Making low value air core inductors

[jujun]

Hello,

For the filter I am building, I need quite low inductor value (3nH) and high Q, for now with the help of this : https://hamwaves.com/antennas/inductance.html I found that I need only 1 turn ! Is it correct to do a low value air core inductor this way ?
Is there any practical tricks to build this?

The filter I am making is a high pass for 1,6 Ghz.

Thank you

J

[mmagin]

I think that it is more popular to use microstrip techniques today, but I think that wouldn't be a strange sight in UHF stuff from 50 years ago.

Considering you probably have to, at a minimum, design a PC board to build stuff at >1.6 GHz (though I think you'd still do okay with ordinary FR4 materials unless your needs are exotic / high power), I wonder if it would be preferable to realize the whole filter with microstrip techniques?

[coppercone2]

did you try varying wire diameter ?

if you don't like 1 turn, I managed to make a 6 turn using 36awg wire and a 0.25mm core

I can see it being made under microscope very carefully if you have a ceramic core to wind it carefully.. just paint the core with a very thin layer of wax before you wrap and heat it a bit to melt the thin layer of wax and it might work. Q is like, 14,

but there is kind of a trick, if you want broadband performance you wind it in a cone/bugle shape.
http://www.piconics.com/conical-inductors/

3nH is really really low though. I wonder if you can get a really good signal inductor by winding a tiny tiny cone out of tiny tiny wire. If I had my HFSS with me right now I would try to simulate it maybe, its super interesting.
40ghz performance.

They also make them really small too:
http://year2000.manufacturer.globalsources.com/si/6008848736032/pdtl/Choke-coil/1156304522/Choke-Coils.htm

Maybe you can outdo PCB material this way.

there is also this
https://www.digikey.com/products/en/inductors-coils-chokes/fixed-inductors/71?k=inductor&k=&pkeyword=inductor&pv2087=u0.1nH&pv2087=u0.2nH&pv2087=u0.3nH&pv2087=u0.33nH&pv2087=u0.39nH&pv2087=u0.4nH&pv2087=u0.47nH&pv2087=u0.5nH&pv2087=u0.56nH&pv2087=u0.6nH&pv2087=u0.68nH&pv2087=u0.7nH&pv2087=u0.8nH&pv2087=u0.82nH&pv2087=u0.9nH&pv2087=u1nH&pv2087=u1.1nH&pv2087=u1.2nH&pv2087=u1.3nH&pv2087=u1.4nH&pv2087=u1.5nH&pv2087=u1.6nH&pv2087=u1.65nH&pv2087=u1.7nH&pv2087=u1.8nH&pv2087=u1.9nH&pv2087=u2nH&pv2087=u2.1nH&pv2087=u2.2nH&pv2087=u2.3nH&pv2087=u2.4nH&pv2087=u2.5nH&pv2087=u2.55nH&pv2087=u2.6nH&pv2087=u2.7nH&pv2087=u2.8nH&pv2087=u2.9nH&pv2087=u3nH&pv2087=u3.1nH&pv2087=u3.2nH&pv2087=u3.3nH&pv2087=u3.32nH&pv2087=u3.4nH&pv2087=u3.5nH&pv2087=u3.6nH&pv2087=u3.7nH&pv2087=u3.8nH&pv2087=u3.85nH&pv2087=u4nH&pv2087=u4.1nH&pv2087=u4.2nH&pv2087=u4.3nH&pv2087=u4.4nH&pv2087=u4.5nH&pv2087=u4.55nH&pv2087=u4.6nH&pv2087=u4.7nH&pv2087=u4.8nH&pv2087=u4.9nH&pv2087=u5nH&sf=0&FV=ffe00047&quantity=&ColumnSort=0&page=1&pageSize=25

They looks like their made with a few turns in an acrylic resin, their like 3x3x3mm
https://www.digikey.com/product-detail/en/wurth-electronics-inc/744913050/732-7280-6-ND/5353422

#1

Inductance    5nH    
Tolerance    ±5%    
Current Rating    4A    
Current - Saturation    -    
Shielding    Unshielded    
DC Resistance (DCR)    1.8 mOhm Max    
Q @ Freq    140 @ 150MHz    
Frequency - Self Resonant    6.5GHz



#2
Inductance    3.85nH    
Tolerance    ±5%    
Current Rating    1.6A    
Current - Saturation    -    
Shielding    Unshielded    
DC Resistance (DCR)    6 mOhm Max    
Q @ Freq    100 @ 800MHz    
Frequency - Self Resonant    7.5GHz

[T3sl4co1l]

3nH is the stray inductance of a single 1206 chip component.

I think you've asked about this before, and it was noted that absurd values like this indicate a change in strategy is required.

Tim

[coppercone2]

what is your opinion of the inductors of 3nH that they have for sale?

I am actually curious in comparison to stripline and such

my simulations showed that most of the stripline filters have modeing that occurs at 2x the design frequency so you for sure need to follow them up with subsequent filters if you want a response similar to that of low frequency filter circuits (no one expects their filter to have a fit at 2x its design frequency, which is what I see with microwave stuff, same with waveguide/cavity filters, you need to find the bode plot to be sure of anything, the ones sold by pasternack seem to account for it).

IIRC the hairpin was kinda OK but compared to the bode plot i got from simulations to something like one of those modules from pasternack the response looked like dog shit. I mean the s21

[Wolfgang]

I may be old fashioned, but I am very sceptic about discrete components with values in the same order of magnitude as their parasitic inductances and capacitances.
If you can, try to make a stripline design with tunable elements.

