Connect with us

Nice criticism of modern digital audio

Discussion in 'Electronic Design' started by martin griffith, Nov 26, 2003.

Scroll to continue with content
  1. I've been a long time meddler in (pro) audio, and over the last few
    years I've been getting irritated by the stupid specmanship on audio

    My annoyance really started a year or so ago, when I was mucking
    around with TI's PGA2310, a digitally controlled audio fader, with
    the late Baz Porter, on a digitally controlled analogue parametric
    equaliser. The PGA is a reasoable little IC for audio, but the stupid
    way it was designed..... 0.5dB steps down to -95.5dB.

    Who on earth would want to fade something that low. Most mixing
    console faders tend to cut to infinity at -60dB... dumb, why not finer
    steps, down to -60. It just seems so sensible. When I was a spotty
    lad, -60dB was a tolerable level for distortion in a system!

    On, ( a very busy NG,but very little wheat in the chaff
    for me ) there was a thread about 384KHz PCM audio systems, and one
    reply really hit the mark
    I think it's this:
    Message-ID: <>
    but I'm not too good on googling, so I post the whole message here,

    Quote "

    My name is Dan Lavry, and I hope you do not mind that I barge in. I do
    want to say a few words on that 192Khz. As a designer, I have long
    realized that faster sampling means less accuracy. You can get 1 bit
    at a GHz, 8 bits at 100MHz, 12 bits at 30Mhz…. Nearly 24 bits at a few
    Hz. Well, I am not suggesting to sample at 10Hz. We need to cover the
    audio bandwidth. Going too slow is bad. Going too fast is also bad.
    Let's face it, you want to say charge a cap, or settle an amp, and it
    does dot happen at zero time. Nothing does. Weather you get there
    exponentially or you have that step with some ringing… whatever it is,
    if you wait till it settled you get it closer and more accurate. These
    are practical limitations. The thing that gets me is that all the EE's
    I talk to make jokes about 192Khz – a common one is "my dog can not
    hear it". But many in the recording and mastering side are a part of a
    different crowd. There seem to be a gap between the EE designer types
    and the recording types. I am not suggesting that EE's are better or
    worst. There have been times when the good ear is right, and the EE
    learns new things. But there should be times such as right here and
    now for the recording types to be open-minded. And folks, that 192KHz
    is a crock.

    I have wondered about how is it possible that a whole industry can go
    in the wrong direction. After all, Few musical instruments will under
    special cases yield a -50dB harmonic at 50Khz, and almost nothing at
    100KHz. Experiments showing that special case required special mics
    and gear. Most recording mics will not pick up a thing at 60khz… The
    speakers will not play it, the ear will not hear it, no matter how big
    the ego is!

    So the sounds are negligible at those high frequencies, the mics will
    not pick them, the speaker will not play it, and you can not hear it.
    So how can such a crock take place? And let me assure you the folks
    that made it happen at the semiconductor houses and workstation houses
    did not include the EE department... I think it was and is about
    making money. I talked to many engineers that stated privately that
    they are not talking because they need their jobs! Other "did not want
    to make waves", "stir the pot"…

    So is everyone trying to cheat? Of course not. It started with 48Khz
    better than 44.1, which is true. Also, while a bit excessive, 88.2 and
    96KHz are in some cases a better compromise than 44.1. That is due to
    the old FIR pre shooting (pre echo) that can happen under some cases
    with a lot of processing. By the time you are at 60KHz it is so far
    down you will not hear it, so 70Khz is a pretty reasonable place to
    be. So after we have improved some as we went higher, the expectation
    was than to continue upwards. Where does it stop? 384? Why not 1Ghz?
    Oh well, the will be pretty bad, will it not?

