Stagger-tuning of FM stereo valve IF stages

[regenfreak]

Gents/Madams
Suppose I have a three-stage, double-tuned LC IF filters for FM valve tuner, is there a simple way of estimating the frequencies ( fo1, fo2 and fo3) so that the overall response has a bandwidth of 250KHz centered at 10.7MHz? Please see attachment. Or there is a rule of thumb? I cannot find any information, protocal or good practice on stagger tuning of FM stereo recievers (e.eg. service manuals)

I have looked up a few technical papers on the design of stagger-tuned amplifiers, the maths is beyond my ability. I dont understand the complex maths of poles and Butterworth filters....

I am interested in doing a visual sweep of FM stereo valve tuners; in my case a DIY FM valve stereo tuner. I have built the chinese FM multiplexing tuner kit and it works perfectly giving awesome sound. Although I have a perfect overall response centered at 10.7MHz and BW=250kHz, it has taken me a long, long time by trials and errors to get the perfect stagger-tuned response curve in the visual sweep. Is there an online calculator/software or simple formulas for estimating fo1, fo2 and fo3, it would save me lots of time mucking about, adjusting the ferrite cores at each stage.

[Radio Wrangler]

Welcome to the deep end of filter theory.

Each double-tuned IFT produces two poles in the response. So, three IFTs can give 6 poles

It isn't so easy as one slug tunes one pole and the other tunes the other. Each slug affects both. The spacing between the poles is strongly influenced by the amount of coupling between the two resonators, and the Q's of the poles are set by the coupling into the source and terminating impedances coupled to the resonators (and to the coupling between them if this is asymmetric)

So, each IFT will create a pole pair, so your filter will have to have an even order.

The Butterworth filter is free of ripple and accelerates smoothly downwards to an eventual slope of 20dB per decade per pole. The Chebyshev filter has ripples in its passband response but as a reward for this compromise gets the jump on the Butterworth filter by starting to fall quicker... it's a more abrupt filter. Both these filters exhibit a lot of change in their time delay versus frequency, the Chebyshev worse than the Butterworth.

Change in the filter's time delay versus frequency doesn't sound too bad, but with an FM signal, it introduces distortion.

Serious FM tuner designs opt for so-called linear phase filter types. This is mis-leading. None are truly linear, just some are better. The Gaussian shape is usually touted as the most linear, but it has very little real selectivity. The usual choice for good FM is a Bessel shape of filter.

Had I to design your IF, I'd open my copy of Zverev (Handbook of filter synthesis) and look at the pages of tabulated data for Bessel filters and look for the 6th order entries.

The tables are normalised, so I'd have to scale them for centre frequency and -3dB bandwidth. These tables contain coordinates in 2 dimensions for the poles needed to make that filter (they also contain Q's and coupling factors for making a block filter without amplifiers between stages)

So I now have the coordinates of 6 poles. Knowing how IFTs make them in pairs, I pair them up. Plotting their coordinates on a piece of paper helps. The outermost ones go as a pair, the innermost ones as a pair and the two remaining for the last pair. Notice this is all symmetrical. I now design the Qs and coupling factors of each IFT to position its pole pair.

So what I've done is to break up the wanted 6-pole IF shape into a trio of independent 2-pole filters. Each will be different. Together they create the 6 poles with the right frequency separations and Q's to create the Bessel filter.

There isn't a lot of arithmetic involved, but unless you're used to them there are some utterly weird concepts involved.

You'll have come across equations for the reactances of inductors, capacitors and know that there are 90 degree phase shifts involved (Quite a wild concept in itself, but everybody is so used to it that no-one bats an eyelid. Wild maths hidden in full view!) Someone took the mathematical idea of using a wild card number, j (mathematicians use i-for imaginary) as a 90 degree shift indicator. Real numbers are the sort we know and love. imaginary numbers are marked by being multiplied by j. A complex number is one with both real and imaginary components for example: 17+0.12j

This is useful. We can keep 90degree-shifted capacitive and inductive impedances separate from the real stuff, but we can also keep tabs on what's going on. If multiplying something by j turns thar thing round by 90 degrees, then doing it again turns it around by another 90 degrees. So 180 degrees... that means the phase has been exactly reversed. The same as multiplying it by -1. So imaginary things can gang up and interact with the real world. so multiplying by j twice is multiplying by j squared, and is the same as multiplying by -1. So j squared is the same as -1. So j must equal the square root of -1. BUT the square root of -1 does not exist. That's OK, that's the whole point. It is imaginary! Plenty of things don'e exist and are fictional, but they can have real effects. Hercule Poirot is 100% imaginary fiction, but the money Agatha Christie, her publishers and theatres made out of him was quite real and spendable

Similarly, the use of imaginary and complex numbers for impedances etc will keep our bookkeeping honest in terms of phase shifts.