[coppercone2]

any bored people  wanna look at those parts and measure some filters?

[jujun]

So I explain a little bit why this question, and what I did.

I really wanted to be able to record the data of a NOAA weather satellite on 1.7Ghz, the pass was in a few hours, so I needed a working filter in a few hours. So I choose the lumped components way instead of the microstrip or cavity to be able to have something very quick.

Because now I understand what cause losses in filters (thank you all ), I choose a 7th order high pass. I found that I needed 1pF and 1,5pF luckily I had them in my SMD C0G capacitor bag. Then I used the kicad calculation tool to have the size of the track at 50ohm. Then with my swiss army knife I made the pcb on scrap dual side FR4. Then for the coil I first tried with a longer coil, thinking that I could space the winding for tunning the filter. I ended to build several smaller coils. Each time I was able to see the filter band pass going higher in frequency. At the end I only had a little pice of wire of maybe 4mm. And the match was good, maybe -25dB RL, and in the band pass (of interest) the losses are only 0.8dB to 1,5dB.
For a quick little filter it's good
But I managed to miss the satellite pass because of the power supply of the LNA ...

[coppercone2]

 

good ' battle engineering' story though

[Gribo]

There are wire wound air core inductors at these values, but expect them to be in 0402 or smaller packages. The pads parasitic capacitance will ruin your day at these frequencies. 0402 pads have about 1pF of parasitic capacitance each.

[coppercone2]

with the inductors I linked above, are you supposed to budget with them or just use air construction?

[G0HZU]

It's only 1.7GHz! There's nothing 'absurd' about using (0603 or 0402 SMD) inductors of just a few nH up at 1.7GHz. For many years now the manufacturers have been releasing wideband s2p models of their SMD inductors and it's possible to simulate a PCB based design like this using an EM simulator such as Sonnet.

On a typical design you would probably have to select a 2.7nH SMD shunt inductor (in the place of a theoretical 3.3nH) to allow for the ground via hole inductances but a typical design would use multiple ground vias per shunt inductor and a very thin PCB substrate. To save frustration and time it really is best to simulate the PCB layout and use accurate models for the inductors. The best SMD lumped component models come from companies like Modelithics but you have to pay for these. For DIY, even a simple tiny loop of skinny wire would be OK for a 3nH inductor in a HPF design at 1.7GHz.

A lot depends on how high something like a 1.6GHz HPF needs to work in terms of passband response. It should be possible to get good passband performance to about 10GHz or so but at some point the response will degrade due to transmission line effects in the components etc. But Jujun only needed operation to 1.7GHz

However, I think it would have been worth trying for a 1.7GHz bandpass filter rather than a 1.6GHz highpass. The BPF could have been designed as a lumped design using SMD LC parts or it could have used microstrip for some of the design.  A reasonable target for a 3rd order lumped BPF would be 1dB insertion loss at 1.7GHz and maybe 200MHz bandwidth. It wouldn't give the steep LF rolloff of the 7th order HPF but it would give some useful rejection of 2.4GHz wifi etc.

[jujun]

However, I think it would have been worth trying for a 1.7GHz bandpass filter rather than a 1.6GHz highpass. The BPF could have been designed as a lumped design using SMD LC parts or it could have used microstrip for some of the design.  A reasonable target for a 3rd order lumped BPF would be 1dB insertion loss at 1.7GHz and maybe 200MHz bandwidth. It wouldn't give the steep LF rolloff of the 7th order HPF but it would give some useful rejection of 2.4GHz wifi etc.

Yes, I must definitely try to build a band pass. It would be a lot cleaner.
I choose an HPF also because there is a lot of strong signals that I need to filter that are under 1.7Ghz, typically FM broadcast and GSM, TV. Above 1.7Ghz there is less strong signals. (but still 3G)
Maybe the best would be to build a BPF with a stub to block the only one very strong FM broadcast signal I have.

[G0HZU]

There's lots of ways to design a fairly narrow BPF up at 1.7GHz but you need to be careful how you build and test it because it's easy to get confusing or false results due to a poor choice of filter topology, or poor component choice or poor layout or poor test gear. If you want to make a lumped BPF I'd recommend you do it on a conventional (thin) PCB with microstrip lines leading into the filter and use good quality SMD caps (suitable for use at 1.7GHz) and you can either wind the inductors yourself or use SMD. But SMD inductors won't be adjustable. The nice thing about doing it on a PCB is that the SMD caps will be easier to fit and so will the inductors. But the effect of the PCB tracks and pads will need to be modelled on a decent simulator like Sonnet. Otherwise, you would have to be prepared to do a fair bit of 'suck it and see' in terms of choosing the correct component values that offset the strays in the PCB layout. Any sloppy practices or poor component choices will result in poor results and frustration and anything you build will need to have a tight layout if it is to work well.

If the circuit and PCB layout is simulated properly beforehand it should work first time or at least be very close. Ignore all the comments above that say it isn't possible to use SMD inductors (of just a few nH) to make lumped filters up at 1.7GHz. These comments are from 'experts' who either have no practical filter design experience at 1.7GHz or they tried and failed and assumed their failure meant that everyone else must therefore fail too 

[KJDS]

If the circuit and PCB layout is simulated properly beforehand it should work first time or at least be very close. Ignore all the comments above that say it isn't possible to use SMD inductors (of just a few nH) to make lumped filters up at 1.7GHz. These comments are from 'experts' who either have no practical filter design experience at 1.7GHz or they tried and failed and assumed their failure meant that everyone else must therefore fail too 

I would prefer to do it as an interdigital, possibly on suspended substrate and would be confident that it would work first time with excellent performance, but given the timescales I know I could get a lumped bandpass filter to work first time.

