TF 2002B plus TF 2170 for intermodal tests

[Draker4c]

I have a Marconi TF2002B with a TF 2170 syncroniser which locks down to 10 Hz. Does know the feasibility of using this in conjunction with a second 2002B/ 2170B for
RF intermodulation tests on Racal RA17 / RA1772 type receivers. I don’t seem
to find any published specs regarding noise floor, stability and harmonic distortion
to see if this would be possible... or worth trying. I am considered purchase of a the second combo if it is.
Thanks! Hugh G1AUR

[loulou31]

Why not?
The problem to measure intermod is to have clear two tone clear frequency generation, without intermod in the source. Be carefull when you combine the outputs of two generator : you have to isolate one to the other using attenuator the output of the generator and the combiner. You can also use isolator but at these frequency not some much easy to find. The problem is more when you combine high level signal together because you return signal on the output amplifier from the other, and could create intermod.

Jean-Louis

[Radio Wrangler]

Jean Louis is exactly right.

There are two intermod mechanisms when you combinr two signal generators. Both get worse at higher output levels where the sig gens switch out most or all of the output attenuator sections.

The first is where one sig gen drives the output amplifier of the other backwards and intermod happens. Needless to say this is a two-way action.

The other is less obvious. Lab grade sig gens have level control of their outputs. Before the big step attenuator, there is a diode detector monitoring the output of the output amplifier. This detector is used to regulate the output level as a feedback controller. This levelling loop is used to implement fine control of the output level, and is often used to apply low-distortion AM when selected.

Any reverse RF from the other sig gen will give a beat with the generator at the levelling detector, which will detect the beat envelope, and the levelling loop will modulate the output of the sig gen in an attempt to cancel the beat. Oooops! this makes AM modulation sidebands right on the intermod product frequencies. Exactly where you don't want them.

Cures:

It may help to use a hybrid combiner which nulls the feedthrough of one genreator back to the other. You won't get the specified isolation though, unless the load impedance is exactly right. It helps, but it may not be a complete cure if you're hoping to test something good.

The other method is to follow each sig gen with a fairly powerful amplifier giving say +20 to +30dBm and then an attenuator down to the wanted level. Then you can use a resistive, broad-band combiner if you wish. The very large amounts of attenuation help with the RF cross-coupling and you get the better intermod performance of a big amplifier run gently. You lose the accuracy of the sig gen levelling loop, so you need a spectrum analyser to check the actual levels reaching the device under test.

Intermodulation is a classic RF performance test, and one that old hands have to teach relative newcomers how to do. You really have to get everything right to get a good result and there are some subtle effects at play.

David

[loulou31]

Usually you work with low or moderate level to test intermod on receiver and you could solve easily the problem. You have also to use combiner with good isolation. Avoid resistive combiner that does not provide good isolation.
However is it always possible to find such old Marconi generators? I am located in France and I suppose you can find it more easily this kind of equipment in England.
Regards

Jean-Louis

[Radio Wrangler]

Hybrid combiners offer useful isolation, but they do use ferrite cores and they create some IMD. I've come across ones which limit the level of IMD you can measure down to. In the most difficult cases you get forced into the amplifiers plus attenuators and resistive combiner route.

If you're lucky, a transformer-based combiner can give you enough isolation, but it all depends on how far you want to go.

David

[Draker4c]

Jean-Louis and David many thanks for your observations.I was planning to use the above with Marconi 2370 or 2381 SA via an appropriate combiner so the signal level would be fairly low. As per the noise and harmonic performance would generators of this vintage pass muster in performance measurement of receivers such as RA1772 and RA17?

[Radio Wrangler]

Possibly. The RA17, I think so. The RA1772 has the Rafuse mixer which is a good deal tougher. The limiting factor becomes the phase noise sidebands on the signal generator output. You can't say how low the intermod product is if you lose it in the noise. There is an insidious little effect when noise mixes with noise and many times your receiver bandwidth winds up IN your receiver bandwidth.

