If you didn't brick-wall-filter (anti-alias filter) a recording, would it REALLY sound bad?*

Discussion in 'Audio Hardware' started by head_unit, Mar 29, 2014.

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

    head_unit Senior Member Thread Starter

    Location:
    Los Angeles CA USA
    As an electrical engineer I'm quite familiar with Shannon's sampling theory. For those who aren't, let me crudely recap that when recording digital data in a "PCM/multibit" system, say at 48 kHz,* you are supposed to put in a super steep "brick wall" filter to cut out incoming analog above half the sampling (recording) frequency. This is to prevent aliasing, i.e. a 25 kHz tone would be miscoded as 25 - 24 = 1 kHz. The tradeoff is that the brick wall filters introduce time distortions and allegedly make bad sound.

    What I wonder, though, is-what if you DIDN'T "brick wall"? Would it really truly sound bad?

    I would kinda think there is very little sound power at or above 25 kHz, and that if it did alias, it would be at a pretty low level. So actually, it might not matter much. If most people have trouble hearing above say 16 kHz, that means the aliased frequency would originally have to be at 32 kHz or above to be audible.

    Further:
    1) If you record at 96 kHz, couldn't you dispense with the filters, or use a very shallow filter? My brain offhand tells me that to alias down to 16 kHz, the original tone would have to be at 112 kHz, unless I'm not remembering the math right.

    2) Couldn't you oversample the incoming stream, filter it shallowly, and then downsample it before committing it to storage? At least with today's computing power?

    3) In today's implementations, which 99% actually 1-bit before committing to multi bit storage, how is all this different if at all?

    4) IIRC, the D/A converter is then supposed to have a filter to "smooth out" the output. So another question would be, what if you didn't have this filter either?

    Besides comments, feel free to link to any research or other comments, or information about filterless A/D or D/A converters.

    *48 kHz just to make math easier. Yeah, everything is more tightly constrained for CD @ 44.1 kHz.
     
  2. SBurke

    SBurke Nostalgia Junkie

    Location:
    Philadelphia, PA
    Why is it that introducing a brick-wall filter supposedly results in bad sound? Why would it?

    I've no training or expertise in this area, by any means, but my first thought was to wonder if complaints result from confusing the filter with brick-walled mastering, which is a different thing.
     
    BGLeduc likes this.
  3. TarnishedEars

    TarnishedEars Forum Resident

    Location:
    The Seattle area
    I think that it would be helpful for somebody to explain precisely what the term "brick wall" means to the members of this forum, because he used the term correctly from an engineering point of view. I also was confused as all heck about what the people on the forum meant by the term "brick wall" when I first joined.

    FWIW: On this forum the term "brick wall" is generally understood to refer to some of the extreme forms of dynamic range compression, not brick-wall low-pass filtering.
     
  4. Jim T

    Jim T Forum Resident

    Location:
    Mars
    I know of some early cds I owned that I would have like to thrown against a brickwall. Not quite the same thing, though.
     
  5. Alice Wonder

    Alice Wonder Active Member

    Location:
    Redding, CA
    My understanding - and this may be incorrect - is that the way most ADCs currently work is to oversample, allowing them to use a low-pass filter at a much higher frequency than the target Nyquist. Then they downsample, using digital methods to remove the frequencies above the target Nyqyuist - resulting in a perfect representation of frequencies below target Nyquist without any roll-off that typically happens with the initial analog low-pass.

    Is that correct? Is that how they currently work?

    -=-

    Secondly, is how he used brickwall equivalent to low-pass filter? low pass filter meaning frequencies below are allowed through but higher are not.
     
  6. Vidiot

    Vidiot Now in 4K HDR!

    Location:
    Hollywood, USA
    Read Ken Pohlmann's book Principles of Digital Audio, which goes into the math and the reality pretty well. My memory is that the distortion goes into the whole numbers if you don't filter the frequencies over 22kHz. Read Wikipedia's entry on anti-aliasing filters:

    https://en.wikipedia.org/wiki/Anti-aliasing_filter
     
  7. head_unit

    head_unit Senior Member Thread Starter

    Location:
    Los Angeles CA USA
    Hmm, maybe I gotta dig that out and re-read that. I don't remember a reality based discussion, but it's a big book. I've no doubt you could in theory get big distortion numbers, I just wonder in real recordings in real rooms if it's a big deal.

    I'm also curious about oversampling on the A/D side; I don't recall hearing about that. (Well, you could argue most ADCs oversample in that they run a high frequency 1-bit converter, but that's not really what I mean, I mean something designed to sample at twice the final frequency etc)
     
  8. head_unit

    head_unit Senior Member Thread Starter

    Location:
    Los Angeles CA USA
    Ah, yes, let me explain "brick wall": it means something really hard.
    So if you are talking about dynamic range, it means the dynamics have been compressed so that everything is really loud (e.g. Metallica's Death Magnetic and far too many others).

