6-gang FM stereo tuner heads

[G0HZU_JMR]

Here's some numbers to play with based on the Siglent datasheet. The spreadsheet below attempts to predict the SFDR at the settings shown in my previous post.

Based on the datasheet, this is what I'd expect to see from the Siglent analyser based on the datasheet info. You can see that I've deliberately set the input tone levels to -16.67dBm and this generates IMD tones (the green text) at the same level as the -90dBm DANL in a 10kHz RBW. The SFDR works out to be about 73dB.

[regenfreak]

Originally Posted by G0HZU_JMR

Another way to look at this is to forget about the TOI point and just inject two clean distortion free tones into the analyser. Set the tone level really low to start with. Choose a convenient RBW, say 10kHz and set the front-end attenuator to 10dB and the reference level to -5dBm. On your Siglent analyser, I think that the noise floor (DANL) would be at about -90dBm with these settings if you have the preamp turned off.

In theory, you should be able to keep increasing the input tone levels up to a point where the IMD tones would have the same power as the displayed -90dBm DANL. In practice, it will be difficult to judge where this happens, you could briefly select a narrower RBW to make it easier to see the -90dBm IMD tone level as it begins to compete with the DANL and become visible.

When you reach this point, the difference between the input test tone levels and the -90dBm DANL is the spurious free dynamic range with a 10kHz RBW.
You would obviously have to set the analyser to a span where the phase noise of the analyser doesn't mask the DANL. Maybe try a 5MHz span with 1MHz tone spacing to try and minimise the impact of the LO phase noise. Set the RBW to 10kHz. I'm guessing that your analyser phase noise will be -120dBc/Hz at 1MHz offset. 10kHz RBW is 40dBHz so you should be able to see about an 80dB dynamic range if there was no IMD. However, I think the Siglent will probably have a SFDR of about 73dB with 10kHz RBW because of the IMD it will generate.

Alternatively, if you know the input TOI of the analyser and the DANL from the datasheet, you can predict the SFDR with a reasonable amount of accuracy without having to do any hardware tests. Just use the equations you already have. The equations allow you to work backwards from the published TOI and noise floor data to work out the SFDR at a given RBW setting.

This generally works well for classic spectrum analysers that use a diode pair or diode ring mixer at the front end. I don't know if the Rigol and Siglent analysers will follow the same classic equations. There are solid state devices in the front end of these analysers and the linearity of these devices probably won't be as predictable. Therefore, the tests outlined above might not give the expected results.

Thanks for that. Your numbers are close to what i have got:

At 10kHz , 10Mhz, the front-end attenuator set to 10dB the noise floor. the DANL of the Siglent is measured to be -94dBm with 100 averaging samples.

The phase noise is <112dbc at 1Mhz offset from Sigent's test data.


The Siglent SSA3021X and SSA3032X use HM488 GaAs Schottky diode ring 1st mixer:

https://www.analog.com/media/en/tech...ets/hmc488.pdf


IP3 = +15db, NF = 7, 1db compression = +8dbm.

There is an excellent video on the walkthrough teardown of SAA3021X from EEVblog:

https://youtu.be/fvTfBwRzpdo


I will try to do the above test tomorrow assuming that the two-tone test is done without DUT and LP diplexers.

One of the key things that no one explain is that what is the effect of tone spacing f2-f1? Does it matter it is 100Khz or 1MHz? With 100KHz, I can use much smaller RBW and it is much easier to see the IP3 products as they are often near the noise floor, particularly with DUP with high TOI.


In the attachment, there is a clear, 1:1 linear relationship between DANL and internal attenuation for a particular unknown spectrum analyzer given by Anritsu. I am guessing that the attenuation increment is positive and it is equal to -1x mixer level decrement, hence there is a -1 slope for DANL vs mixer level straight line. Probably I have answered my early question. The y-intercepts for IMD2, IMD3 and DANL straight lines are dependent on the particular model of the analyzer or some kind of frudge factors ...

[G0HZU_JMR]

Here's a demo using my HP8566B. I've warmed it up and performed the internal automated cal routine on it.

See below for a spreadsheet showing the settings and the predicted performance and also a screenshot.

