Two Tone Test with Scope and SA

[G0HZU]

It's always good to revisit stuff like this. As you can tell my experience is really with spectrum analysers, either old school swept types or modern versions that can do FFT or both swept and FFT. So I'm used to using the classic equations for stuff like this.

I've never had much joy using scopes to measure IMD but then again I've only ever tried this using fairly old scopes and the results were usually quite grim. Some of the screenshots of the modern scopes on this thread look to be very impressive in terms of what the FFT mode can display.

I'm not sure where the significant IMD gets generated in modern scopes. In a typical scope there is usually an impedance converter stage followed by a selectable preamp and then on to the ADC. I think some scopes also split the signal path into low frequency and high frequency paths so it's difficult to know how much the front end affects the IMD generated by the scope.

[2N3055]

It's always good to revisit stuff like this. As you can tell my experience is really with spectrum analysers, either old school swept types or modern versions that can do FFT or both swept and FFT. So I'm used to using the classic equations for stuff like this.

I've never had much joy using scopes to measure IMD but then again I've only ever tried this using fairly old scopes and the results were usually quite grim. Some of the screenshots of the modern scopes on this thread look to be very impressive in terms of what the FFT mode can display.

I'm not sure where the significant IMD gets generated in modern scopes. In a typical scope there is usually an impedance converter stage followed by a selectable preamp and then on to the ADC. I think some scopes also split the signal path into low frequency and high frequency paths so it's difficult to know how much the front end affects the IMD generated by the scope.

Scopes definitely have DC-low frequency/high frequency dual path, you're right about that.
And Mike's intention wasn't to really use (though you could, within limits, of course) scope for measuring IMD of devices.

Two tone technique is routinely used to test linearity of ADCs (that is something Mike knows "a bit" about  ), and he really wanted to test how linear is front end and ADC in new Siglent scopes. This has been ongoing discussion. Results show that they are, in fact, very good...

Thank you for your comments.. You really are RF guru like Performa said, and your contributions are very valued. Even if we get lost a bit sometimes.. 

Best,

[David Hess]

In the case of a DSO I guess a lot depends on the signal handling performance of every stage of the analogue front end before it gets into the digital domain. All these stages can generate IMD ahead of the ADC.

The impedance buffer and first stage contribute most of the distortion because they have to operate over the widest range.  There are circuit techniques like cascodes and bootstrapping to improve this considerably but they increase noise and limit bandwidth.  Integrated front ends have the additional problem of more parasitic elements and coupling; see below about dual integrated JFETs.

I've never had much joy using scopes to measure IMD but then again I've only ever tried this using fairly old scopes and the results were usually quite grim. Some of the screenshots of the modern scopes on this thread look to be very impressive in terms of what the FFT mode can display.

I'm not sure where the significant IMD gets generated in modern scopes. In a typical scope there is usually an impedance converter stage followed by a selectable preamp and then on to the ADC. I think some scopes also split the signal path into low frequency and high frequency paths so it's difficult to know how much the front end affects the IMD generated by the scope.

The impedance converter is what limits performance.  One reason DSOs for a long time were limited to 8-bits, and largely still are, is that the demands on the analog section limit performance to that level or lower anyway without a heroic design, and maybe even then.  Settling time tests of 12-bit DSOs show that they do *not* have 12-bit performance in all respects.  Usually they are not even close, but 12-bit operation is useful in other respects like noise, dynamic range, and marketing.  The performance of some modern DSOs operating as a spectrum analyzer is amazing except for low frequency noise.

One thing modern DSOs can do, and have done for quite some time at the high end, is apply digital correction but this brings up calibration issues.

Scopes definitely have DC-low frequency/high frequency dual path, you're right about that.

All budget oscilloscopes have used a two path design going back to at least the first integrated JFET operational amplifiers.  It removes the expensive requirement for grading JFETs (1) to get good DC performance and later had the advantage of providing AC/DC coupling without a relay.  Some modern instruments might still use a single path design to support faster overload recovery.

(1) An integrated dual matched JFET cannot be used because of cross-coupling.

[Performa01]

In the case of a DSO I guess a lot depends on the signal handling performance of every stage of the analogue front end before it gets into the digital domain. All these stages can generate IMD ahead of the ADC.

The impedance buffer and first stage contribute most of the distortion because they have to operate over the widest range.  There are circuit techniques like cascodes and bootstrapping to improve this considerably but they increase noise and limit bandwidth.  Integrated front ends have the additional problem of more parasitic elements and coupling; ...

Well, at least in case of the Siglent frontends discussed here (which have of course been designed with the assistance from LeCroy), all my tests strongly indicate that the input buffer / impedance converter is not the limiting factor. At least not the HF path, which becomes effective above a few kHz...

