6-gang FM stereo tuner heads

[G0HZU_JMR]

I would be interested to see the measurement of IIP3 and OIP3 of a ring mixer,e.g. ADE-1 or SBL-1 since you have the right equipment to do it. It will need +7dbm signal to drive the LO oscillator.

I've got quite a few diode ring mixers here but some caution is required when testing them. The results will depend on how well terminated the mixer ports are. Ideally a 50 ohm termination impedance is desired on all ports. If this isn't achieved then the IP3 results can be affected quite significantly.

Because of this, the results I achieve might not be replicated in a real circuit design where the ports might not have a good wideband termination on all ports.

I've got several level 7 Minicircuits mixers here including a very old SBL-1, a TUF-1 and an SRA-3. The others are level 10, level 13 and higher. The old rule of thumb for the VHF level 7 mixers was that the input 1dB compression point should be about 6dB below the LO drive level. So the input P1dB should be about +1dBm. The input IP3 should be about 15dB higher than the input 1dB compression point but this is just a crude rule of thumb.

So I'd expect to see an input IP3 of somewhere around +16dBm for these mixers. However, this will vary a bit depending on the RF, IF and LO frequencies. Also, the LO drive level will affect the result and this seems to be more critical for the higher level mixers, eg level 13.

[Radio Wrangler]

You can boost the IMD performance of a diode ring mixer if the LO is not a 50 Ohm sine but a current squarewave! The dv/dt through the switching region being the key.

David

[regenfreak]

Originally Posted by Radio Wrangler

Sig gens are usually somewhat more complicated than Marki suggests, though they're closer in terms of microwave band generators where most of their mixers are aimed. For the 100MHz band you're most likely to have an output level control loop with an RF detector on the output of the amp of the sig gen, certainly any generator with an output level controllable enough not to spoil any accuracy of these tests.

Cross-coupled signal from one generator to another will go through whatever value of output attenuator the generator has engaged and will beat with the generator's own signal in the levelling loop detector. This will AM modulate the generator's output forming just the third-order products you're trying to measure.

To get reasonably useable results coutermeasures like disabling levelling, or adding power amps and external attenuators are often needed.

I did a lot of measuring aviation COM and NAV radio designs through the last decade and they were not quite in the class of a basic packaged diode ring mixer (mini TFM-2 for example) and quite a bit of elaboration was needed to get specified TOI results. Quick tests before elaborations were introduced could easily be several dB out, and this was with low power transistor tree mixers (Called Gilbert cells by everyone except Barrie Gilbert himself and those who know the real Gilbert Cell analogue computing element)

The choice of spacing for a TOI test is also important in receiver tests. It sets how far down the structure of the receiver both tones will penetrate and how much removal of one or both happens due to the selectivity of different stages. Tone spacing sets which aread of the set get their linearity measured. For narrow spacing on an FM set, you'll eventually start measuring the linearity of the limiter stages! and they are intentionally not at all linear. Wide spacing can give too encouraging results because RF selectivity enters the scene and controls what reaches the mixer.

If you were measuring just a diode ring double balanced mixer alone, there is no selectivity at play, and so tone spacing ceases to be so important.

Just being a few dB off in TOI has a larger effecto on your SFDR calculations.

David

Thanks for enlightening us on the complex subjective. According to Marki's article, the results are still acceptable using "quick and dirty method". But it get very complicated when different sources of errors are being considered besides what you mentioned about the issue with leveling loop detector. The front end mixer of the spectrum analyzer, and the dynamic ranges of the spectrum analyzer and three signal generators are the main limiting factors. The use of RF amplifiers in front of the external attenuators would recover the dynamic range. The low pass filters can be used to remove the 2nd order harmonics.

Below is the test procedure of radio receiver. It states that the IP3 measurements depends on the procedures, the frequencies, spacing,,,the procedure should be designed in such a way that it is independent of the receiver design:

https://www.itu.int/dms_pubrec/itu-r...8-I!!PDF-E.pdf


Sometimes the levels of the two tones are different, this calculator would be handy:

https://rfresponse.com/calculators/ip3_2tone_chart.php

Anyway this is all complicated stuff, probably too hardcore for most hobbyists and any serious attempt to do it properly without the right equipment is beyond the realm of possibility.

