almikel Posted June 30, 2016 Posted June 30, 2016 Applying "Nyquist" to the real world isn't as simple as some of you seem to think: http://www.audiostream.com/content/sampling-what-nyquist-didnt-say-and-what-do-about-it-tim-wescott-wescott-design-services#Mgw8vTt0e6hPJzxZ.97 I read it - for a paper that has good technical content the author ruined it all by the stair-step "reconstruction" diagram in figure 5. The Lavry paper linked below does a better job of showing how all the wiggles in a band limited signal sampled high enough does get replicated in the ADC to DAC process perfectly http://lavryengineering.com/pdfs/lavry-sampling-theory.pdf Another Lavry paper describes how sampling well above what we want to capture (eg 192kHz) can reduce accuracy - although as electronics gets better this is likely less of an issue, but keeping ultrasonic information in the signal can cause issues (eg spurious non audible content hitting a titanium dome tweeter's resonant frequency). http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf Tim Wescott's points in the Audiostream article on perfect band limiting and aliasing is all well understood and catered for in good digital design, and his examples of poor application/understanding of Nyquist are valid. His point on signals extending infinitely in time is shown graphically in the Lavry paper with errors at the start and finish. Extract from Lavry Sampling Theory paper page 7 Other than the Stairstep Reconstruction diagram I thought the Audiostream paper was a good article, but it doesn't justify any requirement to sample at 192kHz. cheers Mike
eltech Posted June 30, 2016 Posted June 30, 2016 (edited) so sample at 192khz and bandwidth limit at 88.2khz or some frequency above audibility. Problem solved, and you still get more samples at wanted frequencies Edited June 30, 2016 by eltech
almikel Posted June 30, 2016 Posted June 30, 2016 so sample at 192khz and bandwidth limit at 88.2khz or some frequency above audibility. Problem solved, and you still get more samples at wanted frequencies Hi eltech, like bandwidth limit at 22kHz which is above audibility and use oversampling to push the aliasing frequencies out to make filtering easier? Good Redbook implementations do that now. Sure sampling at higher than 44.1kHz is likely a good thing, but 88/96 kHz is more than required, and 192kHz sampling is like wanting your TV to display X-rays - not great for the user. Mike
eltech Posted July 3, 2016 Posted July 3, 2016 Hi eltech, like bandwidth limit at 22kHz which is above audibility and use oversampling to push the aliasing frequencies out to make filtering easier? Good Redbook implementations do that now. Sure sampling at higher than 44.1kHz is likely a good thing, but 88/96 kHz is more than required, and 192kHz sampling is like wanting your TV to display X-rays - not great for the user. Mike You don't need to oversample with 192khz because you have the high sample rate making the low pass filtering easier anyway. You have the added advantage of more samples to more accurately capture the waveform. Seems illogical that digital audio enthusiasts when presented with a higher resolution format want to bemoan it. It's not at all like wanting your tv to produce x-rays. It's simply about capturing and reproducing all the overtones of a musical instrument. I can't see how capturing a sound more accurately can be a bad thing because isn't that what high fidelity sound is supposed to be about? 1
Newman Posted July 3, 2016 Posted July 3, 2016 I have no objection to that, but the criticisms (your criticisms) of 16/44 are completely inaccurate, false, wrong, don't know what else to say.
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 Applying "Nyquist" to the real world isn't as simple as some of you seem to think: http://www.audiostream.com/content/sampling-what-nyquist-didnt-say-and-what-do-about-it-tim-wescott-wescott-design-services#Mgw8vTt0e6hPJzxZ.97 Indeed, but many people also misunderstand the implications of this. It is an issue for the practical implementation of devices (ie. engineering). It does NOT mean that there is a "problem" with sampling theory, or digital audio systems, per se .... within the caveats mentioned in the link, something sampled into a digital representation is a "full representation" of the signal. People who believe there is "a problem" with digital audio, see (what they think is) evidence which shows the "sampling theory is flawed", and that's not really the case. Like everything there are caveats, and practical limitations in real devices. The caveats in the link show how someone working with digital audio can "do the wrong thing".
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 See the post above and you might understand what I am talking about. The post from Firedog? Anyone who is going to comment on digital audio, should already be aware of everything in that link.
