Phase Test

Well I have to much time on my hands so I did this little experiment …
I created a basic loop ( couple of tracks ) and rendered it to selection .
Normalized the volume and resampled this new loop again ( render selection to sample )
Then resampled this loop/normalized etc …
I did this like 10 times …so theoratically the last rendered should be the same as the first one right ?
Inserted the first rendered loop in a track and the last one , reversed the phase of the last one , rendered both to selection =silence ( as it should be )
Now when I normalized the silent sample ( both phases canceled each other out ) I get static and crackles …?
Whereas normalizing complete silence should be complete silence :blink:

Before we figure out anything else, you’re gonna have to go into a lot more detail about exactly what you’re doing here :)

There could be several things responsible for creating tiny differences in each render, caused by slightly different random values used in effects, dithering, etc. Give some more details about how you have set up and created your loop, and then we can try to eliminate some of the variables.

ok …step by …
sample rate 44.1 Khz 16 bit
Take a random loop from your hard disk or create one .
1Insert this loop in pattern editor
2.render this loop to selection
3.Maximize the volume of the rendered loop in the sample editor ( it’s now 32 bit )
4.Erase the previous loop from pattern and insert the newly frshly rendered loop in the pattern editor
5.Render this loop to selection , maximize it’s volume in sample editor
6.Delete previous loop from pattern , insert newly rendered loop and repeat steps …8 times ( or whatever)
Now …
7.Take the first rendered loop and the last rendered loop , put them each in a separate track and inverse the phase on of the loops …
I used antares filter for this ,( have also tried with generic gain effect )
8.Select both tracks and render to selection
9.result = Silencio …Good…
10.maximize the silence in the sample editor = crackles/poples …
Normally silence multiplied by factor x ( normalizing) should be silence

No effects , a clean renoise template , meters set to 0dB.
The basic loop is of no importance , any audio file will do , cause eventually it will be rendered to 32 bit file as a starting point and from there on we just render a new file .

Each time you render selection you are succumbing to the -6dB that Renoise has built into it inherently. This means you are effectively not using one of the bits, so only really 15bit sample (as far as noise floor etc.) Also it’s not exactly 6dB to a bit (think it’s 6.02dB or something) so there will be extra quantisation noise when you normalise it up to 0dB.

Personally I don’t think it’s a surprise at all that you are finding some noise normalising from the phase cancellation of this.

EDIT: Although I would be interested in how different the results are if you put a Gainer at 6dB boast in the chain.

i would be interested in what happens if you do this in any other DAW.
as long as there is no noise in my music everything is fine :rolleyes:

Also think about dithering. One you dither a sample that is all zeros you are going to get a resultant sample which is a combination of 0s, 1s and -1s. Of course when you normalise this you are going to get nasty noise. It is normal and to be expected.

The reason why I tested this ( mind you I am not a specialist in this field , just merelly interested )is how many times we can rerender a sample/ rebouncing ( taking a clone of a clone of a clone of aclone etc…)before degradation starts …
Normally it should never happen as it’s all taking part in the digital domain …that’s why I was surprised that in the end the normalized phase cancelled sample ( which was total silence before normalisation ) actually contained data ( a mixture of the original soundfile and noise/pops/crackles )

ahh ok
There is no diference when I put a gainer on the final rendered sample …hell I put like 30 gainers( seriously) on the final result it’s still remains silent …it’s just after normalization the crackles comes out
Picture below is the normalized phased out sample

Try with Dithering turned off and see if it makes any difference. With a Gainer at +6dB and dithering off I would of thought you would get flattest result.

The problem here is simply due to the fact that gain adjustments are being applied at various points, and this is of course a destructive process that will absolutely result in changes to the data itself. As kazakore has pointed out, Renoise is applying a -6dB gain adjustment during rendering which is changing the sample data. Then you are normalising that result back to 0dB which is changing the data once more. Every time you do this you are degrading the original sound by a very, very, very tiny amount, so there will be very tiny differences in the new render.

These tiny differences are typically caused by rounding errors that occur when performing calculations on floating point numbers, and they are honestly completely expected in the world of digital audio. A number can only be represented digitally with a limited amount of precision, so every time calculations are performed on values which are not perfect integer/whole values, then some tiny amount of precision will be lost somewhere in the process.

When you finally compare the original sound to the Nth rendered version by using the phase cancellation technique, the noise that you are experiencing is the result of these types of rounding errors. In reality they are almost completely insignificant, and will certainly never be heard or noticed by anyone that is simply listening to the sound itself. But when you normalise those errors back to 0dB, they suddenly seem incredibly bad and look ugly as hell. You’re wondering where the hell this hideous square/pulse wave noise came from, but the truth is that the noise is actually some ridiculously tiny value like 0.000000000000000000000000000001, and is nothing at all to worry about.

Bottom line: you can quite safely ignore this phenomenon. It’s just one of the many strange things that computers do, haha.

.

Thanks dBlue for putting it more clearly than I ever could. All I could really say is look into quantisation (errors) and dithering and you will see that this is totally expected and nothing to worry about.

Thanks everybody for the replies , I know those differences are inaudible and neglectable ,I was just wondering where those noise came from …and your explanation shed some light on this .
Thanks :)

Just to really put this into perspective… you could probably perform these kind of edits millions of times and it still wouldn’t be audible in the final result.

Some interesting conclusions about the so called 'transperancy /neutrality ’ that some vst eff.cts .claim to have .
Before rendering a sample just insert a vst effect one 1 track that has the possibilty to turn all parameters to zero/neutral state value , but the main output to 0.0dB
When phases are inverted on both audiofiles ( then rendered ) and there is no silence , means that the vst effect in the signal chain has coloured the sound file .
And there are a lot …antares vst ( flat filter mode , )fabfilter …and a bunch of others …I guess that’s why we have a bypass mode :D
I even managed to crash renoise several times .Verry hard to go into detail cause probably no one will ever come into this situation

Great research here gentleclockdivider! Interesting results. I might do some tests myself to see if there’s a ‘feel’ difference as well. After all, I was finding that -14dB on the master gave a different poorer feeling compared to 0dB - I wonder if I can verify this fuzzy feeling in anyway.