[regenfreak]

I've got quite a few diode ring mixers here but some caution is required when testing them. The results will depend on how well terminated the mixer ports are. Ideally a 50 ohm termination impedance is desired on all ports. If this isn't achieved then the IP3 results can be affected quite significantly.

Because of this, the results I achieve might not be replicated in a real circuit design where the ports might not have a good wideband termination on all ports.

I've got several level 7 Minicircuits mixers here including a very old SBL-1, a TUF-1 and an SRA-3. The others are level 10, level 13 and higher. The old rule of thumb for the VHF level 7 mixers was that the input 1dB compression point should be about 6dB below the LO drive level. So the input P1dB should be about +1dBm. The input IP3 should be about 15dB higher than the input 1dB compression point but this is just a crude rule of thumb.

So I'd expect to see an input IP3 of somewhere around +16dBm for these mixers. However, this will vary a bit depending on the RF, IF and LO frequencies. Also, the LO drive level will affect the result and this seems to be more critical for the higher level mixers, eg level 13.

I see. Cheers.
Check out this guy's measurement of OIP3 of the AED-1, his set-up is fairly simple:
http://qrp-popcorn.blogspot.com/2014...ip3-notes.html

I followed his blog and learnt a few tricks from him..

[G0HZU_JMR]

One thing to bear in mind is that the Rigol spectrum analyser in that blog demo is being asked to cope with all of the mixer terms being produced at the mixer output port. This will include the two wanted test tones and also the two image tones and quite a few other mixer terms. This can be risky, especially if the analyser is being used close to its own distortion limits.

I see he's used the IF port as the output port even though the system is an up-conversion.

The usual aim is to try and keep the analyser IMD3 terms at least 20dB lower than the predicted IMD3 terms from the device under test. This will keep the uncertainty (caused by the analyser's own IMD) below about 1dB.

If you plan to do lots of mixer testing with an IF of 10.7MHz I'd recommend you make a narrowband diplexer at 10.7MHz. This will correctly terminate the mixer from DC up through VHF and it will also provide a bandpass response at 10.7MHz.

The diplexer consists of two inductors, two capacitors and two 50 ohm resistors so it is fairly quick and easy to make. It would protect the analyser from all the other mixer products arriving at the IF port. A typical 10.7MHz narrowband diplexer would have a response as in the plot below:

[Radio Wrangler]

Originally Posted by regenfreak

It states that the IP3 measurements depends on the procedures, the frequencies, spacing,,,the procedure should be designed in such a way that it is independent of the receiver design:

That can't often be achieved. Different tone spacings will heve the test tones passing different amounts of the way down the receiver strip, tenderising different combinations of stages. So a complete receiver can only be seen as a number of TOI figures relating to different offsets. These then need interpretation in order to see if goals are beingmet after all, and if not, which areas are implicated. This whole business has to be applicable to all sorts of receiver structures. I've had to use it on basic single-superhets as well as hex-superhets, along with variants converting signals to digital at various stages. Having a few figures of merit which can relate the performance with one structure to those of other structures allows you to make meaningful comparisons and stay sane (or reasonably sane).

Once you get fairly involved, you start using spreadsheet models with TOI noise gain and compression figures for each stage in a structure. It is usual to express impairments in terms of calculated equivalents at some convenient place, say the input. With one of these you can compare the impairment contributions of different stages to each parameter on a level playing field and see where improvements are needed. You can see progress towards overall system specs. You can break an overall system spec down into budgets for each impairment in each stage. When you find you're needing impossible or unaffordable stages, you go back and re-jig things.

I use a group of spreadsheets having figures for different offsets from the receiver centre frequency so I can not only see wanted signals grow down the structure, I can also see unwanted signals diminish.

Analysis, synthesis, and procurement all start to get blurred. Somehow you have to navigate to an acceptable solution, or you gather evience as to why it's impractical.

David

[G0HZU_JMR]

I had a quick go at making the diplexer over lunch today. The response is a bit narrow for wideband FM so I redesigned it to have a bit more bandwidth at 10.7MHz. This makes it easier to tune up and it also performs better up at VHF in terms of the port match.

