Sample Rate

Discuss music production with Ableton Live.
TobiasHahn
Posts: 127
Joined: Fri Nov 14, 2008 11:15 am
Location: Ableton Headquarters

Re: Sample Rate

Post by TobiasHahn » Sun May 22, 2011 3:47 pm

OneZeroMusic wrote: The reconstruction of the signal would be based upon the algorithm used to calculate the interpolated points outside of the sampled points, would it not? And while we may be able to accurately calculate the interpolated points there still exists a possibility of small variation. Which would lead to the conclusion that the greater the amount of sampled points there are, the lower the amount of interpolated points that would be required and reduce the risk of error in the interpolation calculation?
Using "ideal" gear (whatever that means...), what you get from a higher sample rate at the same bit depth is a better signal to noise ratio. So, yes, a signal sampled at 24 bit 192 kHz contains less noise (given the right equipment) than a signal sampled at 24 bit 48 kHz. A signal sampled at 23 bit 192 kHz, however contains as much noise as a 24 bit 48 kHz recording. (Thanks docprosper for the link to the wikipedia oversampling article.)

So if you want to look at the interpolation error you always need to consider bit depth and sample rate.

Best,
Tobias

Akshara
Posts: 377
Joined: Tue Dec 21, 2004 10:16 pm

Re: Sample Rate

Post by Akshara » Sun May 22, 2011 5:51 pm

A question then, for clarity. So when one edits audio at the sample level, does having a higher sample rate increase the number of samples per second available, or not?

docprosper
Posts: 1193
Joined: Sat Apr 24, 2010 11:20 am
Location: Fredericksburg, VA

Re: Sample Rate

Post by docprosper » Sun May 22, 2011 6:52 pm

Yes, the number of samples per second in the edited audio should be proportional to the sample rate.
-hamish
Funk N. Furter wrote:Post properly.
Ableton Live Suite | M4L | Powerbook | Launchpad | APC40 | Faderfox | 2x1200 | Xone:96 | ...
---> http://soundcloud.com/kilcraft

Akshara
Posts: 377
Joined: Tue Dec 21, 2004 10:16 pm

Re: Sample Rate

Post by Akshara » Mon May 23, 2011 5:42 am

Then it follows that a 2x or 4x higher sample rate allows for a much finer resolution when manipulating audio at the sample level.

I'm pointing this out because most of the talk is about aliasing, audible frequencies, filters, headroom, etc., with the implication that there is no benefit with using a higher sample rate for obtaining a finer resolution of the captured audio signal; yet my understanding through the years has been that having 192,000 snapshots per second of 32-bit length words is considerably more data to work with than having 44,100 snapshots per second of 16-bit length words, allowing for finer resolution with editing and processing.

If this is inaccurate, then I am open to learning why.

invol
Posts: 604
Joined: Sat Dec 11, 2004 6:47 am
Location: Brooklyn, NY
Contact:

Re: Sample Rate

Post by invol » Mon May 23, 2011 5:56 am

Always record at 24bit. Always export at 24 bit, unless you understand why to use 32bit (not common), or you are done mixing and mastering and ready to burn a CD.

The only reason for 192 kHz is for certain formats that are not common. 384 kHz is for DSD transfer compatibility. Again, not common. Yes, 192 Khz can sound better, but ONLY if you have the best room, the best mics, the best pre's, etc... 96 kHz will be just fine too, btw.

44.1 kHz is for CD audio
48 kHz if for DV / Video projects

Use the multiples of those for higher sample rate versions - e.g., 88.2 for music, 96k for DV, etc.

Dan Lavry, who makes arguably the best converters on the planet (used by many top mastering engineers), does not make anything that does higher than 96 kHz. He has a white paper floating around somewhere as to why anything higher is just a waster of time and resources.

dancerchris
Posts: 343
Joined: Wed Oct 25, 2006 4:48 pm
Location: Los Angeles, CA USA

Re: Sample Rate

Post by dancerchris » Wed May 25, 2011 3:17 pm

It is important to understand that Nyquist theory says that only the frequency content of the material bellow half the sampling rate is reconstructed. It is reconstructed perfectly. It is done with sinc (not mispelled) functions which is not quite the same as sine waves. The confusion the OP seems to have is in the reconstruction of only the data below half the sampling frequency. The stuff above that is hopefully filtered out otherwise aliasing occurs. Therefore the content is not reproduced exactly, but in essence it doesn't matter from unmodified recording point of view because that inaudible material will be lost in the output stage (whether you play it on speakers or bounce it down to a lower frequency format neither is capable of producing the frequencies in question).

