6-gang FM stereo tuner heads

[regenfreak]

've got myself a few nice toys. Something quite off topic outside and a few spectrum analysers inside. The star is an HP 8566B 200Hz to 22GHz 10Hz to 1MHz BW The portable one I assembled myself, an HP8591E with tracking generator and a lot of options - some options which never got released for sale. This does 9kHz - 1800MHz with BW down to 10Hz (crystal) Does up to 5MHz BW and has a fast digitiser as well as AM and FM demods.

Yes I had a job I loved, but due to recent changes, the UK is no longer a member of any civil aviation certification body, so the firm has had to move most operations to a country which is still a member of a certification body.... there went my job! The IET reckon 600 companies were affected and have either moved out or closed, Rolls Royce aero engines being the best known, now a German business. So I never intended retiring, just a long tapering off. But t'was not to be. No discussion, please, the base cause is definitely OT for the forum.

Lucky you!!! I have given up the idea of owning an expensive spectrum analyzer. I dont want to spend money and I am contented with NanoVNAs, Tiny SA and NWT200. Cost of living crisis! Beggars can't choosy!! The attached screenshot shows the noise floor for S11 and S21 of my new toy NanoVNA V2 plus 4 which has the S21 noise floor of near -100db with 20-sample averaging at the low end.

I'm sorry to say that, but your interpretation of result is incorrect.
The cited formula gives an absolute value of insertion gain (here 0.49).
Insertion loss expressed in dB should therefore look as follows:

IL[dB] = -10 log {( 1 - QL/Qu)^2]} or
IL[dB] = -20 log ( 1 - QL/Qu).

The minus sign is to denote insertion loss as opposite of insertion gain.
For your data we have
IL = -20 log (1 - 0.3) = 3dB.
This is not a drama in terms of S/N ratio, as the noise present at the antenna terminal will experience the same treating.

Adam

I copied the equation from Pioneer TX-8500 technical notes service manual (see attachment) and later realised too it is wrong. So the Pioneer engineer made a typo in the equation

The high impedance inputs and outputs of a dual gate MOSFET have a strong influence on the sorts of resonator Q's you feel steered into using with them, and the antenna connection is the only low impedance around. So for a front end with 4 tuned coils and variable/varactor capacitors to tune them, although you can allocate the coils before and after an RF amplifier as wanted, the pattern of the bandwidths tends to be foisted on you.

I was experimenting with 5-gang varactor valve FM tuners a few months ago. It was a bit of "pain in the butt" to tune and get the tracking in synch. I left it aside and moved on to other projects.

I have been playing with VHF dual gate mosfet double-tuned amps lately. I am still learning to get them "well-behaved". I am new to dual gate mosfets. VHF bandpass behaves like cavity resonators and there is lots of things can catch me out unexpectedly when it comes to stray inductance and capacitance.

The attached photo is the variable hi_Z probes for my NWT-200. It does not work for VHF because I made the mistake of using Manhattan island pads that have a stray capacitance of a few pF to a ground plate. It is not a problem for HF frequencies but it becomes a low impedance path to the ground at VHF.

[regenfreak]

The 6-gang valve tuner works so well that it can pick up weak stations without an antenna. It can separate a very strong local station very close to distant station. Now this is before I do any critical alignments with hi-Z probes.

[nemo_07]

Originally Posted by regenfreak

The 6-gang valve tuner works so well that it can pick up weak stations without an antenna. It can separate a very strong local station very close to distant station. Now this is before I do any critical alignments with hi-Z probes.

Can you post a circuit diagram of this valve version https://www.vintage-radio.net/forum/...7&d=1653147111
? Clever build.
From the photo I can only infer a rough picture as follows:
single tuned input -> 1/2 ECC85 gain -> double tuned BP -> 1/2 ECC85 gain -> double tuned BP -> 1/2 ECC85 mixer + 1/2 ECC85 LO. Right?

