In Perspective

Analog audio is generally an electrical representation of changing air pressure. In its most basic form (the electrical signal just out of the microphone), analog audio’s resolution is not too shabby: a set of huge ranges (10,000,000:1 amplitude range --130 dB, and 10,000:1 frequency range -- flat from 10 Hz. to 100 KHz.). By investing cubic dollars, you can increase that resolution by a factor of 10 before you run into what engineers like to call “significant physical limitations.”

Now, this resolution range is not too different from the resolution of our hearing, so analog audio, in its basic form, seems really quite well suited to our hearing (subject to the qualification that the ear is not, repeat not, a microphone and doesn’t work according to the same rules). The problem with analog audio lies in its storage (the actual recording/playback process). The resolution limits of the analog storage medium are significantly poorer than the limits of analog transmission (without noise reduction the amplitude range of analog storage is down to about 10,000:1 [80 dB] and the frequency range is typically 800:1 [not really flat from 30 Hz. to 25 KHz.] with most of the loss coming in low frequencies). Therefore, any analog signal that is stored and then reproduced has significant, audible errors included in its reproduction.

Digital audio, on the other hand, stores perfectly (in theory). What you store is what you get back. Exactly. After all, numbers are numbers, right? Actually, digital audio is no more perfect than analog audio, and in many respects less so, except that it records better. What it does is introduce its errors as a function of the conversion of the signal from analog to digital rather than as a function of recording/playback. The resolution limits are defined not by the precision of the physical hardware but by the mathematical constraints imposed by the sampling rate (which determines frequency range), the number of bits (amplitude range) and the timing and quantification accuracy of the A-to-D and D-to-A converters. So, Neil may have an argument after all when he says the sampling rate is too slow, because there are in fact limits and errors introduced into digital signals by virtue of the sampling rate limits. But is it really too slow? And what about all that crummy analog tape hiss? Everybody knows that digital is quieter.

This is where it gets interesting. In the digital realm, if you really wanted to throw some money at the problem, you could really come up with a like totally awesome system.The limits are mathematical and financial more than physical. There is no technical reason why we couldn’t build a digital system with a sampling rate of 200 KHz. and 24 bits of resolution. This would give us a frequency range of 100,000:1 (flat from 1 Hz. to 100 KHz.) and an amplitude range of 10,000,000:1, which is 140 dB (enough greater than the human hearing range to really quiet the debate). Aside from money, memory becomes a big , I said big,problem. Digital audio is a memory hog now (10 megabytes of memory is required for each minute of digital stereo audio). Well, with the system limits described above, we would need about 72 megabytes of storage for each minute of digital stereo audio, or 4.3 gigabytes for an hour-long Compact Disc. So much for Sound Tools and digital sampling without a mainframe computer. A lot of current music production would have to shrivel up and die if we really required that kind of memory.

So, maybe the digital nerds are right after all. Memory is still expensive enough that we really need to avoid storing anything not really important, and if you can’t reallyhear anything above 20 KHz., we can't afford to store it. Also, nobody (like, nobody!) listens to a full 120 dB dynamic range, so why store that either? 16 bit resolution (65,000:1) is 96 dB, which really seems to be more than adequate for a world with a working noise floor of 40 dB SPL and a hearing system with a threshold of pain at 120 dB SPL. Besides, analog decks can’t even manage 96 dB of resolution (though with Dolby® SR noise reduction they begin to get close). This last item has been one of the big sales arguments for digital recording, and is the basis for the myth of “no tape hiss.” There is hiss, actually, but it is about 10 dB softer in any 16-bit digital recording than in the best analog recording.

And this is where science gets into the act. To make sure that these things are so, engineers and scientists test and measure their assumptions. People with nerd-packs in their shirt pockets have measured perceived noise floors and thresholds of pain and have studied listeners in controlled situations and have tested listeners’ responses to a wide variety of digital and analog storage and processing schemes. The data seems pretty solid, verifiable and conclusive: humans don’t hear stuff above 20 KHz. or below 20 Hz., and the dynamic range of our noisy post-industrial cyber environment (30 dB SPL to 120 dB SPL -- 90 dB) suggests that 16 bit digital audio is plenty adequate.

But this is where science gets to be really fun. Science is about knowing things and verifying things and successfully predicting things. It is blind, it does not take sides, and it does not give a flying fuck who is right. Maybe we have to ask some different questions, measure some different things. To date, we have asked listeners to listen for differences and we have relied on their telling us what they hear. We have asked them to listen to recordings with no information above 17 KHz. and compare them with recordings that do have information above 17 KHz. and tell us if they hear a difference. We have used so-called double-blind AB testing, where listeners can switch back and forth between two different systems at will, to listen and compare as carefully as they want, to the limits of their abilities. All sweetly reasonable, rigorous, fair and objective. But maybe wrong.

A fundamental rule in scientific measurement is the Range Rule: you must make your test fit within the range of your hypothesis. And if you are testing music, as listened to at home, then your test must involve music listened to at home (or some reasonable approximation). Where we tripped up was in the AB test. We don’t normally listen to music in an AB situation, switching back and forth between systems. We usually just sit there like good couch potatoes, with some recreational refreshments in hand and the level cranked to the JBWMNC* Threshold, and wallow in the stuff. And after we’ve wallowed, we feel satisfied. But some of us, after listening to digital recordings, apparently don’t feel quite so satisfied.