It is the PGA that is following the buffer stage. And how could we be surprised? It is a challenge to have a fully integrated variable gain amplifier with up to >2 GHz bandwidth (in case of the SDS6000), programmable in e.g. 2 dB gain steps over a range of 40 dB.

Like you said, true 12-bit performance in a wideband general purpose oscilloscope isn't going to happen anytime soon. But as I demonstrated, there can be sweet spots (read: PGA settings), where the distortion is at its minimum. Alter the gain by just one step and the distortion might rise by several dB immediately.

EDIT: look at the attached datasheet of a typical 900 MHz PGA like it my have been used for the current SDS2000X series. Look specifically at the distortion figures published on page 6. This proves that we actually need to look for sweet spots whenever we want a 12-bit like distortion performance near -70 dBc.

[Performa01]

If it helps I have an early/old Tek spectrum analyser here that has a digital IF. When used down at below 40MHz it is effectively a 40MHz baseband scope sampling at 102.4MHz but the display is trapped in the frequency domain. This is not a swept analyser, it samples the input and generates an FFT just like a scope.

If I inject two clean tones I can generate IMD in this analyser at about -83dBc if I drive it towards FSD of the ADC.

This performance got me curious. Of course, general purpose oscilloscopes cannot compete with dedicated higher end spectrum analyzers, even if they’re old. Yet I wanted to know how close a SDS2000X HD can get.

We want “near full scale” signals now, so the signals are 10 MHz with 20 kHz frequency offset and -10 dBm = 70.71 mVrms = 200 mVpp each, which results in a total amplitude of 400 mVpp.

First test at 50 mV/div. Visible screen range is exactly 400 mVpp. Yet the DSO already indicates possible overrange in its automatic measurements: >100.080001mV(↨)

See 1^st screenshot.

SDS2504X HD_IMD_10MHz_-10dBm_50mV

IMD is about -70 dBc, so it is still worthy of a 12-bit oscilloscope. Even though it cannot touch this particular Tek instrument, there are quite some SA out there which would struggle to achieve a result like this at an input level as high as this.

The point is, that the SA is measured way below the 1 dB compression point of its input amplifier and/or mixer, whereas the SDS2000X HD has the maximum input at 100 mV/div (above that, the first attenuator kicks in). Consequently, 50 mV/div is just 6 dB below the designed “full scale” of the input buffer and PGA input. If we look at the datasheet I’ve attached in my previous post, we can see that the maximum permissible input of the typical PGA is 1V. Consequently, 400 mVpp (after the unity gain buffer) is already close to the permissible input voltage of that stage – certainly not 30 dB below.


Let’s try to get rid of the overrange warning and lower the input gain to 55 mV/div. Now we’ve found a sweet spot at IMD -77 dBc, which is not that far off when compared to the -83 dBc of the dedicated Tek SA.

See 2^nd attached screenshot:

SDS2504X HD_IMD_10MHz_-10dBm_55mV

[RoGeorge]

where the significant IMD gets generated

Distortions can only appear when a circuit is not linear.  A perfectly linear circuit (an ideal LTI - Linear Time Invariant system) can not have any distortions.  In frequency domain, an ideal LTI system will show at its output exactly the same spectrum as it was fed at its input.  The output can only differ in amplitude (the ideal LTI system can amplify or attenuate, but it can not produce new frequencies).  Or else said, an ideal linear system won't have any intermodulation products.

In this topic, intermodulation is only a technique that can evidentiate very small non-linearities that might happen inside the oscilloscope, distortions so small that otherwise would be close to impossible to notice by just looking at the waveform in the time domain.

Intermodulation products get generated anywhere there is a non-linearity.  In the case of an oscilloscope for example, the frontend amplifier is not ideal, it will have some small nonlinearity, therefore some intermodulation products will appear.  Later, the signal get sampled by the ADC then displayed as a spectrum, where we notice the intermodualtion products.  Or, it can be some other components than the input amplifier, where the signal is distorted, even the ADC itself will add some distortions.  It's a test for the whole oscilloscope (assuming the input signal is perfectly clean, with only 2 frequencies).

The TL;DR is, new spectral components will appear anywhere there is a non-linear response.  Non LTI system means the circuit will distort the signal, and we will notice that as intermodulation.  No intermodulation can appear in an ideal, linear amplifier.

I like this video about "Linear and Non-linear Circuits" from "The Signal Path" channel because it tells about intermodulation both in theory and in practice, with experimental measurements to build an intuition:

[G0HZU]

That's very impressive from the Siglent scope. Below 40MHz the Tek analyser runs the input direct to the ADC via an attenuator and presumably a preamp when needed. If I were to try for sweet spot drive levels I think I could get better than -90dBc IMD but I'm not sure I'd trust what is displayed. The main niggle with the Tek analyser is that it has a few low level internal spurious signals even with no input so it has to be used with some care and sympathy for its limitations.