[G0HZU]

Suspended stripline is a bit exotic 

But yes, it could also be done as an interdigital filter in microstrip although this would be a fair bit bigger than a lumped BPF. At work, I would traditionally design a 1700MHz BPF as a combline filter to get the size down.

Getting back to the original 1600MHz HPF I did a quick simulation in Genesys and Sonnet this afternoon using basic models for the caps and the inductors. I also managed to find an old dev board that I hacksawed to just contain the artwork for this HPF. This is a Rogers 4003C PCB 0.02" thick. See the simulation below and then I built it on the dev board. I used Kemet 0603 Hi Q capacitors rather than waste ATC caps but it still gave good agreement with the simulation for S21 and S11 up to about 5GHz or so. I made the inductors by hand as a tiny loop of wire.

The simulation is run in Genesys but it exports the layout to the Sonnet EM simulator to simulate the effect of the PCB layout. Hopefully you can see the VNA display of the real thing is quite similar to the simulation. The real filter has some cheapo SMA PCB end launchers and the loss and mismatch of these will degrade the result a bit up by 8GHz. But the loss is still only about 0.6dB by 8GHz. The VNA was calibrated with a 2 port ecal module so this should be a decent measurement in terms of overall uncertainty.

[jujun]

It's a very clean and professional work !
Can we see a picture of the filter ?

I really need to find some SMD hi Q capacitors and some roger substrate.

Because the filter I build was on FR4 from aliexpress and SMD capacitors from aliexpress too ... so not very good quality.

Any clues on how and where to find high Q capacitors and roger substrate ? My guess is that there is no really cheap way to get them.

I need to improve my knowledge in Genesis, because I never generated the layout with it.

[G0HZU]

See below for a picture. It's a bit scruffy because it is a crudely hacksawed section from a larger PCB. I had to bodge the connections for one of the SMA endlauncher connectors as you can see in the picture. Also, this PCB wasn't really designed for UHF so the 0603 parts aren't fitted as snugly together as I would like. But it was good enough. I didn't try and make it look neat so the soldering could be better.

But if something bodged together like this can work OK then a tighter PCB layout with better quality caps should be slightly better.

[G0HZU]

Any clues on how and where to find high Q capacitors?

I used caps from this Kemet dev kit from Farnell.

https://uk.farnell.com/kemet/cer-eng-kit-34/rf-microwave-capacitor-kit-0603/dp/2456895?krypto=nRv4lv5p7oYuI8syLvS3aGyGpmHYylT15G8zId6kXkBgSMcxMJojjX7pVrkdcJoNsUtY14BJf%2BbOEW2fEKs7sQ%3D%3D&ddkey=https%3Aen-GB%2FElement14_United_Kingdom%2Fsearch

Farnell stock number 2456895. I think it was about £50 for the kit. Not cheap but not exactly expensive either.

These caps aren't as good as ATC but they are going to be better than jellybean ceramic 0603 caps and you get lots of small values, eg 0.3pF to 1pF in 0.1pF steps and then the steps get a bit larger. But you still get lots of values up to 47pF.

Correction: Actually, I think the two outer 2.2pF caps are ATC 600S caps ( I forgot that I had loads of these to hand) but the centre one is definitely a 1pF Kemet cap from the kit.

[jujun]

I dont understand why your layout (in simulation) dont have a bigger ground plane on the top.

Yours coils look like mine !

[G0HZU]

I dont understand why your layout (in simulation) dont have a bigger ground plane on the top.

Because I'm lazy and I did it in a hurry I didn't think the extra ground artwork on the top was relevant so I didn't bother with this. I missed out a lot of via stitching as well but I don't think it matters much in this case. I just did a quick and dirty Sonnet simulation of what I thought was relevant. It simulates much faster like this.

[jujun]

Is it a good idea to buy for for example this https://www.ebay.fr/itm/100-pcs-CERA-TRIM-Johanson-M-C-0-6-2-5pF-250V-2320-RF-Trimmer-Capacitor/253027684130 instead of a rf smd capacitors kit  ?

The main problem is the footprint that is different, so I need to take it in account in the design and simulation, and I cannot use standard SMD capacitor on the board instead of a trimmer.

[jujun]

Also, I see that you don't add the SMA edge connector in the layout, and you don't simulate it too.
So I wonder if it's because it's rated to be 50ohm so you count it as not really part of the circuit,
or because you didn't took the time to add it ?

I always wonder how the layout under an SMA edge connector should be to have a really good match,
and I feel that the way the track enter the connector is very important to have a good match.

[G0HZU]

Yours coils look like mine !

Yes, just a bit of wire. I did use 0603 SMD inductors at first but I only had 2.2nH parts here and I had to fudge the cap values to keep the cutoff near 1600MHz. But it still worked OK. But I thought it was more relevant to the thread to show it working with handwound coils.

Is it a good idea to buy for for example this https://www.ebay.fr/itm/100-pcs-CERA-TRIM-Johanson-M-C-0-6-2-5pF-250V-2320-RF-Trimmer-Capacitor/253027684130 instead of a rf smd capacitors kit  ?

At work, trimmer caps aren't allowed for various reasons so I don't use them much here at home either. But I suppose you could try them as the price looks cheap, but I'd expect to see issues at these frequencies with strays etc so I can't really recommend them as an alternative. I'd really rather have the Kemet kit of fixed parts.... or buy both?

