16 bit to 24 bit.....What’s the other 8 bits?

Discussion in 'Audio Hardware' started by timztunz, Oct 15, 2017.

Thread Status:
Not open for further replies.
  1. anorak2

    anorak2 Forum Resident

    Location:
    Berlin, Germany
    The analogy is wrong though. In the video realm, technology hasn't reached the resolution of human perception yet. So adding pixels still makes sense, until it is reached. In the audio realm, the resolution of human perception has been reached with 44/16 for all practical purposes. We are there. Adding more bits beyond that is pointless.
     
    GetHappy!!, tmtomh, Robert C and 3 others like this.
  2. missan

    missan Forum Resident

    Location:
    Stockholm
    Yes, that is the voice of reason. But if we think 24/192 will sound better, this is what it will do. So here we have the advantage in a nutshell. Personally I normally use 24/88 when recording records, so I´m guilty of this thinking.
     
  3. Grant

    Grant Life is a rock, but the radio rolled me!

    Why feel guilty?
     
  4. Higlander

    Higlander Well-Known Member

    Location:
    Florida, Central
    I think he means he is guilty of thinking something is better based on numbers and not reality.
    Maybe I am wrong.
     
  5. missan

    missan Forum Resident

    Location:
    Stockholm
    Yes. But the reality is it will sound better to most if they think it will sound better. It´s not so people are delusional, it will actually sound better. That is how it works, at least for 50% of us. It´s not an imagination as some think, it´s real. So when people are saying 24bits sound better than 16bits, we should just accept that.
     
    Higlander likes this.
  6. missan

    missan Forum Resident

    Location:
    Stockholm
    I don´t feel guilty, I´m just human. ;)
     
  7. Kyhl

    Kyhl On break

    Location:
    Savage
    16 bit padded to 24 bit raises an interesting question that maybe someone with more knowledge can answer.
    This is where things break for me. If we pad that max 16 bit number into a 24 bit container as 0111 1111 1111 1111 0000 0000, wouldn't this create a lower maximum output from the DAC than say, 0111 1111 1111 1111 1111 1111, the maximum 24 bit output?

    Is the padding ones, 1111 1111, instead of zeros?
     
  8. Schoolmaster Bones

    Schoolmaster Bones Poe's Lawyer

    Location:
    ‎The Midwest
    SandAndGlass and Higlander like this.
  9. Higlander

    Higlander Well-Known Member

    Location:
    Florida, Central
    It starts out sort of okay, then goes off the deep end into all the wrong things I see repeated on audio forums...Sad:sigh:
     
  10. krlpuretone

    krlpuretone Forum Resident

    Location:
    Grantham, NH
    Space is cheap these days...
     
  11. anorak2

    anorak2 Forum Resident

    Location:
    Berlin, Germany
    I'd consider a software that implements "upsampling" simply by padding bits wrong. It has to multiply the old value by a given factor, and if also the sampling range is to be changed it has to interpolate. Any app worth its salt would recompute the new values according to where it "should be" in the new sampling rate. Of course this doesn't add any information, it's akin to enlarging a photo in an image editing software. Early, crude image editors would do that by simply repeating pixels. That is just as wrong, it has to be done by interpolation.
     
  12. Claude Benshaul

    Claude Benshaul Forum Resident

    The padding exist for two likely reasons. One is the case of a DAC that can only accept specific bit depth and the other is to trick the DAC and prevent it from up-sampling to its native bit depth.
     
    tmtomh likes this.
  13. SandAndGlass

    SandAndGlass Twilight Forum Resident

    I don't know about the visual resolution thing. Apple seems to feel that their "Retina Display" attends to the maximum resolution that the human eye can discern.

    Note, how close to your eyes are you holding your phone? Move your phone further away and then try to discern those tiny little dots.

    [​IMG]

    The important thing to consider about resolution is DISTANCE.

    What the human eye can resolve close up, it cannot resolve at further distances.

