RF/Microwave experts: Identify these mystery PCB structures

[macboy]

I came across these two mystery PCBs with strange structures on them, and I am curious what they are. My suspicion is that they are filters of some kind probably for Wi-Fi band(s), but I'd like to hear the opinion of those in the know. They have U.FL connectors on them (spaced 1.5" or 38 mm for reference), they are about 0.6 mm thick, and the back of each board is a ground plane.

[G0HZU]

I think that the 2 little mushroom shapes in the top PCB will form a 2 section notch filter. They will typically be spaced apart by 90 degrees at the notch frequency. I'd expect to see quite a deep notch but not a very sharp one centred somewhere around 2.5-3GHz  assuming the PCB has a dielectric constant somewhere about 4.

The bottom headphone shaped filter looks like a very compact elliptic lowpass filter. The two big squares that are very close together will provide series and shunt capacitance and the bypass trace will give some inductance. So the circuit will look like a 3rd order elliptic lowpass filter with a cutoff somewhere around 2.5GHz at a guess. The advantages with this approach include size reduction and also it may suffer less with re entry modes up at 7-10GHz or so when compared with more conventional printed elliptic LPFs. But it will still need some kind of roofing filter to prevent re entry modes up at higher stopband frequencies.

[German_EE]

We really need either Dave or Shariar to do a Fundamentals Friday on lumped element filters. I've been asking for it for ages but neither of them have stepped up and volunteered.

[Wirehead]

The Rigol RF generator teardown mentions the subject, take a look

[tggzzz]

You might like to speed read
http://ntuemc.tw/index.php?node=microwave-filter
https://www.jlab.org/accel/eecad/pdf/050rfdesign.pdf

[T3sl4co1l]

We really need either Dave or Shariar to do a Fundamentals Friday on lumped element filters. I've been asking for it for ages but neither of them have stepped up and volunteered.

What would an LC Fundamentals consist of?

You can slap a few parts together and get the desired asymptotic response, but choosing the right values in the transition band is something people are still working out!

Applications range from power distribution networks (just one place people rarely realize its presence!) to digital signaling, RF cavity filters and everything inbetween.

Theory is dense, ranging from the earliest m-derived and constant-k filters (which do okay, and introduce important concepts in analysis), to modern network synthesis using standardized polynomial roots.  Stuff that isn't even taught in school [anymore?].

Even just introducing AC analysis could be troublesome; it's hard to do AC steady state without complex numbers, which aren't nearly as complex or imaginary, or mysterious or enigmatic, as their name suggests.  Schools spend an entire class on this, and kids still don't get it..

So I'm not really sure where to start.  If it's just to cover a few basic sticking points, everyone's going to have their own different questions that need answering.

If no one will make a fundamentals video... perhaps these are among the reasons, and then, perhaps we can design one here?

Tim

[G0HZU]

If no one will make a fundamentals video... perhaps these are among the reasons

I think the reality is that there are very few (genuinely) experienced RF design engineers doing video blogs. There are quite a few bloggers who may appear to be experienced in this field but I don't think this is the case.

[macboy]

Thanks G0HZU for your analysis. This fits my suspicions.

[KJDS]

I suspect that to be able to impart enough knowledge to allow someone to design that sort of filter would require a twenty hour long university course, starting with fundamental filter design and putting all the foundations in place for lumped element design, before going on to add elliptic functions, and then classic distributed forms and more complex realizations and analysis using EM tools.

Here's a possible syllabus that assumes a student has a sound grasp of complex numbers and maths and a sound grasp of basic circuit theory and components.

301  Lowpass design and denormalization
302  Butterworth, Chebychev and Bessel
303 Affect of lossy components
304 Bandpass design
305 More bandpass design
306 Affect of lossy components
307 Affects of SRF limitation of L and C components
308 Impedance transformers
309 Quasi-elliptic elements to improve stopband
310 Richardsons transformations
311 Coupled transmission lines
312 Basic coaxial filters
313 more complex coaxial filters
314 Coupled microstrip
315 Basic microstrip filters, coupled line, hairpin etc
316 Complex microstrip filters, interdigital etc
317 A worked example of Denigs approach to interdigital design
318 Using a planar EM solver to analyze a microstrip filter
319 Modern synthesis approaches
320 Using modern synthesis approaches to optimize a quasi-elliptic bandpass

. ...and if anyone fancy's using that to make a video on how to design the sort of filter shown in the original post they are more than welcome to use it, however it may well take a while.

