Difference between 16 and 24 bit

[alehel]

I understand (I believe so anyway) the difference between a file that is 44.1 khz and 192 khz. What I don't really understand is the difference between 16 and 24 bit. What does this effect, and is there any point to a recording that is 44.1/24 compared to a 44.1/16? It seems like most explenations of highres music I find go into what the khz mean, but not much mention of what bit actually involves.

 

What I've found so far is "The more bits there are meanwhile, the more accurately the signal can be measured in the first place, so 16-bit to 24-bit can see a noticeable leap in quality.". What is a bit in digital music?

[esldude]

Well imagine your DAC puts out a max signal of 2 volts.

 

With 16 bits you can have 65,536 steps from 0 volts to 2 volts. The very smallest difference in two levels with 16 bit with this DAC is something like 30 microvolts. 30 millionths of a volt. This for the smallest bit. The next to the smallest bit is twice this big or 60 microvolts. So on and so forth. The largest bit is a 1 volt difference.

 

With 24 bits you can have 16,777,216 levels between 2 volts and 0 volts. The smallest difference is 1.2 nanovolts or 1.2 billionths of a volt. The smallest bit. The next largest bit is twice that or 2.4 nanovolts. The largest bit once again is 1 volt. But there are more bits of more sizes in between.

 

Now in reality once we get converted to analog various noise sources will prevent you from getting performance at 24 bit. But you can maybe get 20 bit performance though more commonly 18 volts if you are careful.

 

Now chances are very good you won't hear any difference between the two.

 

Another way to characterize them is decibels. Each bit gives a 6 db difference. 16 bits gives a total dynamic range of 96 decibels. 24 bit gives theoretically 144 db. But again it is very difficult to get analog results better than 20 bit or 120 db of dynamic range.

[outofthebox]

esldude has pretty much nailed it. the two main differences are better signal to noise ratio and a better dynamic range.

 

this wikipedia page on audio bit depth explains these two differences between 16 and 24 bit audio very well.

[bdiament]

I understand (I believe so anyway) the difference between a file that is 44.1 khz and 192 khz. What I don't really understand is the difference between 16 and 24 bit. What does this effect, and is there any point to a recording that is 44.1/24 compared to a 44.1/16? It seems like most explenations of highres music I find go into what the khz mean, but not much mention of what bit actually involves.

 

What I've found so far is "The more bits there are meanwhile, the more accurately the signal can be measured in the first place, so 16-bit to 24-bit can see a noticeable leap in quality.". What is a bit in digital music?

 

Hi alehel,

 

In plain terms, bits can be thought of as the letters that make up a digital "word".

 

In more practical terms, in a 16-bit recording, one is making the most of all 16 bits when the signal is loudest -- at the top 6 dB of the possible levels (6.02 dB for those who want more precision). Now of course, with real music the signal is not always at the top of the level scale. In other words, not every part of the song is at the loudest point. Further, there are things in the background that are lower in level than things in the foreground. Instrumental harmonics (the things that differentiate a Steinway from a Baldwin or a Gibson from a Fender) are considerably lower in level and spatial cues (whether real or studio generated) are lower still.

 

To put it into numbers, the harmonics might be 20 dB lower in level than the loudest sounds. Spatial cues might be 40 dB lower in level. (I'm just picking the numbers arbitrarily to illustrate the point.) With a 16-bit recording, those harmonics that are 20 dB lower in level will be encoded using about 13 bits, not 16. The spatial cues that are 40 dB lower in level will be encoded using about 10-bits, not 16. This accounts for the coarsening of the sound and thinning of instrumental harmonics many have notices with 16-bit audio since the beginning.

 

Now let's look at the resolution of the same items with a 24-bit recording. The harmonics, at -20 would be encoded using about 21 bits. The spatial cues, at -40 would use about 18 bits. Both still exhibiting more resolution than a full level 16-bit recording. So if you make the same recording at 16/44 and at 24/44, you'll find the complexity of instrumental harmonics much better preserved on the latter. Same with the spatial cues: where the 16-bit version defocuses the space, the 24-bit version makes the room boundaries clear and easy to hear by comparison.

 

Perhaps an easy way to think of it is sampling rate is horizontal resolution while word length (i.e., number of bits) is vertical resolution.

Hope this helps.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[InfernoSTi]

Barry,

 

Thank you for this post. It may be the most helpful description of 16 vs 24 bit audio I have read. I have such respect for the knowledge you impart to us on this forum…you are very much appreciated.

 

Best,

John

[wkhanna]

+1

[PewterTA]

The answer to the question is 8 bit... how come all you guys got this wrong????

