Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

How does a spectrum analyzer measure Noise Spectral Density

  1. Mar 30, 2016 #1

    I'm a noob when is comes to these kinds of measurements on a spectrum analyzer.

    Take for example two LDOs:

    1. http://www.analog.com/media/en/technical-documentation/data-sheets/ADM7154.pdf
    2. http://www.ti.com/lit/ds/symlink/lp5907.pdf (see figure 17)

    They claim to be ultra low noise LDOs and they measure the noise on the output of the LDO by feeding the output into a spectrum analyzer.

    My questions are:

    1. How does a spectrum analyzer measure Noise Spectral Density. Does the analyzer put a bandpass around a particular frequency and measure the amplitude? How can it do so very well for a signal that is in the signle digit Hertz as well as MHz.
    2. Why are the units in V/sqrt(Hz).
    3. If I integrate the curve for the noise spectrum density over a particular frequency range do I get the total power of the noise signal?

  2. jcsd
  3. Mar 30, 2016 #2


    User Avatar

    Staff: Mentor

    I only skimmed the links, but you can certainly compare the unconnected (but terminated -- connect the probe tip to the probe ground) noise floor of the spectrum analyzer to what you measure probing the DC output of the LDOs. You would need to upload the data and do your own calculation to plot the spectral noise density, but that seems pretty straightforward.
  4. Mar 31, 2016 #3


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    Basically, Noise Power is the most important thing in most systems, when you are trying to measure the performance of a communications channel. Noise is a random process and truly random noise can take any value at a given time (∞ in theory) but you can rely on noise power, because it is averaged over a longer period than a noise spike. Assessing channel noise with an old analogue Spectrum Analyser was very dodgy because you could only look at the 'grass' at the bottom of the trace and come up with some approximate estimate of a mean value.
    A noise measuring algorithm in a Spectrum Analyser will look at the signal volts which come out of the i.f. filter and do an MS calculation with a bunch of samples of the filtered signal to give the Power admitted through the filter over a time. If the i.f. filter has Δf noise bandwidth then the noise power density will be Pn/Δf Watts per Hz. That can be stated in terms of Volts (Power = V2/R), rather than the Power as 'Watts per Hz' can be re-stated as 'Volts per √Hz' (square rooting the answer to give the RMS Voltage value). This isn't intuitive but it's a good system, once you get used to it. The resistance involved should be known if you want Watts from Volts and it would often be 50Ω in a comms system.
    The advantage of using Volts, rather than Watts is that the Impedance of the system doesn't matter and a power supply will have a low / unspecified impedance and the noise volts on the line will tend to be independent of the load current.
    If your analyser gives you the noise per root Hz then yes, you can scale up the total noise for any bandwidth. (Square root the frequency ratio, of course, if you want RMS volts). A few 'private' calculations on the back of a fag packet could help you get familiar with this process of hopping between Power and Volts.
  5. Mar 31, 2016 #4
    If you are interested in what goes on under the hood when we make noise measurements with a spectrum analyzer, you should read HP's classic app note (attached).

    Attached Files:

  6. Mar 31, 2016 #5


    User Avatar
    Gold Member

    I am not convinced that a linear envelope detector can measure the power of an arbitrary or noise-like waveform, irrespective of any processing applied after it. I am sure that a square law detector is required for power measurements in these circumstances.
  7. Mar 31, 2016 #6


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    That's fair comment when you are dealing with an arbitrary waveform. I think the value produced by an envelope detector can be manipulated to give a good estimate of RMS for a gaussian waveform in the same way (although not as accurately, possibly`) as it works for a sinusoid. It would be easy to invent a waveform that a simple envelope detector would be completely hopeless for. But, even then, if all that was being changed was the amplitude of the waveform and the designers knew the shape, you'd be back in business again.
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook

Have something to add?
Draft saved Draft deleted