Hi -
I'm reading through the paper on this site about modeling nonlinear magnetics: http://www.designers-guide.org/Modeling/mag.pdf
and I've noticed a few glitches.

In listing 2, for the winding, the electrical terminals are declared as ep, en, but then one of the lines in the module has
 V(ep,e2) <+ turns * (r*I(ep,en) - ddt(Phi(mp,mn)));
Obviously, that "e2" should be "en"

Then, comparing Figure 2 to Listing 4, the topology is the same, but the elements have been renamed: gaps G1 and G2 in the figure are named G2 and G4 in the Spectre netlist; windings W3 and W4 in the figure are named W3b and W3a (respectively) in the netlist.

Finally, where I'm stumped, is that Figure 5 purports to show results from simulating Listing 4. However, the plot shows a node voltage "n4" but there is no "n4" in the netlist, and while there is an "e4", it's not listed on the "save" line, so I woudn't expect to be able to plot it.

Has anyone reproduced the plot in figure 5?

In Verilog-A the port directions are effectively just comments. The simulator itself ignores the directions you specify. Port directions only become significant in Verilog-AMS when it comes time to insert connect modules.

I think the netlist is not used to generate Figure 5. Instead, I think it is intended to produce something similar Figure 4. I infer that from the PWL source and the polynomial controlled source that are used to create a stimulus whose amplitude varies over time. But Figure 4 is just a cartoon, not real simulation results. I believe the netlist in the paper may not correspond to any of the results given in the paper.

-Ken

The loops aren't as nicely spaced as in Figure 4, but the general idea is there.

I'm playing with these models again, and it's too bad there aren't units specified for any of the parameters:

module core(p,n);

magnetic p, n;
parameter real len=0.1 from (0:1000);   // effective magnetic length of core
parameter real area=1 from (0:inf);     // magnetic cross-sectional area of core
parameter real ms=1.6M from (0:inf);    // saturation magnetization


Spectre now has built-in components for "core" and "winding" and I noticed that area and len are in cm^2 and cm, respectively, for the built-in.

The first core and winding models in Spectre pre-dated Verilog-A, and I think they were purely linear.  When Mark joined the team we added the nonlinear core model in Spectre and created Verilog-A version of all the magnetic models and wrote the paper at roughly the same time. At least that is the way I remember it, but that was a long time ago.  Anyway, all that probably suggests that the units you found in the Spectre documentation are probably the right ones.

-Ken

There's a nice demo at Wolfram:
 http://demonstrations.wolfram.com/JilesAthertonModelOfMagneticHysteresis/
which includes units for the Jiles-Atherton parameters:
ms in A/m
a in A/m
k in A/m
c and alpha are dimensionless