I had a go at retesting the single/dual ERA1 MMIC amplifiers this evening and this time I did it quite formally by measuring each one for gain with a VNA and also I checked they were the same for IMD. I also used a precision 0.1dB step attenuator to set up the correct attenuation between the two amplifiers. See the images below. On the VNA the small signal gain was about 12.3dB for both amplifiers and the IMD performance seemed to be identical. The spectrum analyser plot below shows the IMD for both ERA1 amplifiers when tested standalone and trace 3 is when they are in series with the 12.3dB attenuator in between them. The marker shows the IMD change was very close to 6.0dB when the two amps were in series.

I've used a decent Agilent lab VNA for the gain tests and I set the drive level correctly to capture the correct small signal gain for the IMD tests. I also used the VNA to set the step attenuator within 0.1dB of 12.3dB loss.

So I think this is a good experiment and demonstration of the equations provided by Agilent and R&S. It shows the expected 6dB increase in IMD.

[rf-messkopf]

This thread got me curious. To see where I'm at, I first tried two ways to combine the outputs of two signal generators (a R&S SML and a SMU200A, both with ALC switched to sample-and-hold to avoid any interference between them): A Mini-Circuits 3 dB combiner/splitter, model ZFSC-2-6-N+, with a 10 dB pad at each input, and a 6 dB resistive combiner, also with 10 dB pads at the inputs. I tested at 10 MHz with a frequency difference of 1 kHz, and with 10 dBm input power for the 3 dB combiner, and 13 dBm for the resistive combiner.

With the 3 dB combiner, I measure the 3rd order products at 75 dBc, and with the resistive combiner, they are more than 10 dB higher on a high dynamic range spectrum analyzer (R&S FSIQ 26).

I'm pretty confident that in the case of the 3 dB combiner, the 3rd order products are for the most part not generated by the analyzer, because I choose the input attenuator setting high enough that they are no longer moving when I vary it. Here the 1 dB attenuator option comes in handy.

I haven't investigated this any further, but I suspect that the higher IMD with the resistive combiner is due to the lack of isolation (only the 10 dB pad plus the 6 dB insertion loss of the combiner) between the two signal generators, and the IMD is actually produced in the generators.

Then I tried two tones 20 kHz apart at 1 MHz on a R&S RTM 2054 oscilloscope. That is an older model with an 8 bit ADC. Also, its built-in FFT is a bit compromised, so that for more comprehensive tests I would have to pull the samples on a PC and do the FFT there. However, at that frequency with 100 mV/div I measure the 3rd order products at about -60 dBc (see attachment).

I also verified with the analyzer that they are at least 80 dB down at the scope input.

[David Hess]

How would this measurement be made while avoiding the limits of the analyzer?  Notch out the fundamentals?

[RoGeorge]

The two tone testing is particularly useful when the bandwidth is limited, because this technique is shifting the harmonics close to the two main tones, so their amplitude is preserved.  Otherwise (with a single tone) the harmonics will fall out of the DUT's band and will be attenuated by the DUT's frequency response, appearing much smaller than they really are, or not visible at all.

In the case of a 10MHz test signal a combiner and an SA, there is no bandwidth limitation (I expect the passive combiner to work up to many hundreds of MHz, same for the SA) so the SA can observe the harmonics of a single tone directly, at 2f, 3f, etc. without the need of a second frequency.

It might worth measuring the harmonics of each tone alone, one by one (observed at 20, 30, 40MHz, etc), with the second generator disconnected, in the hope that this measurement will show only the purity of each signal, and will exclude any intermodulation it might happen in a second generator.

An alternate way to know if the intermodulation is produced by the outputs of the generators (acting as mixers) might be to add a directional coupler between each generator and the combiner, so to reduce any signal from outside that is trying to travel into the generator.

In theory, it should work.  In practice, I have close to zero experience or equipment to try this. 

[G0HZU]

Usually the way to get good results is to choose a combiner with high isolation and to try and maintain that isolation through having a good port match at the sum port of the combiner. Also, try and use a lowpass filter on the sig gen outputs. Try and use a spacing frequency that is higher than the ALC bandwidth of the sig gens as this can minimise crosstalk effects between the sig gens.

Another thing to try is to use the attenuator lock/hold feature on the sig gen (if it has this feature) as this can keep some internal attenuation inline when trying for higher output levels from each sig gen.

