EXFO FastReporter 2 - Manual (Page 372)

Measuring Polarization Mode Dispersion: Theory
362 FastReporter 2
In a long fiber, numerous mode-coupling events occur along the fiber
length, so that light emerging from the output end is the superposition of a
number of pulses with different delays. Nonetheless, it turns out that for
any given optical frequency, , one can always find two orthogonal input
principal states of polarization (PSPs) such that a light pulse with the same
input SOP as the input PSP, undergoes no spreading. For a single wave
plate, the PSPs are the two birefringent axes, whereas for a concatenation
of wave plates, neither the input nor the output PSPs correspond to the
alignment of the birefringent axes anywhere.
Contrary to the case of a wave plate, the DGD and PSPs of a long fiber are
dependent on wavelength and fluctuate in time as a result of
environmental variations such as temperature, external mechanical
constraints, etc. Their behavior is random, both as a function of wavelength
at a given time and as a function of time at a given wavelength. Fortunately,
this behavior can be characterized statistically. It can be demonstrated that
the probability density function of  is Maxwellian and, by definition, PMD
is its rms value, that is:
Note: PMD is sometimes defined as the mean value of the DGD, which for a
Maxwellian distribution yields a value 17 % lower than the rms definition.
If the average is calculated over , PMD is stable in time, provided that the
averaging window is sufficiently large (Dwdt >> 1).
PMD DGD
2
=