Difference in resistance of in dividual "steps" of digital pot

A digital pot was never intended to divide 5V down to 2.5 volts at 1% accuracy. In this case there are only 64 steps. About 78mV/step.

I have used digital pots to get the 2.5V at 1% another way.
Assume the 5V is +/-5%. (4.75 to 5.25V)
Using a divider set the top end of the pot to 2.7V and the bottom end to 2.3V. We don't know if he 5V is high or low but either way 2.5V is somewhere with the travel of the pot. Now we have 64 steps from 2.3 to 2.7V. About 6mV/step. There will be about 8 steps from -1% to +1% of the 2.5V. With a step of 6mV who cares if one step is 6mV and another is 7mV and another is 5mV you only need to get with in 25mV.

This is the same game we do with mechanical pots. To set a pot to 1% is not easy or repeatable. Most post have a end to end 20% or 30% and a linearity of 5%. But with the pot spanning from-7% to +7% it should be easy to hit with in 1%.

Like ronsimpson said, you guys should read-up on DACs and how to use and how they are spec'ed. After all, DCPs are not much different. The DCPs are guaranteed to be monotonic (meaning no dead spots, and no values which will make them deviate from a linear asymptote). The end-to-end values are spec'ed, tested, and also guaranteed. All the specs are indicating is that there are some variances when the material is processed which will lead to some uncertainty. If you think the uncertainty on the DCPs is much worse than the one on regular potentiometers then it would be nice if you show the actual test data to back that up. If you set a DCP to code 128 you will always get the same value, which may be a bit off from the theoretical value, but it will always be the same; try to do the same with a regular potentiometer. Even with the fanciest linear potentiometers, you may get "linear" behavior, but do you always get the same value at the same spot regardless. DCPs are not the panacea, but for some things (like repeatability) they will run circles around the "old" regular potentiometers (read "analog" pots); for other things (like power handling, very low distortion) the DCPs may not be the best choice.

Some good information from Microchip and Analog Devices:
Comparing Digital Potentiometers to Mechanical Potentiometers
**broken link removed**

I'm sure you can get as many of these you want from the same batch. So I still would test one any one would do that way you can see how its going to work in your hardware. The like in post 21 you just add a little to the pot like maybe a real pot to trim it to where you need that honey spot.

thanks, the link on post #23 from AD doesn't show how to calculate the accuracy for a digi pot used in pot mode.

That is,
We have a 1V2 reference voltage , which we wish to divide down ( 0 to 1v2) with a digi pot (eg mcp4013 2k1 version)
What would be the accuracy of the divider voltage.?

This is an absolutely basic and essential question, but none of the vendors tell how to assess this accuracy.

Maybe the following link can help: AN691 - Optimizing the Digital Potentiometer in Precision Circuits.

Looking at your AN691, it appears that the potentiometer is extremely accurate, it refers to the table of fig , page 2 but does not say how these accuracies were calculated....also, it is not shown how to calculate accuracy for the MCP4013 in potentiometer mode.

For example, if I had gone up 30 of the 64 bits with the wiper of the 2k1 version of MCP4013, and my reference voltage was 1v, what would be the accuracy of the divider voltage, what would be its lowest possible minimum, and highest possible maximum?.... the datasheet just leaves us for dead in relation to calculating these essential things.

Not sure what you are really trying to achieve with all these questions; but all the answers are really on the datasheet. You just have to read it. You have all the necessary elements to add to an Excel spreadsheet and go all out on "what-ifs". I have never seen such information available for an analog potentiometer; so I am not sure what are you even trying to compare to. If you need better "steps" you get one with better resolution. It is really that simple.

I just whipped the info below really quick in Excel, so there may be some slight errors. But I used the AC/DC Table, Fig 2.5, Fig 4-1, Eq 4-1, and Eq-4-2.

For the 2.1k device, at +5V, in rheostat mode;
Rwb should be 40_ohms +/- 8.3_ohms for code=00h => 20.8% variance
Rwb should be 1054.3_ohms +/- 8.3_ohms for code=1Eh => 0.8% variance
Rwb should be 2170_ohms +/- 8.3_ohms for code=3Fh => 0.4% variance

For the 2.1k device, at +5V, in potentiometer mode mode the wiper resistor goes out of the equation (assuming you are feeding a high impedance circuit - which would make sense if you are using it in potentiometer mode);
Rb/(Ra+Rb) ratio should be 0.000 +/- 0.004 for code=00h
Rb/(Ra+Rb) ratio should be 0.476 +/- 0.004 for code=1Eh
Rb/(Ra+Rb) ratio should be 1.000 +/- 0.004 for code=3Fh

