Why transient simulation results vary when the EM model stays the same in ADS

[xiaowenrun]

Hello everyone
I have been bothered by such a problem:
I simulated a transformer using ADS momentum to get its EM model. When I changed the frequency plan, for example, from 70GHz-84GHz to 70GHz-104GHz, its S parameter at 77GHz did not change. But when I simulated the EM model with terms in a schematic, I found that the results differed. The excitation signal is 77GHz. I wonder why transient simulation results vary when the EM model stays the same.
Thanks
Wenrun

 

[BigBoss]

S-parameters are discrete type of parameters so they are not continuous. That's why the simulator while it simulates them in a circuit, it uses interpolation ( or extrapolation ) to find the response. While doing this, it converts these frequency domain parameters ( and others ) onto time domain equivalents by applying convolution techniques ( different techniques are applicable ) and therefore sometimes some convergence problems occur.
In order to prevent such convergence problems, the s-parameter block may be represented by its equivalent "continuous components". I mean if you use equivalent circuit that consists RLCk for your coil, transient simulation will not create problem.
It's not difficult to find a continuous type equivalent circuit for a coil. Research a bit literature.

 

[xiaowenrun]

S-parameters are discrete type of parameters so they are not continuous. That's why the simulator while it simulates them in a circuit, it uses interpolation ( or extrapolation ) to find the response. While doing this, it converts these frequency domain parameters ( and others ) onto time domain equivalents by applying convolution techniques ( different techniques are applicable ) and therefore sometimes some convergence problems occur.
In order to prevent such convergence problems, the s-parameter block may be represented by its equivalent "continuous components". I mean if you use equivalent circuit that consists RLCk for your coil, transient simulation will not create problem.
It's not difficult to find a continuous type equivalent circuit for a coil. Research a bit literature.

Thanks for your reply. But the equivalent model is not simple. Not to mention the need to adjust it. I intend to control my frequency plan in a reasonable range, such as 1.15 times the frequency band. So I should be able to get an accurate result. As you said, there will always be errors. So I don't require that I can get an absolutely correct result. I need that it would not be changed unless I change my design. In addition, I am more concerned about the change in the result when I modify the circuit.

 

[volker@muehlhaus]

I wonder why transient simulation results vary when the EM model stays the same.

This is unexpected indeed.

Is that schematic a simple linear simulation or some non-linear simulation where DC response of the EM Model matters?

EDIT: You already wrote transient simulation, so that answers my question. Make sure that your EM Model includes frequency range down to DC. Check response at DC, it seems to be different between your EM extractions, and that has an effect on your active components connected to the transformer. In case of doubt, you might want to add extra elements such as DC-block.

 

[xiaowenrun]

This is unexpected indeed.

Is that schematic a simple linear simulation or some non-linear simulation where DC response of the EM Model matters?

EDIT: You already wrote transient simulation, so that answers my question. Make sure that your EM Model includes frequency range down to DC. Check response at DC, it seems to be different between your EM extractions, and that has an effect on your active components connected to the transformer. In case of doubt, you might want to add extra elements such as DC-block.

Thanks for your reply. The simulation results cover the DC case, and the circuit does not contain transistors. But your advice reminds me that maybe this situation is related to certain kinds of circuits, so I simulated an inductor. And the problem reappears. The schematic is shown below. The spectrum analysis of out_A is done, and its value at 77GHz is checked. When the frequency plan is set as 70-84GHz, the spectrum density at 77GHz is -6.775dB. But when the frequency plan is set as 70-104GHz, the spectrum desity at 77GHz is -6.790dB. what is even weirder is that when freqency plan is set as 70-124GHz, the circuit oscilates diverges. ):

 

[volker@muehlhaus]

I really think that your EM frequency settings (only DC plus narrowband without covering harmonics) are not appropriate for use with transient analysis. Please try again with frequency range DC to 240 GHz (full range).

That said, differences of 0.0x dB are nothing I would worry about.

 

[dick_freebird]

Transient analysis tends to run with pretty relaxed tolerances,
with numerical methods (default) that have kinda poor
stability, and elements with a high "numerical gain" can turn
low level numerical noise into significant (bogus) "signal".