A pair of HP 8640 ought to be good enough. 2002s I'm not sure about, but I have done such tests with the HP606 valved sig gen and had good results. Very few synthesised sig gens are up to the RA1772 standard for intermod tests. I bought a pair of HP 8662s at work for intermod testing with mini circuits amps and combiner.

Yes, the RA1772 synthesiser isn't particularly low noise, but you don't see as far down if you also have two sig gens making similar contributions.

I was once involved in the development of a very high dynamic range receiver. The tale is told in the HP journal April 1982
https://www.hpl.hp.com/hpjournal/pdf...Fs/1982-04.pdf


David

[Draker4c]

Thanks very much David I should have defined my terms better when referring to “ noise floor” earlier the phase noise of course being the limiting factor ....and phase noise spec does not feature in any specs I have seen on the TF2002B. Do such figures exist?

[loulou31]

If you read the specification of the TF2002A you can find "residual FM in CW" .It should be lower using the frequency synchronizer.
However if you measure intermodulation with frequencies separated of some hundred of KHz or more the phase noise is not a limitation.

Jean-Louis

[Draker4c]

Thanks Jean!

[Radio Wrangler]

The TF2002 is a traditional LC oscillator and its noise behaviour is reasonable, and as you move further out, gets better.

The HP 8640 is a UHF cavity oscillator with a chain of dividers. For a free-running oscillator its noise sidebands are exceptionally good.

Most modern synthesised signal generators are pretty poor with added noise form the synthesiser. Some of the later, more economical ones use a single loop and a 'fractional-N' technique these are very noisy with an exaggerated noise sideband stretching 100kHz or more either side of the carrier. It's true that for an intermod test you can get lower noise intrusion by separating the tones further, but with this class of generator, the separation you need may be more than you want to do.

So the 2002 ought to be reasonable.

David

[G0HZU_JMR]

To get some idea of the phase noise performance of the Racal 1772 you could make a few assumptions based on the reciprocal mixing spec that is given in the manual.

This implies that there is a 70dB + 20dB = 90dB difference in a 3kHz bandwidth at a 20kHz offset.

3kHz = 3000Hz = 10*log(3000) = 35dBHz.

Therefore, the phase noise of the Racal 1772 might be in the ballpark of -(90+35) = -125dBc/Hz at a 20kHz offset.

The other way to look at it would be to try and predict the phase noise performance of a fairly modest LC oscillator that runs at the same frequency as the first LO in the Racal receiver. Up at the top of the frequency range the first LO in the Racal will presumably be running at about 65MHz because the first IF appears to be at about 35.4MHz. If I throw a few numbers into Leeson's equation for an LC oscillator at 65MHz with maybe 2mW power from the resonator arriving at the active device and a fairly modest loaded Q of 25 I get the phase noise curve in the image below. This also assumes a flicker corner frequency of 5kHz for the JFET device in the oscillator and you can see a knee in the phase noise curce at 5kHz because of this. Below 5kHz the phase noise rises at approximately 30dB/decade and above about 10kHz it will fall at about 20dB/decade.

Note that this oscillator analysis currently has no noise contribution from the varactor diodes and the associated steering resistance. Therefore, it would be reasonable to assume the phase noise would degrade a bit with a real world VCO. This is the contribution from the yellow curve in the graph below and you can see I have deliberately forced this contribution to be negligible.

Also, it is quite easy to throw away noise performance with a JFET oscillator if the large signal condition causes excess saturation in the active device when it is conducting (during part of the RF cycle). So the phase noise performance of a real oscillator might not achieve theory. Also when the oscillator reaches a steady state the JFET device only conducts for part of the RF cycle so this makes things a bit vague when trying to apply Leeson's equation. However, in my experience it is possible to predict the phase noise of a JFET oscillator once it has been optimised. The graph below predicts -135dBc/Hz at 20kHz offset for the Racal LO and the spec sheet in the manual says it will be better than -125dBc/Hz. In reality it might be something like -130dBc/Hz at 20kHz offset?

