Highpass microstrip filter

Hello everyone,

I am learning the quirks (and headaches) that microstrip can be for the RF engineer.
I am currently making a highpass filter but I can't seem to be doing it right.
My guess is that although simulation looks wonderful,
when I actually build the PCB is nothing of what I calculated.

So, to make this thread more productive I include my mwo file so anyone who's an expert on both EM simulation and microstrip
could take a look at it and tell me if I am going in the right direction or if i have to start from scratch.

I appreciate the time you have taken to read this thread.

Here's the file:

Could it be your dielectric assumptions were incorrect? What did you use? ceramic? Polyamid? Etched Nitride?

Etch process is not ideal unless SS foil only with no dielectric. Did you make an impedance coupon for Ω test purposes? That is essential to verify and is often put in the perimeter scrap area prior to depanelization. Some vendors can test, calibrated and ensure perm. is correct with coupon test and adjust. https://microlab.berkeley.edu/labmanual/chap1/JMEMSEtchRates2(2003).pdf

(One of...) the worst things about microwave simulation is how happily the simulator/optimizier will converge to a coupled line spacing value such as 0.1378543392 mm without conveying the hypersensitivity of the design to this parameter to the designer!
When you've completed your design, either do a parameter sweep/monte carlo simulation of the critical dimensions of overall structure or (preferably) a sensitivity analysis to check that you're not relying on an un-manufacturable mechanical dimension/tolerance for your network to function. I've found coupled resonant architectures are particularly fiddly in this regard.

Echoing SunnySkyGuy - dielectric variations have MASSIVE influences on the actual performance (read: tuning/matching/resonance) of any physical network. One of the main attributes (other than loss tangent) that distinguishes a "high frequency" PCB material from a bog standard substrate (such as fibreglass FR-4) is the manufacturer's control of the relative permittivity (also called 'Dk' in PCB parlance). Sub-% variation (seen in Rogers/Arlon/Taconic etc's products) is considered good, while FR-4 exhibits nearly a factor of 2 spread in Dk depending on composition and manufacturer.

Looking to your schematic I see that you are using big series capacitors (C1=50pF, C2=150pF, C3=150pF), and in the simulator you use ideal capacitors with no SRF and losses vs frequency defined.
If you don't define these parameters of the capacitors (in the simulator), when you build the circuit you get totally different results.

I used the RO4003 substrate for the design.
This is the second time I build it.
the first time I built it, I got a little bit more bandwidth.
With this one I got little close to nothing.

Which element is the one that you use on the circuit instead of the ideal capacitor in MWO?

Use CHIPCAP, or the best would be to get from the manufacturer the S-parameters of the capacitors, and use them in the simulator.
For Murata capacitors you can use their software to download the S-parameters:

https://www.murata.com/products/design_support/mcsil/index.html

For an accurate description of SMD components, there is a great component library from Modelithics which can be used with MWO, ADS and other simulators. This also includes the parasitics from pads on different substrates.

https://www.modelithics.com/products.asp?a=view&id=1

Can you tell us the Self resonating freq of all your RF parts? (SRF ought to be >>2x operating f)

Ok.. so I followed the advice of vfone (thank you very much!).
I changed it to CHIPCAP and I was able to see almost the same type of response that I am getting from the board.
So, I worked the traces a little bit more, worked the capacitor value and got a more realistic approach to what I expect to see in the analyzer.

I am using DLI capacitors. I got the datasheet for the ones I use and I applied the right Q, SRF of the component and that made a H-U-G-E difference.
I attach a new revised version of the filter and I think this one has a higher probability of being the right one.

Also, the size makes sense regarding the wavelength in the traces.

Please let me know what you think.
Am I going in the right direction or I am just building paperweights? hehehe......

Thank you.

View attachment highpass-real.zip

Yes, now looks much better. DLI makes good quality microwave capacitors. I think they provide also S2P parameters for their capacitors. Would be better if you can use this in your simulation.
For the new 0.3pF capacitors that you are using now, also would be better to use in the simulator the pads parasitics, as was stated above.

SRF, Q , ESR, & loss tangent, are all critical non-ideal characterisitics in all RF passive parts. Also electro-mechanical properties such piezo vs vibration, temp, voltage and stress.

