FPGA Central

I've to calculate the phase difference between 2 quasi-sinusoidal 50Hz
signals (one signal is a distorted version of the other, the source
signal).
I will use the method suggested in a previous post (Clay S. Turner),
using analogy with the "power factor" (I don't need the sign of phase
difference):
<quote>
form the three following scalar dot products:
a = E dot E
b = I dot I
c = E dot I
then
c
-------- is cos(phase difference)
sqrt(ab)
</quote>
This method is very suitable for me because I also have to compute the
RMS values (of the fundamental) of the two signals.
Before any computation I filter the two signals using a bandpass FIR
with a very narrow band (2 Hz) centered at 50Hz.
My question is: how can I estimate the accuracy of this method?
I don't need very high accuracy (0.1-0.5 degrees is enough), but how
can I ensure I have this accuracy?
Thanks,
Stefano.