I am certainly not an audiophile but I remember when CD’s got rolled out in the 1980’s they were heralded as the ultimate recording medium? Now, I here talk of upcoming formats that are supposed to be “better” than CD’s for music reproduction. Then, there are the diehards that still insist that records have something that CD’s never will.
What’s the Dope?
CDs can basically reproduce any sound with frequencies below 22 KHz (since they sample at 44.1 KHz). Since the average human is only sensitive to frequencies between 20 Hz and 20 KHz, this is fine for nearly everyone.
Newer formats like SACD and DVD-Audio improve on CD audio in a couple ways: first, CDs only have two channels (stereo), but other formats can have more (5.1 surround sound). They also have higher sampling rates, which I don’t think really make a difference (see above), and they have higher resolution (24 bits per sample vs. CD’s 16 bits), but I’m not sure how noticable that is either.
Vinyl records are analog recordings of the sound, so at first glance, one might think they can reproduce the sound more accurately than CDs. However, CDs can already reproduce the audible frequencies quite well, and records degrade as you play them, so even if it’s a more accurate reproduction the first time you play it, it won’t be for long.
I believe there are some other arguments against CDs based on things like dynamic range, but someone else will have to explain those.
Compact disks have a better frequency range than what most humans can hear. They also have a very good dynamic range (the difference between the quietest sound they can reproduce and the loudest sound they can reproduce). Any new format could have a greater frequency range and dynamic range, but very few people would even be able to hear the difference.
Vinyl and tape (both cassettes and even “studio quality” from way back when) both distort the signal enough that many people, especially those with musical training, can hear the distortions and noise added onto the signal. Anyone who tells you that they make a “better” reproduction of the sound are quite simply full of hooey.
Some people are used to the particular distortions caused by vinyl (or vacuum tubes, which you didn’t mention), and think that these types of systems sound “better.” In some cases this is because the distorted sound is more pleasing to the ear (or mind). For example, if you play an electric guitar through a tube amp, most people will tell you it sounds “better” than an electric guitar played through an amp that doesn’t distort the signal. Play a bass guitar through the same amp though and many folks won’t like the sound at all.
CD’s use a sampling frequency of about 44 KHz. This means that they can reproduce sounds up to about 22 KHz (half the sampling rate, do a web search on Nyquist’s theorum if you want the gory details) and they have a 16 bit depth, which means the quietest sound they reproduce is about 1/64 thousandth as loud as the loudest sound they can reproduce. It’s actually fairly easy with today’s technology to create something with a higher sampling rate and a wider dynamic range, but your ear isn’t going to hear it. Most people can’t hear much about 10 KHz, and very few can hear anything at all above 15 KHz. I think the highest frequency that anyone has been tested at and found capable of hearing is about 24 KHz, but folks who can hear out to that range are exceptionally rare.
(minor nitpick)
20 Hz to 20 KHz is commonly used as the frequency range that humans can hear (mostly because it makes for nice even numbers), but the average human can usually only hear about 100 Hz to about 15 KHz. The frequency range that you can hear also decreases with age, especially on the high side.
Audiophiles say that digital music sounds digital, where analog sounds warmer and more natural. I personally can’t tell the difference. But then again I can’t tell the difference between Dom Perignon and Boone’s Farm 
.
But sampling rate affects more than just what frequencies are captured. An analog signal synthesized from a digital capture of an analog sound only approximates the original analog waveform. Some people’s ears might actually be physically sensitive or trained well enough to appreciate the differences.
This article on howstuffworks.com gives a pretty accessible explanation.
You can apply the same concept with videotape vs film as viewed via television, for example. Most people cannot tell the difference, but for many of us the differences are quite noticeable.
Cite? 10kHz isn’t really very high. I would think that most people with undamaged hearing could hear to at least 15kHz, though I have no stats at hand. My own hearing drops off somewhere in the 16-18kHz range,* and I don’t think I’m particularly exceptional.
*based on the noise generator we use when testing sound equipment at work. I can hear 16kHz plain as day, but 18 is iffy.
I don’t think this is necessarily true. With an 8 bit sample resolution, you could still define a sample of 255 to be 64,000 times louder than a sample of 0 - you’d have the same range of volume, but fewer divisions within that range.
