Noise added when audio is exported

Newbie here

I have been using Audacity for a year now, unfortunately when working on a 16 bit wav file and doing a vertical zoom on a section that I had used a noise reduction on, I noticed noise. I decided to create a 20 sec stereo track and generate silence and export it under various formats, both lossless and lossy, and found out the following when importing back into Audacity and vertically zooming in. My guess is possibly that the last 4 bits of the bit depth, (the bottom 24 dB) have been scrambled and producing the noise, e.g. for 16 bit file, 96 dB has been reduce to 72 dB.
16 dB.png
24dB Plot Spectrum.png
16 bit wav files show a peak vol. of about -72 dB with a noise floor of -82 dB
24 bit wav shows a peak of about -120 dB
32 bit wav does not show any noise
16dB Plot Spectrum.png
24 dB.png
The Plot Spectrum appears not to be in line with the peak and noise floor values, seems to be off by 24 dB, maybe someone can comment on why this is? and why does the 1st Plot Spectrum show silence at 4,000 Hz

This is the same with all the formats I have tried. Ogg and MP3 do not suffer from this as I guess with Mp3’s they are encoded with Lame.

Maybe someone can show some light on this? Is FFmpeg at fault? Or is Audacity itself creating the problem?

Any idea’s on a workaround?

Bizarrely, that noise (called “dither noise”) actually increases the dynamic range to a little over 100 dB. Information about dither here: Dither - Audacity Manual

Thanks Steve

The link was great. I check the settings, realised the “High Quality Conversion Dither” had been set to “Shaped”, changed that to “None”, and it worked.

I forgot to mention in the original posting, that I had tried it on another computer with the same results, so I guess it installed with “Shaped” as default. Maybe users want “Shaped” as default.