[opus] High Sampling Rates

Abhijit Patait apatait at nvidia.com
Mon Jun 9 01:44:42 PDT 2014

```Ø  Do you have any references for me to investigate, I am trying to understand how noise is reduced by introducing higher sampling rates. (I tried to search, but maybe it is so obvious that nobody even explains it)

This is not very obvious. It requires you to understand basic signal processing theory. I will give some pointers below.

Any physical signal (e.g. audio coming out of speaker, current driving the speaker) is analog and contiguous in nature. On other hand, all codecs (and everything in DSP) works on discrete signals (samples).

When you convert from discrete samples to analog signal, you need to apply an ideal filter, which has a frequency response of 1.0 (0 dB attenuation) in it’s passband and 0.0 (–infinite dB) in its stopband. The most important thing to understand here is that such an ideal  filter does not exist in reality. What we do (most DACs do this) is to approximate such a filter using IIR’s or an FIR with some delay (sufficient delay that the truncation of impulse response does not matter much).

With higher sampling rates, it gives us flexibility to design this filter such that the stopband attenuation is large enough (although not –infinite dB) and the noise is filtered to inaudible levels.

From: opus-bounces at xiph.org [mailto:opus-bounces at xiph.org] On Behalf Of Edwin van den Oetelaar
Sent: Saturday, June 07, 2014 5:22 PM
To: Andrew Lentvorski
Cc: opus at xiph.org; Jean-Marc Valin
Subject: Re: [opus] High Sampling Rates

On Sat, Jun 7, 2014 at 10:58 AM, Andrew Lentvorski <bsder at allcaps.org<mailto:bsder at allcaps.org>> wrote:

On 6/7/14, 1:55 AM, Jean-Marc Valin wrote:
> Actually... no! 24-bit can indeed be useful as extra margin and Opus
> can actually represent even more dynamic range than 24-bit PCM. That's
> not the case for 192 kHz. There's no "margin" that 192 kHz buys you
> over 48 kHz. You can do as much linear filtering as you like, the
But lots of effects are not linear--simulating a tube guitar amplifier,
for example.

Even something as straightforward as resampling a signal to 44.1KHz is
going to benefit from starting at 192KHz rather than 48KHz.

There may not be more signal information but there will be less noise.

Hello Andrew,
Do you have any references for me to investigate, I am trying to understand how noise is reduced by introducing higher sampling rates. (I tried to search, but maybe it is so obvious that nobody even explains it)

In resampling - as far as I understand it - you first limit the bandwidth to the 1/2 the sampling rate to prevent aliasing problems.
If your input signal contains no energy outside the resampled bandwidth to start with, how is this going to increase the signal to noise ratio?

Edwin van den Oetelaar

-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information.  Any unauthorized review, use, disclosure or distribution