Home |
Search |
Today's Posts |
#1
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
Most audio files on the net are recorded at a 44 KHz sampling rate,
but it's mainly referred as "frequency." Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? Also, just where the hell did the number 44,100 emerge from? Why not 40,000? Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? And if this ain't the case, why would the sampling rate be called "frequency?" |
#2
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Industrial One" wrote in message ... Most audio files on the net are recorded at a 44 KHz sampling rate, but it's mainly referred as "frequency." Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? Also, just where the hell did the number 44,100 emerge from? Why not 40,000? Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? And if this ain't the case, why would the sampling rate be called "frequency?" Sampling theory tells us that it takes at least two samples per cycle, hence the 44.1 KHz sample rate. The highest frequency that can be captured is 22.05 KHz (Nyquist frequency); frequencies higher than that will create alias frequencies below 22.05 For example, an audio frequency at 30 KHz would produce an alias frequency component at 14.1 KHz (44.1 - 30). It also produces one at 44.1 + 30, but who cares? The 20KHz audio upper limit allows for comfortable guard band to the Nyquist frequency. DVD audio is just for marketing. No one, with the possible exception of a few young people who can hear above 20 KHz, and many dogs, can hear the difference between regular 44.1K 16-bit audio and 96 or 192K sampling and 24 bits -- it's been proven, though some will tell you they can. It's something they call "resolution" for which they have an altar, dogma and lots of ritual. They get this dreamy look in their eyes. Challenge it, and their veins pop out and they go on rampages. It's likely that much of the stuff you get on DVD-audio discs is better stuff, and has been more meticulously recorded, hence the good sound of many of them. It ain't the extravagant bit depth and sampling rate. There are some damned good-sounding CDs too. Even if you had a regular CD version and a DVD-audio version, and the DVD-audio version sounded better, would you actually believe that the improvement was because of the bit depth and sample rate? Couldn't be anything else, could it? How are they going to sell DVD-audio discs if they let the CDs sound the same? Hope this helps. -- Earl |
#3
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Industrial One" wrote ...
Most audio files on the net are recorded at a 44 KHz sampling rate, The "Red Book" convention for making audio CDs was developed back in the early 1980s and established 44.1KHz as the sampling rate. In order to maintain forwards and backwards compatibility, all CDs must use that sample rate. but it's mainly referred as "frequency." Any periodic occurance can be referred to as "freqency". Whether it is something that happens every femtosecond (like light) or every 1000 years (like the century). Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? The Nyquist-Shannon sampling therom tells us that you must sample at *twice* the desired highest frequency to adequately reproduce the original waveform. It is said that 22KHz was selected as the top end (x2 = 44KHz) because of the state of the art in filters back in those days. Obviously, one can notice the difference if the song was downsampled to 22, You notice it because reducing the sampling rate to 22KHz actually reduces the top end to 11KHz which many people can detect. http://en.wikipedia.org/wiki/Compact_disc http://en.wikipedia.org/wiki/Red_Boo...CD_standard%29 http://en.wikipedia.org/wiki/Nyquist...mpling_theorem Also, just where the hell did the number 44,100 emerge from? Why not 40,000? It is said that 22KHz was selected as the top end to give some space between the theoretical maximum "hi-fi" frequency of 20KHz and the filter frequency (22KHz to allow room for the slope of the filter. Modern techniques make most of the original parameters moot. Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? No. And if this ain't the case, why would the sampling rate be called "frequency?" Any periodic occurrence can be referred to as "frequency". Whether it is something that happens every femtosecond (like light) or every 1000 years (like a new century). Most of us have to pay for electricity and our billing cycle happens with a frequency of one month. This is common scientific/engineering terminology. No great mystery. |
#4
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Richard Crowley" wrote in message ... Any periodic occurance can be referred to as "freqency". Whether it is something that happens every femtosecond (like light) or every 1000 years (like the century). Only *one* century every thousand years where you live? Snip Any periodic occurrence can be referred to as "frequency". Whether it is something that happens every femtosecond (like light) or every 1000 years (like a new century). Obviously not a typo then. MrT. |
#6
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Industrial One" wrote in message ... Most audio files on the net are recorded at a 44 KHz sampling rate, but it's mainly referred as "frequency." Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? Also, just where the hell did the number 44,100 emerge from? Why not 40,000? It was because they used video recorders for mastering prototype and first generation CDs, and it was the nearest available frequency that was greater than 40KHz needed to meet the sampling Nyquist requirement of at least two samples for the highest frequency to be recorded (20KHz). See also: http://www.cs.columbia.edu/~hgs/audio/44.1.html Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? Marketing. There is no defensible mathematical requirement for it. And if this ain't the case, why would the sampling rate be called "frequency?" The sampling rate is a frequency. CD audio is sampled at a frequency of 44.1KHz. |
#7
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
On 2008-07-06, Edmund wrote:
On Sun, 06 Jul 2008 04:18:32 +0000, Earl Kiosterud wrote: DVD audio is just for marketing. No one, with the possible exception of a few young people who can hear above 20 KHz, and many dogs, can hear the difference between regular 44.1K 16-bit audio and 96 or 192K sampling and 24 bits -- it's been proven, though some will tell you they can. I heard about that tests and it was criticized because the music was played over a pair of passive loudspeakers with passive filters that where nowhere near phase linear same problem with electrostatic speakers with step up transformers . So no matter how much better SACD or DVDA can be, played over such loudspeakers all the advantages are down the drain. I assume this refers to the Meyer & Moran paper in JAES (see http://www.aes.org/e-lib/browse.cfm?elib=14195). If so, I understand it was conducted over a number of different "high-res" systems but each test using the same 44.1 kHz/16 bit A/D - D/A loop. For example here's a subsequent comment from E. Brad Meyer: "... But it was a near certainty that someone in that part of the industry would claim that with a 'real audiophile system' the differences would have been obvious. So we found such a system and gave its owner and his friends a chance. We conducted that test with the same rigor as the others; levels of the two signals were matched within 0.1 dB at 1 kHz, and then the subjects were asked to choose their best material and listen however they usually do, to maximize their aural acuity." -- John Phillips |
#8
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Edmund" wrote in message
I heard about that tests and it was criticized because the music was played over a pair of passive loudspeakers with passive filters that where nowhere near phase linear As a rule, speakers are nowhere near phase linear, regardless of the implementation of the crossover. However, similar tests have been done with transducers that have better phase response, and same results. Furthermore, you are ignoring the fact that linear phase microphones are only a little bit easier to find, and as a rule they are not used to record music. same problem with electrostatic speakers with step up transformers . Same problem with 99,9% (more or less) of all loudspeakers ever made. So what? So no matter how much better SACD or DVDA can be, played over such loudspeakers all the advantages are down the drain. Even if you were right, you're basically admitting that SACD and DVDA have no real world application. Don't know if this story is true but I very much like to attend such a listening test an judge for myself. I doubt that, the tests are blind tests. Did anyone here did attend such a test and on what kind of speakers was it played? I can guarantee you that they weren't phase linear. |
#9
