Conference cookie::notes$archive:cd_v1

    I'm sure there is another note on this subject, but rather than
    trying to find it, I will describe it a bit here.
    The audio signal (which is of course a varying voltage) is converted
    into 16-bit numbers, that is, 44.1K times/second, the voltage is
    sampled and a number which represents this voltage is stored.  It
    is assumed (by the analog-to-digital converter) that no frequency
    components higher than 22.05 kHz exist in the signal.  Therefore,
    an analog low-pass filter must be used when encoding.  If these
    higher components do get to the converter, aliasing occurs, which
    ultimately results in harmonic distortion.  A great example of aliasing
    is the rotating wagon wheel in a movie.  The spokes sometimes seem
    to be turning backwards, or not at all.  This is because the
    projector's frame rate is limited.  The frame rate of a movie projector
    is analagous to the sampling rate of the digital encoder.  If too
    high a frequency is fed in, (i.e., if the wagon wheel is turning
    fast enough), we hear that frequency component in the wrong place
    (we see the wagon wheel turning at a lower rate than it really is
    turning).
    With that clumsy description of aliasing out of the way...
    The 16-bit numbers are stored on the disc as one long bit-stream,
    starting in the center of the disc and going out (a spiral, like
    LPs).  Encoded along with the raw data are track and index numbers,
    time stamps, etc.  The coding is fairly complex, because there has
    to be extensive error-checking capability in reading the disc. 
    I find it amazing that it works at all, considering that the laser
    has to focus and remain focused on a micron-wide track of pits as
    the disc spins at 200 to 500 rpm, and that the reader has to follow
    the track as it wobbles back and forth many track widths because
    of a mis-centered hole, etc.
    Well, if the CD player is successful in getting the 16-bit words
    decoded, a digital-to-analog converter produces a varying voltage
    output.  This output needs to be low-pass filtered to remove
    high-frequency components produced in the conversion process.  Some
    players 'over-sample' the data, which simply means that the DAC
    samples at 2x or 4x the regular rate of 44.1 kHz.  This makes it
    'easier' to filter the output.  That is, the response of the low-pass
    filter required does not have to be as steep.  Oversampling buys
    you a bigger "don't care" region in the frequency response, so you
    can use lower-order (less-steep) filters that won't affect the phase
    response as much (a messy subject in itself).
      The idea behind CD is simple and has been around for many years.
    But the mechanical difficulties in building readers and developing
    mass-production facilites prevented CDs from surfacing until this
    decade.
      Please forgive my rather sloppy descriptions, I wrote this fast.
    Let me know what's unclear (or wrong).
    
    		Steve

    	Ok, so the 16-bit number actually represents an analog voltage
    that in turn is re-analog-ed and fed to the speaker to determine
    the change in distance from the magnet to the core which over time
    changes the density of the air so we can hear it??  Great!  If that's
    more or less true, that is very convenient.  If you didn't understand
    any of what I just "summed-up" (like me), I'll re-write it.  I was
    under the impression that it was much more complicated and higher-level
    "commands" in the "language" of music were used (i.e. waveform,
    envelope, pitch, etc.)  Thank you for your input!
    
    				Mike Bell

    Yep, that's about the size of it.  Of course its the details of
    the storage and conversion that make it complicated.  The phone
    company does use other more complex methods, like linear prediction
    coding (LPC).  You can store (or transmit) speech at about 1/100
    the bandwidth (bits/sec) of direct digital coding, but this would
    not be acceptable for music.  You could store about a week's worth
    of continuous speech on a CD!  (But it would sound like the telephone
    does.)
    				Steve (signal processing enthusiast)
    
    

    	Ok, so if I were to...  Actually, let me tell you what I am
    in the process of planning.  I want to program an Amiga 2000 to
    accept an outside signal and then process the sound as a pedal does
    (for guitar or voice) in either real-time or post-recorded time.
    The latter involves (for me) digital encoding.  But to incorporate
    effects or enhancements, it would seem necessary to re-convert to 
    analog.  Does anyone know any algorithms for digital effects?
    
    				Mike Bell

    re.4	Oh Yeah, we know ALL about THAT stuff, s'easy.  You
    don't HAVE TO do the processing at the analogue level, in fact its
    generally more flexible/economic to crunch numbers these days.
    You should probably check out the computer music conference too,
    there are lots of effects boxes ('THEY' call the FX boxes) on the
    market for munchin' up music and making it into "marketable, modern
    sounds".  NOVA::COMMUSIC for details, enjoy.

About oversampling...

