Which topology to choose for 7-11GHz microstrip bandpass filter

[prutser]

I want to design a microstrip bandpass filter from approx. 7-11GHz.
I have a few constraints:

- Want to use free tools
- PCB design in KiCad
- PCB JLPCB PTFE (still a reasonable price compared to Rogers)

At this moment I have more questions than answers ?
Which topology is best suitable for this and why ?

Are there any open source (EM simulation) tools which have a reasonable integration with KiCad ?
I was looking into Qucs using QucsRflayout, but the latter seems to have limited support for certain microstrip options.

Any help or suggestions would be appreciated.

[nctnico]

I'd use Sonnet Lite (a free to use but limited EM field solver package) to simulate the layout of the filter and then implement using Kicad. Keep in mind that etching typically has large tolerances so you better not use traces that are narrower than 3 times the minimum width the PCB manufacturer supports. You might need to go back & forth a bit to find the most optimal dielectric height.

[RoV]

Give also a look at https://rf-tools.com/

[rf-fil]

The better simulator you use, the better the accuracy. Maybe you have a friend that lets you use ADS or Microwave Office? EM solver is ideal, but you can also do it with RF simulator which has analytical coupled transmission line models, like baseline ADS without the EM licence.

I would look into hairpin or coupled line topologies. They don't rely on one side of the resonators being grounded, which would make it difficult to design.

Rogers is not that expensive - I've just quoted hundreds of small PCBs for work, for < 1k USD, from one of the other big PCB places. In any case, pick a material that your PCB fab has in stock.

Don't use solder mask, of coursel, and use thick via stitching on your ground pours, less than 1/10th of wavelength if possible.

Allow yourself time for a few iterations to refine the design.

[prutser]

@nctnico
Thanks for your reply. I was also looking at Sonnet lite. How is your workflow for filter design ? e.g. Import/export file formats, or drawing in each application by hand ? As far as I know Sonnet lite only supports DXF import/export.

[prutser]

@rov
rf-tools for Marki was one of the first online calculators for Microstrip filter I noticed. When I tried it, it seems not to work. When selecting Microstrip filter design, it shows an example interdigital
bandpass filter from 1900-2100 MHz. But when changing something (such as upper corner to 2200MHz) after pushing calculate it turns into 'loading' which never finishes. 
Also tried other micostrip filters but it almost always hangs when starting the calculation. 

[RoV]

I have just replicated your problem with Edge. I normally use Firefox and the tool works fine with it: I have tried several parameter changes and compute commands. I don't have other browsers installed.

[prutser]

Thanks,
I had tried with Chrome and Edge, both same result (but they use the same backend  )
Just installed Firefox, first time same behavior, but all successive attempts were ok.

[nctnico]

@nctnico
Thanks for your reply. I was also looking at Sonnet lite. How is your workflow for filter design ? e.g. Import/export file formats, or drawing in each application by hand ? As far as I know Sonnet lite only supports DXF import/export.

Typically I use some kind of calculator to determine the layout and then redraw in Sonnet Lite. You'll need to choose the grid setting wisely because there are a limited number of 'cells' you can calculate using the free version. A hairpin filter is a good start for a bandpass filter.

[Gerhard_dk4xp]

> I was looking into Qucs using QucsRflayout, but the latter seems to have limited support for certain microstrip options.

<   http://qucsstudio.de/de/herunterladen/    >

qucsstudio is not the same as qucs. No more open, but still free.
From the same author who started it, and it still gets updates now & then.

from the download readme:

The source code of ICARUS, GHDL, KiCAD, MinGW and Octave can be downloaded from the official homepages. QucsStudio uses the unchanged binaries.

[ That would imply that it has a link to kicad  ]

ADMS (VerilogA converter from the ADMS project)
QucsConv (file format converter)
QucsFilter (filter synthesis tool)
technical documentation (circuit simulation, models etc.)

The web site can be switched to English.

I have installed a version 2.x.y.z but methinks before
updating anything I need to clean up my virtual machines
and the Linux they run on. Smells like work.

7 to 11 GHz seems a bit wide for hair pins. A highpass-
lowpass combination would be more flexible.


Gerhard

[BigBoss]

If this filter will be designed for a professional purpose, I'd use a professional filter designer program like Ansys NuHertz Filter Solutions.
Free Trial is available. But I don't know Filter Solutions has a capability to export the layout DXF format.
Also, you can simulate this filter in Sonnet Lite (free with limited features) by exporting the layout in DXF format.
Sonnet lite permits to use up to 6 ports and memory restriction 64MB.
https://www.sonnetsoftware.com/products/lite/download.html

[mick_lee]

For linear modelling Qucs should have you covered (not the Ubuntu snap install - which seems to be spice only!)

