Issue with Lumped RLC in HFSS

[jpotter]

I am using HFSS simulating a 3D coil with an external Lumped capacitance. I want to calculate the change in frequency with a change in the external capacitance. My calculation of the internal stored energy depends on the difference of two frequencies from a change delta_C in the external capacitance. However, HFSS recalculates the mesh when the external capacitance changes even though there is no change in the geometry. To get sufficient precision I need a fairly big mesh. If I could simply keep the mesh from the first value of the capacitor I would eliminate the time to recalculate and reduce the noise that comes from using a different mesh with the same convergence criterion. There seems to be no way to do this, and I can't find any workaround. I'm looking for ideas or a solution. Thanks, Jim

 

[volker@muehlhaus]

Is this low frequency, wireless charging or similar? The best method is to not include the lumped capacitor in EM simulation, and instead include it by nodal circuit simulation later. In that case your EM model covers only the inductor and possibly some interconnect.

 

[jpotter]

Thanks for the reply. That is an interesting idea. My problem is a high Q resonator. I have been using the eigenmode solver. I want to know the inductance and the effective capacitance at the terminal. If I can accurately get the reactance as a function of frequency I can do a least squares fit to get a value for L and C. I am trying to resolve some issues with the source impedance. My answer shouldn't depend on the source impedance and I thought I turned renormalization off, but I am getting admittances 2*10^26. My expected admittance should be around 10^-6-j0.002. Clearly, I don't know what I am doing. Most of my HFSS experience has been with high Q resonators at 3 GHz.

 

[volker@muehlhaus]

As I wrote above ... the proven solution is to exclude the lumped C from your model. You then have a low Q problem that is easy to solve, and results are valid because the lumped C has no field intercation with the EM structure anyway. It is purely nodal connection, and you should use that to your advantage.

That's my advice, based on 20+ years in EM support for other EM solvers.

 

[jpotter]

I like this idea because the mesh doesn't change between frequency steps, which is a problem with my present approach. This gets rid of the noise due to mesh variations between steps since I want to look at the change in reactance with frequency. If I take two frequencies the change in reactance lets me calculate the internal LC circuit, which is composed of a coil with turn-to-turn capacitance and coil-to-wall capacitance. What I am trying now is a current excitation. If you have suggestions to How I should be doing this I'd like to hear them. What I want to do is calculate Y at two frequencies, f1 and f2 and get two admittances Y1 and Y2. I'm not concerned about the resistive component. The shunt resistance is about 1.3 MOhms and I have a handle from the eigenmode solutions.
I would have Y1=2PI*f1*C - 1/(2PI*f1*L) and Y2=2PI*f2*C - 1/(2PI*f2*L). I can then solve for the effective L and C at the terminal. With multiple frequencies I can do a least squares fit and further reduce noise in the calculation. Set the energy change convergence factor to 0.001 J. This took about 6 minutes to get results at 3 frequencies. I am going to look at the results now.

 

[jpotter]

I get nonsense answers with a 1A current source. I get a gap voltage of 0.8721+j.00312 with f=13.46 MHz. The reactance of the coil including the internal capacitance ought to be around 500 Ohms, The effective inductance is about 6 uH. So a 1A excitation should yield a voltage of about 500 V. Obviously I am doing something wrong. I would like to get this figured out because the fact that the mesh is constant should improve the noise in my calculations.

 

[jpotter]

J now get sensible answers by looking at the coil admittance with a lumped port.

 

[volker@muehlhaus]

J now get sensible answers by looking at the coil admittance with a lumped port.

Yes, that is what I was going to suggest. You can always get Y or Z parameters from the resulting S parameters.

 

[jpotter]

Yes, that is what I was going to suggest. You can always get Y or Z parameters from the resulting S parameters.

Thanks, I appreciate you taking the time to respond to my question. I spent my career looking at resonant cavity circuits. It never occurred to me to look at the problem non resonantly. Now I am looking at the affect of this analysis on my results.

 

[volker@muehlhaus]

It never occurred to me to look at the problem non resonantly.

This method of breaking up resonant structures was invented by an EM solver company maybe 20 years ago, to reduce the number of frequency points required to cover a frequency band. One requirement is that port calibration is very accurate. I hope that it works ok for your case!

 

[PlanarMetamaterials]

However, HFSS recalculates the mesh when the external capacitance changes even though there is no change in the geometry. To get sufficient precision I need a fairly big mesh. If I could simply keep the mesh from the first value of the capacitor I would eliminate the time to recalculate and reduce the noise that comes from using a different mesh with the same convergence criterion. There seems to be no way to do this, and I can't find any workaround. I'm looking for ideas or a solution. Thanks, Jim

As I recall (I no longer have access to HFSS), you can export the mesh from one simulation, and then simply instruct the solver to use this mesh every time going forward.

 

[jpotter]

As I recall (I no longer have access to HFSS), you can export the mesh from one simulation, and then simply instruct the solver to use this mesh every time going forward.

Thanks for the reply. That would be a solution, but I haven't been able to figure how to do it.

 

[30 dBw Basho]

Hi All,

If one is setting up a parametric sweep in HFSS, you can check "Copy Geometrically Equivalent Meshes" in the last page of the parametric sweep to simulate without re-meshing each time.

However I think this is more of an XY problem and Volker has the right answer, there's no reason to re-solve the inductor in FEM each time, if the inductor isn't changing geometry. Simply drag and drop the HFSS design, simulated with a port across where the capacitor should be, into a circuit design and add a capacitor which you tweak in the circuit design, tuning instantaneously rather than re-solving the FEM each time. You can even push excitations back to HFSS from the circuit simulation and look at the fields corresponding to different capacitor values.

 

[jpotter]

Thanks for the reply. I implemented Volker's suggestion manually. It gives me answers for ReY and ImY that are close to my expectations. I am now working on a version of my pyaedt Python code. My stumbling block is implementing the lumped_port method. I am not understanding all of the arguments. It appears that the method will implement a Modal Port or a Terminal Port. A 50 Ohm terminal port seems to give me the results I'm expecting. I am using only 3 steps in the frequency sweep. I need only two, but three gives me a consistency check. I am not looking to do a circuit simulation. I am optimizing the resonant shunt resistance as a function of the coil parameters at the desired frequency for a fixed external capacitance.

I have attached a copy of the description of pyaedt Lumped Port method as a pdf file. I am finding pyaedt useful for automating parametric designs and analyzing the results of an HFSS calculation. This is much easier than trying to turn recorded HFSS operations with numerical dimensions into a Python program, but there is a learning curve..