Comparator (op amp) Power supply drift problem

Thread Starter

Gearbreaker

Joined May 23, 2019
24
I have been designing one of my first op amp based projects and everything has come together perfectly.... except for a problem with temperature drift. I am far from an expert and have not been able to figure out a way to get around this problem.

Regarding the schematic attached, I am comparing a voltage that runs from about 5 millivolts through 550mv. I am setting a pot to trigger an op amp and am using a small signal diode to get my pot into the necessary range. The circuit is exposed to about 200 deg. F and from room temp to 200 I get a terrible voltage increase on the pot of about 18mv. I need to stabilize this and can live with a 1mv change. I have been looking into stabistors but can't see a way that they will work in this case. I like this design because of its simplicity and tidiness. Is there any way that this can be made to be thermally stable with the addition of a few components? Perhaps another diode to counteract the thermal drift? If not, I am open to suggestions on how I may accomplish this task.My Schematic.jpg
 

crutschow

Joined Mar 14, 2008
34,285
Yes, a silicon diode is a poor voltage reference as its forward voltage changes about 2 mV/°C.
Instead you could use a voltage reference shunt regulator.
So you want 1mV of stability?
An LT1634-1.25 should exceed that.
It's a 1.25V shunt reference with a 10ppm/°C maximum drift.
To get 0.65V at the pot maximum with that reference, connect an additional 200kΩ resistor in series with the top of the pot connection (below).

What is the plus supply voltage?
upload_2019-9-14_8-21-3.png
 

Thread Starter

Gearbreaker

Joined May 23, 2019
24
Thank you for the info. I'm at about 6 Volts in each direction. I could bump it up to +7 and -7 if necessary. I would think that the LT1634 has a sweet spot on its knee.
 

crutschow

Joined Mar 14, 2008
34,285
I would think that the LT1634 has a sweet spot on its knee.
Not really.
It's not like a Zener.
You could increase the value of the 10kΩ resistor if you like, and operate it at a lower current (say below a mA).
If it operates at more than 1mA you need to add a 100nF capacitor across it to insure stability.
upload_2019-9-14_12-3-51.png
 
Last edited:

Thread Starter

Gearbreaker

Joined May 23, 2019
24
Ah, I am getting it. Been reading up on how these things work. Lots more complicated inside than a zener. I would like to run it at as low current as I can within reason. I have limited power supply current available. Not sure if I need the 10 or 25 PPM/C version but I am thinking I should go with the 10 because the more accurate the better. Also, I have read that the temperature stability is only guaranteed up to about 185 deg F (85 C) This is a little troublesome but I am hopeful that the part will hang in there at the actual 200 deg F operating temperature.
 

crutschow

Joined Mar 14, 2008
34,285
I would like to run it at as low current as I can within reason.
The device can operate down to 10μA.
The pot circuit will draw 1.25v/400k = 3.125μA.
So allowing 20uA total gives a value for R2 (my schematic) of (6v-1.25V/20μA = 230kΩ
I have read that the temperature stability is only guaranteed up to about 185 deg F (85 C)
Good reason to get the 10ppm device, since it's likely to stay more stable above that temperature.
 

TeeKay6

Joined Apr 20, 2019
573
The device can operate down to 10μA.
The pot circuit will draw 1.25v/400k = 3.125μA.
So allowing 20uA total gives a value for R2 (my schematic) of (6v-1.25V/20μA = 230kΩ
Good reason to get the 10ppm device, since it's likely to stay more stable above that temperature.
I have been designing one of my first op amp based projects and everything has come together perfectly.... except for a problem with temperature drift. I am far from an expert and have not been able to figure out a way to get around this problem.

Regarding the schematic attached, I am comparing a voltage that runs from about 5 millivolts through 550mv. I am setting a pot to trigger an op amp and am using a small signal diode to get my pot into the necessary range. The circuit is exposed to about 200 deg. F and from room temp to 200 I get a terrible voltage increase on the pot of about 18mv. I need to stabilize this and can live with a 1mv change. I have been looking into stabistors but can't see a way that they will work in this case. I like this design because of its simplicity and tidiness. Is there any way that this can be made to be thermally stable with the addition of a few components? Perhaps another diode to counteract the thermal drift? If not, I am open to suggestions on how I may accomplish this task.View attachment 186165
@Gearbreaker
As you discuss with @crutschow, the 1N914 is a poor choice as a voltage reference.

What opamp are you using? Many opamps would be unhappy at 93C. Input bias current and offset current could be high. Input offset current is not compensated for; the source resistance for the inv input is (assumed) 4.7K, and for the noninv input it varies from 1K to 51K over the range of the pot. Any load driven by the opamp could contribute additional heat to the IC junctions.

How much does the +/- power supply change when going to 93C? The op amp does not have infinite rejection of PS variation.