[nctnico]

We really need either Dave or Shariar to do a Fundamentals Friday on lumped element filters. I've been asking for it for ages but neither of them have stepped up and volunteered.

IMHO one of the problems is that lumped element filters are typically constructed using values from precalculated tables. You can actually buy a book with precalculated filtering tables for various filters and different pass band/stop band ripple figures. However in the current day and age it should not be hard to write a computer program which calculated the filter values on the fly. It does take quite a bit of math though but I think a Bachelor's degree should be enough to tackle it.

[KJDS]

We really need either Dave or Shariar to do a Fundamentals Friday on lumped element filters. I've been asking for it for ages but neither of them have stepped up and volunteered.

IMHO one of the problems is that lumped element filters are typically constructed using values from precalculated tables. You can actually buy a book with precalculated filtering tables for various filters and different pass band/stop band ripple figures. However in the current day and age it should not be hard to write a computer program which calculated the filter values on the fly. It does take quite a bit of math though but I think a Bachelor's degree should be enough to tackle it.

I got bored at a contract I was working on many years ago and put all the equations into a spreadsheet which calculates the values for any chebychev lumped element filter. It took a week or two to get rid of all the bugs and make it look pretty.

[G0HZU]

The lazy alternative is to use an existing filter design program coupled with a linear simulator

 The first one I used was =Filter= produced by Circuit Busters (later known as Eagleware) over 25 years ago. This was relatively expensive and ran with a (feeble) dongle key for security.
I liked it a lot and the filter could be rapidly simulated using their 'SuperStar' linear simulator program. Over the years I've designed hundreds of filters at work using Eagleware software

But loads of freebie alternatives appeared from the early 1990s. I tried out most of them but the Eagleware suite was the best.

In the early 1990s Eagleware added the ability to design distributed filters using =M/Filter= and shortly after this they introduced their first EM simulator called EMPower. This helped reduce the design time of filters a lot back in those days.

But today, the internet is awash with free (and not so free) filter design programs and you can even download a free student version of Sonnet EM to simulate the layout. I would expect that the free version of Sonnet could manage to simulate the PCB structures shown in this thread

The classic old RFSim99 program is also a useful/free linear simulator that has quite a few lumped filter topologies built into the synthesis part of the program.

[G0HZU]

I should point out that the printed filter types shown in this thread are a bit different in that they aren't supported by any of the filter synthesis programs that I've used.

So you do need to have some knowledge of the basic fundamentals to design anything that isn't mainstream. To give an example, I did a fair bit of research many years ago into filter design using defective groundplane technology and I developed some novel filter designs using weird cutout shapes in the groundplane that won't be in any theory books

[nctnico]

So you do need to have some background knowledge to design anything that isn't mainstream. To give an example, I did a fair bit of research many years ago into filter design using defective groundplane technology and I developed some novel filter designs using weird cutout shapes in the groundplane that won't be in any theory books

Just curious: where these bandpass filters?

[G0HZU]

So you do need to have some background knowledge to design anything that isn't mainstream. To give an example, I did a fair bit of research many years ago into filter design using defective groundplane technology and I developed some novel filter designs using weird cutout shapes in the groundplane that won't be in any theory books

Just curious: where these bandpass filters?

Not sure what you mean but you can google for:

defected ground structure filter

To see how this technology has taken off over the last 10 years or so. 

Some of my novel filter designs have been manufactured and used in reasonable volume at my place of work. Eg for Tx and Rx stuff running up at 6GHz or so

[T3sl4co1l]

It's not really so mysterious; a cutout in the ground plane is the image of a (thinner?) trace on the top side.

For example, a ground plane slot is the image of a dipole antenna, and behaves much the same.  I think the impedance is reciprocal (symmetrical around Zo/2?) as well?