See below for the revised schematic and the results. The inductors are made using T37-6 toroids. This gives good Q at 10.7MHz and this also allows some fine tweaking by squishing or spreading the turns slightly. The insertion loss at 10.7MHz was about 0.2dB.

The port VSWR is very good even above 300MHz. The stopband rejection degrades above about 300MHz partly because of the package inductance of the SMD 470pF caps but also because of various series resonance modes in the 2.7uH toroid. However, even 20dB rejection is useful here in terms of protecting the spectrum analyser from large out of band signals.

The plots below compare the very crude circuit model against the measured results using the VNA. The agreement is very good up until about 200MHz.

I think the diplexer would perform better up at UHF if built using SMD parts using a tight PCB layout. The SMD 470pF caps are in a very large package and each have just over 1.5nH package inductance so this affects the stopband performance. Using several smaller caps might help here. However, I think the overall performance of the diplexer is good enough as it is.

It should present a lower VSWR to the mixer than the RF input port of a typical budget spectrum analyser and it will also protect the analyser from the unwanted mixer terms including the large image terms.

[Radio Wrangler]

That diplexer is a jolly good thing to have after a diode ring mixer even in the final receiver. Getting a good very wideband matc at the input of an IF strip is usually overlooked. You want a good wide match to absorb ALL things coming out of the mixer. Any power on these frequencies will only cause more mischief it it gets back into the mixer. You want to terminate LO and RF leakage into the IF, and all their harmonics and intermods.

David

[G0HZU_JMR]

It's also worth exploring the datasheet for that Rigol spectrum analyser as this model doesn't have good distortion performance when compared to an old school analyser from HP. The Rigol datasheet says that the noise floor with 0dB attenuation is typically -135dBm with a 1Hz BW. So that means the noise figure is -135 - (-174) = 39dB. An old school HP analyser typically manages a noise figure of about 25dB.

The input IP3 for the Rigol is quoted as +10dBm with 10dB attenuation so the equivalent IP3 for the Rigol mixer would be 0dBm with 0dB attenuation. These aren't good numbers at all... A typical HP analyser might have a mixer IP3 of about +10dBm but some newer analysers can get close to +20dBm.

If I put these numbers from the Rigol into a basic spreadsheet it is possible to predict the levels of any internal IMD3 terms produced by the Rigol analyser. Obviously this depends a lot on the accuracy of the datasheet but it should give some idea of what to expect.

There's a screenshot of the Rigol below, taken from the blog website showing input tones at -11dBm to the analyser with 25dB internal attenuation. My excel spreadsheet is also shown below and if I boost the Rigol mixer IP3 up by just 1dB to 1dBm then the IMD3 distortion numbers do seem to agree quite closely.

The spreadsheet predicts IMD3 terms at -85dBm and a delta of 74dB. This is very close to the screenshot performance. The analyser has been set to a 30Hz RBW to try and squeeze out this level of performance. You can see the sweep time is a whopping 277 seconds so the operator was desperately trying to get the most out of the analyser here by selecting a very narrow resolution bandwidth of just 30Hz. It looks like it took over 4 minutes for the analyser to complete each sweep so these measurements required a lot of patience...

The distortion performance of a typical spectrum analyser also degrades when used below about 50MHz so the performance of the Rigol may be worse than this when used down at 10.7MHz for example. At low frequencies like this the first IF and the second mixer of the analyser will also have to cope with any leaked LO energy from the first mixer and it also has to handle the nearby image term so the analyser datasheet will often omit to publish distortion data below (say) 50MHz. This isn't because the manufacturer is being sneaky, it's just difficult to spec how much the performance degrades at lower frequencies. This has a lot to do with the mixer port isolation and the IF1 filter stopband performance and these parameters tend to vary even for the same model of analyser. All the top manufacturers tend to avoid quoting the distortion performance of a microwave spectrum analyser below several tens of MHz.

[G0HZU_JMR]

IMD3 spurious free dynamic range (SFDR) is a function of the noise floor and the IP3 so it's not that great that the Rigol analyser has a noise figure as high as 39dB coupled with a mixer IP3 of just 0dBm. The spreadsheet predicts the Rigol can manage a SFDR of just over 80dB with a 30Hz RBW. With a 1kHz RBW this drops to about a 70dB SFDR so you can see why the operator experimented with a 30Hz RBW and was willing to wait over 4 minutes per sweep...

