Bit Depth

The term bit depth refers to the number of bits used in a digital system to describe elementary information. This has an implication on the number of possible values the system can deal with - in essence, the maximum possible number the computer can count to before having to start again with a new word.

For example, a 32-bit system architecture allows the use of 232 different values in its calculations. The logic behind the number 232 is that as it uses the binary, base 2 system, and it uses 32 bits in each word, we have a total of 232possible values (which comes to 4,294,967,295). In any given calculation, for any given parameter, there are 4,294,967,295 potential values the parameter could be. To put that number into context, consider a 16-bit system (which, in the decimal system, is one step down). This system would allow for a potential 216 values, which comes to a comparatively meagre 65,536 potential integer values. This is an order of magnitude lower than the >4 billion values a 32-bit system allows.

These potential values are put to use in digital audio in very tangible ways. As discussed on the PCM wiki page, when taking samples of an analogue signal, the processor has to assign an amplitude value to each sample. Amplitude (analogous to volume, loudness or voltage) is the only parameter which is recorded during the PCM process. This value is stored as a binary number, which uses a fixed number of digits, or bits - this number is the bit depth. The process of assigning each sample an amplitude value is known as quantization, which is the second stage of the PCM process.

When running at a bit depth of 16, an audio codec has a selection of 216 potential amplitude values to choose from. In a 32-bit system, this number jumps to the considerably larger 232, QED. Fig. 1 compares the potential amplitude values for an 8-bit and a 16-bit PCM audio stream. In actuality, the difference between an 8-bit system and a 16-bit system is notably more drastic than the illustration would indicate - the difference between 28 (256) and 216 (65,536). That difference increases exponentially when increasing the bit depth, so the advantage is obvious.

CD audio streams operate at a bit depth of 16, and combined with their sample rate of 44,100Hz, this provides audio playback which is in most cases indistinguishable from the original source recording. Many modern recording systems offer increased bit depths of 24 and 32-bit (floating point, as opposed to using solely whole numbers/integers). There is a lot of discussion among audio engineers as to the efficacy of using higher bit depths and sample rates. One quite convincing idea for future development is that of using only a 1-bit system (with only one relative amplitude value), but running it with an extreme sample rate of 2.8224 MHz. This concept, making use of pulse-density modulation codecs, is copyrighted as Direct Steam Digital by Sony and Philips, and may be an emerging format aimed at audiophiles in the near future.

Considering the concept of the Haas Window (as discussed in more detail on the psychoacoustics wiki page) suggests a figure of 30ms as the threshold within which two distinct sounds will be perceived as one single source by the human brain, an audio stream producing 2,822,400 samples a second would seem to be utterly indistinguishable from a "Direct Stream." The difficulty arises with a problem which is called quantization error, where the imprecise art of roughly estimating amplitude values based on the format's bit depth limitations can cause audible distortion and unwanted artefacts such as aliasing. As an interesting matter of fact, aliasing and quantization error has historically been used as a somewhat-musical element in some genres of music, namely chiptune, where only 8-bit recording resolution is used, harking back to the days of 8-bit computing and old handheld gaming consoles like the Nintendo GameBoy.