Ngspice-42 or 43 Cannot calculate minimum noise figure (NFmin) if S2P file is used

[Atif07Ali]

Environment

Ngspice 42 or 43
Qucs-s 24.3
I am trying to design an LNA at 170 GHz frequency using SG13G2. I tried to build the circuit with transmission lines, using the S2P files of transmission lines. I faced the problem of 'nan' in the minimum noise figure (NFmin). It seems that the simulator cannot calculate NFmin.
if I use lumped elements, the NFmin gives the right value. However, when I use S2P files as the transmission lines, the NFmin does not provide any value. It seems that the simulator cannot calculate noise figure when there are S2P files.

Expected Behavior

The behavior has been simulated using lumped elements only.

Actual Behavior

This is the behavior after using S2P files. The NFmin gives "nan" value.

Steps to Reproduce the Problem

1. Create the circuit like the schematic.
2. Provide any S2P file to the 2-port s-param.
3. Turn on the noise in "S parameter simulation" and simulate. plot nfmin.

[dwarning]

I am afraid that noise contributions by S2P files can't simulate by ngspice, even if the files contain additional noise data.

But NaN looks more as bug. To investigate that it would be easier you could provide the schematic and the four S2P files. Or alternative the spice4cucs.cir file would be helpful too.

[KrzysztofHerman]

@Atif07Ali @dwarning indeed I have reproduced this testcase a few minutes ago and got the same results, I mean NaN.
Here is the complete testcase (one have to update the fixed paths pointing to the S2P files)

[dwarning]

Slack channel is problematic - I believe I removed my membership.

[Atif07Ali]

@dwarning thanks for your reply. I am sharing the test case and the 4 s2p files in zip format. The S2P file names are marked in red within the schematic.
LNA.zip

[dwarning]

Thank you!
We have a problem in principle with simulating noise in s-parameter environment as we can see for resistors lower then system impedance:

After further investigations maybe we should open an ticket in ngspice forum.

[Atif07Ali]

@dwarning could you please mention the ngspice version?

[dwarning]

It is pre-master-44 branch (not actual anymore because new version 44 just released): ngspice-43+.

[Atif07Ali]

@dwarning There is a significant difference in results based on the version of ngspice. I am using both ngspice-42 and ngspice-43 for noise simulation. This is another problem. I think we should create an issue in ngspice forum. I am creating the issue. Please add your concerns in the reply.

[dwarning]

After solving the NaN problem in ngspice coming back to the initial problem of the missing noise results in the LNA stage by using s-parameter files.

I can only recommend converting s-parameter files in general by using vector fitting with the scikit-rf python package: scikit-rf. This has the additional advantage that AC, DC, Transient and Noise analysis are feasible.

Attached are the used python files, converted spice files and the qucs-s schematic.

@Atif07Ali Could you please check the plausibility of the noise and other results.

@KrzysztofHerman We should find a way to support the converting of s-parameter files in general in the design flow.

LNA_spar_noise_prj.zip

[Atif07Ali]

@dwarning thanks. Let me check it.

[Atif07Ali]

@dwarning, I was able to reproduce the figure as you suggested. I have simulated it with ngspice-42 and 43. The 42 version gives the nan value in some frequencies. I have also simulated the circuit with lumped elements (for matching). In that case also, 42 gives nan value in some frequencies. You can go through this previous slack conversation. https://open-source-silicon.slack.com/archives/C086734GNGH/p1734935566080149

[Atif07Ali]

Here if you see the nfmin values, there is a huge difference. ngspice-42 gives more realistic values (nfmin=4.48 @ 170G). The version 43 does not have the problem with nan. But 43 gives impractical values. I am actually confused which version should I use?

[dwarning]

You are mixing lot of things in your discussion which I am not able to follow.
Version 42 is not acceptable. We are now with 44.
Please reduce the diagrams.
Let us start with one kind of matching elements, then the other types.
Last schematic shows new elements, which are not available for me.
It's confusing, sorry.

[Atif07Ali]

I am sorry. Maybe I got your point. I was confused with the version at this time. Your suggestion worked perfectly for simulating s2p files.
Then the question is which version should I use? 43 gives impractical noise figure (nfmin) values. Should I try for 44 then?

[dwarning]

Try to install 44 - needs about 5-10 min. Then we are on same level.
The vector fit approach is new. We have to investigate the noise performance of the generated networks.
Your help is appreciated.
Thanks!

[Atif07Ali]

Thanks. Ngspice-44 give correct results of nfmin using lumped elements.
However, I have checked the vector fitting method in various circuits (consisting of s2p files).It does not provide accurate results.