96k

Forum for discussion of Professional Audio Technology.

Please address all Sales questions to sales @ lavryengineering.com

Please address all Technical support questions to techsupport @ lavryengineering.com

Moderators: Brad Johnson, Marc

96k

Postby nu-tra » Sun Oct 02, 2005 11:11 pm

I just readyour sampling theory pdf... You say 192k is a waist is 96k a waist as well? I'm think 48 would be key since my album is going to be recorded in conuction of a dvd.. I call mastering labsand they say record in 96k...

This all gets confusing.
nu-tra
 

Re: 96k

Postby lavrye » Thu Oct 13, 2005 3:34 pm

nu-tra wrote:I just readyour sampling theory pdf... You say 192k is a waist is 96k a waist as well? I'm think 48 would be key since my album is going to be recorded in conuction of a dvd.. I call mastering labsand they say record in 96k...

This all gets confusing.


The good question to ask is what is the ideal sample rate. I believe the number is in the 60KHz-70KHz range for the optimum - fast enough to contain 30KHz in theory, and certainly more then 25KHz in practice.
Going faster only reduces accuracy, increases file size and demands more processing power for what we do not hear.

We do not have a 60KHz-70KHz standard, so 88.2KHz is a bit fast but not too bad.

But yes, we have 44.1KHz CD's and much of the sound for video is compressed...

I too, wish I could have a better answer.

Regards
Dan Lavry
lavrye
Site Admin
 
Posts: 522
Joined: Sun May 15, 2005 10:36 am
Location: Poulsbo, WA

Postby Mike Derrick » Mon May 01, 2006 4:42 pm

Dan,

You mentioned that going faster reduces accuracy,...just wondering what you mean by that?
Reduces sonic quality?
Would recording at 60 kHz sound better than 192kHz?

~ mike
Mike Derrick
Mike Derrick
 
Posts: 17
Joined: Wed Jun 29, 2005 3:24 am
Location: White Rock, BC

Postby lavrye » Tue May 02, 2006 10:16 am

Mike Derrick wrote:Dan,

You mentioned that going faster reduces accuracy,...just wondering what you mean by that?
Reduces sonic quality?
Would recording at 60 kHz sound better than 192kHz?

~ mike


Hello Mike,

I wish there was more time in the day - I am going to write a paper about that issue.

There are a lot of "facts of life" in engineering. There are always tradeoffs.

For example, there is a tradeoff between speed and power - when "all other things are held constant" the more speed you want, the more energy you use.

Technology does move forward, but for any given state of technology, that trade off between speed and power is a fact of life.

Well, that was just an example for a tradeoff. When it comes to electronics signals (both analog and digital), there is a tradeoff between speed (in this case sample rate and bandwidth) and accuracy (how accurately we can process the signal).

That fact is not restricted to audio. It is true for all electronics.
Therefore, in electronics, one begins by asking "how much bandwidth do we need". The answer is application driven:

A postal weighting scale does not need to be fast - you put a letter on it and 1 second time to stabilize (or so) is fine. What you want is "supper accuracy", because you are charging millions of customers...

A video AD or DA needs to operate at some MHz rates, because the signal requires it. But the accuracy will be less then the slow weighing scale...

A wide bandwidth scope may require GHz bandwidth to measure wide band signals (fast signals). We know that the accuracy is going to be a lot lower then the slower video application...

Look at today’s AD's or DA's or Opamps, or comparators.... and sort them by accuracy. You will see that the more accurate devices are slower, the fastest devices yield less accuracy. Alternatively, sort the devices by speed, and the slow devices are more accurate.

You can do the same study 10 years ago, 20 years ago, 100 years ago or 100 years from now. That speed vs. accuracy will remain a fact of life, just like the first law of thermodynamics will...

The presentation of detailed reasons will take a whole paper. For now, think of charging a capacitor (which is an issue in all electronic circuits). The more time you have, the closer the charge is to the desired value. Try to do it fast, and you must reduce the time constant, which will impact all sorts of accuracy related issues... Think of a voltage step into an OPamp - the longer you wait, the better it settles to where it needs be. Not enough settling time means compromising accuracy.

Aside from the hardware related issues, there are "design concept" issues. Modern AD's or DA's are based on sigma delta concepts. When "all things held constant" - the same IC technology, the same over sampling ratio and the same filter order, the more bandwidth you ask for, the less accurate the conversion.

