Hi, I don't understand how a sample rate of say 44.1kHz can be any good at all at higher frequencies. I guess I am missing something but surely if the frequency being sampled is 20kHz then each cycle would be sampled only twice (approx). If for instance you had a sine wave at 20kHz and happened to sample on the peaks (approx) then a sine wave would look just like a square wave, ie, not representative of the waveform at all! Is it true that the higher the frequency the less accurate the sampling? Adam.
Adam Skeggs wrote: > > Is it true that the higher the frequency the less accurate the sampling? Exactly right. There are lots of good tutorials on this, but put very simply, it was thought that it simply didn't matter if there were a square wave at 20KHz because you couldn't hear it. A square wave is a sine wave + harmonics, and if you can't hear the harmonics, then all you hear is the fundamental = the sine wave. Now that's turned out to be not quite true, for lots of reasons, not least of which is that the harmonics in that square wave create unpleasant results with the harmonics of other square waves in the sample -- causing resulting tones (subharmonics). One way to get around this was brick-wall filtering at the high end, but this caused some measure of phase-shift that was perceived by some as grittiness and loss of soundstage in early digital recordings. So higher sampling rates have been used and more advanced filters. That's just a start, somewhat misleading if you take it as the complete answer. Look around for tutorials online ... there are lots of good ones! Dennis
On Sat, 8 Nov 2003 13:58:42 +1100, "Adam Skeggs" <> wrote: >I don't understand how a sample rate of say 44.1kHz can be any good at all >at higher frequencies. I guess I am missing something but surely if the >frequency being sampled is 20kHz then each cycle would be sampled only twice >(approx). Correct. You have to sample at twice the signal rate to get anything useful. Nyquist invented the idea. >Is it true that the higher the frequency the less accurate the sampling? No. Unless your sampling hardware is lame. Some consumer stuff may not work well at high rates, but the consumers won't care. Pat http://www.pfarrell.com/prc/
Pat Farrell wrote: > > On Sat, 8 Nov 2003 13:58:42 +1100, "Adam Skeggs" > <> wrote: > > >Is it true that the higher the frequency the less accurate the sampling? > > No. Unless your sampling hardware is lame. Some consumer stuff > may not work well at high rates, but the consumers won't care. I read his question the other way around -- higher the frequency being sampled, not higher the frequency of sampling! Language, darn it. Dennis
On Sat, 08 Nov 2003 04:08:59 GMT, Dennis Bathory-Kitsz <> wrote: >Pat Farrell wrote: >> On Sat, 8 Nov 2003 13:58:42 +1100, "Adam Skeggs" <> wrote: >> >Is it true that the higher the frequency the less accurate the sampling? >> No. Unless your sampling hardware is lame. Some consumer stuff >> may not work well at high rates, but the consumers won't care. > >I read his question the other way around -- higher the frequency being >sampled, not higher the frequency of sampling! Language, darn it. Had not thought of it that way. Lets try again, with a little more crispness in the language. The CD standard rate is 44.1kHz. Its Nyquist rate is 22.05 kHz. It can theoritically capture signal rates as high as 22kHz. Which is higher than the 20-20k Hz range of "human" hearing that audio engineers have used for decades. But you must have an anti-aliasing filter at the Nyquist rate, that is, a low pass filter that throws away everything above 22kHz, because you will not be able to identify and recreate signals (for example 24kHz) when sampled. In reality, filters are more like 6dB per octave, or 12dB per octave, so real world analog filters start lower, say at 15kHz. Starting lower is bad, we pay big bucks for mics and preamps that go up. Or you can try using a brick wall filter that starts at 20kHz, but they are evil, cause all sorts of phase problems in the audible range. So, what do you do if you think that interharmonic sounds exist in the 15kHz to 30kHz range that you have to capture digitally? You have only one option, sample at a higher rate. Sampling at 88.2 kHz will let you try to capture signals at 44.1kHz, which means you are likely to actually capture signals at 30kHz. Sampling at really high rates, say 192kHz would let you capture any signal that exists in the 80kHz range. My personal belief is that humans can detect phase relationships far better than they can detect frequency, and higher sample rates help phase capture as well. I am cautious about prosumer gear that claims to record at high rates, many of them quote a rate but without any supporting information (such as distortion at a given bandwidth, or even frequency response). Pat http://www.pfarrell.com/prc/
