Recording – The How and Why of Dither

Computers store information as binary numbers,
Image placeholder title

 There are three types of POW-R dither.

Image placeholder title

Computers store information as binary numbers, where something is represented in one of two states—one or zero; on or off. Each sample or measurement contains a specific number of ones and zeros that represents the analog voltage level at a specific point in time. This measurement is called quantization, and it determines the dynamic range of a digital recording. The more data we capture per sample, the greater the dynamic range we can record (up to a point), and the quieter the sounds we can capture. In other words, with 0dBfs (decibels relative to full scale; the maximum digital level) as our ceiling, the bit depth we sample determines how far below that ceiling we can accurately quantize, or record, a signal.

Because we’re measuring a continuous voltage in discrete intervals, there is a point below the quantizing resolution where you have steps in the waveform, albeit small ones, and you theoretically cannot capture low-level information below this point. The level has dropped below what the least significant bit (LSB) can register. This is where dither comes in.

Dither is a randomized signal—noise—that is added before the input of the ADC or during bit reduction. It not only mitigates the distortion caused by quantization errors when you reduce the word length of a PCM file, but it can extend the dynamic range of the bit depth beyond its theoretical limits. Adding low-level noise causes the LSB to toggle, which lets you statistically capture input values that are smaller than the LSB: This creates an average value that incudes levels that are below what the LSB would normally be able to encode. In other words, dither allows you to encode less than one bit’s-worth of information. For example, Sony’s 16-bit DAT machines offered Super Bit Mapping (SBM) technology when recording, which yielded a dynamic range resembling 20-bit resolution through noise shaping.

Perceptually based noise-shaping algorithms distribute the spectral energy of the dither signal into frequency ranges where the human hearing range is least sensitive. It’s common for software developers to put the majority of dither noise above 15kHz, moving it towards the Nyquist limit of 22.05kHz for a 44.1kHz sampling rate.

Software products that have dither usually include algorithms that are optimized for different types of program material. There are three types of POW-R dither: type 1 for spoken word; type 2 for music with a limited dynamic range, such as rock and hip-hop; and type 3 for music with a wide dynamic range, such as classical music (see figure 1). The Waves L2 Ultramaximizer offers two types of dither with three levels of noise shaping. The dither portion of WaveArts FinalPlug includes 11 dither/noise-shaping algorithms.

Hearing is Believing You can hear the effects of dither at the end of an audio file. Play a 24-bit file without dither and listen to the way the sound fades out. In a well-recorded file, you should hear the sound decrease evenly to silence. (Turn up the output level of your playback system to hear the final moments clearly, but be careful to reduce the level before further playback, so you don’t damage your speakers.)

Next, create a 16-bit version without dither. Listen to the fade out (again, increasing your playback system’s output level). As the non-dithered signal fades to zero, you’ll hear a distorted crunchiness. Now, create a 16-bit file using dither and listen to the fade once again. At the very end, you’ll hear a smooth transition into hiss-like noise.

Try each type of dither your DAW offers and listen to the way it changes the frequency spectrum and soundstage of your mix, and listen for added distortion or graininess in the sound.

Do not use noise-shaped dither more than once. It should be the final process when converting 24-bit files to 16 bits for use in a CD master, and you shouldn’t process these files any further. Typically mastering engineers apply bit reduction with the proper dither after they’ve completed their work. You’ll get the best results if you deliver full-resolution 32- or 24-bit files of your mixes and let the mastering engineer do the processing.

For a detailed explanation about dither in easy-to-understand language, read Mastering Audio: the Art and the Science, 2nd Edition by Bob Katz (2007, Focal Press).

Gino Robair is a former editor of Electronic Musician.