SFDR = 0.667*(IP3 - noise floor)

An old school HP 8568B analyser would manage a SFDR of about 98dB with 30Hz RBW based on the typical mixer IP3 of 13dBm and a typical noise floor of -135dBm in a 30Hz RBW. A modern Agilent/Keysight analyser would get close to 105dB SFDR with a 30Hz RBW.

[regenfreak]

It's also worth exploring the datasheet for that Rigol spectrum analyser as this model doesn't have good distortion performance when compared to an old school analyser from HP. The Rigol datasheet says that the noise floor with 0dB attenuation is typically -135dBm with a 1Hz BW. So that means the noise figure is -135 - (-174) = 39dB. An old school HP analyser typically manages a noise figure of about 25dB.

The input IP3 for the Rigol is quoted as +10dBm with 10dB attenuation so the equivalent IP3 for the Rigol mixer would be 0dBm with 0dB attenuation. These aren't good numbers at all... A typical HP analyser might have a mixer IP3 of about +10dBm but some newer analysers can get close to +20dBm.

If I put these numbers from the Rigol into a basic spreadsheet it is possible to predict the levels of any internal IMD3 terms produced by the Rigol analyser. Obviously this depends a lot on the accuracy of the datasheet but it should give some idea of what to expect.

There's a screenshot of the Rigol below, taken from the blog website showing input tones at -11dBm to the analyser with 25dB internal attenuation. My excel spreadsheet is also shown below and if I boost the Rigol mixer IP3 up by just 1dB to 1dBm then the IMD3 distortion numbers do seem to agree quite closely.

The spreadsheet predicts IMD3 terms at -85dBm and a delta of 74dB. This is very close to the screenshot performance. The analyser has been set to a 30Hz RBW to try and squeeze out this level of performance. You can see the sweep time is a whopping 277 seconds so the operator was desperately trying to get the most out of the analyser here by selecting a very narrow resolution bandwidth of just 30Hz. It looks like it took over 4 minutes for the analyser to complete each sweep so these measurements required a lot of patience...

The distortion performance of a typical spectrum analyser also degrades when used below about 50MHz so the performance of the Rigol may be worse than this when used down at 10.7MHz for example. At low frequencies like this the first IF and the second mixer of the analyser will also have to cope with any leaked LO energy from the first mixer and it also has to handle the nearby image term so the analyser datasheet will often omit to publish distortion data below (say) 50MHz. This isn't because the manufacturer is being sneaky, it's just difficult to spec how much the performance degrades at lower frequencies. This has a lot to do with the mixer port isolation and the IF1 filter stopband performance and these parameters tend to vary even for the same model of analyser. All the top manufacturers tend to avoid quoting the distortion performance of a microwave spectrum analyser below several tens of MHz.
Attached Thumbnails

It is very observant for you to spot those details based on his screenshot. Once I considered buying the Rigol DSA815-TG but was put off by its lacklustre performance. The model been around for many years and has become dated. I owned the two Siglent SVA1015X for two weeks before i returned them as both of them developed different faults. I really like its VNA and spectrum analyzer features but I found myself used the SVA1015X's VNA feature 99% of the the time and rarely used the spectrum analyzer +TG. The VNA has poor low frequency performance below 10MHz.

I have been promoted at work and is tempted to get the Siglent SSA3032X. But screw it, I don't want to spend that amount of money...I save the money for the rainy days...

had a quick go at making the diplexer over lunch today. The response is a bit narrow for wideband FM so I redesigned it to have a bit more bandwidth at 10.7MHz. This makes it easier to tune up and it also performs better up at VHF in terms of the port match.

See below for the revised schematic and the results. The inductors are made using T37-6 toroids. This gives good Q at 10.7MHz and this also allows some fine tweaking by squishing or spreading the turns slightly. The insertion loss at 10.7MHz was about 0.2dB.

The port VSWR is very good even above 300MHz. The stopband rejection degrades above about 300MHz partly because of the package inductance of the SMD 470pF caps but also because of various series resonance modes in the 2.7uH toroid. However, even 20dB rejection is useful here in terms of protecting the spectrum analyser from large out of band signals.

