The distinction between ADC resolution vs accuracy is the difference between how finely a converter slices its input range and how closely each slice corresponds to the real voltage. Resolution is a property of the number of bits. Accuracy is a property of the reference, the converter’s linearity, the noise floor and everything in the signal path ahead of it. Adding bits improves the first and can leave the second untouched.
What resolution actually buys you
A converter with more bits divides its full-scale range into more steps, so the smallest change it can represent is smaller. That is genuinely useful when you need fine granularity, such as resolving a small change on top of a large signal. But nothing about a step size guarantees that the step boundaries are in the right places, that the full-scale range corresponds to the voltage you think it does, or that the reading will repeat if you take it again. A converter that reports many digits can still be wrong in every one of them beyond the first few. The extra digits are real information only if the errors are smaller than the step size, and that is the condition that usually fails.
The reference sets the ceiling on accuracy
Every converter measures its input as a fraction of a reference, so the accuracy of the result cannot exceed the accuracy of that reference. A reference with an initial tolerance, a temperature coefficient and long-term drift imposes all three on your measurement regardless of the converter’s resolution. Using the supply rail as a reference, which is convenient and common in small systems, ties the measurement to a rail that moves with load and temperature and carries noise. Where absolute accuracy matters, a dedicated reference chosen for its drift specification is the single most effective improvement available, and calibrating the system against a known input removes the initial tolerance while leaving the drift.
Linearity and offset errors inside the converter
Converters specify differential nonlinearity, describing how much individual step widths vary from the ideal, and integral nonlinearity, describing how far the overall transfer function deviates from a straight line. Large differential nonlinearity can produce missing codes, where certain output values never appear. Integral nonlinearity means a reading taken in one part of the range has a different error from one taken elsewhere, which matters when comparing measurements. Offset and gain errors shift and scale the whole transfer function and are the easiest to remove by calibration. All of these are specified in the datasheet in units of the least significant bit, so a converter with many bits and poor linearity may resolve less useful information than one with fewer bits and better specifications.
The circuit in front matters more than the converter
Most measurement errors in practice originate before the converter. The amplifier’s offset, drift and noise are added to the signal and converted faithfully. The source impedance interacts with the converter’s sampling capacitor, and if the amplifier cannot settle the input within the acquisition time the reading is simply wrong. A multiplexer switching between channels asks for a full-scale settle between samples. Supply noise couples into both the reference and the analog input. Thermocouple junctions formed by dissimilar metals at connectors add small voltages that matter at low signal levels. These effects are why a careful design with a modest converter outperforms a careless one with an expensive converter, and why datasheet accuracy figures describe the converter rather than the system.
Establishing what you can actually trust
Measure it rather than assuming. Apply a stable known input and take many readings: the spread tells you the noise, and the difference from the known value tells you the combined offset, gain and reference error. Repeat at several points across the range to see linearity, and repeat at temperature extremes to see drift. Shorting the input to the reference point gives a useful noise floor measurement. From those results you can state how many digits of the reading are meaningful, which is the number you should quote in a specification or display. Displaying more digits than the measurement supports is a common and misleading habit that implies precision the system does not possess.
What to check before you act
- Specify the reference on drift, not just on initial tolerance.
- Read differential and integral nonlinearity, not only the bit count.
- Confirm the driving amplifier settles within the acquisition time.
- Measure a known input repeatedly to find the real noise and error.
Common questions
Should I upgrade to a converter with more bits?
Only after establishing that resolution rather than noise, reference drift or front-end error is the limit. Otherwise the extra bits will simply resolve the existing noise more finely.
Can I use the supply rail as a reference?
It works for ratiometric measurements where the sensor is also powered from that rail, since the errors cancel. For absolute measurements it ties your accuracy to the supply.
The practical takeaway
Decide what accuracy the application needs, then find out which element limits it, because it is rarely the bit count. Fix the reference, the front end and the settling first, verify with repeated measurements against a known input, and display only the digits the system can actually support.
References and further reading
- Department of Energy: how electronics losses turn into wasted energy and heat
- CPSC: safety guidance for electronics and electrical products indoors
- FCC: electromagnetic compatibility and radio frequency safety questions answered
Numerical scenarios are illustrative unless identified otherwise. Follow the exact product instructions; component ratings and local installation requirements can differ.
