• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

A better way to graph Noise? FFT Gain ,(dB)V/√Hz, 1/3 Octave Band Noise

Lambda

Major Contributor
Joined
Mar 22, 2020
Messages
1,809
Likes
1,560
I noticed a lot of people comparing the noise floor in FFT graphs without considering FFT size/gain and scalloping loss.
There must be a better way do it?!

Manufacture datasheets seem to uses Voltage Noise Density:
To me it looks like this is the better way of specifying Noise because the its free from the uncertainty's introduced by FFT size/gain and and scalloping loss.
Screenshot_2021-02-24_22-05-21.png



Alternatively there is this 1/3 Octave Band method:
1614201875834.png

lewitt seems to use this
Screenshot_2021-02-24_22-27-51.png


But its also kind of confusing and i don't fully understand what Lewitt is doing, (actually it looks wrong to me)
is it clever marketing or actually the proper way to do it?
1. there graph is way higher resolution then 1/3 octave.
2. from beeping used to FFTs i would associate a flat line with White noise but in this 1/3 octave view it would correspond to Pink noise?
3. do other manufactures do this?

Is there a good way to show semi or quasi peak values in a noise (density) graph?
A FFT or any method with long averaging would be insufficient so show intermittent noise like clicks and pops every view seconds.
but with some sort of a peak hold it could be visualized?
 
I think the best way to show the real noise floor is to calculate the FFT gain and add a second (very smoothed) curve to the spectrum which is by the amount of the FFT gain shifted up above the visual noise floor.
 
I noticed a lot of people comparing the noise floor in FFT graphs without considering FFT size/gain and scalloping loss.
There must be a better way do it?!

Manufacture datasheets seem to uses Voltage Noise Density:

1. there graph is way higher resolution then 1/3 octave.
2. from beeping used to FFTs i would associate a flat line with White noise but in this 1/3 octave view it would correspond to Pink noise?
3. do other manufactures do this?

1. This plot is smoothed in an "editorial" fashion, by which I mean they take data and fit it to a theoretical line. Nothing wrong with this it is common practice. I prefer nV/rt-Hz with no ambiguity, i.e. a 1K resistor is 4.07nV/rt-Hz at 300K.

2.,3. B&K does this with their microphone measurements, I find it a pain.
 
1. This plot is smoothed in an "editorial" fashion, by which I mean they take data and fit it to a theoretical line. Nothing wrong with this it is common practice.
Thanks!
Just so i understand it better.
1/3 octave is in this case a kind of a kind of arbitrary choice? they could instead have used full octave bands.
But the same graph with full octave bands would be shifted up words?
 
Thanks!
Just so i understand it better.
1/3 octave is in this case a kind of a kind of arbitrary choice? they could instead have used full octave bands.
But the same graph with full octave bands would be shifted up words?

1/3 octave is common industry practice. The thirty band RTA is a nice compromise between resolution and readability (IMO).
 
1/3 octave is common industry practice. The thirty band RTA is a nice compromise between resolution and readability (IMO).
I know its a common industry standards and this is maybe why the chose it.

The question is, is my assumption that "lower the resolution shifts the curve up" correct?
Lower resolution means more bandwidth per bin therefore more noise power per bin?
 
Right the more Hz you average the larger the value.
 
@scott wurcer Thanks for the conformation of my theory.
So this method has the same problems as raw FFT, without knowing the resolution its not comparable.

Lewitt marketing with there graph that there green line is below the white line of the audible threshold.
But this raises my question whats the valid resolution to make this comparison?
Can't you make every noises floor look like its below audible threshold by just incensing the measurement resolution?


Noise Density in V/√Hz dose not seem to have this problem?
i assume dBV/√Hz and dB/√Hz would also be a valid way of expressing the actual noise power independent from resolution.
 
dB needs a reference such as dBV where the reference is usually assumed to be 1V rms.
I know but in case a DAC is tested the reference 0dB can be the max output (0dBFS)
The absolute voltage noise of a DAC is not as imprtend as the signal to noise for me.
For a AMP the dBV noise is more impotent.
 
Back
Top Bottom