Frequency is a friendly concept. 14.060MHz for example. Amongst mathematicians ther are vandals. People who will try anything, no matter how stupid seeming just to see what happens. Do not put any nuclear launch buttons anywhere near mathematicians. Someone mused about what would happen if a complex number was plugged in for frequency into classical equations which analysed circuits. You can't do that! said the sensible people. This only encouraged them. The results were interesting. At some imaginary frequencies a simple passive RC filter ramped up to infinite gain or zero gain.

Infinite gain from a passive RC? Yes, but only at an impossible frequency. We call it a pole. The zero gain point is called a 'zero' and they are not impossible at real frequencies, but they can also happen at imaginary frequencies.

What came out of this was that popular filter shapes became patterns of poles and sometimes zeroes. What was a horrendous mess of nasty high-order differential equations needing to be solved suddenly became the geometry of patterns. And can now fit into human brains

So poles and zeroes are not dragons to be feared. They are cuddly and helpful, once you've got over the weirdness.

So that's a run through of what's involved in designing a stagger-tuned IF. It's a day's work.

I' think I've also explained why, if you watch a sweep, tuning one slug doesn't do what you'd expect. It makes tweaking these things go wrong for a lot of people and RF engineering becomes a very black art. Those that can visualise the maths can use it to cheat

David

[regenfreak]

Dear David
Many thanks for your answers. I always find your posts very helpful and insightful.

Each double-tuned IFT produces two poles in the response. So, three IFTs can give 6 poles

It isn't so easy as one slug tunes one pole and the other tunes the other. Each slug affects both. The spacing between the poles is strongly influenced by the amount of coupling between the two resonators, and the Q's of the poles are set by the coupling into the source and terminating impedances coupled to the resonators (and to the coupling between them if this is asymmetric)

I can easily see the double peaks in a sweep using a nano VNA. I have been using the "damper method" to find the resonance frequencies f1 and f2 of primary and secondary for each IF stage. The damper ( a series resistor of 4.7k and 2.0nF cap) eliminates the peak of the secondary while I sweep the primary, and vice versa (see attached drawing).

The chinese FM stereo kit has no instruction and schematic. I had to do reverse engineering, circuit tracing and drawing the schematic. So i had to do things in a hard way, reading every textbook available on FM receiver design. To make matter worst, the chinese IF transformers use asymmetric double-tuned LC filter (see the attached drawings). The secondary capacitance C2 is twice as the value of the primary C1. I believe a larger secondary inductance L2 will give a larger Q matching the higher impedance of the grid input, and the lower L1 at the primary provides a broader response curve to facilitate stagger tuning.

It is imaginary! Plenty of things don'e exist and are fictional, but they can have real effects. Hercule Poirot is 100% imaginary fiction, but the money Agatha Christie, her publishers and theatres made out of him was quite real and spendable

Brilliant lines Plenty of theory on quantum physics and black holes are like fiction anyway.

nfinite gain from a passive RC? Yes, but only at an impossible frequency. We call it a pole. The zero gain point is called a 'zero' and they are not impossible at real frequencies, but they can also happen at imaginary frequencies.

What came out of this was that popular filter shapes became patterns of poles and sometimes zeroes. What was a horrendous mess of nasty high-order differential equations needing to be solved suddenly became the geometry of patterns. And can now fit into human brains

I remember the poles and zeros in my Avionics control engineering lectures back in 1987 as an undergraduate student. I didnt understand it and i still do not understand it. I get very used to looking at the Smith chart after playing with the Nane VNA v2 for a few months. I am not so much afraid of complex numbers but they are not so intuitive.

For examples, i have been reading these two papers:

https://www.ece.uic.edu/~vahe/spring...2/Bandpass.pdf

https://global.oup.com/us/companion....Appendix_H.pdf

I wish there would be formulas for finding the approximate solutions or someone would write a programe to solve 6th order filters....Of course this seems to be wishful thinking.

