6-gang FM stereo tuner heads

[Radio Wrangler]

If you're doing a VCO and are looking for best RX performance, a trick is to do the VCO for double the frequency you first thought of, and use a logic divide by two. This can give a more symmetrical waveform to the mixer and give you an extra dB or so on TOI.

Having a symmetrical mixer driver with long-tailed pairs that apply a current squarewave to a diode ring mixer transformer is another step along this route.

I used to work with Hugh Walker ("Sources of intermodulation distortion in diode ring mixers" Proc. IRE in the early seventies) and I picked up a few tricks.

David

[regenfreak]

Originally Posted by Radio Wrangler

If you're doing a VCO and are looking for best RX performance, a trick is to do the VCO for double the frequency you first thought of, and use a logic divide by two. This can give a more symmetrical waveform to the mixer and give you an extra dB or so on TOI.

Having a symmetrical mixer driver with long-tailed pairs that apply a current squarewave to a diode ring mixer transformer is another step along this route.

I used to work with Hugh Walker ("Sources of intermodulation distortion in diode ring mixers" Proc. IRE in the early seventies) and I picked up a few tricks.

David

I see. With sinusoidal input of LO and RF, the diodes have a turn-on switching voltage. There will be dead spot near the zero crossing on the waveforms of the mixing products when the diodes are momentarily off. As a result, the waveforms are not as symmetrical as the theoretical mixing output created by square waves. The ring diodes work on the non-linearity behaviors of the diodes (seems paradoxical!). The waveforms of the mixer output look like the spikes on the back of Puff the Magic Dragon with superposition of high and low frequency harmonics contents.

I have attached the wide and narrow spectrum IF output of a +7dbm Mini-circuit ring diode mixer (with FY6900 sig gen).

RF input f1 = 88MHz at -30dbm,
LO f2= 98.7MHz at +7dbm,
and IF = 10.7MHz.

I can see the intermodulation products from some of the markers( I am running out of markers to label every one of them! So I have attached two plots with wide and narrow span):

M3: 2f2-f1= 109.4MHz, 3rd order
M5:2f2 = 197.4MHz, 2nd order
M7: 2f2+f1 = 285MHz, 3rd order
M8: 3f2=296MHz, 3rd order

I would expect 3rd order product, 2f1-f2 = 77.3MHz but I get 62.87MHz instead (see Marker 2 in the 2nd attachment). I dont know why....
I have also attached the harmonics measurements at the IF output of ring diode mixer. The 2nd order harmonics is -63dbc.

At the beginning of this year, I constructed 5-gang varactor tuned FM valve tuner with a pair of E88CC double triodes and a Mini-circuit ring diode ring mixer like the Mrantz 10B schematic attached. However, I struggled with the non-linearity and RF/LO tracking of the varactors and put the project on the back burner. The output port of the mixer is 50ohm, any impedance mismatch will cause reflection of power back to mixer creating more intermod junks.

While the use of ring diodes are common in QRP homebrewers, they are rarely used in commercial broadcast FM tuners. The notable ones are the mega expensive big boys Marantz 10B and Kenwood K917.

I have been toying with the idea of building 7-gang, air variable capacitor tuning, FM tuner using a double balanced mixer with a pair of dual gate mosfets. However, it requires very high drive power which calls for a buffer amplifier for the oscillator and impedance matching baluns or LC matching network. It starts to get a bit challenging to implement in a homebrew project.

[Radio Wrangler]

A diode is a one-port device. Mixers rely on the LO being much larger than the mixture of signals, so that the LO dominates the signals and drives the diodes as much like switches as possible.

The period while the diodes are transitioning between the 'as much off as the drive allows' and the 'as much on as the drive allows' is when they are most amenable to having their switching time modulated by incoming signals. Hence intermodulation.

So we want to get the diodes to go as quickly as possible between the saturated states. If you read data from packaged diode ring mixer manufacturers, they offer mixers with better intermod performance at higher LO drive levels. Essentially they employ measures to increase the LO level at which damage occurs and you have to provide the LO power.

