Hi,
I've done my homework on this one, crawled through the web & talked to
colleagues. If I am missing something obvious please point it out - it's not
for lack of effort on my part!
I'm not really sure what you're expecting. Of course the phase
alignement between the Tx and Rx is going to be random depending on
restart.
Using the same clock will prevent it from drifting, but the initial
phase alignement is random. This is usually resolved by using training
sequence, headers, differential encoding, ...