INA149: Either input offset or gain is out of spec on 5 sample designs.

The most likely cause of the problem you see is that you are operating INA149 outside of its linear input/output voltage range.  For the power supply +/-15V and Vcm=REFA=REFB=0V, the valid input/output voltage ranges are shown below BUT if you change these condition the ranges will also change.  

For default conditions shown in datasheet table above, your circuit should work just fine. - see below. 

However, if you use different supplies or reference voltages, you need to tell us the conditions you use - please use attached Tina-TI schematic to make the changes.

Bob INA149 Circuit.TSC

This is a copy of my design. I am biasing the INA149 with +15 and 0v. I am measuring the voltage across R23 (10K) in series with 2 100K resistors connected to an isolated voltage of 42V. My expectation is that the voltage across R30 (~2.2V) should appear from pin 6 to GND. I have 5 prototype modules with measured results.

I am getting error voltages in excess of the spec. My original design used INA148 with very small errors. Changed to INA149 due to supply change issues.

Marek

I changed the steering resistors to a lower impedance to compensate for the lower input impedance of the INA148. From the simulation the results are acceptable for my application. I modified one of my prototypes with the new values. Module 1 from table above.

Measured results:

R3 voltage = 1.977v   

Vout to Gnd = 1.950

Offset error = .027v

The simulation does not mirror the real device. Error is same 27mv 

Questions

1. How accurate is the spice model? The simulation yields totally acceptable behavior.
2. If this part is truly laser trimmed, why do my original 5 units vary from 27mv to 80 mv? My expectation is that the error would be more consistent. which would solve my problem of coming up with a limit for the processor connected to output.
3. Can you check with factory if this error has occurred before? (all 5 parts have Date code 03L40.) This seems like a component problem.
4. The behavior of previous part INA148 was totally acceptable for my design.
5. Can you forward a phone #?

Marek

I cannot change the design. If I had to buffer the inputs I would not use this part for the price I am paying.  I have 100 modules coming in at any time. I can only work with what I have. This means changing the voltage divider resistors to give the best results.

There is no mention of buffering the part in the spec.

The following sentence is from the INA149 Spec.

Most applications use the INA149 as a simple unity-gain difference amplifier. The transfer function is given in Equation 1: VOUT = (+IN) – (–IN) (1) Some applications, however, apply voltages to the reference terminals (REFA and REFB). The complete transfer function is given in Equation 2: VOUT = (+IN) – (–IN) + 20 × REFA – 19 × REFB

My first choice is lowering the input impedance to 10k 1K 10K. I see that you didn't pick up that change. I believe you will see the error is reduced in the sim but not in the actual design. 

Bob,

I did pick up on your lowering of the input resistor divider values BUT wanted to make sure to communicate why there is still a large error.  The transfer function equation still work and has G=1 but your input voltages are not 20V and 22V because of the error caused by the input currents - see below. This is the reason why difference amplifier cannot have a resistive dividers in front of it and you'd need an instrumentation amplifier, which has much lower IB.

Having said that, with 10k/1k/10k divider you should see perhaps up to 20mV but not 80mV error unless your resistors have up to +/- 5% matching or reference voltages are not directly connected to the ground - see below.

All in all, if you cannot add a buffer, the solution would be to switch to instrumentation amplifier like INA823 or INA121 with 25V-30V single supply - see below.

Marek

We are getting closer to the problem. I cannot change anything except resistors. Parts have been purchased and soldered on boards. The solution using 10K 1K 10k as the dropping resistors giving me smallest error <20mv. This error is a acceptable solution for this application. The resistors have a tolerance of +/-1%.

Lets go on to the real issue. I have populated a module with the new resistor values. When making discrete measurements I do not see an improved error in the output error. This module exhibits the same output error with the 100K/10K/100K (77mv) and 10K/1K/10K (78mv) . This result is extremely surprising to me. There seems to be some internal behavior no accounted for. Input offset???

Most applications use the INA149 as a simple unity-gain difference amplifier. Note: This statement is from the data sheet.

Bob,

I see one more potential issue that may explain why your simulation does not match measurement - the violation of the input range. As you may see below the differential signal is specified as +/-13.5V on +/-15V supplies with VREFs grounded - that's 1.5V from either rail.

However, under your conditions the Vin is less than 1V from the rail, which causes the input stage to operate in a non-linear region - see Vinn=854mV and Vinp=847mV - that is 7mV error that translates into a gain error.  Thus, you would need to raise the Vin voltage above 1V - one way to do so would be to increase the input voltage, VG1, to 63V while changing 100k resistors to 150k (or 10k to 15k) - see below on right side gain goes to 1.

If you cannot change the VG2, you would need to mismatch R6 and R11 resistors in order to increase Vinp=Vinn above 1V - see below.  Doing so eliminates the non-linearity of the input stage - with R11 of 15k and R6 of 5k Vin=~1.5V and G=~1 - this probably is the best way to mitigate your problem as it only requires different resistors.

Marek

Did following on real circuit.

Used 15K 1K 5K per your suggestion to increase Vinp = Vinn > 1.49v . Still have ~30mv offset. 

Raised Voltage source to 49v to increase Vinp Vinn. From sims below yields Vinp = Vinn = 1.75v. Still have ~30mv offset.

Disconnected pin 4 from GND and connected to -15v. Offset is gone, works properly. Offset gone with generator at 42v or 49v.

Before we start "high fiving" Can you provide some explanation?. This fix is not easy to perform.

Bob,

There is no 30mV offset but rather 30mV input voltage error caused by the unbuffered input current; thus, circuit works correctly with G=1 - see below.

Here is the simplified circuit showing the reason for the input voltage error - see  below.  

That's the reason I said at the beginning that difference amplifier must be buffered to eliminate the input error - see below.

Thus, the input voltage error is caused by the unbuffered input to difference amplifier - see below.

Since you said you cannot change anything but the resistor values, you must lower 1k input resistor to ~983ohm to get 2V input - see below.

Marek

I am not sure you are tracking the issue. I am telling you that by changing the power supply feeding the amp to run at a +/- 15 volts and not +15 and 0 volts there is no error. The output tracks the input perfectly. I am using the actual circuit in my prototype. 

Don't you have an EVM module that you can corelate what I am telling you? The issue is now related to the amp. Why does it need +/- 15 rather than +15 and 0v? 

Please provide a contact phone number and we could solve much faster. Send to bbland@ccpower.com. I have product to ship. 

Bob,

You have two issues with your application:

1. The input voltage is not 2V (so the output cannot be 2V) because of the unbuffered resistive divider in front of the INA149 difference amplifier

2. The INA149 Vinp linear range is 1.5V above the negative supply but with 100k/10k/100k or 10k/1k/10k input resistive divider you violate this range.  If you use -15V negative supply you eliminate this issue - see below. Btw, I used below a higher value resistive divider only to make my point clearer because of the higher error it causes but the same problem occurs to lesser degree with 10k/1k/10k resistive divider.

I have sent you my contact info in separate email.