All this assumes sinewave LO.

But the benefit of the higher power becomes more a case of the speed in crossing the active region and less in how hard they are turned on/off.

Having a squarewave drive has the faster dV/dt to get across the transition, but doesn't apply as much heat and stress as a sine with a comparable dV/dt.

Making the LO drive a current source rather than a voltage source also helps manage stress and can lead to a faster transition.

So that's how to soup up a diode ring mixer to the hilt. Guy Douglas wrot it up in the HP journal in April 1982

http://vtda.org/pubs/HP_Journal/HP_Journal_1982-04.pdf

Racal had done a mixer for their RA1772 HF receiver based on a 1968 paper by P R Rafuse which embedded DMOS FETs in a ring like the diodes used. In this they missed a massive trick. The MOSFETs are TWO port devices and give separation of LO and signal paths. There is no need to use the circuit topology optimised for one-port devices.

Several people eventually saw the next step. A colleague at HP came up with it as a suggestion and it got used. A little later someone else contributed it to the 'Tech Topics' column in RADCOM. He named it the 'H-mode Mixer' and it's worth looking at.

David

[regenfreak]

Originally Posted by Radio Wrangler

A diode is a one-port device. Mixers rely on the LO being much larger than the mixture of signals, so that the LO dominates the signals and drives the diodes as much like switches as possible.

The period while the diodes are transitioning between the 'as much off as the drive allows' and the 'as much on as the drive allows' is when they are most amenable to having their switching time modulated by incoming signals. Hence intermodulation.

So we want to get the diodes to go as quickly as possible between the saturated states. If you read data from packaged diode ring mixer manufacturers, they offer mixers with better intermod performance at higher LO drive levels. Essentially they employ measures to increase the LO level at which damage occurs and you have to provide the LO power.

All this assumes sinewave LO.

But the benefit of the higher power becomes more a case of the speed in crossing the active region and less in how hard they are turned on/off.

Having a squarewave drive has the faster dV/dt to get across the transition, but doesn't apply as much heat and stress as a sine with a comparable dV/dt.

Making the LO drive a current source rather than a voltage source also helps manage stress and can lead to a faster transition.

So that's how to soup up a diode ring mixer to the hilt. Guy Douglas wrot it up in the HP journal in April 1982

http://vtda.org/pubs/HP_Journal/HP_Journal_1982-04.pdf

Racal had done a mixer for their RA1772 HF receiver based on a 1968 paper by P R Rafuse which embedded DMOS FETs in a ring like the diodes used. In this they missed a massive trick. The MOSFETs are TWO port devices and give separation of LO and signal paths. There is no need to use the circuit topology optimised for one-port devices.

Several people eventually saw the next step. A colleague at HP came up with it as a suggestion and it got used. A little later someone else contributed it to the 'Tech Topics' column in RADCOM. He named it the 'H-mode Mixer' and it's worth looking at.

David

Thanks. The paradoxical nature of ring diode mixers is that they are supposed to improve "linearity" even they work through the non-linearity mixing action of fast switching diodes.

I have found both Rafuse's IEEE paper and RA1772 ring diode-DMOS schematic. I have only skipped through the articles quickly. Rafuse mentioned that the LO current should be as large as possible to force the series diodes into their linear region. It can be seen from the attached figure LO voltage vs finite rise time from 1982 HP paper that the bigger the slope dV/dt, the less phase modulation.

I cringe every time when I see the terms "current source" is being mentioned. An ideal current source has infinite slope in the V vs I graph (or infinite resistance). Many active transistor configurations can be regarded as a current source with high output impedance with relatively constant current. In this context, the hard switching with square waves (dV/dt = infinity) behaves a constant LO current source. Rafuse wrote that the non-linearity of ring diodes during the OFF state is responsible for ALL of the intermodulation if the IF and LO ports are perfectly matched. The use of 1:1 Baluns are for RF and LO isolation, reducing the intermodulation products.