You wrote elsewhere

For FM tuner, I would opt for slight under-coupling instead of the critical coupling with optimally flat bandpass response commonly used in fixed tuned HF filters. In the case of slight under-coupling, it introduces higher insertion loss but higher filter Q. A more round top response curve is desirable for FM tuner in adjacent channel selectivity. ... A small out of alignment introduces massive insertion losses. This makes alignment very challenging.

If you dispense with benefits of critical or sub-critical coupling, there is a point where a question arises whether a coupled BP still makes much sense.
For double tuned BP with a coupling factor of 0.5 or less the selectivity will approach that of two single tuned tanks of the same QL, separated some way, say by an isolating stage with 0dB gain, all that without incurring any insertion loss because of weak coupling. Additionally, such isolated single tank scheme would much better lend itself to alignments (no interacting).
BTW, for single layer air core inductors the optimal form factor D/L for highest possible Q lies in the range 0.7 - 1.
One valuable feature, exclusive to the magnetic coupling is symmetry, i.e. the -3dB edges of passband are equidistant to the centre frequency.

Before you start building another H-Z-not-so-Hi-Z probe be warned of versions like the widely spread on eBay "RF Active Probe 0.1 - 1500 MHz - 1.5 GHz Analyzer Oscilloscope", originally "Poor Man's 1-GHz" from Elektor (.pdf below).
Depending on the MOSFET type and layout it will show input capacitance of ~0.7pF (~1pF input "PCB trace" coupling capacitor) in series with the impedance Zg at the G1 (Zg = parallel connected ~1-2pF and ~10-50k Ohm), permanently mismatched output (Zout ~25 Ohm) and IL ~10dB (or worse, depending on output loading and frequency). Real part of Zg will drop with a 1/f² slope.
What you actually need is a probe with:
a) very low input capacitance (say <0.1pF) AND
b) well matched 50 Ohm output impedance AND
c) not too high and flat insertion loss,
d) usable BW ~1...100(up to 200-300)MHz.
This "Poor Man's" circuit, requires some mods and adds-on to cope.

Adam

[Radio Wrangler]

Originally Posted by nemo_07

If you dispense with benefits of critical or sub-critical coupling, there is a point where a question arises whether a coupled BP still makes much sense.

If you've got a chain of amplifiers with a chain of resonators, you don't necessarily treat pairs of coupled resonators individually, You might position their poles so that the complete RF chain creates a classical filter prototype like Butterworth or Bessel. I'm assuming you wouldn't want Chebyshev because of the large group delay excursions.

With good resonator design multi-section VHF/FM front ends can be a little narrow for the occupied channel bandwidth in terms of phase flatness. I have one Sony flagship tuner which uses a fast signal from the discriminator to make the RF tuned stages follow the modulated signal, flattening phase variation.

Once, gain was expensive. Now it's cheap, but selectivity and screening remain expensive.

David

[nemo_07]

Originally Posted by Radio Wrangler

With good resonator design multi-section VHF/FM front ends can be a little narrow for the occupied channel bandwidth in terms of phase flatness. I have one Sony flagship tuner which uses a fast signal from the discriminator to make the RF tuned stages follow the modulated signal, flattening phase variation.

Well, you could certainly achieve impressive figures with it in terms of disrortion in a laboratory setup with one and only FM signal.
However appealing this concept may seem to be, I don't like it. Still more, the idea of appliying it to the front end, instead of the IF strip looks to me like ... a bit funny gimmick.
Let's assume reception in the presence of both adjacent channels, both with slightly higher powers.
Now, every time your RF front end selectivity in chase after carrier approaches the edges of the channel band, the IF strip will see one of the neighbours, saying: "Hello neighbour, we live side by side for some time, and you never stop talking or playing records. But now I am speaking and you better stand back or keep quiet, OK?"
Will it stand back or quieten?

Nonetheless, I would be glad to learn the rationale standing behind it. Can you give me a hint/link?

