Some people need hard evidence to make up their minds about what goes on in the digital conversion process. For that reason, there is some hard statistics with visual ques to help. That said here's a 7040Hz sine wave (which is a natural overtone of "A") generated via computer at 2 different sample rates. The first one is 44.1KHz (CD quality) and the second is 88.2KHz (double CD quality). The artifacts of these distortions are within the adult hearing range. Notice how much more the 88.2KHz resembles a sine wave? The total harmonic distortion has been cut in half as well as the dynamic distortion. There still exists some variations in the amplitude of the waveform but it is much less intrusive at the higher sampling rate.
Now below, we have a 10KHz sine wave generated at 44.1KHz during a discussion with somebody who said anything above 44.1KHz sampling was unnecessary. According to this person, 44.1KHz sampling was perfect and that all the problems people percieve in digital recordings lie in the convertors. Just like the above chart, convertors are plainly not an issue as the chart was generated natively in the digital world without the use of analogue to digital convertors. This is here to show that the problem with 44.1KHz sampling not only lies in the format itself, but distortion also increases exponentially as the upper reaches of human hearing is reached. Again, the distortion is not only harmonic in nature but also dynamic.
This is scientific evidence showing up close and personal. Many people have argued that people cannot hear above 20KHz and thus higher sampling rates are not needed. Now, some people argue that humans CAN hear above that level, just not conciously. While there may be some validity to that, this argument will focus primarily on the assumption that 20KHz is in fact the limit of human hearing. As over 150 years in recording history shows, there's more to sound quality than mere frequency response. If frequency response is all that mattered, then yes, 44.1KHz is fine. However, anyone in the industry can vouch for harmonic distortion being an issue and most people will tell you that dynamics are an issue as well.
On another note, if inclined to do some study on Rupert Neve, you'll see that his favorite game is to play a sine wave at 12KHz and switch back & forth between that and a 12KHz square wave. While most people can't conciously hear the artificial harmonics a square wave generates from 36KHz and up, their eyes water when he plays the square wave and not the sine wave. Listening tests show that people prefer sound sampled at a native 96KHz over 44.1KHz. Though at 192KHz, few noticed a difference over 96KHz. That said the ability to sample at 192KHz is still a good option to have because of the ability to record sounds and slow them down for special effects. At 192KHz, audio samples can be slowed down to 1/4 speed and they still sound decent. That could not be done at 44.1KHz or even 96KHz.
Just for good measure, here is a 14KHz sine wave at 44.1KHz. This would look more like the 7KHz tone sampled at 44.1KHz if this example were sampled at 88.2KHz. Again, notice how the higher frequency has a grotesquely exagerated distortion over the lower frequency signals. The character of the wave fluctuates in a sinusoidal pattern. At 14KHz, several subharmonics are generated, the most intrusive of which lies around 2KHz which is near the peak of human hearing. The theorem which states the maximum frequency that can be faithfully be reproduced must be less than half of the sample rate is called the "Nyquist theorem". Though the input sine wave in the example is well below the Nyquist limit, there is still subharmonic aliasing. What's that? Aliasing below the Nyquist limit? Yes, though not to the extreme it would be if it were above the Nyquist limit. There is about a 2dB loss in volume in this particular example. This loss is actually a mild comb filter or ripple effect that spans the entire audible spectrum and not a smooth and predictable role off like you would get on analogue tape. Now this subharminic alias that occurs well within the hearing range is the result of intermodulation between the sound source and the sample rate itself. This will show up on a spectrum analyzer and is actually quite audible. The result is similar to that of a ring modulator with dry sound mixed in. The relationship between the input signal and the alias is inversely proportional. That is, the higher the input frequency, the lower the alias is in frequency. The rate of change is linear rather than logarithmic in response like typical harmonic distortion and is thus, non-musical. Normal harmonic distortion creates false overtones that are logarithmically poportional to the input signal and therefore less offensive. In the example, there's acually a super-harmonic genrated above the sample rate as well as below but that is a matter for another time.
When quality is less of an issue, GCM records at 44.1KHz for CD, 48KHz for video. More important stuff gets recorded at 88.2KHz or 96KHz, especially if there's a lot of processing needed. Believe it or not, I realized purely by happenstance that even material recorded at 44.1KHz, digitally EQed or reverberated at 88.2KHz and then converted back to 44.1KHz, sounded noticably better than material that was recorded and processed completely at 44.1KHz. That is because digital EQ, like reverb is based on complex delay algorithms. Delay demands more time-based resolution unlike amplification and mixing which is primarily dependant on voltage resolution or word length. That said, digital EQ not only causes comb filtering, but it also distorts audio in the time domain. That means that transient spikes in the audio signal get blurred by phantom transients in various parts of the audio spectrum depending on how EQ is applied. The ripple effect of comb filtering is tamed somewhat by the introduction of additional delays which further smears transients.