There is legitimate argument about the higher frequency samples having a "different sound" based on what processing is done with effects (or instruments), but you should ask yourself if it is worth the hassle and increased processing requirements to obtain that "different sound". That is a matter of preference and not science.
Live 8.4.2 / Win 8 Pro 64 bit / Core 2 Quad 2.66 GHZ / 8 Gb ram
Presonus Firepod / Axiom 49 / PadKontrol
Various guitars, keyboards, sax and friends

docprosper
Posts: 1193
Joined: Sat Apr 24, 2010 11:20 am
Location: Fredericksburg, VA

Re: Sample Rate

Post by docprosper » Thu May 26, 2011 11:10 am

Actually aliasing does matter - the signals and/or noise that alias down due to poor filtering before the adc (or a lack of oversampling) roll down into the bin from dc to half the sample rate. This then corrupts the signals in the band of interest, i.e. the audible range. If there is no signal at these higher frequencies or minimal noise then it's a moot point.
-hamish
Funk N. Furter wrote:Post properly.
Ableton Live Suite | M4L | Powerbook | Launchpad | APC40 | Faderfox | 2x1200 | Xone:96 | ...
---> http://soundcloud.com/kilcraft

dancerchris
Posts: 343
Joined: Wed Oct 25, 2006 4:48 pm
Location: Los Angeles, CA USA

Re: Sample Rate

Post by dancerchris » Thu May 26, 2011 2:39 pm

docprosper wrote:Actually aliasing does matter - the signals and/or noise that alias down due to poor filtering before the adc (or a lack of oversampling) roll down into the bin from dc to half the sample rate. This then corrupts the signals in the band of interest, i.e. the audible range. If there is no signal at these higher frequencies or minimal noise then it's a moot point.
-hamish
Where you referring to my post:?: I never said it didn't matter. That's why I said that the upper signals were "hopefully" filtered out.
Live 8.4.2 / Win 8 Pro 64 bit / Core 2 Quad 2.66 GHZ / 8 Gb ram
Presonus Firepod / Axiom 49 / PadKontrol
Various guitars, keyboards, sax and friends

Khazul
Posts: 3185
Joined: Wed Feb 23, 2005 5:19 pm
Location: Reading, UK

Re: Sample Rate

Post by Khazul » Thu May 26, 2011 3:26 pm

Positive side for high sampling rates:
- raises the Nyquist frequency which means reflected overtones from quantising errors during sampling will have diminished to near nothing by the time they impact the audible spectrum, but see the consideration below.
- Do to above, you generally get slightly cleaner sonds from synths and audio processing as you at less at the mercy of randomly good or poor bandwidth limiting and/or anti-aliasing algorithms on whatever synths and processors you use (assuming they have any at all) and particularly less at the mercy of artefacts from non-oversampled slice/grain type processing of audio.

Negative side:
- do you really want to work at such a high sample rate? It requires proportionately more computing resources of all kinds - CPU speed, memory bandwidth, I/O bandwidth to storage. Working at 96K or 192K for eg might mean you are having to freeze bounce alot to keep your computer useable. Bouncing/freezing is a trace off - it trades CPU use for I/O and cache memory use (stream pre-calculated audio instead of real time calculating it). At very high sample rates, your computer may not be able to deal with the I/O side of that trade off.
- eventually you have to sample rate convert back down - so choose you sample rate converter tool carefully if you want to fully gain from recording and working at a high sample rate. A shitty sample rate conversion can easily undo all that effort and pain.

Something to consider:
A modern analog to digital converter uses various techniques to minimise quantising errors anyway. For example, a modern fifth order delta-sigma ADC that can deliver an SNR of 160dB is probably using 64x over sampling - when sampling to 44.1KHz - that equates to the audio signal being effectively sampled at 2.8MHz or so and because of the error/noise shaping that goes on in such converters its actually equivalent to the performance of a dumb ADC operating at a way higher oversampling amount in the order of millions. Because the effectively initial sample rate is so high, then the Nyquist frequency is also extremely high leaving a huge gap between the sample rate limit and the actual filtered output spectrum. Having a huge gap mean a much cleaner and more easily constructed filter can be used to finally deliver the digital signal at the desired bit rate.

So, going back to to reason for using a very high sample rate:
If you record at 192K, then SRC down to say 44.1 before mixing/processing etc - then its net loss due to the use of SRC with a high quality initial converter.
If you record at 192K then stay at 192K for all processing, then finally use a high quality SRC to get down to 48/44.1K as required then its a net gain *if* your computer can handle it.
If you record at 44.1K, then quality SRC up and do all you processing at an exact binary power multiple (say 4x = 176.4K) then finally use a good SRC to get back down again, then its a net gain *if* the processing plugins you are using support the high sample rate and work well at that sample rate as effectively you are exploiting end to end oversampling.

In every day practice - many modern high quality plugins (especially those associated with dynamics/transient and EQ processing) might offer option of oversampling (but even oversampling can be done well or badly), or internally employ oversampling to minimise unwanted distortion/noise, so the benefits of working at a high sample rate are small. With typical careful use of high quality EQ in a well controlled mix, then they may even disappear to nothing. Poorer plugin that can still work properly at a high sample rate OTOH will benefit more from it.

