% over 8 decades
with 12 bits. Your step size is 0.45%.

The technique can do 1% of a 10^-8 signal.

Wrong. Keep thinking, you can, if you use
multiple channels. OK, let me think and
calculate more carefully. The starlight-
to-sunlight version, with 15 parts, is a
simplified version, it's good to about 1%,
degrading somewhat at the bottom of a range.
OK, another hint, there are simultaneously-
overlapping ranges. But no range changing.

Is there a single electrical output that
spans the PD current range?

Yes, the output from the uP and ADC that's
interfaced to the circuit. As I said in a
hint, the ADC converts multiple channels,
coming from the circuit. If you get enough
hints, you may be able to figure it out.

OK three channels, 10^6 could be done with each chan having
output from 100 mV to 10 V. That keeps most offsets small.
With a low offset opamp I guess you could go from 30 mV to 10 V
on each channel.

(You could futz around with math in the uP to make the cross overs look
nicer.)

George H.

--
Thanks,
- Win

 

[Tom Gardner]

On 24/06/19 20:20, Winfield Hill wrote:

pcdhobbs@gmail.com wrote...

You aren't getting 0.1% over 8 decades
with 12 bits. Your step size is 0.45%.

The technique can do 1% of a 10^-8 signal.

Wrong. Keep thinking, you can, if you use
multiple channels. OK, let me think and
calculate more carefully. The starlight-
to-sunlight version, with 15 parts, is a
simplified version, it's good to about 1%,
degrading somewhat at the bottom of a range.
OK, another hint, there are simultaneously-
overlapping ranges. But no range changing.

Old school way of measuring voltage is with
a potential divider and a null meter.

So, the micro causes a current to be generated,
which is balanced against the photocurrent?

Accurate control of current over a wide range
achieved by PWM and filtering, as in modern
voltage calibrators.

 

[Tom Del Rosso]

Winfield Hill wrote:

Winfield Hill wrote...

* A neat TIA amplifier with an 8-decade dynamic
range, 12-bit precision, not a log converter!
Example use: a starlight to sunlight digitizer.

Let's see, 10V out, precision feedback resistor,
10uV input opamp voltage offset, 6-decade range,
with 100% error at the bottom. Raise output to
100V with amp, feedback resistor there, 7-decades.
How get 8 decades with 1% precision (10-decades)?

Maximum current 5mA, minimum FS = 50pA, for 1%
max Ioffset 0.5pA, check!

Maximum Voffset 100V/10^10 = 0.01uV, no way!
Change range resistors, not allowed? Sheesh!
We reveal the secret, a clever, easy-to-use
solution, but it wasn't our idea!

Isn't an idea like that patentable?

 

[George Herold]

On Monday, June 24, 2019 at 4:32:08 PM UTC-4, George Herold wrote:

On Monday, June 24, 2019 at 4:08:50 PM UTC-4, Winfield Hill wrote:
John Larkin wrote...

On 24 Jun 2019, Winfield Hill wrote:

pcdhobbs@gmail.com wrote...

You aren't getting 0.1% over 8 decades
with 12 bits. Your step size is 0.45%.

The technique can do 1% of a 10^-8 signal.

Wrong. Keep thinking, you can, if you use
multiple channels. OK, let me think and
calculate more carefully. The starlight-
to-sunlight version, with 15 parts, is a
simplified version, it's good to about 1%,
degrading somewhat at the bottom of a range.
OK, another hint, there are simultaneously-
overlapping ranges. But no range changing.

Is there a single electrical output that
spans the PD current range?

Yes, the output from the uP and ADC that's
interfaced to the circuit. As I said in a
hint, the ADC converts multiple channels,
coming from the circuit. If you get enough
hints, you may be able to figure it out.
OK three channels, 10^6 could be done with each chan having
output from 100 mV to 10 V. That keeps most offsets small.
With a low offset opamp I guess you could go from 30 mV to 10 V
on each channel.

Or 100mV to 30V. That sounds better.. you can stick an R/ divider
on the output to match the adc V_max, Jfets maybe set the voltage
limits on the two lower current TIAs?
George H.

(You could futz around with math in the uP to make the cross overs look
nicer.)

George H.


--
Thanks,
- Win

 

[Winfield Hill]

Tom Gardner wrote...

Old school way of measuring voltage is with
a potential divider and a null meter.

So, the micro causes a current to be generated,
which is balanced against the photocurrent?

Accurate control of current over a wide range
achieved by PWM and filtering, as in modern
voltage calibrators.