In some way, this is almost opposite to the zero current switching of IGBTs at the zero crossings of Double Resonance Solid Tesla Coils( the OneTesla which I still own) but the goal is the same to reduce power losses:

https://onetesla.com/tutorials/intro..._store=default

[Radio Wrangler]

Consider a simple switch.

It's very linear to the signals which it switches

It's very non-linear from the position of the toggle to the effect on the signals which are applied to the contacts.

Semiconductor manufacturers have never had life easy deciding whether to put mixers in their linear or the non-linear databooks

Of course a pure analogue multiplier is a linear function, but look inside at the non-linear functions needed to implement it! Usually a true Gilbert Cell.

David

[regenfreak]

Originally Posted by Radio Wrangler

Consider a simple switch.

It's very linear to the signals which it switches

It's very non-linear from the position of the toggle to the effect on the signals which are applied to the contacts.

Semiconductor manufacturers have never had life easy deciding whether to put mixers in their linear or the non-linear databooks

Of course a pure analogue multiplier is a linear function, but look inside at the non-linear functions needed to implement it! Usually a true Gilbert Cell.

David

Sounds like an existential identity crisis at the manufactures' data book division.

Perhaps it is easier to visual the issues in time domain. My scopes are not good at triggering wonky mixing product waveforms at low voltage levels from the Mini-circuits SBL-1+ mixer. I have tried to mimic an ideal, hard switching of logic function +1 and -1 using the Math function of my Rigol scope. I input channel A, RF sinewaves at 14MHz and channel B, LO square waves at 10MHz, respectively. You see the low frequency 4MHz IF envelope of the Math multiplication operation: channel A x channel B . There is no way to hide the sine and square waves in the scope. In a real ring diode mixer, you would see clamped positive peaks at the zero crossings of the undulating mixer waveforms when the diodes are off.

I have used my Peak DC pro curve tracer to plot the diode current versus the forward bias voltage by measuring one of the diodes at the IF port of the Mini-circuits SBL-1+. I can say the slope of I/V curve is roughly "linear" when Vf > 0.28V. Of course, there are two diodes in series during switching on and off cycles.

The Peak DC pro incorrectly detects it as Zener diode but I think they have to be Schottky diodes.

[G0HZU_JMR]

About 30 years ago, Minicircuits released some software that could plot the predicted spurious levels of their range of mixers. I still have our works copy of it even though it dates back to the dark days of DOS. In the plots below I've compared my measurement of an ancient SBL1-1 mixer against the Minicircuits SW.

The SW only allows a choice of two RF drive levels, so I chose -10dBm. The RF frequency is 88MHz and the LO frequency is 98.7MHz.

The analyser plots have a start frequency of 8MHz and a stop frequency of 300MHz in both cases. You can see that the old SW agrees quite well with the real measurement.

Note that the amplitude scale of the Minicircuits plot is normalised to the IF output level. Therefore, 0dB on the Minicircuits plot is really -16dBm and this agrees well with the real measurement.

[regenfreak]

Originally Posted by G0HZU_JMR

About 30 years ago, Minicircuits released some software that could plot the predicted spurious levels of their range of mixers. I still have our works copy of it even though it dates back to the dark days of DOS. In the plots below I've compared my measurement of an ancient SBL1-1 mixer against the Minicircuits SW.

The SW only allows a choice of two RF drive levels, so I chose -10dBm. The RF frequency is 88MHz and the LO frequency is 98.7MHz.

The analyser plots have a start frequency of 8MHz and a stop frequency of 300MHz in both cases. You can see that the old SW agrees quite well with the real measurement.

Note that the amplitude scale of the Minicircuits plot is normalised to the IF output level. Therefore, 0dB on the Minicircuits plot is really -16dBm and this agrees well with the real measurement.

