Bit depth

A digital recording represents sample amplitudes using a finite set of numerical values. When a measured value falls between available levels, quantization maps it to a representable value. The difference is quantization error. In a properly dithered system, that error behaves as a low noise floor rather than distortion correlated with the signal. Increasing fixed-point bit depth allows more finely spaced amplitude values over the same full-scale range.

An idealized fixed-point system gains roughly six decibels of theoretical signal-to-quantization-noise ratio per additional bit. Actual recording equipment does not achieve the entire mathematical range implied by a 24-bit file because analogue electronics, converter design and microphone noise impose their own limits. File format capacity and measured converter performance should therefore be discussed separately rather than treated as interchangeable specifications.

Bit depth is independent of sample rate. A file can contain many measurements per second while representing each with relatively few bits, or use a larger amplitude word at a lower rate. These settings address different dimensions of the representation. Changing either after capture cannot repair distortion or recover information already buried beneath the original recording's noise.

A low quantization noise floor makes it practical to record with sensible peak margin. With typical 24-bit recording, an engineer does not need to chase the highest possible meter reading to preserve every quiet phrase. The microphone, room and preamp usually establish a more immediate noise limit. Leave enough headroom for an unexpectedly loud note instead of risking converter overload to use a few more visible meter segments.

This does not mean input gain is irrelevant. An extremely low microphone signal may expose preamp noise or make monitoring inconvenient, while excessive analogue gain can distort before the converter. Good level setting balances the source's expected range against the actual equipment. Bit depth provides numerical capacity, not permission to ignore the electrical stages ahead of the recording application.

Delivery requirements may specify a lower depth than the production session. A project recorded and processed at higher precision can still be prepared for such a destination. The important questions are when the reduction occurs, whether appropriate dither is applied and which file is the approved master. Maintain a higher-precision source so later versions do not begin from an unnecessarily reduced copy.

Fixed-point formats use a defined amplitude range with a hard numerical full-scale boundary. Floating-point formats store values using a significand and exponent, giving a much wider numerical range. A 32-bit floating-point file is therefore not simply a conventional 32-bit fixed-point recording with four billion equally spaced levels across the ordinary playback range. Its behavior and practical advantages require a different explanation.

Inside a floating-point workstation, some signals above the usual zero dBFS reference can be reduced later without the same numerical clipping that would occur in a fixed-point export. However, plugins may have their own limits or deliberately model analogue saturation. Output conversion still requires a valid level. Floating-point processing does not make every route immune to distortion or eliminate the need for gain staging.

Some recorders combine floating-point files with multiple converter paths to accommodate a broad input range. The file label alone does not establish the microphone's overload limit or the recorder's complete analogue capability. Read the actual input specifications. A clipped capsule, overloaded input stage or unsuitable source connection remains a recording problem regardless of how the resulting samples are stored.

For an acoustic recording in Abu Dhabi, the engineer chooses 24-bit capture and asks the player to perform the loudest expected section. Input gain is set conservatively, allowing the musician to become more energetic during a complete take. The engineer checks quiet passages for room noise and equipment hiss instead of assuming any low meter reading represents lost musical detail.

After editing and mixing, the project remains at its normal processing precision. One recipient requests a 16-bit version, so the engineer prepares that version from the approved high-precision source. Dither is applied at the appropriate final reduction stage, with settings suited to the intended use. The resulting file is checked for correct format, clean endings and unwanted noise introduced by an accidental processing choice.

A later revision starts from the preserved session or high-precision master, not from the 16-bit derivative. This keeps the production history clear and avoids treating a distribution copy as the only surviving source. The engineer also records the export settings so another collaborator can distinguish the archive version from a specific delivery version without guessing from the filename alone.

• Recording extremely close to full scale because unused meter space appears wasteful increases the risk of irreversible overload. A clean recording with peak margin is generally more useful than one with slightly higher readings and damaged transients.
• Assuming a 24-bit converter delivers 24 bits of real analogue performance confuses storage resolution with measured noise and distortion. Evaluate the whole capture path, including the source and microphone, rather than relying on the container specification.
• Using floating-point files as a guarantee against clipping overlooks analogue limits and plugin behavior. Identify where distortion occurred before deciding that reducing a file's playback level will restore it.
• Adding dither indiscriminately at every intermediate export can accumulate unnecessary noise. Plan precision changes and apply appropriate dither when reducing fixed-point resolution, rather than inserting it automatically wherever a bounce button appears.
• Increasing the bit depth of an already quantized file does not recreate missing low-level information. A larger working format can be useful for subsequent processing, but it should not be described as recovering the detail of a new recording.