SDG6022X noise with spurious spikes

[baldurn]

Hello

I just got a Siglent SDG6022X waveform generator together with a SSA3032X-R spectrum analyser. I am a newbie so first thing is of course to measure one with the other. During testing the noise generator I noticed two things. First although this is a 200 MHz generator (I have not hacked it yet) it does generate noise through 500 MHz. Not completely flat however.

But what worries me is the two spikes at 300 MHz and 600 MHz. The spikes seems to have a lot of power compared to the noise.

Is this normal or a known problem?

Thanks,

Baldur

[hpw]

Will only comment to ScreenImg (4).png:

I have this closing PN/Jitter signal also from my none RT siglent SSA (reported on this forum).

Just loop back your 10Mhz reference from your SSA and test again.

You will see than to PN/Jitter from the internal synthesizer. An external 10MHz OXCO feed to the SSA will not help in any regards.
Just use the spectrogram plot to see more details. Also the Windows SW do not show 1Hz RBW as reported and waiting since a year for an update 

[gf]

DDS sample rate of this AWG is 300 MSa/s, and 600 is the 2nd harmonic. I don't believe it's just a coincidence.

The spikes seems to have a lot of power compared to the noise.

No, the total noise power is significantly higher, but the noise power is spread over the full ~500 MHz bandwidth, so that the power density (power per Hz bandwidth) is still low at the end. OTOH each spike concentrates its whole power in a very narrow band, so that 100% of its power fall into the RBW. For the noise, the RBW covers only a small fraction of the noise's bandwidth, therefore the RBW filter captures only a small fraction of the total noise power at each frequency.

[TurboTom]

I tried to replicate your findings (SDG6022X, "liberated", F/W 6.01.01.36, H/W 03-00-00-78-38) + DSA815-TG. I couldn't find any of the significant sampling frequency feedthrough superimposed to the signal (100mV standard deviation, 0V offset, 50 Ohms load impedance, no bandwidth limiting). Moreover, at Nyquist frequency (600MHz), it shows a dip and at 300MHz, when deliberately searching for a signal, there was something to be found (please be aware that the DSA815's lowest RBW is 10Hz), but considerably below the generated noise level at 100kHz RBW, that's why it wasn't observable at these settings.

It's difficult to tell exactly what may cause this discrepancy, it could be related to (a change in) Siglent's oversampling approach or possibly a modified reconstruction filter configuration. Can you check overshoot of 1ns and 2ns rise time square / impulse edges on your AWG for comparison?


Edit: I guess I've got to revise my statement from yesterday a little: I re-tested your setup again today since yesterday, my preset noise level was higher than yours. I changed the standard deviation to 48.4mV which should approximately match your settings and -- indeed -- I found some of the feed-through at 300MHz, though considerably less than what seems to be present on your AWG. Moreover, the noise "passband" spectrum on my AWG seems to be slightly flatter and the drop-off shoulder seems considerably steeper which really makes me believe that Siglent may have modified the reconstruction filter. I didn't find any noticeable difference between the two channels, moreover, enabling the "idle" channel with maximum output level of 300MHz sine doesn't change the figures of the noise channel being tested. See the third screenshot for details.


Edit: Typo, Clarification, Add-on

[TurboTom]

Some more reasoning on the findings make me believe that my SDG6000X is sampling (when generating a noise signal) at 1.2GSa/s while @baldurn's has got to sample at a higher frequency, probably 2.4GSa/s, considering the observed sampling frequency (subharmonics / mixing products) feedtrough. A comparison of the hardware revision of the two instruments would be interesting. I reported in the other thread that my SDG6000X came without the four dedicated sampling memory chips installed. It relies on the 16 high-speed lanes form the Zync CPU module to the Kintex main signal processing FPGA to supply the sample data at a sufficiently fast rate.

