REW Beta Release Feature request: RTA with files, adding raw double (plain array of 64bit floating point, *.dbl) input format

KSTR

New Member
Thread Starter
Joined
Jun 7, 2019
Messages
9
If it makes any sense, that is, the input stream to the analyzer is 64bit floats internally, would it be possible to add the raw double file format, as known by other software like Adobe Audition and AudioVero Acourate? This is also the file fomat I use with my own software, both for generators and a block averaging tool... because it is so simple to handle.
That would require a third field "sample rate" in the input dialog, which should allow arbitrary values as is it just a scale factor for frequency display anyway. Also the dBFS display range should extend to lower than -237dB where the display values are cropped currently. A workaround is to apply gain to the file prior to loading but then I would run into the +20dB top limit.
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
1,619
The input sample stream uses 32-bit floats, though subsequent FFT processing is 64-bit if that option is selected in the RTA graph controls.

Where does the requirement for a floor below -240 dB arise in your processing? How far down do you need the floor to be?
 

KSTR

New Member
Thread Starter
Joined
Jun 7, 2019
Messages
9
I'm using an RME Adi-2 Pro and with extremely heavy time-domain block averaging I can get the noise down by 40...50dB to about -160...-170dBFS (~28bits). 1M FFT has noise per bin lowered by ~60dB. This would amount to a total depth of ~230dB which I thought that might be borderline.

Actually, after having played with the file processing option for the RTA (did't know it was that simple, drag&drop) I have to admit that the resolution when fed with 32bit floats should be sufficient. When using a "silence file" with 1bit of dithering to 31bits, saved as 32bit float using Audition, I get a FFT baseline noise of < -230dB which is sure good enough and close to what I calculated. I see you choose the -237 dB limit for good reason.

Using my best effort to create a self-dithering 32bit float sine on an exact bin frequency I get a 200dB range which again seems "good enough" using 64bit FFT, with 32bit bins raise to -160dBFS max.
fft.gif


I will officially withdraw my request for 64-bit input stream and import, not required.
REW actually takes 64bit float wavs input as I found out and I can convert to that from .dbl using SoX so the import thing also isn't a real concern.

And of course, a very big Thank You for making this extremely excellent software available and your continuous effort to improve it (like the Noise Density display option that was proposed by a close friend of mine and which you quickly added) :T
 
Top Bottom