[Don Hills]

... So if you make the same recording at 16/44 and at 24/44, you'll find the complexity of instrumental harmonics much better preserved on the latter. Same with the spatial cues: where the 16-bit version defocuses the space, the 24-bit version makes the room boundaries clear and easy to hear by comparison. ...

 

Barry, do you believe that this is due to a fundamental deficiency of 16 bit versus 24 bit, or due to less than perfect implementation?

[monteverdi]

I find 16 vs 24 bit a more significant improvement than higher sampling rate.

[gmgraves]

I find 16 vs 24 bit a more significant improvement than higher sampling rate.

 

 

I do a lot of recording, as I've said before, and you are absolutely correct. The difference between 16 and 24-bit is profound, the difference between 48 KHz, 88.2 KHz, 96 KHz, and yes, 192 KHz is audibly, not so much. In fact I have recorded different rehearsals on different nights at different sample rates (but all at 24-bit) and frankly, while there is a slight difference between 44.1 and 48 KHz, I have been unable to personally hear any difference between 48 KHz and 96 KHz and neither has anybody I've played the samples for.

[sandyk]

George

Are they reliably able to pick the differences between the 48kHz 24 bit and 192kHz 24 bit samples ?

 

Alex

[firedog]

I find 16 vs 24 bit a more significant improvement than higher sampling rate.

 

If you've listened to many of BIS classical recordings in 24/44.1, you know how good it can sound.

[esldude]

If you've listened to many of BIS classical recordings in 24/44.1, you know how good it can sound.

 

A very good point. I have heard one of those and I don't see how they could be considered anything other than very good.

Would they be better at 48 khz, I don't know. But I have to think the difference if there is one would be small.

 

Very early on when I started digitizing all analog inputs (at 24 bit), we are talking nearly 20 years ago, 48 khz seemed better than 44.1 to me. 48 khz seemed awfully close to full transparency while 44.1 was pretty close, but not quite. 44.1 seems better to me now. Either my hearing has degraded enough (no question HF response has) or 44.1 is closer to theory than it was. I haven't compared any direct microphone feeds at the two rates recently though I will do so in the near future.

 

My opinion is with current abilities to deal with data no reason not to do 24 bit. I would prefer 48 khz be the standard, and am not convinced there is anything to be had with higher. I am equally convinced if in fact higher rates offer anything it is an extremely tiny difference. One that likely few have the equipment to discern.

[sandyk]

My opinion is with current abilities to deal with data no reason not to do 24 bit. I would prefer 48 khz be the standard, and am not convinced there is anything to be had with higher. I am equally convinced if in fact higher rates offer anything it is an extremely tiny difference. One that likely few have the equipment to discern.

 

So all the latest high resolution DSD downloads and the new DSD capable DACs ,well recorded SACD, as well as the 24/96 and 24/192 files from HDTracks and others, and all the 24/192 material from Soundkeeper Records are in your honest opinion a waste of money ? You have already stated in What's Best Forum, that the dynamic range of 16/44.1 is all that is needed.

[jerryct]

This is the explanation i found so far. Bit depth is only about dynamic range if properly dithered.

 

24bit vs 16bit, the myth exploded!

 

Here also some nice test files. You can here the 8 Bit audio file without dither which has artefacts and the other with dither where you only here more noise as the voice level decreases. So it is only about signal to noise ratio.

 

Dynamic Range, Dithering and Noise Shaping

 

Often there is a very misleading analogy to computer vision where more bit depth means better image quality. But this is not how audio bit depth works:

 

Color depth - Wikipedia, the free encyclopedia

 

jerry

[bdiament]

Hi Don,

 

Barry, do you believe that this is due to a fundamental deficiency of 16 bit versus 24 bit, or due to less than perfect implementation?

 

My experience, using different digital formats at what I believe to be the current state of the art, has me convinced it is clearly the former.

To my ears, low level information (and what happens to it at 16-bits) tells the story unambiguously, every time.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[alehel]

Thanks bdiament, that's the best explanation I've heard so far .

[Boris75]

I find 16 vs 24 bit a more significant improvement than higher sampling rate.

 

+1. This is also an opinion that has been expressed by Robert von Bahr of BIS and eClassical fame on this forum (under the name Bissie).

[Boris75]

Hi alehel,

 

In plain terms, bits can be thought of as the letters that make up a digital "word".

 

In more practical terms, in a 16-bit recording, one is making the most of all 16 bits when the signal is loudest -- at the top 6 dB of the possible levels (6.02 dB for those who want more precision). Now of course, with real music the signal is not always at the top of the level scale. In other words, not every part of the song is at the loudest point. Further, there are things in the background that are lower in level than things in the foreground. Instrumental harmonics (the things that differentiate a Steinway from a Baldwin or a Gibson from a Fender) are considerably lower in level and spatial cues (whether real or studio generated) are lower still.