You can achieve great accuracies (near real 23-24 bits) when the bandwidth is say 10Hz. For audio, we are not yet at real 20 bits, at a GHz we are not yet at 6 bits....

Less accuracy means more distortions and noise. So where do we set the sample rate? In theory, at twice the bandwidth we can hear (Nyquist theorem). In practice, we need some reasonable practical margin - that should be left up to the designer. Note that if you believe that we hear say 25KHz, sampling at 192KH means you are taking 300% margin. It is a huge waste of accuracy, disk space.

The "orientation" by the marketing types is "faster, faster and faster". Clearly, everyone would agree that sampling audio at say 1GHz is idiotic, crazy, belongs in the nut house. So how about 1/2GHz? Still crazy. How about 100MHz? Or 1Hz?

The question should be "WHAT IS THE OPTIMAL RATE"!
Just going blindly towards faster and faster is the wrong way to go! The OPTIMUM is always based on the APPLICATION. We do need enough bandwidth to "do the job", and a little margin is a good thing.

If we hear 25KHz, then 60KHz is great. If one believes we hear (or feel) 30-40KHz (I do not) then 88.2-96KHz is fine.

The marketing forces (NOT ENGINEERING) that dreamed up that 192KHz
had no basis for it, other then self serving selling of gear and gaining market position... They did not and still do not have a single explanation why, because there is no such explanation. So they spread a bunch of innuendo - all of which does not hold up to scrutiny.

Regards
Dan Lavry
lavrye
Site Admin
 
Posts: 522
Joined: Sun May 15, 2005 10:36 am
Location: Poulsbo, WA

Postby Mike Derrick » Wed May 03, 2006 4:18 am

Dan

thanks for your detailed and prompt reply.

It's been about a year since I was here at your site reading your papers etc.
I think I'm gonna have to brush up on my converter/sample rate knowledge, as some things are a little foggy to me.
I understand the part about the waste of energy etc for recording at 192.
That did always seem a little ridiculous to me, although I've never recorded at 192, all I had to go on was hear-say and subjective opinions (which are never to be trusted of course.)
So yes, it makes sense that the accuracy is compromised by the speed of having to sample at 192. And I agree, what do we need that information for anyhow?,....radio waves, microwaves?,...definitely not music.

You lost me in this paragraph...

"You can achieve great accuracies (near real 23-24 bits) when the bandwidth is say 10Hz. For audio, we are not yet at real 20 bits, at a GHz we are not yet at 6 bits..."

Is recording with a DAW at 24bit 48kHz not "real" 24 bit? If not can you clarify?
And what did you mean by "for audio, we are not yet at real 20 bits,"

~ mike
Mike Derrick
Mike Derrick
 
Posts: 17
Joined: Wed Jun 29, 2005 3:24 am
Location: White Rock, BC

Postby lavrye » Wed May 03, 2006 9:32 am

Mike Derrick wrote:Dan

thanks for your detailed and prompt reply.

It's been about a year since I was here at your site reading your papers etc.
I think I'm gonna have to brush up on my converter/sample rate knowledge, as some things are a little foggy to me.
I understand the part about the waste of energy etc for recording at 192.
That did always seem a little ridiculous to me, although I've never recorded at 192, all I had to go on was hear-say and subjective opinions (which are never to be trusted of course.)
So yes, it makes sense that the accuracy is compromised by the speed of having to sample at 192. And I agree, what do we need that information for anyhow?,....radio waves, microwaves?,...definitely not music.

You lost me in this paragraph...

"You can achieve great accuracies (near real 23-24 bits) when the bandwidth is say 10Hz. For audio, we are not yet at real 20 bits, at a GHz we are not yet at 6 bits..."

Is recording with a DAW at 24bit 48kHz not "real" 24 bit? If not can you clarify?
And what did you mean by "for audio, we are not yet at real 20 bits,"

~ mike


Hi Mike,

There are some practical reasons why we like to use binary numbers for digital electronics. And there are reasons (some are “historical” now) why we often use them in multiples of 8 digits, which is called a byte. So one byte is 8 bits. One can use 2 bytes for a 16 bit binary number, which was sufficient for a CD red book format (16 bits). But as soon as we wanted better resolution, it was “natural” to add a byte which makes 24 bits.