The RBW is 10kHz, the attenuation is 10dB. The TOI appears to be about 11dBm. The noise floor appears to be about -100dBm in a 10kHz BW as expected. I've used the sample detector. This analyser has 1000 data bins across the span so it's OK to use the sample detector on this span and RBW setting.

You can see that the spreadsheet prediction and the analyser plot agree very well in terms of the IMD levels and the noise floor.

[G0HZU_JMR]

If I then reduce the test tone levels by 5dB the IMD terms should fall about 15dB. In the plot below you can see they have dropped by about this amount, and they are beginning to get lost in the noise floor.

The analyser is very close to displaying its full SFDR. The spreadsheet predicts the IMD will be 78dB lower than the two test tones. However, the IMD term will be riding on the noise and would be bouncing up and down a lot because of this. This will make it appear slightly higher than it really is. it is close enough for a demo though!

I think if I reduced the drive level one more dB it would be very close to demonstrating the true SFDR of 80.7dB with a 10kHz RBW.

Sadly, the noise marker has shifted to the centre of the screen where it is affected by the phase noise of the analyser. It was parked over to the left and it was showing a noise floor of about -139.5dBm/Hz. With a 10kHz RBW this would jump up 40dB to about -100dBm as predicted on the spreadsheet.

[regenfreak]

Originally Posted by G0HZU_JMR

Here's a demo using my HP8566B. I've warmed it up and performed the internal automated cal routine on it.

See below for a spreadsheet showing the settings and the predicted performance and also a screenshot.

The RBW is 10kHz, the attenuation is 10dB. The TOI appears to be about 11dBm. The noise floor appears to be about -100dBm in a 10kHz BW as expected. I've used the sample detector. This analyser has 1000 data bins across the span so it's OK to use the sample detector on this span and RBW setting.

You can see that the spreadsheet prediction and the analyser plot agree very well in terms of the IMD levels and the noise floor.

Thanks. Can you show me the step-by-step breakdown how you get the answers in your attachments? I am confused about the calculations...I only have had the spectrum analyzer for 2 months and has not got a degree or PhD in electronics engineering

[regenfreak]

Finally, I have found the derivations of the equations on page 28 and 30:

https://dl.cdn-anritsu.com/en-en/tes...zer-ee1400.pdf

Now I know they come from!

[Radio Wrangler]

Those give you the derivation of the frequencies of the products. They are simply mixing products between the various fundamental tones applied along their harmonics. The harmonic series are essentially infinite, but their levels tend to decrease with higher order, so at some point you stop bothering with them. There is nothing fancy here, just two sets of harmonic series and all the sum and difference frequencies between every coice of two of them. Often the third order (2.Fa-Fb) and fifth order (3.Fa-2.Fb) are the interesting ones as lying close to the applied pair of fundamentals. You get the pait of intermod products in this area by the ambiguity in which you call a and which b.

As far as the levels go, they just use the relationship to the applied tones = the order number of the product without deriving it or referenceing somewhere with a derivation. This is much more involved to derive, and most people just wimp out and say "It just is" As a student, that looked dangerously like an assumption or something empirical, but the ratios came out to perfect integers at low levels and the mechanism was therefore an intriguing mystery. I once went through the derivation, but have never bothered since and don't have it in my head. It starts with assuming that levels are low enough that a simple power series makes a good model.

We've come an awful long way from FM tuners of any number of RF tuned circuits. Enough time and effort have gone by that you could have hand-filed one out of solid Perhaps a significant fraction of time has gone by towards the powers that by closing down the FM broacast system in the UK. There have been rumblings in that direction, but they seem to have eased off recently in their efforts to shove DAB down our throats.

I don't think your goal is making a superb FM tuner any longer. There are plenty of quite decent models you can find at now-affordable prices and restore to full original condition. I bought one of the top-end Sony 'ES' models for twenty quid, and a Revox B261 (ex-BBC version) for £300. These killed off any feelings I had for making another home-brewed tuner. They were both good enough, and had some intriguing circuitry to keep the theoretician and designer in my head happy.

