I am not very familiar with those functions, but I am wondering if the difficulty is in the fact that 0.1 cannot be exactly represented in binary floating point? And 0.1 is an odd multiple of 0.02 so the two numbers could potentially have different round offs on the occasional sample.
I imagine that the infinite SNR AWGN channel might not be synchronized on the two side, so there could occasionally be slop about exactly which receiving bucket sampled which transmitting bucket. I would be wary about that unless the system was configured for a common clock. Or is an AWGN channel self-synchronizing on the clock signal?