I' think I've also explained why, if you watch a sweep, tuning one slug doesn't do what you'd expect. It makes tweaking these things go wrong for a lot of people and RF engineering becomes a very black art. Those that can visualise the maths can use it to cheat

I have been digging around the service manuals of FM stereo tuners..nothing, absoutely nothing about the stagger tuning. I have seen stagger tuning in service manuals of valve TVs IFs though.

I have noted that the probe loading has significant effect on the response curve shown on the sweep. The probe's distributed capacitance and inductance can flatten the response curves...I need an isolation resistor of 33k or 100k (some service manuals suggest 1M) as the impedance of the probe drops to a few kilo ohms (see the frequency response of my 350MHz probe at 10:1). When i measured the Q and gain of each IF stage, i got much lower Q values due to the probe loading effect.

[regenfreak]

The attached photo is a NanoVNA v2 sweep of the chinese asymmetric IF FM filter. See the two resonance peaks.

Note the measured resonance frequencies f1 and f2 are inaccurate/misleading even the "measurement plane" are corrected at the tips of the crocodile clips during the VNA calibration.

This is due to input and output impedance mismatch, and the loading effects of the clips. Ideally, there should be low loss reactive matching network pads at the input and output of the VNA.

Also coupling caps of very low capacitance value ( 2pF?) will be needed to isolate the LC tank from the VNA terminals for proper resonance frequency measurements.

[Radio Wrangler]

I'm afraid I don't see two peaks there at all on any of the traces - it's difficult being sure of the text at the top.

Most probably the two resonators are being damped a lot by the connection to the VNA.

I'm used to the full size network analysers and they have crazy amounts of sensitivity, so I can connect them via large value resistors (like 10k for example) to very lightly couple to things. Ideally you should make up something to present impedances very close to what the circuit it's to be used in will present.

Coupling between those two resonators in that IFT in the photo should be enough by simple proximity and magnetic coupling.

Let's go theoretical and idealised.

I have two resonators (this goes for LC, crystal, tuning forks, even weights bouncing on rubber bands) Both resonate at the same frequency and have infinite Q

For making a filter, if you look at the ratio of centre frequency to the 3dB bandwidth you are shooting for, this gives you a figure for the overall finished Q of the filter. You use it as a guide number. You want the Q of each of your resonators, isolated from inputs, outputs and other resonators to be significantly higher than the filter Q. The more they exceed the filter Q, the less insertion loss you get and the closer you come to the target shape when you build them up into a filter.

I now make a filter out of my two resonators. Remember that they both start out as identical. If you VNA'd them individually you'd get the same traces. They are tuned to the same frequency. This is the equivalent of showing you that there are no doves in my hat and that my sleeves are rolled up. Don't let my beautiful assistant distract you.

First, I connect one resonator to the signal input from whatever source impedance it was designed around, with the designed amount of coupling. The resistive part of the source impedance immediately damps the Q of the resonator down from infinity to the value arrived at in the design process.

Then, I connect the second resonator to the signal output with whatever load impedance it was designed around with the the designed amount of coupling. It too has its Q brought down from infinity to its planned value (not necessarily the same as the first one)

So I now have two resonators, both resonating at the centre frequency, say 10.7MHz, each with a carefully controlled Q. Notice that the source and load impedances applied are critically important in getting the planned Qs and hence in getting the planned filter shape. Filter shapes are utterly dependent on the right source and load impedances being applied. This is where an awful lot of people go wrong. Note they are still on a perfect 10.7 MHz and each looks like a single pole on that frequency.

So now I bring these two resonators closer together so they couple to the extent the design calls up. This is where the doves appear and the lady gets sawn in two.

The very act of coupling, whether by field proximity or added components causes the individual poles to vanish and to be replaced by a pair which are split apart a little in frequency. It would be terribly easy to say the poles separated, but that would be misleading. It would leave you thinking this pole went thisaway and that pole went thataway. There is NO tracing back of either of the split pair to one resonator or the other. BOTH are created by the whole filter.

This is what makes filter twiddling difficult. Your VNA shows a trace with bumps, but you don't have an individual trimmer for each bump. Trimmers affect multiple bumps in a patterned sort of way. With more involved filters (and I've done 9-resonator ones) the patterns of effects get complicated. Attempts to twiddle them tend to diverge and you get lost. You wind up with a worse shape than you started with, and no knowledge of how even to get back to the starting point. You're looking at a 9 dimensional maze in the case I mentioned.

The experienced wise man knows not to enter the maze. Instead, he cheats.