This will give you some idea of what spec you need from your signal generator if you wanted to do any IMD testing at a 20kHz offset. Ideally, the phase noise of the sig gen should be >10dB better than the Racal LO. Reading between the lines, the reciprocal mixing from the Racal 1772 LO will limit the dynamic range and this might be why they define the IMD performance test at a certain drive level (-23.5dBm per tone?). This will generate IMD terms at -90dBc according to the manual and this is very close to the reciprocal mixing spec given earlier. So you may find yourself exploring the limits of the Racal LO before you find the IMD limits of the front end. One way around this would be to connect a sound card to the AF out of the Racal and display the AF spectrum with a narrow resolution bandwidth. You would then be able to see the IMD term easier. I assume you can turn off the AGC when doing tests like this.

[G0HZU_JMR]

At the lower end of the 1772 receiver range the first LO will run at just over 35MHz so the phase noise might be 6dB lower than the curve given above.
So this means the sig gen has to be cleaner too. However, I would hope/expect the phase noise of the Marconi TF2002 sig gen to be quite good at lower frequencies as it uses a fundamental LC oscillator.

The phase noise of an LC oscillator at a TF2002 sig gen (receiver test) frequency of 3.5MHz should be a low lower than the phase noise of the Racal LO running at 3.5+35.4 = 38.9MHz. However, a lot depends on how careful the TF2002 oscillator designer was. The oscillator in the TF2002 appears to have a varactor diode (or the BJT equivalent on some ranges?) and this may degrade the phase noise quite a bit. Once you start testing up at 29MHz then there is much less margin so the TF2002 phase noise probably won't be 10dB better than the Racal LO at a 20kHz offset.

This is because the difference is only 29MHz for the sig gen vs about 65MHz for the Racal LO at a receiver test frequency of 29MHz.

Has anyone ever measured either of them?

[Radio Wrangler]

I've just realised that it was 28 years ago when I wrote in the ARRL handbook that high linearity amplifiers and mixers along with good filter designs had solved a lot of receiver performance issues, leaving the local oscillator phase noise as the critical limitation and that to go further, this is what we needed to work on.

Varactor diodes aren't too bad on noise. For a number of years the radio astronomer's front end of choice had to be the parametric amplifier, which used a varactor as the active element. Admittedly cooled as far as possible. The noise contribution of a varactor diode in an oscillator comes via two mechanisms. The varactor diode is often the limiting factor in the resonator Q, and the ESR is real resistivity of the bulk material so there is johnson noise associated with this. The effect on resonator Q does most of the damage of the two mechanisms, in an oscillator. It is also pretty much the noise determining thing of a tuneable oscillator. Consequently, a synthesiser's oscillator which has to have significant tuning range under varactor control, is liable to be several dB noisier than one with a mechanical capacitor doing main tuning even if there is avaractor for a bit of frequency modulation.

The Racal, if I remember, carves its 30MHz tuning range up into three segments with three VCOs.

Another ploy in better synthesisers is to switch fixed capacitors as a pretune mechanism, diluting the influence of varactors.

So I think the old 2002 might not be too bad. and it gets better at greater offsets.


If all else fails, build a pair of crystal oscillators in the wanted band with the chosen offset.

David

[G0HZU_JMR]

If we assume that the IMD testing will be done at a 20kHz offset as per the Racal manual then my previous analysis suggests that the sig gen needs to have a phase noise of better than -140dBc/Hz at a 20kHz offset to allow a useful margin against the phase noise of the Racal first LO.

The TF2002B will probably be able to meet this requirement for test frequencies of 15MHz and below. It might also be able to achieve this up at 29MHz but I doubt there will be much margin. In a straight LC oscillator shootout between a 29MHz test oscillator and a 65MHz LO oscillator there will only be a 6dB difference in phase noise if similar technology is used because the LO oscillator is at twice the frequency of the 29MHz test oscillator so it should be 6dB worse. So the TF2002B might not meet the requirement here.

The alternative would be to consider something like the Marconi 2018/2019 sig gen as this should be able to meet the -140dBc/Hz phase noise requirement at 20kHz offset when used at test frequencies across the HF band. At higher offsets the phase noise of the 2018/2019 will probably be limited to -145dBc/Hz (maybe -150dBc/Hz) and this is due to the limitations of the divider system it uses and also the signal path after the oscillator and divider network. This -150dBc/Hz noise floor limitation is fairly typical for a modern lab sig gen.