Is the potentiometer rated for 93C operation (esp any rubber/plastic parts)? What is the tempco of the pot resistance (both as a pot and as a rheostat)?

The effective time constant of your RC filter is likely quite different at 93C than at 25C; tempco's of both resistor and cap could be very significant.

In your post#1 you state that the pot voltage increases as the temperature rises; that is contrary to the behavior of an ordinary silicon diode the voltage of which generally decreases about 2mv/°C for each 1°C temperature rise. That makes me suspicious that the diode forward voltage is not the only parameter that is changing as the temp rises.
 

Thread Starter

Gearbreaker

Joined May 23, 2019
24
TeeKay6, Interesting points. I knew that the 914 would drift but didn't realize, as is actually the case, that it would be so wildly temperature sensitive. I was hoping I would get lucky and if the voltage were to vary with temperature that it would offset another problem area if there were one. No joy here though.

Regarding the voltage going the wrong direction at temperature, I will have to go back and verify it during further testing.

The final two op amps pictured are in an LM158GJB. From what I understand it is a standard but can be easily changed to another that could be better suited if necessary. (Again, I am very new to op amps). Upstream I basically have an INA105 as an inverting buffer. Before that I have an INA106 set up as a 10X inverter. I chose these parts because of their temperature stability. They measure parameters that are fed into the buffer in the LM158. The LM158's final stage is feeding 5ma into an optocoupler. The optocoupler is a Triac unit so I don't need any hysteresis. Once the triac trips all bets are off and it latches until manually reset. So far the first two stages, not visible in the pic are an INA105 and INA106. They seem perfectly happy and the voltage which the LM158 flips at was seeming very accurately at temperature in preliminary testing as well.

Once I saw the apparent set point voltage going crazy (right on the 158's input set pin) I stopped further testing so I don't know what happened with the supply voltages. This week I will set the test subject back up and take it back up to temperature while monitoring the PS. I will research the specs on the pot this week as well. It is a 15 turn WW so I never gave it a second thought.

I understand what you are saying about the RC filter varying. I am using a tantalum cap and a metal film resistor. Further testing on that can't really happen with that until the set point voltage is stabilized. However, your point is well taken and it certainly does need to be tested at temperature.

All of this will happen while I await some LT1634's this week. I am hopeful that Crutschow's suggestion will make things work okay. I really need to get that voltage under control before any serious testing can be done. This certainly is a learning experience.
 

TeeKay6

Joined Apr 20, 2019
573
TeeKay6, Interesting points. I knew that the 914 would drift but didn't realize, as is actually the case, that it would be so wildly temperature sensitive. I was hoping I would get lucky and if the voltage were to vary with temperature that it would offset another problem area if there were one. No joy here though.

Regarding the voltage going the wrong direction at temperature, I will have to go back and verify it during further testing.

The final two op amps pictured are in an LM158GJB. From what I understand it is a standard but can be easily changed to another that could be better suited if necessary. (Again, I am very new to op amps). Upstream I basically have an INA105 as an inverting buffer. Before that I have an INA106 set up as a 10X inverter. I chose these parts because of their temperature stability. They measure parameters that are fed into the buffer in the LM158. The LM158's final stage is feeding 5ma into an optocoupler. The optocoupler is a Triac unit so I don't need any hysteresis. Once the triac trips all bets are off and it latches until manually reset. So far the first two stages, not visible in the pic are an INA105 and INA106. They seem perfectly happy and the voltage which the LM158 flips at was seeming very accurately at temperature in preliminary testing as well.

Once I saw the apparent set point voltage going crazy (right on the 158's input set pin) I stopped further testing so I don't know what happened with the supply voltages. This week I will set the test subject back up and take it back up to temperature while monitoring the PS. I will research the specs on the pot this week as well. It is a 15 turn WW so I never gave it a second thought.

I understand what you are saying about the RC filter varying. I am using a tantalum cap and a metal film resistor. Further testing on that can't really happen with that until the set point voltage is stabilized. However, your point is well taken and it certainly does need to be tested at temperature.

All of this will happen while I await some LT1634's this week. I am hopeful that Crutschow's suggestion will make things work okay. I really need to get that voltage under control before any serious testing can be done. This certainly is a learning experience.
@Gearbreaker
In electronic design, hope gets you nowhere. Electronic components don't care what you think or hope for. :)

All silicon diodes & rectifiers have a similar temperature coefficient for forward voltage. This has the great advantage of allowing the voltage change of one diode to at least partially compensate the change in another diode (or silicon transistor; the base-emitter junction is essentially a silicon diode). Hot carrier/Schottky diodes, GaN, SiC, GaAs diodes are different from ordinary silicon diodes and each material has its own characteristic. None are suitable as voltage references except in a crude sense.