So I'm deducing that you've shifted into a curiosity mode regarding RF design techniques and some of the more intriguing limitations of current circuit techniques?

David

[regenfreak]

Originally Posted by Radio Wrangler

Those give you the derivation of the frequencies of the products. They are simply mixing products between the various fundamental tones applied along their harmonics. The harmonic series are essentially infinite, but their levels tend to decrease with higher order, so at some point you stop bothering with them. There is nothing fancy here, just two sets of harmonic series and all the sum and difference frequencies between every coice of two of them. Often the third order (2.Fa-Fb) and fifth order (3.Fa-2.Fb) are the interesting ones as lying close to the applied pair of fundamentals. You get the pait of intermod products in this area by the ambiguity in which you call a and which b.

As far as the levels go, they just use the relationship to the applied tones = the order number of the product without deriving it or referenceing somewhere with a derivation. This is much more involved to derive, and most people just wimp out and say "It just is" As a student, that looked dangerously like an assumption or something empirical, but the ratios came out to perfect integers at low levels and the mechanism was therefore an intriguing mystery. I once went through the derivation, but have never bothered since and don't have it in my head. It starts with assuming that levels are low enough that a simple power series makes a good model.

We've come an awful long way from FM tuners of any number of RF tuned circuits. Enough time and effort have gone by that you could have hand-filed one out of solid Perhaps a significant fraction of time has gone by towards the powers that by closing down the FM broacast system in the UK. There have been rumblings in that direction, but they seem to have eased off recently in their efforts to shove DAB down our throats.

I don't think your goal is making a superb FM tuner any longer. There are plenty of quite decent models you can find at now-affordable prices and restore to full original condition. I bought one of the top-end Sony 'ES' models for twenty quid, and a Revox B261 (ex-BBC version) for £300. These killed off any feelings I had for making another home-brewed tuner. They were both good enough, and had some intriguing circuitry to keep the theoretician and designer in my head happy.

So I'm deducing that you've shifted into a curiosity mode regarding RF design techniques and some of the more intriguing limitations of current circuit techniques?

David

Thanks. I have figure out how to interpret and use the distortion vs mixer graphs in calculation. Thanks to Anirtsu's example calculations. HP, R&S and all web sites never bother show you one single example.

In the attached screenshot is the TOI taken from Anirtsu. The trick is to transpose y axis of the graph in attachment 2 to be centred at X=0. So X= 0 is relative to mixer input reference level, bingo!
I saw them dropping this "relative to mixer reference level" like cluster bombs articles and never bother to explain the meaning of iy!

The intercept and linear equation for Pmd3 is derived from first principles here (equation 4.116 ):

https://www.sciencedirect.com/topics...lation-product


Yes, curiosity drives most things. I try to avoid collecting and restoring vintage radios (mostly American MW/FM valve radios) because I run out of space to store them.

I no longer like to own something, polish and look at it. I am not a collector but my flat is like an electronic junkyard. I got very bored with swapping capacitors and resistors in restoration.

I found greater satisfaction from building high-performance 6-gang FM tuners from scratch that sounds like CD quality, as good as top-end Kenwood and Pioneer models etc. The homebrew tuners look ugly but the experience gain from building them is valuable.

I only own a £10 Sansui FM tuner but I have the RF/IF modules/circuit boards of top-end FM tuners such as Revox B286 and B261, Technics ST 9030 and JR JR-S600. I got them with the intention of studying their designs, getting them working and sweeping the RF and IF frequency responses. It is a kind of long-life learning journey and masochistic tendency to set myself intellectual challenges and problem-solving tasks.

The spectrum analyzer is an incredibly complex instrument. I have found myself running into rabbit holes when i learn more about it. The attachment 3 and 4 illustrate the complex inter-relationship between mixer limit, dynamic range limit, reference level at the input, mixer and log amp envelope detector from R&S. It is confusing because the 1st mixer is never attached to the log amp envelope detector directly!

[Radio Wrangler]

Originally Posted by regenfreak

The spectrum analyzer is an incredibly complex instrument. I have found myself running into rabbit holes when i learn more about it. The attachment 3 and 4 illustrate the complex inter-relationship between mixer limit, dynamic range limit, reference level at the input, mixer and log amp envelope detector from R&S.

