Ok, so I have an Agilent N1912A power meter with N1921A power sensors on Ch A & B. I have RF power (from signal generator/ amplifier) going into input of a coupler, then Ch A connected to the coupling and Ch B connected to the coupler output. I am trying to measure the loss of the coupling (nominally 20 dB) by just taking the difference in power readings (Ch B- Ch A). I have a VI to read the power from both channels as a signal generator sweeps through different power levels. For the most part it works, except the problem is I am a getting strange "kink" in power reading whenever the power sensor reading is around -10 dBm...see first image attached. Because of this, i get over 0.5 dB change in the coupling loss i am trying to measure, which is unacceptable.
Now with extensive troubleshooting I have ruled out things like amplifier having unstable gain, or faulty sensors. One interesting thing I have noticed - it tends to improve significantly when i change sweepintervals to be much smaller (e.g. 0.1 dB intervals instead of 0.5 dB), which makes me think it is something I'm not setting properly in my VI when I am reading the power from each channel. If i use 0.1 dB intervals the kink is much smaller..resulting in a more stable measurement of coupling loss. See second image attached.
Anyway, wondering if there are any ideas what could be the source of the issue. As you can see from the images, this always happens at -10 dBm..dont' know what's so special about -10 dBm (where's the shrug emoticon?? 
) Anyway, I know can greatly alleviate this problem by always using really small intervals of 0.05 dB or something, but that is just so not practical for my application...(would be too slow). I want to be able to use 0.5 dB intervals and have smooth power readings.