"Pat Farrell" <> wrote in news:: <some snippage> > So, what do you do if you think that interharmonic > sounds exist in the 15kHz to 30kHz range that > you have to capture digitally? > > You have only one option, sample at a higher rate. > Sampling at 88.2 kHz will let you try to capture > signals at 44.1kHz, which means > you are likely to actually capture signals at 30kHz. > > Sampling at really high rates, say 192kHz would let > you capture any signal that exists in the 80kHz range. > > My personal belief is that humans can detect phase > relationships far better than they can detect frequency, > and higher sample rates help phase capture > as well. > > I am cautious about prosumer gear that claims to > record at high rates, many of them quote a rate > but without any supporting information (such > as distortion at a given bandwidth, or even > frequency response). > > Pat http://www.pfarrell.com/prc/ My question with high sample rates is always... how much of this really matters, when you're crunching the final product down to 16 bit 44.1 kHz audio for distribution as MP3's and audio CD's? It's nice to hear (or think you hear) subtle nuances on a high-end set of monitors, in a perfectly tuned room, with your D/A running at 24 bits/96k or 24/192k when you're mixing. But if the extra information doesn't help you do a better mix once you're down to 16/44.1, then what's the point of all the extra CPU cycles and disk storage? I've done some tests comparing 24/95 and 24/44.1 source files after they've been converted down to an audio CD, and that hasn't convinced me that I need to record and mix at 24/96. Maybe my software just isn't good enough to preserve any possible phase information that would survive the conversion, or maybe these old ears have lost too much high frequency range. But whatever the reason, I'm just not hearing enough difference to care about (if I'm hearing anything at all). If I was producing material that would be released on DVDA or SACD, that's another thing. -- Mike Barrs
"Adam Skeggs" <> wrote in message news:3fac635c$0$3501$... > Hi, > > I don't understand how a sample rate of say 44.1kHz can be any good at all > at higher frequencies. I guess I am missing something but surely if the > frequency being sampled is 20kHz then each cycle would be sampled only twice > (approx). If for instance you had a sine wave at 20kHz and happened to > sample on the peaks (approx) then a sine wave would look just like a square > wave, ie, not representative of the waveform at all! > > Is it true that the higher the frequency the less accurate the sampling? > > Adam. > > A 1Hz sine wave would get 44100 samples so you could draw a convincing picture of it. A 20kHz sine wave would get 2 samples. Join up the dots on that. It could be anything. Since demonstrably the sound is OK it must be that at high frequency (and decreasingly at lower and lower frequencies) it doesn't matter what the waveform is. Like a distant object to the eye, we can't see the detail so as long as it looks roughly like a cow it's a cow. If we went for 1000 samples of a 20kHz waveform that would be a 20MHz sample rate. I guess if we were dogs we would complain about the poor quality. Lucky our ears are not much good. Adam.
On Sat, 08 Nov 2003 06:40:04 -0000, foldedpath <> wrote: >My question with high sample rates is always... how much of this really >matters, when you're crunching the final product down to 16 bit 44.1 kHz >audio for distribution as MP3's and audio CD's? If you are aiming at MP3s, I believe that the answer is clearly "not worth it, don't matter." For Red Book audio as a target.... >It's nice to hear (or think you hear) subtle nuances on a high-end set of >monitors, in a perfectly tuned room, with your D/A running at 24 bits/96k >or 24/192k when you're mixing. But if the extra information doesn't help >you do a better mix once you're down to 16/44.1, then what's the point of >all the extra CPU cycles and disk storage? Oh no, an engineering question! Engineering is about deciding what is good enough. When do you add more steel to a building's frame and when it added steel just wasted money. It is not engineering to buy the best of every part to make a cost is no object $10,000 preamp. It is engineering to make a good preamp that sells for $500. In simple terms, no group or body who has come into my studio has an audience who cares. Right now, the CPU cycle budget is fine, and wasting a few cycles is not a problem. In the days of Gigahertz processors, I don't run out of tracks or effects cycles. YMMV, seriously. I don't worry about the disk storage at all. Disks are nearly free. I do worry about disk IO capacity, both during processing and when moving the tracks/mix to another machine. Even over 100baseT, moving a gigabyte takes real time. I was trying to post an explaination of what higher sample rates do. Whether we really can afford them is a totally different topic. One that I think is a lot more useful than how to get surround sound out of Sonar (or Vegas or Logic ...) since all surrond sound is at best medium fidelity. Cycles for nothing and Disks for free.... Pat http://www.pfarrell.com/prc/