Adam

[Radio Wrangler]

No links to anything much known, but it's a Sony 730ES. They seem to have good reputations, good reviews and mine works well, though I tend to leave the amplifier switched to the Revox B261 tuner which I know much better.

David

[regenfreak]

Apology for the late reply. I no longer get email notification of replies from this forum even I checked the option.

Can you post a circuit diagram of this valve version https://www.vintage-radio.net/forum/...7&d=1653147111
? Clever build.
From the photo I can only infer a rough picture as follows:
single tuned input -> 1/2 ECC85 gain -> double tuned BP -> 1/2 ECC85 gain -> double tuned BP -> 1/2 ECC85 mixer + 1/2 ECC85 LO. Right?

I posted the schematic in my first post number 1 of this thread. Indeed I replaced EC81 by ECC85. ( It is because the EC81 produces crazy harmonics with the plate B+ voltage over +19V. Below +19V it is stable and produces good oscillation. This is a mystery to me why EC81 misbehaves in this tuner as I used it many times in other DIY FM tuners.)

If you dispense with benefits of critical or sub-critical coupling, there is a point where a question arises whether a coupled BP still makes much sense.
For double tuned BP with a coupling factor of 0.5 or less the selectivity will approach that of two single tuned tanks of the same QL, separated some way, say by an isolating stage with 0dB gain, all that without incurring any insertion loss because of weak coupling. Additionally, such isolated single tank scheme would much better lend itself to alignments (no interacting).
BTW, for single layer air core inductors the optimal form factor D/L for highest possible Q lies in the range 0.7 - 1.
One valuable feature, exclusive to the magnetic coupling is symmetry, i.e. the -3dB edges of passband are equidistant to the centre frequency.

Indeed I spent ages to experiment and find the optimal form factor D/L that gives the minimal insertion loss for the 2nd order bandpass. The answer is not always clear cut. What works for HF does not mean it is the same at VHF. It seems the higher the the inductance of the coils, the higher the unloaded Q at FM broadcast band (which is opposite to what I expected). At VHF, everything is a compromise. Every component likes to couple to each other either inductively or/and capacitively. I always think of VHF bandpass filters as cavity resonators. Even I use partition walls to separate each stage of the bandpass filter, they are weakly coupled to each other (this can be easily demonstrated with a NanoVNA).

Before you start building another H-Z-not-so-Hi-Z probe be warned of versions like the widely spread on eBay "RF Active Probe 0.1 - 1500 MHz - 1.5 GHz Analyzer Oscilloscope", originally "Poor Man's 1-GHz" from Elektor (.pdf below).
Depending on the MOSFET type and layout it will show input capacitance of ~0.7pF (~1pF input "PCB trace" coupling capacitor) in series with the impedance Zg at the G1 (Zg = parallel connected ~1-2pF and ~10-50k Ohm), permanently mismatched output (Zout ~25 Ohm) and IL ~10dB (or worse, depending on output loading and frequency). Real part of Zg will drop with a 1/f² slope.
What you actually need is a probe with:
a) very low input capacitance (say <0.1pF) AND
b) well matched 50 Ohm output impedance AND
c) not too high and flat insertion loss,
d) usable BW ~1...100(up to 200-300)MHz.
This "Poor Man's" circuit, requires some mods and adds-on to cope.

Interesting, I have that cheap probe from Ukraine for two years but did not used it much. This mod is new to me. The hi-Z probe i mentioned have both adjustable input and output resistance using a series of three trimmer pots 1k, 10k and100K. It uses J310 and 2N3906. I built a few versions of it but the J310 always misbehaves in an unexpected manner at VHF, driving me crazy. I am in the middle of doing mod and optimising it for VHF

With good resonator design multi-section VHF/FM front ends can be a little narrow for the occupied channel bandwidth in terms of phase flatness

What are the physical meaning "phase flatness" and "group delay time" of a filter? For example, the data sheet shows group delay time plots of these 10.7MhZ Murata ceramic filters:

http://www.knap.at/datenblaetter/fil...-alt-p61e5.pdf



BTW I have been playing with a 5th-order Chebyshev bandpass filter of 0.1db ripple, cut-off at 90MHz and 110MHz. The attachements show my response vs the software simulation (https://rf-tools.com/lc-filter/).
My insertion loss is -0.6db versus -0.272 from ideal prediction. It is not bad! The practical difficulty is always the stray inductance and capacitance of the components. I probably do further fine-tuning by replacing the fixed coupling caps by piston trimmers.

