The value of "oversampling" in noise removal is arguable. If you have as much noise in your system, that the "decision logic" "kicks in" regularly, it's too likely you will have data loss, too. Although it is good to have something against spurious spikes, it is also a recommended practice to packetize, checksum, etc.; i.e. be able to recover from data distortion/loss, which is an equivalent measure than the "oversampling".

As its cost in silicon is negligible, all HW UARTs implement the "oversampling", but in bit-banged UARTs it'scost is considerable, so a matter of decision.

Also, a discussion can be open whether the noise suppression is impaired if the 3 samples are taken not at 1/16 of bit ime apart, but using 3 consecutive instructions, which is easier to implement in SW.

There has been a long discussion here on the same topic a year ago or so, maybe it's worth to look it up.

Jan Waclawek