So about two years ago, some Japanese researchers decided to approach things differently. They didn’t use AB testing at first, and they didn’t ask the listeners to report back at first. Instead, they made a recording (of acoustical musical instruments) that they knew by physical measurement had frequencies up to 50 KHz. present, and they then made a copy of that recording that was low-pass filtered to 20 KHz. They played both versions for their test listeners through a monitor system that they knew by physical measurement could reproduce frequencies up to 50 KHz., while observing, via electro-encephalograph machines, their listeners’ brain activity.

Bingo! The recordings that were filtered produced much less brain activity than did the broad-band recordings. So they knew that the brain noticed a difference. Then they asked the subjects to comment on the quality of the recorded sounds, and found that the listeners reported the original broad-band sounds to be interesting, satisfying and beautiful much more than they reported those qualities for band-limited sounds.

From there the researchers worked backwards to resolve these findings with what we’ve already known: that in AB tests listeners can’t hear the difference. They found, interestingly, that it takes a while for the brain activity to change, so after switching from A to B, a change in brain activity wasn’t noticed for 20 seconds or so, and then it came on gradually -- in other words, the brain appears to become conditioned to its listening situation and takes a while to perceive a change. So, AB tests don’t work for testing this hypothesis, but long-term listening trials will reveal it.

So, yup, Neil’s got a point and the audiophiles are right, within limits. Some ultrasonic material nuked by the anti-alias filters used in analog-to-digital conversion turns out to perhaps be significant in the “musical satisfaction” sweepstakes. This can now be scientifically supported. Sumbich!*

What does it all mean? Is digital not boss? Do we beat a hasty retreat to BSR turntables? Do we convert all our CDs to mini-frisbees? Back to razor blades? Ohhhhh, noooo! This, campers, is where you’ve got to reach into your own consciences, and decide for yourselves.

Me? I find it easy to vote for digital. Right now I’m listening to a CD of Mozart’s Mass in C minor and enjoying it greatly. Earlier I was listening to the Cars Heartbeat City on CD and once again admiring it as a wonderful work of art. I’ve got a stack of cassettes that I can’t bring myself to cue up, and a turntable that I’ve got to replace because the drive belt rotted away about a year ago. So, obviously, the limitations of digital aren’t troubling me very much and I’m happy as a clam with CDs. Also, I will share with you some of the differences about digital audio that I do notice and like: digital audio has smoother and extended low-frequency response; it has a speed stability (absence of wow and flutter) that I really appreciate, and there is a transparency of high-level signals due to the linearity (lack of distortion) that I really enjoy. On an operations basis, I like the ease of setup and the stability of the digital recording media.

Actually, I find the audible differences between the two sets of media are pretty small, particularly when compared to the big differences like microphones, loudspeakers and playback rooms. At the same time, I like the convenience of digital audio media -- in production, at home and in my car. There is also an economic benefit derived from digital audio production that we are just beginning to realize: it is far cheaper to do signal processing via mathematical algorithms (as is done in multi-effects processors like the Yamaha SPX-90 family) than to buy hardwired physical units for each function, which is what is needed in the analog realm. As our digital workstation systems mature, that benefit promises to become huge.

But wait a minute! Why, then, is Neil so upset? He must be able to hear that the two systems aren’t that far apart, and that the benefits of digital are pretty significant, even if he prefers analog. Hype like “ice cubes washing over you” and “darkest time ever” are really passionate, heartfelt statements, but they aren’t supported by scientific evidence or common sense observation. The problem may lie in being too close to the recording and the music. I know that I lose my perspective, and minor issues blow up in my mind. When I’m making recordings, I want so badly for the musical illusion to be utterly real that any flaws, including imagined ones, drive me wild. As a result, I generally hate my recordings and music, and can’t bear to listen to them. All I can hear are the flaws. Other people say, “Wow! That’s great, Dave! Where can I get a copy?” Meanwhile, my inner brain is shouting “Omigod! I can’t believe the edit just after the second bridge is so obvious! And the bass truly sucks! Why can’t I sound as good as Steely Dan? Why, God? Why?” This syndrome is fundamental to recording - we are cursed with having to listen to our mistakes forever! It often leads to some truly whacked out, obsessive behavior, and there are many legends from studio lore about obsessive behavior that don’t just border on the lunatic, but absolutely, definitely cross way, way over into the Audio Twilight Zone.

So, I figure Neil has an inner vision of how the music should really feel, and somehow he can’t get it in digital. He knows there is a difference, and it makes him crazy, particularly when its his own work on the ol’ editing block. Hence, the diatribe! Meanwhile, his music is still great, and his fans have no problem with enjoying it at all. They just wallow in it.

So this is what I really think, as a recording engineer: making a great recording means capturing the ineffable spirit, personae and qualities of the performers and music - it is a spiritual exercise as much or more than a technical one. Digital is beginning to be cheaper and easier to use, so use it if you can, and devote the resources you’ve saved by using it to improving the quality of the music. If, on the other hand, digital makes you crazy, use analog and don’t worry about it. The medium is comparatively unimportant - it’s the power and meaning of the music that matters, a power and meaning that transcends the quality and character of the medium.

Dave Moulton is the A-to-D converter (Chairman) of the Music Production and Engineering Department at Berklee College of Music in Boston, MA. Peter Alhadeff teaches a bit of mathematics at Berklee and Alex Case samples production classes there.

COMMENTS

Post a Comment