Is 16BIT enough vs 24BIT

[ullanta]

Petrus, exactly, as your own math shows, the resolution throughout the range doubles for every added bit.

As you know -48dB is very different from 48dB. -48dB = 1/256dB - that is, adding -48dB to a given dynamic range increases that range to about 1.004 times the original range. 24bits gives you 1.004 times the linear dynamic range of 16bits.

Take a look at Moulton's nice article: http://www.moultonlabs.com/more/bits_really_bits/

 

[Petrus]

The SD quote comes from the page where they explain calculating file sizes.

This 24 advantage you quote above is easy to undestand, I have used it several times (saved me several times): Let's say we have a source with 70 dB dynamic range and we have to record it on a 16 bit system. If we, for some reason, set levels 25-30 dB too low, the most quiet part will be squashed by the digital noise of the system (it drops below the 16 system noise floor).

The same thing happens with 24 bit system. There the noise floor is much lower, not the theoretical 48 dB, but in case of SD7xx series recorders about 20 dB lower. So, with this 24 bit signal we can perfectly well normalise the signal and the system noise still does not limb above the -96 dBFS level.

I have used SD722 several time to record from PA mixing table and I had to preset the levels with no possibility to readjust during the show. I set the signals at safe levels knowing that with 24 bits I can amplify the signal up to 40 dB without bringing the noise to unacceptable levels.

If you read carefully the SD page comparing 16 and 24 bits, they say that with full level signals you can not hear the difference, but on those signals recorded at -30 dBFS peak levels and then amplified, the hiss becomes noticable in 16 bits, but not in 24 bits. Of course it does, and the superiority of 24 bit systems only comes evident in situstions like this (mistakes...), when you have to amplfy the signal a lot.

 

[Petrus]

Petrus, exactly, as your own math shows, the resolution throughout the range doubles for every added bit.

No, not throughout the range, just adds 6 dB of dynamic range. Each twice/half size loudness difference is represented by adding/taking away the most significant bit.

If we have a sound bite with a dynamic range of say 50 dB, we do not "stretch" it to cover a 16 bit system or 24 bit system or whatever (I think you see it this way whan you say more reolution for whole range, that would be compressing or expanding the amplitude). We fix the max amplitude to 0 dBFS and see how far the low amplitude end of the signal reaches in the bit ladder before we run out of bits, in this case both 16 and 24 work fine. With 100 dB dynamic range sample 16 bits would run out, 24 bits not.

I give up...

Kind of fun actually...

Where are the drawing tools on this board?

 

[Noiz2]

This is another of those big semantic fights. I have heard the exact opposite argument used. It all depends on what your talking about when you say resolution. More samples is more resolution in time, more bits is more resolution in level. If you think of time as the horizontal axis and volume as the vertical axis then sample rate gives you more horizontal resolution and bit rate gives you more vertical resolution. It doesn't actually give you more dynamic range. That would require that 24 bit get a lot louder that 16 bit. They both top out at zero and bottom out at infinity. One divides that range into 65,536 steps the other into 16,777,216 steps.

Now Quicktime and some other apps "convert" bit rates by truncation (tossing out the bottom) but no professional apps that I know of do that. The higher vertical resolution is spread evenly over the range, not just taged to the bottom.

But we are working with sound here, so listen. If you can't hear the difference then stay at 16 bit. I and every professional that I have worked with can hear the difference so we use 24 bit.
Math, while important can not be used to good effect with out taking the whole picture in. For instance to reach it's full spec a digital system needs to be peaking at 0. However if you do that you will create distortion because of the crest factor. THe peak has to be ~~-3dB or your between sample peaks will exceed 0 and you will get distortion. If you only look at the math of SR and bit depth you would not see this, because it's happening in the DA converters. I guess my point is that science (and I was a physics major once) is great at predicting but you always have to reality check to see if your predictions reflect reality.

The other point is "properly used". In a studio with musicians you might be able to set up a very optimized recording setup. When your trying to get a one time event and your guessing at how loud your source is going to scream etc. you are not working anywhere close to an optimized environment.

Not so. It is true that 17 bits means twice the dynamic range compared to 16 bits, 6 dB more that is. It is still dynamic range, not "resolution". If you have 96 dB of true dynamic range (almost impossbile to get in real life situations) it can be fitted to 16 bit system, and having 24 bits with same peaks would just add 48 dB worth of highly detailed hiss at the bottom, the 96 dB at the top is IDENTICAL in both systems.

