KSTR
Major Contributor
Nice results, interesting stuff.
Looking at the pure symptoms, I think we're looking at a signal integrity problem: GND polluted by shared balancing currents, right in the analog section, including the outgoing RCA cables.
And changing the USB cable modifies those paths and/or the amount of leakage currents.
If I got Amir's setup correctly, his source feed to the USB cable has USB-GND bonded to PE (desktop PC) and the analyzer has its GND at PE too, right? Those PE potentials are dirty and low impedance, giving rise to large balancing currents on the USB shield, across the DAC's PCB (disturbing GND references) then into the RCA shields which terminate to PE of the AP. Changing the USB cable changes the impedance and hence details of the current flow, R*i and L*di/dt errors along the shared GND path will vary accordingly. On top of that we have the mains leakage from the Class-II SMPS wallwart which also returns to the mains via PE and now has two paths available... then again his earlier measurement with a floating source showed less if any leakage than this new measurement, that why I think the PE-to-PE connection dominates the noise picture.
To test the hypothesis the DAC could be fed via TOSLINK for galavanic isolation and then the original balancing current could be re-introduced by using the USB cable's shield for a point-2-point GND-only connect from source to destination USB-Sockets (soldering the connectors together with short wire stubs would be best, but I see the problem...). TOSLINK alone should give a clean spectrum (like in the earlier measurement) but with the GND connection re-introduced I would hope to see siginificant rise in leakage similar to that when the cable did actual data transfer as well. If switching over to the other cable mis-used as a "GND-bleed" the same way again produced a result similar to its actual use-case then we've got it, a classic "ground loop" scenario. If so, the proper solution should be breaking the loop, preferably at the source side.
Looking at the pure symptoms, I think we're looking at a signal integrity problem: GND polluted by shared balancing currents, right in the analog section, including the outgoing RCA cables.
And changing the USB cable modifies those paths and/or the amount of leakage currents.
If I got Amir's setup correctly, his source feed to the USB cable has USB-GND bonded to PE (desktop PC) and the analyzer has its GND at PE too, right? Those PE potentials are dirty and low impedance, giving rise to large balancing currents on the USB shield, across the DAC's PCB (disturbing GND references) then into the RCA shields which terminate to PE of the AP. Changing the USB cable changes the impedance and hence details of the current flow, R*i and L*di/dt errors along the shared GND path will vary accordingly. On top of that we have the mains leakage from the Class-II SMPS wallwart which also returns to the mains via PE and now has two paths available... then again his earlier measurement with a floating source showed less if any leakage than this new measurement, that why I think the PE-to-PE connection dominates the noise picture.
To test the hypothesis the DAC could be fed via TOSLINK for galavanic isolation and then the original balancing current could be re-introduced by using the USB cable's shield for a point-2-point GND-only connect from source to destination USB-Sockets (soldering the connectors together with short wire stubs would be best, but I see the problem...). TOSLINK alone should give a clean spectrum (like in the earlier measurement) but with the GND connection re-introduced I would hope to see siginificant rise in leakage similar to that when the cable did actual data transfer as well. If switching over to the other cable mis-used as a "GND-bleed" the same way again produced a result similar to its actual use-case then we've got it, a classic "ground loop" scenario. If so, the proper solution should be breaking the loop, preferably at the source side.