So an improvement from say 16 bits to 17 bits performance already called for handling 24 bits. It does not harm anything because we are not handling the data a bit a time, we are handling it in multiples of 8 bits…

But that practice make a 17 bit audio data into a 24 bit data? Performance wise, using 24 bit data to express 17 bit audio means you have 17 good bits and 7 useless bits. The useless bits can be set to zero, they can be “jumping around randomly”… they do no harm and they do no good.
In the case of 18 good bits, your 24 bit data has 18 good bits and 6 useless bits.
In the case of 20 good bits, your 24 bit data has 20 good bits and 4 useless bits.
And so on.

So in all cases of 17-24 bits you do have 24 bits. Some carry the audio. I call the useless bits “marketing bits” because the concept of 24 bits is often misrepresented or at least misleading.

If I sold you a 12 cylinder car, with only 8 cylinders connected to the drive shaft, do you have a 12 cylinder car? Is it an 8 cylinder car?

One way to determine if a bit is a good bit (containing valuable data, not “junk”) is to make sure that it contributes about 6dB to the performance (such as improved dynamic range and improved distortions). The simple way to look at it is 6dB dynamic range per good bit. So say you have 20 good bits, then you expect approximately 6*20 = 120dB dynamic range (unweighted measurement). A 16 bit format, assuming all the bits are good yields about 6*16 = 96dB. So 24 good bits would yields 144dB dynamic range.

But, there is no gear capable of 144dB dynamic range. In fact, very rarely can we reach 20 bits. Take any micpre, set the gain to 40, and you are no better then 18 bits because the noise coming into the AD is already high enough to “overpower” the lower 6 bits…

Of course, we also do not need 144dB. We can not hear that much range.

So 24 bits does not really describe the audio quality. It is always less then 24 bits, almost always less then 18 bits, rarely as good as 20 bits…

Regards
Dan Lavry
lavrye
Site Admin
 
Posts: 522
Joined: Sun May 15, 2005 10:36 am
Location: Poulsbo, WA

Postby Mike Derrick » Thu Jul 06, 2006 2:29 am

Dan that all makes sense to me.

And that's all fine for final product right...?
But doesn't multi-track recording at 24 bits allow for a nice wide dynamic range to work with, not only for high track count with a wide range of dynamics within the music but also a chance to use the upper bits to be "further" away from the noise floor???

Maybe 24 bits is excessive, especially considering how compressed music tends to get these days at various points along the chain of recording---to--->consumer.

However, are you indirectly suggesting that multi-track recording could survive just fine in the 16 bit realm with its 96 dB range???

~ mike
Last edited by Mike Derrick on Thu Jul 13, 2006 10:25 pm, edited 1 time in total.
Mike Derrick
Mike Derrick
 
Posts: 17
Joined: Wed Jun 29, 2005 3:24 am
Location: White Rock, BC

Postby lavrye » Thu Jul 06, 2006 9:26 am

Mike Derrick wrote:Dan that all makes sense to me.

And that's all fine for final product right...?
But doesn't multi-track recording at 24 bits allow for a nice wide dynamic range to work with, not only for high track count with a wide range of dynamics within the music but also a chance to use the upper bits to be "further" away from the noise floor???

Maybe 24 bits is excessive, especially considering how compressed music tends to get these days at various points along the chain of recording---to--->consumer.

However, are you indirectly suggesting that multi-track recording could survive just fine in the 16 bit realm with its 96 range???

~ mike


The data may be 24 bits, bit the lower bits are useless, because they carry no music, only noise. The "real bits" are the bits that carry the music.

The marketing "definition" of 24 bits is based on the fact that there are 24 bits there. Using such "definition" I can make you a 124 bits audio gear. All you need is to add 100 bits of random noise on the least significant side (bit 25, bit 26....). Will it improve anything? No way. It will just cost you in data size (storage space...).

The reason we have 24 bit standards is due to the fact that we like to deal with bytes (multiples of 8 bits such as 16 bits, 24 bits, 32 bits...). That is why going from say 16 bits to 17 bits already called for 24 bits (with 7 bits not really used thus "wasted").

I am of course taking about the conversion, and the final outcome (CD, DVD and so on). One needs a lot more bits in the digital world, inside the computer, for signal processing in a DAW. Signal processing bits is a different subject, and it does not yield better resolution then the converters do.