The LM158 (and LM158A) is a very old opamp design that has nevertheless withstood the test of time and remains extremely popular. It was among the very first opamps to have a temperature compensated input bias current. Unfortunately, LM158 datasheets say only that the bias current is "essentially constant" but give no graphs of change vs temperature--a definite shortcoming in the specs. The LM158 is clearly spec'd to allow operation at 93°C and above. Before trying another model opamp, I suggest that the proper course is to understand what shortcoming, if any, of the LM158 is actually causing a problem in your application.

From your statements, I understand that the INA105 & INA106 circuitry is providing a stable signal vs temperature such that your accepting a comparator threshold change of 1mV is appropriate. That is, a change of 1mV at the comparison stage would represent a small fraction of the allowed change in whatever parameter is being conditioned by the preceding INA105/106 stages.

My reference to power supply and pot stability, and to heating due to opamp load, are primarily to be inclusive of all interferences. They could be problems, but would not be the first place I looked. It is certainly unlikely that the tempco of a ww pot would be the cause of much variation (although stability of the wiper setting vs temperature is another question).

As for the RC filter. 4.7K is not a common metal film value; they generally hew to the 1% values: ...
4.53 4.64 4.75 4.87 4.99... Thus I was suspicious that it was a carbon resistor...with a terrible tempco. For the tantalum cap behavior you must carefully review the datasheet for the specific device & manufacturer you are using.

You say "...your point is well taken and it certainly does need to be tested at temperature." Please be aware that when a spec says that a parameter can vary X amount from piece to piece, that is exactly what is meant. You cannot rationally expect that devices that you buy at one time, or that were packaged at one time, etc. will perform any more closely unit-to-unit than is given in the specs! That is, if you build two units, you should not expect operation between the two to be closer than is shown in the datasheet. Also, while typical values shown in datasheets are useful, it is only the min/max specs that are guaranteed by the manufacturer.

When looking for causes, make as few changes as possible between tests or the results will simply be confounding. Good luck.
 

Thread Starter

Gearbreaker

Joined May 23, 2019
24
@Gearbreaker
In electronic design, hope gets you nowhere. Electronic components don't care what you think or hope for. :)

All silicon diodes & rectifiers have a similar temperature coefficient for forward voltage. This has the great advantage of allowing the voltage change of one diode to at least partially compensate the change in another diode (or silicon transistor; the base-emitter junction is essentially a silicon diode). Hot carrier/Schottky diodes, GaN, SiC, GaAs diodes are different from ordinary silicon diodes and each material has its own characteristic. None are suitable as voltage references except in a crude sense.

The LM158 (and LM158A) is a very old opamp design that has nevertheless withstood the test of time and remains extremely popular. It was among the very first opamps to have a temperature compensated input bias current. Unfortunately, LM158 datasheets say only that the bias current is "essentially constant" but give no graphs of change vs temperature--a definite shortcoming in the specs. The LM158 is clearly spec'd to allow operation at 93°C and above. Before trying another model opamp, I suggest that the proper course is to understand what shortcoming, if any, of the LM158 is actually causing a problem in your application.

From your statements, I understand that the INA105 & INA106 circuitry is providing a stable signal vs temperature such that your accepting a comparator threshold change of 1mV is appropriate. That is, a change of 1mV at the comparison stage would represent a small fraction of the allowed change in whatever parameter is being conditioned by the preceding INA105/106 stages.

My reference to power supply and pot stability, and to heating due to opamp load, are primarily to be inclusive of all interferences. They could be problems, but would not be the first place I looked. It is certainly unlikely that the tempco of a ww pot would be the cause of much variation (although stability of the wiper setting vs temperature is another question).

As for the RC filter. 4.7K is not a common metal film value; they generally hew to the 1% values: ...
4.53 4.64 4.75 4.87 4.99... Thus I was suspicious that it was a carbon resistor...with a terrible tempco. For the tantalum cap behavior you must carefully review the datasheet for the specific device & manufacturer you are using.

You say "...your point is well taken and it certainly does need to be tested at temperature." Please be aware that when a spec says that a parameter can vary X amount from piece to piece, that is exactly what is meant. You cannot rationally expect that devices that you buy at one time, or that were packaged at one time, etc. will perform any more closely unit-to-unit than is given in the specs! That is, if you build two units, you should not expect operation between the two to be closer than is shown in the datasheet. Also, while typical values shown in datasheets are useful, it is only the min/max specs that are guaranteed by the manufacturer.

When looking for causes, make as few changes as possible between tests or the results will simply be confounding. Good luck.
Thank you again. I will be sure to stop back here and post my findings as they develop. Perhaps this thread will help someone else someday.
 

