FabFilter User Forum
Samplerates: the higher the better, right?
Hi,
I watched your very interesting video "Samplerates: the higher the better, right?" about sample rates for production, and the purpose of oversampling.
www.youtube.com/watch?v=-jCwIsT0X8M
I have been a big advocate for mixing at 96khz and recording at the highest sample rate possible (96khz or higher), so this video was a bit of a revelation.
The example shown was recreating a sine wave at 'normal' rate of 48khz compared to 96khz, and the lesson was that the sampling process was able to recreate that 20khz tone perfectly with only two samples per cycle, and so higher samples rates are unnesessary, and in fact could actually make things worse. This is obviously the Nyquist Shannon sampling theorem in action.
My question is this, the example demonstrated this perfectly with a nice, predictable sine wave, but how does this work with a complex real-world wave form? How can a complex wave form be reconstructed perfectly with only two samples per cycle, and why would sampling at a higher rate not make things more accurate?
I ask because I am now re-assessing my approach of mixing at 96khz, because I am now questioning whether I am making things sound better or worse.
Given the issues of mixing at 96khz compared to 48khz, and just relying on plugins that employ oversampling to do their job, is it still better to record at 96khz (or higher) and then downsample to 48khz to work on, or just to record at 48khz in the first place?
In other words, is this video suggesting it's better to mix at 48khz but it's still better to record or generate sound at the highest sample rate available (vis-a-vis buying or streaming 'high-res' version of music). Or, are you suggesting there is NO benefit whatsoever in recording at a higher rate than 48khz?
I've heard people say there is no need to record at a higher rate because "you don't need to", or "it makes no difference", but I suspect often they're just saying that because they just can't subjectively tell the difference personally, but in fact they don't actually *know* for a fact.
Can you help please? Maybe Dan would like to make a follow-on video? :-)
Many thanks,
David
Hi David,
This question has been a popular one since the start of digital recording, and we did not intend to end this discussion with the video. We tried to show that it’s not as black and white, and there might be various reasons for using one over the other.
I would always suggest recording in the sample rate it’s going to be mixed at. This eliminates the need of converting between sample rates. The best option for this can differ per project.
Projects that are going to be mixed with a heavy compression and saturation might benefit from recording at 48KHz and oversampling turned on for as much plug-ins possible. However, a jazz quartet that will just need some eq and mild compression during mix down might benefit from recording at 96KHz, as there would be no need to turn on oversampling. However, people might have different workflows, and for different reasons.
Looking forward to see other folks chipping in to describe what they use and why.
Cheers,
Just wanted to say the videos you all produce are some of the most informative I've seen from a technical and use point of view. Keep making them!
A comparison I was hoping to see in the video was between 48kHz and 96kHz where both use oversampling in the plugins. Could there be negative consequences for using the higher sample rate in such a situation or is that somehow alleviated by the oversampling? It's a fantastic video that probably made a lot of people wonder if what they're doing is negatively impacting their mixes.
"My question is this, the example demonstrated this perfectly with a nice, predictable sine wave, but how does this work with a complex real-world wave form? How can a complex wave form be reconstructed perfectly with only two samples per cycle, and why would sampling at a higher rate not make things more accurate?"
David, that's Nyquist Shannon at work again. A complex real-world waveform that is band-limited to 20kHz (this is key) can indeed be reconstructed perfectly with a sampling rate just a little bit above 40kHz. The band limitation ensures that one, and only one, particular waveform can pass through any given sequence of samples. Any line that you could draw passing through the samples that is not the original, would automatically have a larger bandwidth than 20kHz. And that holds true whether the signal is a sine wave or not, provided its bandwidth doesn't go above the sampling rate / 2.
To put it more intuitively, the band limitation ensures a certain "smoothness" in the waveform that forces you to draw one and only one possible line through your samples. Any extra samples that you would get from increasing the sampling rate would necessarilty fall on that one line, so no, they do not provide any extra information about the original waveform, they don't make the representation any more accurate.
In case you haven't seen it, this video is also excellent at explaining and visualizing all this:
www.youtube.com/watch?v=cIQ9IXSUzuM
Hi Cabirio,
Thanks for this. You've hit the nail right on the head with that explanation. What you described, and what the FL Studio guy in the video described was the exact missing piece of the puzzle I've been looking for. The video was especially good for me as, although I have a reasonable grasp of some digital audio concepts, I am not a mathematician and so his practical demonstration triggered a lightbulb moment for me; as you say: it's the band limiting.
That picture of the wave 'scribble' with the dog and the clouds in it represented my misconception exactly!
This video, your explanation, and the FabFilter video that prompted this thread represent a complete u-turn for me about the issue of high-resolution audio.
As we know, there is good reason to employ theoretically unnecessary 24-bit recording, as it allows for a huge safety margin against clipping and so it has a practical use. However, with regards to high sample rate recording, I now gasp in wonder about how an entire industry could have risen up on the basis of "High Res audio", when Sony et al must know that it's effectively nonsense.
Maybe the "Samplerates: the higher the better, right?" video needs a follow up or an update with this?
Thanks,
David