In fact, most of the limitation in dynamic range, thus "real bits" is "cast in stone" by the time you come out of the mic[pre. There is hardly any musical material out there with 120dB dynamic range, which is near 20 bits.

I am not suggesting that multitracks should be limited to 16 bits. I am suggesting that there is no real audio chain in real life situation (mic, mic pre and AD) out there that goes over 20 real bits. So under the best of conditions, the last 4 bits are of no value. That is why I said “there is no such thing as 24 bits performance”.

The difference between real 16 bits and real 24 bits is 8 real bits, which is 48dB dynamic range. In a linear scale, the real 24 bits performer (not available in the real world) is 256 times more accurate then a 16 bit performer.

Better then 16 bits? Yes! As good as 24 bits? No!

Regards
Dan Lavry
lavrye
Site Admin
 
Posts: 522
Joined: Sun May 15, 2005 10:36 am
Location: Poulsbo, WA

Postby Mike Derrick » Sun Jul 09, 2006 5:31 pm

Alright, let me see if I can glue this all together and see if I've understood everything you've been saying so far,...

Considering the science of digital recording, would the optimum system for multitrack recording be a 20bit 60kHz standard? ..with the higher standards (ie 24bit 192kHZ) being a waste of data/processing/etc, and the lower standards (ie 16bit 44.1kHZ) being a compromise sonically due to the technical problems associated with the conversion process?

~ mike
Mike Derrick
Mike Derrick
 
Posts: 17
Joined: Wed Jun 29, 2005 3:24 am
Location: White Rock, BC

Postby lavrye » Mon Jul 10, 2006 9:26 am

Mike Derrick wrote:Alright, let me see if I can glue this all together and see if I've understood everything you've been saying so far,...

Considering the science of digital recording, would the optimum system for multitrack recording be a 20bit 60kHz standard? ..with the higher standards (ie 24bit 192kHZ) being a waste of data/processing/etc, and the lower standards (ie 16bit 44.1kHZ) being a compromise sonically due to the technical problems associated with the conversion process?

~ mike


You have it right.

20 bits is fine for the ear, because it yields near 122 dynamics, which is about the range between a sound isolated room to standing next to a 747 jet engine. 20 bits is also not easy achievable in the "real world", because the mic noise and micpre input stage noise, when amplified by a micpre gain of say 30, make for a lower dynamic range the 120dB.

But, I do no particularly object to 24 bits, because digital electronics in general tends towards use of bytes (multiple of 8 bits) so 24 bis is a "natural" thing to do. Say you just let those lower bits "flop in the wind" - carry a random pattern... the energy the lower 4 bits carry is so small you will not have a problem.

Regarding the point about 60KHz sampling: I will not argue that it is precisely at 60KHz, it may be at 70KHz. In an "ideal world" the sampling rate should be set at some OPTIMAL RATE, not faster and faster, and too slow is not good either. In theory, it has to be fast enough to include all we hear. In practice, we need some reasonable margin, because we can not design on the "theoretical edge". But too much margin will cost in performance accuracy, data size, processing requirement.

I am not against 88.2KHz or 96KHz, because they are not too far from the optimal sample rate. The margin is a bit high, but I can live with it. Going to 192KHz data rate is ridicules.

The industry salesmen that tried to promote 192KHz did a disservice to audio and to their customers. Digidesign was pushing the 192KHz DAW a lot, but their brand new system for live performance is at 48KHz, not even at 96KHz.

I am not into mp3's. They will never be able to compete with uncompressed (or lossless compression) material. While still poor quality mp3's have been getting better. Why? Because the people that do mp3 concentrate on what we do hear. I, too, concentrate on what we do hear. The 192KHz advocates were pushing a whole range of frequencies we DO NOT hear...

Regards
Dan Lavry
lavrye
Site Admin
 
Posts: 522
Joined: Sun May 15, 2005 10:36 am
Location: Poulsbo, WA

Postby Mike Derrick » Wed Jul 12, 2006 12:10 am

Dan wondering if you can clear something up for me regarding bits.

Are they exclusively for dynamic range?

If yes,...would a selection of recorded music with a dynamic range of let's say 80dB sound identical at both 16bits(96dB range potentaial) and 24bits(144dB range potential)? Assuming the recording level was set so that the maximum peak for loudness reached near 0 for digital recording.

