MrJoshua

Are you running out of processing power? If you are, consider converting the files to 48 or 44.1kHz. If not, why bother?

They had an 18 year old friend make the purchase. This criminal may very well have done the same thing. The 18-year-old then sold or gave those guns to the younger kids, without the consent of their parents. At that point it rather ceases to be a legal transaction, no? If I buy a rifle, well and good. If I then sell that rifle to a felon, it's no longer a "legally-purchased" rifle. edit: man. I have GOT to start reading the whole thread before I reply. nothing to see here. move along, folks...

At the risk of offending, that is such a frustrating cop out of an answer. :/ I promise you, it isn't as frustrating as someone comparing an apple to an orange as a method of deciding whether pears are better than bananas, which is essentially what you're doing here. Were you hitting the cymbal in exactly the same spot with exactly the same force and holding the stick exactly the same way on each take? Of course not. But each of these things can have exactly the sort of effect you were describing as the difference between these two "sample rates". Hit it harder, get a brighter ping with more ringing. Hit it a little higher toward the bell, get a bright sound with less ringing and lower volume. And so on and so forth. It's good that you hear a difference. I'm not arguing that there's no difference. But drawing a conclusion based on a test like this is a good way to fool oneself into believing something that may not be true.

What differences exist would more likely be caused by the performances being different as opposed to the sample rates being different.

Yeah, that seems to be the general consensus by most of us. Well, to an extent. With a properly-designed, well-implemented filter you don't really gain much by going to 96kHz. Even if I were going to use a higher sample rate (and I don't - I still use 44.1kHz) I'd still go with 88.2kHz as a maximum. That gets you WELL above the point at which a low-pass antialiasing filter should be fairly easily designed and built, aliasing should be reduced to the point that it essentially ceases to be a problem, and any objections over frequency response or fidelity at that point are just sheer bloody-mindedness. Any higher than that and you're just making it harder on your computer without gaining much of anything.

This will be the last time I bump the thread, honest! I really do want to see AN's oscilloscope results, though. I'm very curious as to how they turned out.

Well, the problem with HarBal (for the sake of THIS discussion, anyway) is that we're sampling again... I have an old oscilloscope around, but it doesn't work right. I really ought to tear that thing apart and try to fix it.

Not to dredge up an old thread, but I really am curious as to how those scope readings came out. I need to get myself a decent scope.

