96khz Sample Rate

  • Hah.


    The motion of your speaker will look like 'that ugly thing' because it has inertia - there is no way it can cease to exist in one location then appear in another as a square wave would demand.


    Try putting a synthesized square (at relatively low frequency) wave through a one pole low pass filter (to emulate the limited speed at which your speaker can move). This is what happens.


    You should read about taylor series or whatever you want to call them - the fact is the theoretical harmonic components of a square wave go up to infinity, and considering harmonic components is pertinent since the instrumentation (ear drum -> bones in your ear -> hairs in cochlea) in your ear effectively performs a wavelet transform on the sound waves coming into your ear. Obviously you can't represent all of these harmonics in a finite bandwidth system (i.e. human hearing, sampled audio, or in fact any moving system with real mass) - and the loss in these higher frequency components from a 'true' square wave creates the ripples you see in a band limited square wave. On the high end of the scale, C9, 10, you only get a small number of audiable/transmitted harmonics, so what is seen will truly look like only a combination of 2 or 3 sine waves...


    Hmm, that was a bit of a ramble but I think it's all in there...

  • Well, for starters the filtering you are talking about is only for ADC, not DAC - so it's entirely independant of the virus.


    Maybe true with a filter for ADC but I wasn't about that. I was about reconstruction that needs a lowpass to avoid the backfolding issue evil from digital domain.



    Second of all - using a decimating input (sima-delta or whatever) gives you a really high bandwidth to start with, now you can filter that digitally or with an analogue filter - incidentally 96dB/octave wouldn't be a great effort. Analog Devices quote for one of their off the shelf AC'97 CODECs a pass band (+/- 0.09dB) of 0.4*fs (for 44.1kHz that's just shy of 18kHz) and a transition band from there up to 26kHz. (-74dB at the nyquist point - full scale sinusoids at just over nyquist point could only occupy last 4 bits, by the way thats over 150dB/octave, so the aliasing could only occupy 2 bits at best by the time you ramped up the frequency of the sinusoid so you could hear its aliasing). 48kHz makes aliasing disappear underneath quantization noise by the time it folds back into audiable territory - and this is with a full scale sinusoid, which would be unlikely to occur ever.


    Not sure if you try making things much more complicated to cover that you are not so sure - or whatever?
    I also didn't say it's impossible to get closer to what I showed up with as an example to make the basic effect clear - I just said you can't simply expect the theoretical limit the theorem promises... or a great construction from the two magic numbers.



    Also, realtek, just because they quote 24/96 it doesnt mean you're actually getting that resolution - its like when they stick 12MP digital camera sensors in mobile phones, the lens instrumentation just doesnt have that kind of resolving power. BUT that's the standard sensor they are flogging, so they stick it in anyway because it's cheap (especially if you are sony and you are making millions of them).


    Practical numbers are fun...


    Oh, now you contradict yourself a way, referring to real hardware and "practical numbers", that's indeed funny. :D


    I do not like to jump into the digital camera thing here - also there are some parallels - and some even more complicated other things ....

  • blah blah blah, etc....


    The camera thing reinforces my argument because the point is the theoretical extra precision you get is swallowed up by real world conditions and the limited nature of human perception. The numbers I gave were a specific measurement of the aliasing (or folding back as you say) and calculating the potential impact on the measured signal which I estimated (it was back of envelope stuff so I shan't be quoted on it) to be negligible.


    Heres the arrogant bit, but whatever...


    The fact that you don't understand the calculations (arm waving ones that they were) involved only demonstrates that you are not qualified to give an informed opinion on the issue. You can't discount a point because it's too precise and directly related to the discussion, thats crazy.


    Anyway, I officially don't care any more and I'll leave it at this. The guys that designed this thing - they aren't so dumb, they do actually know what they are doing (strike that, they REALLY know what they are doing) and it's also not my job to teach you it all so that is that.


    Roby31 - lol. Interestingly there's a whole can of worms about what humans perceive to be 'warm' or 'thin'. Probably more to do with non-linearities than bandwidth, though...

  • Finally I think we have over complicated a simple argument we basically agree about.


    "The fact that you don't understand the calculations (arm waving ones that they were) involved only demonstrates that you are not qualified to give an informed opinion on the issue."
    I am not sure what calculations you refer to... you didn't present these. You quoted some Analog Devices spec sheet data and concluded something... My message was basically quite simple: Quality of real world constructions decide what you get - not just two numbers.


    "the theoretical extra precision you get is swallowed up by real world conditions"
    That's exactly what I was about - a real world construction is just as good as it's weakest parts - usually not reaching theoretical limit. That's all I was about. 4x.xx sample rate can give us sufficient audio quality if the real wold construction is done proper.


    "limited nature of human perception"
    Perception is more difficult to discuss as it is different for every human, I wasn't really about that - also I agree that many people may not be able to hear very high frequencies directly.


