I had my sketch aligned to the edge of the sync, but, after the sync part, through the message, it always moves in steps of one quarter of wave. That means that if the delays I'm using are not accurate enough (and they probably aren't), I could be out of sync after a few cycles. As the message is quite short (32 bits), the code worked and I got away with it, but if the conditions change then it could break.
I've changed my code to solve this. First, I don't look for the preamble at all, I just go for the sync. In fact, the code works if you only look for the second part (the high zone) of the sync, but I detect too the first low part because there were too many fake positives (although the message for them is invalid). Then align to the falling edge of the second zone, wait for what I think is the delay of the third pulse (5100 microseconds) and start a loop to read bits: wait one quarter, read possible value, wait for edge (this realigns the timings), wait one quarter, confirm the value has changed and then wait for another quarter to be at the beginning of the next bit. This should be much more robust than before.
Have you improved the reliability?
Cheers, Rob