Note that I got three S11 dips instead of fives dips in the ideal simulation. It is tricky to tune 5 resonators as the tuning is very sharp. The NanoVNA V2 plus 4 shows S21 dynamic range over 90db in the lower band stop region.

PS. here is a repair video and an interesting comments about the top-ended design features of the mighty 9-gang Kenwood KR 917. It uses saw IF filters and dual conversion.

https://youtu.be/orsah0Erv7w

Insane engineering designs!

[regenfreak]

The attached photos show that one of the weirdest oscillation waveform that I have come across from an oscillator! Before I replaced the EC81 by ECC85, the EC81 develops instability and crazy harmonics at normal plate voltage of 90-100V. I had to drop the plate voltage to 17V before the EC81 produces stable output. The valve was running at space charge voltage! I dont have an explanation for this. Maybe EC81 is a "hot rod", excitable valve or the oscillation coil couples inductively with the 5 other resonators before i installed the shields. I used silvered plated coils and the 6-gang variable capacitor are also silvered plated so the unloaded Q of the LC tank can be quite high.

At the moment, I get excellent stereo fidelity and sensitivity with the 6-gang tuner even it is not yet properly aligned.

[regenfreak]

from the photo I can only infer a rough picture as follows:
single tuned input -> 1/2 ECC85 gain -> double tuned BP -> 1/2 ECC85 gain -> double tuned BP -> 1/2 ECC85 mixer + 1/2 ECC85 LO. Right?

Please find the updated schematics.

It should be

single tuned input --> E88CC cascode amp-->double tuned BP--->E88CC cascode amp ---->double BP--->1/2 ECC85 mixer + 1/2 ECC85 LO

Now I use both gimmick wires for stage 1 and (0-1pF) piston trimmer for stage 2 of the coupling capacitors. Note that in the ideal bandpass design calculation using Zverev's method i assumed the input and output impedance are identical (symmetrical). In reality, the BF filters have asymmetrical input and output impedance with the cascode and mixer valves.

[Radio Wrangler]

Originally Posted by regenfreak

What are the physical meaning "phase flatness" and "group delay time" of a filter? For example, the data sheet shows group delay time plots of these 10.7MhZ Murata ceramic filters:

http://www.knap.at/datenblaetter/fil...-alt-p61e5.pdf

Ideally, for FM, the group delay should be constant across the band occupied by an FM signal's sidebands. Any variation translates into distortion of the demodulated audio.

Constant group delay is the same thing as a linear-phase characteristic in that same region. The slope of the phase/freq plot = group delay. so the phase slope can have ant value, but the plot must be a straight line. Deviation from straightness again makes audio distortion. I would have been clearer if I'd said straight rather than flat, sorry.

So posher ceramic filters will claim linear phase, but deliver something with less variation from true linearity than lower grade filters. Ceramic filters can be a bit off on centre frequency and you may need to match the things for best overall performance, also some seem to age.

That Revox tuner I have uses LC filtering in its IF, not a ceramic filter to be found. It also uses separate stages of IF limiter-amplifier. The LC filter can be adjusted to get the required bandwidth and still allow adjustment of phase linearity. The adjustment labour is a lot more expensive than ceramic filters, so this isn't a common approach. In any other company the bean counters would have such a fit of the heebie-jeebies that it would take months for their golf scores to recover.