Thanks. Thats very interesting. Perhaps, your DOS relic is a reminder that not everything was rosy back in the 80s ( I remember I used the BBC Micro computer, ICL mainframe, 5.23" floppy disk, waist bumbag and DOS in those days).

Regarding the SBL-1+ mixer measurement, I was like an occult numerologist, trying to identify the intermodulation products from the spectrum markers. I was permutating n*f1+/-m*f2 until my head started to spin and gave up.

[regenfreak]

I have bumped into this "Mixer Spur Web" chart used in pre-PC days equivalent to the paper Smith charts:

https://www.rfcafe.com/references/el...r-spur-web.htm

The following link explains why 10.7MHz IF is used simply based on the compatibility condition that the image is always lies outside the broadcast band 87-108.5MHz regardless of the LO frequency being above or below the IF:


https://www.radartutorial.eu/09.receivers/rx06.en.html

The attached figures (from Rohde and Schwarz) explain why modern spectrum analyzers often use high IF for up-conversion in the first mixer stage (I was wondering why). By making the IF higher than the input frequency range, the input frequency and LO range do not overlap. Therefore, it is easy to get rid of the images at the top range using fixed low pass filters (e.g. bow ties filters with distributed LC elements ). On the other hand, it is very hard to design tunable bandpass filters to reject the images when input, LO and images are all overlapping with low IF.

[Radio Wrangler]

Without up conversion and wanting general frequency coverage down to almost DC, you wind up with the LO crossing the IF frequency and giving you a monster spur somewhere in your frequency coverage. Trying to isolate LO to IF to the required extent is impossible.

With the up conversion scheme, you still get the spur, but it is located precisely at 0Hz. Everyone using spectrum analysers knows it's there, but not all know why. You could looka at it as a true response to DC, in this case a response to DC ofset/imbalance in the mixer, which is one component in the leakage path, but not the only one.

David

[regenfreak]

Originally Posted by Radio Wrangler

With the up conversion scheme, you still get the spur, but it is located precisely at 0Hz. Everyone using spectrum analysers knows it's there, but not all know why. You could looka at it as a true response to DC, in this case a response to DC ofset/imbalance in the mixer, which is one component in the leakage path, but not the only one.

David

I didnt know. Most spectrum analyzers/VNA cannot go down to zero Hz but may be lower limit around 9kHz due to the low frequency limitations of the first stage mixer, directional coupler, the DC blocker cap etc. at the input.

Let's say I have a hypothetical spectrum analyzer can go down to 0Hz with high side conversion, assume

Input RF range: 0 to 3GHz
IF = 3.6GHz,
LO range: 3.6 to 6.6GHz

The mixing products would be:

mLO+nRF and

mLO-nRF (LO > RF)

mLO+/-nRF are families of straight lines with slopes m or n, and x- or y- intercepts either mLO or nRF, by permutating different values of n and m.

If you plot output (mixing product freq ) vs input frequency, you will get families of straight lines. Or you call it Spur Web in numerologist's jargon.

For the low side of the mixing product, they would be mLO-nRF;

Likewise for the high side: mLO+nRF

When n =1 and m =1, this corresponds the desired IF frequency for the difference plot; 1xLO-1xRF= IF for high sideband conversion IF.
Any products other than m=1 and n=1 are unwanted junks.

I have played with the Marki online spur and spectrum analyzer spur plotters and got the attached results.

https://www.markimicrowave.com/spur-calculator.aspx

In the label. LO0xINPUT1 means m = 0, n =1, -26dBC spur power with 10dbm mixer input.
In the third attachment, I just put RF = 1G and LO = 4.6G, I can see some strong spurs below 3G in the input RF range.

[Radio Wrangler]

9kHz is a value decided upon as both a lower limit for EMC measurements which were legislated in Germany and later became European requirements. Also 9kHz is one of the specified EMC measuring bandwidths, though there is no causal relationship.

