I am trying to measure the noise of a photo detector at different frequencies, using an RF spectrum analyzer. The measurement range was spanning from 10 Hz to 20 MHz. To cover not only the higher frequencies with a sufficient amount of points, but also the lower range, I repeated the measurement several times while varying the span (10 Hz to 20 MHz, 1 MHz, 100 kHz and 10 kHz), but keeping the amount of points constant. Initially, the instrument set the sweep time automatically, resulting in the following graph: enter image description here Here the transitions are clearly visible at places where the noise level decreases.

To avoid such rapid transitions, I tried to keep the sweep time constant. This resulted in enter image description here Here for all measurements (except the one going up to 20 MHz) the sweep time was fixed, but I still see transitions between the measurements.

Where do those transitions come from, and how can I avoid them in future measurements?

  • \$\begingroup\$ what "transitions" are you referring too? all I see is a graph that looks pretty plausible to me. \$\endgroup\$ Oct 7, 2020 at 10:09
  • 1
    \$\begingroup\$ @MarcusMüller: I was referring to the sudden drop of noise at (for the first image) 1 kHz, 10 kHz, 100 kHz and 1 MHz. \$\endgroup\$
    – arc_lupus
    Oct 7, 2020 at 10:11
  • \$\begingroup\$ Which sudden drops? Sorry, I'm really not following you. I don't see any sudden drops! \$\endgroup\$ Oct 7, 2020 at 10:12
  • \$\begingroup\$ When I look at the noise level to the left of 10 kHz and to the right, the right side is significantly clearer, as far as I can see. I would define that as a "drop". \$\endgroup\$
    – arc_lupus
    Oct 7, 2020 at 10:13
  • 1
    \$\begingroup\$ no, that's not a "drop"; a drop would be if the value dropped. That's a reduction in shown variance, which probably just stems from your instrument observing a larger bandwidth at once at higher frequencies, which makes sense. Don't forget: you're observing noise, that is a random entity. You can always have either low variance of your noise power estimate, or high spectral resolution, never both, based on the same data. At higher frequencies, you get less spectral resolution, but lower variance. Neat! That's how you know the math works. \$\endgroup\$ Oct 7, 2020 at 10:15

1 Answer 1


The spectrum analyser is making choices for various things like resolution bandwidth, video bandwidth, number of averages, according to an algorithm decided by the people who specified the instrument firmware. This was to optimise their idea of the SNR/time tradeoff in each offset span.

You need to force the instrument to make your choice of those things if you want visual continuity between the spans. You may be able to do this, it depends whether the firmware gives you that option.

Notice I talk about 'visual continuity' in the previous paragraph. This to point out that this is all you would be doing, prettifying the display. If you want to understand the measurements of the diode under test, then you've a lot more work to do to determine how much noise is from the analyser, how much from the diode, and what the spread of noise on the analyser display is really telling you.

In a way, the display transitions have helped you, by forcing on you that the display needs interpretting.


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.