At what point does noise floor and dither noticably change the the original sound wave?
And does noise floor/dither only effect the quiet passages of the music? or can it effect the louder passages as well,...or does it effect the entire sonic quality regardless of loud/quiet?

Are there reasons for DAWs to be recording and editing at 24bits?
Or could we record in DAWs at 16bits and achieve the exact same sonic quality?

~ mike
Mike Derrick
Mike Derrick
 
Posts: 17
Joined: Wed Jun 29, 2005 3:24 am
Location: White Rock, BC

Postby lavrye » Wed Jul 12, 2006 10:20 am

Mike Derrick wrote:Dan wondering if you can clear something up for me regarding bits.

“Are bits exclusively for dynamic range?”

In math and engineering, the word is orthogonal stands for “totally independent” or “having zero influence on each other”.
So in theory, dynamic range and bandwidth are orthogonal.

So yes, fundamentally, bits are for dynamic range.

Some comments
1. It is easy to convert between dynamic range and bits. But dynamic range is a “single” number, which is appropriate for a flat noise floor, which is not always the case.

2. In practice, the picture is also more complex, then “a single number”. One can measure the performance with zero signal, and that is important to know. But you have no guarantee that the amount of “idle” noise with no signal, will reflect what happens when you apply a signal. The unwanted signal (noise) may grow when a signal is present. In fact it may be highly dependent on the input waveform (input signal).

“If yes,...would a selection of recorded music with a dynamic range of let's say 80dB sound identical at both 16bits(96dB range potentaial) and 24bits(144dB range potential)? Assuming the recording level was set so that the maximum peak for loudness reached near 0 for digital recording.”

If the noise floor is reasonably flat, that 80dB number means that the bits 14, 15 and 16 are rather useless, and there is no point in going to 24 bits.
But say you have a lot of noise at frequencies that are too high for hearing (say near 20KHz), and very little noise at the hearing sensitive range (say 1-4KHz). Then having more bits is of great value, because they will help retain the range where is counts.

In fact, say you have material with 20 bit performance, and you want to release it as a 16 bit CD. By using noise shaped dither, you can keep the 20 bits performance over the range the ear hears well (say 1-4KHz), by shifting the noise to frequencies the ear does not hear well (say 18-22KHz). So you can effectively have 20 bits psychoacoustic result in a 16 bit format. It will measure poorly, because the dynamic range will “report” on the high noise level near 20KHz, while it does not matter to the ear…

“At what point does noise floor and dither noticeably change the original sound wave?
And does noise floor/dither only effect the quiet passages of the music? or can it effect the louder passages as well,...or does it effect the entire sonic quality regardless of loud/quiet?”

I do not know how to answer “At what point does noise floor noticeably change the original sound wave?” I would say, at the point you hear a change.
But I can say the following: A bit reduction from say 18 or 20 bits to 16 bits, assuming the bits were of real value (carrying musical content) requires dither to make sure that the sound does not change, other then some limited increase in noise. Yes, dither does increase the noise floor, but without dither, at low level signals, the unwanted distortions peaks due to quantization will be HIGHER
Then with no dither. Dither raises the AVARAGE noise, and it is spread across the range. With no dither, the total noise is less, but the problem is due to concentration of energy at specific frequencies, making both the ear and the specification worse. Of course a noise shaped dither is the best way to go.
At what point does dither play a role? It is not an "on or off" issue. The closer you get to the noise floor, the higher the impact. I can measure the distortions when there is no dither while the signal is at 80dB level on a 16 bit system.

“Are there reasons for DAWs to be recording and editing at 24bits?
Or could we record in DAWs at 16bits and achieve the exact same sonic quality?”

We use 24 bits because electronics and computers rely on bytes (multiples of 8 bits thus 16,24,32… 48…64 bits), so anything over 16 bits "automaticaly" becomes a 24 bits format. Many of the processes in a DAW require much more then 24 bits, but that is for processing. That is a whole different subject. You may need a lot of digital processing bits. It has little to do with conversion bits or the final result of a processed mix.
Conversion (and final product) bits are different then processing bits.

Regards
Dan Lavry

~ mike
lavrye
Site Admin
 
Posts: 522
Joined: Sun May 15, 2005 10:36 am
Location: Poulsbo, WA

Postby lfeagan » Thu Jul 13, 2006 4:00 pm

I am not Dan. However, I can take on this part of the question adequately.

