Recently I bought a used Keysight U1620A scopemeter from their official eBay webstore.
It came with ~0.1div CH2 offset, which goes away after warm up. When unit runs hot, the offset goes back on other direction.
CH1 is fine without any temperature dependent offset.
Needless to say, there's something that is temperature dependent, and I need to find it out.
My first attempt was to replace voltage reference chip ADR360, which didn't solve the problem.
I then traced the signal path back from AFE output. The AFE output has an AC path, which went through 2 HV caps to isolated LV side. This meter uses differential capacitive isolation. The DC path went to a differential to single ended converter, then went to an OPAMP to drive a high linearity (feedback-based) optocoupler. One PD output went to LV side, the other went to feedback of optocoupler driver.
I tried to short the input of the optocoupler OPAMP driver, and the result is great -- no offset anymore over cooling and heating. That means the optocoupler driver, the optocoupler and the voltage reference are good.
I then traced back to the differential to SE converter. With its input shorted, the offset went away as well, so either the converter is fine, or it injects offset current to its source.
I then traced further back to the PGA that drives both isolation caps and the DF-SE converter. This alleviates some issue, but the problem persists.
My initial conclusion is the PGA is faulty. To verify this theory, I desoldered its signal source (an FDA pre-driver), and shorted the output of the pre-driver (input of PGA). I cooled and heated the boards, and the offset persists and varies with temperature.
However, I did notice some residual offset from the FDA, so I figure I can replace the FDA as well.
So, I ordered the PGA (AD8370), the pre-driver (LMH6552) and the DF/SE converter (AD8130). Unfortunately I missed today's cutoff time, so I will only get the parts on Friday.
Anyway, the signal path is interesting, and I'm working on tracing the board.