6-gang FM stereo tuner heads

[regenfreak]

Originally Posted by Radio Wrangler

Congratulations! You've just discovered the path which if followed leads you into broadened band, low-Q distributed structures like horns.

An infinite number of infinitesimal value elements, but following a graded density. Sneaking up on them from the lumped-model side.

David

Cheers. These graded-density structures sound like something veneering into the twilight zone of microwave-horned deities in mythology.

[G0HZU_JMR]

Note that there are some tiny/unrealistic component values in those lumped element matching networks. This stuff would often be done using printed/distributed traces on a PCB up at 2GHz.

I think the Q of 1.5 can be done quite neatly using 5 lumped elements and the element values are much easier to realise with this network. The shunt/series inductors could also double up as the bias feed to the amplifier. It's also possible to do it with 4 elements with a Q of 1.5 but the first shunt capacitance value becomes quite small.

The network below has a wider response with a lower insertion loss and VSWR over a wider bandwidth. However, this might not be the design goal.

[regenfreak]

Originally Posted by G0HZU_JMR

Note that there are some tiny/unrealistic component values in those lumped element matching networks. This stuff would often be done using printed/distributed traces on a PCB up at 2GHz.

I think the Q of 1.5 can be done quite neatly using 5 lumped elements and the element values are much easier to realise with this network. The shunt/series inductors could also double up as the bias feed to the amplifier. It's also possible to do it with 4 elements with a Q of 1.5 but the first shunt capacitance value becomes quite small.

The network below has a wider response with a lower insertion loss and VSWR over a wider bandwidth. However, this might not be the design goal.

yes I noticed that too; two of the arcs are very short in my 6-element Smith chart which are dead giveaway but it was my first attempt to force through with 6 elements. It will start to get harder when Q<1. Those tiny component values are definitely realised by microstrips etc. It is possible to try the other alternative configuration by flipping the networks over.

Upon hearing graded-density, distributed structures for the first time, I have accidentally bumped into this scary book:


https://www.microwaves101.com/uploads/MYJ-part-2.pdf

[Radio Wrangler]

Originally Posted by G0HZU_JMR

The network below has a wider response with a lower insertion loss and VSWR over a wider bandwidth. However, this might not be the design goal.

You may have missed a trick by always arcing off to the same side and back to the real axis.

Try it with arcs to alternating sides and back to real. Let there be an equal number of arcs on each side. Make the ratiometric change in impedance be equal for each arc and back. So is the ratio of Z overall is x and you have y arcs, then the ratio of Z for each arc and return arc becomes the yth root of x.

Try this and watch the Smith chart as you sweep the frequency variable. The arcs to the opposite sides vary inevitably, but tend to cancel much of each others disruption. The end point lurks around your target Z over a wider frequency range. So you get a better breadth of match.

You can use Q arcs to decide how far your trips out each side go.

One of those tricks to keep up your sleeve for special occasions.

David

[G0HZU_JMR]

I used a similar ladder network in the drain circuit of the CPH3910 JFET LNA that I designed to vover the VHF FM band. JFETs generally have fairly low gm so then tend to get used in narrowband amplifier circuits. I had to design a wideband matching network in order to get about 12dB gain across the VHF band with a ~2dB noise figure.

I had a quick go at measuring the gain and noise figure with my 50 ohm spectrum analyser and the result is shown below. It looks like it is slightly detuned at the top end but probably still OK. This is partly due to the LNA being designed for a 75 ohm system. The design goal was to achieve a sensible gain (12dB) across the whole of the FM band whilst also achieving some selectivity against out of band signals. The noise figure is about 2dB when measured in a 50 ohm system as in the plot below. It should be about the same in a 75 ohm system but the gain response should be improved slightly when tested in a 75 ohm setup.

The LNA runs from +10V but would work anywhere from about +8V to +12V. It draws about 10mA from the power supply.

[G0HZU_JMR]

Originally Posted by Radio Wrangler

You may have missed a trick by always arcing off to the same side and back to the real axis.

Try it with arcs to alternating sides and back to real. Let there be an equal number of arcs on each side. Make the ratiometric change in impedance be equal for each arc and back. So is the ratio of Z overall is x and you have y arcs, then the ratio of Z for each arc and return arc becomes the yth root of x.

Try this and watch the Smith chart as you sweep the frequency variable. The arcs to the opposite sides vary inevitably, but tend to cancel much of each others disruption. The end point lurks around your target Z over a wider frequency range. So you get a better breadth of match.

You can use Q arcs to decide how far your trips out each side go.

One of those tricks to keep up your sleeve for special occasions.

David

Thanks, David. Can you post up an example network to show this?

[dmowziz]

Originally Posted by G0HZU_JMR

I used a similar ladder network in the drain circuit of the CPH3910 JFET LNA that I designed to vover the VHF FM band. JFETs generally have fairly low gm so then tend to get used in narrowband amplifier circuits. I had to design a wideband matching network in order to get about 12dB gain across the VHF band with a ~2dB noise figure.