Yes this is a tough question.On Sat, 08 Nov 2003 01:59:34 -0500, "Pat Farrell" <> wrote: >On Sat, 08 Nov 2003 06:40:04 -0000, foldedpath ><> wrote: >>My question with high sample rates is always... how much of this really >>matters, when you're crunching the final product down to 16 bit 44.1 kHz >>audio for distribution as MP3's and audio CD's? > >If you are aiming at MP3s, I believe that the answer is clearly "not >worth it, don't matter." For Red Book audio as a target.... Agree. > >>It's nice to hear (or think you hear) subtle nuances on a high-end set of >>monitors, in a perfectly tuned room, with your D/A running at 24 bits/96k >>or 24/192k when you're mixing. But if the extra information doesn't help >>you do a better mix once you're down to 16/44.1, then what's the point of >>all the extra CPU cycles and disk storage? Ok many will disagree but here goes. use the highest rate format your gear will handle. Why? Because that will give you the cleanest, clearest original signal. If 10 years from now you decide to put out a DVD or SACD you will have a better product with higher original sample and bit rates. Now Relaize that if you are doing nmodern music (rap hip hop and such) most of which is compressed to beat the band and left with a 3 to 6 db dynamic range you will not notice much difference. In the end the cleaner and clearer the original source material is the better I.E.(cleaner and clearer) end product you can have (if that is your desire). When you start to process individual tracks the higher the sample and bit rates mean that you have better headroom and again less distortion from rounding errors and such. I am one of those folks who could hear a distinct difference between 16 and 24 bit and 48k, and 96k sampling. To me things had more air and were clearer. Many people hear nothing. I think that means that some peoples ears are more well trained or hear higher freqs but some may say it is just BS. I participated in some blind tests in a top flight studio and only once in 40 examples was I wrong. So for me the answer is simple. Use the highest rates you can get away with. Of course if you have a large number of tracks and high bit and sample rates you may run out or low on hard drive through put. Good luck, I hope this makes was close to what you were looking for. regards, dh
In <news:3fac9b55$0$3498$>, Adam Skeggs said: > > If we went for 1000 samples of a 20kHz waveform that would be a 20MHz > sample rate. > > I guess if we were dogs we would complain about the poor quality. > Lucky our ears are not much good. I can clearly hear the diff between a 44.1K and a 96K sampled sound if I listen to it really closely a few times. But does anyone really listen that closely to music over and over and over just to decide how "good" the recording is? ((U)) M
Hi On Fri, 7 Nov 2003 23:43:35 -0800, "Michael" <> wrote: >I can clearly hear the diff between a 44.1K and a 96K sampled sound if I >listen to it really closely a few times. > >But does anyone really listen that closely to music over and over and over >just to decide how "good" the recording is? Not the point. The point is that the cleaner and clearer it starts out the cleaner, clearer and more accurate your end result is. Now whether you find that to be a good thing is another completely different conversation. regards, dh
In <news:>, said: > Hi > > On Fri, 7 Nov 2003 23:43:35 -0800, "Michael" <> > wrote: > >> I can clearly hear the diff between a 44.1K and a 96K sampled sound >> if I listen to it really closely a few times. >> >> But does anyone really listen that closely to music over and over >> and over just to decide how "good" the recording is? > > Not the point. The point is that the cleaner and clearer it starts > out the cleaner, clearer and more accurate your end result is. Now > whether you find that to be a good thing is another completely > different conversation. > > regards, > > dh Yeah... thanx, Dave. I did read further ahead and realized I was shortsighted about what I hear on the first go 'round. A cleaner, more accurate sound can be fudged and fiddled to make it less so if you like, but a screwed one can't be really unscrewed. ((U)) M