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 (edited) I read it - for a paper that has good technical content the author ruined it all by the stair-step "reconstruction" diagram in figure 5. I don't really agree. "The result of using a zero-order hold is shown in Figure 5." The picture shows exactly that. The target audience for the paper isn't someone who is going to go "ooooh stairsteps" (if you know what I mean). .... but it doesn't justify any requirement to sample at 192kHz. Agreed... aside from the desire to avoid issues with equipment or process that may fall foul of any of the mentioned caveats when operating at lower sampling rates. Edited July 4, 2016 by davewantsmoore
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 ....and you still get more samples at wanted frequencies They are not required You don't need to oversample with 192khz because you have the high sample rate making the low pass filtering easier anyway. Many DACs will oversample 192khz internally, as they run at much higher internal rates. For example the ESS Sabre DAC will oversample 192khz digital audio by 8 times to a sampling rate of 1.5mhz. You have the added advantage of more samples to more accurately capture the waveform More samples don't add any accuracy. Seems illogical that digital audio enthusiasts when presented with a higher resolution format want to bemoan it. That isn't what is happening. Simply repeating over and over that higher sampling rates offer a better representation of a (bandlimited) signal, doesn't make it true. You have no doubt heard with your own ears that audio with a higher sampling rate sounded good. There ARE (possible) reasons for that, but they are NOT the reasons you state. I can't see how capturing a sound more accurately..... High sampling rates don't do that. Higher sampling rates do this: Allow higher frequencies. The "overtones" you mentioned. They are not audible (it is trivial to test) Allow the (easier) avoidance of some of the caveats in the link Firedog posted (which is why most digital converts already run at high rates)
eltech Posted July 4, 2016 Posted July 4, 2016 (edited) @@davewantsmoore why would you want to oversample if you can record at 192Khz? I'd like to draw your attention to this article which states "digital audio will capture any signal, no matter how complex, as long as it can be represented as a set of sine waves lower than the Nyquist frequency." and "However, not all signals can be represented accurately by a set of sine waves below 20kHz. In particular, sawtooth waves (also known as triangular waves) and square waves can only be represented by an infinite set of sine waves with frequencies extending all the way to infinity. When you capture either sawtooth waves or sine waves digitally, you don't get back a perfect waveform when you play it back due to filtering." and "some musical instruments have harmonic characteristics very similar to sawtooth waves. And pop/rock music often contain music generated by synthesizers - sawtooth and square waves are fundamental building blocks for digitally synthesized music." You may also wish to check out this article http://www.cco.caltech.edu/~boyk/spectra/spectra.htm#I Edited July 4, 2016 by eltech
Newman Posted July 4, 2016 Posted July 4, 2016 And the audible part of all those waveforms is fully captured at 44k (and if I sound like a broken record, it is only because you keep repeating your mistakes) Sent from my HTC_0P6B using Tapatalk 2
Happy Posted July 4, 2016 Posted July 4, 2016 Wonder if there has ever been a proper DBT outcome where the majority felt the substantial superiority of the Hi-Rez over 16/44.1?
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 @@davewantsmoore why would you want to oversample if you can record at 192Khz? You seem to misunderstand the difference. .... most DACs take the rate they are fed, and oversample it internally to a much higher rate. If it is the case that oversampling performed outside the DAC was higher quality than the DACs own oversampling, then the user could oversample the data (with a higher quality method), and feed that higher rate data to the DAC. You see audiophiles doing this a lot. It has merit, especially when filters used are specific for the DAC in question (MQA are doing this). Why would I want to record in 192khz? Becuase I want to capture frequencies up to 96khz (not an issue for audio, as it is inaudible) Becuase my hardware falls into the traps posed in the previously linked article .... and recording at a higher rate, may afford an opportunity to avoid them Why would I want to distribute audio at 192khz? Because it was recorded at that rate, and changing it is an opportunity to harm something (can be overcome) Why else? <shrug>
Happy Posted July 4, 2016 Posted July 4, 2016 Why would I want to distribute audio at 192khz? Because it was recorded at that rate, and changing it is an opportunity to harm something (can be overcome) Why else? <shrug> Because the audiophiles would dance and throw their hard earned dollars? 1
