Professional Analog Mastering.
Professional Analog Mastering.
226K+ subscribers
Aliasing is created when the signal goes above the maximum supported frequency.
For example, if the sampling rate is 44.1 kHz, the maximum supported frequency is 22,050Hz.
Anything that goes above 22,050Hz will be reflected down the spectrum in equal measure.
So, if a frequency occupies 25000Hz, meaning it’s over the max supported frequency by 2950Hz, then it will be reflected down the spectrum by 2950Hz from the highest supported frequency.
22050 minus 2950 equals 19100Hz. This is where that signal will end up - back in the audible range.
To show this in real-time, I’ll use a sine wave generator before a sample rate reduction plugin.
First, I’ll generate a high-frequency sine wave.
As I reduce the sample rate, in turn reducing the highest supported frequency, notice how the signal is reflected down the spectrum.
The more I reduce the sampling rate, the lower the frequency of the reflected signal.
But you’re probably not using a sine wave generator and sample rate reduction in your project,
In short, if you use a saturation plugin, compressor, distortion plugin, or anything that generates harmonics, it’s likely that the signal will go over the max supported frequency.
Going back to the sine wave example real quick, look what happens when I saturate a high frequency signal in a 44.1 kHz session.
Even without reducing the sampling rate, harmonics that form above the max frequency or 22050Hz are reflected down the spectrum.
As you can imagine, this occurs with just about any signal not just sine waves.
If I saturate 7500Hz, and a 3rd order harmonic is formed, that would create a signal at 22500Hz, or 7500 times 3. Meaning, that harmonic would be reflected and end up at 21600Hz. That’s not a big issue, but if it’s a 5th-order harmonic or maybe 7th, it’ll go over the max frequency by a greater amount, resulting in a reflection that ends up in a more audible range.
You might have noticed that I’ve used the term harmonic when talking about multiples of a saturated frequency, but when I’m talking about aliasing, I’ve said something like a reflection, or the signal, or something along those lines.
That’s because a harmonic is a whole-number multiple of the original signal. As a result, they often occupy frequencies that are notes.
For example, if I saturate a signal that’s the note A2, or 110Hz, and I get a 2nd order harmonic, that would be 220Hz or 110Hz times 2.
220Hz is A3, or a perfect octave of A2.
A 3rd order harmonic formed from 110Hz would be 330Hz. This is note E4.
It doesn’t always work like this, but it often does.
Just as importantly, instruments have natural harmonics or overtones that are created by the instrument, not a plugin or saturator.
These harmonics, make up a large portion of the instrument’s perceived timbre.
So, how does this relate to aliasing?
In short, aliasing generates frequencies that are not musically related.
Whereas a saturator generates harmonics, again whole number multiples of frequencies that are present, aliasing breaks this relationship by occupying unrelated frequencies.
When it occurs to an individual instrument with natural overtones, the same thing occurs. The overtones of the instrument that are related to the played note now include a signal or multiple signals that no longer relate the what was played.
If aliasing is high enough in amplitude, this becomes noticeable and sounds disharmonious, because it is, by definition.
But there’s another way it can have a negative impact on a signal, so let’s cover
You may have heard that aliasing can cause phase cancellation in the highs. I wanted to measure this to be sure, but it’s true. When the signal is reflected down the frequency spectrum, it appears to have a rotated phase of 90 degrees.
I took a sine wave, reduced the sample rate to cause aliasing, and then used an HP to remove the original 1 kHz sine wave.
When measuring the phase rotation of the aliasing, it appears to be 90 degrees. To be sure that this wasn’t an effect of the HP filter, I measured the aliasing without the HP filter, and again, when isolated, they’re 90 degrees out of phase.
Tried again with a linear phase filter and got the same result.
So, it’s safe to say that when the signal folds back, its phase is rotated.
If these are multiples of the original signal, and rotated, they’ll likely cause phase interference.
If they overlapped with the original and rotated 180 degrees, they’d cause complete cancellation. Since they’re rotated 90-degrees, and may only overlap with a portion of the original signal, they’ll cause partial cancellation.
Regardless, the phase cancellation can be noticeable depending on the amplitude of the aliasing, the frequencies they occupy, and if there is any signal in that area with which it can interact.
Oversampling is the most common method for reducing aliasing.
By increasing the sampling rate, the maximum supported frequency is increased. In turn, signals can occupy higher frequencies before aliasing begins.
For example, if I’m running a 48 kHz session and use 2x oversampling, the maximum supported frequency has increased from 24 kHz to 48 kHz.
Oversampling also introduces a low-pass filter right below the highest supported frequency.
This occurs in tandem with saturation, distortion, or the type of effect to ensure no harmonics can exceed the supported frequency.
So, it seems like there’s no reason to avoid oversampling, however, there are a couple of drawbacks.
The obvious one is the extra CPU. Running the internal processing of a plugin at a higher sampling rate is harder on your computer.
The less talked about con is the linear phase processing needed to create such a steep low-pass filter without introducing phase rotation.
As a result, oversampling, if it includes this filter, which most do, introduces pre-ringing distortion.
The greater the oversampling rate, the greater the latency introduced from the filter - resulting in higher amplitude pre-ringing distortion.
So, if you’re not noticing aliasing distortion, it may be best to leave oversampling alone, since the cons from pre-ringing may outweigh the benefit.
Since oversampling has some drawbacks, what are some better solutions?
If you’re computer is capable of handling a full session with a higher sampling rate, then that's a good idea.
By increasing the sampling rate of the session to 96 kHz, aliasing will be reduced without the need for any linear phase oversampling.
Next, you could use frequency-specific saturation plugins.
If you keep the saturation band lower, then the harmonics that form will be in the low mids to mids.
It’s only when you saturate high frequencies or aggressively saturate high-mids that aliasing is possible.
Lastly, if you’re introducing aliasing during mastering, maybe from aggressive clipping, it may be best to use an RMS compressor prior to clipping.
RMS compression will control the dynamics with significantly less distortion than clipping. By controlling dynamics in a clean way before clipping, you reduce the amount you need to clip to get the signal to a louder level.
So, just to sum everything up from the video, since we covered a lot:
Aliasing is caused by a signal crossing the maximum supported frequency. It’s reflected down the spectrum by the same amount it went over.
For this reason, it often occupies disharmonious frequencies. During the process, its phase is rotated 90 degrees, meaning it’ll likely cause phase interference.
Oversampling reduces aliasing at the expense of additional CPU and pre-ringing distortion.
And lastly, alternatives to reducing aliasing include using a higher sampling rate for the session, saturating only low frequencies, and using alternatives to clipping or reducing the need for clipping with RMS compression.