If we didn't have the couplings, each resonator would resonate at the centre frequency... Hmm, convenient, eh?

So we kill the couplings (somehow) not by breaking them, but by killing the resonators on either side of the one we're concentrating on. We very very gently insert a little signal through a big value resistor, and monitor the response through another. We tune it up to peak on the filter centre frequency. We do this process to every resonator in turn, then we remove the shorts we applied to kill the resonators, and voila! the filter springs into life giving the planned shape, bandwidth and centre frequency. The lady is rejoined and is helping pull doves out of the hat.

It looks like magic, but it's simply there in the design process!

You don't have to do this in real life to study it, you can do it in an LTspice simulation.

The maths of poles and zeroes are normally taught from the control theory perspective and students are both bored by the lectures and are scared off in panic. They have plans to dodge those subjects come exam time.

I saw this, and realised that university output was mostly people lacking these skills. They are necessary, so people with them will be at a premium... hmmm... I see a plan!
I waded in , thought about it and found it was nowhere near as bad as everyone assumed. In fact, it paid bonuses. The lecturers expected no-one to choose their questions unless they'd really run into trouble on the rest of the paper and were just hoping. So they not only wrote nice straight-forward questions, I think they marked generously in the delight of someone helping justify their existence.

After getting a job, yup, it worked. I have the tools to design PLL control loops, amplifiers with stable feedback, oscillators, filters and a great many other things. It wasn't even hard. It was just a case of not being scared of something that looked like alien technology.
If you don't let on that it was a lot easier than it looks, you appear to be a mage of at least the eighth order, with optional illuminated magic wand and cloak.

David

[regenfreak]

I really like your lady in half metaphor, David The RF Magician! I will need to go away to think about what you have written and will sleep over it tonight.

I'm afraid I don't see two peaks there at all on any of the traces - it's difficult being sure of the text at the top.

Most probably the two resonators are being damped a lot by the connection to the VNA.

Above VNA sweep of a FM IF photo is not clear because the primary and secondary are de-tuned. I found some old photos of 470kHz IF sweep without and with damper (see attached ). You can trick the scope to show double humps in a sweep if you attach a series resistor that is large enough(in the example 470K) to reduce the damping effect of the signal generator. I set my Rigol signal generator to default " hi-Z output" and connect a 470K resistor to provide some isolation. Most signal generator has 50 ohms or 75 ohms output impedance that would damp out the resonance peak whichever side of the LC double tuned filter attached to.

First, I connect one resonator to the signal input from whatever source impedance it was designed around, with the designed amount of coupling. The resistive part of the source impedance immediately damps the Q of the resonator down from infinity to the value arrived at in the design process.

Then, I connect the second resonator to the signal output with whatever load impedance it was designed around with the the designed amount of coupling. It too has its Q brought down from infinity to its planned value (not necessarily the same as the first one)

So I now have two resonators, both resonating at the centre frequency, say 10.7MHz, each with a carefully controlled Q. Notice that the source and load impedances applied are critically important in getting the planned Qs and hence in getting the planned filter shape. Filter shapes are utterly dependent on the right source and load impedances being applied. This is where an awful lot of people go wrong. Note they are still on a perfect 10.7 MHz and each looks like a single pole on that frequency.

So now I bring these two resonators closer together so they couple to the extent the design calls up. This is where the doves appear and the lady gets sawn in two.

In my thought experiment(see my drawing below), i have two church bells loosely coupled by a rope. Now there are are several scenarios ( i dont know if they are appropriate analogy?):

1) I hit one of the bell by a hammer, they will oscillate freely in damped oscillation.

2) I pull the string of one of the bells, the string is equivalent to loose capacitive coupling-so that my hand has little influence of the natural frequencies of both bells.

3)I replace the light string by a metal chains, my excitation has tightly coupling to the bells and hence affects the resonance of the bells.

4) I hold and shake one of the bells by hand. My hand will become part of the resonators and lower the resonance frequencies of both bells

5) I send both bells by a rocket to space and eject them into deep space and excite by a baton round; both of oscillating perpetually.

So the proper way of measuring the sweep response of a double LC filter would be like in my drawing? Some Pi or L resistive matching pads for both VNA 50 ohms input and output ports. However, resistive network introduces lots of insertion loss. Maybe some form of inductor and capacitor reactive network may be better.

The maths of poles and zeroes are normally taught from the control theory perspective and students are both bored by the lectures and are scared off in panic. They have plans to dodge those subjects come exam time.

