BigRay
New member
from www.24bitfaq.org :
One of the most important concepts in recording is called dynamic range, and is measured in dB. The dynamic range of a recording is the difference between its loudest point and its quietest point.
To elaborate further, each bit gives us the ability to represent about 6dB of dynamic range. A passage that is 6dB louder than another passage is said to be twice as loud as the other passage. In the 4-bit example, we theoretically have 24dB of dynamic range that can be used. But what if recording doesn’t take advantage of all that dynamic range? What if the recording never peaks beyond 6dB of its maximum possible limit? In this case, the recording would only take advantage of 3 of what we call the least significant (or left-most) bits, meaning 18dB of dynamic range. 16-bit recordings are capable of a theoretical maximum limit of 96dB of dynamic range. This means that a single wave could have up to 65536 discrete values that can be used to represent it. But if the same wave recorded at 16-bit peaks at 48dB below its maximum possible limit, then there would only be 256 discrete values that can be used to represent it, taking advantage of only 8 of the least significant bits. The 8 most significant bits would contain no information whatsoever, and would remain unused. In the case of 24-bit recording, you’d have a maximum of 16,777,216 values to choose from, and in the case of a wave peaking at 48dB below its maximum possible limit, the wave would still have 65536 possible discrete amplitude values that could be used to represent it.
Now, have you ever heard any of the early 8-bit computer recordings that floated around in the early days of home computers? Didn’t they sound just awful? I mean, you were impressed because you had a snippet of music that you could recognize playing from your computer, but you wouldn’t want to listen to it for more than a minute or two. I personally remember playing back an 8-bit digitized 5 second snippet of Van Halen’s rendition of the Kinks’ “You Really Got Me” over and over again on my Atari 800 until I couldn’t take it anymore. The thrill soon had me building an 8-bit digitizing device with a microphone input jack and a connector for the joystick port. Ah, those were the days… but I digress.
Perhaps many are more familiar with 8-bit audio from real-time internet sources like RealAudio. It’s good enough for speech recognition, but leaves all too much to be desired for music.
Now here’s the kicker in the16-bit realm. While the volume level of a recorded low-E note struck on an acoustic guitar might take advantage all 16 available bits (for instance, where the peak on the DAT deck reaches 0dB), the squeak of the fingers on the string, the scratch of the pick hitting the string, and the 5 or 10 audible harmonic overtones of that note may never reach a point beyond 48dB shy of the 96dB maximum. Yes, all of these additional by-products of that low-E string that make the guitar sound alive and compelling receive all of the fidelity of that scratchy, distorted, computerized sound of that 8-bit sample from long ago. And as the basic low-E note fades out, it too gets the same butcher treatment from the ever decreasing number of discrete amplitude values. Yikes!
Now record with a 24-bit word length, and put the CD quality back into those string squeaks, pick scratches, and overtones. With 24 bits, you can hear the clarity of the cymbals decaying as they keep ringing smoothly down to complete silence. The little low-level smack of the bass pedal head hitting the bass drum skin that sounded barely like a small click before (if audible at all) now sounds like a smack, complete with its own smoothly reverberating decay. Even the low-level acoustical reflection from the wall behind the band now contributes to the experience with added detail and a sense of ambience, not simply low-level distortion. Finally, because of this improvement, no more does the recordist have to risk overloading and clipping the recording in effort to achieve maximum fidelity. Levels can be set conservatively with the assurance that a high degree of fidelity is maintained.
One of the most important concepts in recording is called dynamic range, and is measured in dB. The dynamic range of a recording is the difference between its loudest point and its quietest point.
To elaborate further, each bit gives us the ability to represent about 6dB of dynamic range. A passage that is 6dB louder than another passage is said to be twice as loud as the other passage. In the 4-bit example, we theoretically have 24dB of dynamic range that can be used. But what if recording doesn’t take advantage of all that dynamic range? What if the recording never peaks beyond 6dB of its maximum possible limit? In this case, the recording would only take advantage of 3 of what we call the least significant (or left-most) bits, meaning 18dB of dynamic range. 16-bit recordings are capable of a theoretical maximum limit of 96dB of dynamic range. This means that a single wave could have up to 65536 discrete values that can be used to represent it. But if the same wave recorded at 16-bit peaks at 48dB below its maximum possible limit, then there would only be 256 discrete values that can be used to represent it, taking advantage of only 8 of the least significant bits. The 8 most significant bits would contain no information whatsoever, and would remain unused. In the case of 24-bit recording, you’d have a maximum of 16,777,216 values to choose from, and in the case of a wave peaking at 48dB below its maximum possible limit, the wave would still have 65536 possible discrete amplitude values that could be used to represent it.
Now, have you ever heard any of the early 8-bit computer recordings that floated around in the early days of home computers? Didn’t they sound just awful? I mean, you were impressed because you had a snippet of music that you could recognize playing from your computer, but you wouldn’t want to listen to it for more than a minute or two. I personally remember playing back an 8-bit digitized 5 second snippet of Van Halen’s rendition of the Kinks’ “You Really Got Me” over and over again on my Atari 800 until I couldn’t take it anymore. The thrill soon had me building an 8-bit digitizing device with a microphone input jack and a connector for the joystick port. Ah, those were the days… but I digress.
Perhaps many are more familiar with 8-bit audio from real-time internet sources like RealAudio. It’s good enough for speech recognition, but leaves all too much to be desired for music.
Now here’s the kicker in the16-bit realm. While the volume level of a recorded low-E note struck on an acoustic guitar might take advantage all 16 available bits (for instance, where the peak on the DAT deck reaches 0dB), the squeak of the fingers on the string, the scratch of the pick hitting the string, and the 5 or 10 audible harmonic overtones of that note may never reach a point beyond 48dB shy of the 96dB maximum. Yes, all of these additional by-products of that low-E string that make the guitar sound alive and compelling receive all of the fidelity of that scratchy, distorted, computerized sound of that 8-bit sample from long ago. And as the basic low-E note fades out, it too gets the same butcher treatment from the ever decreasing number of discrete amplitude values. Yikes!
Now record with a 24-bit word length, and put the CD quality back into those string squeaks, pick scratches, and overtones. With 24 bits, you can hear the clarity of the cymbals decaying as they keep ringing smoothly down to complete silence. The little low-level smack of the bass pedal head hitting the bass drum skin that sounded barely like a small click before (if audible at all) now sounds like a smack, complete with its own smoothly reverberating decay. Even the low-level acoustical reflection from the wall behind the band now contributes to the experience with added detail and a sense of ambience, not simply low-level distortion. Finally, because of this improvement, no more does the recordist have to risk overloading and clipping the recording in effort to achieve maximum fidelity. Levels can be set conservatively with the assurance that a high degree of fidelity is maintained.
I'm guessing it has something to do with the oversampling architecture used to achieve normal performance levels in the first place.. but theoretically, shouldn't it be possible if you don't do oversampling?
