Siglent SSA3021X Measurement Anomolies

[dnessett]

I have run into a few measurement anomolies during some testing I am doing. These are specific to the Siglent SSA3021X Spectrum Analyzer (SA). I am posting this message to get some advice on what may be causing these problems.

The test setup is shown in figure 1. There is an FEI FE-5650 Rubidium oscillator sourcing a 10 MHz sine wave (the red box at the lower left corner of the figure). It is connected to a Rigol DS1104Z (middle of the figure) by a coax and BNC T-connector, which forwards the signal to another coax connected to the SA (upper right of the figure). The SA connects to the coax through a N to BNC adapter. I warmed up the Rubidium oscillator for an hour before making measurements.

Figure 1

I first looked at the peak power level of the sine wave. The plot has the following parameters: start Freq - 8 MHz; stop Freq - 12 MHz; RBW - 30 Hz; SA internal attenuator - 20 dBm; SA Ref level - 0 dBm; Peak detection is turned on as is the Peak Table (figure 2).

Figure 2

The first thing I noticed was the power of the 10 MHz signal kept decreasing. The 10 MHz signal power began at 3.52 and then decreased as the number of sweeps increased. For example, here is the first series I observed: 3.52, 3.51, 3.50, 3.48, 3.47, 3.46 .... I wondered if this had something to do with a problem in the SA, so I turned it off for 10 minutes and then turned it on again without changing the other parts of the setup. However, the first measurement after 2nd turn on was 3.41 dBm, so turning the SA off and on didn't help.

I then disconnected the coax from the SA and reconnected it while the setup was running. The power of the 10 MHz signal went up to 3.49 dBm, but then started to decrease as before. I disconnected the coax a third time and the 10 MHz power went up to 3.46. It then remained stable at 3.44-3.46 dBm.

I have no idea what caused this. Perhaps some sort of static charge accumulated at the SA input. Perhaps it has something to do with the coax cables, the Rigol DS1104Z inputs, the BNC T-connector, the N to BNC adapter or the Rubidium oscillator input. I am totally stumped by this behavior. It is the first issue concerning which I am seeking commment.

Next, I used the channel power option of the SA to measure the power in a 4 MHz band around the 10 MHz fundamental frequency. The results are shown in figure 3. The first thing to notice is the measured channel power is 3.66 dBM. I used this calculator to compute the difference between the channel power and that of the 10 MHz signal power: -9.83 dBm. This should be the value of the noise power in the 4 MHz band around 10 MHz.

Figure 3

However, notice in figure 3 the value circled in red at the top right of the picture. This shows the value of the center frequence marker, which according to the figure is -18.83 dBm. Huh? Well, if you look at the second thing circled in red on figure 5, you see that when I used the Channel Power option, the SA changed the Detect Type from Peak Power to Average Power. OK, but the two figures are over 20 dB apart. Surely, averaging the power, rather than detecting the peak shouldn't decrease the measured value by that much. (Ignore the value circled in blue, it relates to something I document in my internal notes)

The next day, I decided to reproduce the peak power level test and then instead of selecting the channel power option to simply switch the detect mode to Avg. and see what happens. (I mention that these tests took place the next day, since I ran into the decreasing power level problem again. The peak power value is different than the 3.44-3.46 dBm observed the day before. In fact it was 3.51 dBm)

Figure 4 shows the results. It shows the 10 MHz power as -78.43 dBm. This is obviously wrong, so I turned the SA off and then on again. Before selecting the power spectrum plot, I switched the detect mode to Avg. The results were effectively the same.

Figure 4

Maybe I misunderstand the meaning of Avg. power detect mode. I thought it would average the values within a 30 Hz bin instead of using the maximum of those values. In any case, since the SA automatically switches to Avg power detect mode when using the Channel Power option, I can't get a simultaneous value for both the channel power and the center frequency power in order to compute the noise power (i.e., channel power - center frequency power).

Comments?