If you get not-credible results, a thing to try is changing
METHOD (from TRAP to Euler, for example - depending on
simulator the "METHOD" may be a text or numerical value).
Can also examine the results of making TRTOL tighter - if
nothing changes, that wasn't it - but does it?

Frequency domain analysis ignores one key thing - phase.
Transient will show any phase dependent goings-on, AC
only takes the DC solution as given and constant.

If I had to pick one to believe in, it'd be TRAN with a few
settings crunk down.

 

[volker@muehlhaus]

Frequency domain analysis ignores one key thing - phase.

I don't understand what you are talking about. The baseline data here are ADS EM-results that include phase response, and you would get that exact data (including phase) from the EM Model block by using S-Parameter controller for simulation.

The issue here seems to be that he tries convolution from S2P to time domain using rather narrowband S2P input. That is expected to fail.

 

[xiaowenrun]

I really think that your EM frequency settings (only DC plus narrowband without covering harmonics) are not appropriate for use with transient analysis. Please try again with frequency range DC to 240 GHz (full range).

That said, differences of 0.0x dB are nothing I would worry about.

Since the circuit is linear, why should a large frequency range like 240GHz be used? The circuit will not produce harmonics.

 

[volker@muehlhaus]

Since the circuit is linear, why should a large frequency range like 240GHz be used? The circuit will not produce harmonics.

Your circuit might be linear, but if you run time domain simulation using S-parameter (frequency domain!) data, that involves heavy math behind the scenes to do the FFT. Your input to the FFT is only narrowband data plus DC, so I would not expect accurate response in time domain, even for a linear circuit.

If you want to EM only narrowband, and your device is linear, you should use S-Param simulation. That is valid because it only requires the actual frequency range used in S-Param controller, and does not need convolution/FFT.

Best regards
Volker // Keysight Certified Expert EDA

 

[xiaowenrun]

Your circuit might be linear, but if you run time domain simulation using S-parameter (frequency domain!) data, that involves heavy math behind the scenes to do the FFT. Your input to the FFT is only narrowband data plus DC, so I would not expect accurate response in time domain, even for a linear circuit.

If you want to EM only narrowband, and your device is linear, you should use S-Param simulation. That is valid because it only requires the actual frequency range used in S-Param controller, and does not need convolution/FFT.

Best regards
Volker // Keysight Certified Expert EDA

Thanks for your reply. I don't know much about running time domain simulation with S parameters. I understand after listening to your explanation.
Best regards
Wenrun

--- Updated Sep 19, 2022 ---

Your circuit might be linear, but if you run time domain simulation using S-parameter (frequency domain!) data, that involves heavy math behind the scenes to do the FFT. Your input to the FFT is only narrowband data plus DC, so I would not expect accurate response in time domain, even for a linear circuit.

If you want to EM only narrowband, and your device is linear, you should use S-Param simulation. That is valid because it only requires the actual frequency range used in S-Param controller, and does not need convolution/FFT.

Best regards
Volker // Keysight Certified Expert EDA

What do you mean by "that involves heavy math behind the scenes to do the FFT"? Does that mean the EM S-parameter model is firstly transferred into an FFT model? And then the FFT model is used for transient simulation? I haven‘t found explicit material describing the principle of time domain simulation with EM model. Could you show me some? Thanks.
Best regards
Wenrun

 

[volker@muehlhaus]

Yes, S-parameters (response in frequency domain) cannot be used directly in time domain simulation, so it must be converted behind the scenes. Quality of that conversion depends on bandwidth of the data.

I have searched for a basic document on that topic, maybe this gives an idea:

https://pdfs.semanticscholar.org/3d18/bec0ab18f72bcef1ca5f98a0cd1c11f8076d.pdf

In my EM support work, using S-Parameters in time domain simulators was always an issue for customers. Tools like Broadband SPICE were usually the solution, to extract an equivalent RLC model that is well behaved (causal and passive) and simulates efficiently in time domain simulators. Using the S-Parameter data directly with the simulator's built-in conversion was not always satisfactory.