My point is that 16 bits properly used is so good that it practically always surpasses the source, reproduction and precieving qualities of the audio chain from location to brains (noise floor at location, mics, analog electronics, loudspeakers, viewing room, human hearing). Using 24 bits, but only with great analog front end, gives sound engineers easier times only. For that reason it in not something to loose sleep over.

 

[Noiz2]

Petrus,
Sorry for the repeats in the post above, I responded before reading the rest of the thread so most of my points had already been made.

But if I concede completely to your explanation then I would ALWAYS use 24 bit for production audio. The whole being able to set your levels safely to avoid overloads because of unexpected events would pretty much nail it. And in many cases FX recording and music would not need 24 bit because often the situation is controlled. It's sort of strange then to think that music was the first to push 24 bit, then FX now maybe dialog. One would think ( going strictly by your explanation) that the opposite would be true. So are Music folks just being stupid or is there maybe a little more than theory on paper involved?

 

[Alan H. Chang]

Sometimes a talent may scream during a dialogue track and it may certainly reach higher than 96dB! Even an entire stadium can hit higher than 100dB! A symphony orchestra can get really loud too. This is where fine resolution comes into play, and you want the most detailed approximation your converters can take (20-22bits)

 

[Petrus]

Alan, dB can mean absolute loudness (db SPL) or, the way we use it here, used to denote relations. 96 dB dynamic range does not necessarily have anything to do with loud sounds, it could be voltages, whatever. It is pefectly possibe to capture jet takeoffs with 16 bit 96 dB system, just have your low levels around 50 dB SLP.

Ullanta, I propably figured out what you mean about "more resolution across the whole range" or how you put it.

In 16 bit system there are only 2^15 dicrete levels between the 0 dBFS and -6 dBFS levels. In 24 bit sytem there are 2^23 levels, much more. Does this give more "resolution" for loud signal, more accurate representation. Yes, it does, and it is called MORE DYNAMIC RANGE! Why; bacause wider range of amplitudes can be measured and preserved.

Back to audo physics: complex waveforms can be reduced to a set of pure sine waves (fourier transformation). What some of us call "high resolution" is the ability to record minuscule amplitudes; those sine waves of this set which have only small variations. Those sounds are the low level components, others, like me, call it having high dynamic range to be able to record these low level componets. Having a 24 bit system instead of 16 bits means smaller variations in waveform can be captured. This does not mean that high level signals are more accuratelly described, but that also small variations can be recorded. This ability has a technical name: dynamic range...

For this reason you never find any measure or unit for this "resolution" in technical papers because there isn't one, more bits are expressed as more DR expressed in dB, 6 dB for each additional bit.

 

[ullanta]

Ullanta, I propably figured out what you mean about "more resolution across the whole range" or how you put it.

In 16 bit system there are only 2^15 dicrete levels between the 0 dBFS and -6 dBFS levels. In 24 bit sytem there are 2^23 levels, much more. Does this give more "resolution" for loud signal, more accurate representation. Yes, it does, and it is called MORE DYNAMIC RANGE! Why; bacause wider range of amplitudes can be measured and preserved.

Hey, we're getting close here!

Given your example above - the dynamic range between 0dBFS and -6dBFS is the same in 16 and 24 bit (or any sample size) systems. Having 2^15 or 2^23 steps does not affect range, it affects resolution within that range.

What's important to dynamic range is the SIZE of each step (a value which is constant throughout the range). In a 16-bit system, each step is -96dBFS - or 1/65536. In a 24-bit system, each step is 1/16,777,216 - or -144dBFS. This indicates the smallest measurable signal, indeed the "bottom end" of the dynamic range.

But, here's the thing: in a 24 bit system, there are (to be repetitive) 16,277,216 steps between 0 (silence) and 1 (0dBFS).

of these, only 256 steps cover the portion of the dynamic range that's below -96dB. The entire range of (-144dBFS to -96dBFS) is broken into 256 steps - just .00002 of the overall linear dynamic range. So while resolution is dramatically improved EVERYWHERE, the dynamic range extension is a pretty small amount. And this is also sorta misleading - because 0 is 0, so the real dynamic range is always exactly the same - 0 to 1 (0 dBFS). It's only the resolution - that is, the smallest step above 0 that we can measure - that changes. So, there is a (small) extension of "measurable dynamic range" when going from 16 to 24 bits - but it's not 256 times the dynamic range, or double the dynamic range - it's 1.00002 times the dynamic range. That's not meaningless - it can certainly be apparent for very quiet signals with very high-quality gear. But rarely a major factor. The increased RESOLUTION throughout the range benefits all recordings, and allows one to leave more headroom without as much worry about quality degradation - which I think concensus has shown is the main benefit of 24-bit recording.


