stability for an op-amp integrator

[starfire]

Hi all -

I'm working on a project that requires a stable, repeatable integration of a
35ms, 4.1VDC pulse input to produce a negative-going ramp output.

I've constructed an integrator using an LF347 quad, JFET input, op-amp with
a 0.01uF cap in the feedback (output to negative input) and a 1.82Mohm input
resistor into the negative input. I've also tied a 1.82Mohm resistor to
ground from the non-inverting input. The source signal is a negative-going,
TTL pulse so I ran it into the gate of an N-channel MOSFET. The source of
the MOSFET is grounded and the drain is pulled up to a regulated 4.1VDC
supply through a 100K resistor. The "on" resistance of the MOSFET is about
5 ohms (2N7000). I tapped the drain of the MOSFET into the positive input
of one stage of the LF347 set up as a voltage follower. The output of the
voltage follower drives the input resistor to the integrator. The output of
the integrator goes to another voltage follower and a 10K resistor to
ground. I'm using 1% carbon resistors and a 10% polyester capacitor.

The circuit will be used to generate multiple consecutive ramps, so an input
reset signal is received prior to the pulse associated with the ramp. This
reset will be used to enable an analog switch or a MOSFET to discharge the
capacitor in a short period prior to starting the ramp.

The circuit needs to be very linear and repeatable.

Is there something I can add to this circuit to make it more stable and
repeatable?

What will be some of the major points to be concerned with for reducing
noise?

Does this sound like a viable approach or is there a better way?

Thanks for any constuctive input.

Dave

 

[John Larkin]

Hi all -

I'm working on a project that requires a stable, repeatable integration of a
35ms, 4.1VDC pulse input to produce a negative-going ramp output.

I've constructed an integrator using an LF347 quad, JFET input, op-amp with
a 0.01uF cap in the feedback (output to negative input) and a 1.82Mohm input
resistor into the negative input.

I've also tied a 1.82Mohm resistor to
ground from the non-inverting input.

Don't do that; just ground the n.i. input. The resistor will make
noise.

The source signal is a negative-going,
TTL pulse so I ran it into the gate of an N-channel MOSFET. The source of
the MOSFET is grounded and the drain is pulled up to a regulated 4.1VDC
supply through a 100K resistor. The "on" resistance of the MOSFET is about
5 ohms (2N7000). I tapped the drain of the MOSFET into the positive input
of one stage of the LF347 set up as a voltage follower. The output of the
voltage follower drives the input resistor to the integrator.

If you make the mosfet pullup resistor smaller, it will switch a lot
faster and you can eliminate the buffer stage. Say, a 10K pullup, then
directly into the 1.8M thingie. Better yet, use a SPDT CMOS analog
switch instead of the fet and resistor; you can use the other switch
sections to do your reset thing.

The output of
the integrator goes to another voltage follower and a 10K resistor to
ground. I'm using 1% carbon resistors and a 10% polyester capacitor.

Get rid of all that, too. So it evolves to a single opamp section, a
dual SPDT CMOS analog switch, one resistor, and one cap.

The circuit will be used to generate multiple consecutive ramps, so an input
reset signal is received prior to the pulse associated with the ramp. This
reset will be used to enable an analog switch or a MOSFET to discharge the
capacitor in a short period prior to starting the ramp.

Get a switch with low charge injection.

The circuit needs to be very linear and repeatable.

Is there something I can add to this circuit to make it more stable and
repeatable?

Use a low TC resistor and a cap with low TC and low dielectric
absorption. Polystyrene or PPC or maybe polycarb; polyester sucks. For
extreme precision, you'll have to correct for the cap's negative TC,
probably by adding a small temperature correction to the 4.1 volt
reference.

What's this for?

John

 

[starfire]

Thanks for the feedback, John.

I'll try everything you're suggesting. I need to reset the integration
capacitor completely in less than 1 microsecond so I need an analog switch
with a very low "on" resistance. I have found the Maxim MAX4608 with a 2.5
ohm "on" resistance but, unfortunately, they don't seem to be in stock at
DigiKey. Do you have a part number for an analog switch that may work in
this instance?

This circuit is for timing measurements in a nuclear instrumentation
application. When an external pulse is received, the integration capacitor
needs to reset within a microsecond then allow the integrator to ramp to a
repeatable voltage, based on the width of an input square pulse. If another
pulse is received prior to the completion of the integration period, the
capacitor is reset again and the integration cycle begins again.

Thanks, again, for your input, especially your insight into the capacitor
types. I'm starting to order parts to experiment with and for the final
design (hopefully) and it sounds like I should delete the polyester parts.
Is it still a problem with 5% parts. I've found polystyrenes in 2%, though,
and it sounds like that may be a better selection.

Dave

 

[John Larkin]

Thanks for the feedback, John.

I'll try everything you're suggesting. I need to reset the integration
capacitor completely in less than 1 microsecond so I need an analog switch
with a very low "on" resistance. I have found the Maxim MAX4608 with a 2.5
ohm "on" resistance but, unfortunately, they don't seem to be in stock at
DigiKey. Do you have a part number for an analog switch that may work in
this instance?

The HC4066 (and 405x) types are fast and have pretty low on
resistances, but only work to +-5 volt rails and, like all low Ron
switches, will have lots of charge injection. Paralleling sections
improves Ron and makes the charge thing worse. But, then, .01 uF is a
pretty big cap.

Analog devices has some nice switches. Try Vishay (Siliconix) too. Or
a gaasfet if things get extreme: a CLY2 goes to about 2 ohms at zero
gate voltage.

This circuit is for timing measurements in a nuclear instrumentation
application. When an external pulse is received, the integration capacitor
needs to reset within a microsecond then allow the integrator to ramp to a
repeatable voltage, based on the width of an input square pulse. If another
pulse is received prior to the completion of the integration period, the
capacitor is reset again and the integration cycle begins again.

Thanks, again, for your input, especially your insight into the capacitor
types. I'm starting to order parts to experiment with and for the final
design (hopefully) and it sounds like I should delete the polyester parts.
Is it still a problem with 5% parts. I've found polystyrenes in 2%, though,
and it sounds like that may be a better selection.

If you discharge a mylar (polyester) cap for 1 usec, some amount of
the previous charge will bounce back from dielectric absorption, very
roughly a percent or so. That could make an error that's a function of
the voltage history of the cap.

If you need extreme precision here, say in the 0.2% or better range,
it won't be trivial. A better opamp wouldn't hurt, either.

Burr-Brown used to make a dedicated analog integrator thing, I think.

John

 

[Walter Harley]

I'll try everything you're suggesting. I need to reset the integration
capacitor completely in less than 1 microsecond so I need an analog switch
with a very low "on" resistance.

What's "completely"?

You get about 69% discharge per RC time constant, right? So getting to .001
the original value is about 20 time constants, so with 10nF cap you should
only need 5 ohms. (10nF * 5 ohms * 20 constants = 1usec.) Unless
inductance, or dielectric absorption, are getting in the way. Maybe the
switch's on resistance isn't what's actually causing the problem?

I wonder, though, whether a counter mightn't be a better option for this
whole problem. E.g., take a 10MHz signal, AND it with the input pulse, and
count it. Reset the counter when needed. I'd think that would give you
better accuracy, better repeatability, and lower price.

But that probably just means I don't understand the problem you're solving.
Do you actually need the ramp, or just the voltage at the end?

 

[John Larkin]

What's "completely"?

You get about 69% discharge per RC time constant, right? So getting to .001
the original value is about 20 time constants,

Um, more like 7 taus. e^-7 is about 0.001, assuming the discharge
device is actually resistive. CMOS switches aren't, so some spare time
constants would actually be a good idea.

Yeah, the required precision is unknown to us. Measuring this won't be
easy.

John

 

[Fred Bloggs]

Does this sound like a viable approach or is there a better way?

Yeah- well since you asked- you don't want an integrator per se, you
want a time to voltage converter. You can do this by driving a ground
referenced capacitor with a precision current source, and use a
transistor to short the capacitor for reset. Buffer the capacitor
voltage for output however you like, and the advice concerning low
dielectric absorption continues to apply- the reset function is vastly
simplified.

Thanks for any constuctive input.

Sounds like you have an attitude.

 

[Spehro Pefhany]

Thanks for the feedback, John.

I'll try everything you're suggesting. I need to reset the integration
capacitor completely in less than 1 microsecond so I need an analog switch
with a very low "on" resistance. I have found the Maxim MAX4608 with a 2.5
ohm "on" resistance but, unfortunately, they don't seem to be in stock at
DigiKey. Do you have a part number for an analog switch that may work in
this instance?

Have you looked at these?
http://www.fairchildsemi.com/ds/NC/NC7SZ66.pdf

These are really amazing:
http://www.onsemi.com/pub/Collateral/NLAS4684-D.PDF


Best regards,
Spehro Pefhany

 

[John Larkin]

Yeah- well since you asked- you don't want an integrator per se, you
want a time to voltage converter. You can do this by driving a ground
referenced capacitor with a precision current source, and use a
transistor to short the capacitor for reset. Buffer the capacitor
voltage for output however you like, and the advice concerning low
dielectric absorption continues to apply- the reset function is vastly
simplified.


Sounds like you have an attitude.

Yeah, his attitude is "polite"; pretty rare around here.

John

 

[Fred Bloggs]

Yeah, his attitude is "polite"; pretty rare around here.

John

Not really- when he says thanks for "constructive" input then he is
pre-qualifying his thnx to those he deems useful. I take this to be
judgmental and offensive- Shall I continue? Lessee, works with nuclear
instrumentation...I will abstain from more colorful descriptors just
yet. But you're still a plagiarizing hack

 

[John Larkin]

These are really amazing:
http://www.onsemi.com/pub/Collateral/NLAS4684-D.PDF

What a beast! "Speaker switching"!

The charge injection (fig 10) does get pretty grim, though.

John

 

[John Larkin]

Not really- when he says thanks for "constructive" input then he is
pre-qualifying his thnx to those he deems useful.

Makes sense to me. You think maybe he should thank the people whose
chief contribution is sarcasm?

I take this to be judgmental and offensive-

Oh, go on, you say that to all the girls.

Shall I continue?

Yes! Please! Tell us more!

Lessee, works with nuclear instrumentation.

So do I; lots of money and great fun!

But you're still a plagiarizing hack

Huh? I thought most of my hacking was fairly original.

John

 

[Ben Bradley]

Hi all -

I'm working on a project that requires a stable, repeatable integration of a
35ms, 4.1VDC pulse input to produce a negative-going ramp output.

It seems like it will be more of a stairstep. It will ramp for
35mS, then stay at a constant voltage when the pulse goes away
(assuming the pulse goes to zero volts).

I've constructed an integrator using an LF347 quad, JFET input, op-amp with
a 0.01uF cap in the feedback (output to negative input) and a 1.82Mohm input
resistor into the negative input. I've also tied a 1.82Mohm resistor to
ground from the non-inverting input. The source signal is a negative-going,
TTL pulse so I ran it into the gate of an N-channel MOSFET. The source of
the MOSFET is grounded and the drain is pulled up to a regulated 4.1VDC
supply through a 100K resistor. The "on" resistance of the MOSFET is about
5 ohms (2N7000). I tapped the drain of the MOSFET into the positive input
of one stage of the LF347 set up as a voltage follower. The output of the
voltage follower drives the input resistor to the integrator. The output of
the integrator goes to another voltage follower and a 10K resistor to
ground. I'm using 1% carbon resistors and a 10% polyester capacitor.

The circuit will be used to generate multiple consecutive ramps, so an input
reset signal is received prior to the pulse associated with the ramp. This
reset will be used to enable an analog switch or a MOSFET to discharge the
capacitor in a short period prior to starting the ramp.

The circuit needs to be very linear and repeatable.

Is there something I can add to this circuit to make it more stable and
repeatable?

What will be some of the major points to be concerned with for reducing
noise?

It seems the value of those resistors for the integrator make it a
fairly high-impedance circuit which would easily pick up noise. I'd
lower their value by maybe a couple orders of magnitude, and likewise
raise the capacitor value a couple orders of magnitude.

Does this sound like a viable approach or is there a better way?

Okay, I used the 'M' word above - you could do this either analog
the way you're doing it, or with digital stuff, and each will have
different characteristics. One may be in some way 'better' than the
other for one characteristic, but the other may be better as far as
another characteristic.
So if you don't want to tell us the application, you at least need
to give real specifications. "Very linear and repeatable" isn't a
spec, it's more of a sales blurb. Give us numbers...

 

[Tony Williams]

This circuit is for timing measurements in a nuclear
instrumentation application. When an external pulse is received,
the integration capacitor needs to reset within a microsecond
then allow the integrator to ramp to a repeatable voltage, based
on the width of an input square pulse. If another pulse is
received prior to the completion of the integration period, the
capacitor is reset again and the integration cycle begins again.

The problem you describe is straightforward period
measurement, which can be reliably done with a gated
crystal clock and a binary counter. A 100KHz clock into
a 12-bit counter would have a full scale of 40.96mS.
If fed to a 12-bit DAC with your 4.1V as Vref then that
would be a ramp with 1mV steps. (Actually 1.00098mV.)

Hmmm.... 12-bits is sounding suspiciously tidy. Do you by
any chance have a following 12-bit ADC, which also uses that
4.1V as Vref?

 

[John Woodgate]

I read in sci.electronics.design that Fred Bloggs <[email protected]>

Sounds like you have an attitude.

PKB par excellence!

 

[Bill Sloman]

Not really- when he says thanks for "constructive" input then he is
pre-qualifying his thnx to those he deems useful. I take this to be
judgmental and offensive- Shall I continue? Lessee, works with nuclear
instrumentation...I will abstain from more colorful descriptors just
yet. But you're still a plagiarizing hack

Plagiarists are writers who copy other people literary efforts.
Engineers who copy successful circuits are called experts who have
done their research.

People who can't be bothered to do that sort of research are called
nit-wits, or "inventors". I get very nervous if there is a known good
solution to a problem, and I find that someone has invented a "better"
solution.

 

[Fred Bloggs]

Makes sense to me. You think maybe he should thank the people whose
chief contribution is sarcasm?

Right- you would see nothing wrong, you're another worthless yuppie snob
turning old- a gruppie I guess. You're not worth the effort it would
take to incinerate you in a back alley garbage can.

So do I; lots of money and great fun!

-Usually at enormous expense to the taxpayer. Well that honeymoon is
about to come to a very rude end. The era of government subsidization of
quite a bit of worthless high-tech boondoggles promulgated by
unemployables 1960's style is in a mach-3 tailspin to hell- fasten your
seat belt, chump.

 

[John Larkin]

-Usually at enormous expense to the taxpayer. Well that honeymoon is
about to come to a very rude end. The era of government subsidization of
quite a bit of worthless high-tech boondoggles promulgated by
unemployables 1960's style is in a mach-3 tailspin to hell- fasten your
seat belt, chump.

The seat belt I'm tightening up is for the ride on the F-35 Joint
Strike Fighter; $170 billion or so, and there will be tons of VME in
there. This bird is heavy on DSP, and is already gobbling lots of gear
for simulation, and will need more for field support.

Yee-haa, y'all.

John

 

[Spehro Pefhany]

The seat belt I'm tightening up is for the ride on the F-35 Joint
Strike Fighter; $170 billion or so, and there will be tons of VME in
there. This bird is heavy on DSP, and is already gobbling lots of gear
for simulation, and will need more for field support.

Yee-haa, y'all.
John

Heh. Good on ya!

http://www.acespilotshop.com/images/products/gifts/bumper-sticker-jet-noise-sound-freedom-b.jpg

Best regards,
Spehro Pefhany

 

[John Larkin]

Heh. Good on ya!

http://www.acespilotshop.com/images/products/gifts/bumper-sticker-jet-noise-sound-freedom-b.jpg

Best regards,
Spehro Pefhany

Cool.

Word on the streets was that VME was dead, and CPCI would rule the
bus-and-boards world in the future. Well, the telecom bubble popped,
and the aerospace/scientific people kept buying VME, so today VME
outsells CPCI by roughly 2:1. GSE (ground support equipment) is a big
market, and it's full of VME.

The big selling point for CPCI is that it would leverage regular PCI
chips and reduce costs. They forgot to mention that the lifetime of
PCI chips is typically a year or two, and people produce
instrumentation products for a decade or more.

The PCI bus is *such* a dog.

John