Starlight Tools

Audio Bit Depth Calculator — Dynamic Range, Quantization Levels & Noise Floor

Calculate the exact number of integer PCM code values, LSB size, ideal full-scale-sine SQNR, integrated quantization noise, and the effect of recording below 0 dBFS.

Audio quantizer settings

Whole-number PCM resolution from 1 to 64 bits. See the 32-bit float note below.

0 dBFS is full-scale peak. Use a negative value for normal recording level.

Common integer PCM depths

Your values stay in this browser. They are not uploaded, saved, or added to the URL.

Bit-depth result

Ideal full-scale sine SQNR 146.26 dB 24-bit ideal uniform quantizer; integrated over the Nyquist band
Quantization levels16,777,216
One-LSB range144.49 dB
Ideal noise floor−146.26 dBFS
SQNR at entered level128.26 dB
One LSB of full-scale span0.00000596046%
RMS quantization error0.288675 LSB
Signed PCM code range−8,388,608 to 8,388,607
Approx. precision used at signal level21.01 bits

Theoretical result: This is an ideal integer quantizer model, not a measured converter or complete recording-chain noise specification.

How to use this audio bit depth calculator

  1. Enter the stored integer precision. Choose 16 or 24 bits for common PCM recording, or enter another whole-number depth.
  2. Set the sine-wave peak level. Use 0 dBFS to see the formula's full-scale limit. A value such as −18 dBFS shows how lower signal level reduces SQNR relative to the same fixed quantization noise.
  3. Read levels and LSB size. Every extra bit doubles the available code count and halves the ideal step size.
  4. Treat the dB result as a ceiling. Real A/D and D/A converters add analog noise, distortion, jitter, and implementation limits. Consult measured dynamic range, SNR, SINAD, or ENOB for hardware comparisons.

For example, 24-bit integer PCM has 16,777,216 codes and an ideal full-scale-sine SQNR of 146.26 dB. With a sine-wave peak level of −18 dBFS, the signal SQNR is 146.26 − 18 = 128.26 dB. Lowering the signal reduces its power relative to the modeled quantization noise. The stored bit depth and quantization step remain fixed, so the code count and modeled integrated quantization-noise power remain unchanged.

Bit depth, dynamic range, and quantization noise formulas

For an ideal, uniform, linear N-bit integer quantizer with a sufficiently varying signal, step size q, and error assumed uniformly distributed from −½ LSB to +½ LSB:

Quantization levels = 2ᴺ
One LSB as a fraction of full-scale span = 1 ÷ 2ᴺ
One-LSB range = 20 log₁₀(2ᴺ) = 6.0206N dB
RMS quantization error = q ÷ √12 = 0.288675 LSB
Ideal full-scale sine SQNR = 6.0206N + 1.7609 dB
SQNR at signal level L dBFS = 6.0206N + 1.7609 + L

The 1.7609 dB term is not an extra bit of resolution. It comes from comparing the RMS value of a full-scale sine wave with the RMS value of the ideal uniform quantization error. This is why a one-LSB span calculation gives 6.02N dB while full-scale-sine SQNR gives 6.02N + 1.76 dB. The displayed SQNR is a model estimate under the stated assumptions. At very low bit depths, or for very quiet or insufficiently varying signals, the error may not follow the assumed uniform distribution and may be correlated with the signal, so actual SQNR can differ. The noise figures do not include added dither or noise shaping.

On this page, “ideal noise floor” is the negative of the full-scale-sine SQNR and refers to total integrated quantization noise across the Nyquist band. The input level uses the sine-wave peak relative to full-scale peak, whereas the displayed negative-SQNR noise floor uses the RMS value of a full-scale sine as its reference. If full-scale peak is normalized to 1, that sine has RMS 1 ÷ √2, and the displayed noise floor is 20 log₁₀[(q ÷ √12) ÷ (1 ÷ √2)]. Thus the displayed −146.26 dBFS for 24 bits uses a full-scale-sine RMS reference; expressing the same noise RMS relative to peak amplitude 1 would give a value about 3.01 dB lower. Match both the reference and measurement bandwidth when comparing noise figures. This integrated result is not a per-Hz noise-density value or the height of one FFT bin. FFT noise-floor appearance changes with sample rate, transform length, window, and bandwidth.

