The Way Things Are

Digital audio has been around for twenty-plus years now, and has become both a basic part of virtually all audio production as well as an old friend that we have come to take for granted. Our thinking has, in general, gone on to newer things, and we’ve stopped thinking quite so intensively about how digital audio works. Naturally, gaps in our understanding have occurred as the technology has progressed while we weren’t paying attention. Also, misconceptions that we’ve had from the beginning and have never really unlearned may have begun to present some serious confusions and problems.

So, it is useful to review the terrain, and see where some of these things are these days. Recently, I’ve been reading a lot of stuff and talking with a lot of people about various aspects of the digital recording production process, and it has become clear to me that a lot of misunderstanding exists (including plenty of my own, I’m sorry to admit!).

With that in mind, let’s conduct a brief refresher course on sample frequency rates, word-lengths (er, number of bits), and dither – and find out where we are on these things these days.

About Sampling Rate

Sampling Rate is reasonably well-known and understood. We all seem to have gotten the idea about it and it’s related Nyquist Frequency – you know, the frequency that’s half the sampling frequency and the highest frequency the system can reproduce. Some of the surrounding issues seem to still give us some trouble though, and are worth going over once again. This is particularly true because we are now contemplating a 96 kHz. sampling rate format.

Meanwhile, the benefits of 96 kHz. are not necessarily obvious. Given the pretty obvious 20 kHz. upper limit to our hearing, the 48 kHz. Nyquist frequency obtainable with 96 kHz. sampling seems like overkill.

However, there are some strong arguments for a 96 kHz. sampling rate, now that the technology is available. First off, there is considerable evidence that acoustical information above 20 kHz. may play an important, if poorly understood, role in what and how we hear, even though we don’t directly perceive it as frequency information (most of us stop hearing frequencies as such – you know, sine waves – between 15 and 17 kHz.). Secondly, some of the time information that we are capable of detecting with our hearing involves intervals considerably shorter (down to 3 microseconds, according to some sources) than the smallest time interval that can be detected by a 44 kHz. sampling rate (which is about 45 microseconds), and there is even more evidence that such time information may play a significant role in our hearing. So, in a direct way, the octave above 20 kHz. may very well contain acoustical information that would be quite beneficial to include in recordings.

But more than that, there is a fundamental issue about what I call the audio window. The audio window is the range of magnitudes of frequency, amplitude and time within which sounds that we can hear exist. It is obviously and intuitively important that we maintain an adequate “space” or “guard band” around the limits of that window to reduce distortion or error at or near the limits of the window. In the frequency realm, an octave seems reasonable to me – which would mean a bandwidth of 10 Hz. to 40 kHz. Cool!

Meanwhile, our beloved 44.1 kHz. sampling rate yields an audible upper frequency limit of, in theory, about 22 kHz. Definitely less than an octave (in fact, about a minor second!). Not much of a guard band. The actual audio behavior in the octave below that limit (i.e. the top octave of the audible spectrum) is subject to amplitude and time errors, not to mention aliasing as we approach the Nyquist frequency.

We have used a variety of engineering techniques to reduce the badnesses that occur in the top octave, including oversampling, so-called “brickwall” filters, etc. However, there is no free lunch. All such work-arounds introduce their own side-effects and badnesses, which we have lived with, mostly without too much pain. However, the Internet gossip (which you should note with interest, but also regard with caution and skepticism) is that when you properly implement a 96 kHz. sampling rate, and listen carefully to the difference, there is a difference and it’s a good one.

Whatever. It seems reasonable to me that we should, when technology permits it, include all possible acoustical information (including ultrasonic and infrasonic) in our recordings, even when we aren’t sure how important some of that information is. Some humility about what we don’t know is called for here.