The 9kHz limit is essentially the DC block capacitor. Some analysers don't have one and their first mixer really does go down to DC. You have to be very careful with these instruments like the HP8568A/B

How low you can go without a DC block is a function of your narrowest resolution bandwidth allowing you to resolve close to the IF crossing (zero hertz response) as well as the LO phase noise at that offset/IF bandwidth. THe 8568 is spec'd down to 200Hz.

David

[regenfreak]

Originally Posted by Radio Wrangler

9kHz is a value decided upon as both a lower limit for EMC measurements which were legislated in Germany and later became European requirements. Also 9kHz is one of the specified EMC measuring bandwidths, though there is no causal relationship.

The 9kHz limit is essentially the DC block capacitor. Some analysers don't have one and their first mixer really does go down to DC. You have to be very careful with these instruments like the HP8568A/B

How low you can go without a DC block is a function of your narrowest resolution bandwidth allowing you to resolve close to the IF crossing (zero hertz response) as well as the LO phase noise at that offset/IF bandwidth. THe 8568 is spec'd down to 200Hz.

David

I see the Germans did the RF hobbyists lots of favor by keeping the audiophiles out. Anything the audiophines touch, they will drive up the prices...see what have happened to valves and audio amps.

Houston i have a problem here, the maths does not add up in the plots, if OP is the output frequency, the equation of the straight line would be:

OP = -nRF + mLO
-n is the slope , mLo is the y-intercept.

For input RF = 0 hz and m = 1, OP = mLO=3.6GHz. In the graph, the lines all intercept at 6GHz, I am confused now and need to go away to do some exercises to clear my head.

[regenfreak]

Houston i have a problem here, the maths does not add up in the plots, if OP is the output frequency, the equation of the straight line would be:

OP = -nRF + mLO
-n is the slope , mLo is the y-intercept.or input RF = 0 hz and m = 1, OP = mLO=3.6GHz. In the graph, the lines all intercept at 6GHz, I am confused now and need to go away to do some exercises to clear my head.

Ok, the muppet overlooked the LO slider. Now the numbers add up.

This old paper based web works similarly except they non-dimensionalize by dividing the output OP and RF frequencies by LO:


https://www.rfcafe.com/references/el...r-spur-web.htm

OP = mLO + nRF

Op/LO = n RF/LO + m

n = slope

m = y-intercept

[regenfreak]

Originally Posted by Radio Wrangler

THe 8568 is spec'd down to 200Hz.

David

In the spec: the bandwidth selectivity:
60 dB points on 10 Hz bandwidth are separated by < 100 Hz.

How was it possible to make a physical bandpass filter for RBW = 10Hz before the implementation of digital IF filters? I would love to see a photo of the IF bandpass filters of the HP 8568 on the circuit board.

[regenfreak]

On page 434 of the HP 8563 service manual, it seems to be crystal IF filters at the smallest RBW 300Hz-10kHz:

https://www.testunlimited.com/pdf/an/08563-90214.pdf

[G0HZU_JMR]

If you plot out the various mixer spurious terms for a typical (upconverting) spectrum analyser you should find that the main problematic mixer terms are:

(0*LO + n*RF)
(1*LO - 2*RF)

Other major terms include (n*LO - n*RF) although the two terms above are usually dominant.

I can remember using the 'spider's web' mixer charts about 30 years ago. I didn't find them very intuitive to use. It can all make sense on the day you draw them, but it requires some effort to process the results when the plots are re-visited a few months later.

Back in those days, the company purchased some mixer analysis SW from Synergy Microwave called SPECT and LOCUS. This was also DOS based but it was quite powerful. SPECT was very similar to the Minicircuits program I showed earlier, but LOCUS could present data across the whole of the RF and LO ranges. I still use LOCUS occasionally although it is quite clunky to use.

Back in about 2005, Agilent added 'WhatIF?' to the Genesys CAD suite. This is meant to assist in the design of frequency converters. 'Spectrasys' also proved fairly powerful system analysis within that suite.