[tautech]

Member hendorog has mine ATM so I can't check specifics for you ATM sorry.

Not being very experienced with SA's has me leaving mine to start on factory presets as I've found myself 'bushed' in the past until you've got to know how to set them correctly.
Averages could well have something to do with what you see and they're running @100 IIRC in factory default.

There's some useful operating tips and app notes on the US website:
https://www.siglentamerica.com/spectrum-analyzers/ssa3000x-series-spectrum-analyzers/

[hendorog]

I'll try and replicate what you are seeing - my feeling is that there is nothing wrong here, but won't spout off until I've first checked my facts

[hendorog]

Here are my thoughts. I didn't spend too much time on the power issue as that will be time consuming to replicate with such a slow sweep.

This is a guess and hopefully a guru will chime in here. I'm thinking that using such a wide slow sweep might be causing this due to a bit of drift between the two devices, and the finite frequency resolution of the instrument and shape of the RBW filters. It is a very small drift in power that you are talking about.
I wonder what you would see if you checked with the rubidium output split, and fed into the input, as well as into the 10MHz reference input in the back to eliminate the drift?

Also I noticed that your peak of +3.5dBm is slightly above the reference level of 0dBm that you have set, even though it appears on the screen to be below 0dBm?

Edit: The below is wrong, the trace changes into Average detector mode when you switch on a Noise marker, or are in Channel Power mode.
The channel power reading shows the value referenced to 1Hz (dBm/Hz). If you play around with Noise Markers on the device you will see that the _entire trace_ changes to units of dBm/Hz when you switch a marker to this mode. So this reduces the peak value accordingly. One issue - I think the units shown against the marker at the top of the screen should change to dBm/Hz but they don't.

I think this is normal for the device, but is different to how other SA's work, they leave the trace alone and just show the dBm/Hz value at the marker.

To show both peak and channel power at the same time I think you need to turn on a second trace (Trace button, then select trace B and press Clear Write)
Put a marker on the peak on trace B and leave trace A in channel power mode.

[rf-loop]

Comments?

You use FEI for level reference. So, with SSA you are more like testing FEI level drift than SSA drift.
But also need note SSA specifications for level accuracy (what also include some short time drifting)
but here clearly DUT is not SSA, DUT is your signal from totally unreliable source what have nothing to do in class of level references.

It is also good to really understand even basics how different detectors works before start talking SSA itself anomalies. Anomalies are now mostly outside of SSA in user side (My suspect is that FEI is total crap for use as level reference and perhaps also need more some study. I have previously used several FEI RB's and still have some in home in some junk box. Of course they are perhaps somehow ok for some kind low level frequency reference if short time stability needs are not important at all but in any case they are not reliable level reference)!

This Keysight AN-150 tell also lot about how SSA3000X works when you look full digital IF systems. Least this AN or other basic fundamentals is good to know and understand enogugh deeply before start claiming something works wrong or right.

[dnessett]

...
I wonder what you would see if you checked with the rubidium output split, and fed into the input, as well as into the 10MHz reference input in the back to eliminate the drift?

Thanks for your comments.

Instead of splitting the signal to the Scope and SA, I will try connecting one branch to the 10 MHz input of the SA and the other to the SA signal input. I will report back the results.

Also I noticed that your peak of +3.5dBm is slightly above the reference level of 0dBm that you have set, even though it appears on the screen to be below 0dBm?

I think that is just an artifact of the display, since there is no screen real estate above 0 dBm to show the marker symbol. I don't think this affects the measurement.

To show both peak and channel power at the same time I think you need to turn on a second trace (Trace button, then select trace B and press Clear Write)
Put a marker on the peak on trace B and leave trace A in channel power mode.

I will try using a second trace and see what happens.

[dnessett]

You use FEI for level reference. So, with SSA you are more like testing FEI level drift than SSA drift.

Thanks for the comments.