I had a quick go at measuring the gain and noise figure with my 50 ohm spectrum analyser and the result is shown below. It looks like it is slightly detuned at the top end but probably still OK. This is partly due to the LNA being designed for a 75 ohm system. The design goal was to achieve a sensible gain (12dB) across the whole of the FM band whilst also achieving some selectivity against out of band signals. The noise figure is about 2dB when measured in a 50 ohm system as in the plot below. It should be about the same in a 75 ohm system but the gain response should be improved slightly when tested in a 75 ohm setup.

The LNA runs from +10V but would work anywhere from about +8V to +12V. It draws about 10mA from the power supply.

Please professor Jeremy? The S parameters

and can you post the schematic?

Thanks

[G0HZU_JMR]

In case anyone is interested, I'm using the classic Fritz Dellsperger Smith chart program here:

https://www.fritz.dellsperger.net/smith.html

This program has been around for many years. The demo version is limited to just five circuit elements which is a bit of a shame.

I can get a slightly better response with the revised network below but the centre capacitor value value reduces to about 0.35pF.

I can't show a six element network with this smith chart program, but I can draw in the sixth element by hand on the chart. See the diagram below. I think this is the kind of thing David was referring to.

[Radio Wrangler]

There's Lance Lascari's LLSmith programme. Last seen on this RFDUDE website. That's pretty useful, rater old in operating system terms, but isn't hobbled in terms of elements.

David

[regenfreak]

G0HZU_JMR
I can get a slightly better response with the revised network below but the centre capacitor value value reduces to about 0.35pF.

I can't show a six element network with this smith chart program, but I can draw in the sixth element by hand on the chart. See the diagram below. I think this is the kind of thing David was referring to.

I have tried your 5-element approach and got similar component values. But it shows a horrible lopsided sweep response.
My first attempt with the 6-element networks had a much better symmetric sweep response.

I didn't read the instructions of Simsmith and still managed to get it working. I have had very little experience using it.

[regenfreak]

Here is my 1st attempt to try Q<1 with 8 elements, following the intuition of the mole in the aforementioned Hunger Games; matching Z = 61-j90 to 50 ohm, centred at 1Ghz, sweeping from 500Mhz to 1.5GHz.

In the first arc, it is inside the Q-circle of 1.5. The other arcs are inside the Q=0.64 circles. The Q circles can be variable in the steps...

[regenfreak]

Here is a quickie multi-stage matching of a real broadband power amp based on the Z = 10.5 + 1.25j, Q = 0.85 of this random paper that I have come across, centred at 1.6G, sweep 0.5-2.5G:

https://link.springer.com/chapter/10...319-61382-6_21

The SWR looks a bit wonky. This one involves the crossing of the real axis.

[Radio Wrangler]

That's almost the trick. You get better frequency compensation with an equal number of arcs to each side of reality.

For a power amp, it's beneficial to do a lowpass section first, then there is less harmonic energy radiated by the later sections, and less gets induced in the final output. Also keeps loops carrying harmonic current smaller.

There are plenty of tricks.

David

[G0HZU_JMR]

Originally Posted by regenfreak

I have tried your 5-element approach and got similar component values. But it shows a horrible lopsided sweep response.
My first attempt with the 6-element networks had a much better symmetric sweep response.
.

OK but the circuit I gave was based on real world experience in terms of what could actually be transferred to a PCB using real lumped SMD components. The smallest inductance was 8nH and the smallest capacitance was 0.5pF and the network topology was chosen to make the design as realisable as possible. The network I provided also gave a bias feed connection for the amplifier, a dc blocking cap and an L match at the output that could all be absorbed into a real PCB layout such that the PCB interconnection pads would not upset the frequency response. A real PCB will have interconnection pads that have distributed properties. I chose a circuit topology that could be absorbed into a real PCB with minimal impact caused by the small interconnection pads on the PCB.

You need to think about those little connection pads on the PCB that connect those lumped parts together. The network I provided takes all this into account.

Try building the 2GHz networks on a real PCB and you will find that my original network will perform well. The inductances will have to be tweaked slightly and the 0.5pF cap might become 0.35pF but the network will still end up looking like the original network. The final L match could be flipped to a shunt C series L to give a better lowpass response if that was desired? It would also increase the bandwidth slightly but it would still give a fairly lopsided response.

[G0HZU_JMR]

Here's the 5 element network with a shunt C series L as the final two elements. This trades the low side rejection for a less lopsided response. I think I've copied your 6 element circuit across OK and I've compared the response out to 10GHz. Both simulations assume lumped components and no PCB layout effects.

Of course, there's no right or wrong network here because there is no spec target for symmetry or bandwidth. Sometimes asymmetry can be a good thing or a bad thing and having some extra bandwidth can be a good thing or a bad thing.

I think all of the shunt capacitors in your network would have to be replaced with printed shapes if it were built on a PCB. Both networks would lose bandwidth once transferred across to a real PCB layout so the network design might have to be revisited to correct for this.

I've changed the trace colours to green and blue in the plot below. This makes it easier to view.