It would be very interesting to know if @baldurn's AWG came with or without the dedicated sampling ram and if this may possibly mean that the instruments without it, sample at the lower rate for all signals. This could possibly also be an explanation for the substantial overshoot at steep (1ns) edges. If that's the case, Siglent must have either installed the dedicated sampling RAMs to approach these problematic findings or otherwise changed the architecture on recently produced AWGs. Moreover, it would mean that the instruments provided without the memory poulated, don't meet their specs (Sampling frequency). I won't right now pretend that that's the case, I'm just suspicious that we may be at something here that could possibly require further investigation...

[egonotto]

Hello,

my SDG6000Xalso dont show spikes.

Best regards
egonotto

[TurboTom]

Egonotto -- Thanks a lot for testing yours!

Yes -- seems like our AWGs perform similarly. Yours also shows the "dip" at 600MHz which should be located right at the Nyquist frequency. So yours will be sampling in "Noise Mode" at 1.2GHz as well.

Hint:

- If you use Trace -> Power Average on your SA, you'll get much cleaner visualization and it will permit to find peaks that may otherwise hide in the "lawn" of fluctuations...

- Especially when measuring noise signals, the RBW setting on the SA is very critical. It's not always "narrower is better" since the signal power is distributed over the whole spectrum, and the narrower the (SA's) filter, the less signal will be available for evaluation, i.e. when you reduce RBW, you will notice the noise floor is dropping (and this happens regardless of the noise source, exernal or from within the SA's frontend), while superimposed "frequency-discrete" signals will prevail. So when replicating such tests involving noise (or possibly other wide-band) signals, it's good practice to copy the instrument's settings as accurately as possible. 

[baldurn]

Will only comment to ScreenImg (4).png:

I have this closing PN/Jitter signal also from my none RT siglent SSA (reported on this forum).

Just loop back your 10Mhz reference from your SSA and test again.

You will see than to PN/Jitter from the internal synthesizer. An external 10MHz OXCO feed to the SSA will not help in any regards.
Just use the spectrogram plot to see more details. Also the Windows SW do not show 1Hz RBW as reported and waiting since a year for an update 

I just got my Mini GPSDO from Leo Bodnar and measured the 10 MHz signal from that. There appears to be about 4 Hz of phase noise at the bottom before the noise floor. Which could originate with either instrument - not too bad?

[hpw]

Will only comment to ScreenImg (4).png:

I have this closing PN/Jitter signal also from my none RT siglent SSA (reported on this forum).

Just loop back your 10Mhz reference from your SSA and test again.

You will see than to PN/Jitter from the internal synthesizer. An external 10MHz OXCO feed to the SSA will not help in any regards.
Just use the spectrogram plot to see more details. Also the Windows SW do not show 1Hz RBW as reported and waiting since a year for an update 

I just got my Mini GPSDO from Leo Bodnar and measured the 10 MHz signal from that. There appears to be about 4 Hz of phase noise at the bottom before the noise floor. Which could originate with either instrument - not too bad?

Well, you used 100 times average... how about several single shots... so you should get random side loop changes (Jitter)

Hp

[Performa01]

I've checked my (rather old by now) SDG6052X (HW 01-00-00-76-38) at three different noise levels and could not find any spurious signals from the generator.

See attached screenshots for FFT analysis up to 1 GHz at 1, 10 and 100 mV AC-RMS noise.

[baldurn]

Well, you used 100 times average... how about several single shots... so you should get random side loop changes (Jitter)

Ok here are a few random screenshots without averaging. This time I have both instruments using my GPSDO as 10 MHz reference and measuring a 100 MHz sine generated by the AGW. I do not know what to expect but if I have any jitter it appears to be -70 dB down and close to the noise floor. Otherwise it appears rock stable and does not move much at all. My setup is a bit sketchy due to lacking enough cables and converters so this is the best I can do for now.

If yours are different perhaps some improvements were made?

[hpw]

@baldurn

I did similar tests without AVG and equal results. Even using an external OXCO connected to 10MHz ref and input.

Conclusion it's the internal synthesizer (PLL or alike) who produces this. Even using spectrogram you will not see that much.

The 3D waterfall PC SW is limited to 3Hz BW... just waiting that one in China moves on..