What is Aliasing & Why we should talk about it?

[El°HYM]

The Problem
https://youtu.be/2k_mX7BJ3_Q

The Solution
https://youtu.be/JVa2FkBP9wA

Both videos are very good in explaining what the often referred 'bad' digital - sound actually is; its harshness in the Area our Ears are most sensible to, about 2KHz to 4Khz (Fletcher Munson Curve).


So in conclusion there is no such thing as bad 'sounding' digital - audio, its basically Aliasing from plugins adding up and getting too prominent in our most sensitive hearing - range. Thus causing ear - fatigue and an overall unpleasing 'harsh' digital - sound.

An interesting read about all of this (for those who have time) is over at GS.

Technical background:

Simulation of analog sound like saturation from transformers, compressors, tape, EQs,... makes the creation of harmonics necessary.

Hardware:
When analog gear is recorded, it doesn't matter if that gear creates harmonic content exceeding Nyquist frequency, because all A/D-converters use steep lopass filters and get rid of any frequencies above Nyquist frequency (sampling frequency/2) and therefore prevent aliasing artifacts.
48 kHz sampling f = 24 kHz Nyquist = highest detectable frequency


ITB:
If harmonics are created by plugins, frequencies within the plugin are created which easily could exceed the current Nyquist frequency of the project.

For example smashing hihats or cymbals (which have significant amplitudes in the 10 kHz range) with a compressor and/or tape saturation, means - for example - the 3rd harmonics of the 10 kHz already appear @30 kHz.

That means, if the DAW runs @44 kHz (Nyquist = 22 kHz) and the plugin developer does not care abut that problem, that 3rd harmonics created in the plugin will appear below Nyquist in the audible treble range at 14 kHz.

example 4th harmonics of 10 kHz = 40 kHz -> that inaudible 4th harmonics of 10k reappears at 4 kHz in the audible range...


So an aliased 14 kHz / 4 kHz signal is added to the cymbals track (ofcourse not only these two frequencies, because the cymbals are not only @10 kHz, but all harmonic multiples of the frequencies in the original signal above Nyquist are folded down into the audible range).

In contrast any recorded analog gear would be free of these aliasing "signals".



What makes things worse: the aliased, downfolded, unwanted frequencies add unwanted signal energy to the frequency range they appear in. They add "treble" energy by adding garbage. They increase signal energy in the mids, but its garbage.



If a SNR of 65 dB is assumed as technically acceptable minimum between the signal and any unwanted "noise" (aliasing frequencies technically are not noise, but are unwanted garbage just like noise), the anti-alias filtering taking place in the plugin should keep any harmonics below that level.


It should also be kept in mind, that usually lots of additional processing is taking place with the signal after harmonic content has been added by a plugin: once a "analog" plugin has injected audible aliasing garbage into the signal, its not possible to get rid of it ever again.

https://gearspace.com/board/music-compu ... d2c10554dd

Furthermore there are a few other Techniques to add (next to the more classical approaches from above) like ultrasonic - filtering, using a Wrapper with OS or bumping your Project's Samplerate up, even exporting at higher SR's and then downsampling can work.

The most easiest and effective Way yet, might still be to just use Plugins with OS including internally oversampling the Delta - Path or other nice Geekery.

[ampetrosillo]

Aliasing is such an overblown concern. Where you have aliasing due to harmonics generated by distortion, compression etc. bouncing off the Nyquist frequency, you will also have intermodulation distortion that is just as enharmonic. That intermodulation distortion is indistinguishable from aliasing essentially, and intermodulation distortion is what gives distortion its character ultimately. Nobody is running sines and distorting them, you are distorting real life signals with lots of frequency components all over the place, many of which enharmonic to the fundamental (if present at all!).

Of course, less aliasing may be a sound concept, because it's still added distortion that might be just a bit too much, but aliasing mostly is significant in the upper treble (because the energy of real-life signals in the treble is low, therefore it takes a while for aliasing to manifest significantly in the mid and high-mid region). How often do you have purely high mid and treble signals saturating at more than 3% THD or so, so that their upper harmonics are strong enough to be over -60dB relative to the original signal below 10kHz or so? (A hint: if you fear this is the case, run a high shelf pair, a kind of filter that is very light-headed on phase, before and after the non-linearity, cutting then boosting the treble by the same amount, like 20dB or so, or just distort the treble cut and mix it with the original, or with the original with a bass cut).

This doesn't mean that you shouldn't oversample, but we've used non-OS plugins for decades with lots of satisfaction. Decapitator is the most famous example, but there are countless others, synths included (which is where aliasing manifests itself the clearest, due to the comparatively simple signals synths generate, especially on glides and pitch modulation). If a plugin can't oversample, and you are dead set on not getting any aliasing, work at 88.1k and upwards, encapsulate it in a plugin manager (such as Metaplugin) or use Reaper which can oversample plugins externally.

[plexuss]

^ what he said. And... "saturation" aka distortion: This to me is the largest contributor to bad digital sound. Producers adding in saturation to the point where its just audio haze caking the dynamics and transients and muddying up the response. It amazes me when people say they put such and such a saturator on a track or mix and it was "magic". It was sh*t more like it. Its another area, like loudness, where your brain leds you to THINK it sounds better when in reality it doesn't (eg. when you listen outside of the context of the engineer). I basically can't hear aliasing unless its the raw single track. In a mix, I never hear it. But my mixed are dense so that may be why.

[crickey13]

Negative effects of linear filtering, be it linear EQing or linear OS filtering, can be way worse, imparting that awful digital harshness more readily IMHO. Add several plugins using linear OS filters and just listen to how much crap congests in the 2-4 kHz range.

[briefcasemanx]

I just export mixes at the highest sample rate and don't worry about it at all.

[fese]

The problems with all those “bad” aliasing plugins is that they are not old enough. The digital samplers and synths from 30-40 years ago though are vintage and the aliasing gives them their special mojo! Just wait another twenty years and we will get emulations of the original Decapitator plugin, because that sounded “so much crunchier” than the new 16xoversampled follow up…

[Aiynzahev]

The problems with all those “bad” aliasing plugins is that they are not old enough. The digital samplers and synths from 30-40 years ago though are vintage and the aliasing gives them their special mojo! Just wait another twenty years and we will get emulations of the original Decapitator plugin, because that sounded “so much crunchier” than the new 16xoversampled follow up…

Surely some of the waves plugins are now old enough to give us that vintage flavor?