Now time to get a bit more personal. When I was directly involved in the electronics industry, we had a 3M rep that would come in once & a while that would "educate" us about Electro Static Discharge (ESD) safe materials. He would go on & on about how people spend so much money for the ESD work mats from their competitors that "didn't work". He kept saying that other companies' mats didn't work and how 3M's were the only reliable way to go. He would go around with a meter showing us that "this computer monitor has 3000 volts of static build up" and "that table is 600V". I said "why don't you test our ESD stations, which he refused to do. He said the static meter wouldn't work on our stations. I asked why, to which his response was "there won't be any build up there". So I said that the mats DO work and that they are safe to use. He grudgingly agreed with the stipulation "if you follow the proper procedure", which is to place items on the mat and grounding one's self BEFORE handling the sensitive devices. I then said that the same procedure had to be followed with his mats as well and that making false claims about their competition to boost his own sales was unethical. Why bring this up? Don't believe what reps say. Many people from reputable companies claim that digital is the ideal way to record and there is no need to record at rates higher than 44.1KHz. I've heard such claims from reps at Sony and Digidesign among others. That does not mean they are lying. Most of the time the reps themselves are misinformed. You have no idea how many times I asked questions from company reps about their own products which they couldn't answere. By the way, if Sony doesn't think higher sampling rates are necessary, why did they invent Delta Stream Digital (DSD) based recording like the Super Audio Compact Disc (SACD) which samples at 2.8224MHz?
Let's use some logic here guys. It has been argued that testing with sine waves is not an accurate means of measuring performance. Well all sound can be created with combinations of simple waves like sine waves, triangle waves, square waves and sawtooth waves. Sine waves are by far the most commmon occurring in nature. They are also the most easy waves in which to measure distortion. So tell me, if you recording machine can't represent a sine wave which is the most simple form of sound that ever existed, how can it represent the much more complex sounds like music accurately? An article on SACD will be quoted where the guy who helped create DSD recording is interviewed. He makes comments that intermodulation CAN be heard on music and that the Pulse Code Modulation used in CDs and other common digital encoding is very "alien" and detrimental to sound. The following is an excerpt about him talking about their absolute best Delta-Sigma convertors used for traditional PCM digital sampling.
"Meitner: But if you look what an interpolator like an eight times oversampled interpolator will do, it will give you a flat frequency response. Which means that it deals with the sine-x-over-x error. It pushes it out eight times. So now the 3.2 or 3.4 dB down at 20 kHz is now 0.04 dB down at 20 kHz, okay?
Pappas: Okay.
Meitner: But it only knows to do that on a sine wave. Imagine a transient coming along.
Pappas: It won’t be able to figure it out.
Meitner: It can’t do it."He continues stating that their best convertors (which are 1 bit Delta-Sigma convertors with 5th order analogue filters cause a "ringing" only after transients. Digital filters cause ringing BEFORE AND AFTER transients which is "another very alien part" of more common convertors. I really suggest everybody read this interview. It is incredibly informative on the shortcommings of normal digital signals. Now to finalize the statement, included are some shots of a sweep recorded to 1 track of GCM's own 1" 16-track. This was recorded at +3VU with no Noise Reduction. The Total Harmonic Distortion (THD) is at -75dB. THD at 0VU is more like -76dB with NR. Notice that it looks like a sine wave.
This is 14KHz on the 16-track. Notice the dots representing digital sampling which overlays the actual sine wave to show how the computer portion of the test compares to the "original". This was recorded at 24-bit 96KHz by the way. Even though there is technically a higher THD on the analogue deck, the dynamic distortion is much lower and there is no alias.
Remember that this is a TASCAM deck which is good but not top of the line. An added plus is that the inputs and outputs are Class-A FET based which is nice to have. Not even the 192KHz A/D convertors with 128x oversampling employed by GCM have that. As far as people not being able to hear the different between analogue sound and digital recording? Consider this... I have heard Dire Straits "Brothers in Arms" on 3 different media. Vinyl, cassette and CD. I can tell you straight out that this album was recorded on a DASH. It sounds digital regardless of the medium on which I've heard it. While it's one of my favorite albums, I find the brittle sound irritating. Same thing with the Rush albums of the mid 80's. They are also obviously digital. As well as Metallica. They all wreak of "DASH". Granted the convertors they used back then were 16-bit PCM convertors and newer digital recordings sound a lot better. I can still often hear the difference.
In conclusion, if you want to record music, choose the medium with the highest possible resolution. A wide tape format like 1 inch or 2 inch is desirable. However for many, digital is the only option. In that case, do not tell yourself that since it will end up on CD, you only need to record 44.1KHz 16-bit. There is a great benefit from recording at 88.2KHz and higher. Most modern analogue to digital convertors will output a 24 bit signal. Use that option. If you don't hear the difference after it has been converted to CD, somebody else will. Save the original high resolution master. Having that higher resolution recording will allow for a better transfer to DVD audio or SACD (whichever becomes the new standard) than merely transferring from CD quality audio.
If you do not understand terminology in this site, go hereEncyclopedia
To E-mail the author, type what you see in the image below into another window. Sorry for the inconvinience.