I have lots of RF combiners here and for use below 200MHz my favourite one has 6dB through loss but over 48dB isolation from LF through to about 200MHz. If I have to use a resistive combiner for wideband stuff I have one here that has about 9.5dB through loss and just under 20dB port isolation.

Don't overlook that fact that a typical microwave spectrum analyser will typically have compromised distortion performance below about 20-50MHz. This is because of the image and LO breakthrough that will degrade the distortion performance of the analyser below about 20-50MHz. So the analyser might become the weakest link unless care is taken with the attenuator settings within the analyser.

It should be possible to get >100dB SFDR using the above techniques if care is taken with the setup.

[rf-messkopf]

Usually the way to get good results is to choose a combiner with high isolation and to try and maintain that isolation through having a good port match at the sum port of the combiner. Also, try and use a lowpass filter on the sig gen outputs. Try and use a spacing frequency that is higher than the ALC bandwidth of the sig gens as this can minimise crosstalk effects between the sig gens.

Another thing to try is to use the attenuator lock/hold feature on the sig gen (if it has this feature) as this can keep some internal attenuation inline when trying for higher output levels from each sig gen.

The datasheet of the ZFSC-2-6+ I was using states about 30 dB isolation at 10 MHz typical. Plus there was a 10 dB pad at each input, so I would expect about 50 dB of isolation. No idea if that is enough. I haven't gotten the VNA out to actually measure it though. The ALC was turned off on both generators, so that it would not cause any trouble.

Don't overlook that fact that a typical microwave spectrum analyser will typically have compromised distortion performance below about 20-50MHz. This is because of the image and LO breakthrough that will degrade the distortion performance of the analyser below about 20-50MHz. So the analyser might become the weakest link unless care is taken with the attenuator settings within the analyser.

True, but I have set the attenuator of the analyzer high enough that the IMD products would not move with a variation of the attenuator. Therefore I suppose that what I'm seeing is not generated within the analyzer.

It should be possible to get >100dB SFDR using the above techniques if care is taken with the setup.

I wonder if the IMD products are generated in the splitter. It is a core and wire model, rated from 2 kHz to 60 MHz, and the core might introduce some distortion. I should check that by varying the drive level at the inputs.

How would this measurement be made while avoiding the limits of the analyzer?  Notch out the fundamentals?

If you can make a filter that is steep enough to notch them out without disturbing the 3rd order products.

The two tone testing is particularly useful when the bandwidth is limited, because this technique is shifting the harmonics close to the two main tones, so their amplitude is preserved.  Otherwise (with a single tone) the harmonics will fall out of the DUT's band and will be attenuated by the DUT's frequency response, appearing much smaller than they really are, or not visible at all.

In the case of a 10MHz test signal a combiner and an SA, there is no bandwidth limitation (I expect the passive combiner to work up to many hundreds of MHz, same for the SA) so the SA can observe the harmonics of a single tone directly, at 2f, 3f, etc. without the need of a second frequency.

It might worth measuring the harmonics of each tone alone, one by one (observed at 20, 30, 40MHz, etc), with the second generator disconnected, in the hope that this measurement will show only the purity of each signal, and will exclude any intermodulation it might happen in a second generator.

This is true, at least in theory. If you model the output voltage of the two-port DUT as a nonlinear function and expand it as a Taylor series, i.e.,
$$v_{\rm out}(v_{\rm in})=\sum_{k=0}^\infty a_kv_{\rm in}^k,$$
and consider it up to order 3, then in a single tone excitation the products at \(3f_0\) are proportional to \(a_3\), and the same is true for the third order products at \(2f_1-f_2\) and \(2f_2-f_1\) in a two-tone test. So the harmonics contain information about IMD products. However, this assumes that the Taylor series is an accurate model, and there are no frequency dependent effects. In practice, I suppose there is no way around a two-tone measurement.

An alternate way to know if the intermodulation is produced by the outputs of the generators (acting as mixers) might be to add a directional coupler between each generator and the combiner, so to reduce any signal from outside that is trying to travel into the generator.

In theory, it should work.  In practice, I have close to zero experience or equipment to try this. 

True, but I guess that a highly isolating combiner is more promising.

[rf-messkopf]

Okay, I had another go at it.

The datasheet of the ZFSC-2-6+ I was using states about 30 dB isolation at 10 MHz typical. Plus there was a 10 dB pad at each input, so I would expect about 50 dB of isolation. No idea if that is enough. I haven't gotten the VNA out to actually measure it though. The ALC was turned off on both generators, so that it would not cause any trouble.

I measured the isolation of the ZFSC-2-6+, see the attachment. I get about 33 dB at 10 MHz with the sum port terminated in a high quality load (the orange Trc3 trace), and about -18 dB isolation with 12 dB return loss at the sum port. provided by a 6 dB attenuator with one port left open (the blue Mem4[Trc3] trace). So I can confirm the 50 dB of isolation with the two 10 dB pads.