For both modeling and synthesis, the added degrees of freedom (trace width and length, and ground slot width and length, and also the coupling of traces crossing regions of open ground) make the problem much harder to solve, so it's kind of a curse, at least until analysis and algorithms advance enough to harness such things.

Tim

[DanielS]

Just curious: where these bandpass filters?

In OP's image? At microwave frequencies, traces can act as inductances, resistances and capacitors depending on shape and what other nearby structures they can couple with through magnetic and electric fields. The "mushrooms" act as high-Q LC circuits (fat traces are inductive and if there is a ground plane at the end of the mushrooms, that makes them capacitive) and a high-Q LC circuit between a signal and ground has its lowest impedance at resonance, which would 'notch' the signal near the LC's resonant frequency.

[G0HZU]

It's not really so mysterious; a cutout in the ground plane is the image of a (thinner?) trace on the top side.

For example, a ground plane slot is the image of a dipole antenna, and behaves much the same.  I think the impedance is reciprocal (symmetrical around Zo/2?) as well?

For both modeling and synthesis, the added degrees of freedom (trace width and length, and ground slot width and length, and also the coupling of traces crossing regions of open ground) make the problem much harder to solve, so it's kind of a curse, at least until analysis and algorithms advance enough to harness such things.

Tim

It has been possible to model DGS stuff like this for many years using tools like Sonnet EM. I've used Sonnet to model PCBs using multiple layers with defected grounds in various locations on the PCB since we bought Sonnet at work >10yrs ago

The simulation typically takes a long time (usually many minutes, even hours) so it isn't a real time simulation but there are ways to speed up the simulation if certain assumptions about the overall structure are made. So complex filters that would require many gigabytes of memory to simulate (and many hours) can be simulated in a few minutes with a bit of experience.

Sonnet EM is a fabulous design tool because it allows you to draw and experiment with filter structures in a few minutes. If it is used as an external simulation engine with something like Eagleware/Agilent Genesys you get a very powerful design suite indeed because you can include SMD components both active and passive in the simulation very quickly and easily

[KJDS]

It's not really so mysterious; a cutout in the ground plane is the image of a (thinner?) trace on the top side.

For example, a ground plane slot is the image of a dipole antenna, and behaves much the same.  I think the impedance is reciprocal (symmetrical around Zo/2?) as well?

For both modeling and synthesis, the added degrees of freedom (trace width and length, and ground slot width and length, and also the coupling of traces crossing regions of open ground) make the problem much harder to solve, so it's kind of a curse, at least until analysis and algorithms advance enough to harness such things.

Tim

It has been possible to model DGS stuff like this for many years using tools like Sonnet EM. I've used Sonnet to model PCBs using multiple layers with defected grounds in various locations on the PCB since we bought Sonnet at work >10yrs ago

The simulation typically takes a long time (usually many minutes, even hours) so it isn't a real time simulation but there are ways to speed up the simulation if certain assumptions about the overall structure are made. So complex filters that would require many gigabytes of memory to simulate (and many hours) can be simulated in a few minutes with a bit of experience.

Sonnet EM is a fabulous design tool because it allows you to draw and experiment with filter structures in a few minutes. If it is used as an external simulation engine with something like Eagleware/Agilent Genesys you get a very powerful design suite indeed because you can include SMD components both active and passive in the simulation very quickly and easily

When it works, Agilent Momentum within ADS is brilliant. I've done many matching circuits for power amps using it, no problem at all getting good accurate results matching to the typical 0.5 ohm input impedance of a big power transistor, but it's harder work than it should be as it's so so bug ridden.

[Deathwish]

Y'all mean they aint some fancy Chinese Mahjong tiles ! 

[G0HZU]

When it works, Agilent Momentum within ADS is brilliant. I've done many matching circuits for power amps using it, no problem at all getting good accurate results matching to the typical 0.5 ohm input impedance of a big power transistor, but it's harder work than it should be as it's so so bug ridden.

At work I can use ADS Momentum within Agilent Genesys and it definitely works well here 'when it works' as you say. It is much quicker to arrange a simulation with Momentum and it is usually quicker to run for this reason too.

