Ok, here is what I see.
1. The (M = R1 - 1) difference signal goes low and high due to the phase difference in the acquisition of the live signal copy stored as the reference signal vs the live signal realtime capture.
2. The phase difference arises from the fact the the digital snapshots that are occurring at a 1GS per second rate are asynchronous to the 10MHz input signal. The capture drifts within the time span of at least a 1LSB x 1nS window.
3. When the input signal stored in reference memory was captured, the trigger point (where the input signal exceeds the trigger level) was at one position within the 1nS span between sample points.
4. When the LIVE input signal was captured, the trigger point (where the input signal exceeds the trigger level) was at another position within the 1nS span between sample points.
5. (IMAGE DS0010.PNG) When the first live capture of the input with the reference and difference signal was taken, the live trigger point was very close to the reference copy but not at the same instant resulting in a spike in the difference signal
6. (IMAGE DS0011.PNG) When the second live capture of the input with the reference and difference signal was taken, the live trigger point was much further from the reference copy resulting in wider pulse (~4.8div x 2nS/div = ~9.6nS) in the difference signal
7. Referring the image (IMAGE DS0010.PNG), the interval between the rising edge of R1 (WHITE) and the rising edge of the live signal (YELLOW) causes positive difference signal M (800mV) since R1 is at +400mV amd the live signal is at -400mV. (+400 - (-400)) = +800
8. Again, referring the image (IMAGE DS0010.PNG), the interval between the unseen falling edge of R1 (WHITE) and the unseen falling edge of the live signal (YELLOW) causes negative difference signal M (-800mV) since R1 is at -400mV amd the live signal is at +400mV. (-400 - (+400)) = -800
9. You should try the same experiment with longer acquisition memory up to 10Mpts. The pulse width span from test to test will occupy a narrower range.
10. Who can figure out the upper bound on M 's pulse width since it is around 10 times greater than the 1nS sample interval in the second figure?
Make any sense - it's late.
Regards,