Bit depth → dynamic range

SNR and dynamic range by bit depth

With ideal PCM quantization, every extra bit buys you a fixed amount of signal-to-noise ratio. The relationship is linear: SNR ≈ 6.02 · n + 1.76 dB, where n is the number of bits per sample. So 16-bit audio lands around ~98.1 dB, which is the CD standard and plenty for most rooms you'll ever listen in. Push to 24-bit and you get roughly ~146.3 dB, well past anything an analog noise floor can produce; that extra room is what lets you mix and edit without hearing quantization creep in. 32-bit float sits in a different category altogether. Because it stores values as an exponent plus a mantissa, there's no real SNR ceiling, and a signal that clips past 0 dBFS can still be pulled back down intact. Dither is the last piece: a sliver of low-level noise added before you quantize, which breaks up the correlation between the quantization error and the signal and effectively buys you another 1–2 bits of perceived depth down at the quiet end.

Applications: recording, mastering and broadcast

Studios track at 24-bit for two reasons: it leaves room when someone sets a level wrong, and it keeps quantization out of long signal chains. When the project is done, CD masters get dithered back down to 16-bit, and video broadcast (TV, streaming) usually goes out as 16-bit AAC. Inside the box, modern DAWs like Pro Tools and Reaper do their summing in 32-bit float, so a track clipping on its own never destroys data. The one place you still have to watch your gain is the master bus.

FAQ

Is 24-bit audibly better than 16-bit? On playback, most people won't hear a difference. Dithered 16-bit already covers more dynamic range than a normal room can use. Where 24-bit actually earns its keep is during recording and processing.

Why use dither? Skip it and the quantization error tracks the signal, which shows up as audible distortion when things get quiet. Dither trades that distortion for harmless white noise instead.

Does 32-bit float really have no limit? Functionally, no. The exponent range spans something like ~1500 dB, which is so far past what any analog source can deliver that clipping a float recording just doesn't happen.