Your "null meter" is going to hae 10nV DC offsets?


--
Thanks,
- Win

 

[Phil Hobbs]

On 6/24/19 3:35 PM, John Larkin wrote:

On 24 Jun 2019 11:51:16 -0700, Winfield Hill <winfieldhill@yahoo.com
wrote:

John Larkin wrote...

On 24 Jun 2019, Winfield Hill wrote:
John Larkin wrote...
Winfield Hill wrote:
Winfield Hill wrote...

* A neat TIA amplifier with an 8-decade dynamic
range, 12-bit precision, not a log converter!
Example use: a starlight to sunlight digitizer.

Let's see, 10V out, precision feedback resistor,
10uV input opamp voltage offset, 6-decade range,
with 100% error at the bottom. Raise output to
100V with amp, feedback resistor there, 7-decades.
How get 8 decades with 1% precision (10-decades)?

Maximum current 5mA, minimum FS = 50pA, for 1%
max Ioffset 0.5pA, check!

Maximum Voffset 100V/10^10 = 0.01uV, no way!
Change range resistors, not allowed? Sheesh!
We reveal the secret, a clever, easy-to-use
solution, but it wasn't our idea!

Hint, starlight to sunlight circuit, 15 parts:
three opamps, eight Rs, two Cs, and two JFETs.
The input opamp is an LMP7701, Vos = 200uV Ib.
One S2387–33R detector. Readout, 12-bit ADC in
uP. 60pA (10% is 6pA) to 3mA, 5 x 10^8 range.
Can you do it?

You can dump both ends of a photodiode into two
TIAs, with different gains. There are also other
uses for that idea.

Yes, that's cute, but this one needs more range.

Suppose you build a TIA with high gain. When it
rails, one could diode-couple the (former)
summing point into the next, lower gain TIA.
With very good diodes of course.

Or connect the first summing point into an emitter
or a fet source, which would conduct when the
TIA gives up. Similar idea.

Or use a nonlinear breakpoint thing in a TIA feedback.

What do you have against log amps?!

Can't see how any of your ideas could be made
to work to 10% 1% let alone 09.1%, exactly.
Getting close maybe, but no cigar.

The progressive-railing thing would be as good as the parts you care
to buy. 0.1% should be possible.

How did your guy do it?

Plessey used to sell RF amp chips with 10 dB gain that dropped to 0 dB
when they railed. The idea was to put N of them in a row to make a log
video amp with a range of 10*N dB. I used them in my heterodyne laser
microscope in grad school, circa 1985, with a medium-fancy AM detector
following, together with a closed-loop calibration system based on the
ring-down of a crystal oscillator.

Once I figured out why the crystal didn't ring down starting at full
amplitude, it all worked great.

Cheers

Phil Hobbs

(It was because the oscillator wasn't running quite at the crystal's
mechanical resonant frequency.)

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Phil Hobbs]

On 6/24/19 3:37 PM, John Larkin wrote:

On Mon, 24 Jun 2019 13:54:54 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 6/24/19 11:01 AM, Winfield Hill wrote:
Winfield Hill wrote...

* A neat TIA amplifier with an 8-decade dynamic
range, 12-bit precision, not a log converter!
Example use: a starlight to sunlight digitizer.

Let's see, 10V out, precision feedback resistor,
10uV input opamp voltage offset, 6-decade range,
with 100% error at the bottom. Raise output to
100V with amp, feedback resistor there, 7-decades.
How get 8 decades with 1% precision (10-decades)?

Maximum current 5mA, minimum FS = 50pA, for 1%
max Ioffset 0.5pA, check!

Maximum Voffset 100V/10^10 = 0.01uV, no way!
Change range resistors, not allowed? Sheesh!
We reveal the secret, a clever, easy-to-use
solution, but it wasn't our idea!

Hint, starlight to sunlight circuit, 15 parts:
three opamps, eight Rs, two Cs, and two JFETs.
The input opamp is an LMP7701, Vos = 200uV Ib.
One S2387–33R detector. Readout, 12-bit ADC in
uP. 60pA (10% is 6pA) to 3mA, 5 x 10^8 range.
Can you do it?



Sure. Run the PD open circuit and measure the voltage. Back in 1990 I
built a beam aiming detector that I still use based on that trick. It
uses a Ge quad cell and uses two op amps to form V1-V3 and V2-V4. Two
centre-zero, edge-reading taut-band panel meters read X and Y.

It works over about four decades of photocurrent because the shunt
resistance of the Ge quad cell is the pits, but I've used it from < 400
nm to >1600 nm.

