I decided to check and there is actually no aliasing. I used following parameters:
And here is what my oscilloscope showed:
Note that there is only one peak at 53 Mhz.
However if you connect TX and RX channels on LimeSDR and plot FFT you will see an alias. This happens because 50 MHz will be shifted to 0 MHz before calculating FFT and because of the way FFT works it will mirror all positive frequencies in the negative part of the plot.
So far it looks like my original theory was more or less accurate (except the CMIX part - it doesn’t look like it changes anything).