The testing I am trying to do is characterize the performance of a 10 MHz signal distribution amplifier. I need to get the baseline data of the input in order to see how faithfully the distribution amplifier replicates its signal. So, my original intention was not to test the SA. I created this post because I was having trouble getting reliable data for the input (the FEI RB oscillator).

But also need note SSA specifications for level accuracy (what also include some short time drifting)
but here clearly DUT is not SSA, DUT is your signal from totally unreliable source what have nothing to do in class of level references.

It is also good to really understand even basics how different detectors works before start talking SSA itself anomalies. Anomalies are now mostly outside of SSA in user side (My suspect is that FEI is total crap for use as level reference and perhaps also need more some study. I have previously used several FEI RB's and still have some in home in some junk box. Of course they are perhaps somehow ok for some kind low level frequency reference if short time stability needs are not important at all but in any case they are not reliable level reference)!

The ~3.5 dBm figure from the peak power measurement is in the ball park for the expected value. The RB oscillator outputs a ~ 1 V p-p signal, which is about 3.98 dBm. So, the SA seems to work fine using peak power detection and not so well using averaging. It is when averaging is turned on that problems appear.

This Keysight AN-150 tell also lot about how SSA3000X works when you look full digital IF systems. Least this AN or other basic fundamentals is good to know and understand enogugh deeply before start claiming something works wrong or right.

Thanks for the reference. I will look it over and see if it helps explain the problem.

[rf-loop]

The ~3.5 dBm figure from the peak power measurement is in the ball park for the expected value. The RB oscillator outputs a ~ 1 V p-p signal, which is about 3.98 dBm. So, the SA seems to work fine using peak power detection and not so well using averaging. It is when averaging is turned on that problems appear.

I do not understand what you want with average for single frequency constant carrier. What you want measure? If you want measure this kind of (CW) carrier level, Peak detector is for this.

Usually when measure carrier power level it is suitable to use Peak detector. It is right detector for measure carrier level.
If you have random noise plus "sinusoidal type" carrier you can reduce noise using example trace average and/or adjusting VBW more narrow than RBW.

Other type detectors need carefully think in what situation and for what they are good to use and why. It need real reason.

Note that also can use up to 4 traces and all may have different detector and many other settings so you can combine these to same display. Also trace draw mode can be diferent for all traces, as example trace average, max hold, min hold and so on.
 

[dnessett]

I wonder what you would see if you checked with the rubidium output split, and fed into the input, as well as into the 10MHz reference input in the back to eliminate the drift?

To do this I had to change the test setup. Both the 10 MHz and RF input to the SSA3021X are 50 ohm. If I just used a T-connector, the RB oscillator would have been driving a 25 ohm equivalent load.

So, I used the distribution amp that is the ultimate objective for testing. I connected the output of the RB oscillator to the input of the distribution amp (a HP5087A) and connected two of its channels to the SA, one to the 10 MHz input and one to the RF input. However, the dist. amp is set up to amplify the input signal, so the input to the SA is now 8 V P-P, which means I had to modify the SA parameters. For this test these were: Ref - 30 dB and Attenuation - 40 dB. When I ran the Peak power test, the results were as shown in figure 5.

Figure 5

I then delected Avg Power mode using the detect menu. The results are shown in figure 6. As you can see driving the SA with the 10 MHz signal from the RB oscillator did not solve the problem.

Figure 6

To show both peak and channel power at the same time I think you need to turn on a second trace (Trace button, then select trace B and press Clear Write)
Put a marker on the peak on trace B and leave trace A in channel power mode.

I tried this, as well (using the original test setup: no connection to the SA 10 MHz external input and the RB oscillator connected directly to the SA). The results are shown in Figure 7. You can see the problem persists. Also, the SA did something strange when I tried this out. I first selected trace A and set up the parameters for the Peak Power test. Trace A showed P-PK as the selected detect mode. I then selected Trace B and set up the channel power option.

Figure 7

However, the SA seemed switched the configurations of the two traces. Trace A displayed the Channel Power results and Trace B the Peak-Power results. This suggests that the SA can only do the Channel Power measurement on Trace A.

