So I have been using my Picoscope and taking FFTs to help troubleshoot my ground loop problems. It has helped tremendously in understanding the root causes. one thing I have been wondering though is should I be applying a weighting curve on my measurements. I realize it's not necessary to get meaningful results but I have noticed that when I get noise reduced way down it still measures 20 db higher than I see against commercial equipment and higher than it sounds. Should I be applying A weighting? Don'[t the commercial designs use A weighting to make things measure better? I haven't used Rewind but does it allow you to apply curves to FFT measurements? Notice that Audiotools on my phone lets you apply A or C to the FFT.
Also to noise levels integrated over the whole audio band, but not to FFT plots and usually not to hum measurements.
If you are using spectrum analysis, you don't want to affect levels. You want to see real levels.
If you are creating advertising copy - yeah. Apply curves all you want. That data is invalid. At that point, export the real data into excel (or whatever) and run math functions to properly assess what you want to see. Your weighting curve must be correct.
If you are creating advertising copy - yeah. Apply curves all you want. That data is invalid. At that point, export the real data into excel (or whatever) and run math functions to properly assess what you want to see. Your weighting curve must be correct.
I have used FFT's in my oscilloscope software design and around DC can often read higher than it really is.
I cant answer why.
I did have a struggle finding a bug free FFT algorithm and some were terribly noisy.
I remember using one from Microchip application note in the 1990's for a RPM counter for a car.
That was very poor.
I had to debug it myself and found they hadnt handled the maths divide or multiply signs correctly.
Once that was fixed the display was much better.
I cant answer why.
I did have a struggle finding a bug free FFT algorithm and some were terribly noisy.
I remember using one from Microchip application note in the 1990's for a RPM counter for a car.
That was very poor.
I had to debug it myself and found they hadnt handled the maths divide or multiply signs correctly.
Once that was fixed the display was much better.
Do you mean that the value at 0 Hz doesn't match the time average, or that you see too high values in bins adjacent to 0 Hz? Some problem with windowing or some mix-up between RMS and peak values maybe?