-----


decibels are a logarithmic scale. Indeed, when dealing with positive dB numbers, linear dynamic range doubles with every 6 dB. The sum of (1dB + 2 dB + 3 dB ...) is infinitely large. But adding NEGATIVE dB numbers does not DOUBLE linear dynamic range. The sum of (-1dB + -2dB + -3dB + -4dB ...) approaches 1.

 

[Alan H. Chang]

Alan, dB can mean absolute loudness (db SPL) or, the way we use it here, used to denote relations. 96 dB dynamic range does not necessarily have anything to do with loud sounds, it could be voltages, whatever. It is pefectly possibe to capture jet takeoffs with 16 bit 96 dB system, just have your low levels around 50 dB SLP.

You're right. I think I had a misconception with acoustical pressure and electrical current.

 

[filmelife]

I'm doing all my dialogue at 24bit, then editing at 32bit float, all imported music for and fx are also converted to 32bit for editing. I dont wanna state the obvious but for the people that dont know, you have a lot of room to play around with 24bit files and very generous dynamic range from little whispers to loud screams.

Camera audio maybe good for smaller projects. But if you are shooting for the big time in hopes of a distributor picking up your production, i suggest the highest quality, clean 24bit recording you can get.

 

[Petrus]

There is no such thing as "RESOLUTION throughout the range". Better amplitude detail means you can record queter sounds, and this is measured in dB:s of dynamic range. 24 bit system can (in theory) record signals which are 1/256 in amplitude of the quetest signals 16 bits can handle, which is 48 dB difference.

As I tend to write dymanic range all the time, it is time to give up. =

 

[ullanta]

In 16 bit system there are only 2^15 dicrete levels between the 0 dBFS and -6 dBFS levels. In 24 bit sytem there are 2^23 levels, much more. Does this give more "resolution" for loud signal, more accurate representation. Yes, it does, and it is called MORE DYNAMIC RANGE! Why; bacause wider range of amplitudes can be measured and preserved.

Petrus... are you really saying that a 24 bit system has more dynamic range between 0dBFS and -6dBFS than a 16-bit system? Please explain this! How much more?


MORE RESOLUTION THROUGHOUT THE RANGE. Dude, any two signal levels (e.g., -12dBFS and -6dBFS). How many different discrete signal levels can you resolve in thwe range between them in 16 bits? in 24 bits? The difference will always be: there are 256 times as many discrete signal levels resolvable in a 24 bit system than in a 16 bit system. As you said yourself, in the range of 0dBFS to -6dBFS: 24 bit audio has 256 times the number of discrete resolveable signal levels (we can just say resolution for short). Between 0 and -96dBFS: 24 bit has 256 times the resolution of 16-bit. Between -20dBFS and -13dBFS? 24 bit has 256 times the resolution of 16-bit. A range between any two signal levels is just that - a range. It does not get larger or smaller when you subdivide it.... between -6 and 0dBFS is always 6dB of range. Wheteher divided into 32000 steps or 8 million steps, it's six dB. WITHIN THE SAME RANGE, the larger sample size only increases your ability to precisely resolve signal strength.

Does this apply only to quiet signals? No. In the range between the loudest possible signal (0dBFS) and the next loudest possible signal in a 16-bit system (-0.00013dBFS), a 24-bit system can distinguish 256 different signal levels. This is what is meant by "throughout the range". A 24-bit sample is 256 times more accurate than a 16-bit sample, regardless of whether the material being sampled is very quiet.

 

[Oleg 2]

kill film i dont have any doubt that noone ever would hear any difference recorded dialog inside intire movie , if you really need to make your sound better invest in good analog stage and ato d converter , that would be worth much more then reording another 8 bit of garbage ( fans ,street noise , camera noise , self recorder noise and your booman hands which would be high resolution recordings ). waist of space , by the way , i just finish my 9 feature ( 7 minutes in heaven - transfax productions , omry givon- director )with hd for cinema relice and yes it is 16 bit and noone have any trouble with ( f-900)

 

[Noiz2]

Sorry kid no mater how loud you SHOUT what you are describing is dynamic resolution not range. Range is the difference between the top and the bottom. In the case of most audio situations its the difference between clipping and the noise floor.

