INA350: Inconsistent gain over output range

Hey AJay,

Thanks for posting on the E2E forums! Could you please provide the sampling rate you are using on this ADC? Additionally, is INA Vout measured with the ADC, or is it while the ADC is off and this is measured via DMM on TP1?

Best,
Jerry

Jerry,

Great question. I'm not completely sure what the actual sampling rate is, but I have checked to see if sampling is causing a discrepancy. When I'm measuring that 3.5556 output voltage from the INA, if I turn the ADC reading on, the result begins to fluctuate by +/-0.0001V. I also checked with the oscope to make sure that TP1 isn't swinging, and it's holding a steady voltate, with or without the ADC. I tried this test at both the high end and the low end and it seems the sampling doesn't effect the output.

Hey AJay,

You may be running into a limitation in common-mode vs Vout for this device. Are you able to increase Vref to confirm this?

Using the Analog Engineer's Calculator, the mathematical limit from the typical specs of the device can be found.

In case 3 in your second table, here is the resultant boundary plot for the amplifier. The common mode was calculated to be (2.5+2.1305)/2 = 2.315. The max output value here should only be 3.122 (you are seeing a little bit more output voltage here)

Bumping the reference voltage to 2.5V increases your output swing range drastically.

Best,
Jerry

Jerry,

I believe you are correct. Initially when I thought I was within the proper VCM vs VOUT region I was looking at the 5.5V graph in the datasheet. After I designed the circuit, in a hasty decision I changed the power supply to the INA from 5V down to 4.096. I didn't think how that would change the output.  With a little help from the calculator, it's easy to see where I need to be. 

Bumping VREF up to 2V5 falsely widens my output range, in actuality it halves my range. The reason is I will always be "on top" of VREF because my differential input is always positive. That would bump my useable range to 2.5-4V or so, or a range of ~1.5V.

Instead I removed my bias resistor R52. I replaced the RDUT with a 25 turn 20 ohm potentiometer. My VCM is now ~2.4V and the Vout is < 2V, and this is well within chart from the calculator. 

I used the potentiometer to precisely control the output voltage of the INA350 to see where the gain was the most accurate. Note: VREF is reported at 2.4996, ADC measuring was OFF during this test.

I'm still getting really bad gain error! The second to last column shows the gain error % from the last to result to the current, and the last column shows the percent of gain error over the spread of voltages. Below 1V we have as much as a 3.7% spread, and from 1V to 2V we have a 0.8% gain error. This is still greater than the maximum listed gain error of 0.6%. I'm assuming there are other errors that weigh in too, but this just seems like far too much.

Hey AJay,

Good point, I was only thinking from an output swing range and not from the valid range of inputs!

How are you measuring the INA VOUT? If the gain error is being calculated and only measuring to the 100mV level, your measurements can be off greatly. In the case that the differential voltage is 0.0418V, and the output voltage is measured to be 0.4V. If your actual value is 0.449V, your calculated gain is now the same as if your actual value was 0.351. However, the calculated gain in both of these scenarios would be 10.74 and 8.4 respectively. Yet, the equivalent INA Vout would be 0.4 in both cases, if only measured to the tenth of a volt.

I'd be interested to know what the INA output value is on a 6.5 digit DMM.

Best,
Jerry

Excel shortened them... All measurements are done with a BK5491B 4.5 digit. 

All measurements are 4 digits. With the 25 turn pot, I was able to dial in the INA output until my DMM said 400.00mV or 0.6000V exactly, usually with the last digit flickering. I understand the increments may be off slightly - I'm measuring to the best of my equipment. Four and a half digits is definitely enough to show the wild 3.7% spread.

Hey AJay,

Sorry for the confusion, you are correct that 4.5 digits will be sufficient.

Just trying to wrap my head around the test procedure here. You are adjusting the input pot until a target output voltage, then measuring the input voltage value, and calculating the differential voltage gain error from this differential value and the output voltage value?

Please confirm if my understanding is correct.

Best,
Jerry

Jerry,

That's it!

Hey AJay,

I believe that the error that you are seeing is the offset voltage of the amplifier. The typical offset (referred to input) is 200uV, however in a gain of ten will typically be 2mV. In your case I believe it is roughly -1.8mV when referred to the output.

As your input is lower in lower in differential amplitude, your offset voltage is going to make up a larger percent of your error, which accounts for why your larger output values show better gain error than your smaller output values.

When gain error is calculated, it is calculated as the mismatch between the slope of the output divided by the slope of the input. This cancels out any DC errors attributed to anything other than the mismatch in gain setting resistors.

For example, for your first two points, your gain for 0.4V and 0.3V of output voltage will be:

(0.4 - 0.3)/(0.0418 - 0.0318), this yields 0.1/0.01, which is exactly a gain of 10.

Best,
Jerry

Jerry,

Could you share how you came to -1.8mV?

Why I'm so confused by this number, is because it would seem like there's a input offset of 1.8mV. If you subtract 1.8mV from the differential voltage, all gains instantly go to nearly 10 and the gain variance is less than 0.05% from 0.1V to 0.2V.