This was exactly what happened in my Avionics examination.

[John_BS]

Looking at your first post, is it the case that this IF strip has three sets of double-tuned (coupled) inductors, which are each buffered / separated from each other by an amplifier stage or similar?

If so, the problem becomes much simpler: if you want -3B points at say +/-125kHz, each double-tuned stage should be tuned to 10.7MHz and configured (loaded) to be -1dB at these offsets. As David has said, it's the group delay which is actually more important, but you do tend to get less GD variation with Butterworth / Guassian (monotonic) amplitude responses compared with Cheby equi-ripple. The penalty you pay is less steep out-of-band attenuation skirts. Also bear in mind that the working Q's required for these coils operating at 10.7MHz means you generally end up with significant loss in (say) a six-pole design (where all the coils are intercoupled in one filter).

In my distant past I did design a 6-pole 10.7MHz L-C filter which incorporated additional coupling between non-adjacent resonators: this gave rise to very useful notches on either side of the passband, and in an astonishing byline, the group delay was improved, not degraded.

John

[John_BS]

PS Regenfreak: you have my sympathy regarding the maths etc of filter design. David mentioned Zverev, and if you can find a copy it's a real bible.

The approach I found most useful was to follow through the design process starting from what is termed a low-pass prototype. Zverev (and others) tabulate normalised values for these low-pass filters according to order (2, 3 ...6 etc) and type (Butterworth, 0.1dB ripple, linear-phase etc).
The component values are normalised to 1 ohm and 1radian/s. To design say a simple low-pass filter, you just de-normalise the element values; e.g L is multiplied by Ro/w and C by 1/(Ro.w) where Ro = input and output termination and w= cut-off frequency in radians.
For a band-pass filter, each low-pass element is further transformed into a tuned circuit: an L becomes a series tuned circuit, and a C transforms directly to a parallel tuned circuit. In the vast majority of cases this results in an unrealisable filter, as the series circuit will contain unfeasible component values, and the L will, of course, have a parasitic C across it which can't be absorbed into another component.

This is where a bit of magic comes in: something called the j-inverter.
A quarter-wave transmission line, when shorted at the far end, looks like an open circuit at the other, and vice-versa. If we connect a series-tuned circuit to the output, it looks and behaves like a parallel tuned circuit at the input. What's more, we can transform the impedance slope = L-C ratio by changing the Zo of the transmission line.

Now, it is generally not convenient (nor practical) to use actual transmission lines, but a j-inverter can be constructed from a variety of networks, the caveat being that the transform is only valid over a limited frequency range.

The simplest and most used inverters (at these IF frequencies) are a) Mutual inductance and b) a PI network comprising

-C' to ground, +C' series (coupling), -C' to ground.

If you do the sums on the input and output impedance of the latter, you'll see it yields the required short/open behaviour, and it has a characteristic impedance of jwC.

When the -C',C'-C' network in placed between two parallel tuned circuits, the -C's are conveniently absorbed into the main tuning C's and vanish. The tuning C's are therefore made C' smaller then they would have been to make the filter resonate at Fo. We now have a filter which is realisable, and because we are free to chose C', we can make the two inductors the same value, as we're scaling the impedance of the series L-C thro the j-inverter.

Hope this helps!
John

[regenfreak]

Looking at your first post, is it the case that this IF strip has three sets of double-tuned (coupled) inductors, which are each buffered / separated from each other by an amplifier stage or similar?

If so, the problem becomes much simpler: if you want -3B points at say +/-125kHz, each double-tuned stage should be tuned to 10.7MHz and configured (loaded) to be -1dB at these offsets. As David has said, it's the group delay which is actually more important, but you do tend to get less GD variation with Butterworth / Guassian (monotonic) amplitude responses compared with Cheby equi-ripple. The penalty you pay is less steep out-of-band attenuation skirts. Also bear in mind that the working Q's required for these coils operating at 10.7MHz means you generally end up with significant loss in (say) a six-pole design (where all the coils are intercoupled in one filter).

Thanks John. Each stage is separated by a pentode amplifer (see attached schematic). So the primary is connected to the plate of previous stage pentode; the secondary to the input grid of the succesive pentode.

Now the complications are twofold:

(1) Each LC filter has asymmetric design, different L,C and Qs for primary and secondary tank circuits.( I believe the design is a compromise between gain and broader bandwidth. I spent a long time scratching my head and thinking why it is asymmetric before arriving at this conclusion).