Well, I'm not an expert in signal processing - a lot of my knowledge came from college courses I took years ago, and more of it is from me wanting to be more educated about this as it relates to music. Quite a bit of what I've said in this thread has been gleaned from textbooks. So, as with anything you read on the Internet, it should be taken with a grain of salt - I might think I know what I'm talking about, and sometimes I'll even be right, but any subject you're really interested in should be pursued through proper educational channels. That said... Aliasing occurs when you have frequency content in your signal above the Nyquist restriction. In other words, if you're sampling at 44.1kHz and you have some 24kHz content in the signal you're sampling, you're going to get some aliasing. Now, while I have a loose understanding of this, I'm not entirely confident that I understand it well enough to actually try explaining it in coherent fashion. I'll give it a shot, but if it all goes down in flames, don't say I didn't warn you. A sampled signal shows up in the frequency domain as a repeating function. Let's say you record a bass note on the A string, open A, 55Hz fundamental. You'll have harmonics present at (for the purposes of this not-very-realistic example) 110Hz, 220Hz, 440Hz, etc... Putting a single sample of that note on a graph in the frequency domain would show spikes at 55Hz, 110Hz, 220Hz, 440Hz, and anywhere else that the signal had strength. But more than that, if you were sampling (for some crazy reason) at 2kHz, then you'd also show spikes at 2055Hz, 2110Hz, 2220Hz, 2440Hz, and so on. It shows up as a periodic function, you see, even in the frequency domain. Now, unfortunately, bass contains harmonics well above 1kHz, so our sampling rate of 2kHz is going to be a real problem. You see, not only is a 2kHz sampling rate not going to accurately capture frequencies above 1kHz, but we have another problem. It's going to capture those frequencies inaccurately, but they'll still be there, and it's going to make a mess. Because now we also have content at 1100Hz, and 2200Hz ... and that 2200Hz signal is going to be overlapping that 2220Hz signal most likely that we talked about above. Even worse, there will be interference from the repeating signal overlapping our original period - our 2200Hz signal is going to show up as an overlap back in our audible frequency range! So we've really messed up here, because we've really made our signal capture highly inaccurate. And it's going to show up as noise and weirdness when we try to convert back to analog. How do you combat this? Two ways. First, we use a low-pass filter (also referred to in this application as an anti-aliasing filter) BEFORE we sample the signal. You'll have noticed several references to these filters throughout this thread. The point of the filter is to keep inaudible high-frequency content from getting into your signal and causing sampling inaccuracies that will muck up your audible content. One problem (as you mention) is that there are no perfect filters. We can reduce the ultrasonic stuff enough that it isn't much of a problem, but we can't make it go away entirely. Also, steep filters can cause "ringing" or strange artifacts. Filter design is a branch of engineering I haven't delved into very deeply, and I'm not going to pretend to be an expert on it. Second, we sample at a rate a high enough rate to help avoid the issue. That does NOT mean that a higher rate will always be better - you reach a point where it's as good as it's going to get, really. If we were to sample right at 40kHz, then we'd have issues with any noise above 20kHz (audible or not) getting into our systems and mucking up our audible samples due to aliasing. That's one reason why we sample at 44.1kHz or higher - it's easier to design the filters to keep stuff above 20kHz out of our systems and avoid aliasing because we have a little room there to work with. A little 22kHz inaudible noise isn't going to wreck our recording because it's still under our Nyquist limit, and above that our filters should be keeping everything knocked down low enough to minimize the problem. I honestly believe that some converters MIGHT sound better at higher sampling rates, but that's NOT because of the conversion being more accurate, the higher sampling rate capturing more detail and space, or any of that stuff. It's because their filters aren't designed properly to prevent aliasing when you record at 44.1 or even sometimes 48kHz, so you still wind up with some ultrasonic junk in the samples that turns into noise and funkiness in the signal (and not the good P-Funk kind of funkiness; the kind that smells bad). In the end, the point is to make music. If you feel like you make better music at 96kHz sampling rates, or by moving your keys over to your other pocket when you play a solo, or with a lava lamp setting the mood and the overhead lights off, then DO IT if you're willing to take the hit on the number of plugins and such you can use - it's going to be a lot harder on your computer. I have no issue with people doing whatever they think it takes to make good music. After all, that's what it's all about. I just enjoy conversations like this because they tend to teach me something, whether through someone else explaining something I didn't know or through forcing me to delve into a textbook and learn something so I can explain it without making a fool of myself (or at least, no more of one than normal). edit: I hope I didn't mess that up too badly. It's a LOT easier to see with some graphs and a little calculus than it is to explain with text.

I'm still looking forward to seeing his oscilloscope results, because I just can't figure out what he's talking about when he says "parallel waveforms". Because there pretty much has to be only one waveform. How would you display more than one waveform on an oscilloscope anyway, unless you were using multiple test leads and what sense would that make? So I just don't know what he's saying. So I'm hoping to clarify that anyway. I don't mind people disagreeing with me. I just don't like it when we can't get our terminology on the same page, and I feel like that's happening here.

What would be nice is 64 bit 320khz sampling. That would theoretically be better then anything ever. It would also allow vastly higher level of detail. I can here a difference bwetween 24/48 and 24/96 and even 16/44 and 16 at various sampling rates. . but beyond that it becomes a moot difference. All right, here's where I have to point out yet again that there's absolutely no point to using 320 kHz sampling, at all. Any differences you hear between 24/48 and 24/96 are purely due to low-pass filter design, not because of the sampling rate. A properly-designed and well-implemented filter will allow a converter to capture every bit of the audible detail at 44.1kHz that it will at any higher sampling frequency.

I will borrow the big scope this weekend and get out the record player. I will also do a audio capture of the same source on digital playback. I will post up the results. I would appreciate it, because unless I'm much mistaken you should be seeing the same thing on either one. Now, if the digital audio has been "digitally remastered" then there may be differences there, but if it's a straight conversion to digital it should be the same. It will be interesting to see the results. edit: If you have the time, it would also be interesting to see what you would get if you recorded some audio from your record player through a decent converter at 44.1kHz/24-bit, then played back that recording through the scope. That way you know you're not getting any differences through CD remastering or anything like that. Especially useful would be if you could record the same pass of the record as you were monitoring on the scope, so we'll see the exact same pass of the exact same record, if you see what I'm getting out. Just to eliminate as many variables as possible. But I know that would be time-consuming, so if you can't don't worry about it. Just a thought.