David

[regenfreak]

Ideally, for FM, the group delay should be constant across the band occupied by an FM signal's sidebands. Any variation translates into distortion of the demodulated audio.

Constant group delay is the same thing as a linear-phase characteristic in that same region. The slope of the phase/freq plot = group delay. so the phase slope can have ant value, but the plot must be a straight line. Deviation from straightness again makes audio distortion. I would have been clearer if I'd said straight rather than flat, sorry.

Thanks. I have plotted the phase (Arg S21) and S21 for the 5th order Cheryshev bandpass and compared its with the phase and group delay from software. They look very similar and slopes of the phase are quite straight. But the group delay has two peaks and is far from flat. Bear in mind this is a wide band filter though (probably can be used for SDR FM reception with 50 ohms antenna input).

If slope of the phase/freq plot = group delay, it should have unit of degrees per unit Hz. In the rf-tools.com software, it uses time unit of ns. Dimensionally it is consistent as the inverse of frequency is time. NB. Note the large phase noise near the noise floor of S21 at -90-100db of the NanoVNA V2 plus 2. The plot has no sample averaging.

In any other company the bean counters would have such a fit of the heebie-jeebies that it would take months for their golf scores to recover.

David

[Radio Wrangler]

Looking at the spectrum picture in the second photo in post 48, that looks like your oscillator may be squegging, or if it's in a PLL, the loop could be unstable. The spacing of the spurs will tell you the frequency of the instability. The number of harmonics of the self-modulation says that something is hitting the end-stops, or going into complete cutoff.

David

[nemo_07]

Originally Posted by regenfreak

If slope of the phase/freq plot = group delay, it should have unit of degrees per unit Hz.

The phase is represented in radians, not in degrees. Also, the frequency omega is understood as 2*Pi*f [radians/second]. So the derivative of slope will have the unit [1/Hz] = [s].

Originally Posted by regenfreak

The attached photos show that one of the weirdest oscillation waveform that I have come across from an oscillator! Before I replaced the EC81 by ECC85, the EC81 develops instability and crazy harmonics at normal plate voltage of 90-100V. I had to drop the plate voltage to 17V before the EC81 produces stable output. The valve was running at space charge voltage! I dont have an explanation for this. Maybe EC81 is a "hot rod", excitable valve or the oscillation coil couples inductively with the 5 other resonators before i installed the shields. I used silvered plated coils and the 6-gang variable capacitor are also silvered plated so the unloaded Q of the LC tank can be quite high.

At the moment, I get excellent stereo fidelity and sensitivity with the 6-gang tuner even it is not yet properly aligned.

I would suspect, that the triode in the LO circuit, with its inherent Ca-g capacitance prefers to look at the 3uH at its anode as the perfect candidate for a Colpitts oscillator, ... and here we go.
So you should probably switch to Colpitts from the start.
... But now, if I see right the anode blocking capacitor to be of leadless type soldered directly to ground plane, I take back the explanation.

[regenfreak]

I would suspect, that the triode in the LO circuit, with its inherent Ca-g capacitance prefers to look at the 3uH at its anode as the perfect candidate for a Colpitts oscillator, ... and here we go.
So you should probably switch to Colpitts from the start.
... But now, if I see right the anode blocking capacitor to be of leadless type soldered directly to ground plane, I take back the explanation.

I thought exactly the same and these were the first obvious things I checked. I removed the 3uH choke as it was not really needed but it didn't cure the problem. I added double chokes for filament leads etc. All ground solder joins and RF bypass caps were triple checked. The EC81 is a solid UHF oscillator valve. In this chassis, the unstable oscillation stopped at plate voltage 17.5V and it still oscillated strongly and stably at plate voltage of 12V! Once I replaced EC81 by ECC85 without changing anything else, voila problem solved! I still think there is more to do with the weak coupling of the oscillator coil with other coils with the absence of shields ( I tested the oscillator with both E88CC removed.). Some UHF triodes are prone to self-excitation but i have many successfully experience of using EC81 as oscillator in FM homebuilt sets.