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 (edited) In particular, sawtooth waves (also known as triangular waves) and square waves can only be represented by an infinite set of sine waves with frequencies extending all the way to infinity That's correct.... but (as they go on to mentioned in their conclusion) that's not particularly an issue. Audioholics probably should know better than to leave that open ended - as it invites unnecessary debate However, not all waveforms, even deceptively simple ones, can be represented accurately once they are filtered<snip> It could also be argued that the frequency components above 20kHz for sawtooth and sinewaves are not important, since we don't hear them, so arguably what our ears are hearing are represented accurately in Figures 6-8. This is perhaps true This IS true. Many people have commented that CDs do not seem to reproduce high frequencies as well as analog sources such as vinyl and magnetic tape. Perhaps the above could at least partially explain the subjective impressions? This would also suggest one benefit of sampling at higher rates such as 96kHz or 192kHz even though our ears cannot hear past 20kHz. A higher sampling rate can help preserve amplitude and harmonic accuracy for non-sine waves at high frequencies. It is not that difficult for people to do a practical test of this for themselves. Carefully convert some high rate audio to 44.1khz (and then back to the original rate), and see if you can find a difference. Converting back to the original rate, is to nullify any differences in how the playback hardware operates at low vs. high rates, which may cloud a conclusion. Edited July 4, 2016 by davewantsmoore
Guest Posted July 4, 2016 Posted July 4, 2016 My $0.20 24bit 96Khz sounds better to me (than 16 bit 44Khz) as long as it is the native resolution of the master. Higher is very difficult differentiate and practically the same in sonic value, to my ears (that turned 46 today)
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 Wonder if there has ever been a proper DBT outcome where the majority felt the substantial superiority of the Hi-Rez over 16/44.1? There has been AFAIUI. There have been many the other way. The critical part though is WHY. Many explanations you typically see for "why 192khz matters" lie somewhere between "a misleading snippet of truth" .... and a "complete misunderstanding of how digital audio works" Depending on exactly how the test is orchestrated, you can control for all the variables. For example: If you wanted to test only the digital system (and not the issues inherent in a specific DAC) .... then you could convert 192khz to 48khz and back to 192khz, and compare that to the original 192khz. This is to avoid something specific your DAC might do which might make it work better at 192khz, than it would at 48khz..... which might prevent you from drawing a clear conclusion from the results. ie. how can you tell the difference between "audio captured at high rates is a more accurate representation" .... and "my DAC works better when fed high rate audio"
davewantsmoore Posted July 4, 2016 Posted July 4, 2016 (edited) My $0.20 24bit 96Khz sounds better to me (than 16 bit 44Khz) You use a miniDSP in your system which is resampling all inputs to 96khz. By feeding the miniDSP a 96khz signal, it may "work better" as it does not resample the incoming data. as long as it is the native resolution of the master. This can be quite important. If the audio is originally recorded at a certain rate .... then resampling it to another rate can only damage it. [in theory it is always best to keep audio in the original rate] that turned 46 today Edited July 4, 2016 by davewantsmoore
Guest Posted July 4, 2016 Posted July 4, 2016 You use a miniDSP in your system which is resampling all inputs to 96khz. By feeding the miniDSP a 96khz signal, it may "work better" as it does not resample the incoming data. This can be quite important. If the audio is originally recorded at a certain rate .... then resampling it to another rate can only damage it. [in theory it is always best to keep audio in the original rate] and my Wife forgot same as last year.
eltech Posted July 5, 2016 Posted July 5, 2016 That's correct.... but (as they go on to mentioned in their conclusion) that's not particularly an issue. Audioholics probably should know better than to leave that open ended - as it invites unnecessary debate Allow me to try to explain where I'm coming from. I don't like the use of the word "perfect" when discussing digital sampling. Since I believe nothing is perfect in this world, I am more comfortable describing digital sampling as "good enough" as was the phrase used by the person who wrote the article I linked a few posts up. That said, I feel a more comprehensive description is to say its "good enough for most people, but some people are still unsatisfied" this is based on theory, practice, and percieved sound quality. I think it is fair to say that there are people who prefer the sound of DSD recordings. Despite being a different technology - it uses a much higher sample rate than any PCM recording and captures sounds above the range of average human hearing. Technology does move with the times, and digital storage is cheap these days so a 192Khz recording is comparatively small and cheap when compared to early digital recordings using Beta tape and a PCM adaptor. I dont think the file size detracts from what I see as its benefits. I think hifi should be about capturing as much of the musical information as possible including the information above average human hearing. Finally, I have no specific personal experience, or inside knowledge, and I conceed that not all new things are better than old things, but if there is no improvement of any sort to be gained from higher sampling rates, I ask the question why DSD is gaining popularity in recording studios, and why we now have manufacturers making ADCs and DACs which record/play at 384khz. I hopefully assume that there is some good reason for it, and not simply to sell something new?