    That is why, stores selling large screen TV's, get people as close to the picture as possible, to view the resolution.

    Contrast that to old CRT type TV sets. In a store, they would demonstrate them at a normal viewing distance. You wouldn't think of looking at these older technology TV sets up close, cuz the image would not appear to be very attractive.

    I doubt if you could tell the difference of a 4k TV, when viewed at a distance of 10', vs a the same size 1080i set, viewed at the same distance, as far as resolution goes.

    I have a 4k set and that is my viewing distance. Yes, the picture appears superior, in many ways, due to newer and more improved technologies, than my older 60" 1080i set.

    I think it is fair to say, that we have now achieved the amount of detail in TV technology that allows the maximum resolution of the naked eye, when viewing a large TV at normal viewing distances.

    Since, I can listen to well engineered and mastered resolutions of compressed music and find it to be agreeable with my sensibilities, I don't personally worry to much about hi-res music.
     
  14. darkmass

    darkmass Forum Resident

    While the padding, as you mention, could be with a lower eight bits of 1111 1111 rather than 0000 0000 (and I won't rule out some choosing to actually implement the padding using all '1' bits), for maximum DAC output, look at things this way...

    Look at the "all 1 bits" proposed padding value of:
    0000 0000 0000 0000 1111 1111

    If you add:
    0000 0000 0000 0000 0000 0001 to that,

    you get
    0000 0000 0000 0001 0000 0000, in other words, the minimum 16 bit value of:
    0000 0000 0000 0001 xxxx xxxx

    So whether all '0' bits or all '1' bits are used for padding, the effect on maximum DAC output is quite small.
     
    Kyhl and Vincent Kars like this.
  15. GT40sc

    GT40sc Senior Member

    Location:
    Eugene, Oregon
    When you work in the studio and listen to a band playing live in the room,
    you know that even your best recording is not going to capture the full impact of their performance.

    Analog, 24-bit, DSD, whatever. Doesn't matter. One can make a very good recording in any format,
    but we still cannot capture the full mental and physical experience of being there.

    Make a nice hi-rez mix, and watch it go downhill by the time it gets to the consumer.
    When you get down to 16-bit, you will question so many of your choices
    that were originally made at a higher resolution.

    Reverb, delay, EQ...
    the entire sense of space in the recording is so different that the songs will feel "wrong."

    I know it is the best we can do right now.
    But don't tell me that's all there is...
     
    SandAndGlass likes this.
  16. timztunz

    timztunz Audioista Thread Starter

    Location:
    Texas
    ......and so were born mind altering substances.
     
    SandAndGlass likes this.
  17. Kyhl

    Kyhl On break

    Location:
    Savage
    Are you suggesting that the least significant 8 bits of a 24 bit word are not significant enough to matter if they are zeros or ones?

    I'm fine if that answer is yes. It was just something I was wondering about while thinking of how the padding would work.

    I think of digital working as values less than maximum. Back to my original example where the maximum output of a 24 bit wave would be:
    0111 1111 1111 1111 1111 1111. Now assume that this max value outputs 2v from my DAC.
    Likewise, the DAC also outputs the max 2v from the maximum of a 16 bit file being:
    0111 1111 1111 1111.
    If I pad that 16 bit with zeros to 24 bits it becomes:
    0111 1111 1111 1111 0000 0000.

    Here is the problem, the new padded value will not create a 2v output from the DAC. It will be something less. Is that something less, so minuscule that it does not matter to the listener?
    Or, this could be resolved by padding with ones instead of zeros. So I was curious to know, how does the padding actually work.
     
  18. Ken_McAlinden

    Ken_McAlinden MichiGort Staff

    Location:
    Livonia, MI
    It would not be padding. It would most likely be scaling up so that full scale at 24 bits = full scale at 16 bits with some kind of interpolation filling in the gaps.
     
