Impossible RF filter values

Thread Starter

atlantis

Joined Apr 17, 2008
5
Hi

I'm new here, and the first question I'm asking is the following:

I use "RF Circuit Design" by Chris Bowick to design rf filters. It happens that the calculations turn out impossible values, like capacitors of less than 1pF or inductors of less than 1nH, in other words, values that can hardly be realized in the real world.

What should best be done about it? To build it these values must either be downsized to 0 or be sized up to the next possible value. How to do that?

Thanks for your help.

Kind regards

Martin
 

Wendy

Joined Mar 24, 2008
23,415
Generally when that happens you need to gang several filters together, or redefine what you're trying to do. Another option is to change the type of filter.

Basically not everything can be done the way we want to, but usually there is a work around somewhere. If you have a specific example post it up here and you'll probably get what you're looking for.
 

Thread Starter

atlantis

Joined Apr 17, 2008
5
Hi Bill

Yes I have an example and I'm happy to post it.

I'm trying to design a 4th order Butterworth as a front-end bandpass for amateur radio use in the UHF band. 3db bandwidth B was chosen as narrow as possible, so 20MHz (430-450MHz). Input and output resistance are the same (50R). Center frequency fo is 439.9MHz. Then the multiplicators can be read out of Bowick's table as C1,L1,C4,L4 = 0.765 and C2,L2,C3,L3 = 1.848 and transformed into values with the formulae

parallel-resonant

C = Cn / 2*Pi*R*B
L = R*B / 2*Pi *(fo^2)*Ln

series-resonant

C = B / 2*Pi*(fo^2)*Cn*R
L = R*Ln / 2*Pi*(fo^2)*Cn*R

where Cn and Ln are the multiplicator values.

I get the following resulting values:

C1 = 122pF
L1 = 1nH
C2 = 0.178pF
L2 = 735nH
C3 = 294pF
L3 = 445pH
C4 = 0.43pF
L4 = 304nH

where C1/L1 is the leftmost parallel-resonant element and C4/L4 the rightmost series-resonant element. About the same happens if I calculate the corresponding filter for VHF (144-148MHz, B=4MHz, fo = 146MHz). I wrote a spreadsheet to calculate order 2-7 and in whichever order there's always at least one odd value.

It's also possible that there's a mistake from my side, but if there is I don't see it.

Any help is appreciated, thanks.

Kind regards

Martin
 

Wendy

Joined Mar 24, 2008
23,415
I'm going to have to refresh myself somewhat on what you're talking about, I let my ham ticket expire over a year ago, and it wasn't well used even then.

Given that, try expanding your bandwidth, and then think of putting several filters after each other, with an amp between them if possible (even unity gain is good). If not, then a small attenuator/matching bridge to minimize interaction between them. Two filters with a 100Mhz bandwidth stacked will narrow the bandwidth somewhat.

I will say this, I've tuned my share of RF filters at Collins Radio, even with the right test equipment it is a bear to do. Those nice frequency responses rarely just fall into place, I've worked almost an 8 hour shift getting a not so basic bandpass filter to meet the specs. It looked really good when I was finished, but before I started... nasty.

It may be a day or so before I find the info, I suspect some of the more up to date guys will jump in here first. If I was still using my old TRS80 I had written some software to design what you're talking about, but that was 20 years ago and 10 years after school for me.
 

SgtWookie

Joined Jul 17, 2007
22,230
I put your numbers into AAED Filter Design, see the resulting plot.
You can download this freeware program here:
http://www.aade.com/filter32/download.htm

Something is a bit out of whack (see the lump on the low side) and the filter isn't quite symetrical.

I have also spent many hours in front of a network analyzer tuning filters. Bill's almost making it sound easier than it is. :eek:

Something that you'll find REALLY frustrating at those frequencies is the reactance of capacitors, even 0603 series SMT's. Down around 10MHz, a 47pf cap might measure very close to it's rating, but by the time you approach 500MHz, it's measuring triple that. What's worse, it's a non-linear response curve. Lacking a $20,000 material analizer, you'll have to experiment.

At those frequencies, layout will be critical, along with lead length. Even a short piece of straight wire measures a few nH.

As Bill mentioned, widening the passband will help things quite a bit.

Your filter is very narrow for the center frequency.
 

Attachments

SgtWookie

Joined Jul 17, 2007
22,230
Threw a few more numbers at AAED. Widened out the passband to 40MHz, centered around 439.9MHz. If only real-world return loss looked that good ;) Of course, this is with idealized components; your Q won't be nearly that good.

Did another with a 22MHz passband.

Download the program and experiment with it. You'll find it's a lot easier than trying to crunch those numbers out by hand.

Did another for 20MHz passband. The numbers were close to what you calculated, but your L1 should've been 1.076nH. That was off enough to throw the filter way out of tune, with very poor return loss.

If you round the numbers off, you'll get a very nasty looking filter output. You have to be really precise.

Tuning the physical filter is going to be lots of fun, even if you happen to have a spectrum analyzer and a HF signal generator. It's easier if you have a network analyzer with an S-parameter test set, but those can get pricey even in the used low-end varieties. You might see if your local college or university's EE department has a student that wants to try their hand at it.
 

Attachments

Thread Starter

atlantis

Joined Apr 17, 2008
5
This program is great, I installed it and am fiddling around with it ;-) Yes this is a lot easier than doing those calculations by hand, thank you for the tip!