I know, I spent my last several years at HP as senior design engineer in the spectrum analyser division. Designing one is an interesting juggling act making sure you don't spoil any of a very wide variety of uses. Chunks of them also got used in things like the Noise Figure Analyser as well as custom receivers for spooks and military.

David

[G0HZU_JMR]

Of course, all this stuff carries across to the front-end design of your FM tuner. You can apply the same equations for SFDR etc.

The spectrum analyzer is an incredibly complex instrument. I have found myself running into rabbit holes when i learn more about it. The attachment 3 and 4 illustrate the complex inter-relationship between mixer limit, dynamic range limit, reference level at the input, mixer and log amp envelope detector from R&S. It is confusing because the 1st mixer is never attached to the log amp envelope detector directly!

With the older analysers like the HP8566B and 8568B, the log amp only had an effective range of just over 90dB. So, this complicates things when measuring noise near the bottom of the display. In theory, you are supposed to keep the noise at least 10-14dB above the bottom of the log display if you want to measure noise power fairly accurately. This is because the old school logamp can't handle noise properly near the bottom of its log range. This means the operator has to be very careful when measuring phase noise on these old analysers. It's easy to get a false result.

Newer analysers with a digital IF perform much better here. However, the digital IF is going to be prone to clipping problems in the ADC with signals near the top of the display. This is due to the overall PEP of the waveform. Complex waveforms can exceed the ADC input range. Older swept analysers with analogue RBW filters are much more tolerant of this.

[regenfreak]

Originally Posted by G0HZU_JMR

Here's some numbers to play with based on the Siglent datasheet. The spreadsheet below attempts to predict the SFDR at the settings shown in my previous post.

Based on the datasheet, this is what I'd expect to see from the Siglent analyser based on the datasheet info. You can see that I've deliberately set the input tone levels to -16.67dBm and this generates IMD tones (the green text) at the same level as the -90dBm DANL in a 10kHz RBW. The SFDR works out to be about 73dB.

Thanks I have only done the test you suggested last night:

fundamental 10Mz, 11MHz, BW 10khz, 10db attenuation, pre-amp off, input -18dm,
second test input 5dbm down. The noise floor is about -130dbc/Hz at 4MHz span, 10KHz RBW and 8-12Mhz, 100 samples. I cant find the noise figure for the Siglent. But its 1st mixer HMC499 has IP3 = +15dbm.

I got quite a few strong spurs from the sign gen which are stronger than IMD3 which may invalidate the tests. They disappear completely if I increase the attenuation from 10db to 30db, meaning they are generated inside the spectrum analyzer mixer. I dont have the chance to make the LP diplexers that I was supposed to make last night. But the spurs are below the fundamentals! I am still trying to make sense of so many complicated things I have stumbled across.

I dont have all the equations used in your excel spreadsheets so I wiil dig around online. The muddy water is getting clearer, I start to see fainted stars amid the nebulous clouds.

[regenfreak]

The troublesome spurs come from the BY6900, so i have replaced it by my cheapo 10MHz OCXO and the spurs are completely gone. So the SFDR is around 70db vs your guess of 73db, very close!

[G0HZU_JMR]

Probably a good place to start would be with your signal source(s). It might be better to make something cleaner for f1 and f2.

Maybe make a couple of 5th overtone crystal oscillators to get you up to 100MHz. Then filter and buffer each with a high reverse isolation amplifier and then combine in a classic 6dB combiner.

Otherwise, it's hard to comment on your analyser plots. Hopefully, the numerous spurs are due to your current signal source and combiner setup and not from something inside the analyser.

Whoops, I just saw that we double posted. You already tried a cleaner source.

[regenfreak]

Originally Posted by G0HZU_JMR

Probably a good place to start would be with your signal source(s). It might be better to make something cleaner for f1 and f2.

Maybe make a couple of 5th overtone crystal oscillators to get you up to 100MHz. Then filter and buffer each with a high reverse isolation amplifier and then combine in a classic 6dB combiner.

Otherwise, it's hard to comment on your analyser plots. Hopefully, the numerous spurs are due to your current signal source and combiner setup and not from something inside the analyser.