But this doesn't make sense because there only be a ~200uV input offset. That's why 1.8mV was confusing, it works perfectly but in the wrong way.

Hey AJay,

Since you are adjusting and targeting the input until seeing a desired output voltage, then looking back at the input, this looks like an "input offset" voltage, when in reality it is just the DC adjusted voltage due to the output referred voltage.

Working from the output back to the input can skew how the offset is perceived.

Let's look at this starting from the input voltage and working our way to the output. Here we will target a controlled diff voltage, 30mV, 40mV, 50mV, etc.

Once these values are found, we will measure the output voltage. You will find that if you measure this value it will be 0.2982V. This is the gain of the 30mV difference, minus the output referred offset voltage of 1.8mV.

The input offset voltage is not the difference in voltage between IN+ and IN-, it is the difference of the two input gain stage amplifiers, added with the offset of the output stage amplifier.

The only error that can be seen on the actual chip is the output referred error.

Best,
Jerry

Jerry,

Thank you for the more complete understanding of what's actually going on, however I would still say I'm lost. What you said does make sense on paper, but is more than confusing for me in my lab. I tried what you mentioned adjusting for an exact differential voltage of 30mV then measuring the output.

Jerry Madalvanos said:

You will find that if you measure this value it will be 0.2982V.

Except for 0.0300V -> 0.28089V

I tried it again adjusting for an exact differential voltage of 0.1V

0.1000V -> 0.9813V

In both cases, assuming a gain of 10, there would seem to be a DC offset of 18-19mV. I'm assuming a gain of 10 because you showed in an earlier post that it would appear to be spot on.

Jerry Madalvanos said:

The input offset voltage is not the difference in voltage between IN+ and IN-, it is the difference of the two input gain stage amplifiers, added with the offset of the output stage amplifier.

Thank you getting into the weeds, I do appreciate that, as well as the opportunity to learn! However, I'd like to pull us back to surface level. Unless I'm still completely misunderstanding what you've been trying to tell me, two blaring questions that remain.

1) Any way I slice this pie, it seems like the errors I'm getting (due to gain or offsets or whatever) seem like they are an order of magnitude too large. Even though my understanding of how things work was wrong, I believe the measurements still show these discrepancies. 

2) What actionable items can I do to remedy this situation? 

Jerry,

Thanks for the suggestion. That's actually what I wound up doing with R52, the bias resistor. With this method, the results come out fairly accurate. I will mark this answer as resolved, because in the context of my circuit, it's working good enough after the bias resistor and dc error subtraction.

However, I feel like a DC error of 10x greater what it should be falls squarely on the designer and some oversight. If you are so inclined, I'd really appreciate you taking it step further and helping me analyze why my error is so bad. I'd like to design circuits in the future with confidence. Not knowing why my error is so terrible is killing me!

Thanks!

Hey AJay,

I'd be happy to help you find the source of the error! Are you able to try to source 2.5V from a benchtop supply? I want to isolate just the INA device. I know that shifts in ref will be common to both inputs, I just want to fully evaluate the INA in a standalone way.

Another test: are you able to short the inputs of the INA together, and either shift the supplies or the ref voltage to a valid linear range for the amp?

The most straightforward measurement would be putting a -1V supply on V-, and tying both of the inputs to GND.

Just trying to establish a control measurement, then add pieces back in and see if we can pinpoint where the error is coming from.

Best,
Jerry

Jerry,

I was not able to use the exact same INA from our earlier calculations. Instead, I used an INA from a duplicate circuit. Before I used it however, I wanted to compare it against the INA that has been featured in this post. I did the exact same measurement process as before. The only difference I saw is that the duplicate INA has an offset of -21mV instead of previous 18mV, so basically the exact same situation. 

I built the below circuit with the duplicate INA on a SMD to DIP breakout.

I daisy chained two benchtop supplies to achieve the -1V. Here are the exact measurements: 2.5V -> 2.4984, -1V -> -1.003.

Correspondingly Vout was -15.26mV. After this, I got a brand new INA out of an ESD bag and repeated the test, Vout was -4.70mV. 

Hey AJay,

You mentioned grabbing a new INA, what has been done to the INA's that show a higher offset? Are you performing any aging stresses to these devices similar to HTOL or BHAST? If these stresses are being applied to the units, could you provide a schematic for how the devices are being biased in these stresses?

Best,
Jerry

Jerry,

Nothing of the sort. Both INA's were measured on the same PCB, both have been unbagged within several weeks of each other. 

Hey AJay,

One last test, and sorry for the back and forth on this. Are you able to replicate the datasheet conditions? Can you create symmetrical +/- 2.75V supplies. IN+/IN-/REF should all still be GND. The output should have a 10kOhm resistor connected to ground.

In this case the output/10 should be less than 1.2mV. If it is, we can add one potential error source back in at a time and see where the DC error is coming from.

Best,
Jerry