    But there is more to that than that "trendy upwards worked before so
    lets continue". The common sense tells you that more is better. For
    example, more pixels yield more detail. More bits gives better sample
    accuracy. So if we take that analog wave, made out infinite points
    connecting into a line, will we not benefit from a better "tracking"
    of the wave? Well here is the news, and it is pretty old stuff, though
    it may not be easy to grasp with simple common sense. More pixels for
    better picture- yes! More bits for better accuracy (assuming
    theoretical case - no noise) – yes! More sample density (higher sample
    rate) for more accuracy –NO! This is the beauty of the sampling
    theory. Nyquist did not say: We take more points and we will get
    closer approximation. What he said – and it is a FUNDUMENTAL THEORY,
    is that once we agree to deal with a limited bandwidth (called the
    Nyquist bandwidth), all we need to do is sample at a tiny amount
    greater than twice that bandwidth. This will yield 100% of the
    waveform information in the data stream. We may need to filter a
    signal (anti alaising) to make sure we do not have energy over
    Nyquist, but than we are home free. Taking 4 times as many samples
    does not yield 400% of the information. You can only have 100%! How do
    you retrieve the information? You use a filter and it connects the
    sample points in such a way that that you get the original wave shape.
    A filter does not connect the dots (sample points) with straight
    lines, or parabola… It recreates the original wave! You do not need to
    help things with extra point in between. It buys you nothing!

    I also see a lot of confusion regarding that Nyquist, upsampling,
    oversampling, gradual filters… Some folks think that a 96KHz AD will
    require a sharper anti aliasing filter than say 192KHz. This is
    typically very wrong. The 96KHz or 192KHz AD refers to the OUTPUT RATE
    of the converter. The antialias requirement is determined by the INPUT
    SAMPLING RATE which is usually way beyond 192Khz. This days, most
    modern AD's are running at input sampling rates of 3-12Mhz!!! DSD is
    64fs and many mulibit IC's run even faster… So even with 50Khz audio,
    Nyquist is so high the a gradual 3 pole will yield 120dB at the input
    rate Nyquist. The outcome of the high rate modulator (input) is than
    down-sampled to whatever – 44.1, 96, 192… Of course, when the sales
    guys try to stick you with it they tell you need more bandwidth, but
    they also ALL I saw regarding semiconductor and gear makers alike:
    specs for 192KHz device are with A weighting – which states
    (indirectly?) that you do not even need to measure flat to 20Khz. So
    is it a crock? It is!

    Theoretically, there is "no harm" in more points, and there is "no
    added good". But as I pointed out, faster is less accurate! And yes,
    you double the sample rate and the processing power requirement, and
    so is the file size. These are serious draw backs! Don't you say you
    do not care about file size: The DVD audio has 12 to 1 compression
    (Dolby AC), We do not even get near a 1 to 1, and we want to push it

    I realize that with all that reasoning and science and engineering,
    someone is going to tell me that they hear it and like it. In fact,
    someone told me that they still hear that high frequency in the
    44.1KHz CD. I will not dignify that impossibility. If you hear some
    distortion you like on the 44.1K CD, you did not need to go any faster
    than 44.1KHz to generate it. I am not arguing against controlled
    distortions (such as tube sound and what not). If you like it is fine.
    It may be artistic decision. If you feel like you need to go to 1Mhz
    than down to like it, fine. I think you are letting the gear control
    you instead of the other way around, but fine! Just as long as I get
    you to realize that you can get those distortions with a 44Khz… And we
    do not all need to double the file size and processing power, and buy
    new gear that is less accurate.

    192 is a crock! 382 is a super crock! 88.2/96Khz is a bit excessive,
    but not too far from a good rate. I too can glue a faster IC on the
    board and make more money. My 192 DA prototype is not bad, but the
    96KHz bits it by a lot.
    Anyone telling you that more points will give better aproximation is
    lacking lacking some know how.

    Thanks for your patience.

    Dan Lavry
    Lavry Engineering
    end quote

    "When all else fails, digitize everything, use fiber optic cable and enter a
    whole new realm of problems."
    <Found on the Rane tech page>
  2. Why is it stupid? You can still make the step from -60 to
    infinity in the software that controls it. Perhaps there
    are other applications (other than audio, but within audio
    frequencies) where the -95.5db is welcome. Do you really
    need smaller steps than 0.5 db? You can put 2 in series ;)

  3. Martin's application was a parametric equaliser rather than a straight
    I guess the eq was quite sensitive to the fader step size.

