Hi @blurpy, the link @VAR posted is the best starting point. In your specific case, you have made a divider with a 10K and 220K resistor, so the impedance of the divider is about 10k (10k parallel with 220k).
If you go to a resistor calculator such as HERE and plug in your values, you will see the noise from a 10k resistor in 20 kHz bandwidth is about -115 dBV, which might seem fine since that has the same noise floor as the QA401. But, in practice you need your divider to have 5-10 dB less noise than the analyzer, otherwise you will degrade the noise floor of the analyzer. So, if we cut our divider impedance in half, we’d see the noise drop by 3 dB. So, you’d really want your divider to have an impedance of no more than 2500 ohms, but ideally around 1k.
In addition to the resistor noise, you need to consider the input opamp bias current noise. For the OPA1612, as it’s a bipolar opamp, its “self noise” will be considerable with source impedances rising beyond 1k. But around 1k or lower, the impact will be modest.
So, thus far, there are two good reasons to keep your source impedance below 1k ohm.
Finally, the higher your source resistance, the more susceptible your setup will be to picking up radiated energy all around. And in your case, that looks to be the greatest offender. Your 10k ohm effective impedance means it’s easy to pick up powerline. If you used a 100k effective impedance, the powerline spikes would be worse still, and a 1k would be better still.
So, when picking a source impedance for a divider, aim to keep the effective resistance below 1k. And the lower the better.
When measuring power amps, the best of all worlds is to build the attenuator directly into the load as the article @VAR linked to shows.