and there is the diferenece between analog and digital. Analog represent all waves that may exist simultanoeusly unless they overlap. Whereas digital must SUM the waves into distinct events. Also they do come in sucession. Grab a osicliscope and have a look. You will find more wav patterns in a good analog record then in a digital one.where you will see them is in the parrelel. You won't see parrelel waves in a digital recording. Parallel waves? You're telling me that you're seeing multiple waveforms on a oscilloscope screen?

they can sum signals.Correct, what they can't do is play back 2 simultaneously different signals. Hence the difference between analog and digital. I guess I'm still not following you. If they can sum signals, why would they need to play back two simultaneous different signals? Why not just sum them together and play the one resulting signal?

thats exactly what it is doing. Its just doing it very very very fast. So fast that it is inaudable. But this is where analog wins out. analog can have multple sinusoidial waves simultanoeusly but the caveat is that when waves overlap that one will moentarily cancel the other. This is why analog seems to GEL better. I'm pretty sure that's inaccurate. I mean, think about it. If it worked that way you'd be unable to mix anything larger than, say, a dozen tracks without running into some pretty serious problems. You certainly wouldn't be able to mix sessions with a hundred-and-some-odd tracks like you see in some places. It just doesn't make sense. Your DAW software adds the waves together digitally and then sends the resulting waveform data to the DA converter, which converts it into one single analog waveform. It doesn't switch back and forth between them. It wouldn't make sense. Your computer is basically a big adding machine, and summing waves together is exactly what it's good at - addition.

I guess I'm not following exactly what it is you're trying to say, AN. You're saying that a digital system can only generate "one wave at a time" but what wave are you talking about? Say you record a guitar track and a bass track. Are you saying that when you play those back at the same time, the computer is only playing back one at a time, switching between the two? I'm not trying to nitpick; I just want to make sure I understand what you're saying because I'm a little confused right now.

You're missing a fundamental part of the process here, AN. Your asymmetrical wave is simply a bunch of symmetrical sinusoidal waves stacked on top of each other, of varying frequencies and lengths. As long as your sampling rate is above twice the frequency of the highest-frequency component of that weird shape, you'll reproduce it with 100% accuracy. I'm not sure exactly what you're trying to say where you talk about a bit here and a bit of the next... Are you saying that, for example, an 8-track converter only samples one track at a time? I could be wrong, but I don't think that's accurate. In fact I'd be VERY surprised if that was true, just on the face of it, as it seems like it would be much easier to design eight channels of conversion working in tandem off a single clock than it would be to cascade it. Could there be design ramifications that would call for a cascaded design? Sure. And I'm not a converter designer so it's entirely possible that I'm ignorant of something here. It just seems like since they're all acting off the same clock signal (they pretty much HAVE to be), it would be easiest to have them acting together.

Ya know.. you're right. I'm an engineer and also have a DSP textbook up in the attic somewhere. That was a hard freaking class. When i pictured say for example a lumpy sine wave, i was thinking that one sample per half period wouldn't capture the lumps. But of course those lumps are coming from higher frequency sine wave components and so the Nyquist freq would be respective to those higher freq sine waves, not the big lumpy sine wave i was picturing. I forgot that all signals can be constructed by the addition of multiple sine waves of various frequencies. So with this in mind, you just need that highest freq component, and sampling it at double the freq is all the accuracy you need. It's ok - we're in the same boat. Sorry if the previous post came off a little testy. It's been a long thread, lol. I've been having to dredge a lot of stuff up from memory, and finally figured I better grab the book and make sure I wasn't misremembering something.

