# Diff. Amplifier with Single Supply Opamp - Strange Behavior

I'm playing with some "precision" (as claimed by manufacturers) opamps and DACs to test how "precise" they are.

So, the components I'm testing are:

• DAC7612U by TI: A two channel 12-bit (1mV/LSB, up to 4095mV) DAC with internal reference and buffered outputs.
• OPA2335 by TI: A single-supply opamp with very low input offset voltage (5uV max).

For testing purposes, I've built the following circuit (Note that all resistors are 0603 case SMD ones with %1 tolerance and the output sections of DAC is taken from datasheet as they're shown):

simulate this circuit – Schematic created using CircuitLab

As you can see, the first block after the DAC is a differential amplifier with unity gain and the last block is a non-inverting amplifier with a gain of 1 or 2 (depending on the state of the analog switch, 1G1357). $VO_A$, $VO_B$, $V_{diff}$ and $V_{buff}$ are test/measurement points.

And finally, here are what I got:

• When I reset (zero) both outputs of DAC, I measure about 1mV on both $VO_A$ and $VO_B$. Quite normal, because zero-scale error is given as 1mV in the datasheet. And, as expected, voltages on $V_{diff}$ and $V_{buff}$ are in uV range. No problem at this point.
• When I set OUT_A to about 500mV and leave OUT_B untouched, the measurements are correct on all test points.
• This is the one that I have question about: When I set OUT_A to about 3500mV or higher, and leave OUT_B untouched, I measure about 4mV on $VO_B$, but it should be about 1mV! If I disconnect the differential amplifier, the measurements are correct on both $VO_A$ (the code I loaded in mV) and $VO_B$ (about 1mV). If I reconnect the differential amplifier and use lower resistances (e.g. 15k) for all of R1, R2, R3 and R4 then the measurements get worse: $VO_B$ increases to 8mV even if the loaded code is zero! One more interesting thing: $V_{diff}$ shows $V_{diff} = VO_A - VO_B + V_{err}$; where $V_{err}$ is the error voltage that I measured on $VO_B$ for $VO_A \ge 2500mV$ (It's 0mV for $VO_A \le 2000mV$).

How can it be? Why does this happen? What should I do?

Sorry for long post. Hope I could explain the problem.

• Can you post a picture of your layout? Is the second stage at $\times$ 1? 3.5 V $\times$ 2 = 7 V outside the supply range. – skvery Mar 17 '17 at 9:23
• @skvery It's not about saturation. The switch is closed for small (in mV range) voltages. It's normally open, so the last stage acts as a follower. – Rohat Kılıç Mar 17 '17 at 9:29
• ($3 500~mV\times 2 > 5~V$ on my calculator? :-) – skvery Mar 17 '17 at 10:00
• @skvery You're talking about the last stage, but my problem shows itself on the first stage. As I said, the switch on the last stage is "normally open" (look carefully), so the gain of the last stage is 1 thus $V_{buff} = V_{diff}$. It's closed when measuring a few (1 to 50) millivolts. Again, my problem is on the first stage (differential amplifier with unity gain). If $VO_A$ is greater than 3500mV then $VO_B$ unexpectedly becomes 4-6mV even if the loaded code for DAC_B is zero. Also, the difference voltage, $V_{diff}$ shows some error voltage that is zero when $VO_A \le 2500mV$. – Rohat Kılıç Mar 17 '17 at 10:07
• Do you still have the problem if you disconnect the second stage? The $2500~mV\times 2$ is very close to 5 V? You can also compare values with the switch open or closed. – skvery Mar 17 '17 at 10:11

The act of setting OUTA high lifts the inverting input of the diff amp high and forces a current into OUTB output. This will produce an offset error on OUTB's output.

If you look on page 6 of the DS it shows this effect (output swing capability) but it doesn't really indicate how poor this will be with a 200 kohm. It can be seen at 1 kohm (about 0.1 volt offset induced).

Given also that to produce close to 0 volts on an output when the device itself receives no power rail below 0 volts is a tall ask and it will be susceptible to pull-up effects.

Please also ensure you measure the output relative to the common 0 volt pin on the device.

As a general rule I never rely on the top or bottom 20 mV range on dacs or adcs.

• Pull-down Voltage vs Sink Current graph! How couldn't I see it! The lower the resistances in diff. amp, the higher the sink current into OUTB, thus the more shift on VOUT_B. Thanks for showing that. And yes, the measurements are relative to the GND pin of the corresponding device. Anyway, maybe using a DAC which can produce an output in +/-Vref range (via a split supply) would be a solution. But the minimum required reference voltage is 1mV (that's why a 12-bit DAC is selected) and the system will be supplied from a single 5V source. – Rohat Kılıç Mar 17 '17 at 11:02
• Try a 10 kohm pull down on the pin to see if things improve. – Andy aka Mar 17 '17 at 11:56
• Tried it before with several resistors from 1k to 270k but no improvement... – Rohat Kılıç Mar 17 '17 at 12:27
• That does surprise me. I would expect some improvement. – Andy aka Mar 17 '17 at 13:29
• I loaded both DACs with 1k to 270k, but as I said, no improvement. If I disconnect the opamp stage, I see no problems with or without a load (pull-down) resistor. – Rohat Kılıç Mar 17 '17 at 14:14