Parallel coupled line filter and high insertion loss

[paszczakojad]

Hi,

I designed a few parallel coupled line filters (at 2.1 GHz and 10 GHz) and after manufacturing they have very high insertion loss, much larger than in simulations. Simulation in HFSS gave 2 dB loss, while the real filter has 7-8 dB loss!

S11 is quite good - between 15 and 20 dB, so I don't think it's caused by connector soldering or other reflections. I thought it's because of radiation or substrate loss, but I'm using Arlon 25FR with loss tangent of 0.0035 and whole filter is shielded (top cover is 5 mm above the substrate).

I noticed that simulations using ADS Momentum gave S21 closer to what I measured (4.5-5dB) and I was unable to optimize it to perform better. But I thought HFSS would be more accurate.

I found an article (**broken link removed**) about increasing coupling between coupled lines by cutting out ground plane under the coupling sections, but that doesn't change much in the HFSS simulation (because the S21 is already good there).

I attached my HFSS project - could you take a look at it? The 'cutouts' of the enclosure are meant to reduce design complexity and prevent box resonances (the real-life enclosure looks more or less the same). There is also mesh seeding used but the result is the same with regular adaptive meshing (but the simulations takes much longer to finish).

I used 0.254mm thick Arlon25FR substrate (because in other projects 50 ohm line width was narrow enough to connect to MMICs without any tapers) - maybe this thickness contributes to the loss? (the ground plane is closer to microstrips than coupled microstrips). But in such case why HFSS didn't show that? And in case of other structures - like stepped-impedance filter or directional coupler - measurements match HFSS simulations.

P.

 

[volker_muehlhaus]

But I thought HFSS would be more accurate.

Why do you think so?
Planar MoM solvers like Momentum or Sonnet are optimized for planar problems, and solve the complete EM problem.

I never understand why people try to solve planar problems with 3D volume meshing tools. The planar solvers have excellent loss models for planar conductors, more accurate port calibration etc. You need a lot of experience and a lot of simulation time to get similar accuracy from a volume meshing tool.

 

[paszczakojad]

Why do you think so?

I _thought_ it would be In case of some designs HFSS was closer to reality than Momentum, but not this time. I think I took into consideration all requirements like boundaries, wave port dimensions, proper meshing, etc., but apparently something is wrong.

But anyway - why this filter has such insertion loss in the passband when simulated using Momentum? I copied that HFSS filter to Momentum (attached). The return loss isn't very good, because it was optimized using HFSS and Momentum results are a little bit different, but that shouldn't cause such high loss.

P.

 

[volker_muehlhaus]

What is your metalization thickness?

 

[paszczakojad]

What is your metalization thickness?

17 um. But in Momentum I'm simulating with flat layer (no thickness).

P.

 

[volker_muehlhaus]

But in Momentum I'm simulating with flat layer (no thickness).

Sure, but Momentum thin metal needs the thickness to map the correct surface impedance onto the conductor.
From what I have seen in your files, you have set 1µm copper thickness in Momentum !?

I exported your ADS data to Sonnet EM and simulated with Sonnet. With 17µm copper, the passband loss calculated by Sonnet is ~3dB (mid-band, with metal top cover at 5mm above the filter).

 

[paszczakojad]

Sure, but Momentum thin metal needs the thickness to map the correct surface impedance onto the conductor.
From what I have seen in your files, you have set 1µm copper thickness in Momentum !?

That was set as default. I thought Momentum treats 'sheet' as PEC. But with 17 um copper the insertion loss is the same (4.6 dB or more).

I exported your ADS data to Sonnet EM and simulated with Sonnet. With 17µm copper, the passband loss calculated by Sonnet is ~3dB (mid-band, with metal top cover at 5mm above the filter).

So Sonnet gives ~3dB, HFSS ~2dB, Momentum ~5dB, measurement is ~7dB. Let alone differences between tools, my primary concern is this real-life measurement. Where did the energy go and how can I improve that?

The 2.1 GHz filter has exactly the same problem. A stepped impedance filter has low insertion loss, so I suppose there's something wrong with these coupling sections.

P.

 

[volker_muehlhaus]

I thought Momentum treats 'sheet' as PEC.

There are multiple models for metalization, and this does make a difference in results.

So Sonnet gives ~3dB

... or 4dB when the metal is simulated with 2µm metal roughness (EDC copper) on the bottom side of the conductors.
However, I did not include possible edge roughness on the sides (from possibly bad etching).

I checked metal cover vs. free space top cover, and the results don't differ much, so I think that radiation is not the issue here.

The 2dB from HFSS must be wrong (insufficient mesh density or incorrect material properties), or you have compared different layouts, different reference planes etc.

Let alone differences between tools, my primary concern is this real-life measurement. Where did the energy go and how can I improve that?

Did you measure exactly what you have simulated? How did you measure? What calibration, TRL in microstrip plane or just coax so that coax to MSL adapters are included in the measurement?

There is a lot what can go wrong with measurement. My EM results are usually accurate within < 0.5dB insertion loss or better, if measured by an expert user who understands VNA calibration.