Oversampling, do we really need it?

In plugins we sometimes see the 'HD' Button or an x2 / x4 ... options.

This means that the plugin will oversample the signal before processing it,

And will downsample it back to the Session's sample rate before sending the signal. back to the tack.

Here we can see a distortion plugin that creates harmonics with and without the 'HD' feature

I used sine wave as a test signal and passed it through Klirrton Grindstein ( a great plugin for all you metal head out there!)

Oversampling VS No oversampling

The same process without the oversampling introduced alias frequencies

And we can see that we have harmonics BELOW our fundamental sine waves.

Does it necessarily mean that we will not like the none HD process with a guitar running through it?

The answer is NO!

While we can prove scientifically the difference between the oversampled and the none oversampled process it does not mean that in real-world scenarios we would feel the need for oversampling.

The cost of oversampling in the example above is more latency (0.1ms) and more CPU (over x5 times increased CPU consumption)

We will need to use our ears to determine which process sounds better to us on a case-to-case basis as it depends on the content that we are using.

So one more question to be asked is:

Why we have multiple oversampling values?

Oversampling help us get the Ani aliasing filter farther away from our recorded signal.

Every signal processing we introduce comes with a price. 

Here are few examples of a sine wave going through an EQ with a HiCut a filter 

In the scope the original sine in green, the signal after the EQ in blue.

anti aliasing filter (low oversampling values)

As we can see even when the Filter value is higher than the sine wave we can introduce phase shift to our signal.

So when we are using it on an individual track of a multi mic'ed performance 

We can introduce phase issues between the tracks.

The steeper our filter slope will be the harsher the phase issue will be.

frequncy cancelation (no oversampling)

The further away the filter will be from our signal the fewer issues it will introduce 

with oversampling

The same applies in oversampling the higher the sample rate is

We can have a more relaxed filter slope and introduce fewer phase issues to our signal.

Another factor in our decision will be how intense the process we introduce to the signal is.

If it is a mild change that introduces fewer harmonics, or that the content of our audio consists of mainly low-end information we may not need to oversample it or oversample to a lower value.

For example:

We have 2 identical tracks with pink noise (the same file on both tracks) with the same plugin, all the params ware the same aside of the oversampling value.

The phase is flipped on one of the tracks to reveal if we have any difference between the tracks.

Mild processing

mild process

We can see that X2 and X8 with no clipping results in the same processing 

the signal that was X2 and X8 are identical.

On the other hand X1 and X8 (bellow) will not

mild processing no over sampling (NOL test)

Heavy processing.

heavy processing

However the moment we introduce heavy processing we can see the difference between X2 and X8 (no cancelation).

To conclude.

Oversampling helps us avoid issues related aliasing & the anti-aliasing filter.

We pay for it in CPU and latency.

In the real world, it does not make sense to always use the oversampling as it might not be noticeable enough to justify the cost.

I hope this gave you a better perspective on the issue.

Cheers Lemmy