Also, I see that you don't add the SMA edge connector in the layout, and you don't simulate it too.

Yes, I just simulated the filter section and I did expect to see some extra loss up at 8GHz because of my fudgy SMA to PCB interface. But I think it is OK for a demo

[jujun]

Is it true to say that the lumped componement aproch for this frequency is only valid for RX and TX at low power ?
For example If I wanted to be able to use this filter on a 10W TX, capacitor that could handle this tension would be very big, and because of theire size it would act as an inductor.
So is it true that for TX, the microstrip or the interdigital would be a lot better ?

What is the maximum allowed power for a filter with small SMD capacitor like this ?

[Wolfgang]

Only sick people even think of making a transmit filter from 0603 components.
Reasons are:
- in a filter the voltages across components can be several times larger than the applied voltage due to resonances
  This is especially true in case of mismatch conditions. Ultra small SMD cases just can stand a few ten volts maximum, often a lot less.
- The loss tangent of small lumped SMD parts is poor. This means that a significant part of the input power is converted to heat,
  and the dissipation properties and heat conductivity of small parts is also very small.
- At 1.6GHZ, also substrate losses can become significant. This lowers Q and also generates heat.

In order to play safe:
- Simulate your filter with the maximum power level you have and all output terminations including, opens, shorts and other mismtaches.
- Find the maximum voltage levels for all caps. If your SMD stuff is overloaded, you need RF-grade caps with voltage margins. They will
  probably be larger than 0603, howver.
- Use a quality dielectric instead of FR4

[coppercone2]

when it comes to power you would need to have a reason for the electrical properties to change, such as thermal drift or some kind of saturation parameter. It won't have more inductance for no reason. They are C0G and stable.

[Wolfgang]

one of the obvious reasons is maximum voltage. Stability does not help when a cap just burns out due to arcing.
Have a look at antenna tuners or filter banks if you want to see burnt caps, and those are a lot larger than SMD C0Gs.
Simulation is key !!

[ogden]

Only sick people even think of making a transmit filter from 0603 components.

What you say about TI engineers? Sick? PCB picture, 2.4GHz TRX matching network:

http://www.ti.com/lit/ml/swru528/swru528.pdf

[T3sl4co1l]

That's a considerable amount less than the 10W mentioned above.

Tim

[ogden]

That's a considerable amount less than the 10W mentioned above.

I was kind of looking for power mentioned but managed to miss it. Thank you for pointing out.

10W is way too much for 0603 size components. Agreed.

[edit] High altitude, thus low pressure can make things even worse.

[Wolfgang]

At 10W your 0603 SMD stuff would blow up even at the bottom of the Marianas trench. 
For transmit filters, good Q and robustness is of prime importance, size is secondary.

[G0HZU]

Is it true to say that the lumped componement aproch for this frequency is only valid for RX and TX at low power ?
For example If I wanted to be able to use this filter on a 10W TX, capacitor that could handle this tension would be very big, and because of theire size it would act as an inductor.
So is it true that for TX, the microstrip or the interdigital would be a lot better ?

What is the maximum allowed power for a filter with small SMD capacitor like this ?

A lot depends on the filter type. If you want a narrow bandpass response from a lumped filter then it's generally best to design for power levels of <<0.25W. This is because of the high impedances within the filter. So you have to be wary of the RF voltages that might get generated within the filter at higher power levels as the voltage can easily zap the lumped components.

So if you definitely wanted a narrow bandpass filter at 1.7GHz at a 10W power level you would typically be looking at an expensive inline cavity filter from the likes of K&L or BSC or Spectrum/FSY.

But you can run fairly high power into a basic cheby1 low pass or high pass filter and still use lumped parts. It would generally be OK to use 0603 caps at 10W in a typical LPF or HPF as long as you used decent quality caps. For example, from ATC. The Kemet HiQ-CBR 0603 caps I linked to earlier might be OK in a 1.6GHz HPF at these power levels but that is just a guess.

A lot will depend on the frequencies you expect to use and the range of load (antenna) impedances it might have to work into. A HPF ought to be OK up to frequencies of several times the cutoff frequency but the inductors will see some thermal stress when transmitting very close to the cutoff frequency as in your case of Fc =1600MHz and an operating frequency of 1700MHz. I really don't know how good the Kemet parts are but I'd be disappointed if they failed when used in a basic LPF or HPF at these power levels into a 50 ohm load for example. ATC 0603 caps would be a better option.

[jujun]

Also to return to the initial topic, it would be very interesting to have a comparison of handmade coils Vs air core SMD coils.
What is the maximum Q possible by hand ? Is it better than air core SMD coils ?

On a test filter what will be the differences between a filter that use hand made coils instead of RF SMD air core ?

[G0HZU]

I'd guess that ultimately there would be no difference because you could copy the SMD coil dimensions with a hand wound coil. However, you would need a microscope, plenty of dedication and and steady hands! For a 3.3nH coil at 1.7GHz I'd think that a Q of 200 would be a practical maximum either for hand wound or SMD if you optimise the dimensions.

The main advantage of handwound coils is that they are adjustable. 

[T3sl4co1l]

SMD coils are only limited by what's on the market; whatever's out there, is what's out there.

You can make arbitrarily good inductors, give or take how much space you can spare and relative to the wavelength of interest.  At most frequencies I think a Q over 1000 or so is rather difficult, using regular metals and insulators.

Also as you go up in Q, sizes get larger and larger, until eventually the component is more like a piece of transmission line or resonator, and the geometry of your circuit overall matters very much (like "plumber's special" designs), so that it's not very meaningful anymore to speak in terms of single components soldered onto pads.  Which really shouldn't be a surprise anyway, as maintaining a Q so high is certainly a whole-circuit consideration!