    "The guys that designed this thing - they aren't so dumb, they do actually know what they are doing"
    Absolutely - but you imply that highest quality is the only design goal - and that's too often not the case - cheap make is often enough the goal. This includes also combining elements that really do not match like your camera example. But marketing knows about the magic of numbers to people so they still request impressive ones - regardless if this has any practical relevance. Another marketing method is to put numbers into an irrelevant context to make them look good. It's often difficult to rate something from paper - hence my warning just to trust two numbers...


    My non arrogant point:
    You showed off a great sweeping blow - why not presenting this in a more friendly way? This is discussion and no war. Just "beating" people does not make a community better.

  • Perhaps its a latent over-emphasis from dealing with too many people for whom 24/96 or whatever is a 'must' without any understanding of how these numbers actually affect what you get, they just hear that it is somehow 'more' and want it.

  • Perhaps its a latent over-emphasis from dealing with too many people for whom 24/96 or whatever is a 'must' without any understanding of how these numbers actually affect what you get, they just hear that it is somehow 'more' and want it.


    again, wrong. if you bothered to read the beginning of the thread you would know that i already had this arguement with TiUser over the 96khz request just me saying "omg i want my sample rate # to be higher then it will be better." no. I already explained that TO ME, 96khz is a HUGE difference in sound ( big enough to hear and feel the difference drastically), if FOR YOU, reduced polyphony/latency isnt worth the higher sampling rate, then fine, glad that works for you. however, i was requesting the feature because I feel a huge difference in sound from 44.1 to 96, no matter what all the theories and bs explainations have to say. you can explain it and try to refute anyway you want, but for me it doesnt matter

  • I understood you motivation.


    The shannon theorem clearly allows higher frequency response with 96kHz than 44.1kHz - theoretically and practically too - in terms of technical frequency response - the closer the better the constructions are.


    Other arguments have been found why the difference for the final audio quality might be less relevant than the price you have to pay on the dsp.


    D/A converters have their own properties - i.e. own "sound" - depending on construction and independent from the converted digital signal. This depends on many factors - some have been quoted here, others not.


    The whole audio passes a chain of elements and the weakest part will affect the result most. So quoting things like hearing abilities, speaker flaws and other issues is all right. However mixing these can easily lead to a lot of confusion where single points get lost.


    Finally even if there would be no measurable difference that does not mean that perception can not be different. Placebo in medicine is also something that should make no difference as it has no medical ingredients - but practically it does even no one knows the reason yet.

  • I understood you motivation.


    The shannon theorem clearly allows higher frequency response with 96kHz than 44.1kHz - theoretically and practically too...


    ...But not for frequencies already captured - the representation of a 1kHz signal is not improved by increasing the sampling rate beyond 2kHz (plus a tiddly bit) - thats the whole point.




    Quote

    Finally even if there would be no measurable difference that does not mean that perception can not be different. Placebo in medicine is also something that should make no difference as it has no medical ingredients - but practically it does even no one knows the reason yet.


    True, the determined enough listener can hear whatever they want to if they try hard enough. Also, happier people have better immune systems - being given medicine and the promise of a possible cure makes you happier. Also, the placebo effect has never been proven to exist.

  • "...But not for frequencies already captured - the representation of a 1kHz signal is not improved by increasing the sampling rate beyond 2kHz (plus a tiddly bit) - thats the whole point."
    Did I say this anywhere? If that's what bothers you than calm down, that's totally true - no objection.


    "...Also, the placebo effect has never been proven to exist. "
    This is definitely wrong. Medical studies regularly do double blind studies and the placebo effect has been proved to exist. It's so shocking significant that medical research goes more and more into this. With audio so far I haven't heard of similar research... :D

  • I contest the point again. Yes it is true that medical studies make allowances for the so called placebo effect by giving blanks. BUT there is no scientifically valid proof of its existence.


    Also, if the increased sample rate doesnt give additional resolution to frequencies already captured (as you conceded) when why choose a sample rate in excess of what is necessary (i.e. to accomodate frequencies within the human hearing range).

  • That's somewhat silly. What prove do you miss with placebo experiments with clearly significant effects and results? What does it need to make it proven in your opinion? I am not talking about woodoo experiments - that was serious science - done by people like you who where more likely willing to prove that such effects do not exist... but they were scientists and could not deny clear results.


    I give up with the sampling rate thing.


    A) We agree about shannon theorem defining the maximum theoretical limits for digital capture and reconstruction. If I understand right you also agree that real hw is never better than the theory - more or less - and depending on the converter principle. Until now this all is not about anyone can hear any frequencies or artefactst. That's just the technical properties that apply to an audio card as well as to a GHz range digital oscilloscope or similar devices.


    B) Now you quote hearing abilities and I again agree that hardly anyone can hear frequencies of 20kHz or above. But this is a biological constraint and does not influence shannon or the capability of a technical system.


    C) Then you put that together and conclude you do not need more than 44.1KHz sample rate to transport frequencies usual humans can hear - and I again agree that this is true if the converters are constructed well.


    D) Then you state you can not increase information already captured by using a better resolving system - and again I agree in this circumstance.