In technical papers about theoretical performance this is may well be talked of as dynamic range because in a theoretical environment you will gain dynamic range at the bottom of the levels because you can describe lower levels with a higher resolution system than you can with a lower resolution system. This however has no practical application because in the real world your range is not set by your resolution (16 VS 24) but by the noise floor of your analog section and the clipping point of your recorder. Since neither of these are effected by your bit rate your dynamic range doesn't change in the real world. But your resolution does.

You can argue theory all you want but this isn't really the place for it. The folks here are dealing with the real world not some perfect one in a scientists mind.


AND dB is ALWAYS relative and never absolute. SPL is relative to a reference level as is dBu and all the other dB's.

And while your correct in saying that what you get is better description of small loudness changes you are quite wrong to assume those are by definition quiet sounds. You are getting a more accurate description of ALL level changes including small level changes near the top of the range.

Alan, dB can mean absolute loudness (db SPL) or, the way we use it here, used to denote relations. 96 dB dynamic range does not necessarily have anything to do with loud sounds, it could be voltages, whatever. It is pefectly possibe to capture jet takeoffs with 16 bit 96 dB system, just have your low levels around 50 dB SLP.

Ullanta, I propably figured out what you mean about "more resolution across the whole range" or how you put it.

In 16 bit system there are only 2^15 dicrete levels between the 0 dBFS and -6 dBFS levels. In 24 bit sytem there are 2^23 levels, much more. Does this give more "resolution" for loud signal, more accurate representation. Yes, it does, and it is called MORE DYNAMIC RANGE! Why; bacause wider range of amplitudes can be measured and preserved.

Back to audo physics: complex waveforms can be reduced to a set of pure sine waves (fourier transformation). What some of us call "high resolution" is the ability to record minuscule amplitudes; those sine waves of this set which have only small variations. Those sounds are the low level components, others, like me, call it having high dynamic range to be able to record these low level componets. Having a 24 bit system instead of 16 bits means smaller variations in waveform can be captured. This does not mean that high level signals are more accuratelly described, but that also small variations can be recorded. This ability has a technical name: dynamic range...

For this reason you never find any measure or unit for this "resolution" in technical papers because there isn't one, more bits are expressed as more DR expressed in dB, 6 dB for each additional bit.

 

[Noiz2]

Alan, you folded too soon. He's not right, dB is always a relative measure.

Alan, dB can mean absolute loudness (db SPL) or, the way we use it here, used to denote relations. 96 dB dynamic range does not necessarily have anything to do with loud sounds, it could be voltages, whatever. It is pefectly possibe to capture jet takeoffs with 16 bit 96 dB system, just have your low levels around 50 dB SLP.

You're right. I think I had a misconception with acoustical pressure and electrical current.

 

[Noiz2]

A good point. As I sort of said above your "effective" dynamic range is not determined by your bit depth. If my choice were between a really good analog section and a 16 bit recorder or an ok analog section and a 24 bit recorder I would go with the first always.
There are advantages to 24 bit but not necessarily big one (depending on what your recording and what with). I personally think it is probably unnecessary for dialog. I would certainly spend money elsewhere first. THe waste of space argument sort of depends on what your recording on and what the work flow is. If post is in 24 bit then it's not really a waste of space, if it's in 16 bit then it is. If your recorder has limited storage it may be a pain for you and a time waster in which case the benefits may well not be worth the effort. If your working in broadcast and or some multi track nightmare like a reality show then it is probably a waste of space and time (1 1/2 transfer time). If your analog section is not so hot then it also may be a waste of space. If your on a noisy set however that "high resolution" noise you record may be a godsend to post who needs some good noise prints to optimize noise reduction and now has more bits to work with and a better chance of recovering better sounding dialog.