I also measured the IMD again, with two 10 dB pads at the inputs of the ZFSC-2-6+, plus a 6 dB pad at the sum port to improve termination. The generators were set to a level of 15 dBm, and the RF attenuation of the analyzer was set to 40 dB. Again, I observe no dependence of level of the 3rd order products when I very the attenuator around 40 dB.

I also tried two analyzers, a R&S FSIQ26 (20 Hz to 26.5 GHz) and a FSEA30 (20 Hz to 3.5 GHz), with consistent results, see the attachments.

I'm still not sure though if the 3rd order products might to some extent be generated at the IF stage of the analyzer, or if it is all due to the splitter. When I set the FSIQ26 to FFT mode, I get the same results (only much faster, with 400 ms sweep time).

[G0HZU]

I can only really offer general advice to the forum on stuff like this as I've not used the R&S gear you have there. Trying for a narrow tone spacing like 1kHz is quite challenging if you want to try for (say) -80dBc IMD. This is well inside the normal ALC bandwidth of a sig gen so this means the sig gens will be very prone to crosstalk issues. One option is to turn off the ALC or to select a lower ALC bandwidth. It looks like your sig gen can change the ALC in other ways as well.

I would try and avoid doing IMD tests at such a narrow tone spacing but if it has to be done then I'd recommend a high isolation combiner, try using the attenuator hold option in the sig gen and turn down the ALC bandwidth if the sig gen permits. The attenuator hold option can typically introduce an extra 10dB isolation per signal generator if used correctly.

See below for a plot from my old Tek analyser. This analyser doesn't sweep the LO and the plot below is taken using an FFT. To get about -90dBc IMD the tone levels have to be 10dB below the reference level on the display. This is why the tones are so low compared to the reference level. I don't really trust this analyser when exploring IMD levels as low as this. I'd normally use something else.

I've added a second plot below where I've turned up the sig gens such that I test with two tones at 0dBm each into the analyser. This gives similar results.

[rf-messkopf]

One option is to turn off the ALC or to select a lower ALC bandwidth. It looks like your sig gen can change the ALC in other ways as well.

I used the sample-and-hold function of the ALC, i.e., it measures the output level once, adjusts, and then holds that level, with the control loop turned off.

I would try and avoid doing IMD tests at such a narrow tone spacing but if it has to be done then I'd recommend a high isolation combiner,

One option might be to use a 4-way splitter, because it has higher isolation between opposite ports than adjacent ones. With this splitter one could achieve more than 50 dB of isolation between opposite ports at 10 MHz, which could be further increased with attenuators. Unfortunately I don't have a suitable 4-way splitter here. And most datasheets don't quote the IMD performance of splitters. I've only seen that for high power splitters intended for cellular radio applications.

See below for a plot from my old Tek analyser. This analyser doesn't sweep the LO and the plot below is taken using an FFT. To get about -90dBc IMD the tone levels have to be 10dB below the reference level on the display. This is why the tones are so low compared to the reference level. I don't really trust this analyser when exploring IMD levels as low as this. I'd normally use something else.

I've added a second plot below where I've turned up the sig gens such that I test with two tones at 0dBm each into the analyser. This gives similar results.

Thanks for running it. I also didn't see much difference with different drive levels.

But I don't mean to hijack the thread with this discussion. Maybe the additional data point for the R&S RTM2054 is of some interest.

[Performa01]

I’ve never tried to use RF-signal generators for dual tone tests, even though I do have two of them available. A single modern dual channel AWG is just so much handier for this and up to now I’ve not had any desire for such tests at frequencies higher than 500 MHz.

I would expect a R&S FSEA30 measurement to provide much better results than what has been shown here. Since I’ve never experienced any problems and always assumed that the analyzer is the weakest spot in tests like this, it becomes obvious that the output stages of a modern AWG must be much less prone to intermodulation distortion from injected signals. Most likely there is no ALC either, even though the accuracy of the output levels can easily compete with the best RF-signal generators, like a specified accuracy of 1 % and +/-0.3 dB flatness over the entire bandwidth.

As usual, actual performance is much better than this and there are sweet spots, like 100 kHz - 500 MHz  +0.06 / -0.04 dB absolute error at -30 dBm. See this old reply #162 in the thread linked below, showing the response of my (original) SDG6052X and comparing it to an older high performance signal generator.

https://www.eevblog.com/forum/testgear/siglent-sdg6000-series-awg_s/msg2621457/#msg2621457

Long story short, the signal quality, stability and accuracy can easily compete – and the insensitive (to intermodulation from outside) output stages are a big plus.