Cheers

Phil Hobbs

Could you squirt a little current into the PD occasionally, to make a
tempco compensation measurement?

Sure. Anything with a 10**8:1 range is going to be slow as molasses
down at the bottom, so why not?

If you're okay with millihertz bandwidths, NIST folks have done 10**14:1
with a single photodiode. (Gary Oppeldauer et al.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Phil Hobbs]

On 6/24/19 3:20 PM, Winfield Hill wrote:

pcdhobbs@gmail.com wrote...

You aren't getting 0.1% over 8 decades
with 12 bits. Your step size is 0.45%.

The technique can do 1% of a 10^-8 signal.

Wrong. Keep thinking, you can, if you use
multiple channels. OK, let me think and
calculate more carefully. The starlight-
to-sunlight version, with 15 parts, is a
simplified version, it's good to about 1%,
degrading somewhat at the bottom of a range.
OK, another hint, there are simultaneously-
overlapping ranges. But no range changing.

TIAs with input resistors can share photocurrent--the current just
divides according to the conductance of the various paths. But that's
super slow and noisy.

Cheers

Phil Hobbs


--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Winfield Hill]

George Herold wrote...

George Herold wrote:

OK three channels, 10^6 could be done with each chan
having output from 100 mV to 10 V. That keeps most
offsets small. ...

Or 100mV to 30V. That sounds better...

A TIA has an opamp SJ, feedback resistor, and output.
How'ya gonna get extra output channels?


--
Thanks,
- Win

 

[George Herold]

On Monday, June 24, 2019 at 8:47:16 PM UTC-4, Winfield Hill wrote:

George Herold wrote...
George Herold wrote:

OK three channels, 10^6 could be done with each chan
having output from 100 mV to 10 V. That keeps most
offsets small. ...

Or 100mV to 30V. That sounds better...

A TIA has an opamp SJ, feedback resistor, and output.
How'ya gonna get extra output channels?

OK this is probably stupid, but I was going to feed all the outputs
back to one common summing junction. (All the opamps tied together..
right that makes the input capacitance three times worse, slower)
The current has to go somewhere.

GH

--
Thanks,
- Win

 

[John Larkin]

On Mon, 24 Jun 2019 21:16:14 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 6/24/19 3:35 PM, John Larkin wrote:
On 24 Jun 2019 11:51:16 -0700, Winfield Hill <winfieldhill@yahoo.com
wrote:

John Larkin wrote...

On 24 Jun 2019, Winfield Hill wrote:
John Larkin wrote...
Winfield Hill wrote:
Winfield Hill wrote...

* A neat TIA amplifier with an 8-decade dynamic
range, 12-bit precision, not a log converter!
Example use: a starlight to sunlight digitizer.

Let's see, 10V out, precision feedback resistor,
10uV input opamp voltage offset, 6-decade range,
with 100% error at the bottom. Raise output to
100V with amp, feedback resistor there, 7-decades.
How get 8 decades with 1% precision (10-decades)?

Maximum current 5mA, minimum FS = 50pA, for 1%
max Ioffset 0.5pA, check!

Maximum Voffset 100V/10^10 = 0.01uV, no way!
Change range resistors, not allowed? Sheesh!
We reveal the secret, a clever, easy-to-use
solution, but it wasn't our idea!

Hint, starlight to sunlight circuit, 15 parts:
three opamps, eight Rs, two Cs, and two JFETs.
The input opamp is an LMP7701, Vos = 200uV Ib.
One S2387–33R detector. Readout, 12-bit ADC in
uP. 60pA (10% is 6pA) to 3mA, 5 x 10^8 range.
Can you do it?

You can dump both ends of a photodiode into two
TIAs, with different gains. There are also other
uses for that idea.

Yes, that's cute, but this one needs more range.

Suppose you build a TIA with high gain. When it
rails, one could diode-couple the (former)
summing point into the next, lower gain TIA.
With very good diodes of course.

Or connect the first summing point into an emitter
or a fet source, which would conduct when the
TIA gives up. Similar idea.

Or use a nonlinear breakpoint thing in a TIA feedback.

What do you have against log amps?!

Can't see how any of your ideas could be made
to work to 10% 1% let alone 09.1%, exactly.
Getting close maybe, but no cigar.

The progressive-railing thing would be as good as the parts you care
to buy. 0.1% should be possible.

How did your guy do it?


