This question takes off from this question regarding syncronization of BPSK signals. For the sake of argument let us assume 10 samples make up 1 bit.
One method currently suggested is break up the envelope of the downmixed and matched filtered I and Q signals into multiple 10 sample bit intervals, Then, the average of those multiple 10-sample long vectors is taken, to yield a final 10-sample long vector. The position with the largest peak is taken to give an estimate of the optimal sampling position for a bit.
The idea in the averaging cancel out the effects of the noise in this case. Can this method be counted on to also mitigate multipath effects? If not, why not? Is it because the position will 'slip' over time? I tend to not think this will happen, but would like your inputs. Thanks