( 2) The Miller capacitance affects the resonance frequency of the primary LC.

I wish each double tuned stage would be tuned to 10.7MHz and offset then by -1db. But the problem is that finding the values of f1, f2 and f3.

f1 is the combined effect of f1p and f1s (primary and secondary) ; f2 from f2p and f2s; f3 from f3p and f3s.

The experienced wise man knows not to enter the maze. Instead, he cheats.

If we didn't have the couplings, each resonator would resonate at the centre frequency... Hmm, convenient, eh?

So we kill the couplings (somehow) not by breaking them, but by killing the resonators on either side of the one we're concentrating on. We very very gently insert a little signal through a big value resistor, and monitor the response through another. We tune it up to peak on the filter centre frequency. We do this process to every resonator in turn, then we remove the shorts we applied to kill the resonators, and voila! the filter springs into life giving the planned shape, bandwidth and centre frequency. The lady is rejoined and is helping pull doves out of the hat.

It looks like magic, but it's simply there in the design process!

You don't have to do this in real life to study it, you can do it in an LTspice simulation.

There was thread a year ago about the use of damper from me, Jeremy G0HZU made a lovely video about this technique if that is what you meant:

https://www.youtube.com/watch?v=FsAq...ature=youtu.be

That was the thread:
https://www.vintage-radio.net/forum/....php?p=1203168

For a band-pass filter, each low-pass element is further transformed into a tuned circuit: an L becomes a parallel tuned circuit, and a C transforms directly to a series tuned circuit. In the vast majority of cases this results in an unrealisable filter, as the series circuit will contain unfeasible component values, and the L will, of course, have a parasitic C across it which can't be absorbed into another component.

This is where a bit of magic comes in: something called the j-inverter.
A quarter-wave transmission line, when shorted at the far end, looks like an open circuit at the other, and vice-versa. If we connect a series-tuned circuit to the output, it looks and behaves like a parallel tuned circuit at the input. What's more, we can transform the impedance slope = L-C ratio by changing the Zo of the transmission line.

Now, it is generally not convenient (nor practical) to use actual transmission lines, but a j-inverter can be constructed from a variety of networks, the caveat being that the transform is only valid over a limited frequency range.

The simplest and most used inverters (at these IF frequencies) are a) Mutual inductance and b) a PI network comprising

This is really abstract stuff to comprehend. I will leave the clever people to do all the hard work for me. I am more hands-on type of person perfering to get my hands dirty, messing around with electronics as a hobby. Unfortunately I also got a curious mind driving myself crazy.

You know children have fear of clowns.The poles and zeros evoke my irrational phobia of control engineering theory in my undergraduate years. But I keep bumping into these hard stuff whenever i start to ask questions about RF.

[Radio Wrangler]

There are two design routes into bandpass filters.

The obvious up-front one that turns up on a number of web calculators is to design a lowpass filter (which is really your wanted IF filter, but centred on zero hertz and extending equally into positive and negative frequency) which you then transform to your wanted centre frequency.

This isn't too bad to do. All the filters yu need are available in pre-computed tables, you just have to scale them to suit your choice of frequencies.

There is a snag. A big snag. As you try to make filters this way, going below 10% relative bandwidth you find some inductors start getting very big and are resonated with tiny capacitors - so you have trouble getting the stray C in the inductor low enough - inother words, the inductor's self-resonant frequency gets in the way. Other inductors become very small and are resonated with huge capacitors (for the frequency you're working at) and then lead inductance does you in. THEN you also have series-resonant tanks which you find are ultra sensitive to stray C to ground. A freacion of a pf can wreck the filter.

Transformed lowpass filters have their place, where widths are definitely above 5%, preferably above 10% of centre frequency. They fall apart if you go ultra-wide, so you use separate highpass and lowpass up there.

In the 10% and down, coupled resonator structures come into their own.

In Zverev, they come later, look for the tables entitled "3dB k and q" these allow you to think of a chain of resonators. The tables tell you what Q to load the end stages to, and what coupling coefficients to use between each pair of resonators.

With these, you can go very narrow as long as you can do resonators with enough Q

Zverev has a handy but little diagram of what filter styles are on song for different widths at different centre frequencies. It's hidden away and you have to find it.

Cross coupled filters to create zeroes are pretty much where you have to wing it with a simulator. My trick was to use mixed forms of coupling to create arithmetically symmetric filters. They're a lot better fit for superhet use than geometric symmetry.