Plessey used to sell RF amp chips with 10 dB gain that dropped to 0 dB
when they railed. The idea was to put N of them in a row to make a log
video amp with a range of 10*N dB. I used them in my heterodyne laser
microscope in grad school, circa 1985, with a medium-fancy AM detector
following, together with a closed-loop calibration system based on the
ring-down of a crystal oscillator.

Once I figured out why the crystal didn't ring down starting at full
amplitude, it all worked great.

Cheers

Phil Hobbs

(It was because the oscillator wasn't running quite at the crystal's
mechanical resonant frequency.)

I invented the successive-detection log idea when I was a kid. And I
invented the dual-slope ADC. Coulda got rich on either one.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

 

[Winfield Hill]

John Larkin wrote...

The progressive-railing thing would be as good
as the parts you care to buy. 0.1% should be
possible.

OK, precision railing to say 1%, OK. But at
the highest gains, the fixed-value input-stage
feedback resistor, and the input offset ...
Explain that for us.


--
Thanks,
- Win

 

[Tom Gardner]

On 25/06/19 01:42, Winfield Hill wrote:

Tom Gardner wrote...


Old school way of measuring voltage is with
a potential divider and a null meter.

So, the micro causes a current to be generated,
which is balanced against the photocurrent?

Accurate control of current over a wide range
achieved by PWM and filtering, as in modern
voltage calibrators.

Your "null meter" is going to hae 10nV DC offsets?

Modulate/chop that balancing current?

I don't think you've mentioned bandwidth,
so I can fall back on "the solution to all
signal processing problems is to integrate
for longer"

 

[Jan Panteltje]

On a sunny day (24 Jun 2019 11:48:23 -0700) it happened Winfield Hill
<winfieldhill@yahoo.com> wrote in <qer5tn0prp@drn.newsguy.com>:

Jan Panteltje wrote...

On 24 Jun 2019, Winfield Hill wrote
George Herold wrote...
On June 24, 2019, Winfield Hill wrote:

Hint, starlight to sunlight circuit, 15 parts:
three opamps, eight Rs, two Cs, and two JFETs.
The input opamp is an LMP7701, Vos =3D 200uV Ib.
One S2387-33R detector. Readout, 12-bit ADC in
uP. 60pA (10% is 6pA) to 3mA, 5 x 10^8 range.
Can you do it?

Well only if I'm allowed to change feedback R's. :^)

Nope, only one input stage allowed, and it takes
all the photo-current. You can play games with
the feedback resistor, but you're not allowed to
disconnect one and connect another, with a switch,
because that'd interrupt the measurement.

Sometimes, once you know something is possible,
you can figure out how to do it. But I don't
think I could have solved this puzzle, even with
that knowledge. Once I saw it done, I had to
think carefully to understand it. But others
are smarter than me, and may get the answer.

You would not be using a JFET as AGC element?

As an AGC, no, could that be accurate to 1% or better?

That depends, in radios measuring the AGC voltage gives a good idea of signal strength from uV to volts on a S meter,
but it depends on the drift on the gain components what the accuracy is of course.
S meters give just a gross indication.
But I see no reason that would not work with a high precision gain system.
Sort of a log scale.

As to 'starlight to sunlight',
have you ever used the Sony Super HAD cameras?
I have 3 for those as security camera, super picture with only starlight and same in bright sunlight,
analog out NTSC/ PAL. 35 $ I think I payed for each ebay, 0.01 Lux...
Tested one once hanging out of the window looking straight up in a cloudy sky, sees stars, satellites, planes...
has color. maybe something to monitor your Bs?

 

[George Herold]

On Monday, June 24, 2019 at 10:19:26 PM UTC-4, George Herold wrote:

On Monday, June 24, 2019 at 8:47:16 PM UTC-4, Winfield Hill wrote:
George Herold wrote...
George Herold wrote:

OK three channels, 10^6 could be done with each chan
having output from 100 mV to 10 V. That keeps most
offsets small. ...

Or 100mV to 30V. That sounds better...

A TIA has an opamp SJ, feedback resistor, and output.
How'ya gonna get extra output channels?

OK this is probably stupid, but I was going to feed all the outputs
back to one common summing junction. (All the opamps tied together..
right that makes the input capacitance three times worse, slower)
The current has to go somewhere.

OK that is dumb.
Can you chain together TIA stages? Lowest gain first.


+---R2---+
| |\ |
+---R1---+-|-\ |
| |\ | >---+
I_in---+-|-\ .-|+/
| >---+ |/
Gnd-|+/
|/

I'm not sure that works.. and I have to do something when the 2nd TIA
rails... short it's inputs or something.

George H.

GH


--
Thanks,
- Win