  19. bryantn3

    bryantn3 New Member

    Location:
    Atlanta, GA, USA
    My point is that at a distance, yes, video has reached the limit of human perception. You can't see the individual pixels in a well rendered, say, video game, on a 4k display. It may not look identical reality, but that's not up to the display, that's up to the art style. Yet without any antialiasing the image will look like a poorly rendered mess. It's because when an unequal amount of pixels are within the source textures and such than are in the display, it looks messy. Antialiasing ups the internal resolution before it's downsampled to the display resolution. Hi-res audio is like really good antialiasing, whereas very highly dithered audio is like a rather off, badly aliased video.
    Believe me, I listen to cd quality audio as well, in fact for most Genesis records, I much prefer the original cd masters to the new mixes AND the 94 remasters, and some of them I even prefer to the vinyl releases, because many of the vinyl releases were too lengthy for one LP, and when they squeeze them on one, it makes the bass response terrible. But I would take a well mastered hi res file over the cd any day. I really wish Nick Davis, or someone would do a flat hi res transfer of the original Genesis masters. I think the demand is there.
    Anyway, I'm getting off on a rabbit path, so I must have run out of things to say.
    Oh yes, and
    this is correct in most cases when audio is up converted. it usually gets dithered (upscaled). Not always the case at the sound card level, but usually it is with software conversion.
     
    SandAndGlass likes this.
  20. bryantn3

    bryantn3 New Member

    Location:
    Atlanta, GA, USA
    Also, as far as sample rates go, I would agree that it's far more important than bit depth, because 44.1 kHz can only render frequencies up to 22,050 Hz, which is well within the range of what someone with good hearing can hear. Also 16 bit can only render 96 db of volume range, while humans with acute hearing can hear about 120 db of volume range. Yes 24 bit goes above the audible resolution, but 16 bit is below the audible resolution for people with good hearing, like me. Also, cats probably think cd quality transfers sound like nothing but a garbled mess, which would explain why mine gets really still and reflective-looking when I play hi res files, and runs from the speakers when I play cd quality audio. :D
     
    Last edited: Oct 18, 2017
  21. SandAndGlass

    SandAndGlass Twilight Forum Resident

    Yes, definitely, if there has been some cross breeding with bats in their genome.

    Yes and no. These figures are the maximum that a human being who has perfect hearing under ideal conditions.

    In the real world...

    I'm in a quiet room, it is almost 11 PM and outside is quiet. I have a split-pack air conditioning unit in my listening room, which I just set from medium fan to quiet fan. Measuring the SPL, I am reading roughly 32-dB. This is the minimum noise floor (note, this is with the dB meter set to "A" rating, which disregards lower frequencies and frequencies above 8k).

    If we subtract the noise floor from the maximum human hearing range of 120-dB, that would leave us with a hearing range of 88-dB.

    If a CD has 96-dB of dynamic range, we have 8-dB to spare.

    What real world music is anyone listening to that even comes close to an 88-dB of dynamic range? Nothing that I am aware of. Disclaimer; knowing the maximum dynamic range of every piece of music, is not in my set of limited skills. Please feel free to correct me on this point.

    Moving the fan speed to medium, and switching to the "C" scale, yields 52.-dB.

    On the "A" scale, it reads around 40-dB. The quietest that a normal house, say in the bedroom at night, would be about 30-dB.

    Another thing to point out, if you are listening to music at loud concert levels of 115-dB for, say fifteen minutes, your ears would not be sensitive enough to pick up anything below 30-40 cycles, until your ears (and your brain processor) have time to reset.

    Same thing with your eyes. If you are in bright daylight approaching a tunnel, all you can see is a large dark spot. Once you leave daylight, then you can see detail inside of the tunnel.

    If I enter a dark movie theater, from being in bright sunlight, I have to stand against the back wall before I can see enough to find a seat. Otherwise, I might end up trying to sit on somebodies lap.
     
