Post Go back to editing

6/1/2016: Noise analysis in precision Analog Designs

Hello Scott Hunt,

I listened to your web cast the other day with interest. I have a question which I only thought of afterwards!

How exactly do you make accurate noise measurements? This has been a changing problem for me over the last 45 years. As time goes on, the technology available has changed. In 1971 I used Vpp/7 on an analog scope.

Now days, I use a low noise pre-amplifier with a gain of +40dB or +60dB with a 2 pole LC 1MHz lowpass filter to raise the signal level above that of my scope's noise floor.  The output is AC coupled into an Agilent DSO which can measure RMS. I turn off averaging and rotate the timebase control to get a maximum noise reading. How do you do it?

Thanks for the information about the equivalent bandwidth factor for various filters. This is in many noise documents that I have read. They failed to point out that if you use it as a noise correction factor for the filter bandwidth, then you have to square root it. This is something that I misunderstood.

Another error that I have made is to use a X1 scope probe to make noise measurements. These probes can have a 400 Ohm end to end resistance, which can contribute 2.6nV/rootHz .

Best regards,

Tim Orr

  • Hello Tim,

    Thanks for listening to the webcast. I'm glad it cleared up some confusion for you about the noise equivalent bandwidth.

    Noise measurement is a big topic and there are many ways to go about it, the important thing is that your measured noise has to dominate over the instrument noise. As you noted, if you're using an oscilloscope for this, you may need quite a bit of gain to be sure that the noise you are measuring will overcome the noise floor of that scope (even if you can use the BW limit on the scope). In the best case, you just have a very low noise digitizer and you can eliminate the gain stage or at least reduce the gain needed.

    In general, we try to find something much lower noise than the device under test (DUT), for example, the AD8429 has been a common choice for a low noise preamp around here (we often need the differential inputs, but if not, you can always tie one input to ground). If the DUT is an amplifier, for example, we might consider increasing the DUT gain to be sure that its noise is dominant. You may also add filtering after this depending what you're trying to measure. Then you measure the noise floor of your scope or acquisition instrument for comparison purposes. Now you can take enough gain with your low noise amplifier so that your DUT noise dominates (10x the instrument noise is a good goal just to be certain, and it might require multiple gain stages to get the required bandwidth), and you acquire the noise waveform just as you described. After acquisition, you can do whatever post-processing is necessary to arrive at RMS or NSD.

    As a small aside: I might disagree with you on the 1X probes vs 10X probes. Attenuation is generally bad for noise purposes, since it reduces the noise you're trying to measure with respect to the noise of the next stage. If you're using a 10X attenuating probe, you conceptually have to multiply the scope noise floor by 10 to refer it to the input signal. Of course, the scope does all of this scaling for you. With the bandwidth limit and no averaging, my scope has a noise floor of ~300µVrms in the smallest range with the 1X probe and ~2mVrms with the 10x probe. Usually if we choose the 10X probe over the 1X probe, we do it to reduce capacitive loading on the circuit.

    Sorry for the long answer. I hope it helps.

    Best regards,