I have a question regarding white noise generated by an arbitrary waveform generator (AWG). I'm working with one (AWG5014C) that has a sampling rate of 1.2 GSamples, with which I want to produce white noise that is relatively flat over some range, for example up to 300 MHz.

However, when I have it produce a white noise signal, it is not flat up to these kinds of frequencies at all, as shown in this figure enter image description here

Here I show the PSD as measured with a spectrum analyzer; a clear dropoff is visible starting at relatively low frequencies compared to the sampling rate (or maybe not?) and I am wondering what the cause of this is.

I should note that I am not so much looking for a 'fix' for this; I assume that this is just a consequence of how the machine operates. My question is of what it is a consequence, so that I can understand it better. One clue that I have is that I can for example use the spectrum analyzer to extract a filter that will offset the dropoff; treating the AWG as some sort of black box that applies a filter |H(omega)| to my time series, applying |H(omega)| gives me a flat spectrum.

As a short aside, you can ignore the first points that are way off. Although I don't understand where they come from (another question?), they seem to be artifacts of the spectrum analyzer.

  • \$\begingroup\$ I would think that the sin(x)/x attenuation which is done by the generator and bandwidth limitations of the generator can cause this. Also: what is your spectrum analyzer, what are its settings ? The strange points near 0 Hz: Spectrum analyzers always have issues near 0 Hz so I would just ignore these points. \$\endgroup\$ – Bimpelrekkie Jun 28 '16 at 14:30
  • \$\begingroup\$ If you could go into a bit more detail on why that type of attenuation would occur, then that would be amazing, as it does not seem very intuitive to me. But it definitely sounds plausible! As for the spectrum analyzer, it is a E4407B Spectrum Analyzer, span is 300 MHz, CF is 300 MHz, Avg = 100, RBW = VBW = 1 MHz. \$\endgroup\$ – user129412 Jun 28 '16 at 15:23
  • \$\begingroup\$ That sin(x)/x is for preventing aliasing of the signal, since it's a sampled signal measures have to be taken to prevent "folding" of the signal around the sample frequency. It like a sort of windowing function. Have you tried increasing the RBW and VBW to 10 or 30 MHz ? The power level is quite low, are you using the generator's attenuator or an external one ? I would try to generate a 1 V noise signal and check that that is flat. If not, sinewaves at 100, 200, ...600 MHz and see the response. If that response is not flat then there you go. \$\endgroup\$ – Bimpelrekkie Jun 28 '16 at 15:38
  • \$\begingroup\$ Okay, interesting. Kind of suboptimal if I want to produce a flat band. As for the RBW, I haven't tried that yet, I can do that next. I'm not attenuating the signal externally, I'm just setting the variance of the gaussian noise that is being produced, which is 0.15V in this case. But generating a 1V noise signal is a problem, as that will exceed the maximum V_pp of the AWG. \$\endgroup\$ – user129412 Jun 28 '16 at 17:37
  • \$\begingroup\$ maximintegrated.com/en/app-notes/index.mvp/id/3853 seems to have a lot of information on what you mentioned (which I found thanks to your comment) \$\endgroup\$ – user129412 Jun 28 '16 at 17:43

This is a sinx/x roll off. You can compensate

  • mathematically in the waveform files
  • Some AWG have List mode". You can perform amplitude compensation by writing into the AWG List entries, amplitude fields, first by determining if you are sending a signal in the band edge.
  • Some commercial vendor comes with automated software to perform frequency flatness calibration (which they write the compensation table as per item 2).

enter image description here

Example: Keysight M8195A sampling rate at 65 GSa/s and its compensation for sinx/x roll off.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.