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Industrial One" wrote in message ... Most audio files on the net are recorded at a 44 KHz sampling rate, but it's mainly referred as "frequency." Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? Also, just where the hell did the number 44,100 emerge from? Why not 40,000? as others have noted... Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? not really for listening as least, but for "audio as data", higher sampling rate and bit depth makes sense. of course, it is rather unlikely that end users would be doing the levels of processing (notable doppler shifting, signal analysis, ...) that would justify this. of course, in my case, I am usually dealing with lower-quality input, so it works fairly well to just use 44.1kHz 16-bits anyways (one gets a better payoff trying to write good quality filtering functions). likewise, IMO, for MP3s much above about 128 or 192 kbps, I don't personally hear any real difference (actually, for internet radio at least, I prefer a little lower bitrates, as at least then the stream has less stalling and buffering issues). for something unrelated it is a thought that, for "audio as data" uses, one could stuff a little bit higher-range data into the 16-bit samples, by representing a slightly bigger range (say, 24 bits), as log-scaled values within the 16 bit range. of course, this would likely reduce quality a little, and add a little noise, if played back as 16 bit audio (mostly, really quiet stuff that would normally be quantized away is left). log10(8388608)=6.9236899, log10(32768)=4.51545 6.9236899/4.51545=1.53333333 .... actually, recently I have been suspecting that, mathematically (and for a defined dynamic range) log-scaled values may be much more accurate for a given number of bits than an actual floating-point format (in particular, I think that splitting the mantissa and exponent wastes some amount of the value range). of course, a claim like this would require testing, and even then, who would really care?... it only really makes that much difference for 24 bits and lower (where FP-style encoding starts breaking down anyways). yes, the accuracy of 16-bit half-floats suck, not much debate here. I guess the question is if log-scale values would be more accurate when covering the same range (2^16 to 2^-16). it may have practical relevance though, since log-scale values are quicker and easier to encode and decode than hfloats (given modern HW tends not to support them anyways). would still need to be verified though. And if this ain't the case, why would the sampling rate be called "frequency?" as others have noted. police also buy doughnuts with a high frequency... doesn't mean they are chirping at the cashier... |
#10
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
Industrial One wrote:
Most audio files on the net are recorded at a 44 KHz sampling rate, but it's mainly referred as "frequency." while 'frequency' is true of any periodic function (as noted in other posts), 'sampling rate' is much more precise. Essentially, the sample rate is the carrier of the information in the waveform. Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) technically, any waveform requires at least two samples per wave. I say 'technically', because to fully replicate the wave, with all of its harmonics, takes many more samples than two. at two samples, for example, you really cannot tell whether the waveform is sine, square, sawtooth, or whatever variant the harmonics may imbue to the waveform. Even at four samples, the original waveform is only approximated. (i.e. sine can look square or sawtooth depending upon at what degrees of arc the sample is taken.) that said, few humans can detect the difference between a sine, square, or sawtooth above 10k. But there is a difference. How we as humans interpret that difference is not easily quantifiable--some may speak in terms of clarity, crispness, 'air', 'musicality', or what have you--not very useful terms to the engineer. As well, sampling at a fixed frequency any other frequency pattern will result in sampling-induced harmonics, non-musical, that are in fact lower than the fundamental of the wave. the amplitude of these harmonics reduces greatly with increased number of samples per wave, so at under, say, 5K, they are not noticable, when sampling at 44.1k. finally, the lower the sampling frequency, the increased number of artifacts created when resampling. So, if one recorded some cuts at 44.1 and others at 48k, the movement back and forth to digitally (or analog) mix will have the effect of altering the higher-frequency waveforms in progressive generations of resampling. the bottom line is that sampling at 44.1k has a noticable and significant impact on all waveforms over 11k, and some modest impact over 5.5k. Impacts increase with successive generations of resampling (any time bit rate changes, and in particular, compression. Measuring whether humans can detect such a change is torturously difficult. Sampling at 96k raises the affected frequencies to the top of the hearing range, and 192k well beyond. to me, the greatest benefit of higher sampling rate is in the recording and mixing stages, where the higher rate eliminates any detectable resampling distoration (detectable=by a scope of by a computer image of the wave). prior to the 96 and 192 rates, i found that recording and mixing all within 44.1 resulted in less resampling bias and artifacts than switching between 48 and 44.1. Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? Also, just where the hell did the number 44,100 emerge from? Why not 40,000? again, sampling rate is NOT audio frequency. Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? again, recording at 96/192 may be a good idea, but it is definitely not required for playback. And if this ain't the case, why would the sampling rate be called "frequency?" because the frequency of a sine waveform involves a continuous function that goes both positive and negative during one complete 'wave'. There are generally two zero crossings during a wave. Let's say, for a moment, that I sampled the audio energy at each zero crossing--exactly two samples per waveform, or at exactly 22.05 KHz. Then, upon playback, i iterpolate a straight line between the samples. What waveform would result? -steve |
#11
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
steve wrote:
Industrial One wrote: Most audio files on the net are recorded at a 44 KHz sampling rate, but it's mainly referred as "frequency." while 'frequency' is true of any periodic function (as noted in other posts), 'sampling rate' is much more precise. Essentially, the sample rate is the carrier of the information in the waveform. Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) technically, any waveform requires at least two samples per wave. I say 'technically', because to fully replicate the wave, with all of its harmonics, takes many more samples than two. at two samples, for example, you really cannot tell whether the waveform is sine, square, sawtooth, or whatever variant the harmonics may imbue to the waveform. Even at four samples, the original waveform is only approximated. (i.e. sine can look square or sawtooth depending upon at what degrees of arc the sample is taken.) that said, few humans can detect the difference between a sine, square, or sawtooth above 10k. But there is a difference. How we as humans interpret that difference is not easily quantifiable--some may speak in terms of clarity, crispness, 'air', 'musicality', or what have you--not very useful terms to the engineer. As well, sampling at a fixed frequency any other frequency pattern will result in sampling-induced harmonics, non-musical, that are in fact lower than the fundamental of the wave. the amplitude of these harmonics reduces greatly with increased number of samples per wave, so at under, say, 5K, they are not noticable, when sampling at 44.1k. finally, the lower the sampling frequency, the increased number of artifacts created when resampling. So, if one recorded some cuts at 44.1 and others at 48k, the movement back and forth to digitally (or analog) mix will have the effect of altering the higher-frequency waveforms in progressive generations of resampling. the bottom line is that sampling at 44.1k has a noticable and significant impact on all waveforms over 11k, and some modest impact over 5.5k. Impacts increase with successive generations of resampling (any time bit rate changes, and in particular, compression. Measuring whether humans can detect such a change is torturously difficult. Sampling at 96k raises the affected frequencies to the top of the hearing range, and 192k well beyond. to me, the greatest benefit of higher sampling rate is in the recording and mixing stages, where the higher rate eliminates any detectable resampling distoration (detectable=by a scope of by a computer image of the wave). prior to the 96 and 192 rates, i found that recording and mixing all within 44.1 resulted in less resampling bias and artifacts than switching between 48 and 44.1. Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? Also, just where the hell did the number 44,100 emerge from? Why not 40,000? again, sampling rate is NOT audio frequency. Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? again, recording at 96/192 may be a good idea, but it is definitely not required for playback. And if this ain't the case, why would the sampling rate be called "frequency?" because the frequency of a sine waveform involves a continuous function that goes both positive and negative during one complete 'wave'. There are generally two zero crossings during a wave. Let's say, for a moment, that I sampled the audio energy at each zero crossing--exactly two samples per waveform, or at exactly 22.05 KHz. Then, upon playback, i iterpolate a straight line between the samples. What waveform would result? -steve Utter cock. And of course sampling theory states that you need MORE THAN two samples per wave, not at least two samples per wave. As long as you have more than two, the wave is uniquely and totally described - any further samples are unneeded and superfluous. The situation described in the paragraph above is exactly the reason why you need more than two samples; at exactly two the wanted and the first alias collide, and as they are 180 degrees out of phase with each other, they cancel. d d |
#12
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
Don Pearce wrote:
steve wrote: because the frequency of a sine waveform involves a continuous function that goes both positive and negative during one complete 'wave'. There are generally two zero crossings during a wave. Let's say, for a moment, that I sampled the audio energy at each zero crossing--exactly two samples per waveform, or at exactly 22.05 KHz. Then, upon playback, i iterpolate a straight line between the samples. What waveform would result? -steve Utter cock. ok. And of course sampling theory states that you need MORE THAN two samples per wave, not at least two samples per wave. ....so, if i require "more than two per wave", then are you saying that the highest frequency distinguishable from 44.1k is something less than 15k (44.1 divided by 3)? As long as you have more than two, the wave is uniquely and totally described - any further samples are unneeded and superfluous. only if you assume a perfect sine wave with no overtones. To achieve faithful replication of up to 20k, then at least 3 samples per wave at 20k would be necessary, true? I mean, 44.1 and 48k wouldn't cut it. only something greater than 60k. e.g. 96k. The situation described in the paragraph above is exactly the reason why you need more than two samples; at exactly two the wanted and the first alias collide, and as they are 180 degrees out of phase with each other, they cancel. not exactly. sure, the resultant wave would be 'nothing', but not because the sampling rate is out of phase with the frequency, but rather, perfectly in phase with the zero crossings. the sound pressure level at the zero crossings is zero (or -infinity from nominal). So, if the sample rate is, say 22k exactly, and the frequency rate is 11k exactly, depending upon when the sample measures the spl at that moment, you might get a perfect sawtooth or you might get nothing. of course, such a situation is not likely in the 'real world', because no natural sound source would be that perfectly in lock step with the sampling frequency. Nevertheless, 'just two' samples per wave is insufficient to faithfully replicate a waveform, and invites distortion. i would agree that 'more than two' are thus necessary. which, in my klunker way of putting it, was what i was trying to say in the first place. regards, -steve |
#13
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
On Jul 6, 3:27 pm, steve wrote:
Don Pearce wrote: steve wrote: because the frequency of a sine waveform involves a continuous function that goes both positive and negative during one complete 'wave'. There are generally two zero crossings during a wave. Let's say, for a moment, that I sampled the audio energy at each zero crossing--exactly two samples per waveform, or at exactly 22.05 KHz. Then, upon playback, i iterpolate a straight line between the samples. What waveform would result? -steve Utter cock. ok. And of course sampling theory states that you need MORE THAN two samples per wave, not at least two samples per wave. ...so, if i require "more than two per wave", then are you saying that the highest frequency distinguishable from 44.1k is something less than 15k (44.1 divided by 3)? Try 44.1k divided by 2.1, or 2.0000000001. Both are, mathematically speaking, more than 2 and fully meet the requirements of the Bysquist sampling theorem. Given the first case, that would indicate that the widest bandwidth possible if 44.1kHz/2.1 or 21 kHz. The latter would indicate the widest bandwidth is 21.0499999... kHz. As long as you have more than two, the wave is uniquely and totally described - any further samples are unneeded and superfluous. only if you assume a perfect sine wave with no overtones. No, the ONLY assumption of Nyquist Shannon is that ANY waveform, sinew wave, periodic, impulsive, noise- like, ANY WAVEFORM WHATSOEVER, be band-limited to less than 1/2 the sample rate. That's it. To achieve faithful replication of up to 20k, then at least 3 samples per wave at 20k would be necessary, true? False I mean, 44.1 and 48k wouldn't cut it. only something greater than 60k. e.g. 96k. If the bandwidth is limited to 20 kHz, and that is the bandwidth of ANY waveform, then a sample rate greater than twice that bandwidth is ALL that's needed. A 20 kHz waveform will NOT be better catpured at a sample rate of 48 kHz than at 44.1. The situation described in the paragraph above is exactly the reason why you need more than two samples; at exactly two the wanted and the first alias collide, and as they are 180 degrees out of phase with each other, they cancel. not exactly. sure, the resultant wave would be 'nothing', but not because the sampling rate is out of phase with the frequency, but rather, perfectly in phase with the zero crossings. the sound pressure level at the zero crossings is zero (or -infinity from nominal). So, if the sample rate is, say 22k exactly, and the frequency rate is 11k exactly, depending upon when the sample measures the spl at that moment, you might get a perfect sawtooth or you might get nothing. And you have just violated completely the tenets of the Nyquist-SHannon sampling theorem. Why would anyone reaosnably expect any usable results if you violate the operating principles? of course, such a situation is not likely in the 'real world', because no natural sound source would be that perfectly in lock step with the sampling frequency. Nevertheless, 'just two' samples per wave is insufficient to faithfully replicate a waveform, and invites distortion. And that's why NO ONE who knows anything says "exactly two samples per cycle." It's ALWAYS "more than two samples." And "more than two DOESN'T mean three, it means ANY number more than two, like 2.000000000000000000001. |
#14
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
On Jul 6, 12:58 pm, steve wrote:
technically, any waveform requires at least two samples per wave. No, "mathematically", to fully replicate a waveform requires MORE than two samples of the highest component of that wave. at two samples, for example, you really cannot tell whether the waveform is sine, square, sawtooth, or whatever variant the harmonics may imbue to the waveform. Even at four samples, the original waveform is only approximated. (i.e. sine can look square or sawtooth depending upon at what degrees of arc the sample is taken.) More than two, that's all you neeed, for the highest component of interest: end of discussion. that said, few humans can detect the difference between a sine, square, or sawtooth above 10k. I would challenge you to find said few humans. This claim is made over and over again and it's flawed beyond utility. The test is almost always made with a standard lab function generator whose output waveforem are normalized to have the same peak aimplitude regardless of the waveform., e.g., the sine wave is 1 V peak-to-peak and the square wave is 1 v peak-to-peak. Set the frequency to 10 kHz, and it's actually pretty easy for most people to tell the different between the switch set to sine wave and the switch set to square wave, but they are NOT hearing the difference between a 10 kHz sine and a 10 kHz square wave: They're hearing the difference between a 10 kHz sine wave with a peak-peak amplitude of 1 volt and a 10 kHz sine wave with a peak amplitude of about 1.28 volts. That 1.28 volt sine wave is the amplitude of the 10 kHz sine wave fundamental of your 1 volt peak square wave. That difference, almost 2 dB, is actually quite EASY to hear. Now, go find us these few humans that can hear the difference between a 10 kHz 1 volt P-P sine wave and a 10 kHz 0.786 volt P-P square wave, and now you might have something. But since no one else has survived the challenge, I'd not place any hard cash on you being the first. But there is a difference. Not when the signal is PROPERLY band-limited to less than 1/2 the same rate, there isn't, other than simple in- band amplitude differences. How we as humans interpret that difference is not easily quantifiable--some may speak in terms of clarity, crispness, 'air', 'musicality', or what have you--not very useful terms to the engineer. It's not very quantificable because those making the claims have never quantified it. As well, sampling at a fixed frequency any other frequency pattern will result in sampling-induced harmonics, non-musical, that are in fact lower than the fundamental of the wave. the amplitude of these harmonics reduces greatly with increased number of samples per wave, so at under, say, 5K, they are not noticable, when sampling at 44.1k. Complete and utter nonsense. As long as the signal is bandlimited to less than 1/2 the sample rate, no such artifacts exist. Period. That means any, repeat ANY waveform that is bandlimited to less than 1/2 the sample rate will be captured with NO additional artifacts. finally, the lower the sampling frequency, the increased number of artifacts created when resampling. Not as long as it is greater than twice the abndwidth, it won't. So, if one recorded some cuts at 44.1 and others at 48k, the movement back and forth to digitally (or analog) mix will have the effect of altering the higher-frequency waveforms in progressive generations of resampling. Not for ANY waveform band-limited to 20 kHz, it won't. the bottom line is that sampling at 44.1k has a noticable and significant impact on all waveforms over 11k, and some modest impact over 5.5k. Again, complete nonsense. Impacts increase with successive generations of resampling (any time bit rate changes, and in particular, compression. Measuring whether humans can detect such a change is torturously difficult. Actually, it's not. prior to the 96 and 192 rates, i found that recording and mixing all within 44.1 resulted in less resampling bias and artifacts than switching between 48 and 44.1. What on earth is "resampling bias?" And if this ain't the case, why would the sampling rate be called "frequency?" because the frequency of a sine waveform involves a continuous function that goes both positive and negative during one complete 'wave'. There are generally two zero crossings during a wave. Let's say, for a moment, that I sampled the audio energy at each zero crossing--exactly two samples per waveform, or at exactly 22.05 KHz. If you do this, you have violated Nyquist/SHannon. You have a broken, defective sampler. Why then take something that's broken and try to make any sense or draw any conclusion from the result: IT'S BROKEN! Then, upon playback, i iterpolate a straight line between the samples. During playback, you NEVER "interpolate a straight line." What waveform would result? Why would anyone care what waveform results from your broken sampler? IT'S BROKEN! |
#15
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
And of course sampling theory states that you need MORE THAN