Important: 32-bit float is not a 32-bit integer quantizer

Choose 32 bits here only for 32-bit integer PCM. IEEE 754 single-precision floating point uses a sign, an 8-bit exponent, and 23 stored fraction bits with an implicit leading bit for normal values. That gives about 24 bits of significand precision while the exponent supplies a far wider representable level range. Therefore, applying 6.02 × 32 + 1.76 to a complete 32-bit float format is not a meaningful fixed dynamic-range specification.

Floating-point DAW files can preserve values above 0 dBFS internally, but the analog input converter's real capture range is still limited by its hardware. Bit depth also does not determine file loudness or prevent analog clipping.

Common integer PCM bit depths compared

Bit depthExact levelsOne-LSB rangeIdeal full-scale sine SQNRTypical context
8-bit25648.16 dB49.93 dBLegacy or deliberately lo-fi integer PCM.
16-bit65,53696.33 dB98.09 dBAudio CD and many delivery files.
20-bit1,048,576120.41 dB122.17 dBSome legacy professional converters and formats.
24-bit16,777,216144.49 dB146.26 dBCommon production recording and interchange depth.
32-bit integer4,294,967,296192.66 dB194.42 dBInteger processing or storage; distinct from 32-bit float.

These are mathematical limits for the stated model. They are not claims about the acoustic noise, analog electronics, or effective performance of a product.

What bit depth changes in real audio work

  • More capture margin: a lower quantization step lets engineers leave sensible headroom without bringing the ideal quantization floor close to ordinary recording levels.
  • More storage: uncompressed PCM data rate grows linearly with bit depth. Use the Audio File Size Calculator to calculate the storage impact.
  • Dither behavior: properly chosen dither decorrelates quantization error and makes very low-level distortion noise-like, but it does not create more PCM code values.
  • Hardware reality: nominal bits and effective number of bits are different. Analog noise and distortion usually make a converter's measured performance lower than the ideal N-bit limit.

Audio bit depth FAQ

How many quantization levels does an N-bit audio format have?

It has 2N code values. For example, 16-bit PCM has 65,536 possible values per sample and 24-bit PCM has 16,777,216.

What is the theoretical dynamic range of 16-bit and 24-bit audio?

Using ideal full-scale-sine SQNR, 16-bit audio is about 98.09 dB and 24-bit audio is about 146.26 dB. The simpler one-LSB span is about 96.33 dB and 144.49 dB respectively. Neither is a measured hardware specification.

Why do some calculators use 6.02N and others use 6.02N + 1.76?

They use different references. The first is the amplitude ratio from full-scale span to one LSB. The second is RMS SQNR for a full-scale sine wave against ideal uniform quantization error.

Is quantization noise floor the same as measured interface noise?

No. This result includes ideal quantization error only. An audio interface also contains analog circuit noise, distortion, jitter, filtering, power-supply effects, and other implementation limits.

Does recording at −18 dBFS waste three bits?

No stored bits are removed. The displayed “Approx. precision used at signal level” is an approximate amplitude-resolution equivalent: 24 − 18 ÷ 6.0206 ≈ 21.01 bits for a 24-bit signal with a sine-wave peak of −18 dBFS. It describes the lower signal amplitude relative to the fixed quantization step. The format remains 24-bit PCM with 16,777,216 available codes; the estimate does not mean that three stored bits have been removed. It is also not measured converter ENOB, which reflects actual converter performance including noise and distortion. With 24-bit storage, the ideal mathematical margin remains very large; actual analog-system noise is usually the more relevant limit.

Does 32-bit float audio have a fixed 194 dB dynamic range?

No. That figure comes from treating 32 bits as a linear integer code. Single-precision float has about 24 bits of significand precision plus an exponent, so it behaves differently and has no single fixed LSB across its full range.

Does dither increase bit depth?

No. Dither does not increase the code count. It trades correlated quantization distortion for a controlled noise characteristic and can make signals below one LSB statistically recoverable over time.

Are my settings uploaded or saved?

No. The calculation, clipboard report, and CSV file are created locally by JavaScript in your browser.

Method and references

Formula and assumptions checked 2 August 2026. Analog Devices derives the RMS quantization error and the classical ideal full-scale-sine result. Texas Instruments documents the same ideal ADC equation, distinguishes effective bits from nominal resolution, and explains that thermal noise and jitter also contribute to real converter noise.

Explore more tools