But I still like Sonnet because it has helped me understand (and correct) subtle layout issues so many times.

eg it has been able to accurately model subtle resonances in the stopbands of cascaded filters that can be seen on a VNA.  i.e. the classic wobbly stuff that appears in the stopbands when the lid is fitted over the filters. This can be a showstopper if the filter cascade has to meet a mask spec and there is an unwanted bump somewhere deep down in the stopbands.

If I had to criticise Sonnet, it would be that the user interface still looks like something from the 1990s. However, the latest version is (finally) supposed to be a lot better in this respect. I haven't upgraded yet at work so I can't comment much on the latest version of Sonnet.

[nctnico]

I did some simulations in Sonnet recently and editing the layers has been improved indeed.

[KJDS]

Y'all mean they aint some fancy Chinese Mahjong tiles ! 

I went through a stage of making PA impedance matching networks look like space invaders.

[T3sl4co1l]

It has been possible to model DGS stuff like this for many years using tools like Sonnet EM. I've used Sonnet to model PCBs using multiple layers with defected grounds in various locations on the PCB since we bought Sonnet at work >10yrs ago

The simulation typically takes a long time (usually many minutes, even hours) so it isn't a real time simulation but there are ways to speed up the simulation if certain assumptions about the overall structure are made. So complex filters that would require many gigabytes of memory to simulate (and many hours) can be simulated in a few minutes with a bit of experience.

Yeah, if it's doing full 3D E&M, then layers aren't going to matter.  But a careless model will result in tons of wasted CPU time, as has always been the case with modeling.

In terms of generating structures, I don't know that there are any tools where you can say, I want this impedance and frequency response here, and another there, and have it solve the problem for you.

I've seen some tools that vary component values for optimal parameters, tolerances, etc., though that's with lumped elements.  Don't know if the same exists for printed format (or others, e.g., cavity, waveguide, etc.).

If that's out there, too, then fantastic can't wait until it trickles down to regular PCB designer availability

Tim

[G0HZU]

've seen some tools that vary component values for optimal parameters, tolerances, etc., though that's with lumped elements.  Don't know if the same exists for printed format (or others, e.g., cavity, waveguide, etc.).

Eagleware produced =M/FILTER=  back in the early 1990s that allowed several common printed filters to be designed. eg there were several lowpass, lots of bandpass and some bandstop topologies.

It proved invaluable at my place of work but sadly, Eagleware (and later Agilent) never really developed it further in their Genesys suite. So the latest version of this SW has pretty much the same range/capability for printed filter design as the old version. The later versions have a useful optimiser function and I think the design algorithms are a bit better but essentially it's the same thing.

You enter your chosen substrate (usually from its library) and then it asks for filter order and enter the passband response/ripple and you also get control over what impedance traces it uses in parts of the filter.

See below for a quick screenshot of a combline BPF design using the 2004 version of Genesys. I can probably find an old version of Eagleware from around 1995 that would look very similar.

It is a very slick process and any user edits and resimulations are essentially in real time even if you change the order of the filter. It self generates the PCB layout and sets up the simulation in a very fluid manner.

It only takes about 1 minute to start the SW suite and design that combline filter to the stage shown in the image below if you are experienced in using this program. However, in reality a certain amount of manual tweaking/experience is needed beyond this. Eg the default via hole placement is poor wrt the caps (as you can see below) and usually the tap point isn't ideal for best return loss. Also, it needs to be simulated in a decent EM simulator and this would be the next stage. i.e. I'd add a Sonnet layout simulation in the workspace tree and run an EM simulation. The height of any screening lid over the filter will affect the passband and the Sonnet EM simulator can predict this effect very accurately.

In order to get really precise results I would manually tweak the length of the combline fingers to optimise the design for a single value of ATC capacitor and this takes some experience.

But because I've been designing filters like this for over 20 years I can confidently design a combline filter and go straight to an expensive and fully populated Rogers PCB without even prototyping the filter. But usually I would mill it on a T-Tech or LPKF machine and test it