Choosing an ADC: why more bits do not automatically mean more accuracy

Sinuss Redactie

Vertical bit bar with the upper part marked as usable bits (ENOB) and the lower part fading into a hatched noise floor, beside a signal path from source and reference into the ADC.

A 24-bit ADC does not automatically provide 24 stable measurement bits. Output word length describes the digital scale, while noise, distortion, reference and input circuitry determine how much information is usable in practice.

Check these eight points

  1. Signal range: define minimum, maximum, common mode and unipolar or bipolar input.
  2. Resolution and LSB: calculate the smallest theoretical step required.
  3. ENOB and noise: use datasheet values at the relevant input, frequency and data rate.
  4. Sample rate and bandwidth: include anti-alias filtering, settling and multiplexing.
  5. Architecture: SAR, delta-sigma and other types offer different strengths and latency.
  6. Reference: accuracy, noise, drift and decoupling affect the entire scale.
  7. Analogue input: verify impedance, driver, acquisition time and permitted overvoltage.
  8. Interface and system: channels, SPI/I2C/parallel, clock, supply, package and software matter.

Do not compare ADCs only by bits and samples per second. Check SNR, SINAD, ENOB, offset, gain error, INL, DNL and noise under conditions close to your application.

Define signal, required real resolution, speed, channels and interface. Then search for ADCs and voltage references and verify the complete signal chain.