    In the case of A/D-D/A, it means a very steep filter. In CD, they are trying to let 20 kHz (the supposed upper limit of human hearing, which maybe many cannot hear, but still regarded as a high fidelity benchmark) pass through at 0 dB. But by 22.05 kHz (= 1/2 of the sampling frequency of 44.1 kHz), the input is supposed to be cut something like -80 dB. That equates to roughly -480 decibels per octave!

    Why is that bad? Well, in the analog domain at least, it means time response problems. And at the A/D side you need at least some amount of analog filtering; I don't think you can have pure digital right at the input. Plus, I *think* steep digital filtering comes with unnatural pre-ringing. Can we hear that? Good question.
     
    felix.scerri likes this.
  9. kevintomb

    kevintomb Forum Resident

    The "Funny" thing, many speakers, many recordings, and to a much higher degree, most all male listeners over puberty, can not hear much at all over 14-17khz at best.

    When they mention human hearing, they are factoring in, even very young children etc.

    Adult men, (contrary to what many will report anecdotally) usually can not hear even close to 20khz.
     
    The FRiNgE likes this.
  10. Vidiot

    Vidiot Now in 4K HDR!

    Location:
    Hollywood, USA
    My memory back when I read the first edition around 1988 or so was that with the 22kHz filter, you get distortion around .01% or so; without the filter, you get like 1% or 2% distortion, but it's non-linear. I also seem to recall at an AES demo Sony showing that you also wound up getting weird harmonics and kind of a "heterodyning" effects on test signals. I dunno how bad that would be with actual music, but it was nasty stuff. They wouldn't throw the filter in just for laughs -- it does add to the cost of the circuit, and I'm sure they'd love to leave it out if they could.

    The book does go into anti-aliasing filters very well, as does the link I provided. So the circuit does serve a purpose, and it's not a good idea not to use it.

    That's somewhat true, but: the aliasing effects I heard had all kinds of schmutz -- that's a technical term -- that kind of splattered out here and there as harmonics. And those were low enough that you could hear them all over the place, particularly on a variable sine-wave signal. Again, with music... I dunno how audible it would be. I know that several high-end manufacturers (even in Sony's very esoteric CD players) have provided different kinds of anti-aliasing filters that do steep, graduated, and even analog vs. digital filters, allowing the user to choose which they preferred. I could never hear a gigantic or unsubtle difference either way... but they didn't provide a means to bypass the filter altogether. What I remember at a 1980s AES meeting was we heard a bunch of weird out-of-band signals that sounded pretty ugly -- without the filter -- on an experimental test unit.
     
    Last edited: Mar 31, 2014
  11. The FRiNgE

    The FRiNgE Forum Resident

    The CD player would probably repel all the common household pests!
     
  12. missan

    missan Forum Resident

    Location:
    Stockholm
    I would say that those that can clearly hear over 12kHz and are over 30 is a minority.
     
  13. felix.scerri

    felix.scerri Forum Resident

    G'day all, an interesting thread although the term 'brickwalling' can have different meanings 'depending'. When I first read this thread title I thought it referred to the unfortunate 'loudness war' where all apparent dynamic range is squashed out of existence in the 'interests' of 'loudness'. Yuk!

    The other meaning refers to brickwalling filters used to remove all digital artifacts beyond 22.05 KHz in the D to A convertor as defined by the Nyquist Sampling theorem. I do recall a certain Pioneer CD player that incorporated a 'more loose' brickwall filter in the 'interests' of a more natural sound and better 'fidelity'. Anybody remember it? Regardless, I can say that D to A convertors have come a long way since the early days. Regards, Felix.
     
    Last edited: Mar 31, 2014
  14. The FRiNgE

    The FRiNgE Forum Resident

    Yes, not very many men over age 20 can hear beyond 16,000 cycles. There are very few acoustic instruments that have any harmonics beyond 16K anyway, except for a triangle or a cymbal. Another exception would be a synthesizer. Even for those percussion instruments that do go to 20k maybe 22k, there is very little acoustic energy above 16k. When music is compressed, the relative ultrasonic levels increase, so in certain program material there will be significant high frequency content above 16k. But any excessive harmonic levels at and above 16,000 cycles can sound grainy and tiresome.. too much high frequency and/or ultrasonic energy feels unnatural, and can be unpleasant.

    In perspective, the highest fundamental piano note is C8, 4186 hz.
     
    Vidiot likes this.
Thread Status:
Not open for further replies.

Share This Page

molar-endocrine