Why Bit Depth Matters
While discussions of bit depth may seem a bit esoteric, the subject has real-world ramifications for those of us working with audio.
Let’s start with the basics. While the sampling rate of a digital recording is primarily concerned with the theoretical frequency response of the recording, The Shannon-Nyquist Theorem states that frequencies below half the sampling rate can be reconstructed perfectly, so a 44 kilohertz (kHz) recording can accurately capture all frequencies below 22.05kHz and a 48kHz recording can capture all the frequencies below 24kHz — the bit depth of a recording is primarily concerned with dynamic range.
The theoretical dynamic range of a 16-bit signal is 96 decibels (abbreviated as ‘dB’) and a 24-bit signal has a theoretical dynamic range of round 144dB — far more than is usable or practical. These figures are theoretical, though. In the real world, converters aren’t perfect. Even the best 24-bit converters are typically closer to 22-bit converters. But what’s a marketing bit or two among friends?
What’s a Decibel?
Oh yeah, since this section is going to be using decibels as a system of measurement, we should talk about what they are. The short (and extremely unhelpful) version is that a decibel is a tenth of a Bel. The term itself was first used in the 1920s. It was developed as a ratio of two power levels by Bell Telephone Labs, and was named after Alexander Graham Bell. A decibel is a tenth of the original measurement unit. The decibel works like this: 1db represents approximately the smallest volume change that can be heard (when listening carefully). It represents a change in pressure (or volume, or power) of about 12%. So, 3dB would represent a small but noticeable change in volume, and 6dB represents a change in level of twice (or half) as much. These aren’t absolute numbers, though, they’re a ratio that represents the amount of change. Since the decibel scale is based on human perception, the quietest sound that a human can hear represents 0dB; normal conversation measures about 60dB, and prolonged exposure to 95dB can result in hearing loss. A rock concert can hit 120dB.
For a little historical perspective, let’s look at the dynamic range of some of the older delivery formats. For perspective, remember that CDs have 16 bits, or 96dB of dynamic range. The early Edison cylinders were, quite frankly, pretty bad. They were extremely fragile and took up a lot of space. From a sonic perspective, Edison cylinders were severely band-limited and any quiet sounds could get lost in the surface noise — they weren’t much of a delivery format. Compared to cylinders, 78 RPM records were a great improvement. They had an effective dynamic range of 30–40dB, from the softest sounds to the loudest sounds the record could hold. This gave 78s an effective bit depth of less than 6 bits — and folks bought them like hot cakes.
Long Playing records (LPs) were an improvement over 78s. With a good stereo and a good pressing, LPs had an average dynamic range of 60–70dB. Yep, vinyl, the beloved format of audiophiles everywhere, has about 11 bits of resolution. On average, mass-market 45s weren’t as good as LPs. And then came tape.
Remember cassettes? Commercially made cassettes had about 40dB of dynamic range, or about 6 bits of resolution. To be fair, though, if you bought high-quality blank tapes and were careful, you might get as much as 70dB of dynamic range — about the same as decent vinyl. Reel to reel tape, with its greater track width (and speed) could capture from 60dB to more than 100dB of dynamic range. Though to be fair, the high number came at the expense of distortion — a lot of it.
What it all means
What all of this means is that a CD is capable of greater dynamic range than all but the best analog tape recordings, and better than any delivery medium that came before its introduction. Which brings us to another interesting point: through most of recording’s history, the recording medium — whether it was the master acetate (later duplicated for mass-market sales) or tape — was higher quality than what was sold to the public. This was due to a certain fidelity loss when copying in the analog domain, or (before tape) the generational loss that comes from simply making the stampers. Even in the glory days of multi-track recording, the mastering process necessarily involved limiting the dynamic range of the final mix to fit the lower dynamic range of the album. But once CDs were introduced, the delivery format (full bandwidth and full dynamic range) was just as good as what the engineers and musicians were used to hearing in the recording studio.
With 0dB being the softest sound that average humans can hear, normal conversation being from 60 to 70dB, and sustained exposure to 95dB being capable of causing permanent hearing damage, it’s easy to see that the average recording will need nowhere near 96dB of dynamic range. But there’s more to it than that. Every time a mathematical function is performed on a digital audio signal (and every thing you do with a digital audio track is a mathematical function, from changing the level to adding a compressor plug-in), the bit depth of that track increases. A number of different approaches to dealing with this increased bit depth have been implemented by software manufacturers. Some will use 32-bit floating-point processing and some use fixed-point processing with intermediate bit depths up to 56 bits. In any case, at the end of the process, the file needs to be reduced back to the original 16- or 24-bit depth.
And this is why the extra bit depth matters: after every mathematical calculation — whether a volume increase or a bit of EQ, the bits at the end increase — and they have to be removed from the file at the end of the process. The more bits you have, the less likely it is those bits will have an audible impact on the audio — especially when the bits are removed.
Want to learn more? Visit Sweetwater Sound’s SweetCare article on this subject.