If you want to perform EM analysis on the structure - you can use sonnet lite.  Some other open source EM softwares are MEEP (although this is a bit more aimed at the photonic end of the EM spectrum, I believe) or openEMS.  Both of these need a bit of effort - they are certainly NOT plug and play or for the faint hearted.

https://meep.readthedocs.io/en/latest/

https://www.openems.de/

There's an example filter quoted here:
https://docs.openems.de/python/openEMS/Tutorials/index.html

Failing that, you could have a go a writing your own! No seriously, the finite difference time domain (FDTD) technique is conceptually quite simple and is reasonable simple to code up.  Have a look here at Sullivan's text https://ieeexplore.ieee.org/book/5263542

Have fun and let us know how you get on!

[mick_lee]

"7 to 11 GHz seems a bit wide for hair pins. A highpass-
lowpass combination would be more flexible."

Nope - you can do octave bandwidth filters using coupled lines.  It does require the introduction of extra elements and the use of Kuroda identities - but it is possible.

Here's the paper: https://drive.google.com/file/d/1HT5qHt0q4k2b4w2G3o_q_TbpJNRsJl6C/view?usp=drive_link

[mick_lee]

Here's the book commonly referred to as "the Bible" by microwave engineers - Matthei, Young and Jones!  This will tell you everything you need to know to design a half decent filter - it's in two parts

https://www.microwaves101.com/uploads/MYJ-part-1.pdf
https://www.microwaves101.com/uploads/MYJ-part-2.pdf

[Weston]

I have been working on a project modeling transmission line transformers and been looking at free EMS simulators.

I started with Sonnet Lite. Its a good tool, but the free version is a bit limiting and I dislike learning tools which cost more $$$ than I would be willing to spend. For a basic structure it should work though.

Recently I have moved to openEMS. The documentation is a bit rough, but it's a free and open source tool that is quite powerful. Most EMS simulation software uses the same FDTD technique for simulation, so the main difference between the free and commercial software is GUI and some possible performance improvements. I only previously did some rather basic stuff in Sonnet Lite, but after a few days I am successfully generating, simulating, and optimizing, rather complex structures in openEMS.

So far I like openEMS and am getting pretty good results. I recommend using it. Here is a project using openEMS and kicad to make hairpin filters that you might find useful: https://charleslabs.fr/en/project-Hairpin+filter+design
Someone also has a cool project on github using openEMS to model a SMA to microstrip transition: https://github.com/toammann/Multilayer_SMA2Microstrip

Most knowledge about how to set up a problem for simulation, how to effective mesh the structure for simulation, and boundary conditions, should broadly transfer between software.

I am unsure of the current state, but some level operability between kicad and openEMS exists (there is also an import flow from FreeCAD). However, due to the peculiars of setting up simulation, you are probably better off generating the structure for simulation and the PCB layout independently. An approach I have seen is to have a high level description for the structure and one script to generate the simulation mesh and another script to generate the PCB layout.

[BigBoss]

Another topology might be "Edge Coupled Filter" that is frequently used in many MW systems. I have demonstrated a samPle over Rogers 5880 and simulated in AXIEM.
The results are pretty promising.

[mick_lee]

Here's the problem - wideband filters like this need quite high coupling (from your derived K's and Q's).  This drives us to coupled lines with unrealisable odd/even mode impedance ratios (you have to work to a minimum gap of around 0.004" when using standard wet etch technology).  The coupled line designs submitted here, whilst well intentioned, are unfortunately NOT manufacturable.

There is a solution - use the technique in the paper by Minnis that I put up earlier - this uses Kuroda transformation to make realisable elements.

[Gerhard_dk4xp]

"7 to 11 GHz seems a bit wide for hair pins. A highpass-
lowpass combination would be more flexible."

Nope - you can do octave bandwidth filters using coupled lines.  It does require the introduction of extra elements and the use of Kuroda identities - but it is possible.

Here's the paper: https://drive.google.com/file/d/1HT5qHt0q4k2b4w2G3o_q_TbpJNRsJl6C/view?usp=drive_link

Nope. At 3*fc there starts a new passband and it is typically just 10 dB weaker.
And with the wide bandwidth and the tight coupling required to achive it
the pass bands won't  be far away. OK, you may switch on tunnel vision and admire
the lowest pass band.
Gerhard