ADC Averaging / Oversampling Calculator

Estimate effective resolution gain (ENOB) from averaging or oversampling + decimation. Each 4× oversample adds 1 effective bit for white noise sources.

Calculator Electronics Updated Oct 15, 2026
How to Use
  1. Enter original ADC resolution (N bits) and number of samples averaged.
  2. The tool computes ENOB gain, effective resolution, and required sample rate.
  3. Averaging works only when noise is random (white) and uncorrelated with the signal.
  4. Rule of thumb: 4× samples adds 1 effective bit; 16× adds 2; 64× adds 3.
Input
bits
N (4, 16, 64, 256, ...)
Hz (kHz, MHz OK)
Hz
Presets
ENOB Gain Curve
ENOB Gain
—bits
Effective Res
—bits
Noise Reduction
—×
Output Rate
—

Worked Example

An Arduino's 10-bit ADC (5 V reference) reads a slow temperature sensor, averaging 64 samples taken at 10 kS/s. Noise falls by √64 = 8×, and every 4× more samples adds one bit, so 64 = 4³ adds 3 bits: 13 effective bits. The effective step shrinks from 4.88 mV to 5 ÷ 2¹³ = 0.61 mV.

The cost is speed: one averaged reading per 64 samples is 10,000 ÷ 64 = 156 readings a second, plenty for temperature but too slow for audio.

The common mistake: averaging a perfectly quiet signal. If the noise is well under 1 LSB, all 64 samples return the same code and the average is that same code, with no extra resolution at all. Averaging needs about 1 LSB of random noise (or added dither) to work.

Show Work

Enter values to calculate ENOB gain.

Formulas

ENOB Gain
ΔN = ½ × log₂(N)
Extra effective bits from averaging N samples.
Effective Resolution
N_eff = N + ½ × log₂(M)
Total effective bits after averaging M samples.
Noise Reduction
σ_avg = σ / √N
RMS noise drops by √N.
Oversample Ratio
OSR = fs / (2 × fbw)
Sample rate relative to Nyquist of signal.
Output Sample Rate
f_out = f_s / M
After decimation by M.
Quick Reference
4× = +1 bit
16× = +2, 64× = +3, 256× = +4, etc.

History of Oversampling and Averaging

The idea that averaging N noisy samples reduces RMS noise by √N goes back to Carl Friedrich Gauss's 1809 work on least-squares estimation. The specific application to ADCs — trading bandwidth for resolution — was developed in the 1960s–70s alongside the rise of digital signal processing. The √N rule gives one effective bit per 4× increase in sample count because each doubling of N reduces noise by √2 = one half-bit.

Σ-Δ (sigma-delta) converters take this to the extreme: a 1-bit modulator running at several megahertz, with noise-shaping to push quantization noise out of the signal band, produces 20+ bits of effective resolution at audio rates after digital decimation. Work by James Candy at Bell Labs and others through the 1970s and 1980s formalized the architecture that now dominates audio, precision measurement, and sensor interfaces.

The √N rule assumes white (Gaussian, uncorrelated) input noise. Real ADCs have 1/f (flicker) noise at low frequencies and systematic errors (INL, offset, gain) that don't average out. These place a practical ceiling — typically 3–4 effective bits of gain through simple averaging — beyond which better ADCs or more sophisticated architectures (chopper amplifiers, auto-zero circuits) are required.

About This Calculator

Enter the ADC's native resolution in bits and the number of samples you plan to average (or oversample + decimate). The tool computes the effective resolution gain (ΔN = ½ log₂N bits), new effective bit count, noise reduction factor, and — if you supply a sample rate and target signal bandwidth — the output sample rate after decimation.

Remember: averaging only gains you ENOB if the input has at least ~1 LSB of uncorrelated noise. If samples are identical (noiseless input), averaging returns the same value every time with zero improvement. Intentional dither (adding small white noise) is standard practice in precision DAQ. All math runs client-side.

Frequently Asked Questions

How does averaging add bits?

Noise on each sample is random. Averaging N samples reduces RMS noise by √N. A 4× reduction in noise is equivalent to a 2× reduction in LSB size — one extra bit of effective resolution. Every 4× increase in sample count adds one ENOB.

What kind of noise works?

White (Gaussian, uncorrelated) noise. Requires noise on the input that's at least 1 LSB RMS — otherwise all samples are identical and averaging does nothing. Thermal noise, flicker noise, or intentional dither provides this.

How is this different from oversampling?

Oversampling = sampling faster than needed, then decimating (low-pass filter + downsample). Mathematically similar to averaging — 4× oversample with a matched decimation filter gives +1 ENOB. Σ-Δ converters work this way built-in.

What's the practical limit?

Eventually you hit the 1/f (flicker) noise floor, after which more averaging doesn't help. Typical SAR ADCs can gain 2-4 effective bits through averaging before this limit. Σ-Δ converters are designed to overcome 1/f with chopping and calibration.

How long does averaging take?

Averaging N samples gives one result per N samples, so the output rate is f_s ÷ N. Averaging 64 samples taken at 10 kS/s gives 156 results a second, and anything changing faster than about half that (78 Hz) gets smeared. Each extra bit costs 4× the time, which is why averaging suits slow signals such as temperature, weight and battery voltage.

How do I use the ADC Averaging / Oversampling Calculator?

Simply type your numbers and read the result, which refreshes the instant you change something. There is nothing to submit and nothing to wait for.

Does it cost anything or need an account?

No. The tool is completely free, there is no account to create, and it keeps working offline after the page first loads.

Is anything I type uploaded?

No. The tool works entirely on your device, so the values you enter never leave your browser.

Common Use Cases

Temperature Sensor Smoothing

Average 64 samples to get +3 ENOB — a 10-bit Arduino reads a thermistor with ~13-bit effective resolution.

Weight Scale (Load Cell)

A 24-bit sigma-delta converter made for bridge sensors oversamples internally and delivers roughly 19–20 noise-free bits at slow output rates, enough to weigh to a fraction of a gram on a 1 mV/V load cell.

Battery Voltage Monitor

Low-speed averaging of battery rail: 256 samples at 100Hz gives very stable 1 mV resolution from a 10-bit ADC.

Audio DAC Dithering

Intentionally add 1 LSB RMS noise before quantization to allow averaging-based recovery of sub-LSB signal detail.

Strain Gauge Bridge

Signal levels <1mV need maximum averaging. 24-bit ADC + 16× oversampling yields usable ~18 ENOB.

Last updated: