Op Amp Selection For Analog Computing

[rstofer]

There must be thousands of op amps to choose from and I'm not quite sure where to begin looking...

My application is conceptually simple:  I want to build precision integrators for analog computing.  High speed is not a consideration!

Here are some of the parameters:  For the most part I want to run in real time so the integration time constant is 1 second.  There will be the usual 10x inputs as well.  I have been using 1 ufd capacitors and 1 M input resistors (100k for 10x).  Due to the resistor value, I think I need to worry about input bias current.  In a perfect world, I wouldn't need trimmer capacitors to balance the input offset voltages.  The op amps will work on +- 15V supplies and the signal swing will be from -10V to +10V.

I have been using TL081 Op Amps and they have been satisfactory even without offset adjustment but if there is a better choice, I want to use it.  Cost isn't a really big issue as I only need a dozen as long as the price isn't completely insane.  In a perfect world, the device would be available in a DIP package to allow replacement after an "oopsie!".

Second topic:  Traditionally, timing capacitors are switched to provide a 100x speedup and this was often done using a relay.  I get it, no injection current.  It would be much nicer if there was an analog switch suitable for switching the capacitors without mucking up the input current.  Any thoughts?  A similar device could also be used to apply the initial condition voltage which basically just charges the capacitor to some initial voltage.  SPDT is handy since it is common to disconnect the op amp inputs and ground the summing junction while making potentiometer adjustments.

I have been kicking around the idea of building a second analog computer with, perhaps, 12 amplifiers, any of which can be used as an inverter, summer or integrator by switching components in and out of the circuit.  That seems kind of handy because it allows a left to right program without patching all over the place to find an inverter.  It would be very cool to do all the switching with analog switches and even better to control the switches with SPI (IO Expander).  Relays seem kind of clunky...

I have been thinking about this project for quite a while.  It seems like a good time to get started.

[BrianHG]

For the analog switch, your 20v range, 30v supply...  You might be best off with DMOS switch, or JFet/Mosfet.  However, have you considered an optocoupler with an AC Mosfet output.  You get a few 100Meg ohm when off, around 10 ohm down to 1 ohm depending on type, when on.

For the OPAmp, you might like the low power, low offset precision of these:
http://www.ti.com/product/opa4192

I'm sorry about not having DIP packages, I would recommend those small cheap SMD-DIP PCBs you can get in bulk.  SOIC and 5 pin SOT-23 (Stay away from TSSOP) are easy enough to weld and remove from the PCBs, ok, not as easy as having the DIP in the first place...

[BrianHG]

These are a little pricey as switches, but they go to 2.5 ohms when on:
https://www.maximintegrated.com/en/products/analog/analog-switches-multiplexers/MAX4669.html

Unlike the optocoupler, your getting at least 2 switches per package and the logic control input is CMOS, so it wont consume current like the LED in the optocoupler.

They still cost less than 2 reed relays, or 2 optocouplers which 1 would replace...
And they come in DIP.

[rstofer]

For the analog switch, your 20v range, 30v supply...  You might be best off with DMOS switch, or JFet/Mosfet.  However, have you considered an optocoupler with an AC Mosfet output.  You get a few 100Meg ohm when off, around 10 ohm down to 1 ohm depending on type, when on.

I hadn't considered that but I'll look into it.  The output has to be perfectly linear over the +- 10V signal range but that may not be an issue.

For the OPAmp, you might like the low power, low offset precision of these:
http://www.ti.com/product/opa4192

I had seen that op amp when I went searching; among others, of course.  I like the idea of using a quad device but I'm not sure of the wisdom.  I could get 4 amplifiers, 8 timing capacitors and the associated switches on one PCB and, as it turns out, ExpressPCB makes my boards in groups of 3.  So, 3 boards, 12 amplifiers, sounds good to me!

Twelve amplifiers is probably overkill.  Equations of motion may only need two integrators (acceleration->velocity->displacement) plus summers and inverters.  I don't see many problems using higher order integration but with a dozen amplifiers, I should be set for a lot of problems.

My current computer has 2 integrators, 2 summers and 1 inverter.  It has been just sufficient for problems like damped harmonic motion, pendulum motion, simple predator-prey and similar equations.  Moving up to 12 amplifiers more than doubles my capability.

And, yes, Matlab with the Simulink plug-in does a magnificent job of analog computing.  But it isn't hardware!

[BrianHG]

I hadn't considered that but I'll look into it.  The output has to be perfectly linear over the +- 10V signal range but that may not be an issue.

Analog MOSFET/DMOS based switches act as a strict open or closed bidirectional series resistor.  With the optocoupler, the output impedance may sway with the amount of current you are feeding the LED, but, it wont skew switch side A-B voltages.  As for the Maxim part, it will have a better fixed 2.5 ohm on resistance, however, if either side, A or B exceeds the input +&- supply range, they are diode protected and will clamp.  This should not be a concern with your +/- 10v swing and +/-15v supply.

As for analog computing done the real way, take a look at GK's threads, here is his latest:
https://www.eevblog.com/forum/projects/oscilloscope-pong-for-1-or-2-players/

[rstofer]

As for analog computing done the real way, take a look at GK's threads, here is his latest:
https://www.eevblog.com/forum/projects/oscilloscope-pong-for-1-or-2-players/

Thanks for the link, that's a great project.  And, no, I'm not going to build my own op amps!
There's another thread around here where a fellow is building an old-school 100V analog computer.
https://www.eevblog.com/forum/projects/home-brew-analog-computer-system/

Both of these projects are well beyond my level...

My interests are modest:  My grandson will be taking Differential Equations in the Spring next year.  I would like to be able to demonstrate many of the problems.  One of the big stumbling blocks when I took the course (prior to '73) was not being able to 'play' with the equations.  I like to touch and feel and, frankly, the analytic solutions weren't satisfying.  Especially since we had to hand plot the output!

We can talk about damped harmonic motion all we want, describing the cosine inside the e^-(gamma*t) but there is nothing so impressive as seeing it in real life with variable parameters.  Under-damped, critical-damped and over-damped are easy to simulate with an analog computer.  Yes, it's probably overkill...

[David Hess]

I often prefer using dual instead of quad operational amplifiers because of easier layout.

The TL051/2/4 is the old precision upgrade for the TL071 and TL081 series.  The TL031/2/4 is the lower power alternative to the TL051/2/4.  The TL032 and TL052 are available inexpensively.

I definitely think you want to stay with FET input operational amplifiers for low input bias current although there *are* suitable low input current precision bipolar parts like the descendants of the LM308.  Modern single examples include the LT1008, LT1012, and LT1097.  Duals and quads include the LT1112/14 and LT6011/12.

With an unlimited budget, I would use the LT1012 or LT1097 everywhere because they support overcompensation.

[rstofer]

With an unlimited budget, I would use the LT1012 or LT1097 everywhere because they support overcompensation.

The LT1012 is definitely within budget and the LT1097 is even cheaper.
I'm not hung up on single versus dual versus quad.  I haven't even begun to think about PCB layout.  There are certain advantages to 8 pin DIP whether in single or dual configuration.

What I want to do is pick some analog switches and op amps and breadboard a full circuit.  Then I want to run it and verify the nature of the integrations.  I'll probably use in my Analog Discovery.  14 bits of ADC seems like a lot...

Since I can't buy capacitors better than 1%, I am trying to keep the integration within, say, 2%.

[JXL]

Second topic:  Traditionally, timing capacitors are switched to provide a 100x speedup and this was often done using a relay.

Why not switch the resistor(s) for the 100x speedup.  In the integrators I built, I used 100-mil jumpers to short out resistors to get 1x, 10x, and 100x "speedup".  Analog muxes may work as well.

Since I can't buy capacitors better than 1%, I am trying to keep the integration within, say, 2%.

I would pick any available capacitors which are stable over voltage and tune the integrator ramps with 0.1% resistors.  You can even use trim pots for maximum flexibility.

BTW, I use LF353 and TL082 dual opamps for my circuits.  They are cheap so I buy a bunch and screen them for lowest bias current and offset voltage to use in precision circuits.  The "bad" ones are used for non-critical applications.

[rstofer]

Second topic:  Traditionally, timing capacitors are switched to provide a 100x speedup and this was often done using a relay.

Why not switch the resistor(s) for the 100x speedup.  In the integrators I built, I used 100-mil jumpers to short out resistors to get 1x, 10x, and 100x "speedup".  Analog muxes may work as well.

I would pick any available capacitors which are stable over voltage and tune the integrator ramps with 0.1% resistors.  You can even use trim pots for maximum flexibility.

BTW, I use LF353 and TL082 dual opamps for my circuits.  They are cheap so I buy a bunch and screen them for lowest bias current and offset voltage to use in precision circuits.  The "bad" ones are used for non-critical applications.

I guess the only reason for switching the capacitors is because it is traditional.  i haven't really thought thru why capacitors are switched but there must be some reason, the high dollar precision caps of yesteryear were a lot more costly than precision resistors yet most designs I have seen used two or three capacitors.