 

To put it into numbers, the harmonics might be 20 dB lower in level than the loudest sounds. Spatial cues might be 40 dB lower in level. (I'm just picking the numbers arbitrarily to illustrate the point.) With a 16-bit recording, those harmonics that are 20 dB lower in level will be encoded using about 13 bits, not 16. The spatial cues that are 40 dB lower in level will be encoded using about 10-bits, not 16. This accounts for the coarsening of the sound and thinning of instrumental harmonics many have notices with 16-bit audio since the beginning.

 

Now let's look at the resolution of the same items with a 24-bit recording. The harmonics, at -20 would be encoded using about 21 bits. The spatial cues, at -40 would use about 18 bits. Both still exhibiting more resolution than a full level 16-bit recording. So if you make the same recording at 16/44 and at 24/44, you'll find the complexity of instrumental harmonics much better preserved on the latter. Same with the spatial cues: where the 16-bit version defocuses the space, the 24-bit version makes the room boundaries clear and easy to hear by comparison.

 

Perhaps an easy way to think of it is sampling rate is horizontal resolution while word length (i.e., number of bits) is vertical resolution.

Hope this helps.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

 

Many thanks for a very clear, very instructive post. I am learning a lot thanks to you.

[jerryct]

Sorry, but the explanation is not really correct.

 

An example:

 

Your loudest sound gives you 2 Volt

For the sake of simplicity we use full scale for 2 Volt in the digital domain.

I.e. 2V are represented as 65535 with 16 Bits.

 

A quieter part gives you about -32dB. I.e. 0,05V (calculation: 20*log(2V/0,05V)=32,041dB)

An even more quieter sound gives you about -72 dB. I.e. 0,0005V

 

Then 0,05V is represented as 65535/2V*0,05V = 1638,375. As you can only represented natural numbers (no float or fix point) this can be represented as 1638 or 1639.

The rounding error (or quantization error) is 0,375 or 0,625.

 

0,0005V is represented as 65535/2V*0,0005 = 16,38375. So it is represented as 16 or 17.

Rounding error: 0,38375 or 0,61625

 

The rounding error for the quieter part is almost the same. You don't use 4Bits for -72dB (i.e. 0,0005V) or 11Bits for -32dB (i.e. 0,05V).

 

You only have rounding errors which are not random. So you hear artifacts. Dither adds random noise. So you don't have artifacts. If you use 24Bit you have smaller rounding errors.

 

jerry

[bdiament]

In a straight 16-bit recording vs. a straight 24-bit recording, there is no application for dither, which is useful when reducing word length.

 

I won't argue a theoretical analysis. Rather than discuss the theoretical "map", I'm much more interested in the experience of walking the real territory. (Unfortunately, the numbers tell us absolutely nothing about what happens to the sound of a musical instrument or to the sound of the room around it.)

 

What I suggest is folks listen for themselves to a good variety of samples and draw their own conclusions. The problem is, most folks don't have access to making the same recording at straight (i.e., no additional processing) 16-bits and at 24-bits. When this is done on an equal footing, again, with no word length or any other sort of manipulation, the differences are quite plain to hear.

 

Even if one starts with a 24-bit source and applies the most transparent dither and noise shaping possible to create a 16-bit version, the limitations of 16-bit are still quite plain to hear. Good dither mitigates them to a degree but the experience of hearing a recording at 16-bits vs. hearing it at 24-bits and listening to what happens to low level information will speak for itself.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[alehel]

How much an effect on file size would a 24 bit wav file have compared to a 16 bit one with, for example, 60 min of music?

[bdiament]

Hi alehel,

 

How much an effect on file size would a 24 bit wav file have compared to a 16 bit one with, for example, 60 min of music?

 

At the same sample rate, the 24-bit file will be 50% larger.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[bdiament]

It is all too common to confuse the signal-to-noise ratio specification with real dynamic range capability.

This seems to occur quite frequently when the subject of word length comes up.

 

The thing is, a low level signal -- or the low level components of any real music signal -- might be well above the noise floor of a given format. This has no bearing whatsoever on the *quality* of the signal compared to the same sound recorded at a higher level.

 

As with anything in audio, some will hear it and some will not.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[gmgraves]

George

Are they reliably able to pick the differences between the 48kHz 24 bit and 192kHz 24 bit samples ?

 

Alex

 

 

No. That's the point. I can't, and my audiophile friends can't. But we can fairly reliably tell the difference 16 and 24-bit (keep in mind that there is nothing formal about these listening tests, so no statistical analysis is possible. Just a group a friends doing some relaxed listening.