Audio Advice: Digital Audio Basics (page 2)

As you can see, analog audio has its share of imperfections in the way it stores sound, many of which digital fixes. Instead of trying to pass along a continuously varying current or magnetic field, digital audio converts these vibrations into a string of numbers easy to store and replay at a later time.

The conversion from analog to digital takes place by "sampling" (measuring) the air pressure or current level at regular intervals. We call the frequency of these intervals the "sample rate." The sample rate relates almost directly to the bandwidth of the sound recorded, bandwidth being the difference between the lowest and highest frequencies a system can reproduce. The more frequently a digital audio device samples the moment-to-moment level of the sound, the more accurately it can capture and describe higher frequencies and the finer nuances of sounds.

The highest theoretical frequency a digital system can capture is half of the sample rate, since you need to measure both the positive and negative positions of a fluctuation to know one occurred. In reality, the highest practical frequency captured is a bit less than half that. Audio CDs have a sample rate of 44,100Hz; divide this by two and you see it nicely contains the human hearing range.

As mentioned, each of these samples (or measurements) becomes a number. The more resolution--the wider the range of values these numbers can have--the more accurately the system can store the sound. It would be nice if a recorder could just count from zero to infinity, using whatever value it needed. This isn't practical--digital audio systems have limits like everything else. In the case of digital, this limit is in the number of bits available (bits being the little on/off switches that store all digital information). The more digital bits devoted to each sample, the more accurately the system can store each measurement. For example, using 16 bits for each sample gives you 65,536 different possible values. CDs use this 16-bit audio system, and the results sound pretty good.

There's a disadvantage attached to using fixed numbers instead of continuously-variable voltages or magnetic fields. Because the recording system has a limited amount of resolution, it has to approximate levels by picking the nearest number that fits. For example, an 8-bit digital audio system says that the loudest possible signal can fluctuate between the extremes of -128 and +127. If the fluctuations swing past that, then they get clipped off and distort, just like with analog audio. And if a given sample measures +124.379, for example, the recorder has to pick +124--it can't record the actual value. It rounds down in this case, causing something called "quantization" error. The result is distortion in the sound that quite often sounds like noise.

Quantization distortion is particularly devilish, since it increases as the sound gets quieter. Compare the percentage of error in rounding 1.379 down to the nearest whole value (1) with the above example. This is why the apparent noise level seems to rise as the sound itself gets quieter!

The advantage of using numbers to represent audio is once you have captured them, they usually don't degrade any farther. The methods used to record digital numbers onto tape are much more reliable than those used to record analog fluctuations, because random tape noise has no effect on the system's ability to read the numbers. The result is far fewer worries about audio quality being reduced as it passes through parts of the signal chain, or to and from tape.

You can also pass multiple channels of audio down one wire (i.e. "the first number is the left channel, the next number is the right channel") rather than the usual requirement of one wire per channel of analog audio. It's also possible to combine digital video and digital audio down the same cable simply by knowing which number represents which signal.

The new DV Format (digital video format) is interesting in that it has two different options for digital audio: two channels of 16-bit 44.1 or 48kHz audio, or four channels of 12-bit 32kHz audio. The first one is similar to that used by audio CDs, and is great for commercial video releases and professional applications. It's the second format that offers some interesting options and tradeoffs available with digital audio.

Four-channel audio allows you to record the location sound that occurs while you're shooting the video, and then later add music, sound effects, or narration on top. The problem is how to shoe-horn four channels into the same space reserved on tape for two. If you drag out a calculator and run the numbers mentioned above, you see how the math works out: two channels x 16 bits x 48,000 samples per second = 1,536,000 samples per second. Four channels x 12 bits x 32,000 samples per second = 1,536,000 as well. The DV format can store either in the same amount of space.

So what's the tradeoff? Bandwidth and resolution. Remember our method for dividing the sampling rate by two to get the highest frequency the system can record? Notice that 32,000 Hz/2 = 16,000 Hz, which even under the most optimistic conditions is starting to crimp in on the upper limit of our hearing. This is not great fidelity, but in practice is still better than most other consumer videotape formats--let alone TVs--can comfortably handle. And what about 4096 values (12 bits per sample) versus 65,536? Again, not perfect. But in this case the DV system distributes the 4096 numbers in a special way to use more of them for lower sound levels. This reduces apparent quantization distortion and noise. All in all, not that bad a set of compromises.

It's inevitable that we'll be seeing more and more media--pictures, words, video, and sound--in digital form. Dealing with digital audio and video may seem new to many of us, but it certainly is the future--one that offers us a lot of creative options.

• Digg This!
• del.icio.us
• Technorati
• StumbleUpon
• Reddit