two samples per wave, not at least two samples per wave. ...so, if i require "more than two per wave", then are you saying that the highest frequency distinguishable from 44.1k is something less than 15k (44.1 divided by 3)? Any decent CD player has a frequency response which is flat to within a small fraction of one dB, up to 20 kHz, and accurately resolves frequencies within that bandwidth. As long as you have more than two, the wave is uniquely and totally described - any further samples are unneeded and superfluous. only if you assume a perfect sine wave with no overtones. If it has overtones (i.e. harmonics), then by definition these harmonics have to be taken into account as part of the "highest frequency distinguishable". For example, a non-sinusoidal 20 kHz signal contains overtones (e.g. at 40 and 60 kHz, the second and third harmonics) and is actually a composite signal which has 60 kHz of bandwidth. This cannot be accurately sampled and reconstructed at a sampling rate of 44.1 ksamples/second. That's why signals must be low-pass-filtered before being sampled. It's an essential part of the process. To achieve faithful replication of up to 20k, then at least 3 samples per wave at 20k would be necessary, true? No, not true. You don't need an _integral_ number of samples per cycle (e.g. 3 or 4) to accurately distinguish, and reproduce, the signal. All that's required is that you have somewhat more than two. With CDs, a 20 kHz audio bandwidth, and a 44.1 kHz sampling rate, gives you a minimum of 2.2 samples per cycle (at the 20 kHz bandwidth limit) and more than that at lower frequencies. This is sufficient to accurately reproduce the signal. This may seem counter-intuitive, but it actually does work (both in practice, and in the underlying mathematics). I mean, 44.1 and 48k wouldn't cut it. only something greater than 60k. e.g. 96k. This turns out not to be the case. 44.1 ksamples/second *does* allow the accurate sampling, and reconstruction, of an audio signal with 20 kHz of bandwidth. There are two *essential* steps in this process. You *must* filter the incoming continuous signal before you sample it, to ensure that it actually has no more than 20 kHz of bandwidth (i.e. you must filter out any individual signals, or harmonics/overtones which lie above 20 kHz). This is usually known as the "anti-aliasing filter" step. Then, when you convert the samples back to continuous form, you *must* run the samples through another bandwidth-limiting filter (again, DC to 20 kHz in the case of CDs) to eliminate the image frequencies lying above 20 kHz. This is usually referred to as the "reconstruction filter". of course, such a situation is not likely in the 'real world', because no natural sound source would be that perfectly in lock step with the sampling frequency. Nevertheless, 'just two' samples per wave is insufficient to faithfully replicate a waveform, and invites distortion. i would agree that 'more than two' are thus necessary. which, in my klunker way of putting it, was what i was trying to say in the first place. You're correct. You need more than two. You just don't need all that much *more* than two. You don't need three. 2.2 turns out to be sufficient, if you do a proper job implementing the anti-aliasing and reconstruction filters. -- Dave Platt AE6EO Friends of Jade Warrior home page: http://www.radagast.org/jade-warrior I do _not_ wish to receive unsolicited commercial email, and I will boycott any company which has the gall to send me such ads! |
#16
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
On Jul 5, 11:47 pm, Industrial One wrote:
Most audio files on the net are recorded at a 44 KHz sampling rate, but it's mainly referred as "frequency." Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) Others have chimed in (regretably, some incorrectly), but the short answer is: the nathematical basis behind periodic sampling tells us that if we have a singnal whose bandwidth is, oh, "x", to capture that sample that signal with no loss of time-domain information or have no unwanted artifacts, we are required to sampke that signal at MORE THAN twice "x". For example, if you assume the bandwidth of human hearing is 20 kHz, you must sample at more than 2*20 kHz or GREATER THAN 40 kHz to ensure that everything within that 20 kHz bandwidth is captured and not lost. One of the MOST important parts of a properly implemented sampler is the preceeding band-limiting filter. ALL operational sampler (no exceptions) provide some means of ensureing that NO components outside that bandwidth reach the sampler. What it comes down to is this: IF you can assume, a priori, that your bandwidth is LESS THAN 1/2 the sampling rate, ALL waveforms within that bandwidth can and are uniquely identified by all available samples. If you have, say, 2.0001 samples per cycle of, say, a sine wave, there is exactly one and ONLY one waveform whose bandwidth is less than 1.2 the sample rate that can pass through those samples. Adding more samples will NOT make the representation of that waveform ANY more precise: it will simply waste data. Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? Because, if you follow the discussion about the requirements of a sampler, when you have down-sampled to 22 kHz, you have to first filter EVERYTHING that's at or above half that, or 11 kHz. And many people have no problem hearing the intrusion of a 11kHz low pass filter on the right kind of musical material. And, to repeat, it's not doubled, it's multiplied by SLIGHTLY more than double. Also, just where the hell did the number 44,100 emerge from? Why not 40,000? Way back in the late 1970's, the only form of portable recordable storage that was affordable for the kind of data rates needed for digital audio were video tape recorders. The samplers used the vertical modulation from white to black to store 1's and 0's. To ease the designof the samplers and the synchronization and to meet the bandwidth limts of the recorders, it was decide to put an intergral number of samples on each scan line. For 60 Hz/525 line NTSC, you have 35 blanked lines, leaving 490 lines per frame, 245 line for field. Storing 3 samples per line, the resulting sample rate becomes: 60 field/second * 245 lines/field * 3 samples/line = 44.1 kHz samples per second. Similar calculations yield sample rates of 48 kS/s and 50 Hz/625 PAL video can also accomodate these rates in similar fashions. One very interesting side benefit is not only could you store wide-band audio this way, you could also transmit the resulting digitized audio over normal broadcast TV channels with no loss. During the '80s' a number of stations did just that: if you had a compatible D/A converter, you could listen to full bandwidth digital audio at home from your TV set. Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? And if this ain't the case, why would the sampling rate be called "frequency?" Becasue "frequency, in the technical parlance, means quite precisely "per unit time" Whether it's cycles per second or sample per second or high tides per day, they all describe the frequency, or how often, at which some semi-periodic event happens. |
#17
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
steve wrote:
Don Pearce wrote: steve wrote: because the frequency of a sine waveform involves a continuous function that goes both positive and negative during one complete 'wave'. There are generally two zero crossings during a wave. Let's say, for a moment, that I sampled the audio energy at each zero crossing--exactly two samples per waveform, or at exactly 22.05 KHz. Then, upon playback, i iterpolate a straight line between the samples. What waveform would result? -steve Utter cock. ok. And of course sampling theory states that you need MORE THAN two samples per wave, not at least two samples per wave. ...so, if i require "more than two per wave", then are you saying that the highest frequency distinguishable from 44.1k is something less than 15k (44.1 divided by 3)? No, less than 22.05. I said more than 2, not more than 3. As long as you have more than two, the wave is uniquely and totally described - any further samples are unneeded and superfluous. only if you assume a perfect sine wave with no overtones. To achieve faithful replication of up to 20k, then at least 3 samples per wave at 20k would be necessary, true? I mean, 44.1 and 48k wouldn't cut it. only something greater than 60k. e.g. 96k. This is where you need to understand a bit of theory, and maybe it isn't all that intuitive. If you have a signal that is limited (by filtering) to a bandwidth less than 22.05, then those two-and-a-bit samples per wave describe it perfectly and unambiguously. There is only one solution possible for the DAC to work on, and that is the identical wave that was present at the input. The situation described in the paragraph above is exactly the reason why you need more than two samples; at exactly two the wanted and the first alias collide, and as they are 180 degrees out of phase with each other, they cancel. not exactly. sure, the resultant wave would be 'nothing', but not because the sampling rate is out of phase with the frequency, but rather, perfectly in phase with the zero crossings. the sound pressure level at the zero crossings is zero (or -infinity from nominal). So, if the sample rate is, say 22k exactly, and the frequency rate is 11k exactly, depending upon when the sample measures the spl at that moment, you might get a perfect sawtooth or you might get nothing. Could be zero crossings, or any other point up and down the wave - depends on where the sampling happens during the cycle. of course, such a situation is not likely in the 'real world', because no natural sound source would be that perfectly in lock step with the sampling frequency. Nevertheless, 'just two' samples per wave is insufficient to faithfully replicate a waveform, and invites distortion. i would agree that 'more than two' are thus necessary. which, in my klunker way of putting it, was what i was trying to say in the first place. No, you were saying that much more than 2 were needed. That simply isn't so. d |