Now, I may be missing something. Wikipedia says “The more bits that are used to represent the amplitude, the greater the dynamic range that can be represented, with each bit providing a gain of approximately 6 dB.” But I don’t see how the greatest expressible difference between the quietest and loudest samples can be a function of the sample resolution, instead of a function of the decoding software.
I don’t think this is true either in practical use (though I may be on a losing streak now). The sample resolution does limit the accuracy of the reproduction, but the digital nature of the signal doesn’t. If you sample at 44,100 Hz, and the signal has been filtered to remove inaudible frequencies above 22,050 Hz before you sample it, then you should be able to exactly reconstruct the signal (limited only by the sample resolution).
You’re perfectly correct, if by ‘reproduce’, you mean ‘render as a square wave of the same base frequency’. At 11 KHz, the CD format only permits 4 samples per cycle, and at 5.5 KHz, well within everyone’s hearing, you get 8 samples per cycle. That’s not bad, enough to tell the difference between square, sine or triangle waves, but it’s not perfection.
DVD audio, sampled at 96KHz, likely does a better job of reproducing tones at 5Khz and higher.
Similar but not the same. Conventional videotape vs. film is not digital vs. analog; both are analog media.
Conventional film is quantized, whereas analog sound is not. That is, film shows you a series of still pictures which create an illusion of motion because they are presented to the eye at a rate just faster than the eye’s effective sampling rate. Tape does the same thing, basically.
I don’t know what causes the perceived difference between the two but it is true that they are very distinct. Now you have me wondering how a filmmaker would show a scene that looks like it was shot on tape.
A square wave isn’t considered a single frequency for this. If you filter out all sinusoidal waves at >=22,050 Hz, you can perfectly reproduce the signal by sampling at 44,100 Hz. Such filtering may make square waves appear less square, though.
Mr2001, sure, I was using a square wave in it’s communication (on/off), rather than audio sense there. The point was that with only two samples per cycle, every waveform with that frequency will look and sound the same (excepting amplitude). You can distinguish more different waveform shapes with 8 samples per cycle, but it’s still not good enough to distinguish all types of waveform.
So are you suggesting the Shannon-Nyquist sampling theorem is invalid?
I may be getting even further out of my league here, but I believe human ears basically just detect sine waves. A square or sawtooth wave can be decomposed into a set of sine waves, and the human ear can distinguish the shape of the waveform because it can hear those extra sine waves. Those extra waves, according to the theorem, will be preserved as long as they’re below 22,050 Hz – and if they’re higher than 22,050 Hz, they’re inaudible anyway, so you wouldn’t even be able to hear the shape of the wave in the original analog signal.
There are lots of issues with digital reproduction that can effect the sound. For example, the quality of the filtering - as noted above, before you digitize the signal you want to filter out all frequencies above the Nyquist limit. But filters are not perfect. They don’t cut off frequencies perfectly, but roll off on a slope. They can introduce distortion.
Then there’s the quality of the digital to analog converters and other electronics in the analog stage. And if the DACs aren’t matched well, you can have problems with jitter.
Finally, digital distortion is very ‘harsh’. I had an el-cheapo first generation CD player that was horrible. Sibilants would crack and hiss.
And even though humans can’t hear above 20K, they can hear lower-order artifacts of higher frequencies. For example, a single square wave is made up of many, many sine waves of many frequencies. If you filter out the high frequencies, the square wave will not be truly square any more.
This is all in theory. The real question is how much of it is actually audible. I used to think “not much”, until I heard the difference between a regular CD and SACD and DVD-A. There is a noticeable improvement with the newer formats, even though all are digital. The difference may be primarily the difference between 16 bit sampling and 24 bit. Or maybe it’s the sampling rate. Or a little of both. But clearly, there’s more to it than just saying, “As long as you sample at 44.1 KHz you’ll perfectly reproduce anything people can hear” - professional audio gear uses 24 bit, 96 kHz sampling. For one thing, it makes it easier to make accurate filters.
Now I’m dredging up some old theory, so I could be a bit off base here, but I seem to recall that the problem with filtering is that ideally you want to pass everything up to the Nyquist limit, and cut off everything after it. Ideally, you’d like a perfect vertical slope. In reality, it’s very hard to make a good filter that has a steep cutoff slope. So there will always be some frequencies that sneak in which can cause distortion when the waveform is reconstructed. But if you sample at 96 khz, you can start your filter rolloff at 20khz, and as long as you have a perfect cuttoff by 43khz, you can reconstruct the waveform accurately.
