"John's 1 MHz BPF"

I know he has bought a LPF for his application, but this is for kicks.

He had done a 4 resonator BPF design. It had a high spread of inductor valu es. It was canonical and had over 80 dB of suppression at the 2nd harmonic (ignoring loss).

What I have here is a 4 resonator tubular design where all inductors are 10 uH. (Keeping them equal, it is possible for the 50 ohm filter "class" I ma de to have the inductors anywhere between 6.8 and 27 uH, using the Norton t ransformer and R-scaling.) Tubular filters are not canonical. The design th eoretically has over 90 dB of attenuation at the 2nd harmonic for the Q val ues given in the ASC file. It is physically symmetrical. The heavy bias of poles at infinity was intentional because the main purpose is harmonic filt ering.

formatting link
formatting link

Reply to
Simon S Aysdie
Loading thread data ...

What the heck is a "tubular" filter? Where does that come from? Is it just descriptive of one method of building filters? --

formatting link

Works nicely in any case, and the only way you could do better is adding some zeroes at harmonics (by coupling from, say, L6 to L10 or L14, and the complementary pairings from the other side).

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
 Click to see the full signature
Reply to
Tim Williams
?

Yup. A cylinder with a connector on each end.

Cheers

Phil Hobbs

Reply to
pcdhobbs

(ignoring loss).

to have

transformer and R-scaling.)

dB of attenuation

physically symmetrical.

purpose is harmonic filtering.

Nice job Simon! I have a couple of questions...

You have various calculations in .PARAM statements, which looks like you're trying to calculate the C values... but it's not. I was hoping the sim might have the actual design equations you used to derive this filter, so I can derive one for e.g. 3.6MHz.

Can you share the equations please?

Clifford Heath.

Reply to
Clifford Heath

If you want the exact same (same fractional bandwidth, same synthesis type, same order), just scale all L and C by the frequency ratio. Ditto for impedance.

The frequency transform is: L --> L_old * (F_old / F_new) C --> C_old * (F_old / F_new)

The impedance transform is: L --> L_old * (Z_new / Z_old) C --> C_old / (Z_old / Z_new)

The filter topology is that of coupled resonators, using the Norton transform, as described.

A regular bandpass filter (that you get from any calculator) has alternating series and parallel resonators. The coupling factor is determined by their impedances relative to Zo (the resonant impedance being sqrt(L/C)). The further this is away from Zo, the lower the coupling. To lower the coupling, the resonant impedance is made lower for parallel resonators, or higher for series resonators.

As long as the Q is reasonably high (which it is, here), you can series-parallel transform whichever branches you like by stringing that branch up inside a capacitor or inductor divider.

This is the discrete LC version of the crystal filter: a series chain of crystals, with parallel caps to ground between each one (and occasionally a series cap to tune each crystal slightly). A crystal is best described as... a series resonator!

After transforming the parallel branches into series branches, joined with capacitor networks, values likely need to be tweaked because of various assumptions made in the transform. This is tedious, of course, but you'll have to align a real filter anyway.

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
 Click to see the full signature
Reply to
Tim Williams

For some reason, when I run that first one, I can't probe any nodes.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

alues. It was canonical and had over 80 dB of suppression at the 2nd harmon ic (ignoring loss).

10 uH. (Keeping them equal, it is possible for the 50 ohm filter "class" I made to have the inductors anywhere between 6.8 and 27 uH, using the Norto n transformer and R-scaling.) Tubular filters are not canonical. The design theoretically has over 90 dB of attenuation at the 2nd harmonic for the Q values given in the ASC file. It is physically symmetrical. The heavy bias of poles at infinity was intentional because the main purpose is harmonic f iltering.

It's because of the .SAVE line. To keep the RAW file down in size, only the listed elements are saved. It isn't needed. Just a habit of mine when do ing s-parameters.

Reply to
Simon S Aysdie

ust

You got it, as shown in the link you found. "Tubular" started as a descript ive term for the physical tube used in construction. But the tubular ladder pattern for BPF (and LPF) is pretty much nailed down, so nowadays tubular defines the schematic too. In a tubular L are all series, and there is a c ap-pi separating every coil. The end cap-pi make the end inductors "coercib le" in value and this implies it is a "parametric filter" (it is). Also, al l but one pole is at infinity.

I used the same number of inductors as John did, but a lot more caps. His i s N=8 and mine is N=10 (the "free parameters" of the parametric have th is cost of two orders higher, but those are cheap caps). One of the main ad vantages is that every node has design capacitance to ground. This means th e topology can "absorb" any parasitic capacitance (of any node) to ground.

e

Oh, those cross coupling zeros will probably sneak in whether one wants the m or not. lol. I got the pictures to prove it.

10

ade

s

Reply to
Simon S Aysdie

values.

e 10 uH.

All the .PARAM stuff was done so that the Q for an element could be defined without having the user calculate R. LTspice does not have a Q entry---we need R.

You can scale it to 3.6 MHz, as Tim replied. For me, I'd just start from sc ratch. (This is a pretty simple filter.) While I have worked out the math f or this style of filter, I didn't use that. I simply entered the basic para meters and let the synthesis tool spit it out. Then I did some post-synthes is manipulation (in the same tool) to make all the inductors 10 uH.

If I get time, I will recount the steps needed to produce a filter of this kind. Not this week tho. :-(

Reply to
Simon S Aysdie

Yes, obviously. But if I want different bandwidth, how do I go about that?

Ok, I can probably figure that out :)

Above 20MHz the inductors can be little air cores you can squeeze and stretch, which is probably enough tweakage.

Clifford Heath.

Reply to
Clifford Heath

e,

ing

ir

r

a

h

l

Great points, Tim. One general side note I forgot to make in another respon se is my design ripple was 0.1 dB and John's was 1 dB. That's another way o f saying the tubular I showed has much better return loss in the passband.

I put Q in this way so the user can experiment with its impact.

I also want to make a note about the realization (going from a driving poin t immittance function to an actual filter made of L & C) of this filter. No te the response of the "lossless reference filter." You can see that the 0.

1 dB band corners are *exactly* at 900 & 1100 kHz. The ripple (return loss) is also dead on to the spec.

I mention this because if the filter was realized according to the low-pass prototype, and then transformed via the approximations inherent to a non-i deal immittance inverter (to realize the classic coupled resonator filter), then it would actually be off frequency a bit (on paper). I think Cohn's p aper /Direct-Coupled-Resonator Filters/ includes a tweak to help the matter . In any case, I believe the tool I used synthesizes the whole thing in the bandpass domain. That is, it doesn't use the immittance inverter approxima tion, and so is dead on. I don't think this exactness matters in the end, but I think it is interesting. Most filter design programs produce the sche matic more or less in the manner you described.

Reply to
Simon S Aysdie

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.