RS485 is not fully differential.
- The receivers need to see a signal that's within some volts of their supply voltages (usually +/-14V or so).
- The transmitters electrically connect the transmission cable to +V/GND alternately, to send data. This fixes the transmitter output's common mode voltage firmly at (VCC + GND) / 2.
The oscilloscope is not fully differential.
- Each probe input channel can only resolve signals between the maximum and minimum range the ADCs can measure. This is usually not far beyond a "full screen" reading, at whatever V/div setting is in use. Signals outside of that range are ignored: limited to MAXVAL or MINVAL.
- Each probe will have a different ground return impedance, because they are not in exactly the same physical locations, nor are they linked together along the way (according to your description). Thus, they will measure different (high frequency) signals as well. That is to say, the CMRR of the probe configuration itself is poor.
- To fix the latter condition, you can join the probe grounds to each other. Now, at high frequencies, a step change at one probe tip, will cause a current into that probe, which is carried out of the other probe. As long as the probe impedances (from tip to tip-ground) are equal, the differential (and common mode) voltages will also be symmetrical. This preserves CMRR, but does nothing about the former problem (scope input range).
- To fix both problems, ground the probes to the transmitter's ground reference.
One more important factor to note:
- Isolating the oscilloscope will have very little effect, if any at all, on the high frequency CMRR of the measurement. The isolation transformer itself will have quite a low common-mode impedance to ground. That is to say, it isn't isolating high frequencies. And even if the transformer were ideal, the fact that the oscilloscope, and any attached cables (including the probes themselves), all take up physical space, means they also have equivalent capacitance or impedance to free space. This provides a current-return path for common mode currents. Thus, you will have little or no change in high frequency errors. (The only way to truly solve that, is an ideal isolated probe, where the isolation barrier has extremely low capacitance, and the isolated probe section is no larger than the probe tips and handles themselves!)
When is it okay?
- If you don't need to resolve high frequencies (such as the risetime of the signal, or high data/clock rates), you can still use the ungrounded or bridged-ground probe method, and filter the high frequency content by enabling input bandwidth filters (probably for a cutoff frequency of 20MHz or below). You may still see artifacts due to clipping. You will have to decide what is real and what isn't, and mentally subtract that from your measurement.
Tim