S parms are also essentiall vs f for simulations and choosing a qualified second source part.

When I was Ops Mgr for microwave R&D, we created a master parts catalog with multiple qualified P/N's for every part. It is the designer's responsibility to ensure only these part numbers are used in BOM's for ordering purposes and never specify a generic part unless it is truly non-critical.

I wonder how your filter performs with worst case or monte carlo simulations. :?:

S parameters of the capacitors. Definitely the right way to go.
I noticed the website may have them. I will look for them. I made this simulation first so you could get and idea of where I stand.

Now, forgive my ignorance, but could you explain to me what a Monte carlo simulation is?

Thank you very much.

vfone, whe you say calculate the parasitics of the pads, you are talking about leads that are used to solder the capacitor itself?
Or are you suggesting I make sure all 3 capacitors have a pad to solder the entire body on top of the trace?

I appreciate your help with this. I am in the process of improving my knowledge about transmission lines and others
in order to become more experienced with RF. Thank you very much!

- - - Updated - - -

Ok. So, I simulated with the S Parameters of the 0.3 pF from DLI, horizontal assembly.
I think this looks pretty realistic as far as response.
I don't expect much from the VSWR to look all nice and pretty, but I do expect it to be under the -15 dB (hopefully....).

I think this simulation is closer to reality than the ones I've had before.
So, for future simulations, let see if I got this right.
The higher the frequency, the higher the probability that I should use the s parameters instead of the CHIPCAP model, correct?
Or does this rule of thumb apply to all possible scenarios?

Thank you very much to all of you who have kindly provided feedback on my situation.

Here's the attached with the s-parameter simulation.

Thank you very much!

That's fine. Luckily DLI provides S-parameters of the caps up to 40GHz. I think these parameters contain all the component parasitics including pads, so at the moment you can let as it is.
CHIPCAP should be close, but S-parameters of the caps is much better choice (if they are available).

vfone said:

That's fine. Luckily DLI provides S-parameters of the caps up to 40GHz. I think these parameters contain all the component parasitics including pads, so at the moment you can let as it is.

Click to expand...

The pad parasitics depend on the substrate. That's why it is better NOT to include them in the component manufacturer's S-parameters. If they are included, the component manufacturer has hopefully defined what measurement conditions where used (substrate and exact location of reference planes). The better approach is the substrate-aware component library by Modelithics.

Of course DLI (Dielectric Labs) provide all the information regarding substrate.
Using the component model directly from the manufacturer, most of the time is better than using from a third-party company.
Anyway, before existing all these component models RF engineers use successfully just SRF and Q, usually given at low frequency.

vfone said:

Of course DLI (Dielectric Labs) provide all the information regarding substrate.

Click to expand...

Excellent!

vfone said:

Using the component model directly from the manufacturer, most of the time is better than using from a third-party company.

Click to expand...

From my experience, many other manufacturer don't provide enough detail, and often the data is not good.
Modelithics are doing precise and well defined measurements. EDA software companies and also some component manufacturer now use that data, because it is accurate.

Monte Carlo methods (or Monte Carlo experiments) are a class of computational algorithms that rely on repeated random sampling to compute their results. Monte Carlo methods are often used in computer simulations of physical and mathematical systems. These methods are most suited to calculation by a computer and tend to be used when it is infeasible to compute an exact result with a deterministic algorithm. This method is also used to complement theoretical derivations.

This simulation is not simply the worst case of all input variables but a random combination to determine the performance loss in the real world using the specified uncertainty of assumptions for constant values. There are many.

Taguchi method is another popular way of detemining design goodness by determining the Cpk or ratio of upper and lower control limits to the 3 sigma variation, using Design of Experirments, one can reduce the number of computations or tested experiments from millions of permutations to perhaps 10 experiments with 10 input variables by changing the tolerance of many variables simultaneously. A theoretical nominal result is useless except for knowing one sample works. THis is what experts do for high volume critical production in Japan and many American factories.

Hello again.
Just wanted to say thank you for the advice and suggestions.
I did a new simulation with the S parameters of the DLIs I was going to use and that had a significant impact because unlike
the previous boards, this one was dead on the frequency. VSWR has to to be worked out, but the resonance of the circuit is exactly where
I wanted. Thank you very much!!