Inductor question . . . permeability vs. AC flux

["Tesla list" <tesla-at-pupman-dot-com>]

Original poster: dhmccauley-at-spacecatlighting-dot-com 

When designing a DC filter inductor the other day, I noticed something
interesting.

The first thing was expected.  With an iron powder (distributed air gap)
core, as you increase the amount of DC bias on the core, the permeabililty
of the core decreases.
So for example, an inductor with zero bias might read 50uH but with 20A DC
bias on it, its down to about 10uH.  Okay, pretty normal.

Now, the second interesting phenomenon is that if you have a core with some
DC bias on it, and now you have some AC ripple on that bias, it turns out
the more
AC ripple (or AC flux) there is on the core at that point, the permeability
starts increasing dramatically again.
So my particular core is almost saturated with 20A of DC current and
permeability is about 20% of nominal at this point on the B-H curve.
However, if there happens to
be 10A ripple current (which there is) riding on this 20A of DC current, the
permeability increases by over 250%.

Can anyone explain this phenomenon and what is causing it?