David

[John_BS]

Mods; can you please correct my post 8 above: should read



...an L becomes a series tuned circuit, and a C transforms directly to a parallel tuned circuit.


I was looking at band-stop transforms!

John

Done... David

[ms660]

That schematic looks like a no frills IF strip and ratio detector to me, I would peak them up to 10.7MHz if that's supposed to be the frequency then tweek one one way and one the other way if you need to increase the bandwidth, forget trying to figure out the resonance points by coupling into the tuned circuit because it will have shifted again when you remove the 'scopes connection, typical probe input capacitance is several pF so at the frequency concerned it will have a considerable effect on the tuning.

Seem to remember some design procedures for FM IFT's in RDH4:

http://www.tubebooks.org/books/rdh4.pdf

Lawrence.

[John_BS]

I agree with Lawrence.

You have band-pass pairs operating at very high impedance; output terminated in c.100k and grid capacitance, input is the plate resistance of the pentode in parallel with the anode capacitance. Plus stray wiring capacitance everywhere. The only way to tune each stage is to leave it unmolested and measure its response after the following valve, having clobbered the following stage so it's no longer narrow-band. EG strap 1k across the following anode coil and measure across that. Have you opened the cans to see what capacitor values are used? I assume there are two tuning slugs, top & bottom adjust?

John

[regenfreak]

Have you opened the cans to see what capacitor values are used? I assume there are two tuning slugs, top & bottom adjust?

The kit comes with the ferrite cores uninstalled (see photo). You have to start from scratch to get the baseline positions of the slugs. The primary C1 is 22pF and secondary C2 = 10pF (C2 = 2 C1). Therefore it is an asymmetric double-tuned LC filter. The key point is that L2 is not equal to two times L1. It is not that simple. Q is proportional to L/C. Whoever designed the FM filter knew exactly what thay were doing for a good reason I suggested above. There is no instruction and no schematic diagram provided...

hat schematic looks like a no frills IF strip and ratio detector to me, I would peak them up to 10.7MHz if that's supposed to be the frequency then tweek one one way and one the other way if you need to increase the bandwidth, forget trying to figure out the resonance points by coupling into the tuned circuit because it will have shifted again when you remove the 'scopes connection, typical probe input capacitance is several pF so at the frequency concerned it will have a considerable effect on the tuning.

Yes it is very simple and classic American design. Thats exactly how i started out peaking them at 10.7MHz but i endeded up falling flat on my face. I soon realised that the standard approach of a mono FM using synchnonous tuning is useless in staggered tuning of FM stereo. I could end up like a dog chasing its tail when you have 6 slugs to permutate.

Zverev has a handy but little diagram of what filter styles are on song for different widths at different centre frequencies. It's hidden away and you have to find it.

For clever people who has an inquisitive mind to read and undersatnd Zerev:

https://ia803101.us.archive.org/20/i...0Synthesis.pdf

I have used an online calculator to work out what the impedance matching LC network would be (drawing) for the VNA measurement of resonance frequency. I am aware of the fact that the out-of-circuit LC filter in a VNA test rig would be very different from the dynamic behaviour after they are installed into valve FM tuner. But it would be an interesting thing to do. The problem is that the impedance of the LC varies sharply near the resonance frequency, therefore, there cannot be a perfect matched input or output network. Unlike the VNA sweep of ceramic filters that manufacturer would specfiy the input impedance in the datasheets.

ou have band-pass pairs operating at very high impedance; output terminated in c.100k and grid capacitance, input is the plate resistance of the pentode in parallel with the anode capacitance. Plus stray wiring capacitance everywhere. The only way to tune each stage is to leave it unmolested and measure its response after the following valve, having clobbered the following stage so it's no longer narrow-band. EG strap 1k across the following anode coil and measure across that. Have you opened the cans to see what capacitor values are used? I assume there are two tuning slugs, top & bottom adjust?

It is very diffcult and have to keep flipping the PCB around during the bottom and top slug adjustments. ..the valve transformer input has 550V input!!!! One must wear gloves and use isolation transformer. 1k is not enough isolation. I would say 100k to 1M.

I have not connected to the tuning magic eye came with it. As i said the stereo sounds audiophile quality, beating my DAB and SDR FM out of the water.

[John_BS]

Originally Posted by regenfreak

1k is not enough isolation. I would say 100k to 1M.