davewantsmoore Posted July 5, 2016 Posted July 5, 2016 Allow me to try to explain where I'm coming from. I don't like the use of the word "perfect" when discussing digital sampling. Definitely. It's not a good term.... and it is things like that where discussions can fall over. The way a digital system differs from "perfection" can be quantified. It is not about believing whether it is, or isn't "good enough" (for what?!) .... but about actually understanding precisely how it performs, and what that means. I think it is fair to say that there are people who prefer the sound of DSD recordings. There are many reasons why this may be the case, that do not support any conclusions about whether or not low bitrate and high sampling rate is a "superior" way to audio digitally. HINT: Both high and low bit and sampling rates are a zillion times better than generally good enough to represent digital audio. I say "generally", because the devil is in the details of how the audio has been made, treated, and played-back. However, there formats themselves are not the culprit, per se. Despite being a different technology ... but not as different as most people think. What many people do not realise, is that many DACs today convert everything into "DSD like" formats internally. Technology does move with the times ... but low bit, high rate isn't "new". It isn't some new discovery. "PCM"-type digital systems were invented nearly 100 years ago, and they all used low bitrates.... it's simply a trade off with noise. Anyways. If we examine how the digital representation (as different from how to build a practical converter, or how people make/release content for a specific format) ..... we see they perform close enough to each other not to matter. I dont think the file size detracts from what I see as its benefits. There are potential advantages to operating at high rates.... but capturing frequencies above audibility is not one of them. Those potential advantages can also be achieved in other ways (at lower rates) .... and it is important to note, that operating at high rates, doesn't guarantee any advantage (so we can't say "ohh high rate = must be good") I ask the question why DSD is gaining popularity in recording studios If audio was created in a DSD system, then resampling it into another system can only harm it. Analogue to digital converters are example of converters operating at a very high rate (because it's easier to build them that way). You can either resample their output into high bit system ..... or leave them in the 1bit system (ie. as "DSD") DSD can store sounds quieter and higher in frequency than CD audio.... and arguably the limits of CD could be reached in very (very!) corner case situations. DSD / SACD also offers copy protection and surround. ...however post #339 contains most of the reasoning.
JSmith Posted July 5, 2016 Posted July 5, 2016 There is a "sweet" spot with digital sampling... 192kHz is too high and 44.1kHz is too low. There are audible liabilities at both ends. JSmith
eltech Posted July 5, 2016 Posted July 5, 2016 understanding how it performs and how it is perceived to perform are separate things. I perceive analogue to out perform digital based on what I've heard in spite of its numerous technical documented and acknowledged flaws. So I approach the topic based on the criterion of personal perception of sound quality. I have to assume that digital sampling is imperfect, flawed and has room for improvement in spite of its theoretical "perfection". Indeed you may be correct about inaudible higher frequencies, or not. But I still search for the reasons for what I hear and perceive. I am not satisfied with digital audio nor the mathematics and theory behind it. I didn't want to, and don't want to distract this thread from its original topic, but the only way for me to further state my position was to say the above. 1
davewantsmoore Posted July 5, 2016 Posted July 5, 2016 There is a "sweet" spot with digital sampling... 192kHz is too high and 44.1kHz is too low. There are audible liabilities at both ends. I really don't see what the harm is in using very high rates (aside from data rates being too high for via the internet) 44.1 isn't too low (for what?!) ..... given proper band limiting, and equipment which works well ..... which obviously hasn't always been the case, but those caveats apply for any rate, not just 44.1. I perceive analogue to out perform digital based on what I've heard I don't necessarily disagree. .... but you are providing explanations for what caused the difference which are incorrect. I have to assume that digital sampling is imperfect, flawed and has room for improvement Why? There are many other reasons why a difference was heard which don't rely on the assumption you are making here. Some of those reasons are specific problems for digital .... but they're issues to do with implementing a real device, and not because "digital sampling is flawed". It's a complex topic, which really takes pages to unpack. But I still search for the reasons for what I hear and perceive. I am not satisfied with digital audio nor the mathematics and theory behind it. I didn't want to, and don't want to distract this thread from its original topic, but the only way for me to further state my position was to say the above. Heh. I think you're on-topic You should look into it further. All the explanations you seek are there, and not particularly controversial.
Recommended Posts