The designer of my existing AC used two resistors in series at each input to allow for the possibility of trimming the time interval.  I just used the nominal values and everything came out ok but there was always the possibility of tweaking things a bit.  I need to remember this idea when I lay out the PCB.  I'm also thinking about using a jumper to select between +15 or -15 for the offset correction pot.  Different devices have different ideas.

My existing machine is based on the work of Dr. F Vogel.
http://blog.analogmachine.org/wp-content/uploads/2012/03/VogelAnalogComputer1.png

A photo of my front panel is attached

[David Hess]

With an unlimited budget, I would use the LT1012 or LT1097 everywhere because they support overcompensation.

The LT1012 is definitely within budget and the LT1097 is even cheaper.
I'm not hung up on single versus dual versus quad.  I haven't even begun to think about PCB layout.  There are certain advantages to 8 pin DIP whether in single or dual configuration.

The TL051/2 and TL031/2 are obvious and inexpensive upgrades for more precision to the TL081/2/4 you have been using.  Beyond that whether to use a low input bias current bipolar like the LT1012, TL1097, LT1112/14, or LT6011/12 or a more modern +/-15 volt precision JFET or MOSFET input operational amplifier will depend on error analysis.  The bipolar ones have better precision at the cost of slightly higher bias current.

What I want to do is pick some analog switches and op amps and breadboard a full circuit.  Then I want to run it and verify the nature of the integrations.  I'll probably use in my Analog Discovery.  14 bits of ADC seems like a lot...

I like discrete JFETs but there are some viable +/-15 volt analog switches like the DG411 series from Maxim and ADG411 series from Analog Devices.  Not all are suitable because they many higher leakage than even low input bias current bipolar operational amplifiers so watch out for that.

If you want to go for lower bias current operational amplifiers, then lower leakage analog switches need to be used also.

Since I can't buy capacitors better than 1%, I am trying to keep the integration within, say, 2%.

Tektronix used home made oil filled film capacitors for precision applications and graded them into ratio matched sets.  I think you could do the same thing and then trim the integration constant out through a gain adjustment.

[David Hess]

Second topic:  Traditionally, timing capacitors are switched to provide a 100x speedup and this was often done using a relay.

Why not switch the resistor(s) for the 100x speedup.  In the integrators I built, I used 100-mil jumpers to short out resistors to get 1x, 10x, and 100x "speedup".  Analog muxes may work as well.

I think the reason for this is that if the capacitors are switched on the output side of the operational amplifier, then the output signal can be taken from the capacitor and errors from the analog switch are removed.  That is not such a big deal now that we have much better and lower resistance analog switches but it mattered in the past.

[Zero999]

I have been using TL081 Op Amps and they have been satisfactory even without offset adjustment but if there is a better choice, I want to use it.  Cost isn't a really big issue as I only need a dozen as long as the price isn't completely insane.  In a perfect world, the device would be available in a DIP package to allow replacement after an "oopsie!".

If you've found the TL081 adequate so far, then try the OP07. It's fairly cheap and common, so if you blow one up, it's easy to replace.

Second topic:  Traditionally, timing capacitors are switched to provide a 100x speedup and this was often done using a relay.  I get it, no injection current.  It would be much nicer if there was an analog switch suitable for switching the capacitors without mucking up the input current.  Any thoughts?  A similar device could also be used to apply the initial condition voltage which basically just charges the capacitor to some initial voltage.  SPDT is handy since it is common to disconnect the op amp inputs and ground the summing junction while making potentiometer adjustments.

How about the DG419, DG303 or MAX312? SPDT or two NC and two NO switches in a package. Low on resistance. The downside is they're expensive.
http://www.e2v.com/shared/content/resources/File/documents/QPSemi/DG417-DG418-DG419_datasheet.pdf
https://datasheets.maximintegrated.com/en/ds/MAX312-MAX314.pdf
http://www.e2v-us.com/shared/content/resources/File/documents/QPSemi/DG300A-DG301A_datasheet.pdf

[rstofer]

How about the DG419, DG303 or MAX312? SPDT or two NC and two NO switches in a package. Low on resistance. The downside is they're expensive.

The price is not out of range.  I don't need all that many.  There are some applications where the MAX314 (dual SPDT) would work really well.  One example is grounding the summing junction while disconnecting the connection to the op amp.  SPDT type of thing.  Initial conditions and capacitor selection can probably use the quad SPDT NO.

We aren't even up to $25/amplifier yet so I'm not worried about cost.  Functionality is what I'm really after.