Kind regards
Martin
 

SgtWookie

Joined Jul 17, 2007
22,230
You're welcome, Martin!

As an experiment, design a filter that has a narrow passband like yours (where the passband is roughly 4.5% of the Fc) and try modifying the numbers a bit.

Then design a similar filter with a much wider passband, and see how much less of an effect the variables have on it. The return loss plot is the best indicator of how well-tuned the filter is.
 

Thread Starter

atlantis

Joined Apr 17, 2008
5
I have played around a bit, but the initial problem stays the same; as soon as I have a design which makes sense from the response curve and I start to replace the values for available ones the curve turns from fine to absolute crap.

How the heck is it possible to bring available component values and design together? The values still have a tolerance, and, as you say, the filter will, in reality, not be as good as the simulation looks. Therefore the design should look absolutely immaculate on the screen to have chances it will be useful when built, shouldn't it?

Kind regards
Martin
 

SgtWookie

Joined Jul 17, 2007
22,230
That's the problem!

The engineer designing the filter has to continually re-tweak the basic design with available values, and then the techs have to deal with the results.

You can't forget things like lead lengths, either. A section of paperclip wire about 3/4" long has a value of about 8pH. I discovered this when investigating zeroing errors on a material analyzer; one of the engineers decided that a paperclip would have negligible inductance. I replaced the paperclip with a solid sheet of copper, re-calibrated, and tried the paperclip again.

Designing filters that are actually buildable/tuneable in that frequency range is not an easy task. You'll find that physical components don't respond like they do near DC levels (below 200kHz). If you can actually get one built and working within a month, you will be doing better than a number of filter engineers making good money designing them.

It's not going to be easy.
 

SgtWookie

Joined Jul 17, 2007
22,230
For your caps, you might look at Johanson's trimmer cap offerings.
http://www.johansonmfg.com/
Newark stocks them.
Look at air and nonmagnetic.
They will cost quite a bit more than you would pay for an ordinary ceramic cap, but the ordinary variety aren't adjustable.

You could try to do it "on the cheap" with a "gimmick" capacitor. Gimmick caps are short pieces of wire or narrow, thin flat copper stock that the tuner trims to size while carefully bending back and fourth to observe what makes the filter "look" better, or closer to specifications.

Coils are another thing. At those values, you can use Wheeler's formula to calculate your single-layer coils; that will get them close. It's likely that you will have to open them up (stretch them longer) quite a bit to get them where they need to be. Larger diameter wire will give you a better Q, but you'll wind up with a much larger inductor, and corresponding parasitic resistance/capacitance.

You CAN actually tune filters like this using an old-fashioned meter that has a dBm scale along with a frequency sweep generator. YMMV.

Construction will be somewhat problematic. At these frequencies, parasitics even from objects a ways away come into play. Building your circuit using non-copper-clad perfboard, and suspending it inside a copper enclosure may make your tuning task a lot easier. The response will be different depending upon if the cover is on or off. How deep your circuit is in the box will affect that response.
 

Wendy

Joined Mar 24, 2008
23,415
The real world solution for filters is variable components, which is why techs generally don't like them much. Most real world filters that have more than 10 components more than half are variable, and that especially includes inductors. Makes life easier for engineers, and the techs doing the job quietly make a new stock of engineer jokes.

I had to design and build a filter at Collins one time, so for fun I built two identical versions, one with surface mount and one with leads. It's nice having a really deep parts bin. The SMT version was a lot cleaner because of lack of leads, and it's rejection of the off frequencies was well over 10db greater. This was an easy filter too, a simple low pass in the RF range.

Another approach to designing filters, one that is somewhat easier, it to break them down into parts. Design a high Q low pass filter, isolate it with a unity gain amp, then design a high Q high pass filter. You tweak each one independent of each other, then combine them. By isolating the two curves the controllability goes up somewhat. The reason I keep mentioning the unity gain amp is isolation is important. The interaction of the components is what drives the complexity of tuning any filter through the roof.
 

Thread Starter

atlantis

Joined Apr 17, 2008
5
You're pretty much discouraging me... I appreciate being told the truth, but, from what you say, and as I don't claim to be a first-class RF engineer, I'm not sure whether I would end up with a good result in less than 2 years :confused:.

The problem with air trimmers, apart from the price, is that they can obviously not be made that small, compared to 0603 components and ceramic miniature smd trimmers they're still big big big. For a repeater frontend this would not matter that much, but here I'm trying to save space.

Even though a lot could be learned I might be better off for the moment looking for some 2 or 3 element helical filters which are often used for this purpose. What do you think?

Kind regards
Martin
 

Wendy

Joined Mar 24, 2008
23,415
Most of the variables in the pF range are pretty small, as are the inductors.

Anytime you can buy premade filters you are ahead in the game, but I would still try to make them. By the way, the bandpass filter I was talking about was an m derived job, which makes for interesting reading.

The other way most filtering is acomplished is to hetrodyne the frequency down, then filter the lower frequencies that are generated. Much easier.

I have found some interesting result with straight coax cut to length, it is a suprisingly sharp filter, although it will also pick up every harmonic too.
 

DickCappels

Joined Aug 21, 2008
10,152
Exactly. You aren't going to be able to make a filter like that with lumped constants.

As SgtWookie pointed out, parasitic characteristics become significant at such frequencies.

A 45 MHz narrow band filter yes; a 450 MHz narrow band filter, probably not with discreet components.
 
Top