The basic principle of a early-late gate timing estimator :
In demodulators we have the matched filters or the correlators...the output of these is to be sampled at the time instant T(duration of the pulse) to get a maximised SNR at the output..
But at the receiver we dont have the knowledge of T...estimating this T is called the timing estimation.
Now whatever is the pulse shaping being used, the the output is even symmetric about T. So to start of with the estimation we assume a value of T....now we sample the output at two instants T-δ (early) and T+δ (late)......
Now if the two sampled values are equal, our T lies exactly in between the two sampling times we have choosen(as the output is even symmetric about T)...i.e our estimate was correct.
But if the sampled outputs are not equal, we shift the value of T in a direction which may make the sampled values equal.....
we continue this until we get the equal sampled outputs....
for more info...you can go through the book---Digital Communication by:John G. Proakis
Regards,
Sumit