The first attached screenshot shows the IMD at 450 MHz with a resistive wideband power combiner (DC – 12.4 GHz) and two additional 10 dB inline attenuators at its inputs, for a total of 26 dB of isolation.

Ref_450MHz_O200kHz_-10dBm_Iso20dB

We get better than -85 dBc IMD – and mind you, this is not an SA with 110 dB third order dynamic range like the R&S FSEA30. Base line is, that I expect the actual signal to be much better than this. Even without additional isolation, i.e. just the 6 dB of the power combiner alone, I could measure -74 dBc.

I can verify that the signal is better than this with another instrument, but only up to 5 MHz:

Pico4262_IMD_4MHz_O100kHz_0dBm

At 4 MHz I can easily demonstrate the IMD to be better than -97.5 dBc – and once again I hit the limits of the analyzer, not the test signal.

The whole point of this posting is to demonstrate, that with the right tools the quality of the test signal needs not be an issue at all.

[rf-messkopf]

The whole point of this posting is to demonstrate, that with the right tools the quality of the test signal needs not be an issue at all.

Okay, but you are testing at 200 kHz and 100 kHz spacing. I was testing at 1 kHz. That's a bit more challenging. But I'll re-run my measurement at >100 kHz and see what I get.

I guess the main problem I'm having with the IMD level is due to the lack of isolation (about 50 dB in my case) between the signal generators. A better splitter (e.g. a 4-way one) should make an improvement. Still  not 100% sure but I don't think that the analyzer is the weakest link in my setup.

[Performa01]

The whole point of this posting is to demonstrate, that with the right tools the quality of the test signal needs not be an issue at all.

Okay, but you are testing at 200 kHz and 100 kHz spacing. I was testing at 1 kHz. That's a bit more challenging. But I'll re-run my measurement at >100 kHz and see what I get.

I guess the main problem I'm having with the IMD level is due to the lack of isolation (about 50 dB in my case) between the signal generators. A better splitter (e.g. a 4-way one) should make an improvement. Still  not 100% sure but I don't think that the analyzer is the weakest link in my setup.

Well, if you think the frequency spacing is supposed to change the distortion characteristics of a nonlinear device - and that's about the only thing we're interested in – then please find attached the equivalent measurements at 1 kHz spacing.

First attachment is the measurement at 10 MHz with the SA, which should be comparable to yours.

Ref_10MHz_O1kHz_-30dBm_Iso20dB

Thanks to your complaint! Because of this I finally found out (after many years of usage) that my SA performs much better at narrow spans – probably because it uses FFT exclusively instead of swept mode then. Consequently, I now can finally measure an IMD of -92 dBc even at high frequencies!

To get a truly comparable result with my earlier measurements, the second attachment shows the 4 MHz test again:

Pico4262_IMD_4MHz_O1kHz_0dBm

Nothing changed. It is still -97 dBc, hence right at the edge of what this 16-bit DSO is specified for.


I know, there are suspicions that it might have something to do with the ALC in the levelled signal generators, but I was talking about an AWG (which apparently doesn’t have nor need that). And then, if I wanted an appropriate test signal to measure distortion products of a DSO, then I would certainly stay away from conditions (like narrow frequency spacing) that might challenge the signal generators and prevent me from getting a clean dual tone test signal.

Yes, your FSEA30 certainly is not a weak link, yet it should be easy to fully exploit its phenomenal intermodulation free dynamic range with proper signal sources and distortion-free power combiners. I wish I had one here to demonstrate just that…

[rf-messkopf]

Well, if you think the frequency spacing is supposed to change the distortion characteristics of a nonlinear device - and that's about the only thing we're interested in – then please find attached the equivalent measurements at 1 kHz spacing.

The frequency spacing would certainly not change the characteristics of the DUT, but it gets harder to measure levels of 3rd order product reliably at narrow spacing.

Anyway, thanks for testing at 1 kHz.

I know, there are suspicions that it might have something to do with the ALC in the levelled signal generators, but I was talking about an AWG (which apparently doesn’t have nor need that). And then, if I wanted an appropriate test signal to measure distortion products of a DSO, then I would certainly stay away from conditions (like narrow frequency spacing) that might challenge the signal generators and prevent me from getting a clean dual tone test signal.

Not all signal generators have an ALC. Some of the old Marconi ones always run open loop, I think the 2024 is an example. I'll repeat the test with two function generators and see if that makes any difference.

And it does not necessarily have something to do with the ALC (which was disabled anyway), but the IMD seen can also be due to nonlinearities in the generator output stage. Also, the two signal generators (R&S SML and SMU200A) both have electronic attenuators. Maybe that has an influence as well.

