Signal Chain Noise Budget and ENOB Calculator
Enter each stage’s gain and input-referred noise. The calculator refers everything back to the input, ranks the stages, folds in the converter’s own quantisation noise, and tells you the effective bits you are actually getting.
| Stage | Gain (V/V) | Input-referred noise (nV/√Hz) | Gain to here | Contribution RTI | Share |
|---|
Runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere.
Noise does not add the way people expect
A signal chain’s noise is not the sum of its stages, and it is not dominated by the noisiest part. Each stage’s noise is referred back to the input by dividing by all the gain ahead of it, so a mediocre stage sitting behind 100× of gain contributes one hundredth of its noise to the total. This is the same result as Friis’ formula in RF, and it has one blunt consequence: the first stage with real gain sets the noise floor, and almost nothing downstream can recover it.
That is why a low-noise op-amp placed after the gain stage is decoration, why an expensive anti-alias filter rarely improves anything, and why the single most effective change in most chains is moving gain earlier rather than making any individual part quieter.
eRTI = √( esrc² + e1² + (e2/G1)² + (e3/(G1G2))² + … )SNR = 20 log10( VFS/(2√2) / eout ) ·
ENOB = (SNR − 1.76) / 6.02The converter is a noise source too
Quantisation noise is real noise, and this tool refers it back to the input alongside everything else. That comparison is the one that saves money. If the ADC dominates, your analog chain is already quieter than the converter can resolve and every further improvement to it is wasted. If the analog chain dominates, buying a higher-resolution converter buys you nothing at all — you will simply digitise your own noise more precisely.
Quantisation noise is LSB/√12 spread across the Nyquist band, which is where
oversampling gets its advantage: the noise is fixed in total power, so sampling faster and
averaging pushes most of it outside the band you care about. Every 4× of oversampling buys
about one extra bit, and only until the analog noise floor stops it.
Reading the result honestly
- The share column is noise power, not voltage. A stage at 40% of the power is about 63% of the voltage. Halving the largest contributor rarely halves the total — it moves you to wherever the next-largest term is, which is why the tool ranks them rather than just totalling.
- Effective bits are the honest figure. A 16-bit converter fed by a chain that delivers 12.5 effective bits is a 12.5-bit system with an expensive part in it.
- Bandwidth is a design variable. Noise scales with its square root, so halving the bandwidth buys 1.5 dB. If you do not need the speed, the cheapest noise improvement available is a lower filter corner.
What else is missing
- Current noise through source impedance, which for a high-impedance sensor can exceed the voltage-noise term entirely.
- Resistor noise in the gain network — a 100 kΩ feedback resistor brings its own 40 nV/√Hz to the party. The divider calculator models both thermal and excess resistor noise.
- Interference, which is not noise. Supply ripple, clock feedthrough and ground coupling are deterministic, do not add in quadrature, and are usually what actually limits a real board.
- Aliasing. Wideband noise folds into the band at every sampling edge. If the anti-alias filter is inadequate, measured noise will exceed anything predicted here.
More free tools
Each of these runs entirely in your browser. Nothing is uploaded, stored or sent anywhere, and none of them needs an email address.