Hi Kai,

For a random data sequence you could look for a quanta period T such that every measured period is equal to n x T for some n = 1, 2, 3, 4, ..., 8, (9, maybe?), within some acceptable %error tolerance (n = 8 or 9 maximum assumes that the random data stream is limited to a single character). Hasn't it been shown in another thread(s) that the unitary period must be within 2% or 3% of the ideal in order for the UART to work. I suspect that would define the acceptable error in any test.

Of course, all of this assumes that one has to be able to "autobaud" any random data stream. Frankly, I like the way Atmel does it. The firmware assumes that the autobaud character is an uppercase 'U', which means that the data stream is a series of alternating 1's and 0's. In other word, every transition occurs after a single (n = 1 above) untary period. That's got to make for the simplest algorithm possible.