Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Best way to get frequency dense results for waveguide based simulation

Please login with a confirmed email address before reporting spam

Hello,

We have a structure where we need pretty dense frequency results, and want to try to optimize it. We would need to run a good amount of iterations. Just trying to do straight forward frequency domain simulation ends up being pretty slow. I was thinking about doing a transient simulation and then FFT into the frequency domain to get more points quickly, but transient in wave optics does not seem very friendly to waveguide type simulations.

Does anyone have a good way to approach this in COMSOL?


1 Reply Last Post 13.11.2021, 11:37 GMT-5
Robert Koslover Certified Consultant

Please login with a confirmed email address before reporting spam

Posted: 3 years ago 13.11.2021, 11:37 GMT-5
Updated: 3 years ago 13.11.2021, 11:49 GMT-5

Just some ideas: (1) If you "need" tightly spaced results in frequency, then presumably this is due to having structures where there are some very narrow resonances (or similar) or a large number of modes (waveguide or cavity modes) present. You may be able to apply an eigenfrequency study, to help you quickly identify those resonances. Then, you can limit your use of dense frequency spacings to frequencies near those values, and use a coarser spacing in frequency elsewhere. (2) Unless you actually want your system-under-design to exhibit highly narrowband resonances (or similar) and/or many different modes, then you might want to reconsider your whole design concept and seek one with less-problematic behaviors! (3) Unless you have a lot of Comsol experience, you probably haven't chosen the fastest solvers, choices of meshing, etc. To get quick preliminary (but sometimes less accurate) answers, set your mesh discretization to linear. (The default is typically quadratic.) Linear will solve much faster and consume far less memory. In regard to mesh, for regions where fields vary slowly on a spatial scale, consider meshing as coarsely as lambda/6, or so. Use fine meshes only in regions and on surfaces/edges where you expect to need to resolve the details. Get to know how the mesh setting parameters work and adjust your meshes carefully to get the most bang for the buck. If you have enough RAM, use a direct solver. I use PARDISO for most of my RF modeling. And... if you don't have enough RAM, consider getting more! Your time is valuable (or should be). (4) Have you already made use of all the symmetries in your model? If there is a plane of symmetry, use it! (5) Are you limiting your model details to only what matters? If you imported your geometry from somebody else's model (e.g., a SolidWorks model made by a mechanical engineer) then there is likely a whole lot of extraneous detail you can throw away. For speed, your goal should be for your model to capture the physics that matters, and only the physics that matters, within only (or nearly only) the volume that matters.

-------------------
Scientific Applications & Research Associates (SARA) Inc.
www.comsol.com/partners-consultants/certified-consultants/sara
Just some ideas: (1) If you "need" tightly spaced results in frequency, then presumably this is due to having structures where there are some very narrow resonances (or similar) or a large number of modes (waveguide or cavity modes) present. You may be able to apply an *eigenfrequency* study, to help you quickly *identify* those resonances. Then, you can limit your use of dense frequency spacings to frequencies near those values, and use a coarser spacing in frequency elsewhere. (2) Unless you actually *want* your system-under-design to exhibit highly narrowband resonances (or similar) and/or many different modes, then you might want to reconsider your whole design concept and seek one with less-problematic behaviors! (3) Unless you have a lot of Comsol experience, you probably haven't chosen the fastest solvers, choices of meshing, etc. To get quick preliminary (but sometimes less accurate) answers, set your mesh *discretization* to *linear*. (The default is typically quadratic.) Linear will solve *much* faster and consume far less memory. In regard to mesh, for regions where fields vary slowly on a spatial scale, consider meshing as coarsely as lambda/6, or so. Use fine meshes *only* in regions and on surfaces/edges where you expect to need to resolve the details. Get to know how the mesh setting parameters work and adjust your meshes carefully to get the most bang for the buck. If you have enough RAM, use a direct solver. I use PARDISO for most of my RF modeling. And... if you don't have enough RAM, consider getting more! Your time is valuable (or should be). (4) Have you already made use of *all* the symmetries in your model? If there is a plane of symmetry, use it! (5) Are you limiting your model details to *only* what matters? If you imported your geometry from somebody else's model (e.g., a SolidWorks model made by a mechanical engineer) then there is likely a whole lot of extraneous detail you can throw away. For speed, your goal should be for your model to capture the physics that matters, and only the physics that matters, within only (or nearly only) the volume that matters.

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.