Hi Michael, Exactly!! regards, dh On Sat, 8 Nov 2003 00:35:57 -0800, "Michael" <> wrote: >In <news:>, > said: > >> Hi >> >> On Fri, 7 Nov 2003 23:43:35 -0800, "Michael" <> >> wrote: >> >>> I can clearly hear the diff between a 44.1K and a 96K sampled sound >>> if I listen to it really closely a few times. >>> >>> But does anyone really listen that closely to music over and over >>> and over just to decide how "good" the recording is? >> >> Not the point. The point is that the cleaner and clearer it starts >> out the cleaner, clearer and more accurate your end result is. Now >> whether you find that to be a good thing is another completely >> different conversation. >> >> regards, >> >> dh > >Yeah... thanx, Dave. > >I did read further ahead and realized I was shortsighted about what I hear >on the first go 'round. > >A cleaner, more accurate sound can be fudged and fiddled to make it less so >if you like, but a screwed one can't be really unscrewed. > >((U)) > M >
"Adam Skeggs" <> wrote in message news:3fac9b55$0$3498$... > > "Adam Skeggs" <> wrote in message > news:3fac635c$0$3501$... > > Hi, > > > > I don't understand how a sample rate of say 44.1kHz can be any good at all > > at higher frequencies. I guess I am missing something but surely if the > > frequency being sampled is 20kHz then each cycle would be sampled only > twice > > (approx). If for instance you had a sine wave at 20kHz and happened to > > sample on the peaks (approx) then a sine wave would look just like a > square > > wave, ie, not representative of the waveform at all! > > > > Is it true that the higher the frequency the less accurate the sampling? > > > > Adam. > > > > > > A 1Hz sine wave would get 44100 samples so you could draw a convincing > picture of it. > A 20kHz sine wave would get 2 samples. Join up the dots on that. It could be > anything. Since demonstrably the sound is OK it must be that at high > frequency (and decreasingly at lower and lower frequencies) it doesn't > matter what the waveform is. Like a distant object to the eye, we can't see > the detail so as long as it looks roughly like a cow it's a cow. > > If we went for 1000 samples of a 20kHz waveform that would be a 20MHz sample > rate. > > I guess if we were dogs we would complain about the poor quality. Lucky our > ears are not much good. > > Adam. > > I did read the later posts I am just posting in this position for purposes of "narative flow". Another perspective: Say your minimum samples (points of sampling) on a sine wave was 10; I picked that because I think you could join up 10 dots and it would look pretty much like a sine wave though it would obviously look chunky (not smooth). It's arbitrary but I have decided to call everything with less than 10 sample points junk. On this "logic" at 44.1Khz the highest frequency to have at least 10 dots per cycle is 4410 Hz. This is getting to the upper limit of a guitar. So the upper limit of my guitar is dodgy. My cymbals are going to be complete crud (visually at least). So far so good. Going to 96kHz lifts my 10 sample point limit to 9600Hz. This is in cymbal land and should give me a quality crash. On the other hand, I use sampled drums which are probably not that good anyway. OK I think I get it now. Adam.
On Sat, 8 Nov 2003 23:44:10 +1100, "Adam Skeggs" <> wrote: >"Adam Skeggs" <> wrote in message >news:3fac9b55$0$3498$... >Another perspective: Say your minimum samples (points of sampling) on a sine >wave was 10; I picked that because I think you could join up 10 dots and it >would look pretty much like a sine wave though it would obviously look >chunky (not smooth). It's arbitrary Good approach to visualize things. In reality, since you know you are going to regenerate a sine wave, you can do it with fewer points. But the analogy is correct. > but I have decided to call everything >with less than 10 sample points junk. On this "logic" at 44.1Khz the highest >frequency to have at least 10 dots per cycle is 4410 Hz. This is getting to >the upper limit of a guitar. >So the upper limit of my guitar is dodgy. My cymbals are going to be >complete crud (visually at least). So far, fine, except that what makes an instrument sound like the instrument is the overtones. A guitar and trumpet play the same note -- the fundamental sine wave is identical. It is the overtones at higher frequencies that let you tell a trumpet from a guitar, and an acoustic guitar from a jazz electric, or shred metal at 11. >OK I think I get it now. BTW, electric guitars are two piece instruments, they have the guitar part, and the amplifier part. The tones are generated by the combination. Many of the high frequency tones in an electric guitar signal are caused by distortion and weirdness of the amplifier and the amp's speakers. Pat http://www.pfarrell.com/prc/