At microwaves, dielectric resonators become interesting, too.  Think total internal reflection but at fractional-wavelength scales.  They can be very good, but obviously break the "component with pins" concept even further.

Beyond that, you need physics.  A transducer to a mechanical resonance, or an electron or nuclear resonance*, or superconductors.

*Although NMR has rather low density to bother with.  You can make an oscillator locked to proton resonance, but I don't think you'd make a real signal filter with much over -60dB insertion loss that way.  EPR is usable, for example YIG spheres.  Note that resonant frequency is proportional to magnetic field, making these useful for variable filters/oscillators but not so much for stable fixed references.

Tim

[G0HZU]

To show the advantage of hand wound coils, here's a simulation of a lumped (SC Coupled) bandpass filter design at 1700MHz. In order to tune this on frequency the inductors need to be tuneable. So it would be hard to make this with fixed SMD coils. The PCB layout will affect the centre frequency as you can see in the graphs. This simulation uses a PCB made from Rogers 4003C 0.02" thick. The graph in the top right corner doesn't take into account the PCB layout but the Sonnet S21 graph does and it shows the filter response is bang on 1700MHz. The insertion loss will be dominated by the Q of the inductors although even the cap Q will have some effect on loss for this type of filter. I chose a Q of 150 for the coils based on experience. You can see the filter has a bandwidth of about 200MHz but because it is only a 3 section filter the stopband performance isn't that great. But add a couple more sections and it would be much better for only a bit more insertion loss.

This bandpass filter would only be suitable for low power, definitely not suitable for 10W signals!

[coppercone2]

what if you just use copper plate and use shim-stock to elevate the parts over a copper plane and solder them in mid air to get rid of all the balony relating to PCB material?

You can get bits of metal to lay the parts down on, solder them together in an air bridge, then remove the shim-stock.

Personally the idea of using a structural reinforcement as a electrical property bothers me.

https://www.ebay.com/bhp/shim-stock

If you really solder it all down flat, maybe you can use your own controlled dielectric material as a sheet to lay the parts down on, if its necessary for some reason?

Or what if you glued the parts to a dielectric, soldered it together, then elevated the dielectric over the ground plane on supports?

[G0HZU]

Well you could just about make it ugly style over a bare copper PCB but it would be a fragile labour of love with 0603 sized parts. Even when fitted to the skinny Rogers 4003C PCB the 0603 caps can crack in an instant if the PCB is only flexed a tiny amount. The Rogers 4003C material in 0.02" thickness is easy to bend or flex and it is also quite brittle. So I had to solder a metal sheet under the earlier high pass filter PCB to keep it rigid enough to allow gentle handling. Otherwise, even the act of fitting and tightening SMA cables to it would probably crack the caps.

The SC coupled BPF is very prone to detuning if there is stray shunt capacitance to ground at the series LC node so this is a major reason why the PCB affects the frequency response. But it's no big deal to model it and correct the design to suit the PCB strays and the component strays.

[coppercone2]

isent the PCB horribly drifty as a dielectric because of internal stress?

[G0HZU]

No, the Rogers 4003C material is very stable. It's just a bit flexible and fragile if it is only 0.02" thick. For real production boards the 0.02" thick 4003C material is usually fitted as the top and bottom layer of a multilayer PCB. The central core of the PCB might be anything from 4 layers through 16 (very skinny) layers of FR4 material. This usually results in a PCB of a couple of mm thickness and it is very stiff.

I borrowed a wideband test RF amplifier from work today that can produce up to about 50W up to nearly 3GHz so I set it up to produce 10W at 1.7GHz and fed this through the little 1.6GHz HPF board I showed earlier. This was fed through the board into a decent (8GHz) Weinschel 30dB 50W attenuator and into a lab thermocouple power meter. With +10dBm on the power meter, there will be at least 10W at the filter. I looked at it on a Flir thermal camera and after a few minutes the inductors were at +56degC and the centre 0603 cap was a few degrees cooler. The main PCB was at about 36degC. I had to put a tiny smear of white Tippex on the inductor to get a reliable reading on the camera. But the caps didn't fail or explode

If I had used a PA with a poor source match and a load with a poor VSWR I'd expect the inductors to get very hot very quickly if the load was at a part of the smith chart that caused a higher voltage at the filter. But I didn't try this. I'd expect it to get very hot but I'd expect it to survive. Normally, there would be a metal heat spreader directly under the PCB and this would help in removing the heat from the board. So with a proper heat spreader, the temperatures on the thermal camera would have been lower.

[G0HZU]

I've also found another Rogers 4003C 0.02" PCB I can cut up with a hacksaw and I'll have a go at making the 1.7GHz lumped BPF. I think I'll have to cut it about a bit to get all the 0603 components to fit and I'll post up the frequency response but it might not be this evening. However, it's going to look just like the simulation as I've made quite a few filters like this over the years.

A microstrip interdigital or combline filter would probably be of more interest to most people. I can simulate designs of both of these and post them up as well? The simulation is usually very close to reality so I don't think I'd want to make them for real.
 

[Wolfgang]

Hi,

in order to make this test meaningful you have to tune the input power source thru all the band of interest and see where hot spots occur. Especially in a higher-order filter the hot spots move thru the structure as frequency is changed. When Qs are high the "hot spots" can be only small fractions of the center or corner frequency.
Another reason for filter destruction is load mismatch.