But we all work in the real world so what might sound great on paper has to take a back seat to what sounds great in the real world and practicality sometimes trump everything.

kill film i dont have any doubt that noone ever would hear any difference recorded dialog inside intire movie , if you really need to make your sound better invest in good analog stage and ato d converter , that would be worth much more then reording another 8 bit of garbage ( fans ,street noise , camera noise , self recorder noise and your booman hands which would be high resolution recordings ). waist of space , by the way , i just finish my 9 feature ( 7 minutes in heaven - transfax productions , omry givon- director )with hd for cinema relice and yes it is 16 bit and noone have any trouble with ( f-900)

 

[Alan H. Chang]

Alan, you folded too soon. He's not right, dB is always a relative measure.

Alan, dB can mean absolute loudness (db SPL) or, the way we use it here, used to denote relations. 96 dB dynamic range does not necessarily have anything to do with loud sounds, it could be voltages, whatever. It is pefectly possibe to capture jet takeoffs with 16 bit 96 dB system, just have your low levels around 50 dB SLP.

Scott, I never folded. Just re-worded by interpretation a bit. I know decibel is always a relative measure. There is a formula to finding out the absolute dynamic range of 16 bits or any bits you want to know it. I'll dig through my paperwork and post.

 

[Petrus]

Study Fourier transformations; small amplitude changes, appearing alone or piggybacking a loud signal, are those quiet sounds that 24 bit system captures better (or at all) compared to 16 bit. That's dynamic range (call it resolution if you wish). This simple fact seems to escape you. It is not only the totally quiet parts, in AD conversion the small changes are low level signals, big, loud changes are high level signals. They are all present at all times in a natural, complex sounds we have.


Quote from ullanta:

"The entire dynamic range of any digital audio measured in dBFS is exactly the same... 0 to 1. A one-bit PCM system has a dynamic range of 0-1, as does 8 bit, 16-bit, or 24 bit. One is always the loudest representable sound. 0 is always no sound. More bits just add resolution to our measurement of signals falling between 0 and 1. Extra bits determine the smallest representable quantity, not the largest - the largest is always 1 (which we call 0dBFS)."

About this ullanta's 0 to 1 alien math: as we use digital (numeric) system, we could use any base, hexadecimal, decimal, dozen... Computers use binary for practical reasons. What does your math look like if we mark down the voltage values in decimal system where max value is 100000, say? Does 0dBFS suddenly become 10? In math the rules stay the same no matter which base you use.

One more thing, if you insist calling this "resolution", what is the measurement used? I have never seen anythig else but dB (as in dynamic range) connected with bit depth.

 

[ullanta]

About this ullanta's 0 to 1 alien math: as we use digital (numeric) system, we could use any base, hexadecimal, decimal, dozen... Computers use binary for practical reasons. What does your math look like if we mark down the voltage values in decimal system where max value is 100000, say? Does 0dBFS suddenly become 10? In math the rules stay the same no matter which base you use.

Petrus, the range of 0 to 1 is not binary, these are real numbers. 0dBFS is represented as 2^(sample size) -1 / 2^(sample size)-1, or 1. 0 (-infinity) is 0/2^(sample size). Between these two are all the interesting values. You're correct that representing the intermediate values in binary, decimal, or trinary doesn't make a difference. We tend to represent these numbers in decibels, as I illustrated in a long boring post with lots of binary and dBFS equivalents.

Again, you're missing a big part of the message here, though... even in a decimal representation, the max value will be 1, and the dynamic range will be 0 to 0dBFS (1). But the intermediate values will look like 32767/65535 (-6dBFS) rather than 0111111111111111/1111111111111111

One more thing, if you insist calling this "resolution", what is the measurement used? I have never seen anythig else but dB (as in dynamic range) connected with bit depth.

Petrus, Petrus - dear Petrus! The measure is exactly "bit depth" or "sample size" or "sample resolution" - it's measured in bits, which determine exactly how many discrete levels can de distinguished in your entire dynamic range. It is by far one of the MOST commonly discussed (along with sample rate) statistics of digital audio. Look, here we have a whole thread (not the first, as we well know) about it. Dynamic range (in digital representation, not analog quality) is an EFFECT of bit depth. Bit depth IS resolution; dynamic range, because of the added few "steps" on the bottom due to increased resolution, is an EFFECT of resolution.

 

[Noiz2]

Alan, you folded too soon. He's not right, dB is always a relative measure.

Scott, I never folded. Just re-worded by interpretation a bit. I know decibel is always a relative measure. There is a formula to finding out the absolute dynamic range of 16 bits or any bits you want to know it. I'll dig through my paperwork and post.

I was joking. I need to figure out the little smiley face things so people can tell...