[G0HZU_JMR]

If the PCB material (and material thickness) was suitable, the other matching option would be to use a simple 145 ohm PCB trace about 80degrees long at 2GHz. This would give a similar bandwidth to your 6 element LC network. It would probably be about 20mm long. The trace width would be quite narrow, especially if the PCB material was fairly thin.

It could be squished into a hairpin shape to save on PCB space. I think I would prefer to use lumped parts for this task though.

[regenfreak]

Originally Posted by G0HZU_JMR

You need to think about those little connection pads on the PCB that connect those lumped parts together. The network I provided takes all this into account.

Try building the 2GHz networks on a real PCB and you will find that my original network will perform well. The inductances will have to be tweaked slightly and the 0.5pF cap might become 0.35pF but the network will still end up looking like the original network. The final L match could be flipped to a shunt C series L to give a better lowpass response if that was desired? It would also increase the bandwidth slightly but it would still give a fairly lopsided response.

Here's the 5 element network with a shunt C series L as the final two elements. This trades the low side rejection for a less lopsided response. I think I've copied your 6 element circuit across OK and I've compared the response out to 10GHz. Both simulations assume lumped components and no PCB layout effects.

Of course, there's no right or wrong network here because there is no spec target for symmetry or bandwidth. Sometimes asymmetry can be a good thing or a bad thing and having some extra bandwidth can be a good thing or a bad thing.

I think all of the shunt capacitors in your network would have to be replaced with printed shapes if it were built on a PCB. Both networks would lose bandwidth once transferred across to a real PCB layout so the network design might have to be revisited to correct for this.

I've changed the trace colours to green and blue in the plot below. This makes it easier to view.

Thanks. Anyway, the impedance for the amplifier was a fictitious value used as an example for learning. Keysight ADS software uses an optimiser to take all the labour and guesswork out of humans. You can specify and reject small component values in the optimisation:

https://youtu.be/XpR6uoKYfF4

I have noticed that the series and parallel inductors combination allows tricks like crossing the real axis. Usually, it comes with the price of disrupting the symmetry of the frequency response. The L-network of the L and C pair tends to preserve the symmetry as they cancel each other disruption in each step of the multi-staging zig-zag walk.

Radio WranglerThat's almost the trick. You get better frequency compensation with an equal number of arcs to each side of reality.

For a power amp, it's beneficial to do a lowpass section first, then there is less harmonic energy radiated by the later sections, and less gets induced in the final output. Also keeps loops carrying harmonic current smaller.

There are plenty of tricks.

Cheers. If the reactive component is relatively small or close to the real axis, it is easier to do a zig-zag walk about the real axis with a low value of the Q-circle. On the hand, if there is a large reactive component for the amplifier and impedance matching becomes difficult in certain cases, it may call for asymmetric networks with a highly unbalanced number of L and C.

The Q may be divided into 3-section transformations with different resistive transformation ratios. So you start off the mole walk with a large Q, and the Q gets smaller in each section. They obtain different values of Qs from ADS optimiser based on purely resistive to resistive transformation. However, I got the equation

Q = SQRT [ SQRT(n) - 1 ) ] from somewhere that I can't remember where...

There is something called Chebyshev broadening technique that I have been messing around with the SimSmith.

[regenfreak]

I have tried to play with the three-section Qs aforementioned.

Here I have applied two approaches using a purely resistive source 5 ohms to 50 ohms, centred at 850MHz, sweep from 700MHz to 1Ghz.

Case 1, two sections, constant Q = 1.5 (attachment 1 and 2)

Case 2: three sections with Q1 = 1.68, Q2= 1.08, Q3 = 0.48 (attachment 3 and 4).

Initially, I got the Qs from the table in attachment 5 based on ADS simulator optimisation, coming from:

https://www.highfrequencyelectronics...09_Bichler.pdf

Resistive ratio is 50 to 5 = 10:1 in the table.

However, the Q3 of 0.58 has not worked out, and I have fudged the Q3 to 0.48 to get the mole back home. I guess how it works, right or wrong. Note that it is less symmetric than the two-section version with a fixed Q , more like a low pass...

The graphical method has taken away all the horrendous and tedious maths behind the transformations, and I can see the beauty of the whole thing.

[Radio Wrangler]

It's one of those things where you start to see the shape of the forest and where you're going, when before you just seemed to be surrounded by trees.

In an RF team I was once part of, there was a bunch of people forming a not-too-serious rock group. I suggested the name: "Crosby, Stills, Matthei, Young and Jones".

David

[regenfreak]

Originally Posted by Radio Wrangler

It's one of those things where you start to see the shape of the forest and where you're going, when before you just seemed to be surrounded by trees.

In an RF team I was once part of, there was a bunch of people forming a not-too-serious rock group. I suggested the name: "Crosby, Stills, Matthei, Young and Jones".

David

yes, off the beaten path, everything looks disorientating and confusing. Once someone can see the intuitive side of the navigating skills, there comes the clarity on the direction of travel. A mole has poor vision and yet it can burrow its way to the nest through complex networks. Isnt it an amazing creature?