IMHO, a proper monte carlo simulation of input frequencies and load conditions could help a lot. Then you know what the worst case really is.

[G0HZU]

Only sick people even think of making a transmit filter from 0603 components.

At 10W your 0603 SMD stuff would blow up even at the bottom of the Marianas trench.

I guess I'm responding to your hyperbole above. Maybe your argument is that if you can cause one 0603 cap to fail in a filter at 10W with a certain PA type and a sweetspot load impedance then the above argument stands?

But I think it depends on how you define the requirements. This especially applies to the range of source and load impedances it is required to work over. It also depends on the filter type. LPF and
HPF are generally very low Q networks and so can operate at higher power levels than a typical narrow BPF.

https://www.atceramics.com/UserFiles/1206_LPF_HP.pdf

ATC make little 1206 packaged lowpass filters using thin film technology and this package contains the equivalent of a complete high order LPF. But they rate them at 12W. So that's possibly the
equivalent of 4 caps and three inductors crammed into this tiny low profile space yet they claim a 12W rating. The 12W rating will presumably apply only if you use the filter within sensible limits for source and load impedance and at sensible test frequencies.

[Wolfgang]

I am maybe a bit conservative, but in my logic a transmit filter must never blow up over the whole frequency range  and all possible load impedances.

Here is why:
- in case of antenna or cabling problems, almost *any* impedance can appear at the filter output, depending where the fault is and how long the cable to the filter output is
- during PA tuning, the feed impedance can also take a vast range of values, so does the power level
- filter components can fail by 1) overvoltage (w.g. du due resonances in mismatch) 2) overcurrent (same reason) and simply too high power loss (bad design or cheap cap)

I simply do not understand why the most important issue is to use minimum size components, not robustness, longevity and fault tolerance. Why create work for yourself by
making a marginal design only working on sundays with good weather.

Are you trying a planned obsolescene approach ?

[G0HZU]

Are you trying a planned obsolescene approach ?

I'm not sure what you mean but I've never used the 1206 ATC 12W LPFs. I've only shown them as an example of a 12W rated filter that uses the equivalent of tiny parts.

[Wolfgang]

... the best of the filters is the LPF1206HP6000, this one even has a negative insertion loss !  at least according to their datasheet.
But seriously: Their power handling data is given with a 50Ohm input and output termination and a CW signal.
Mismatch tolerance is not even mentioned. I guess this stuff is used in mobile phones where power is just a small fraction of the 12W specified.

[G0HZU]

I simply do not understand why the most important issue is to use minimum size components

Here's one example for you. If you use several ATC 600S (0603) caps in parallel in a LPF you get lower overall ESR and lower overall series inductance for the same overall volume as a single big cap. So this means lower losses (better efficiency) and better filter stopband performance.

[Wolfgang]

All agreed for a receive filter. In a transmit filter its a different story.

[KJDS]

Mostly, Tx filters have an isolator on the output, or some other form of antenna fault circuitry, or both.

Having said that, common sense says use a reasonable level of component derating, but for a discussion of what's possible on a forum then whilst it can be worthwhile mentioning something, it isn't beneficial to keep banging the same drum without coming up with something else constructive.

[Wolfgang]

Sorry, I wanted to be helpful in avoiding errors or a nonfunctional design.
I would propose to:
- use a quality substrate (Rogers)
- transmitter grade caps (ATCs are fine, but I would choose larger sizes so ratings are not a problem. 1206 should be fine for 10W)
- air core coils instead of SMD parts where power is at the limit. This also gives some tuning possibilities.
- Use a small box with a top lid ferrite absorber.
- simulate everything with all frequencies, power leves and loads involved.
That should do 

[G0HZU]

OK but ATC600S caps with a small value of a few pF have a typical ESR of 0.05-0.08R up at 0.5-1GHz with low package inductance. At 1.7GHz it is 0.1R ESR.  Put three in parallel and the equivalent goes even lower. The ATC 600S (0603) package rating for voltage at DC is 250V but ATC briefly test them at 2.5x this voltage to prove them.

https://www.atceramics.com/UserFiles/600s.pdf

To imply these 0603 caps are only suitable for receive filtering or low power Tx is not right. They can be used in typical LPFs at 10W even with a relatively poor load mismatch especially if the PA has a sensible source impedance. The old dev test amp I used was a balanced GaN PA between two IPP hybrids.

[G0HZU]

Minicircuits do a HPF in 1206 that is similar to the original 1600MHz filter requirement.

https://ww2.minicircuits.com/pdfs/HFCN-1320+.pdf

If you click on the data here

https://ww2.minicircuits.com/pages/s-params/HFCN-1320+_VIEW.pdf

You can see that their filter gets quite lossy by 8GHz. It even has 2dB loss at about 5.5GHz! It has 5dB loss by 8GHz. But it is OK at 1700MHz. They rate it at 7W max despite the tiny 1206 package but if it was used at 1700MHz at 5-7W I'd expect it to get quite hot. There are several caps and inductors crammed into that package and I think the inductors will generate most of the heat when used at 1700MHz especially into a poor load. If you look back at my homebrew HPF using decent low loss caps and hand wound inductors the loss across 5-8GHz was less than 1dB.

The Minicircuits 1206 HPFs are quite cheap at a couple of dollars and no inductors to find or wind! These are similar to the LPF and HPF filters seen in various RF analysers in youtube teardowns.

[G0HZU]

As promised, here's the response of the 1700MHz lumped BPF. This was built in a hurry using hand wound inductors and it is built the same as the earlier simulation in terms of the component values.