Looking at the spectrum picture in the second photo in post 48, that looks like your oscillator may be squegging, or if it's in a PLL, the loop could be unstable. The spacing of the spurs will tell you the frequency of the instability. The number of harmonics of the self-modulation says that something is hitting the end-stops, or going into complete cutoff.

These peaks have spacing of about 200kHz. As I said it is the strangest thing i have seen.

I read an article about the use of shorten transmission line in a Hartley oscillator for the Leak Troughline. So I built similar co-axial design but I have not got the chance to test it as I have been busy:
http://44bx.com/leak/troughline.html

[Radio Wrangler]

Oscillators are interesting beasties. Their power demands are not linear and not resistive. Start/stop phenomena can cause hysteresis and they can exhibit a negative resistance function. In other words, unless you've been careful with their HT feed decoupling, the little beggers can superimpose a low frequency relaxation oscillation on top of their RF oscillation. If you can trigger your scope on the LF component, you'll see the RF component turning on and off. This has long been known about and has the name 'Squegging'. It works the same way as the old neon bulb + capacitor + resistor oscillator, only with the RF oscillator playing the part of the neon.

There are some other possibilities, but this is the one to check first.

David

[regenfreak]

Oscillators are interesting beasties. Their power demands are not linear and not resistive. Start/stop phenomena can cause hysteresis and they can exhibit a negative resistance function. In other words, unless you've been careful with their HT feed decoupling, the little beggers can superimpose a low frequency relaxation oscillation on top of their RF oscillation. If you can trigger your scope on the LF component, you'll see the RF component turning on and off. This has long been known about and has the name 'Squegging'. It works the same way as the old neon bulb + capacitor + resistor oscillator, only with the RF oscillator playing the part of the neon.

There are some other possibilities, but this is the one to check first.

Thanks. Interesting! Squegging looks like "christmas tree" in the spectrum analyzer.
Indeed HT feed decoupling cap was my first prime suspect. I checked it repeatedly and it was fine.


Previously i encountered feedback, "sqealing" of the oscillator when the oscillator coil was physically too close to the mixer coil in a 3-gang FM tuner with cascode amp. I solved the problem by adding a grounded partition wall between the coils. A long time ago in my MW homebrew valve superhet radio, I encountered squegging and motorboating when I built the valve set using PCB strip board without a good RF ground plane. From there onwards, I avoid PCB strip board for my valve radio project completely.

this is a nice demo of the neon relaxation oscillator:
https://youtu.be/ZNYwIEouvxA

Here is a bit more about group delay for filters of different orders:

https://www.microwaves101.com/encycl...lay-in-filters

[Radio Wrangler]

Be a little careful. That link talks about some filters having more group delay than others.

More group delay as such is OK, it's just more delay from the transmitter to your lug holes, like moving a bit further from the transmitter. It is group delay VARIATION across the occupied bandwidth of the signal which does the damage.

Chebyshev functions try to make a maximally flat amplitude response at all costs. amplitude ripples are the main compromise, but the group delay variation is exaggerated a lot because of some poles being run at rather high Q to get the square shoulders of the cChebyshev response.

The linear phase filters try to get the straightest phase response (= flattest group delay) at all costs, and the cost you pay is in having a soft-shouldered amplitude response and not so good selectivity.

Digital filters can get around this compromise and allow almost independent control of amplitude and phase characteristics. To be honest, analogu filters can do this using a transverse filter topology with plenty of taps, but it's easier to realise this in the digital world.

If you look in the catalogue of normalised filter responses in zverev, you can easily compare phase and amplitude responses of all sorts of filter styles.

David

[regenfreak]

More group delay as such is OK, it's just more delay from the transmitter to your lug holes, like moving a bit further from the transmitter. It is group delay VARIATION across the occupied bandwidth of the signal which does the damage.