As I underdstand it, the DAC does *not* actually sample at 2x or 4x the 
44.1 KHz rate - after all, how could it?  There's no 16-bit data there 
to sample at a higher rate.

What oversampling actually does is to speculate about what the signal 
level *probably* was at a given time.  The following graphs should give 
you some idea.

o   The first graph is a clumsy representation of an analog waveform
    segment. 

o   The second is this waveform superimposed on a 44.1 KHZ DAC's output.
    The stars mark the points at which the ADC changed its digital
    output to reflect what it heard in the original audio signal -
    consequently, they also mark the points at which the DAC changes its
    analog output to correspond with the digital information.  The
    digital-to-analog conversion is quite accurate; but because the 
    samples are gross, the area under the curve doesn't correspond to
    the original analog waveform as closely as desired.  As this signal
    is fed to the analog section of the player, it is again low-pass
    filtered; but the filtering cannot extract all the high-frequency
    components without altering the low-frequency stuff as well.  There
    is audible distortion, depending on just when during the sampling
    cycle the audio began to change and on how good the filters are. 

o   The third is the same DAC's output with 2x oversampling added.  The
    plus signs are the points at which the DAC has used the immediate
    past history of the digital information to guess at where the audio
    signal really was halfway between the last two samples it just read.
    Since the audio input is low-pass filtered, there aren't any spikes
    of high-frequency signal to be missed - the waveform presents a
    pretty smooth picture to the "eye" of the ADC and consequenty to the 
    DAC.  By inserting these output changes based on guesswork, the
    system can produce a far more accurate representation of what the 
    original input was, and there is less need to do extensive filtering.


       Audio               Audio with digital      With oversampling

      _______                  *___*___.               *_+_*_+_.
     /       \                /|      \|              /|      \|
    /         \              / |       *___.         +_|       *_.
   /           \            /  |        \  |        /|          \|
  /             \          *___|         \ |       *_|           +_.
 /               \        /|              \|      /|              \|
/                 \      / |               *     +_|               *

- Dick

    
    
    Hi, Mike
    
      Part of the information you need may be found in note 549.0
    
    Also you may find some information on Amiga in the Amiga notefile,
    BOMBE::AMIGA.   Sorry, I forgot how to add this reply with the KP7
    feature.
    
    If you need more information you can try by sending me Vaxmail.
    
    Don't forget that if you want to store audio information in 
    digital format (on magnetic medium) that each (full) CD have
    almost 20 GigaBits (about six RA81).  That's a hell of a lot
    of diskettes.  I beleive the Amiga is fast enough to produce
    a very good quality sound but to store/retreive that information?
    
    Rgds,  Vince.

re:-1
     Wait a sec, 20Gb?  I think it might be only about 3 or 4 gigabits,
    or 500 Mbytes.  Still a little more than an RA81.
    		Steve

     16 bits per sample
   x 44.1e3 samples per second
-----------
 705,600  bits per second
   x 60   seconds per minute
-----------
42,336,000  bits per minute
   x 75     minutes per disc
-----------
3,175,200,000  bits per disc per channel
   x 2        for stereo
--------------
6,350,400,000 bits per disc
   / 8        bits per byte
---------------
793,800,000   bytes per disc

and that's just for the audible program material - indexing, id, and overhead
amount to ?20%, meaning data storgae capacity is almost 1 GByte.

    
    According to the info I have, Audio is only 32,7% of the disk 
    capacity.  Almost one third.
    
    41.8% is the EFM bits (8 to 14 modulation).  To process a word
    (16 bits, the player has to read 28 bits + mergin bits (17.4%)
    Synchronization (4%), subcode (1.4%) and finaly parity (2.7%)
    
    Wich give 19,918,878,200 bits per disc. Albeit only 6.5 billion
    are strictly audio. (on a standard 74 min. disc)
    
    The eight to 14 modulation bits uses the largest slice of the disk
    in percent but not physically since "0"s and "1"s do not take
    the same space to write.  According to the book "Despite its longer
    words, EFM is actually 25% more efficient than recording done 
    using an 8-bit format...
    
    The "sector" of a disc is as such:
    
    Sync    27 bits
    Subcode  8 bits
    Audio   96 bits
    Parity  32 bits
    Audio   96 bits
    Parity  32 bits
    
       Total = 288 bits + EFM = 588 bits
            Note: EFM is not used on all fields, explaining why
                  % may not match in strict bits numbers.
    
To come back to our Amiga, the information has to come to almost
    1.5 mega bits per second to reproduce only audio.  3 times more
    for the whole thing.
    
    Rgds,   Vince.