I then tried two more variations. I changed the RBW to 300 Hz and then to 3KHz. In the first case (300Hz) the Avg power of the center frequency went to -59.61 dBM from -82.23 dBm (for 30 Hz). In the second case (3KHz) it went to -60.52 dBm.

[hendorog]

30dBm is scary high power, a quick look at the specs and the limit is 33dBm for three minutes with attenuation > 20dB. Personally I'd use something - e.g. external attenuator - to bring the power down before it is fed into the SA. That would reduce the chance of blowing the front end up with what otherwise would be a minor mistake on the keyboard.

One other thing I am confused about :- originally one of the issues you noticed was a small drift down in the power measurement. I don't think you can measure that with a distribution amplifier in the loop, as that might be drifting on its own.

I'm not in front of the SA right now, but I expect you are correct and the power measurement will operate on Trace A. I am not sure if that is configurable but I expect the place to look is in the Measurement Setup menu.

[dnessett]

I do not understand what you want with average for single frequency constant carrier. What you want measure? If you want measure this kind of (CW) carrier level, Peak detector is for this.

Let me try this one more time in a bit more detail.

I purchased a used HP5078A distribution amp from Ebay. It has one 10 MHz input and 12 channels of 10 MHz output. I want to know what sort of distortion each channel introduces to the 10 MHz input signal. Since I will drive this dist. amp with the RB oscillator mentioned in this thread, I need to characterize the RB oscillator output and then measure the output for each dist. amp channel. By comparing the dist. amp input to a channel output, I will get some idea of the channel distortion.

There are several things to consider: 1) inter-channel delay (i.e., the difference between the input-output delay for each channel), 2) differences in the power for each harmonic of the input and each channel output, and 3) differences in the noise near the center frequency between the input and each channel output. The topic of this thread arose when I was trying to measure the parameters for the input related to the third item.

To obtain an estimate of the noise power, I looked at the spectrum of the RB oscillator between 1 MHz and 100 MHz (while getting some data for harmonic power). I eye-balled the hump in the spectrum around 10 MHz and decided to measure the noise power in a 4 MHz band around the center frequency (i.e., 10 MHz). If I measure the channel power in this band and subtract the power for the center frequency, then I will get an estimate of the noise power near the center frequency. (From this point on, I define noise power as this figure).

Before anyone posts a comment pointing out that channel power minus center frequency power contains a bunch of components that are caused by separate factors - I know this. If fact there is a good explanation of these factors here starting on page 13. The principal noise power component I am interested in is phase noise, which is contained in the noise power specified above. Since it would be tedious to attempt to isolate phase noise from the total noise power and since specifying the difference between the noise power of the input and the noise power of each channel output characterizes the introduced noise of the channel, I decided to just use the noise power as the characterizing parameter.

Now to address the question you directly posed above. I am not interested in avg power of the center frequency. I am interested in channel power. However, when I use the SSA3021X channel power option, it changes the SA detect mode to avg power. Since I want to subtract the center frequency power from the channel power and since (as described in my original post) I observe drift in the center frequency power over time, I want to get both values from the same set of data. Of course, I can get a pretty good estimate by first measuring Peak-Power of the center frequency and then quickly measuring the channel power. But, it would be nice if the SA would compute these two values from the same sweep data.

[dnessett]

30dBm is scary high power, a quick look at the specs and the limit is 33dBm for three minutes with attenuation > 20dB. Personally I'd use something - e.g. external attenuator - to bring the power down before it is fed into the SA. That would reduce the chance of blowing the front end up with what otherwise would be a minor mistake on the keyboard.

Yes, you're right. I should have done that.

One other thing I am confused about :- originally one of the issues you noticed was a small drift down in the power measurement. I don't think you can measure that with a distribution amplifier in the loop, as that might be drifting on its own.

I still don't understand what is causing that drift down. However, it isn't one of the things I intend to measure when characterizing the distr. amp.