Whoops, I just saw that we double posted. You already tried a cleaner source.

Yes my cheapo -150dbc/Hz OCXO has saved my day. I let it oven baked for 10 minutes using a rechargeables battery for model cars. I will need to put my signal sources and combiner under greater scrutiny. I measured the power combiner with a VNA a long time ago, it has at least 40db isolation. With two 20db attenuators at the sources, they have 80db isolation in theory.

I will try the Mini-circuit combiner you gave me.

I have messed around with the online calculator so I dont have to use my brain. I have to fudge the noise figure for the Silgent to be 35 (pre-amp off) in order to give SFDR about 77db. With the 10db attenuator, the noise figure will degrade to 45db.i dont really trust the measurements as the distortion products are so close to the noise floor as indicated by the mixer level sensitivity study

https://www.changpuak.ch/electronics/calc_19.php

[G0HZU_JMR]

I will try the Mini-circuit combiner you gave me.

The TSC-2-1 combiner has fairly good performance at 100MHz. The isolation is just over 30dB. It's at its best across the HF bands.

However, I think you can get better isolation performance at 100MHz if you make a 6dB combiner. Try both and see if it makes any difference.

The TFM-2 mixers should perform well for level 7 mixers at 100MHz because these little mixers are good to 1GHz. The port isolation is about 54dB at 100MHz. You should see an output TOI of over +10dBm at 100MHz. With a conversion loss of just 5.5dB the input TOI would therefore be about +15.5dBm or so.

I used the level 10 TFM-2LH version a lot back in the late 1990s and it worked very well. The level 10 ADE-1LH was also a very good performer across 30MHz to 200MHz. The input TOI can be expected to be about 3dB higher with a level 10 mixer.

[Radio Wrangler]

The more cross-isolation a combiner/splitter offers, the touchier it is to having good impedances presented to it. Eventually you hit a limit and get bulldozed into small power amps and attenuators.

David

[regenfreak]

Originally Posted by G0HZU_JMR

The TSC-2-1 combiner has fairly good performance at 100MHz. The isolation is just over 30dB. It's at its best across the HF bands.

However, I think you can get better isolation performance at 100MHz if you make a 6dB combiner. Try both and see if it makes any difference.

The TFM-2 mixers should perform well for level 7 mixers at 100MHz because these little mixers are good to 1GHz. The port isolation is about 54dB at 100MHz. You should see an output TOI of over +10dBm at 100MHz. With a conversion loss of just 5.5dB the input TOI would therefore be about +15.5dBm or so.

I used the level 10 TFM-2LH version a lot back in the late 1990s and it worked very well. The level 10 ADE-1LH was also a very good performer across 30MHz to 200MHz. The input TOI can be expected to be about 3dB higher with a level 10 mixer.

My only signal source can only go up to 100MHz is FY6800. It has spurs in the upper and lower sidebands at very low levels. Normal Butterworth and Cheryshev filters won't work as they reflect all the band-stop frequencies back to the frequency sources creating more distortion products. So I have to make "absorptive" diplexers to bypass the band-stop frequencies to the 50-ohm loads.

I have to look for schematics for crystal oscillators with good frequency stability, low phase noise,low load pushing and pulling.

The Siglent has 751 frequency points per sweep. For RBW = 10Khz at 10MHz, if the noise floor is measured to be 130dbc/Hz and TOI = +15db for the 1st mixer based on HMC499 datasheet:

Maximum dynamic range = 2/3 (TOI-DANL)=2/3[15-(130)]=96.7db

Optimal mixer lever=1/3[2xTOI+DANL] =1/3(2x15-130)=-33dbm.
The mixer level is spectrum input minus the spectrum analyzer's internal attenuation. So the optimal spectrum analyzer input should be around -23dbm with 10db attenuation.

The distortion relative to the mixer level is the distortion power relative to the mixer power input in dBc. The TOI has a slope of 2 in the distortion relative to mixer level graph because 1db increment in the mixer level corresponds to 3db TOI distortion, the increment is 2db. The graph for the DANL relative to mixer level has a slope of -1 ( it is equivalent to N/S). For every 1db increases in mixer input power, the S/N decreases by 1db.

