The OP may have called it quits, but I hate unanswered questions.
But why does XLR fix EMI?
Because it wasn't EMI in the first place, but a ground loop at work.
A ground loop makes itself felt every time the common ground potential between two devices differs, as for an unbalanced connection this looks exactly the same as actual signal voltage.
Now the SP200 should generally be sitting at outlet PE potential, given that the current flowing through its very own PE conductor is best described as "sod all". (PE = Protective Earth.)
If you have PC --> USB --> M500, the common ground on the M500 has to go through:
1. USB cable shield
2. Mo/bo ground planes
3. ATX power connector and power supply cables, power supply internal connections
4. IEC power cable PE conductor
(all of generally low but nonzero resistance)
before we finally arrive at the outlet. #1 is shared with DAC power supply, #2 and #3 may be shared with substantial return currents from PSU and GPU supply, and #4 tends to exhibit substantial capacitive coupling to the L and N conductors (e.g. from PSU-internal filtering) so AC components of power supply mains current will have a tendency to couple into it capacitively. The latter issue may actually be present on the SP200 as well but its power supply in general will be much, much smaller.
This leaves us with an AC voltage between M500 ground and SP200 ground. Now establishing a second connection between the two via an audio cable is going to reduce this voltage but not bring it to zero, as the first path is fairly low-resistance to begin with. This should explain why RCA cables with low shield resistance are preferred.
Balanced connections, properly implemented (
*cough* AES48-2005 compliance *cough*), care precious little about
common-mode voltage differences, just as long as they're within a few volts or so. Shield currents will be flowing past the audio circuitry entirely, which cares only about
differential voltage between the two signal conductors. If some voltage appears on both equally as it would here, it cancels out almost entirely (the relevant performance parameter is called
common-mode rejection ratio or
CMRR, and in this case I would expect about 40-50 dB - which, while not infinite, improves the robustness quite dramatically over the 0 dB of an unbalanced connection). This potentially makes the equipment designer's job a whole lot easier - you can just connect your audio ground to PE and don't have to worry about mains leakage in transformers or mains filters.