Offset Correction for a Fully Differential Signal Chain

[KD494]

Hey all,

So I'm working on a signal chain using two stages of fully differential amplifiers. Essentially the circuit looks like two stages on the picture below one after another.



The part I'm using, pretty much the only one I can find with the right characteristics, has a maximum differential voltage offset of +/-3.5mV. The combined gain of the two stages will be around 1000 in the worst case scenario. Operating around a midsupply cm voltage of 1.65V this should have me hitting the rails. In the past when dealing with single ended circuits I have been able to add a summing node with an appropriate DC cancellation voltage in between the two stages to eliminate the offset. However, without the virtual ground, I don't see how a fully differential inverting stage can implement a summing junction. I'm wondering if there is a similar scheme I can use here, implementing a DAC or some controlled voltage source in a feedback loop to keep my signal centered around the common mode voltage I desire. Thanks in advance.

 

[Warpspeed]

How you go about this depends a lot on the nature of the signal you are trying to amplify.

If the dc input signal can be interrupted, or is known to be at true zero zero at certain times, some kind of long time constant ac coupling with cyclic auto zero, similar in concept to a chopper amplifier or a sample and hold might do it.

If constant true uninterrupted dc amplification is required, a self zeroing system might be useful at initial power up, using a DAC and a low end microcontroller.

The usual way is to use a bipolar current source feeding into a summing junction somewhere, which allows the summing point voltage to move up and down with the signal.

 

[AMS012]

Yes, You could add a DAC at the the node named Va in the above picture, you will have to add current sourcing DACs and current sinking DACs on both sides of the differential halves (at Va nodes) to be able to handle any kind(+/-) of DC offset occurrence.

Another way could be to implement a high pass function in the path with a very low cut-off frequency, but this depends on the nature of signal you are trying to deal with as pointed out above.