#18
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
|
#19
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
Industrial One writes:
Most audio files on the net are recorded at a 44 KHz sampling rate, but it's mainly referred as "frequency." Now, humans can only hear up to 20 KHz, so why would audio be recorded at 44 KHz (twice the audible hearing range?) Obviously, one can notice the difference if the song was downsampled to 22, so why not coin the standard frequency at 22 KHz instead of 44, why is the number doubled? What you're missing is that the bandwidth of a digital system is HALF the sample rate. So sampling at 44.1 kHz passes (potentially) a signal with frequencies up to 22.05 kHz. Also, just where the hell did the number 44,100 emerge from? Why not 40,000? See 44,100 and 44,056 he http://en.wikipedia.org/wiki/Sampling_rate Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? Yes - to make Sony and other media moguls more money (by requiring people to replace their collections). Other than that, no. And if this ain't the case, why would the sampling rate be called "frequency?" Simple laziness - humans get lazy with terms. -- % Randy Yates % "Ticket to the moon, flight leaves here today %% Fuquay-Varina, NC % from Satellite 2" %%% 919-577-9882 % 'Ticket To The Moon' %%%% % *Time*, Electric Light Orchestra http://www.digitalsignallabs.com |
#20
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Randy Yates" wrote in message ... Industrial One writes: [snip] And if this ain't the case, why would the sampling rate be called "frequency?" Simple laziness - humans get lazy with terms. I do not think the term frequency is being used improperly. A sample rate has a frequency: Q: "How frequently are you sampling?" A: "44.1 thousand times a second -- at a frequency of 44.1KHz" |
#21
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
To take a contrarian position he
Here is a good article for understanding sampling theory: http://www.wescottdesign.com/article.../sampling.html When you sample a signal, you have to tradeoff between frequency response, aliasing, and ringing artifacts. For audio I believe it's ok to have ringing since we don't notice it. On reproducing that signal, there is that set of tradeoffs a second time. So you can lose frequency response there again. In practical systems, you aren't working with idealized sinc filters (brickwall) so there is some dropoff in frequency response when you sample that signal and again when you reproduce it as sound. So depending on what analog filters cost, etc. etc. there might be some sense in going with 96khz systems. It definitely does make sense to sample at 96khz at acquisition... the oversampling is beneficial (if you sample at 48khz, you can't get very good frequency response because the analog filters won't let you do that). 2- Anyways this is just speculating. The real way to figure it out is to do a test. Unfortunately I haven't done so myself. But according to one audio engineer, there is an audible difference. So maybe there is merit to 96khz systems. Do read the sampling article as it provides a better understanding of what goes on. http://www.prorec.com/Articles/tabid...8/Default.aspx QUOTE: I know... I know... I can hear many of you saying there is absolutely NO need for recording with a 96kHZ Sample Rate. Two weeks ago, I would have agreed with you! I emphasize *would have* agreed with you! Let me state this very clearly... YOU CAN INDEED HEAR THE DIFFERENCE when recording with a 96kHz Sample Rate! I wouldn't have believed it myself if I hadn't heard the results. Bottom line is that the highs sound more open and detailed. By the way... two other folks here in my studio could pick the 96kHz track EVERY time in a blind listening test (when compared with a 44.1kHz version). To hell with theory, my EARS tell me there is a difference. Want a real dose of Blasphemy? I compared recording at 96kHz and Sample Rate converting down to 44.1, to simply recording at 44.1kHz. I couldn't believe my ears! The track originally recorded at 96kHz and Sample Rate converted down to 44.1kHz had much better sounding highs, maintaining much of the character from recording at 96kHz. This goes against everything that I have learned over the years... and goes against accepted practice. So I don't make this statement lightly! You CAN hear a difference... anyone who tells you otherwise hasn't tried recording at 96kHz! Period. |
#22
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
|
#23
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
On Jul 6, 6:29 pm, wrote:
wrote: Now, go find us these few humans that can hear the difference between a 10 kHz 1 volt P-P sine wave and a 10 kHz 0.786 volt P-P square wave, and now you might have something. But since no one else has survived the challenge, I'd not place any hard cash on you being the first. \ That's a TOUGH test, since the 2nd harmonic (at 20 kHz) is missing from a square wave. It's also missing from a symmetric triangle wave. The lowest harmonic present is at 30 kHz. But it is the test often cited as "proving" that one can hear the difference. It's not my choice of a test, to be sure, and becasue of the way it's normally suggested it be conducted, it's a trivially easy test to shoot down because of its flaws. Yet we see it being brought up over and over again. A better test would be a 10 kHz sine wave with an added second harmonic (i.e. 20 kHz.) Nevertheless, I HAVE heard the difference between a 10 kHz since wave and a 10 kHz square wave with the same amount of the 10 kHz component, on a speaker. AND I actually know WHY I heard it. Using a 100 kHz bandwidth HP instrumentation mike and a computer with a 200 kHz sampling rate National Instruments card, with a 80 kHz multiple low pass filter, it is quite clear the the speaker I used was generating not-harmonicly-related buzzing due to nonlinearities, as well as having the added harmonics change the level of the produced 10 kHz. But, of course, you're not hearing the difference between a 10 kHz sine wave and a 10 kHz square wave, you're hearing the difference between a 10 kHz sine wave and some other wave, BOTH of which have components within the audible bandwidth, and as such do nothing to support the case that the information above 20 kHz has an audble effect. And that's not even worrying about ordinary IM distortion (say, between a 20 kHz sine wave at its alias at 24.1 kHz, which is of course at 4.1 kHz) in speakers. The alias would only occur anyway if the system was badly implemented. Further, on musical material it's not likely that you will EVER find energy at high enough frequencies such that these sorts of artifacts will have a sufficient level to overcome the masking effects of that material that's already there to begin with. Claims of these kinds suffer from all sorts of serious and funcdamental problems. It suggests that many of the claimants and proponents of wide-band audio touting them have very likely NEVER conducted the experiments and very likely never even heard a situation where they MIGHT occur, because the basis of the claims themselves are so dubious and poorly thought out. |
#24
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
On Jul 7, 5:01 am, wrote:
When you sample a signal, you have to tradeoff between frequency response, aliasing, and ringing artifacts. For audio I believe it's ok to have ringing since we don't notice it. A fundamental error here. First, sampling does not cause the "ringing," it's the truncation of the bandwidth. You can have all sorts of ringing in a continuous time analog system. Look at the output, for example, of an old-style analog anti-aliasing filter that preceeded early generation A/D converters. Take a 1 kHz square wave. Band limit it with a 20 kHz low-bass filter. Now, sum the following series: F(t) = sum (sin(x*t)/x) where x = 2 pi * 1, 3, 5, ... 19 and see what you get. Is the ringing in the first case real and in the second case simply a result of truncation of a mathematical series? On reproducing that signal, there is that set of tradeoffs a second time. So you can lose frequency response there again. Why? How? I have inexpensive A/D and D/A chains here that have frequency response from 2 Hz to 20 kHz with less than +-.2 dB total error across the band and with a phase response 20-20 kHz with 5 degrees or 0. What "lost frequency response" are you talking about? In practical systems, you aren't working with idealized sinc filters (brickwall) so there is some dropoff in frequency response when you sample that signal and again when you reproduce it as sound. I don't sample it again and again when I reproduce it. So depending on what analog filters cost, etc. etc. there might be some sense in going with 96khz systems. The analog filter in the BEST 44.1 kHz digital system I have here costs on the order of a buck or two and has almost NO imact whatsoever on the frequency response within the 20 kHz audio bandwidth. It definitely does make sense to sample at 96khz at acquisition... the oversampling is beneficial (if you sample at 48khz, you can't get very good frequency response because the analog filters won't let you do that). That's why NO ONE uses analog filters to do this job. THat's why no one with any competence has used analog anti-aliasing and anti-imaging filters for two decades. And what you're talkig about is NOT "oversampling." In fact, for the last two decades, MOSTY A/D and D/A systems HAVE used oversampling techniques to elininate the issues surrounding analog filters. And the run not at 96 kHz or 192, but are 44.1 or 48 kHz oversampled systems. 2- Anyways this is just speculating. The real way to figure it out is to do a test. Unfortunately I haven't done so myself. But according to one audio engineer, there is an audible difference. And, according to many independent researchers, there is not. |
#25
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