Does that sound right? My electronics theory is almost 20 years old now.
It won’t look truly square on a graph, but the question is, can humans hear the difference? If only inaudible frequencies are filtered out, I suspect not.
It has been several years since I studied fourier analysis, but I seem to recall that what the Nyquist theorem says is that a periodic signal composed of component frequencies with an upper bound of f can be perfectly reproduced by sampling it at a rate of 2f. The sampling determines the amplitudes of the sinusoidal components of the periodic signal and allows one to rebuild the original signal by essentially adding together a bunch of sine waves of increasing frequencies up to one half of the sampling frequency.
Music however is not generally performed with no instrument other than tuning forks. Sine waves are also not the best fit for every sound that goes into music. For example, bessel functions are the solutions to the vibration of a flexible membrane fixed in a circular frame i.e. drums.
I can’t get into the really technical stuff, but I’ve continued to own vinyl all my life and have a pretty good setup now: Denon amp and direct-drive turntable. Sony speakers from '87. “Real” audiophiles would probably sniff, but for someone who wants to hear the music in a nice and clear fashion, it works.
CD just isn’t very good overall, but it can work well for certain types of music. I’ve had little problem with CDs for acoustic blues and general pop, but I’ve found CDs to be pretty bad for symphony music. There simply is no comparison with vinyl chloride.
I don’t know why this should be, but even on a decent car stereo chamber and orchestral music is just missing something that you will get from vinyl. With vinyl, you will hear clearer pizzicato and string harmonics. With CD, the sound sounds washed out and phoney.
We need a better format. Maybe audio DVD will be it.
:eek: Not at all! It relates to frequency, not waveform. If you sample below the nyquist frequency, you get aliasing. But sampling at or even above the Nyquist frequency still doesn’t give you perfect reproduction of a waveform.
[nitpick]
I think you mean polyvinyl chloride (PVC). Vinyl chloride, its precursor, is a gas at room temperature and is toxic with regular exposure, both properties making it rather unsuitable for a commercial recording medium. Just plain “vinyl” still lets everyone know you’re referring to LP (or other) records.
[/nitpick]
When I said “can’t hear much” I was perhaps a bit vague, but I didn’t intend on doing a 3 page description of exactly how the ear hears. This is a typical graph of how a person hears:
http://www.ee.washington.edu/conselec/CE/kuhn/audio/95x32.gif
Note that above 8 KHz the sound level required for you to hear rises rather dramatically, but frequencies out to 16 KHz are still audible as long as the sound pressure level is high enough. Basically, what I was trying to say was that for a typical person, above about 10 KHz your hearing really starts to get crappy, and once you get to 15 KHz and above it’s not really going to work at all.
If you look at the above graph, you are going to be able to fairly easily say that the person has a “good” hearing range of about 130 Hz to about 12 KHz, although they are able to detect frequencies in the range of about 20 to 18 KHz.
If you lower the volume of your noise generator you’ll find that your detectable frequency range also drops.
The only way that would work is if you used a non-linear scaling of your bits. This would mean that with your 8 bit sample the quietest sound would be 1/64 thousandth of the loudest sound, but that you could not reproduce a sound near the high end of the range that was 1/64 thousandth louder than the next quietest sound. You’d end up putting some weird artifiacts into the sound that would be somewhat similar to the hysteresis problem that magnetic audio heads have (audio tape) since their response is also non-linear.
CDs don’t do anything weird like that, and just use a simple linear bit conversion. This means that with 16 bits you have 65,536 (64k) different combinations.
This is basically correct. The human ear has a bunch of cells inside the cochlea called hair cells (IIRC) which act as itty bitty bandpass filters. If the sine waves are too close in frequency they’ll both be detected by the same cells, so the ear isn’t able to perfectly detect frequencies, but it has a resolution of about 1500 different frequency ranges that it can detect, which is pretty good.
That’s correct. However, one thing that CD players do now that I don’t think they did 20 years ago is that they digitally fill in some of the missing data at a much higher sampling rate. They obviously can’t fill it in correctly because the original data is just plain gone, but if you know that sample 1 is value X and sample 2 is value Y, you can do 8 intermediate steps in between X and Y (which requires your D/A to operate 8 times as fast) instead of just doing one big step change between X and Y. This drastically eases the requirements for your analog filters and helps minimize the type of distortion you are talking about.