Yes, your FSEA30 certainly is not a weak link, yet it should be easy to fully exploit its phenomenal intermodulation free dynamic range with proper signal sources and distortion-free power combiners. I wish I had one here to demonstrate just that…

Not so easy with the stuff I have in the drawer it seems. 

[Performa01]

Not all signal generators have an ALC. Some of the old Marconi ones always run open loop, I think the 2024 is an example. I'll repeat the test with two function generators and see if that makes any difference.

I only had (and still have them) 3 RF signal generators in my whole life. So my practical experience with these is limited. Since none of them appears to have any settings for the ALC, I cannot know whether they have it or not. But at least my first generator, a Wavetek 3000, settles its output amplitude in a way that it's quite obvious that there must be some ill-designed control loop for it

Without the ALC, it's hard to see where there is a difference. If we ignore modulation, there should not be any substantial differences between the output stages of a signal generator and an AWG. We have an output amplifier, which is of course prone to intermodulation when fed with external signals and a step attenuator (which usually covers a much wider range in the signal generator, but nothing that cannot be rectified by using an external device for additional attenuation).

My Siglent AWGs are somewhat old school because we hear several relays click when dialing the output amplitude within its range of -56 dBm and +25.5 dBm (at frequencies up to 40 MHz). Yet I suspect that there will be some additional electronic attenuator to manage the fine steps in between. Since this is just a switch, i.e. a negligible serial resistance within the signal path, I do hope that it won’t cause any significant distortion.

BTW, for all these recent measurements, I've not used the SDG6052X where I demonstrated the amplitude accuracy, but the new SDG7102A, which covers the frequency range up to 1 GHz and can still maintain +13 dBm at that frequency. Normally irrelevant, I want to make clear that I do not compare apples to bananas, by using a 20 MHz AWG, rather than something with a proper high power HF output stage. Of course, in this context it is totally irrelevant what’s been used, as long as it results in a distortion-free dual tone signal that's fit for the purpose.

[G0HZU]

If it helps, I've been doing critical IMD tests like this all my career. In my case I would normally testing an RF downconverter. I designed lots of RF downconverters from about 1990 onwards. These were for early versions of SDR receivers for gov/military use back then. The ADC/DSP back end was in a huge rack system in those days. Very different to today. The usual spec for IMD back then varied from -72dBc to -80dBc for the downconverter but obviously some margin was required beyond this. The aim of the overall system was to digitise the RF spectrum up to several GHz as quickly and cleanly as possible. Often a high performance spectrum analyser was used at the IF in place of the ADC/DSP during dev work.

If IMD tests are done at narrow spacing then this can be inside the ALC bandwidth of the sig gen and this will degrade the performance unless extra isolation is included somewhere. Also, a conventional spectrum analyser will usually have degraded IMD performance below about 50MHz and also the IMD of the analyser can be worse on narrow spans as both test tones travel further together down the signal path of the analyser. In those days the top spectrum analysers for IMD performance were the HP 8568B, Advantest TR4172 and the Marconi 2382 analyser.

It's fairly standard to have ALC in a classic RF lab sig gen because the ALC system is often also used for the AM modulation. This usually means the ALC bandwidth is at least 20kHz although some sig gens can turn the ALC bandwidth right down if AM isn't needed. I'm not aware of any mainstream lab RF sig gens that run open loop unless the ALC is deliberately disabled. I guess sig gens like this do exist but I've not used one unless you include really old or low cost RF signal generators or RF capable function generators.

I've got quite a few RF sig gens here and I've used many different types at work over the years. With care it's possible to achieve -100dBc IMD at the RF combiner output. More is possible but it becomes much more challenging to prove it. When testing really high performance HF receivers the IMD3 SFDR can easily be >110dB so this requires a very good test setup indeed.

[rf-messkopf]

Not all signal generators have an ALC. Some of the old Marconi ones always run open loop, I think the 2024 is an example. I'll repeat the test with two function generators and see if that makes any difference.

Correction: From the block diagram, the Marconi 2024 does have an ALC. But it cannot be switched off, I think, I've never seen such an option in the menu of mine.

Without the ALC, it's hard to see where there is a difference. If we ignore modulation, there should not be any substantial differences between the output stages of a signal generator and an AWG. We have an output amplifier, which is of course prone to intermodulation when fed with external signals and a step attenuator (which usually covers a much wider range in the signal generator, but nothing that cannot be rectified by using an external device for additional attenuation).

My Siglent AWGs are somewhat old school because we hear several relays click when dialing the output amplitude within its range of -56 dBm and +25.5 dBm (at frequencies up to 40 MHz). Yet I suspect that there will be some additional electronic attenuator to manage the fine steps in between. Since this is just a switch, i.e. a negligible serial resistance within the signal path, I do hope that it won’t cause any significant distortion.