Mike Derrick wrote:Are there reasons for DAWs to be recording and editing at 24bits?
Or could we record in DAWs at 16bits and achieve the exact same sonic quality?

~ mike


I finally got some dinner and have a better way of writing this post. :D

Setup: You have two or more 16-bit files you want to mix together. In particular you might be interested in altering the levels from their recorded state.

Assumptions:
1) You likely recorded all tracks to maximize the use of the dynamic range where your equipment behaves very nearly linearly in response to signals.

2) Your sources had differing levels in reality, but to maximize the precision of the recordings (to avoid loss of precision in the low order bits) you (logically) raised the gain on a mic recording a quiet insturment. The thought here being that you might want to later raise the volume of the insturment relative to the other volumes you recorded.

Problem: When you mix together 16-bit tracks, you may want to raise or lower the levels of one track relative to another. When you do this, if you are only working with 16-bits, you may/will lose the low-order bits of the signal you lower. Thus, you have suffered a loss in precision.

Solution: Using 24-bits (or more) for the editing (while the signals are resident in memory) is a good idea. This ensures that lower order bits are not truncated. Engineers and computer scientists do not generally like to throw away bits before it is determined that they are not needed. In the case of a DAW outputting to CD, we can safely return to 16-bits when we prepare the 44.1/16 output. Additionally, the use of 24-bits allows any smoothing or more complex additive/multiplicative/convolution operations between tracks to be legitimately useful.

Example: You have 2 bits of resolution. Thus, you can only represent the values 0,1,2,3. I ask you to perform the following calculation:

11 / 10 = ??

In decimal this is 3 / 2, which you realize is 1.5. However, we are in a digital world. So, 11 / 10 = 10. At this point you gasp and realize that I just said that 3 /2 = 2. Surely I must be off my rockers with some sort of new age math. :) No, I am not. The wonders of decimal. I should say that I just did that IEEE754 style with proper rounding. Many embedded systems simply drop the bit necessary for rounding and say 11 / 10 = 01. Yes, 3 /2 = 1;

This alone isn't that bad. However, what if we perform a calculation such as this on a multitude of tracks and then add them together! Now we can start to get answers that are really ridiculous. Lets do the following:
01 / 10 + 01 /10 = ??
In decimal, this is 0.5 + 0.5 = 1
In a system with rounding, we get a result of 2. And in a system that drops bits, we get 0. These are, clearly, undesirable results. Now, lets assume that I were to read in files that were two-bit precision files, but I represented them internally as 3-bit files for the purposes of calculation. To represent them, we take the original 2-bit bit value and add on some padding bits to the left. I will note them with a ,
These pad bits can be used to hold things that would otherwise spill off the end forcing us to round or truncate.

Now lets do that last problem again:
01,0 / 10,0 + 01,0 / 10,0 = ??
Breaking it down, 01,0 / 10,0 = 00,1
So, 00,1 + 00,1 = 01,1
Eureka, we can now complete simple divisions without a loss of accuracy for this case.

Moving Beyond:
Unfortunately, we will need more bits to perform other potential calculations. Consider 01 / 11 (in decimal 1/3). Clearly our single bit pad is no longer sufficient. In fact, we will not be able to achieve an exact representation of this value in our system! As you know, 1/3 is a reapeating fraction that looks like 0.333... (too bad I can't draw a venculum in a forum, oh well). Anyways, with our binary system we can only represent decimal values 1/2, 1/4, 1/8, 1/16, etc... Only by adding up an infinite number of ever smaller values such as these could we achieve an exact representation of the value 0.333... So, with only 1 pad bit, we could only do the decimal values 0.0, 0.5. With 2 pad bits, we could do 0.0, 0.25, 0.50, 0.75. With 3 pad bits, we could do 0.0, 0.125, 0.250, 0.375, 0.50, 0.675, 0.750, 0.875.

Hope this helps. Dan writes very clear explanations and perhaps he will clarify for my perhaps less than entirely elucidating explanation.

BTW, if topics such as this interest you, I might suggest a course on numerical analysis. I find the subject very rewarding.
lfeagan
 
Posts: 1
Joined: Thu Jul 13, 2006 3:21 pm
Location: Lawrence, KS

Postby lavrye » Fri Jul 14, 2006 9:57 am

lfeagan wrote:I am not Dan. However, I can take on this part of the question adequately.