There is some seriously bad info in this thread. First, Nyquist's theorem (sample rate should be twice the signal's frequency) does NOT dictate that required or the best sample rate it dictates the MINIMUM sample rate. Its a rule of thumb not a hard and fast rule. a sample rate of 2*freq simple gives you one high point (+1) and one low point (-1) per sine wave period. It gives nothing in between. Secondly, yes as you double the sampling rate (say from 48k to 96k) you are able to capture twice as high frequencies (by Nyquist's theorem), but you also get double the capture resolution throughout the entire frequency range. So even at 500hz, 96k is twice as accurate than 48k. Its capturing twice as many samples per period of signal. OK, you do NOT get any more accurate by doubling the points. I can't believe I'm getting an old textbook out, but... The following is from "Signal Processing & Linear Systems" by B.P. Lathi, Berkeley Cambridge Press, copyright 1998. Chapter 5, section 5.1 (page 319). Please forgive any typos. Emphasis is my own. We now show that a real signal whose spectrum is bandlimited to B Hz {F(w) = 0 for |w|>2*pi* B} can be reconstructed exactly (without any error) from its samples taken uniformly at a rate Fs > 2 B samples per seconds. In other words, the minimum sampling frequency is Fs = 2 B Hz. It goes on to prove this through the use of some fairly straightforward calculus, which I can also provide if you're interested. Once you've passed that Nyquist required frequency, you're 100% accurate. You can't get any more accurate than 100%. The ONE AND ONLY point to increasing the sampling frequency is that it becomes easier to design a low-pass filter for the circuitry. THAT'S IT. Not because it captures anything more accurately. That "harsh or strange high end" some people talk about? That's from using converters with improperly implemented or designed low-pass filters. It's a converTER problem, not a converSION problem.

Wait, you mean that it's easier to design converter filters for higher sample rates? The only purpose of going to those higher sampling frequencies it to make it easier to design a low-pass filter for the converter. Like the very first thing I said in this thread? Regardless, while a higher sampling frequency makes it EASIER to implement your filter, that doesn't mean it can't be done right at 44.1kHz. A good converter at 44.1kHz is perfectly capable of correctly reproducing all of the frequencies of the audible range. Which is why I track at 44.1kHz.

No one is going to convince me otherwise. And that in a nutshell is exactly why this has been going round and round. You've made your decision and won't listen to anyone saying otherwise. Fine. Just don't expect us to act like the science backs you up.

He says it's 3 times faster than the optimal rate, AND compromises accuracy. How does this back up your claim that it will provide a better waveform?

This actually proves my point Only if you're recording a lot of 20kHz square waves.

Guys, I've had enough of this. You are not getting an accurate reproduction of waveforms at the upper limit of human hearing when sampling at 44.1 or 48kHz and it DOES make a difference to the quality of the audio. This is backed up by Lavry's paper above and is the one point that I've consistently made that none of you have been able to successfully rebut. Lavry himself says that the optimum sample rate for audio is around 62-63kHz taking into account that the only reason that sample rates higher than this are not optimal is due to machine speeds not yet being fast enough so clearly there is a benefit from recording with a higher sample rate. That's all I've been saying for the last 2 days. I've read the math and it backs up what I'm saying so take it or leave it people, higher sample rates = better quality audio But he doesn't say that. He clearly says on Page ONE of the paper that higher sample rates are not useful. To quote, "The above suggests that 88.2 or 96KHz would be overkill." Furthermore, he's not talking about machine speeds i.e. processor clock speeds as you mention in a previous post - he's talking about the reaction time of the electronics in the sampling hardware not being able to react quickly enough, thus having a negative impact on sampling accuracy at extremely high sample rates. Which in fact would indicate to me that after a certain point, higher sample rates results in WORSE quality audio. Furthermore, I disagree with your interpretation of his statement about ~60kHz sampling rate. You're saying he claims it's optimal - he doesn't actually use that terminology. He says, "In fact all the objections regarding audio sampling at 44.1KHz, (including the arguments relating to pre ringing of an FIR filter) are long gone by increasing sampling to about 60KHz." That's not the same as saying 60kHz is necessary, and it's a LONG way from backing up your claim that higher=better period. In fact it directly goes against the idea that there would be ANY improvement from ANY sample rates higher than 60kHz ... and it further implies that even going up to 60kHz is only necessary if you want to be able to remove all doubt and any possibility of anyone being able to bring up any objections at all. A properly-designed 44.1kHz converter is perfectly capable of reproducing any audible or musical signal with 100% accuracy. Going up to 60kHz makes it easier to design the filters for the hardware and avoid any artifacts; that's it. Going to anything above 60kHz is overkill, and going up to 196kHz is just pure marketing unless you can hear ultrasonic waves.

Where does he say 192kHz would be the most desirable sampling rate? Unless I missed something in the linked PDF (which is very informative, by the way, and thanks to rlindsey0 for posting the link - it's a good refresher on stuff I haven't looked at much since college) he says even 88.1kHz is overkill. Right on page 1 - "The above suggests that 88.2 or 96KHz would be overkill."