fbpx

Oversampling vs Aliasing

Test setting

This setting has been chosen to accentuate the aliasing effect (OVERDRIVE channel with PREAMP knob @ 11).

THD with 1KHz 0dB TestTone / SampleRate 88.2KHz

DownFilterTransBand 2.0%

It can be easily observed that alias decreases by increasing the Upsampling factor. In this test, we used UpSampling factors from 1x to 1024x.

DownFilterTransBand 43.0%

We can take advantage of the high Samplerate – 88.2KHz – to reduce the slope of the DownSampling LinearPhase LPF by increasing the DownFilterTransBand to 43%. As it can be seen in the “Oscilloscope” section below, this reduces both the oscillations of the output waveform and the pre-ringing artifacts of the DownSampling LinearPhase filter. Please notice that the softer LPF does not affect the audible frequency range.

THD with 10KHz 0dB TestTone / SampleRate 88.2KHz

DownFilterTransBand 2.0%

DownFilterTransBand 43.0%

A more uncompromising test consists of using an input 10KHz sine wave instead of the one @ 1KHz. However, take into account that this is quite an overkill because a guitar signal will NEVER have so much information at 10KHz (0dB!) – in this range it will be very likely LESS than -30dB. You can roughly imagine that in a real situation the alias is, in the worst case, 30dB quieter than what is reported in each of these graphs.

Oscilloscope

Output with 1KHz 0dB TestTone / SampleRate 88.2KHz

Output with 1KHz 0dB TestTone / SampleRate 88.2KHz - Zoom

Increasing the DownFilterTransBand – by taking care of not affecting the audible spectrum – reduces both the oscillations of the output waveform and the pre-ringing artifacts of the DownSampling LinearPhase filter.

Clash with reality - Would you pass a blind test?

The analysis above suggests to:

  • increase the UpsamplingFactor as much as you can to reduce the aliasing artifacts (for the OVERDRIVE channel, we recommend to render files at 64x if SampleRate is 88.2/96KHz, 128x if SampleRate is 44.1/48KHz);
  • use a high SampleRate (88.2KHz or 96KHz) and reduce the slope of the DownSampling LinearPhase LPF to minimize oscillations and artifacts.

This page could stop here if we were not talking about music. But we are, and we should trust our ears. So, do these numbers and graphs have any connection with reality?

Here is a blind test: SaturnLO – Oversampling vs Aliasing

It contains 2 groups of files:

  • RHY: rhythm hi-gain guitar;
  • LEAD: lead guitar (additional gain has been put by raising the INPUT knob to 12dB).

Cabinet simulations have been done with a minimum phase EQ.

Each group containts 3 files: A, B, C. One @ 1x, one @ 64x, one @ 256x.

Can you tell which one is 1x? And 256x?

Good luck.