I'm not in front of the SA right now, but I expect you are correct and the power measurement will operate on Trace A. I am not sure if that is configurable but I expect the place to look is in the Measurement Setup menu.

I looked in the Measurement Setup menu and the manual and could find nothing about advanced measurements being limited to trace A.

[dnessett]

I thought I might have a workaround for the channel power option changing the detect mode to Avg power. I selected the channel power option and then used the detect menu to reselect Peak-Power. That fixed the marker power value, but the channel power value was then way out of bounds - Figure 8.

Figure 8

I also poked around a bit and found other users who had problems with the Avg power detect option. One, here on the EEVBlog, observed anomalous doubling of sidebands when using the power (rather than log power) sub-selection for Avg power.

Another, here reports the average log power sub-selection of Avg power reported an erroneous result.

For the record, my FW version is 1.2.8.5a.

[tautech]

For the record, my FW version is 1.2.8.5a.

Then I'd advise you to update it and when you ask if the issues you see have been fixed.....I don't know as not all fixes are always listed in the changelog.

Changelog and link available here:
https://www.eevblog.com/forum/testgear/siglent-ssa3000x-spectrum-analyzers/msg1497604/#msg1497604

[hendorog]

I thought I might have a workaround for the channel power option changing the detect mode to Avg power. I selected the channel power option and then used the detect menu to reselect Peak-Power. That fixed the marker power value, but the channel power value was then way out of bounds - Figure 8.

I think it is simple to show both values in one sweep. Use two traces as confirmed by rf-loop. However you need to use them in the reverse order that you tried as Channel Power mode is limited to Trace A only as you discovered yourself.

Trace A = Channel Power mode. Leave detector at Avg. Do not assign a marker to this trace.
Trace B = Peak Power. Leave detector at Peak. Assign Marker 2 to trace B and the marker will correctly read the Peak value.

Or am I missing something? 

In terms of the measurement, the SSA I was using was showing a sensible channel power value. I didn't see the same issue as you. So the firmware upgrade is likely the difference there.

[dnessett]

I think it is simple to show both values in one sweep. Use two traces as confirmed by rf-loop. However you need to use them in the reverse order that you tried as Channel Power mode is limited to Trace A only as you discovered yourself.

Trace A = Channel Power mode. Leave detector at Avg. Do not assign a marker to this trace.
Trace B = Peak Power. Leave detector at Peak. Assign Marker 2 to trace B and the marker will correctly read the Peak value.

Or am I missing something? 

In terms of the measurement, the SSA I was using was showing a sensible channel power value. I didn't see the same issue as you. So the firmware upgrade is likely the difference there.

Thanks to both you and tautech. I upgraded to 1.2.9.1,  followed your instructions and, as you indicated, I now get both the peak center frequency value and channel power value correctly. Just for the record, here is the plot - Figure 9.

Figure 9

I imagine this is an operator error on my part, but when I followed your instructions, the SA changed both trace A and trace B to Peak-Power mode when I set trace B to Peak-Power. I had to reselect trace A and set it back to Avg mode to get the plot shown here.

[hendorog]

Thanks to both you and tautech. I upgraded to 1.2.9.1,  followed your instructions and, as you indicated, I now get both the peak center frequency value and channel power value correctly. Just for the record, here is the plot - Figure 9.

Figure 9

I imagine this is an operator error on my part, but when I followed your instructions, the SA changed both trace A and trace B to Peak-Power mode when I set trace B to Peak-Power. I had to reselect trace A and set it back to Avg mode to get the plot shown here.

Great result.

I'll have another play tonight, there could well be more to selecting different detectors for each trace than what I thought.

[tautech]

 

What we didn't do is ask what FW was on your unit first. 
Sorry it's been a bit of a journey for you but that's how we all learn, isn't it. 

Thanks to hendorog for his help on this one as he's got my demo for a fiddle around against his new/old HP and his SH.

[dnessett]

that's how we all learn, isn't it. 

Yep. Thanks again for your help.