    Higlander likes this.
  22. darkmass

    darkmass Forum Resident

    Okay, I'm not about to state whether something is "insignificant" or not, that's an individual decision, and a seemingly vast number of entire threads have been devoted to parsing the difference between "significant" and "insignificant" in one digital audio domain or another. I'm not in the mood to go there. But I'm hoping it's safe to say that the effect of stuffing the eight least significant bits of a 24 bit word with all zeros versus with all ones is an effectively small difference on a two volt 'maximum voltage' out of a DAC.

    Here's the arithmetic (as produced by the Microsoft calculator in Scientific mode on Windows 7 64 Professional...other calculators may vary slightly; for that matter there could be different results if someone is a bit more careful than I have been)...

    Letting the sign bit be "0", for positive, and assuming the maximum positive 24 bit word corresponds to 2.0 volts out of the DAC, we have:
    0111 1111 1111 1111 1111 1111 = 2v*(2^23 - 1)/(2^23 - 1) = 2.0 volts

    If the eight least significant bits are zero-filled rather than one-filled, we have:
    0111 1111 1111 1111 0000 0000 = 2v*((2^23 - 1) - (2^8 - 1))/(2^23 - 1) = 1.999939203255081564793773268911 volts

    So the effect on maximum voltage between zero filling and one filling the eight least significant bits in a 24 bit word is a small one in voltage terms. Changing that to a difference in dB, and if my dB arithmetic is correct, the zero-filled eight least significant bits version produces a result that is 2.7670211604822758017453972330188e-4 dB down from filling the eight least significant bits with ones. In my mind, that is a fairly small difference.
     
    Last edited: Oct 19, 2017
    Kyhl likes this.
  23. harby

    harby Forum Resident

    Location:
    Portland, OR, USA
    PCM audio is stored as a two's complement number, in 16 bit, a number between +32,767 and -32,767 (-32,768 is also possible in binary, but is invalid audio, as the maximum negative magnitude should be the same as the maximum positive. It can be a clipping indicator.).

    A two's complement 16 bit PCM maximum is 7FFF = 0111111111111111
    A two's complement 16 bit PCM minimum is 8001 = 1000000000000001

    In 24 bit audio, the maximum values are instead +/- 8,388,607:

    A two's complement 24 bit PCM maximum is 7FFFFF = 011111111111111111111111
    A two's complement 24 bit PCM minimum is 800001 = 100000000000000000000001

    To convert from 16 bit to 24 bit without inducing quantizing error at the 24th bit, we simply multiply all 16 bit values by 256 (8 bits) and store as 24-bit two's complement. This actually does "pad" both positive and negative values with 0000 0000 bits:

    011111111111111100000000 = 8,388,352
    100000000000000100000000 = -8,388,352


    You will note that this does not reach 24-bit's maximum waveform sample value. That's OK. The signal integrity is preserved and it can be losslessly converted back to 16 bit without dither by truncation or rounding.

    If instead of multiplying by exactly 256, we instead floating-point multiply by (256.0 * 8388607.0 / 8388352.0) = 256.0077822199164, and convert to 24 bit, we get full-scale 0dB, but mathematical inaccuracy (causing noise at the -144dB level - which is inconsequential compared to the 16 bit audio's quantization noise).

    Either way is fine, the former is less difficult for a programmer or floating point hardware to mess up.
     
  24. timztunz

    timztunz Audioista Thread Starter

    Location:
    Texas
    Well I'm certainly glad we've cleared all of that up! And to think that I was hoping for a simple answer.
     
    Catcher10 likes this.
  25. Kyhl

    Kyhl On break

    Location:
    Savage
    I'll be willing to call it insignificant. :D

    In hindsight I could have just ballparked it as 2v - (2v*(255 /8,388,607)) = 1.99993920326v = Close enough.
    or 2v- (2v* ((0111 1111 1111 1111 1111 1111 - 0111 1111 1111 1111 0000 0000)/(0111 1111 1111 1111 1111 1111)).

    Thanks for walking me through it.
     
    Last edited: Oct 19, 2017
    darkmass likes this.
Thread Status:
Not open for further replies.

Share This Page

molar-endocrine