Bit depth explained: what 16-bit and 24-bit audio change

Levels, steps and the 6 dB rule

A digital recording stores the height of the waveform at each sample as a whole number. With 16 bits there are 65,536 possible numbers, and with 24 bits there are 16,777,216. Each sample is rounded to the nearest one.

Adding a bit doubles the count and halves the size of each step. Halving an amplitude is a change of 6.02 dB, from 20 × log10(2), so every extra bit moves the smallest step about 6 dB further below full scale. The span from full scale, written 0 dBFS, down to the level of that smallest step is the format's dynamic range.

That is all bit depth governs. Which frequencies a file can hold is set by the sample rate. A quiet cymbal in a 16-bit file is stored with the same frequency content as in a 24-bit one. The difference is how far below it the format's own noise sits.

Quantization error, and what dither does about it

Rounding each sample leaves an error of up to half a step. On loud, busy music the error behaves like a faint hiss far down in the range. On fades and very quiet passages, where the signal spans only a few steps, the error follows the music and becomes distortion, a grainy edge on a decaying note.

Dither is a small amount of random noise added before the rounding. It breaks the link between the error and the signal, so what is left is a steady, constant hiss at a known low level, and detail smaller than one step survives inside that hiss. Noise-shaped dither goes further and moves most of the hiss toward the frequencies where hearing is least sensitive, which lowers how loud it seems.

Any time a 24-bit master is reduced to 16 bits for release, it should be dithered, and competent mastering does it as routine. The bit depth calculator has a live demo that plays a dithered signal at the depth your volume setting leaves, beside a 24-bit reference, so you can hear where the hiss sits.

How many bits a real converter delivers

A "24-bit" or "32-bit" line on a spec sheet names the data format a DAC or interface accepts. The analog circuit that follows has noise of its own, from resistors and transistors, and no converter reaches the 144 dB that 24 bits describe. The dynamic range figure tells you how much of the format survives. Dividing it by six gives a rough count of the bits that sit above the analog noise.

• The Yellowtec PUC2 Mic LEA records at 192 kHz / 24-bit and is rated at 108 dB of dynamic range, about 18 bits' worth.
• The SMSL RAW-MDA1 takes 32-bit files and is rated at 132 dB from its XLR outputs, 127 dB from RCA and 122 dB from the headphone jack, so about 22 bits at best.
• The SMSL PS100 accepts 24-bit/192 kHz over optical and coaxial, and SMSL rates its dynamic range at 96 dB, the theoretical figure for 16 bits.

Falling short of the format is normal. Thirty-two bits would describe about 192 dB, and no analog output comes close to that. How far short still counts: the PS100's review lists its modest 96 dB among the cons. Support for a large bit depth tells you which files a DAC will play, and the dynamic range tells you what it does with them.

Why recording and editing use 24 bits or more

A recording engineer can't know the loudest peak before it arrives, so levels are set well below full scale. At 16 bits, leaving 18 dB spare throws away three bits and brings the noise floor closer. At 24 bits the same margin costs nothing you could hear, because the converter's analog noise already sits above the lowest bits.

Editing adds its own demands. Every change of gain, every EQ and every fade produces values that need more bits than they started with, so audio software usually calculates internally at 32-bit floating point or higher. Floating point stores a separate exponent with each value, so a level pushed above full scale inside the computer is kept intact and can be turned back down later. The converter itself can still clip, and a float file holds no more than the converter captured.

The PUC2 shows the other way to handle recording levels. It has no gain knob at all, and its LEA automatic leveling keeps the input in range. The review found it almost impossible to overdrive in normal use, despite being "just" 24-bit where some interfaces offer 32-bit float.

Digital volume control and lost bits

When a computer or phone turns the volume down in software, it scales every sample by a number smaller than one. The music then uses fewer of the available steps, and at a 16-bit output every 6 dB of attenuation costs a bit. Thirty decibels down, a 16-bit stream has about 11 bits left above its floor, roughly 66 dB of range.

Sending 24 or 32 bits to the DAC keeps that rounding far below the converter's own analog noise, so setting your operating system to its highest output depth is a sensible default. A DAC with no digital volume, such as the Henry Audio DA 256, leaves the level to an analog control downstream, so with the computer's slider at maximum nothing is lost. The bit depth calculator shows what your slider position costs at each output depth.

Choosing files and gear by bit depth

For finished music, 16 bits are enough. If the loudest peaks of a 16-bit file played at 110 dB SPL, which is uncomfortably loud, its dithered noise floor would sit below 20 dB SPL, under the background noise of a quiet living room. A 24-bit download is harmless and costs only storage. If the hi-res version of an album sounds different, compare the masters first: a different EQ or level on that release is a far larger change than eight more bits of range.

When you buy a DAC, read the dynamic range or signal-to-noise figure and ignore the bit-depth headline, then check how quiet it stays with your headphones, which the noise floor explainer covers. The DAC guide ranks the converters reviewed here.