Good luck with that. If you still think you don't have the necessary info, then my humble suggestion is to go back an use the analog potentiometer; the DCPs are not for you - either because of the application, or because the lack of understanding (either way you probably have spent more time on this than what it probably takes you to instrument it and let the data speak for itself).

languer, you are taking the typical value of +/-0.25LSB, but I think you should rather use the +/-0.5 LSB worst case.
And that is just the linearity at a stable 25°C. Then there is the temperature coefficient, so you need to account for the worst case at both temperature extremes again, and then there is the wiper resistance range (plus the tempco).
Add the two uncertainty extremes, multiply by the tempco at -40°C and +125°C respectively to get even worse values, and you get the total possible error.

Flyback said:

thanks, the link on post #23 from AD doesn't show how to calculate the accuracy for a digi pot used in pot mode.

That is,
We have a 1V2 reference voltage , which we wish to divide down ( 0 to 1v2) with a digi pot (eg mcp4013 2k1 version)
What would be the accuracy of the divider voltage.?

This is an absolutely basic and essential question, but none of the vendors tell how to assess this accuracy.

Click to expand...

Do you want to output an accurate voltage divided from a reference voltage?
Maybe a multiplying DAC is the solution.

Fig 2.5 does account for temperature. But yes, one could worst case it even more. But I don't need to use the part, Flyback does; so he can do as many analyses as he wants or none at all. Apparently the "trying to teach someone to fish" around these forum turns into "please do this for me". I am all out of free fish.

As a side note we use similar parts from ADI on our designs,; and yes we do have some decent models that we use to account for many things. The temperature variation is really amongst the lowest of our concerns (our design takes care of any such small variations).

Thankyou Languer, though I should point out that your analysis is at odds to what others here have calculated here .
Fig 6-5 shows how badly things could go astray.
Again, using the MCP4013 in pot mode, (2k1 version)
Supposing I am using 32 steps (ie half way)
One step is 2100/64 = 32.8 ohms
…but there is an error in any step of 0.25LSB….so this could be 24.6 ohms or 41 ohms
Then 32 steps could be 32 * 24.6 = 788 ohms
Or
32 steps could be 32 * 41 = 1313 ohms
The 788 ohms gives a divider voltage of [788/(2100)] * 1.2 = 0.45V
The 1313 ohms gives a divider voltage of [1313/2100] * 1.2 = 0.75V
….as you can see, the inaccuracy here is dreadful.
I cannot understand how you have calculated it to be so accurate


Also, we don't wish to use a dac...we wish only to use a digi pot with an "up or down" protocol.

Flyback, I think you are assuming that the 0.25_LSB errors are cumulative. I may be mistaken, but I am pretty sure they are not.

This probably will not help but:

Back in the dark ages when I built radio stations; to get a really good volume controls. They are 100 position rotary switches with 100 resistors. Each resistor was +/-10%. There are used in cases where the control is used every minute 24/7.
**broken link removed**

I think you are assuming that the 0.25_LSB errors are cumulative.

Click to expand...

Lets assume that all resistors are 10% low. (This is very likely in the case of the digital pot.) The end to end resistance will be 10% low. The center point will be exactly center. Each step will be exactly right. Assume all are 10% high = center is center!

A bad case is where half of the resistors are 10% high and half are 10% low. (probably not the case) End to end is right!. In pot mode some of the resistors will have 10% too much or too little voltage across them. So a step will have three sizes. (big, little and just right)

From experiance; it is common to find the end to end resistance to be high or low. (common in ICs)
You will never find resistor 32 to be +max error and resistor 33 to be -max error.
These resistors are made on the same silicon at the same time and should be close to their neighbor.

thanks, certainly page 5 of the following backs up what you are saying....
https://www.onsemi.com/pub_link/Collateral/AND8414-D.PDF

it says...

The linearity specifications of the digital POT are
similar to a DAC, one-half to one LSB

Click to expand...

Thus I believe that the accuracy of a set voltage , whether with the MCP4013 digital pot, or the MCP4716 DAC, will be similar?

That is, if we are saying that the MCP4013 Digital pot might have a cumulative error problem, then so too would the MCP4716 DAC?

MCP4013 Digital pot datasheet:
https://ww1.microchip.com/downloads/en/DeviceDoc/21978c.pdf

MCP4716 DAC datasheet
https://ww1.microchip.com/downloads/en/DeviceDoc/22272C.pdf