Measuring noise

HI there,
I am thoroughly enjoying using my QA401 - great hardware!

I would like to know what are the best parameters for measuring/quantifying amplifier noise? I prefer unweighted measurements and use a 20-20 000Hz bandwidth - I am presuming that because bandwidth is digitally derived this 20-20kHz is genuine noise bandwidth and i do not have to take into account any filter slopes used to define the bandwidth like in an analogue analyser.

Should I use averaging? I am not looking for repetitive phenomena just noise quantification, so it would seem averaging should be off…or does averaging simply affect the display of the noise and not its power/amplitude?

Should I use any windowing? It seems to me that any window (except rectangular of course) will bias the measurement.

What does the N+D Level actually measure?

On a separate note was reading your notch filter post and wondered would it be possible to export FR measurements in the same way so as to use them as self-correcting weighting files? My QA401 is 0.2dB at 20kHz and I would like to get it flat to 50kHz for measuring phono preamps.

That is correct: To measure in 20 to 20K bandwidth the measurement is done in the frequency domain, and the only points used are those within the region you specify.

If you specify weighting, then the weighting is applied to the frequency domain AND THEN the noise measurement is made between the indicated bandwidth points you’ve specified.

If you noise is just noise (and not some signal near the noise floor), then averaging won’t change it much. You can test this by setting averaging to 0, and looking at the noise measurements that are reported over 5 or 10 cycles. And then bump up averaging–it won’t change much.

If it’s noise, the windowing shouldn’t matter. Windowing matters when you are dealing with a signal that might be discontinuous at the window edges.

The N+D (noise + distortion) measures the level of the noise and the harmonics BUT NOT the fundamental. For example, if you connect the L+ output to the L+ input, and short the L- input, and turn the generators off, you will measure a noise level (“RMS L”)around -109.5 dBV or so. This is the noise level of the generator (which is about 6 dB higher than the noise level of the analyzer input). The N+D will be the same because there isn’t a fundamental present.

Then, turn on the generator at -10 dBV. There you will see the N+D increase about 0.5 dB. That is because the N+D measurement has detected a fundamental (there should be a green “F” at the bottom of the display as shown below) and the N+D has thus been degraded about half a dB by the presence of the visible 2H and 3H.

Not currently. You could create a compensation file by hand and specify points at 10, 20, 30, 40 kHz, etc and get pretty close. But I agree it would be nice to be able to export a correction file.

Here are the noise measurements with 1 nV/sqrtHz 80dB preamp with 460 Ohm resistor on the input. The total noise density is 3nV/sqrtHz. It is my understanding that equivalent noise bandwidth correction is not implemented in the software And I should add it to the input gain. Is it correct?

My second question is about different reading on low frequency when rectangle window is selected.

Thank you.

Hi @dimitri, what value have you specified for external input gain? Is that 80 dB? If you have specified the 80 dB, then you value you read can be treated as the noise in 1 Hz (because your Y axis is dBV/rthz). So, we can see roughly that your input noise density at 1 kHz is around -168 dBV = 4 nV, which is pretty close to the 2.7nV expected for that value resistor.

When you change windows, as you see you will sometimes reduce the width of the peak. Each window has a different bandwidth and some are more suited for certain tasks than others. If you aren’t too concerned with ripple and measuring the flatness of a DUT, then Hann is a pretty good choice for just about everything. Around page 35 in the manual you’ll see more discussion.

Rectangular should generally be used only when you have a perfect integer number of cycles. Otherwise, when you get to the discontinuity, you get noise added to the plot. Since you still have tones in your plot, my guess is that there is a break between the phase at the front of the sample buffer and end, and that is resulting in the low frequency noise increase you see. But if there were no tones at all, then the window type shouldn’t matter.

Thank you,

Rectangular should generally be used only when you have a perfect integer number of cycles.

How can I alter the number of cycles in QA401 SW? Bin width is (Fs/2)/(FFT/2), where FFT=64k and Fs=192k

Well, generally you generally can’t. FFT must be a power of 2, and sample rates are limited. Even if you had infinite flexibility, you have enough unwanted higher-amplitude tones where you’d still have a discontinuity and thus noise.

Remember, too, when in dBV/rthz mode, the peaks of tones don’t mean much anymore. Just as visible noise level doesn’t mean much in dBV mode (because the noise level changes with FFT size), the peak levels don’t mean much in noise density mode.

I think you can generally use your eyes to do some filtering here. It’s pretty clear what the noise is at 1k, 10k etc. You can see the 1/f noise too and get a reading at 100 and 10 Hz. So, if you have unwanted tones in the measurements, use Hann to keep the low-freq noise to a minimum and visually extract the noise density measurements at 10, 100, 1K, etc.

Finally, remember the current noise of the QA401. It uses OPA1612 in follower on the inputs. For resistors below 1k (as you’ve done), the noise is dominated by the resistor. But for noise values above 1k or 2k, the current noise of the OPA1612 will start to contribute more than the source.

Here’s some more detail on measuring noise and the importance of resistor value.