Mike Derrick wrote:Are there reasons for DAWs to be recording and editing at 24bits?
Or could we record in DAWs at 16bits and achieve the exact same sonic quality?

~ mike


Very fine explanation.

Many people in audio do not want to study numerical analysis. So let me try to explain things in some intuitive way, to give the less technical users of the gear some “intuitive feel” with no math involved.

Let’s think of the DAW as a page, say a “scratch pad”, such as one uses to write on with a pen. Lets think of a single track as an object to be drawn on the page. Obviously, the track must fit on the page. But should the page be much bigger then the track?

The track is analogous to say 24 bits. The page is analogous to the bits in the DAW, so obviously we are asking if the DAW needs to have more bits then a single track…

Lets say that moving the object on the scratch pad to the right means attenuation of the signal, and moving to the left means amplifying the signal. We can immediately see that the scratch pad should be bigger (extended to the left and to the right) relative to the object.

Next, let us take a number of objects (tracks) and add them together. That by itself is going to make a bigger object, thus require a bigger scratch pad.

More space is required if we shift some object to the left, or to the right, before adding them.. Even more space (bits) are called for when doing both, amplifying some tracks while attenuating other tracks. You do not want to have a part of the picture “spill off the page” because once it is gone, it is gone for ever. You do not want to run out of bits on the left or the right, because some information will be lost.

And adding tracks is not the only reason why the object grows (more bits). Doing various signal processing (such as EQ) can cause it as well.

Of course, once all is done, we may have an outcome where there are too many bits for our scratch pad. The final digital format has 16-24 bits (96-144dB dynamic range). We can "do away" with the lower bits that we do not need.

We end up taking the important bits (such as the 16 or 24 most significant bits), and BTW one may need dither for the final bit reduction, but that is another subject...

Clearly, we need more space on the most significant bits. If you “run out” on the most significant bits side, the distortions are horrible. The signal is badly clipped and the music is not even recognizable.

But while more subtle, we also want more processing bits on the least significant side. Example: Say you have a 20 bit material and 24 bit DAW. You have enough “space” to attenuate the signal by 4 bits (24dB). Say you wanted to attenuate by 6 bits (36dB attenuation). That new attenuated track is 2 bits short. So far we are “OK” because those missing bits are way below the level we can hear. They are in fact bits 25 and 26 (representing dynamic range of 156dB). The signal is now weaker by 36dB so the level of the lower bits is way below our hearing capability.

But say we change our mind, and wish to bring the track back to “the way it was” before attenuation. When you amplify back by 36dB, the 2 lower bits are missing. We lost them for good. When the “object” ran out of room on the page, and we can not ever know what it was.

So other then keeping track of objects (tracks) with respect to the “scratch pad space”, which takes some “study”, many of the DAW makers decided to provide a huge scratch pad. Some are at 32 bit floating, others are at 48 bits or even 64 bits. The “original music” is “parked sort of in the middle” of all that space, giving the DAW operator a lot of room to do all sorts "moving things around" - mixing tracks with or without, EQ boosts, track gain, attenuation and other processing. At the end, we take up to 24 most significant bits (or 16 in the case of CD)…. Most often, we align the most significant bit of the final mix with the most significant bit of the release format, thus "lopping off" the extra space to the right of the most significant bit, and to the left of the least significant bit. There is no need to include huge white space to convey for an object occupying the center of the page...

That is why we need a lot of processing bits. The original tracks have no more the 24 bits (and some lower bits are useless). The final outcome has no more the 24 bits (and some lower bits are useless). But we need more bits available for the processing that takes place between the recording and the final release.

Regards
Dan Lavry
lavrye
Site Admin
 
Posts: 522
Joined: Sun May 15, 2005 10:36 am
Location: Poulsbo, WA

Postby Mike Derrick » Wed Aug 30, 2006 8:22 pm

thanks, good explanations, that all makes sense.

~ mike
Mike Derrick
Mike Derrick
 
Posts: 17
Joined: Wed Jun 29, 2005 3:24 am
Location: White Rock, BC

Next

Return to Audio Technology

Who is online

Users browsing this forum: No registered users and 1 guest

cron