"Pat Farrell" <> wrote in news:: > On Sat, 08 Nov 2003 06:40:04 -0000, foldedpath > <> wrote: >>My question with high sample rates is always... how much of this >>really matters, when you're crunching the final product down to 16 bit >>44.1 kHz audio for distribution as MP3's and audio CD's? > > If you are aiming at MP3s, I believe that the answer is clearly "not > worth it, don't matter." For Red Book audio as a target.... > >>It's nice to hear (or think you hear) subtle nuances on a high-end set >>of monitors, in a perfectly tuned room, with your D/A running at 24 >>bits/96k or 24/192k when you're mixing. But if the extra information >>doesn't help you do a better mix once you're down to 16/44.1, then >>what's the point of all the extra CPU cycles and disk storage? > > Oh no, an engineering question! > Engineering is about deciding what is good enough. When do you > add more steel to a building's frame and when it added steel > just wasted money. > > It is not engineering to buy the best of every part to make > a cost is no object $10,000 preamp. It is engineering to make > a good preamp that sells for $500. Good point. > In simple terms, no group or body who has come into my studio > has an audience who cares. Same here, at least for now. > Right now, the CPU cycle budget is fine, and wasting a few cycles > is not a problem. In the days of Gigahertz processors, I don't run out > of tracks or effects cycles. YMMV, seriously. I don't worry about > the disk storage at all. Disks are nearly free. I do worry about > disk IO capacity, both during processing and when moving > the tracks/mix to another machine. Even over 100baseT, > moving a gigabyte takes real time. In my own situation, moving away from recording at 44.1 kHz to a higher sampling rate would actually mean ditching my current audio recording platform (Emu/Ensoniq Paris), since it's a "vintage" discontinued DAW that will never be updated for higher sample rates. Every once in a while, I re-visit this question... wondering if maybe I'm sticking to 44.1 kHz just to hold onto the sound quality and UI of this system (which I love). I have to be careful that I'm not fooling myself. But so far, I'm pretty convinced that I wouldn't gain anything meaningful by moving up to higher sample rates, and I have many things now in the Paris system that I'd miss if I switched systems. BTW, I do use Sonar... but only as a sequencer and soft synth host. I shoot all the DXi/VSTi audio via ADAT lightpipe into Paris when I'm working with audio. > I was trying to post an explaination of what higher sample > rates do. Whether we really can afford them is a totally different > topic. One that I think is a lot more useful than how to > get surround sound out of Sonar (or Vegas or Logic ...) > since all surrond sound is at best medium fidelity. > > Cycles for nothing and Disks for free.... > > Pat http://www.pfarrell.com/prc/ Gotcha, and that was good info, thanks. -- Mike Barrs
> A 1Hz sine wave would get 44100 samples so you could draw a convincing > picture of it. > A 20kHz sine wave would get 2 samples. Join up the dots on that. It could be > anything. Well... any detail in a 20KHz waveform represents frequencies higher than 20KHz. -S
OK, I'm going to try to clear up all this sampling munbo jumbo. In the first place, the filter in a CD player or soundcard output is the anti-imaging filter, not the anti-aliasing filter (more on anti-aliasing filters later). The fundimental problem with a building a CD anti-imaging filter is trying to go from 0dB in the passband to -80dB (or so) in the stopband in only 4.1 KHz. How did I get this number? Well, when you reproduce samples, the spectrum below the nyquist frequency (0-22.05 KHz) is "mirror imaged" above the nyquist frequency (22.05-44.1 KHz), so your theoretical maximum frequency of 20KHz will appear again at 24.1KHz. The purpose of the anti-imaging filter is to filter out this image. The problem is that it needs to be very steep, and steep filters are hard to make, and even when you do you get lots of ripples in the passband. The steeper the filter, the bigger the ripples, and going from 0 to -80dB in 4.1 KHz is very steep indeed. Steep filters also invariably change the phase of the frequencies as they get near the stopband, and many people think they can hear that effect. There are ways around the problem, a popular one being oversampling, where the sample rate is cranked way up (like 16 times 44.1 KHz) alowing the use of a very gentle analog filter and doing