The TF2002B might be cleaner than the 2018/2019 at 100kHz offsets because it uses a fundamental LC oscillator but this would only be of benefit for a far out blocking test and you already have a single TF2002B for this test.

[Alan_G3XAQ]

I had to look up Leeson's formula (it's on Wiki) but it doesn't help much, in the sense that you need ways to estimate the loaded Q of the resonator and the amplifier noise factor. Is there a way to go about this when an oscillator is running into saturation on RF peaks, as most circuits do?

Alan

[Draker4c]

Thanks very much for your detailed expositions of my questions everyone. None
of your detailed observations and advice would be easily found anywhere but on this forum!
73 Hugh G1AUR

[Draker4c]

I assume use of the 2017B digital syncroniser in conjunction with the TF2002B would lead to
unacceptable phase noise increase. I am basing this on it being a fairly early digital design.

[G0HZU_JMR]

With a typical JFET oscillator I think it is quite difficult to apply Leeson's equation because the active device isn't in conduction all the time once the oscillator reaches large signal equilibrium. I think the small signal (startup) loaded Q will be quite different to the large signal conditions because of this.

Many years ago I designed a 10.7MHz 'trainer' oscillator that was designed to demonstrate Leeson's equation but this used a 50R MMIC device running in class A and this hopefully was in conduction for the whole RF cycle. This allowed the oscillator power at the MMIC input to be measured quite accurately and the loaded Q could be measured by breaking open the loop and using a VNA to measure it based on the slope of the phase through the resonator. The oscillation frequency could also be predicted very accurately in open loop and also on a simulator. I also selected a MMIC that gave very little flicker noise. The results when measured on a E5052A were within a few dB of theory from offsets of 100Hz right out to 100kHz where the phase noise was as low as -170dBc/Hz. There was a step attenuator in the feedback loop and this allowed the loop gain to be changed and this could then be used to look at the effects of saturation on the phase noise. I generally degrade the device noise figure by several dB to allow for some saturation effect and also for losses in the resonator.

With a JFET things aren't so easy and in my experience the DC operating point and the feedback have to be optimised to prevent significant degradation in the phase noise. I have a fair bit of design experience with JFET oscillators across HF and into VHF but I wouldn't want to design an ultra low phase noise JFET oscillator for mass production because of the spread in device characteristics. Once the circuit reaches equilibrium after startup the resonator gets topped up by the JFET with short bursts/spikes of current once every RF cycle. To get the very best phase noise I think this requires quite careful design. By contrast, the MMIC oscillator will give results very close to theory with little or no fiddling at all.

Looking at the Racal circuit they use a current source at the JFET source pin and the current level of this source is controlled by an AGC loop. So presumably the JFET oscillator output level will be controlled by this.

[G0HZU_JMR]

I assume use of the 2017B digital syncroniser in conjunction with the TF2002B would lead to
unacceptable phase noise increase. I am basing this on it being a fairly early digital design.

It could cause poor phase noise very close to the carrier but hopefully the impact on the phase noise at a 20kHz offset would be minimal. I'm really just guessing though.

If you bought a 2019 sig gen instead of another TF2002 you could always arrange the sig gen frequencies such that the cleaner sig gen would be the one closest to the Racal receiver frequency. This would put the dirtier sig gen 40kHz away rather than 20kHz away. This would minimise any reciprocal mixing effects from the dirtier sig gen. Also, the 2019 doesn't suffer much from isolation effects when combining them to deliver -23.5dBm test tones from a typical combiner.

I generally use a classic -6dB hybrid combiner using a toroidal current balun although a resistive attenuator can work well too especially if the receiver has a poor input match. A classic 0.333/0.111 resistive combiner will have 9.5dB loss and about 19dB isolation assuming a 50R DUT impedance. This will be a bit better than a regular 6dB resistive splitter.