TeeKay6

Joined Apr 20, 2019
573
Thank you again. I will be sure to stop back here and post my findings as they develop. Perhaps this thread will help someone else someday.
@Gearbreaker
You say "The LM158's final stage is feeding 5ma into an optocoupler. The optocoupler is a Triac unit so I don't need any hysteresis. Once the triac trips all bets are off and it latches until manually reset." Since you have not provided the schematic for the opamp load (opto coupler) and related circuitry, we can only assume that you have measured that the opto coupler LED current is indeed (limited to) 5mA (and that that current is an appropriate value for your use of the coupler) and that the Triac remains "on" once triggered (that would not be true if it were passing AC current unless the gate current is maintained above the trigger level from cycle to cycle).

While @crutschow 's advice is good and certainly worth implementing, a common multimeter could detect most of the problems we have noted. For example, the reference diode forward voltage vs temp is easily large enough to see on even low cost multimeters, as are the + and - PS voltage changes. As I am sure you know, you cannot depend on components specs alone; you must verify operation with measurements.
 

Thread Starter

Gearbreaker

Joined May 23, 2019
24
Not really.
It's not like a Zener.
You could increase the value of the 10kΩ resistor if you like, and operate it at a lower current (say below a mA).
If it operates at more than 1mA you need to add a 100nF capacitor across it to insure stability.
View attachment 186182
I finally got the time to put the LT1634-1.25 in and take it up in temperature. I was amazed at the stability. From ambient to 235F I got right around a -1mv deviation. I have left the resistor configuration as pictured above. I have yet to put in the resistor divider to get my range back to around 0-600mv. The deviation is about in my spec but I just can't leave well enough alone. I am hopeful that I can place a carbon film resistor in the aforementioned divider that will help offset the temperature coefficient of the regulator. Yes, it is doubtful but it may be worth a try.

The resistor and pot behaved well too. My deviation on the regulator matched the deviation on the pots output when taken up to temperature. However, this was at around a 75 mv pot output setting. I have yet to set the pot to the lower range around 8mv. but will on the next go-round. I would think that the error should be less than on the higher pot settings. I haven't yet fully deciphered the LT1634-1.25 datasheet but do I understand that it will have a better temperature coefficient if it isn't driven as hard? Thank you again. Your suggestion has made the project a reality again.
 

Thread Starter

Gearbreaker

Joined May 23, 2019
24
@Gearbreaker
You say "The LM158's final stage is feeding 5ma into an optocoupler. The optocoupler is a Triac unit so I don't need any hysteresis. Once the triac trips all bets are off and it latches until manually reset." Since you have not provided the schematic for the opamp load (opto coupler) and related circuitry, we can only assume that you have measured that the opto coupler LED current is indeed (limited to) 5mA (and that that current is an appropriate value for your use of the coupler) and that the Triac remains "on" once triggered (that would not be true if it were passing AC current unless the gate current is maintained above the trigger level from cycle to cycle).

While @crutschow 's advice is good and certainly worth implementing, a common multimeter could detect most of the problems we have noted. For example, the reference diode forward voltage vs temp is easily large enough to see on even low cost multimeters, as are the + and - PS voltage changes. As I am sure you know, you cannot depend on components specs alone; you must verify operation with measurements.
 

Thread Starter

Gearbreaker

Joined May 23, 2019
24
You say "The LM158's final stage is feeding 5ma into an optocoupler. The optocoupler is a Triac unit so I don't need any hysteresis. Once the triac trips all bets are off and it latches until manually reset." Since you have not provided the schematic for the opamp load (opto coupler) and related circuitry, we can only assume that you have measured that the opto coupler LED current is indeed (limited to) 5mA (and that that current is an appropriate value for your use of the coupler) and that the Triac remains "on" once triggered (that would not be true if it were passing AC current unless the gate current is maintained above the trigger level from cycle to cycle).


Regarding the optocoupler, yes I have chosen a part that will trip at 5ma. However, I will consider increasing the source to about 7 or 8ma once the final project is completed and I can then measure the current draw. Currently, ;) I am limited to a 10ma power source. A possibility is the replacement of the power transformer with a 30ma unit. (The larger transformers size would be too large for the area allocated on the board I am building). The optocouplers spec is a guaranteed trip at 5ma but as you had said in an earlier post, not all components with the same part number behave exactly the same, especially the current limiting resistor. I wouldn't want to design something on the edge of the spec. Yes, the current passing through the triac side is DC. It is simpler and less power hungry than setting up a latching circuit. So far I am at about 7ma overall draw. Keeping my fingers crossed that I can keep it there.

As I am sure you know, you cannot depend on components specs alone; you must verify operation with measurements.

Yes, I am very old school in that regard. I don't trust any modeling program and have had very good results by designing on a proto board with meters and a scope. I do have a voltage standard and was extremely surprised that the accuracy and linearity of the "free" Harbor Freight multimeters. Although I also have quality meters, they are great for monitoring various voltage points in actual operation. I am currently ;) set up to monitor five voltages at a time, something that would have been nearly cost preventive without the little cheap red meters.
 
Top