However, I used the Kemet CBR caps and not ATC caps and I don't know how accurate they are. The response looks to be very close but if you look closely the return loss is not as good on the real filter and the passband is slightly wider and less rounded. This is because the shunt cap values aren't quite right and probably need to be tweaked a tiny fraction of a pF. But this is good enough I think and the insertion loss is about 1.0dB at 1700MHz. It would improve slightly with tweaked shunt caps but I'm too lazy to bother. I've made loads of filters like this over the years so I'm not so keen to spend too much time and use up any more of my Kemet caps on it. I think it's good enough already

See below for a copy of the Genesys simulation and a slightly fuzzy image of the filter being tested on the VNA. In the lower left corner you can see part of the big old wideband 50W GaN test amp I've borrowed.

[coppercone2]

does anyone think the idea of a rice sized heatsink for a SMD filter is adorable?

Or would it detune and require a ceramic heat sink?

[G0HZU]

Having a heatspreader under the PCB is a good way to get heat away. On the exotic Rogers + FR4 multilayer boards the best way to do this would be to have the PCB manufactured with a pocket or cutout area within the 12(?) layer board that is only 2 layer. i.e. the 0.02" width of the top Rogers layer would be under the parts that get hot. This pocket could be under a filter or a TR switch PIN diode for example. Then make sure this sits against an aluminium heatsink and this can be part of a milled chassis. This would get the heat away really well and would prevent the local PCB area from gradually creeping up in temperature over time like a pedestal. But this PCB will be very expensive!

[G0HZU]

If you want to see a typical heatspreader then look again at my middle image in post reply#53 that shows the VNA plot of the lumped filter. In the lower left corner is the big old prototype GaN amplifier. This is a gold coloured PCB about 22cm x 18cm and I think the PCB is either Rogers 4003C or 4350 and it is only 0.02" or 0.5mm thick. This is very fragile and would snap in half like a cracker if it was flexed because the material is brittle and because of all the via holes in the PCB. If you look you can see it is screwed to a 10mm thick slab of aluminium and this provides support and it also provides a great way to suck heat away from the PCB. The aluminium then sits on another heatsink below this and this can be force air cooled. This really does keep the PCB components much cooler.

I also had a quick go at designing an interdigital BPF at 1700MHz using microstrip. I used a slightly thicker PCB material but it was Rogers 4003C again and you can see the result after a Sonnet simulation. I think the freebie Sonnet Lite program could cope with this filter as it only needed 29Mb of memory. I think the free version of Sonnet allows up to 32Mb.

You can see the insertion loss of the interdigital filter is about 1.6dB at 1700MHz even though the PCB material is Rogers 4003C. I've designed loads of filters like this but this was many years ago now. However, the simulation plot below should be representative of what you might get with such a filter. The simulation includes an enclosure with a metal lid over the PCB and this does affect the response slightly.

To give an idea of size, the vertical microstrip fingers in the filter are each about 28-29mm long.

[yl3akb]

G0HZU, great stuff!

What is the procedure You use to design such microstrip filters? Do You start with some initial model based design (do You use any available calculators for that?) and then move to EM simulation? Do You use any built-in optimization tools or just tune it manually?

Also, what is there at 1.7 GHz band for which this filter is needed? As far as I know, closest stuff is GSM at ~1.8 GHz and GPS/GLONASS at ~1.6 GHz

[yl3akb]

Sorry, I found the answer to last part of the equation at start of this thread

[G0HZU]

To design that interdigital filter in Genesys is a fairly clunky process but here is how it goes...

The design equations used by Genesys for the interdigital filter are fairly basic and are based on a few classic equations and these are usually given in the literature that gets bundled with each version of the software. I don't think this part of the software has changed in 25 years. These equations give the microstrip lengths, widths and spacing and the tap point based on user inputs for start and stop frequencies for the passband and the design ripple and impedance of the resonant sections and the specs of the PCB material. The software automatically generates the complete circuit as a linear model based on the equation results.

However, the equations are crude and the next step in the process is to simulate the design using Genesys' own enhanced linear models for coupled microstrip lines and for the via holes and end effects etc. This takes the accuracy to a much higher level.

This quickly shows that the basic design equations are inadequate on their own because the simulated response will usually be off frequency and will also be ripply with a poor match. So the next thing Genesys does is to automatically assign some optimisation goals to try and drag the response back to where it should be. The user can edit the optimisation goals if required (really easy to do) but usually they are OK. The optimisation is automatic and takes maybe 10 seconds to fiddle about and optimise the response.

This gives a tweaked version of the original design and it is often quite close if then built as a real PCB. But usually there will still be subtle issues with bandwidth, match and stopband performance in the real PCB version when built. So the next step is to set up the PCB layout such that it can be exported to the Sonnet EM simulator. This is mainly a manual (fairly tedious) process for the user and takes a few minutes to set up the filter and the ports in a 3D enclosure suitable for Sonnet.