"cr88192" wrote in message ... "Industrial One" wrote in message ... for something unrelated it is a thought that, for "audio as data" uses, one could stuff a little bit higher-range data into the 16-bit samples, by representing a slightly bigger range (say, 24 bits), as log-scaled values within the 16 bit range. of course, this would likely reduce quality a little, and add a little noise, if played back as 16 bit audio (mostly, really quiet stuff that would normally be quantized away is left). Along a similar line, I've wondered why we didn't non-linearize the the digital audio values for 16-bit audio. You get less noise at low signal levels,, though more at higher levels where it would be masked. Back in the 70's, I was experimenting with 8-bit ADC and DAC chips, which were a bit noisy, as you would expect for 8 bits. The noise was constant with regard to signal level (it sounded a lot like a damaged speaker). I rigged up a non-linearizing circuit before the ADC (it looked somewhat like very soft clipping), and a complementary one after the DAC. The noise was substantially less noticeable. The quiet parts of the audio had less noise, and the louder parts, even though the noise was actually worse, didn't really sound so, because of masking. The function couldn't quite be called logarithmic, as the values went through zero, but were close for most values. Floating-point audio achieves something similar. If nonlinearizing the transfer function was considered when CD audio was being standardized, I suspect that one reason it was rejected was that it would open arguments about the accuracy of the system. The complementary function in a CD player would have to exactly match that of the CD, or there would be amplitude distortion. I think it'd have to have been done in analog -- doing it in the digital domain would ensure compatibility, but wasn't practical back then -- it would have required DSPs and converters with greater than 16 bits, impractical for the time for consumer stuff. -- Earl |
#26
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
Earl Kiosterud wrote:
"cr88192" wrote in message ... "Industrial One" wrote in message ... for something unrelated it is a thought that, for "audio as data" uses, one could stuff a little bit higher-range data into the 16-bit samples, by representing a slightly bigger range (say, 24 bits), as log-scaled values within the 16 bit range. of course, this would likely reduce quality a little, and add a little noise, if played back as 16 bit audio (mostly, really quiet stuff that would normally be quantized away is left). Along a similar line, I've wondered why we didn't non-linearize the the digital audio values for 16-bit audio. You get less noise at low signal levels,, though more at higher levels where it would be masked. Back in the 70's, I was experimenting with 8-bit ADC and DAC chips, which were a bit noisy, as you would expect for 8 bits. The noise was constant with regard to signal level (it sounded a lot like a damaged speaker). I rigged up a non-linearizing circuit before the ADC (it looked somewhat like very soft clipping), and a complementary one after the DAC. The noise was substantially less noticeable. The quiet parts of the audio had less noise, and the louder parts, even though the noise was actually worse, didn't really sound so, because of masking. The function couldn't quite be called logarithmic, as the values went through zero, but were close for most values. Floating-point audio achieves something similar. If nonlinearizing the transfer function was considered when CD audio was being standardized, I suspect that one reason it was rejected was that it would open arguments about the accuracy of the system. The complementary function in a CD player would have to exactly match that of the CD, or there would be amplitude distortion. I think it'd have to have been done in analog -- doing it in the digital domain would ensure compatibility, but wasn't practical back then -- it would have required DSPs and converters with greater than 16 bits, impractical for the time for consumer stuff. This is done already for telephone lines - and very successfully. There are two system - A-Law and Mu-Law; the first is international and the second local to the USA. If there were such a system applied to 16-bit systems, we could be seeing better than 24 bit performance. But first agreeing and then constructing the necessary variable slopes was way beyond the technology of the day. It could still be done very simply during a bit-reduction process from 24 to 16 bits - just a mathematical operation. Then at the player end those 16 A-law bits could be reconstructed back to 24 very quiet bits. Of course all this would save is some real estate on the recording medium. Interesting project, though. d |
#27
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
Don Pearce wrote:
If nonlinearizing the transfer function was considered when CD audio was being standardized, I suspect that one reason it was rejected was that it would open arguments about the accuracy of the system. The complementary function in a CD player would have to exactly match that of the CD, or there would be amplitude distortion. I think it'd have to have been done in analog -- doing it in the digital domain would ensure compatibility, but wasn't practical back then -- it would have required DSPs and converters with greater than 16 bits, impractical for the time for consumer stuff. This is done already for telephone lines - and very successfully. There are two system - A-Law and Mu-Law; the first is international and the second local to the USA. If there were such a system applied to 16-bit systems, we could be seeing better than 24 bit performance. But first agreeing and then constructing the necessary variable slopes was way beyond the technology of the day. That would not work in a single-band system. Consider a piece of music with a high loudness bass drum and a very soft flute. The high absolute value part of the wave of the drum would be in a part of teh amplitude regime where the sampling points (in amplitude) were far apart, and the flute signal might be missed entirely when the bass amplitude was low. That's why MP3 is an agressively multiband system. It **IS** such a system, which works well. Doug McDonald |
#28
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
|
#29
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
On Jul 5, 9:18 pm, "Earl Kiosterud" wrote:
Sampling theory tells us that it takes at least two samples per cycle, hence the 44.1 KHz sample rate. The highest frequency that can be captured is 22.05 KHz (Nyquist frequency); frequencies higher than that will create alias frequencies below 22.05 For example, an audio frequency at 30 KHz would produce an alias frequency component at 14.1 KHz (44.1 - 30). It also produces one at 44.1 + 30, but who cares? The 20KHz audio upper limit allows for comfortable guard band to the Nyquist frequency. Do I understand correct: hz is one sine loop per second, I generate a sine sweep from 0 - 20 KHz with a specified duration, when I view with an audio application and zoom 'till individual samples are visible, I notice that as frequency increases, the sine waves become shorter, and gradually begin to appear more triangular as the smaller sample interval makes a perfect, smooth sine shape impossible. Finally, when it reaches 20 KHz (20,000 sampling rate) the waves have reached their limit on appearing anything that resembles a sine, and is now a perfect triangle: one sample at the bottom, one at the top, and one at the bottom again, like /\/\/\/\/\/\/\/\/\/\. This would technically be the maximum, but instead, as I continue scrolling, I see the waveform look something like a private-case of sine waves. This time, a sine block composed of triangles. What you're saying is that beyond 22.05 is a hack that simulates higher frequencies, but don't technically exist on a digital medium, like the waveform of the sine sweep I created? DVD audio is just for marketing. No one, with the possible exception of a few young people who can hear above 20 KHz, and many dogs, can hear the difference between regular 44.1K Pffft... I'm 18 and I can hear 17 KHz maximum, assuming 20 KHz for anyone is an exaggeration, and whoever claims to tell the difference between 44/96 is some autistic mother****er that probably ****es himself in class 'cuz he keeps hearing the "whistling" from the rat repellant. Also, I've subtracted a 44 track by a downsampled 32 version to hear what EXACTLY is stripped and all I heard was extremely faint clicks, jingles and dings (from the beat of the treble percussion instruments) and had to amplify the waveform up 30 dB to hear it clearly (real headache inducer, yo.) So... what I'm really missing from a 96 KHz track are some faint jingles TWICE as high- pitched and inaudible from the upper freqs of a 44 KHz track? WOW........ is THIS really what those assramming audiophiles bitch about? 