Thanks. Cool, man it always a pleasure to watch the expert's demonstrations. May i ask a dumb question; how come your diplexer has only two ports? Usually a diplexer has three ports, one for low-band and one for high-band. In your case, the response looks like a bandpass filter. I am a bit confused.

That can't often be achieved. Different tone spacings will heve the test tones passing different amounts of the way down the receiver strip, tenderising different combinations of stages. So a complete receiver can only be seen as a number of TOI figures relating to different offsets. These then need interpretation in order to see if goals are beingmet after all, and if not, which areas are implicated. This whole business has to be applicable to all sorts of receiver structures. I've had to use it on basic single-superhets as well as hex-superhets, along with variants converting signals to digital at various stages. Having a few figures of merit which can relate the performance with one structure to those of other structures allows you to make meaningful comparisons and stay sane (or reasonably sane).

Once you get fairly involved, you start using spreadsheet models with TOI noise gain and compression figures for each stage in a structure. It is usual to express impairments in terms of calculated equivalents at some convenient place, say the input. With one of these you can compare the impairment contributions of different stages to each parameter on a level playing field and see where improvements are needed. You can see progress towards overall system specs. You can break an overall system spec down into budgets for each impairment in each stage. When you find you're needing impossible or unaffordable stages, you go back and re-jig things.

I use a group of spreadsheets having figures for different offsets from the receiver centre frequency so I can not only see wanted signals grow down the structure, I can also see unwanted signals diminish.

Analysis, synthesis, and procurement all start to get blurred. Somehow you have to navigate to an acceptable solution, or you gather evience as to why it's impractical.

If we see the receiver as a black box, there should be the overall figure of merit "OIPn" for the whole receiver for the cascade stages of OIP3 (see attached figure). It must be hard to measure OIP3 because the I/O of most mixer or RF amp stages in most superhet receivers do not have native 50/75 ohms impedance. How are you supposed to measure OIP3 of high impedance or non-50ohm stages in real receivers? The impedance matching problems must be a heartache.

[regenfreak]

OK, in your circuit, it has a series LC and parallel LC plus two 50 ohms. At the resonant frequency, the series LC has very low impedance, only a narrow band can pass through. The parallel LC has very high impedance so the two 50 ohms resistors have no effect on the circuit.

Outside the bandpass freqencies, the opposite happens. The parallel LC appears as a low impedance path and they see the two 50 ohms resistors to the ground.

[G0HZU_JMR]

Thanks. Cool, man it always a pleasure to watch the expert's demonstrations. May i ask a dumb question; how come your diplexer has only two ports? Usually a diplexer has three ports, one for low-band and one for high-band. In your case, the response looks like a bandpass filter. I am a bit confused.

OK, in your circuit, it has a series LC and parallel LC plus two 50 ohms. At the resonant frequency, the series LC has very low impedance, only a narrow band can pass through. The parallel LC has very high impedance so the two 50 ohms resistors have no effect on the circuit.

Outside the bandpass freqencies, the opposite happens. The parallel LC appears as a low impedance path and they see the two 50 ohms resistors to the ground.

Yes, that's how it works. That network is commonly described as a diplexer although I can see why it may seem unconventional to call it a diplexer.

The energy from the mixer gets shared/split between the 50R load resistors and the spectrum analyser. At large frequency offsets (far away from 10.7MHz) pretty much all the RF energy from the mixer ends up as heat in the first 50R resistor. So in this sense this resistor is the other port that soaks up energy at the image frequency and at the leaked LO frequency. There probably is a better (correct?) name for this network but it has always been called a diplexer at work and it is often described as a diplexer in amateur radio literature.

[regenfreak]

Yes, that's how it works. That network is commonly described as a diplexer although I can see why it may seem unconventional to call it a diplexer.

The energy from the mixer gets shared/split between the 50R load resistors and the spectrum analyser. At large frequency offsets (far away from 10.7MHz) pretty much all the RF energy from the mixer ends up as heat in the first 50R resistor. So in this sense this resistor is the other port that soaks up energy at the image frequency and at the leaked LO frequency. There probably is a better (correct?) name for this network but it has always been called a diplexer at work and it is often described as a diplexer in amateur radio literature.