However, for a given frequency plan, it's often more powerful to write your own analysis tools. I've created various analysis tools for dual and triple up/downconverter design work. It's also possible to predict the internal spurious signals or 'birdies' that occur in such systems.

[regenfreak]

Originally Posted by G0HZU_JMR

If you plot out the various mixer spurious terms for a typical (upconverting) spectrum analyser you should find that the main problematic mixer terms are:

(0*LO + n*RF)
(1*LO - 2*RF)

Other major terms include (n*LO - n*RF) although the two terms above are usually dominant.

I can remember using the 'spider's web' mixer charts about 30 years ago. I didn't find them very intuitive to use. It can all make sense on the day you draw them, but it requires some effort to process the results when the plots are re-visited a few months later.

Back in those days, the company purchased some mixer analysis SW from Synergy Microwave called SPECT and LOCUS. This was also DOS based but it was quite powerful. SPECT was very similar to the Minicircuits program I showed earlier, but LOCUS could present data across the whole of the RF and LO ranges. I still use LOCUS occasionally although it is quite clunky to use.

Back in about 2005, Agilent added 'WhatIF?' to the Genesys CAD suite. This is meant to assist in the design of frequency converters. 'Spectrasys' also proved fairly powerful system analysis within that suite.

However, for a given frequency plan, it's often more powerful to write your own analysis tools. I've created various analysis tools for dual and triple up/downconverter design work. It's also possible to predict the internal spurious signals or 'birdies' that occur in such systems.

yes the spur web can be confusing to read, making me feel like a fly trapped in a spider web.... kind of experience like listening to my arty-farty friend reading my tarot cards' prediction.

1*LO-2*2RF is one of the troublesome 3rd order intermodulation products. They grow by the slope of 3:1.(n+m)th is the order of mixing products.

To make matter worst, consider the IF down conversion of two tones RF1 and RF2 (RF > LO), these 3rd order products are the troublemakers with m=1, n = +/-1 or n= 2:

low side:
2RF1-RF2-LO
2RF2-RF1-LO
high side:
2RF1+RF2-LO
2RF2+RF1-LO

then the 2nd order
2RF1-LO
2RF2-LO

[G0HZU_JMR]

Yes, second and third order distortion terms tend to dominate real-world performance.

Most lab spectrum analysers will manage a mixer input IP3 of about +20dBm with 10dB input attenuation. A really good one might manage +30dBm. The noise floor might be -100dBm with a 10kHz RBW and 10dB front end attenuation.

The SFDR would then be 0.666*(20 - -100) = 80dB.

The second harmonic intercept might be +60dBm with 10dB input attenuation. The SFDR here would therefore be about 0.5*(60 - -100) = 80dB.

These were the classic benchmark numbers used at work for a decent spectrum analyser. The 1500MHz HP 8568B analyser has similar specs to this.

[regenfreak]

Originally Posted by G0HZU_JMR

Yes, second and third order distortion terms tend to dominate real-world performance.

Most lab spectrum analysers will manage a mixer input IP3 of about +20dBm with 10dB input attenuation. A really good one might manage +30dBm. The noise floor might be -100dBm with a 10kHz RBW and 10dB front end attenuation.

The SFDR would then be 0.666*(20 - -100) = 80dB.

The second harmonic intercept might be +60dBm with 10dB input attenuation. The SFDR here would therefore be about 0.5*(60 - -100) = 80dB.

These were the classic benchmark numbers used at work for a decent spectrum analyser. The 1500MHz HP 8568B analyser has similar specs to this.

I see, the SFDR is 2/3 of the difference between IP3 and the noise floor based on the geometry of three intercepting straight lines for fundamental slope = 1, 2nd order harmonic slope = 2, and third order harmonic slope = 3.

I have a HP 3324A 21MHz sweep generator coming on the way. This will enable me to do two-tune measurements of a mixer. But I must build three diplexer filters for the three signal sources (Rigol DG1022Z, FY9600 100MHz and HP2224A) with a Wilkinson-type combiner.