Usually, rf signal signal generators have a class A output stage, whereas the classical function generator has a class AB complementary output stage with a separate DC path. No idea how the output amplifier in one of these modern x GSa/S arbitrary function generators  looks like. The electronic attenuator/PGA for fine control of the output level usually is further up in the signal path, before the output amplifier, and likely will not see much of an externally applied signal.

If IMD tests are done at narrow spacing then this can be inside the ALC bandwidth of the sig gen and this will degrade the performance unless extra isolation is included somewhere. Also, a conventional spectrum analyser will usually have degraded IMD performance below about 50MHz and also the IMD of the analyser can be worse on narrow spans as both test tones travel further together down the signal path of the analyser. In those days the top spectrum analysers for IMD performance were the HP 8568B, Advantest TR4172 and the Marconi 2382 analyser.

Right. The 115 dB intermodulation free dynamic range of the FSEA30 is specified for inputs from 50 MHz onwards. But as I said above, I would expect that the intermodulation products I have seen would depend on the attenuator setting if they were generated in the first mixer. And if they were generated further down the signal path at an IF, then they would also be a problem for input frequencies above 50 MHz. Perhaps this could be checked by varying the reference level.

Anyway, some further tests are necessary. Currently I suspect the signal generators and/or the splitter.

[mawyatt]

Usually, rf signal signal generators have a class A output stage, whereas the classical function generator has a class AB complementary output stage with a separate DC path. No idea how the output amplifier in one of these modern x GSa/S arbitrary function generators  looks like. The electronic attenuator/PGA for fine control of the output level usually is further up in the signal path, before the output amplifier, and likely will not see much of an externally applied signal.

Interesting discussions!!

The AWG output amps are usually wide-band higher performance op-amp types with very little introduced non-linearity. TI and AD have created op-amps especially for this type of use.

The mentioned "relay clicking" when using built-in two tone signals from some of the modern AWGs (For example Siglent's SDG2042X and 6022X) hint that this is combining the separate tones from each channel signal path in the analog domain and not the digital domain, which may provide better results than a simple digital combine.

An interesting IMD test might employ a couple wide-band quality linear Power Amps buffering each individual tone source before combining into a single two tone signal. Since the required two tone signal is low relative to the linear PA output capability, maybe this PA wouldn't introduce much additional non-linearity and it's output wouldn't be effected as much by the "other" signal, thus reduced IMD. We've utilized high power wide-band low noise linear PA types as receiver front ends for high input dynamic range in "special" type receiver applications, and might prove useful here.

Best,

[2N3055]

Usually, rf signal signal generators have a class A output stage, whereas the classical function generator has a class AB complementary output stage with a separate DC path. No idea how the output amplifier in one of these modern x GSa/S arbitrary function generators  looks like. The electronic attenuator/PGA for fine control of the output level usually is further up in the signal path, before the output amplifier, and likely will not see much of an externally applied signal.

Interesting discussions!!

The AWG output amps are usually wide-band higher performance op-amp types with very little introduced non-linearity. TI and AD have created op-amps especially for this type of use.

The mentioned "relay clicking" when using built-in two tone signal from some of the modern AWGs (For example Siglent's SDG2042X and 6022X) hint than this is combining the separate tones from each channel signal path in the analog domain and not the digital domain, which may provide better results than a simple digital combine.

An interesting IMD test might employ a wide-band quality linear Power Amp buffering each individual tone source before combining into a signal two tone signal. Since the require two tone signal is low relative to the linear PA output capability, just maybe this PA wouldn't introduce much additional non-linearity and it's output wouldn't be effected by the "other" signal, thus reduced IMD. We've utilized high power wide-band low noise linear PA types as receiver front ends for high input dynamic range in "special" type receiver applications, and might be useful here.

Best,

Mike,

SDG series  from Siglent definitelly combine in digital domain...

Relay clicking is from mundane reason. If amplitude of single channel is right there on the edge where it would go between output ranges, once you combine amplitudes it will be large enough for output relays to "switch to higher gear"..

Best,

[mawyatt]

Mike,

SDG series  from Siglent definitelly combine in digital domain...

Relay clicking is from mundane reason. If amplitude of single channel is right there on the edge where it would go between output ranges, once you combine amplitudes it will be large enough for output relays to "switch to higher gear"..

Best,

Thanks, that's why I said "hint" since didn't know for sure. Suspect Siglent had done the proper evaluation for the Digital vs. Analog combining of signals from each channel and weighted the pros and cons. Starting with 16 bit DACs likely helps as well, which gives good resolution.

Best,