Then the design is exported to Sonnet and it gets analysed. This is a very simple circuit for Sonnet and it takes a minute or so to simulate depending on the accuracy required. Usually Sonnet will reveal an issue with slight mismatch and it will give more realistic stopband performance due to the PCB layout and the enclosure. The results are automatically sent back to Genesys and presented to the user in Genesys and the end result you see on the screen is the result of me tweaking the layout (takes a bit of experience to do this quickly) to optimise the match and response such that it looks good after the final simulation in Sonnet

[G0HZU]

The final stage is to then to use the inbuilt PCB layout editor such that it can have extra pads and vias and connectors (and the artwork for a screen/lid)  as you would get in a real test PCB. This can be re simulated by Sonnet if required and the final PCB layout is then exported as a Gerber or DXF file. This can be imported into the SW that controls a PCB mill and the PCB can be milled and drilled and routed in about half an hour. Here at home I have an old T-Tech 7000S PCB milling machine that can do this for me. It isn't as fast as a modern LPKF machine but it can mill the PCB really accurately and it's possible to be testing a real PCB on a VNA within half an hour of finishing the design on the computer

[yl3akb]

Thanks for the insight of Your procedure.
I myself have little experience with AWR microwave office and process is quite similar. First, built in tool (called iFilter) generates perfect model based circuit, then it can be exported to AWR and simulated using AWR models (it takes in to account cross-coupling, vias, etc). I usually optimize the filter at this stage with built in optimizer (usually for particular return loss). Then filter can be simulated and optimized with one of AWRs EM simulators, like Axiem. I never tried more than ~2 GHz, but usually I need to trim resonators of actual filter, because pas band usually turns out little low, even with all the EM sim. which includes vias and enclosure.

You mentioned milling - from experience with such process, milling takes off  a little bit of  material also, not just copper. Does it affects the results of filter, knowing that coupled edge region is very critical for these kind of filters ?

[G0HZU]

You mentioned milling - from experience with such process, milling takes off  a little bit of  material also, not just copper. Does it affects the results of filter, knowing that coupled edge region is very critical for these kind of filters ?

There isn't a definite answer to this but a lot depends on the skill of the operator and the quality of the tools and also on the filter design itself. At work, our LPKF machine is only used by one experienced operator and he is really careful and he has access to new tools all the time for critical work. So he can usually mill a filter like this using a fresh/accurate end mill tool and he can set the depth so it only just skims the copper away with a straight vertical edge. So at work we usually get very good results when we transfer the same gerber to an etched PCB from Labtech or Graphic or Exception PCB.

Here at home I'm a bit less critical and I would probably use a narrow V cutter and a worn end mill to make a filter like this. Otherwise I end up using and wearing too many expensive end mill tools. Once the end mill tools start to get some advanced wear they tend to give a ragged edge and this is worse than using the V tool. Most of my tools are part worn freebies from work. So in my case my results don't look as neat but I'm usually pleased with the RF response. I have sometimes really messed up the depth and it doesn't seem to make a significant difference but maybe it depends on the filter type and specs. I tend to design fairly narrow BW filters so the spacing is wider and I think the milling isn't as critical. At work we sometimes use end mills as narrow as 0.005" for filters with very narrow gaps between fingers. But usually it's 0.015" for stuff like this.

[G0HZU]

To show a real world example where some people really do use 0603 caps at fairly high RF power levels, I had a rummage through some old GaN amplifier app notes at work and found this 30W PA from Cree. This uses a single GaN device and you can see in the output network below that they used 0603 caps in the output matching/filtering. Each capacitance is effectively two 0603 caps in parallel. You can see they are 0603 when compared to the size of the SMA connector. These will be ATC 600S caps and this amplifier is a fairly narrow design up at 1.4GHz. You can also see that the skinny PCB is mounted on an aluminium spreader/heatsink and this helps suck heat away from this filter network.  Obviously, this is an evaluation circuit aimed at demonstrating the capabilities of the GaN device rather than a demonstration of the robustness of ATC 600S 0603 caps in a 30W amp. But it does show that it isn't 'unthinkable' to use these tiny capacitors at 10W in a transmit filter network

[G0HZU]

Another supplier of HiQ ceramic caps is PPI

http://passiveplus.com/

You can get some idea of sizes and prices here:

http://passiveplus.com/kits.php#0603N - (.060 x .030) Ultra-Low ESR

We have various demo kits of these PPI caps at work in large and small packages but I don't think anyone has used them (except me).

Note that the Kemet HiQ-CBR 0603 kit I showed earlier has confusing data for price/qty at Farnell. My kit cost about £50 and it has qty 50 caps on tape for each of 36 values across 0.3pF to 47pF. So that is a total qty of 1800 HiQ 0603 caps for £50. They won't be as good as ATC 600S caps but they are pretty good value.

See below for what the Kemet kit looks like inside. You get two (and a bit) pages of 0603 caps all on tape and clearly marked. It's easy to use but not so easy to return caps to the kit that have been used or have fell out of the tape (when you got two but only wanted one!)

https://uk.farnell.com/kemet/cer-eng-kit-34/rf-microwave-capacitor-kit-0603/dp/2456895?krypto=nRv4lv5p7oYuI8syLvS3aGyGpmHYylT15G8zId6kXkBgSMcxMJojjX7pVrkdcJoNsUtY14BJf%2BbOEW2fEKs7sQ%3D%3D&ddkey=https%3Aen-GB%2FElement14_United_Kingdom%2Fsearch

[yl3akb]

I would also recommend Johanson S-series kits:
https://uk.farnell.com/w/c/passive-components/capacitors/capacitor-kits-assortments/prl/results?brand=johanson-dielectrics|johanson-technology&sort=P_PRICE

They also offer capacitor modeling/database software, which allows to export S-parameters of each capacitor:
https://www.johansontechnology.com/software

[jujun]

I think it's important to have the simulation model of all the capacitors of the kit.
For the Kemet ones, there is a paying option, and also this : http://ksim.kemet.com/

Also about SMD Coils, I noticed that coilcraft do a very cool feature in the design kit : there is a free refill !

Now I need to learn a design flow to be able to simulate layout with real models, and then I will buy a kit.

Thank you all for your precious help