Chebyshev functions try to make a maximally flat amplitude response at all costs. amplitude ripples are the main compromise, but the group delay variation is exaggerated a lot because of some poles being run at rather high Q to get the square shoulders of the cChebyshev response.

The linear phase filters try to get the straightest phase response (= flattest group delay) at all costs, and the cost you pay is in having a soft-shouldered amplitude response and not so good selectivity.

Digital filters can get around this compromise and allow almost independent control of amplitude and phase characteristics. To be honest, analogu filters can do this using a transverse filter topology with plenty of taps, but it's easier to realise this in the digital world.

If you look in the catalogue of normalised filter responses in zverev, you can easily compare phase and amplitude responses of all sorts of filter styles.

I see.
Skipping all the maths, below paper said the only way a filter to introduce group delay is to store energy through reactive components like capacitors and inductors. A notch filter can have negative group delay. Does it mean a notch filter release stored energy? It is desirable to have "minimal group delay distortion" in communication system:

https://www.radio-labs.com/DesignFile/DN004.pdf


Bessel is a maximally-flat-group-delay linear filter with soft shoulders.
I must confess that I skipped 99% of Zverev's book and focused solely on a few pages on 2nd order Butterworth filter design method and the normalised filter tables. I just took the equations and put them into Excel spreadsheet to do my design calculations. I dont really understand all maths behind it and I rather leave it to someone else to figure them out. As I dont get paid to know about these things. It is an intimidating book!

When I tuned the 5th-order Chebyshev filter, I tried to flatten the ripples instead of making the dips (corresponds to the poles?) of S11 deeper-minimizing the reflected power loss.

I have seen SAW filters having similar S11 and S21 shapes as higher order Chebyshev filters. I found SAW filter mysterious like a little black box and they were only used in very few FM tuners like the Kenwood KR 917. These days, saw filters seem to be used often in Helium mining application. For example, see the S11 at 3:40:

https://youtu.be/FF8QenYx0GM

[nemo_07]

Originally Posted by regenfreak

... I tried to flatten the ripples instead of ...

The flatnees of group delay characteristics of a filter has in general little to nothing in common with the flatness of its S-parameters characteristics.
Of the three celebrated generic filter types, the Chebyshev offers the worse flatness of group delay, as Radio Wrangler already pointed out.
Anyway, you should perhaps consider to open another thread for discussion on your filter synthesis as mixing it together with the 6-gang FM project introduces some confusion and unnecessary distraction.

The group delay of basic parallel LC filter (i.e. with lumped L and C elements) is not related to the real time delay encountered (which is null), but to its bandwidth. It is merely a calculated value, used for analysis in most modern modulation schemes, where phase linearity is of merit.
Group delay (precisely its flatness) is relevant to FM systems, independent of filter types used. If you need details, I could put some, but it doesn't go without a bit of basic maths.

(The linked radio-labs page https://www.radio-labs.com/DesignFile/DN004.pdf has highly entertaining qualities, especially the attempt to explain the term "group delay" exploiting the the idea of a filter "waiting" before making decision upon what to pass and what not. And then they found a "negative group delay" through a parallel LC tank and tell, you can't violate the causality principle, yet they didn't mind to try it. However, they would fail though, because their "negative group delay" is a mirage. So, you've had some fun, and the world is saved )

Turning back to the roots ...
It would be interesting to see few sweeps of your current front end' response (from ANT in to IF out), preferably ~1-2(max)MHz wide, in three points of band, say 94MHz, 98MHz and 102MHz, with input level ~100uV (for gain & selectivity check), and few sweeps across the whole band, 88-108MHz, with unit tuned to points as above, with input level ~200mV (for distortions check).
To preserve the selectivity add 25 Ohm in series on the generator hot side to match its 50 Ohm output to ANT input, and for the same reason, another series resistor to the IF out hot side, as required (value depending on IF transformer ratio; ~50-150 Ohm should do).
Can you manage it?