I've not made myself clear. Strap a temporary 1k across the primary coil of the pentode anode circuit which follows the filter section you want to tune. Don't get hung up on the fact that the inductors are not the same value. An air-cored coil, when tuned with a slug, will have a range of inductance ajustment of over two to one.

What you do need to be aware of is that either coil will have the "correct" inductance (to resonate) with the core in two positions: "still entering from the port of entry", and "gone past the middle of the coil and coming out the other end". If either core has done the latter, you will get a) an over-coupled filter, resulting in two peaks, not one, and b) probably difficulty tuning the other coil to the right frequency, and strong tuning interaction between the two coils.

Also bear in mind that the fixed capacitors need to have all the strays added which include coil self-cap, coil-to-can, grid and anode caps + wiring.

John

[regenfreak]

I've not made myself clear. Strap a temporary 1k across the primary coil of the pentode anode circuit which follows the filter section you want to tune. Don't get hung up on the fact that the inductors are not the same value. An air-cored coil, when tuned with a slug, will have a range of inductance ajustment of over two to one.
What you do need to be aware of is that either coil will have the "correct" inductance (to resonate) with the core in two positions: "still entering from the port of entry", and "gone past the middle of the coil and coming out the other end". If either core has done the latter, you will get a) an over-coupled filter, resulting in two peaks, not one, and b) probably difficulty tuning the other coil to the right frequency, and strong tuning interaction between the two coils.

Also bear in mind that the fixed capacitors need to have all the strays added which include coil self-cap, coil-to-can, grid and anode caps + wiring.

Ok your 1k resistor is equivalent to my "damper" consists of a series of 4.7K resistor and 2nF capacitor (see my drawing posted in the previous post). The 2nF capacitor is needed for dynamic tuning to isolate the HV DC.

I checked the Quad FM user manual. I vaguely recall they do static alignment with a 4.7K damper resistor without the series capacitor having the HT powered off.

Yes it is sensitive the coupling coefficient k. I have attached the equations for the two frequencies that I posted in this old thread:
https://www.vintage-radio.net/forum/...=162378&page=2

[peter_scott]

Another vote for Lawrence here. Given that you have a little spectrum analyser just try tiny iterations until you achieve the bandwidth and flatness that you desire. I have done this with a number of TV IFs to get 3MHz bandwith with well controlled step response.

Here are three different televisions tweaked for the above response. Takes patience but not difficult.

Peter

[regenfreak]

Originally Posted by peter_scott

Another vote for Lawrence here. Given that you have a little spectrum analyser just try tiny iterations until you achieve the bandwidth and flatness that you desire. I have done this with a number of TV IFs to get 3MHz bandwith with well controlled step response.

Here are three different televisions tweaked for the above response. Takes patience but not difficult.

Peter

How did you know I have a tiny spectrum analyzer?

The photos above is a Nano VNA V2. The attached photo device on the left is a TinySA spectrum analyzer. I use it for FM oscillator alignment all the time as it has an antenna that will not detune the RF oscillator.

NanoVNA + TinySA = "democratisation of the black art of RF ".making to accessible to newbie like me as an educational tool. Most of us could never dream of having a HP VNA or spectrum analyzer. Imagine thses cheapo chinese devices can open the world of discovery and wonder...

The challenge of staggered-tuning is the divergent problem; it is easy to get lost in a maze using trial and error.

PS It is easy to mix up the double humped response of two over-coupled LC tanks with stagger tuning.

[John_BS]

Ok your 1k resistor is equivalent to my "damper" consists of a series of 4.7K resistor and 2nF capacitor (see my drawing posted in the previous post). The 2nF capacitor is needed for dynamic tuning to isolate the HV DC.

No, as your drawing implies you're damping one half of the filter you're trying to align.

It's important that you don't in any way couple onto the pair you're trying to align. Drive the sweep signal into the grid of the preceding pentode. Monitor the output from the anode of the following pentode. This output is damped as you suggest to prevent the next pair altering the response, but all you need is the 2k2 across the primary (as the coil is s/c DC) and say 1nf @500V from the "hot" end of your resistor to feed whatever device you're using to measure RF voltage.
John
PS v approx expected state of such a filter

[peter_scott]

The first and third screen shot in my previous post were showing 3MHz flat bandwidth in a 16MHz IF. Admittedly the second one was a bit easier showing 3MHz from a 45MHz TRF

You've got it easy with only 250kHz in 10.7MHz.

Peter