Discussion Forum

No one can help me ?

Hi,

I'm no expert, but if I were tackling this, I'd start as follows:

1. For the impulse response, you should define the port to be a "circuit port" ....I've never really played around with that feature much (but still some), but I know you can either user-define voltages (literally write a text file that comsol uses for voltage vs. time information on that port) and just define that voltage in the piecewise fashion they did in the conference paper.

Also, I know at least in 3.5a there's an option to define a spice circuit that drives the voltage at that port, and I'm really rust at spice, but some googling on "spice impulse input voltage" or "spice piecewise voltage excitation" or something like that should give you the process for defining the voltage using spice, then connect that circuit model to be the source port in your model.

2. Insertion loss is essetially what you lose in trying to go from feed line to actual device.

The basic principle would be (Power incident on input port of device) - (Power coupled to device)

Again, I'm pretty iffy on it, but that's how I'd start. Hope it helps at least a little!

--Matt

Thanks you so much, Matt! But I still get stuck. The paper looks like use the time domain (voltage in time domain) but the result (Insertion Loss) is frequency domain. This makes me more confusion *_*

Hello again,

Yes, so they may in fact use the time domain modeling to get the data. However, it's what you can do in post processing that can get you the frequency domain representation.

Since their input to the system is an "impulse" function (in this case, just a very short duration spike in voltage), it acts effectively as a dirac delta function (voltage in this case).

When you hit a system with an impulse (any system, not just this one), the output can be regarded as the transfer function of the system.

That is to say:

Impulse = SYSTEM_INPUT----->SYSTEM---->SYSTEM_OUTPUT = Transfer_Function (for impulse input only)

In your case, the input is voltage, and the output you want to monitor is insertion loss.

In equation form, and generally:

Insertion_loss(t) = convolution [ h(t), V(t) ] (where h(t) is the transfer function from voltage input to insertion loss output)

That's in the time domain.

If you take the fourier transform of both sides of that equation you'd get:

Insertion_loss(OMEGA) = H(OMEGA)*V(OMEGA), where OMEGA is the angular frequency w = 2*pi*frequency

If the duration of you voltage pulse is very very short compared to the natural response of the system, then you can regard even a finite duration voltage pulse as a Dirac delta impulse function. Not strictly true, but as far as the system is concerned the voltage spike occurs all at once in an instant, just like the mathematical impulse function.

This means your insertion loss output as a function of time (insertion_loss(t)) is essentially the transfer function from voltage input to insertion loss output.

Those are really just details....you can find more in any signals and systems textbook....or with a quick google....

Ok, now all this boils down to the fact that you need to convert your time domain in post processing to the frequency domain.

Do you have matlab?

If you export Insertion loss as a function of time you can use the built in function " fft " to do a numerical discrete time fourier transform.

It's been a while since I used that, but I can tell you that "help fft" on the command line will give you all the info you need. I recall the only slightly tricky thing about it was getting the frequency axis correct (x-axis, y-axis would be Insertion loss). However, last I knew the example in matlab's documentation was easy to follow and made it clear what kind of scaling you had to do to get your frequency axis correct.

I also think I remember reading somewhere that Comsol version 4.x had the ability to do this straight from the post-processing options in the GUI....not sure if you're using that version, though....but that might be even handier if you can find the fourier transform option in there somewhere.

Alrighty then....Hope I explained that ok...good luck with it!

--Matt

PS You could also probably do a parametric sweep of stationary models and vary the frequency. At each frequency across the range of your interest, you could pull out the insertion loss...I THINK that might be another means to the same end...

Yes, maybe I run the model in time domain in Comsol, then use the Fourier Transform to change the result to frequency domain. I will try this way.
Thanks you so much, Matt. Wish you have a nice day!