[regenfreak]

The flatnees of group delay characteristics of a filter has in general little to nothing in common with the flatness of its S-parameters characteristics.
Of the three celebrated generic filter types, the Chebyshev offers the worse flatness of group delay, as Radio Wrangler already pointed out.
Anyway, you should perhaps consider to open another thread for discussion on your filter synthesis as mixing it together with the 6-gang FM project introduces some confusion and unnecessary distraction.

The group delay of basic parallel LC filter (i.e. with lumped L and C elements) is not related to the real time delay encountered (which is null), but to its bandwidth. It is merely a calculated value, used for analysis in most modern modulation schemes, where phase linearity is of merit.
Group delay (precisely its flatness) is relevant to FM systems, independent of filter types used. If you need details, I could put some, but it doesn't go without a bit of basic maths.
(The linked radio-labs page https://www.radio-labs.com/DesignFile/DN004.pdf has highly entertaining qualities, especially the attempt to explain the term "group delay" exploiting the the idea of a filter "waiting" before making decision upon what to pass and what not. And then they found a "negative group delay" through a parallel LC tank and tell, you can't violate the causality principle, yet they didn't mind to try it. However, they would fail though, because their "negative group delay" is a mirage. So, you've had some fun, and the world is saved )

Thanks. I was scratching my head after reading the bits in the article https://www.radio-labs.com/DesignFile/DN004.pdf.
Presto! Negative group delay means time travel.

urning back to the roots ...
It would be interesting to see few sweeps of your current front end' response (from ANT in to IF out), preferably ~1-2(max)MHz wide, in three points of band, say 94MHz, 98MHz and 102MHz, with input level ~100uV (for gain & selectivity check), and few sweeps across the whole band, 88-108MHz, with unit tuned to points as above, with input level ~200mV (for distortions check).
To preserve the selectivity add 25 Ohm in series on the generator hot side to match its 50 Ohm output to ANT input, and for the same reason, another series resistor to the IF out hot side, as required (value depending on IF transformer ratio; ~50-150 Ohm should do).
Can you manage it?

I am not interested in making sweeps from antenna input to IF out because my FM IF strips use complicated staggered tuning (in contract to synchronous tuning in mono FM or AM IF strips) to max out the stereo fidelity. The IF responses will mask out the misalignment issues in the individual single and double tuned RF front end stages. It is easy to do frequency sweep of IF stages but it is hard to do visual sweep on the FM front ends.

I would not use 50 input injection from signal gen or sweep generator. I am in the middle of winding my own 1:6 balun (50 unbalanced to 300 balance antenna input).

The most challenging part of the project is to align the double-tuned filters in the cascode stages (powered-on) using 50 ohm instrument. It does not matter if I have a £50 or £3000 spectrum analyzer, the challenges remains the same. In Zverev's method, there are two solutions for the same 2nd order Butterworth filter: 1) one with I/O coupling caps matched to 50 ohms source and load 3) one with high impedance I/O terminations. I have 50 matching ports soldered to the I/O of the bandpass filters of the FM tuners. ( it is also possible to match 50 ohms by tapping the I/O to the coils near the ground ends) I have managed to sweep the bandpass filter S11, S21 and SWVR response using the NanoVNA V2 Plus 4 with both power on and off. However, they may not represent the actual response curve due to the fact that the I/O of the cascode stages do not necessarily have the same I/O impedance of LC bandpass. I have tried high Z probes and variable Z probes, they have not been working well. I am still searching for the solutions. If i am able to crack this, I will be on my way to Nirvana. The other thing is that the 6-gang tuner is already working very well, I am reluctant to touch it further. Sometimes something is working well and you mess it up by changing things around, you can't get back where you were easily (Murphy's Law).

At the moment, I am messing around with dual gate FETs and Gali 51+. I can see the beauty of having bandpass filters with 50 ohms I/O impedance that I can characterise them easily with VNA.