all the really hard filtering with a digital filer, which can be built very nearly perfect. What does this have to do with SONAR? Well, while there are people who claim to be able to hear above 20KHz, I tend to think the the improvement people hear at 96 or 192 KHz is due mostly to gentler anti-image filtering. Why should you sample your tracks at 96 KHz when you're just going to downsample them to 44.1 KHz anyway for a CD? That's where the anti-alias filter comes in. The anti-alias filter is the filter between your microphone and the A-D converter in your soundcard. Its got all the same problems as the anti-imaging filter, but any weaknesses in it will be much more obvious. For instance, any frequencies above the nyquist frequency that get through the anti-imaging filter, well, they're above human hearing anyway, so you probably won't notice them. But any frequencies above the nyquist frequency that get through the anti-alias filter get reflected down into the audible part of the spectrum (i.e. 35 KHz is reflected to 9.1 KHz at a sample rate of 44.1 KHz) where they are very obvious and un-musical. Since it's practically impossible to biuld a filter that goes from 0 to -80dB in 4.1 KHz, you're either going to loose some of your high end because the soundcard anti-alias filter starts earlier than 20 KHz, or you're going to get some reflections on source material that has very high frequency components. So sampling at 96 KHz could make a significant difference depending on what frequencies come out of your microphone (or guitar FX box) just because the anti-alias filters at 96 KHz have a lot more breathing room. As for why you should use 24 bit depth when the CD is only going to be 16 bits, well 16 bits is actually pretty OK for music reproduction, but not so good when you're recording. A CD will be normalized to use the entire 16-bit range, but when you're recording you usually give yourself some headroom because you do not want to peak in digital. With 16 bit recording, if you give yourself 6dB of headroom (which in many cases is pushing it), you're only really recording at 14 bits. Recording at 24 bits allows you to give yourself lots of headroom and still have 16 good bits worth of signal after you've run it through your compressors and other FX. Of course none of this matters if you're just using softsynths and pre-recorded samples. --John On Fri, 07 Nov 2003 23:23:12 -0500, "Pat Farrell" <> wrote: >On Sat, 08 Nov 2003 04:08:59 GMT, Dennis Bathory-Kitsz ><> wrote: >>Pat Farrell wrote: >>> On Sat, 8 Nov 2003 13:58:42 +1100, "Adam Skeggs" <> wrote: >>> >Is it true that the higher the frequency the less accurate the sampling? >>> No. Unless your sampling hardware is lame. Some consumer stuff >>> may not work well at high rates, but the consumers won't care. >> >>I read his question the other way around -- higher the frequency being >>sampled, not higher the frequency of sampling! Language, darn it. > >Had not thought of it that way. > >Lets try again, with a little more crispness in the language. > >The CD standard rate is 44.1kHz. Its Nyquist rate is >22.05 kHz. It can theoritically capture signal rates as >high as 22kHz. Which is higher than the 20-20k Hz range >of "human" hearing that audio engineers have used for >decades. > >But you must have an anti-aliasing filter at the Nyquist >rate, that is, a low pass filter that throws away everything >above 22kHz, because you will not be able to identify >and recreate signals (for example 24kHz) when sampled. > >In reality, filters are more like 6dB per octave, or 12dB per octave, >so real world analog filters start lower, say at 15kHz. >Starting lower is bad, we pay big bucks for mics and preamps >that go up. Or you can try using a brick wall filter that >starts at 20kHz, but they are evil, cause all sorts of >phase problems in the audible range. > >So, what do you do if you think that interharmonic >sounds exist in the 15kHz to 30kHz range that >you have to capture digitally? > >You have only one option, sample at a higher rate. >Sampling at 88.2 kHz will let you try to capture >signals at 44.1kHz, which means >you are likely to actually capture signals at 30kHz. > >Sampling at really high rates, say 192kHz would let >you capture any signal that exists in the 80kHz range. > >My personal belief is that humans can detect phase >relationships far better than they can detect frequency, >and higher sample rates help phase capture >as well. > >I am cautious about prosumer gear that claims to >record at high rates, many of them quote a rate >but without any supporting information (such >as distortion at a given bandwidth, or even >frequency response). > >Pat http://www.pfarrell.com/prc/