    E) Somewhere you referred to limited speaker frequency response to make an emphasis on C) But that usual speakers for music do not transmit such high frequencies does not mean that it is impossible to make such or that they can not even exist. But then you would point to B) again and say it's pointless - again true in this context..


    What about the scenarios you didn't take into consideration? Simple, they are not relevant to you or you find a context to put it into that makes it irrelevant then, right? Because of A, B, C, D, E? More letters in the alphabet? And if I would pick up another hypothetical example just for the sake of giving another rough idea you will tell this is unqualified nonsense and not state of the art - of course it is not, it is about a rough basic idea.


    When I would say F) there can be processing steps wanted in the chain that benefit from even higher sample frequencies to maintain intermediate frequency results that can well be transformed back in the normal acoustic frequency range later an well be important for the sound than you will again find arguments that this is nonsense because of G) Oversampling does the job as well - and indeed I would have to agree again...


    As far as it concerns my intentions this discussion gets out of control and value. I agreed to every single point you made referring to the context you put it in and you still think there is something wrong in the chain?


    You won - even if I do not know what... :pinch:

  • I've found that distortion plugs and some other FX that operate on a sample-by-sample basis produce noticeably different (often smoother) results @ 96khz. Not something I've tested extensively but definitely something I've noted.

  • May be an oversampling or even downsampling thing - highly nonlinear plugins may respond differently to different sample rates - just as a matter of imperfect operation if nothing else.


    Of course on vague inspection you could just be imagining it - the brain plays funny tricks...

  • Hey. I thought add something to the discussion.


    Has anyone thought about the DA converters and how they work? It's not so much the sample rate that matters, but how the DA converters handle the outputting of high frequencies that really matters.


    If anyone has an oscilloscope (even a basic one), I challenge you to output a full 0dB sine wave at half and quarter of the sample rate, and then at random frequencies of your choice. Take a look at how the DA converter deals with smoothing the wafeforms. You will definitely see and may actually hear that there's a fair difference between a cheap on-board soundcard and an audio interface when outputting pure tones. {EDIT: Also try with square waves as they contain a lot of harmonics that when output incorrectly (i.e. several blurry sine waves overlayed on top of each other), will alter how your ears perceive the sound. You will see massive differences in how the DA converters of different interfaces deal with the shapes. It's really quite an interesting experiment when you see it with your own eyes and hear the results directly!}


    This is the key issue, that increasing the sampling rate will actually sound audibly clearer on some older and/or cheap interfaces, yet with others it will be negligable and the placebo effect will far outweigh any audible difference. It also depends on the content of the source material being used to compare interfaces. Most modern audio interfaces have been designed with all the DA flaws in mind, so there will actually be very little to no audible difference when running 48kHz as opposed to 96kHz. I prefer to run my projects at 96kHz because I have OCD and it makes me feel better to know the waveforms are being captured in a higher resolution :whistling:


    The Virus Ti now actually allows me to run projects at 96kHz since OS 4.0.5.01, so I'm very happy (even though the Virus is running internally at 48kHz). With older OSs, it would be out of sync and cause problems on realtime mixdowns so I would be forced to run the project at 48kHz. This is not good for OCD :)

  • This is all very well, but how many of you know actually know what you can measurably hear ? How does your hearing compare to audiometric zero? None, I bet. How do you know your hearing in the high frequencies is normal? ( Chances are it isn't due to the fact you produce music....) Hearing is a subjective phenomena, all this talk of ' sounds better' is purely subjective, so I'm afraid its a somewhat pointless discussion. Do whatever is good for you, if it sounds good, it IS good.

  • The subjective aspect is not the issue - if it can't be measured, then you definitely can't hear it. Everyone knows that any real and finite signal can be decomposed into a set of sinusoids and when a linear transform is applied to them, you can consider the effect on each of the sinusoids separately (principle of superposition). Not only this, but for any linear transform the output will contain the same sinusoids, but maybe at different amplitudes (subtractive synthesis).


    In short, if you can't hear it going in, then you wont hear what comes out as a result of it, and it wont have an effect on what you can hear.


    That is the linear case (ie not including distortion or chorus/flanger effects).


    The interesting thing is that if you did a downward doppler shift on a delay effect working on a higher sample rates, it should cause inaudiable frequencies to become audiable
    (although it would have to be a pretty dramatic shift to bring anything inaudiable down into 'interesting' frequencies, thus making it negligible for chorus/flanger effects)
    which would not become audiable at a lower sample rate.


    Also, for distortion plugins the waveform would not be 'smoother' for higher sample rate signals as there is no concept of 'smoothness' when you are working with quantised and discrete values - you do not increase your spectral resolution for a given chunk of time, only your nyquist frequency. Also, distortion plugins add harmonics, so if the starting point is inaudiable, the result you get out of it certainly will not be!

  • I would also like to work at a higher sample rate. I always work realtime, my sequencer sends midi data to the hardware-instruments and they stream the music back into the converters. Sadly, I can not let them run higher than 48khz when working with the Virus TI. Letting all the converters and plugin calculations run at a higher sample rate does translate into the final result, even after converting the master back to 44.1.