16-bit audio and 96 or 192K sampling and 24 bits -- it's been proven, though some will tell you they can. It's something they call "resolution" for which they have an altar, dogma and lots of ritual. They get this dreamy look in their eyes. Challenge it, and their veins pop out and they go on rampages. It's likely that much of the stuff you get on DVD-audio discs Proof drugs are awesome. is better stuff, and has been more meticulously recorded, hence the good sound of many of them. It ain't the extravagant bit depth and sampling rate. There are some damned good-sounding CDs too. Even if you had a regular CD version and a DVD-audio version, and the DVD-audio version sounded better, would you actually believe that the improvement was because of the bit depth and sample rate? Couldn't be anything else, could it? How are they going to sell DVD-audio discs if they let the CDs sound the same? QFT. Hope this helps. -- Earl Was informative, thanks. On Jul 5, 10:31 pm, "Richard Crowley" wrote: The "Red Book" convention for making audio CDs was developed back in the early 1980s and established 44.1KHz as the sampling rate. In order to maintain forwards and backwards compatibility, all CDs must use that sample rate. Any periodic occurance can be referred to as "freqency". Whether it is something that happens every femtosecond (like light) or every 1000 years (like the century). The Nyquist-Shannon sampling therom tells us that you must sample at *twice* the desired highest frequency to adequately reproduce the original waveform. It is said that 22KHz was selected as the top end (x2 = 44KHz) because of the state of the art in filters back in those days. I see. Obviously, one can notice the difference if the song was downsampled to 22, You notice it because reducing the sampling rate to 22KHz actually reduces the top end to 11KHz which many people can detect. I don't follow. Why 11? On Jul 6, 2:06 am, "Chronic Philharmonic" wrote: It was because they used video recorders for mastering prototype and first generation CDs, and it was the nearest available frequency that was greater than 40KHz needed to meet the sampling Nyquist requirement of at least two samples for the highest frequency to be recorded (20KHz). See also:http://www.cs.columbia.edu/~hgs/audio/44.1.html Damn, I didn't know that. Nowadays, DVD-audio songs are recorded at 96/192 KHz, is there a point? Marketing. There is no defensible mathematical requirement for it. 's what I thought. I wouldn't be surprised if them retard scene rippers start releasing audio in 192 KHz with a bitrate of probably 1024+ and advertise "DVD QUALITY AUDIO!" I'll read the other posts later. |
#30
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
Industrial One wrote:
) Do I understand correct: hz is one sine loop per second, Hz is one whatever per second. Your heart beats at roughly 80 Hz. The moon revolves around the earth at roughly 0.38 uHz (Microhertz) ) I generate a ) sine sweep from 0 - 20 KHz with a specified duration, when I view with ) an audio application and zoom 'till individual samples are visible, I ) notice that as frequency increases, the sine waves become shorter, and ) gradually begin to appear more triangular as the smaller sample ) interval makes a perfect, smooth sine shape impossible. Finally, when ) it reaches 20 KHz (20,000 sampling rate) the waves have reached their ) limit on appearing anything that resembles a sine, and is now a ) perfect triangle: one sample at the bottom, one at the top, and one at ) the bottom again, like /\/\/\/\/\/\/\/\/\/\. This would technically be ) the maximum, but instead, as I continue scrolling, I see the waveform ) look something like a private-case of sine waves. This time, a sine ) block composed of triangles. What you're saying is that beyond 22.05 ) is a hack that simulates higher frequencies, but don't technically ) exist on a digital medium, like the waveform of the sine sweep I ) created? Play your sample at half speed and listen to what happens when you reach 11.025 Khz. ) Obviously, one can notice the difference if the song was ) downsampled to 22, ) ) You notice it because reducing the sampling rate to 22KHz ) actually reduces the top end to 11KHz which many people ) can detect. ) ) I don't follow. Why 11? Because 11 is half of 22. See above. SaSW, Willem -- Disclaimer: I am in no way responsible for any of the statements made in the above text. For all I know I might be drugged or something.. No I'm not paranoid. You all think I'm paranoid, don't you ! #EOT |
#31
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
"Industrial One" wrote in message ... On Jul 5, 9:18 pm, "Earl Kiosterud" wrote: Sampling theory tells us that it takes at least two samples per cycle, hence the 44.1 KHz sample rate. The highest frequency that can be captured is 22.05 KHz (Nyquist frequency); frequencies higher than that will create alias frequencies below 22.05 For example, an audio frequency at 30 KHz would produce an alias frequency component at 14.1 KHz (44.1 - 30). It also produces one at 44.1 + 30, but who cares? The 20KHz audio upper limit allows for comfortable guard band to the Nyquist frequency. Do I understand correct: hz is one sine loop per second, I generate a sine sweep from 0 - 20 KHz with a specified duration, when I view with an audio application and zoom 'till individual samples are visible, I notice that as frequency increases, the sine waves become shorter, and gradually begin to appear more triangular as the smaller sample interval makes a perfect, smooth sine shape impossible. Finally, when it reaches 20 KHz (20,000 sampling rate) the waves have reached their limit on appearing anything that resembles a sine, and is now a perfect triangle: one sample at the bottom, one at the top, and one at the bottom again, like /\/\/\/\/\/\/\/\/\/\. This would technically be the maximum, but instead, as I continue scrolling, I see the waveform look something like a private-case of sine waves. This time, a sine block composed of triangles. What you're saying is that beyond 22.05 is a hack that simulates higher frequencies, but don't technically exist on a digital medium, like the waveform of the sine sweep I created? With regard to your swept sine, as you go towards the higher frequency sines, you'll see fewer and fewer samples per cycle, until there are barely more than two per cycle at 20 KHz. Your audio program should not connect them with straight lines. If anything, it should show them as a post filter (a brick-wall at 20KHz, probably) would see them. That is the waveform of those samples with the above-Nyquist (above 22.05 KHz) frequency components removed. It should draw a sine. Anything above 10 KHz should be sinusoidal, since any other function (waveshape) would need harmonics, which would fall above the 20 KHz point, and could not appear. For example, there ain't no such thing as a 15KHz triangle, sawtooth, square etc., wave in audio. If there were, we'd only hear the fundamental, and that, by definition, is a single frequency component, thus a sine. A sine has only one frequency component in the spectrum, and a single frequency component in the spectrum is a sine. There's your sine. -- Earl |
#32
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
In article ,
Willem wrote: Industrial One wrote: ) Do I understand correct: hz is one sine loop per second, Hz is one whatever per second. Your heart beats at roughly 80 Hz. I think you mean the heart beats 1.333 Hz. PS. |
#33
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
On Jul 7, 11:02 am, Industrial One wrote:
Do I understand correct: hz is one sine loop per second, No, "Hz" simply is the abbreviation for "Hertz" which is the SI unit for reciprocal second. It has nothing to do per se with sine waves, square waves, are waves of any kind,. Rather it is simply the rate at which whatever semi-periodic phenomenon occurs. It could be the rotational period of a neutron star, like 792 Hz. I generate a sine sweep from 0 - 20 KHz with a specified duration, when I view with an audio application and zoom 'till individual samples are visible, I notice that as frequency increases, the sine waves become shorter, and gradually begin to appear more triangular as the smaller sample interval makes a perfect, smooth sine shape impossible. Finally, when it reaches 20 KHz (20,000 sampling rate) the waves have reached their limit on appearing anything that resembles a sine, and is now a perfect triangle: one sample at the bottom, one at the top, and one at the bottom again, like /\/\/\/\/\/\/\/\/\/\. This would technically be the maximum, but instead, as I continue scrolling, I see the waveform look something like a private-case of sine waves. This time, a sine block composed of triangles. What you're saying is that beyond 22.05 is a hack that simulates higher frequencies, but don't technically exist on a digital medium, like the waveform of the sine sweep I created? Uhm, I think, if I am able to parse what you are saying, yes, though your grammar is a bit tortured. But a couple of points: What you "see" in an audio application often does NOT, depending upon the application, reflect what the resulting weaveform actually looks like. The EASIEST graphic display of a series of sample valies is to simply connect the dots, but that is NOT what D/A converters do. If you had a series of alternating +- and -1 values, near 1/2 the sample rate, and if simply "drew" them by connecting the dots, what you would see could be what looked like a triangle wave carrier modulated by a sine wave. FOr example, at 22 kHz, you'd see the modulating wave as having a frequency of 50 Hz. But that waveform, AS DRAWN, has components that extend to infinity. JUstv as there is a component at 22.05 kHz - 50 Hz = 22 khz, there is also one at 22.05+50 Hz or 23 kHz. If you had only three such components, then the waveform would be a 22 kHz SINE wave modulated by 50 Hz sine. But the triangle carrier has compoenents way out as well, so you'll see images at 44.1 +- 50 Hz, and 88.2 kHz +- 50 Hz and so on. ALL these are IMAGES of the original 22 kHz sine wave and, to be reconstructed properly, requires the additon of the necessary anti-IMAGING filter (also known as the reconstruction filter). Now, get rid of all the components at an above 22.05 kHz, and all you're left with is your original 22 kHz sine, just as it was recorded. Now, where OVERsampling techniques come in is HOW you implement this filter. YOu can, as some assumed incorrectly, try to implement it in an analog filter, and that get's real tough. You can also try to implement it as a digital filter at 44.1 kHz, but it's effectively impossible to implement a 22.05 kHz low pass filter at a sample rate of 44.1 kHz. So what we do is we OVERsample the 44.1 kHz stream that we already have: inbetween every real sample, lets insert, oh 63 samples whose value is, let's say, 0. Or anything slse, for that matter. What we've done is that we've taken the original images spaced 22 kHz apart and now spaced them instead 22 kHz * 64 or 1411 Khz apart. Now you can build a VERY GOOD digital filter with very good out-of-band rejection and very good in-band response, and then all you have to do in your final analog stage is make sure 20 kHz is getting through fine, but 1411 Khz is properly attenuated. That's a LOT easier than building a filter which lets 20 Khz through but attentuates 22.05 kHz completely. |