I see, cheers. Your C1 and L1 have the resonance reactance of 181.34 ohms and C2 and C2 reactance of 15.824 ohms. It works like a bridge tee attenuator that maintains 50 ohms for source and load input impedance to attain the design attenuator.

[regenfreak]

I have read online some news group that someone suggested the use of 4:1 transformer in series with a 130 resistor for both ends of the ceramic filter tester. He claimed it would have a more broad band matching. I have used the Mini-Circuits TC4-1T+ to try out the idea and it turns to be inferior and poorer match in the Smith chart. It is better to stick with LC matching pad.

The TC4-1T is incredibly small, I wonder how it is possible to have the inductance of 400 microhenries for the secondary in such micro package.

[regenfreak]

I suddenly remember watching the demonstration of the use of a slide screw tuner to correct the impedance mismatch between the spectrum analyzer and an amplifier, bring the point to the centre of a Smith chart(at 35 minutes):

https://youtu.be/ToVJTKCyIU8

These are very neat looking devices that can work as low as 600MHz:

https://focus-microwaves.com/wp-cont...rew-Tuners.pdf


https://manualzz.com/doc/21325695/sl...rs-description

[G0HZU_JMR]

Thanks. The return loss and smith chart plots looking into that Minicircuits amp look to be very dodgy to me.

At 34:40 onwards the ripple in the return loss and the squiggles in the smith chart response look to be caused by a system calibration issue. The TSP presenter thinks this is to do with the way the amplifier is matched but I doubt it. The regular pattern suggests an incorrectly calibrated system at the end of a cable that might be 35cm long. There does appear to be a cable of that length in the video.

Either that or the Siglent VNA has really poor directivity after calibration. The datasheet suggests >40dB directivity so I'd expect to see much better results than what is shown in the video.

A general rule of thumb states that in order to get < +/-1dB uncertainty ripple in the return loss measurement the directivity of the VNA has to be >20dB greater than the return loss being measured.

So if the VNA works as per the datasheet (if calibrated correctly) then there should be less than +/- 1dB ripple on the return loss plots where the return loss is less than 20dB. But the plot shows up to +/- 2.5dB ripple in the circled region below where the RL is about 16dB. This suggests the effective directivity of the VNA is only about 26dB. I strongly suspect that the VNA hasn't been calibrated properly. The regular peaks and troughs do suggest there is a ~35cm long RF cable at the input and this is a strong clue that there is a significant problem with the setup.

There's no way the Minicircuits amplifier will have regular ripple in the return loss like that so either the VNA directivity is really poor or the operator doesn't know how to calibrate the VNA properly. I suspect there is a calibration issue. Maybe the calibration was carried out correctly but it isn't set to be active. Maybe the calibration is disabled in the menu system somewhere.

[G0HZU_JMR]

The smith chart plot shown later does suggest that there has at least been a partial success with the calibration because there would be lots of rotations around the smith chart caused by the cable length if the calibration was disabled. So the calibration must be active.

There's no way the Minicircuits amplifier will have significant return loss ripple like that with peaks every 300MHz. Therefore, I can't explain why it looks like that unless the VNA is poor/faulty or the calibration hasn't been done correctly.

[G0HZU_JMR]

I had a rummage today and managed to find a few 10.7MHz ceramic filters and measured a few on my VNA. The plot below is of the widest one I could find.

I took a full s2p model of the filter and then matched it using virtual components in a simulator. This method works extremely well when reproduced with real matching hardware. I think I can optimise the matching a bit more than this but see below for the first attempt to match the filter across about a 200kHz bandwidth. This filter isn't as wide as yours but it isn't far off. I can try and find a wider filter at work as there may still be some in a box under my desk.

[G0HZU_JMR]

I let the computer optimiser play with various matching networks and left it to optimise for an hour last night. The result below is quite good but this uses a two stage match on the input and output so it is a bit unrealistic in terms of component count. Also this would be hard to replicate in reality unless tuneable components were used in the matching as the optimiser was set to allow 1% tuning increments on each component in the matching network.