  4. Ban

    Ban Guest

    martin griffith wrote:
    || I've been a long time meddler in (pro) audio, and over the last few
    || years I've been getting irritated by the stupid specmanship on audio
    || IC's.
    || My annoyance really started a year or so ago, when I was mucking
    || around with TI's PGA2310, a digitally controlled audio fader, with
    || the late Baz Porter,

    I love the neteq, sad to hear Mr. Porter has passed away. Do you still
    continue this work and do you need some support? I'm ready to collaborate!
    Drop me a mail in case.

    snipped quote
    || "When all else fails, digitize everything, use fiber optic cable and
    || enter a whole new realm of problems."
    || <Found on the Rane tech page>

    My feeling is that despite a few exaggerated specs digital audio has finally
    reached an acceptable quality level, allowing all-digital amps and a
    reproduction superiour to analog gear. This is mainly due to 24bit and 88.2
    or 96kHz sampling rate. Hopefully the high-end SACD will take off within the
    next years in contrast to this awful MP3 hype, the kids are spreading.
    As much as you don't need the accuracy of a 99.0 to 99.5dB attenuation or
    the 24 vs. 16bit, it doesn't cost anything extra and it is nice to have it
    for maybe some unexpected use.

    In an all digital amp also the volume pot is realized digitally and with low
    settings you would usually loose information because if you have speakers
    with high sensitivity you might need to adjust your volume pot to -60dB and
    10bits are gone, so the remaining 14 will be better than only 6 in a 16bit
    system, and as you see here we better had 32bits available.

    With high sampling rates I do agree, but again a counter example: a constant
    directivity horn requires an equalisation of +24dB at 20kHz of shelving
    6dB/Octave. Now with a 44.1 sampling rate you will have problems to
    implement this digitally and even at 96kHz it is done only approximatly. So
    there is always a pro and a con.
    Todays progress in DSP will get boosted with newer standards and we have to
    create more demanding situations to keep the engine running, as Bill Gates
    does it with Windows.
    Keep on rocking...
  5. Ian Stirling

    Ian Stirling Guest

    A volume control can need that dynamic range.
    Say 110dB on the top end, that's only 50db on the bottom.
    Distortion is different.
    I want to be able to adjust volume from somewhere around 90db, all the way
    down to almost audible in a quiet room (say 5-10db).
  6. I agree.
    There are not that many applications for intelligence as one might
    think. (quote from Dilbert)

    Raymund Hofmann
  7. Tim Shoppa

    Tim Shoppa Guest

    The PGA2310 is a part. The designer of the system including the part
    is responsible for using the part appropriately.

    You may as well complain that a resistor manufacturer would let you put
    two resistors together and make a -95.5dB divider, and that if they
    were a responsible resistor company they wouldn't sell that combination :).

  8. I know there's a smiley there, but still . . .

    That will give you a fader with a theoretical range of 0 to -191 dB, but
    still in 0.5 dB steps. Decibels in series add, they don't multiply. To
    get finer steps, you need a fader with finer steps.

    I believe Martin's point is that the chip has includes all the circuitry
    to give 192 steps, so why not space them over a reasonable range instead
    of wasting them to get a "better" spec?
  9. Dbowey

    Dbowey Guest

    martin posted, among other things:
    That coming from an Engineer appears shallow; why use finer steps when one
    cannot perceive the change of even 0.5 db? I imagine the ability to attenuate
    to -95.5dB enables it's use as a mute device.

    The rest of your post is mostly rant. Try to focus a bit.

  10. Most of the consoles I've worked with don't have 0 as the top of the faders.
    Rather, they go from somewhere around +15dB down to somewhere around -60dB
    and thence -infinity.
  11. Realistically, you need a -inf setting (just like you get from a real pot) -
    you want to be able to turn something OFF, not just down to the point where
    it's theoretically inaudible, because you might be adding a bunch of
    nominally "off" channels together. But if -inf is really -120dB (or
    whatever the i/o bleedthrough amounts to), it doesn't follow that the next
    notch up has to be one notch higher than -120. It is fine to go from "the
    lowest thing you need to fade to" down to "off".

    If I were designing a digital audio attenuator (as opposed to a digital
    potentiometer, which is the OP's problem), I'd want it to have an uneven
    scale: fine gradations around 0, down to about -40, and then coarser
    gradations down to -inf. That would correspond to how pro mixing desks seem
    to work: the physical distance between -30 and -40 on one desk I regularly
    work is about the same as the distance between 0 and +3, for instance.

    0.5dB is a noticeable jump if you're trying to do a mix. (I know, the OP
    was about a parametric EQ, not a channel fader.) I guess that just as with
    pitch, it's easier to tell slight level changes in comparison to a fixed
    reference than by themselves. If you need to get the lead vocal just a tad
    hotter than the guitar, 0.5dB gradations are awkwardly coarse - workable but
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day