#34
Posted to rec.audio.tech
|
|||
|
|||
Frequency/Sample rate
"Don Pearce" wrote in message et... stuff. This is done already for telephone lines - and very successfully. There are two system - A-Law and Mu-Law; the first is international and the second local to the USA. If there were such a system applied to 16-bit systems, we could be seeing better than 24 bit performance. But first agreeing and then constructing the necessary variable slopes was way beyond the technology of the day. It could still be done very simply during a bit-reduction process from 24 to 16 bits - just a mathematical operation. Then at the player end those 16 A-law bits could be reconstructed back to 24 very quiet bits. Of course all this would save is some real estate on the recording medium. Interesting project, though. d I think I read somewhere that AT&T was using digital for long-distance as early as 1970, long before most of use had even heard of digital audio. And it was 8-bit, 8 Khz sample rate, I think. To do the bit-reduction method, it seems that something like a 24-bit DAC would be needed in the player. In my project, I didn't try changing the nonlinearization much, and then found that the low order bit of my 8-bit chips didn't appear to be working correctly anyway. Mine was a continuously variable slope, using transistors in op-amp feedback loops. It was only a for-fun project anyway, so I didn't pursue it further to see how much I could reduce the noise -- I didn't have an application for it at the time. But it did prove the nonlinearization workable, even if it didn't achieve AT&T's quality. -- Earl |
#35
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
Peter Schepers wrote:
) In article , ) Willem wrote: )Industrial One wrote: )) Do I understand correct: hz is one sine loop per second, ) )Hz is one whatever per second. )Your heart beats at roughly 80 Hz. ) ) I think you mean the heart beats 1.333 Hz. D'oh, you're right. Roughly, that is. :-) Did you check the other example (frequency of the moon) ? I made a mistake there too, although it is not by an order of magnitude... Must be a bad day. SaSW, Willem -- Disclaimer: I am in no way responsible for any of the statements made in the above text. For all I know I might be drugged or something.. No I'm not paranoid. You all think I'm paranoid, don't you ! #EOT |
#36
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
Peter Schepers wrote:
In article , Willem wrote: Industrial One wrote: ) Do I understand correct: hz is one sine loop per second, Hz is one whatever per second. Your heart beats at roughly 80 Hz. I think you mean the heart beats 1.333 Hz. PS. Better do some exercise then. Mine is more like 0.85Hz. It has to be said, though, that in virtually every calculation that involves frequency, the number you actually want is radians per second - 2pi * normal frequency. It would make life much easier if the world adopted that measure. d |
#37
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
In article ,
Willem wrote: Peter Schepers wrote: ) In article , ) Willem wrote: )Industrial One wrote: )) Do I understand correct: hz is one sine loop per second, ) )Hz is one whatever per second. )Your heart beats at roughly 80 Hz. ) ) I think you mean the heart beats 1.333 Hz. D'oh, you're right. Roughly, that is. :-) Did you check the other example (frequency of the moon) ? I made a mistake there too, although it is not by an order of magnitude... Well, I originally calc'd .41uHz instead of your number. PS. |
#38
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
In article ,
Don Pearce wrote: Peter Schepers wrote: In article , Willem wrote: Industrial One wrote: ) Do I understand correct: hz is one sine loop per second, Hz is one whatever per second. Your heart beats at roughly 80 Hz. I think you mean the heart beats 1.333 Hz. Better do some exercise then. Mine is more like 0.85Hz. I was using his 80 beats/min, not mine. PS. |
#39
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
In rec.audio.tech Earl Kiosterud wrote:
DVD audio is just for marketing. No one, with the possible exception of a few young people who can hear above 20 KHz, and many dogs, can hear the difference between regular 44.1K 16-bit audio and 96 or 192K sampling and 24 bits -- it's been proven, though some will tell you they can. It's something they call "resolution" for which they have an altar, dogma and lots of ritual. They get this dreamy look in their eyes. Challenge it, and their veins pop out and they go on rampages. It's likely that much of the stuff you get on DVD-audio discs is better stuff, and has been more meticulously recorded, hence the good sound of many of them. Actually, with the advent of DVD-A ripping software, I've found that the stereo mixes on many of them (the rock/pop ones at least) are just as dynamically compressed as their modern CD counterparts. And this at 24 bits! Sheer lunacy. The 'hi rez' format where you are more likely to get 'full-range' dynamics is SACD, due to restrictions built in to the Scarlet Book spec. But even there it's now possible to squeeze the range, from what I hear. In any case, there's no reason why these full-range recordings couldn't be presented on good old 16-bit CD. The perversion of one of the original benefits of Redbook (extended dynamic range) into the 'new' benefit (heretofore unheard-of amounts of compression) is just sad. -- -S Poe's Law: Without a winking smiley or other blatant display of humorous intent, it is impossible to create a parody of a religious Fundamentalist that SOMEONE won't mistake for the real thing. |
#40
Posted to rec.audio.tech,comp.compression
|
|||
|
|||
Frequency/Sample rate
In rec.audio.tech Edmund wrote:
I heard about that tests and it was criticized because the music was played over a pair of passive loudspeakers with passive filters that where nowhere near phase linear same problem with electrostatic speakers with step up transformers . So no matter how much better SACD or DVDA can be, played over such loudspeakers all the advantages are down the drain. Don't know if this story is true but I very much like to attend such a listening test an judge for myself. Did anyone here did attend such a test and on what kind of speakers was it played? No doubt the test is being 'swift-boated' by anxious 'audiophiles', but here are the facts (clipped from http://www.bostonaudiosociety.org/explanation.htm ) The Principal System The playback equipment in this system consisted of an Adcom GTP-450 preamp and a Carver M1.5t power amplifier. Speaker cables were 8 feet of generic 12-gauge stranded wire; the line-level connecting cables were garden-variety. Three different players were used: a Pioneer DV-563A universal player, a Sony XA777ES SACD model, and a Yamaha DVD-S1500. The loudspeakers were a pair of Snell C5s. The CD-standard A/D/A loop was an HHB CDR-850 professional CD recorder .. .. System 2 We also conducted a series of tests at a local CD/DVD mastering facility. I do not currently have a detailed equipment list for this venue, but the speakers were very large and capable high-end monitors, approximately 7 feet tall, and the power amps were sufficient to drive the speakers to very high levels without audible distortion. Some of the source material for these trials was a classical production which was then in process at this establishment. Like all the others, these trials, which were done under a promise of anonymity made to those involved, produced no significant correlations on music at normal levels. .. .. System 3 Another series of trials took place at a facility at the University of Massachusetts - Lowell campus, using students in their recording program as subjects. Their large monitoring room is custom-designed and has very good acoustics, with a system to match. The system has a center channel and surrounds, but as in the other trials we restricted ourselves to the two-channel versions of our sources, so only the left and right were working. The equipment list for the two channels is: Klark Teknik DN-410 custom-modified 2-channel parametric equalizer Stage Accompany PPA-1200 Dig Control class AB amplifier w/ crossover card Stage Accompany ES-20 Class G amplifier w/crossover card SLS S1266 3-way monitor (two 12" dynamic drivers, two 6" dynamic drivers, one 6" ribbon tweeter) Bag End ELF-1 8-Hz 2-channel low-frequency integrator Bag End D18E-I dual-18" ELF subwoofer system This is another professional monitoring system, installed in a large custom-built listening room with auditorium-type seating. It was capable of very high levels with no audible distortion as well as imaging of a quality not usually found in large spaces of this kind. We were interested to find that our informal high-frequency-hearing tests, which we administered to most of our subjects, indicated that these students had taken unusually good care of their hearing. Most of them had an upper limit in our test of 16 to 18 kHz .. .. .. System 4 Another set of trials was performed during the evening at another suburban location . a custom-built listening room with good acoustics (with the help of an assortment of professional absorbers and diffusers), very low background noise, and equipment that we trusted would pass muster with most audiophiles: Denon 2900 Universal Player with full PartsConnexion mods Conrad-Johnson 17 LS line stage preamp Sim Audio Moon 7 monoblock power amplifiers Quad ESL 989 electrostatic speakers Muse Model 18 subwoofer, 24 dB/octave crossover @ 50 Hz Nordost SPM interconnects and speaker cable // -- -S Poe's Law: Without a winking smiley or other blatant display of humorous intent, it is impossible to create a parody of a religious Fundamentalist that SOMEONE won't mistake for the real thing. |
Reply |
Thread Tools | |
Display Modes | |
|
|