Is it worth the pain though for the marginal gain? IMHO if you have a must-use plugin that has a shitty high end, then there are some plugin oversampling plugins around that allow the shitty plugin to get its own over-sampling wrapper and that might help is the SRC on either side is of decent quality.
Nothing to see here - move along!

docprosper
Posts: 1193
Joined: Sat Apr 24, 2010 11:20 am
Location: Fredericksburg, VA

Re: Sample Rate

Post by docprosper » Thu May 26, 2011 6:11 pm

dancerchris wrote:
docprosper wrote:Actually aliasing does matter - the signals and/or noise that alias down due to poor filtering before the adc (or a lack of oversampling) roll down into the bin from dc to half the sample rate. This then corrupts the signals in the band of interest, i.e. the audible range. If there is no signal at these higher frequencies or minimal noise then it's a moot point.
-hamish
Where you referring to my post:?: I never said it didn't matter. That's why I said that the upper signals were "hopefully" filtered out.
sorry, misread... I really should respond after my morning coffee. all love baby :)
Funk N. Furter wrote:Post properly.
Ableton Live Suite | M4L | Powerbook | Launchpad | APC40 | Faderfox | 2x1200 | Xone:96 | ...
---> http://soundcloud.com/kilcraft

nowtime
Posts: 1566
Joined: Thu Jan 11, 2007 7:24 pm
Location: Homefree

Re: Sample Rate

Post by nowtime » Thu May 26, 2011 6:29 pm

I've always played it "safe" using 24/44100. Now I want to use my Toslink connector on my keyboard and I am forced to move to 48000. Are there any hidden downsides to this other than a bit more CPU and disc space?

And when rendering files, do I need to keep everything at 48000, or should I render down to 16/44100 (for CD burning and ipod)?

Tone Deft
Posts: 24152
Joined: Mon Oct 02, 2006 5:19 pm

Re: Sample Rate

Post by Tone Deft » Thu May 26, 2011 6:35 pm

nowtime wrote:I've always played it "safe" using 24/44100. Now I want to use my Toslink connector on my keyboard and I am forced to move to 48000. Are there any hidden downsides to this other than a bit more CPU and disc space?

And when rendering files, do I need to keep everything at 48000, or should I render down to 16/44100 (for CD burning and ipod)?
loosely speaking, just some SRC artifacts down in the -90 - -100dB range. make sure your Live effects and SRC settings are set to Hi-Q. do some tests for yourself.

seems that's the worst thing about 44.1k, it's not a common multiple of the other sample rates, just 88.2k.

iPod conversion is probably more of an issue of the mp3 settings, which is a much noisier process than PCM .wav sample rate conversions. most iPod listening is in a noisy environment to begin with.
In my life
Why do I smile
At people who I'd much rather kick in the eye?
-Moz

dancerchris
Posts: 343
Joined: Wed Oct 25, 2006 4:48 pm
Location: Los Angeles, CA USA

Re: Sample Rate

Post by dancerchris » Thu May 26, 2011 7:38 pm

Sample rate conversion quality is independent of the ratio of the conversion sample rates. It is common misconception that it is better to have whole number multipliers. The math works out "perfectly" because modern algorithms upsample first to the common multiplier and then do the conversion with a whole number factor down.
Live 8.4.2 / Win 8 Pro 64 bit / Core 2 Quad 2.66 GHZ / 8 Gb ram
Presonus Firepod / Axiom 49 / PadKontrol
Various guitars, keyboards, sax and friends

Tone Deft
Posts: 24152
Joined: Mon Oct 02, 2006 5:19 pm

Re: Sample Rate

Post by Tone Deft » Thu May 26, 2011 7:43 pm

then why do comparison tests choose to pick 96k to 44.1k conversion as their test setup? like this site:
http://src.infinitewave.ca/
In my life
Why do I smile
At people who I'd much rather kick in the eye?
-Moz

dancerchris
Posts: 343
Joined: Wed Oct 25, 2006 4:48 pm
Location: Los Angeles, CA USA

Re: Sample Rate

Post by dancerchris » Fri May 27, 2011 1:12 am

Tone Deft wrote:then why do comparison tests choose to pick 96k to 44.1k conversion as their test setup? like this site:
http://src.infinitewave.ca/
From the URL you provided:

"How about conversions between other sampling rates?
To limit the amount of work, we have decided to only test 96 kHz to 44.1 kHz conversion. At other sampling rates SRCs may behave differently. However in many cases they actually behave similarly."

This has to do with the application of poor aliasing filters that causes the problems, not the upsampling/downsampling ratio. Better filters yield "actually behave similarly". No doubth this choice of sample ratio is to catch the lower quality filtering problems.
Live 8.4.2 / Win 8 Pro 64 bit / Core 2 Quad 2.66 GHZ / 8 Gb ram
Presonus Firepod / Axiom 49 / PadKontrol
Various guitars, keyboards, sax and friends

Post Reply