I guess the 1Mhz frequency spacing for the two-tone is industrial standard, regardless it is 1GHz or 10MHz fundamental. However, the phase noise around two tones' double sidebands at a particular frequency offset starts to mask the distortion product as the RBW increases. The phase noise is a horizontal line in the dynamic range chart and it can be added to the broad noise. It is advantageous to use smaller frequency spacing to improve the DANL. So this 1MHz two-tone spacing puts a significant constraint on measuring DUT of very high TOI.

Radio Wrangler The more cross-isolation a combiner/splitter offers, the touchier it is to having good impedances presented to it. Eventually you hit a limit and get bulldozed into small power amps and attenuators.

I suppose the level of isolation between ports of a Wilkinson resistive or hybrid splitter can be upset by any mismatch of one of the ports. It is a fine balancing act.
Perfect isolation only happens when all the ports are perfectly terminated with 50 ohms.

[G0HZU_JMR]

In the case of the Siglent analyser, I don't think you can use the +15dBm input TOI (of the first mixer) for your calculations unless you use really wide tone spacings, eg 20MHz spacing.

This is because the Siglent appears to have been designed with a different gain distribution to a classic spectrum analyser.

The old HP 8566 and 8568 analysers don't have any amplification until after the second mixer. This means the distortion performance on all but the narrowest spans will be dictated by the first mixer. Your Siglent analyser has a lot of amplification after the first mixer and so the input TOI of the second mixer will probably limit the overall input TOI. I think this means that the effective input TOI will be lower than +15dBm with 0dBm input attenuation. The datasheet suggests that it will be about +10dBm.

The noise floor with a 10kHz RBW at 10dB attenuation is going to be about -90dBm and this is equivalent to -130dBm/Hz.

Your SFDR calculation should be done at the chosen RBW setting.

I tried watching the latest eevblog teardown in your link and put together a spreadsheet for the Siglent based on the parts used in the signal path. It did show that the input TOI is affected by the second mixer.

I can remember watching the initial teardown of this analyser several years ago and I recall trying to work out the frequency plan for that analyser.

[G0HZU_JMR]

It is advantageous to use smaller frequency spacing to improve the DANL. So this 1MHz two-tone spacing puts a significant constraint on measuring DUT of very high TOI.

Normally, when measuring the TOI of an external DUT (rather than the TOI of the analyser itself), it's best to do it at a fairly wide tone spacing if allowed and if possible. The HP 8566 and 8568 only have an average input TOI, so one way to get around this is to space the tones a long way apart. eg 1MHz apart. This avoids having high phase noise interference at the 2f2-f1 and 2f1-f2 IMD tone frequencies.

Then set the analyser to a very narrow span of maybe 5kHz so a narrow RBW can be used and then visit 2f1-f2, f1, f2 and 2f2-f1 in turn to measure the amplitude and then work out the TOI.

This method has the advantage of avoiding the worst of the analyser's phase noise and it also allows a very narrow RBW. This means the analyser can be run with a higher input attenuation setting and this helps with the overall measurement uncertainty of the DUT as it improves the input TOI of the HP 8566/8 analyser.

[G0HZU_JMR]

Tp preserve isolation, I always try and fit a 10dB attenuator at the sum port of the combiner. In the case of the 6dB hybrid combiner this should allow just over 30dB isolation even if the DUT has a very high VSWR. If the DUT has a 2:1 input VSWR then the isolation should be just over 40dB.

The TSC-2-1 won't perform as well here, even with the 10dB attenuator at the sum port. The isolation will be about 10dB worse. However, in the past, I've fitted an RC network followed by a 10dB attenuator at the sum port of the TSC-2-1 and optimised the RC values for max isolation for a chosen band. I've done this at the popular satellite IF frequencies of 70MHz and 140MHz in the past. With the RC network suitably tweaked, it's possible to get a very high isolation over a very narrow bandwidth, maybe across a few MHz or so.

I've got loads of TSC-2-1 combiners here in a tube and I'll see if I can do this again at 100MHz. It might not be until the weekend though.