ADC

An ADC, short for analog-to-digital converter, is a circuit that takes a continuously varying analog voltage and converts it into a discrete digital number a processor can actually compute with — the bridge every real-world signal, from a microphone's sound wave to a sensor's temperature reading, has to cross before any digital system can touch it. An ADC's output isn't a perfect continuous representation of its input; it's a finite set of discrete steps, determined by the converter's resolution, and understanding that step-based approximation is the key to understanding both an ADC's usefulness and its fundamental limits. How an analog voltage actually becomes a digital number, from a continuous signal down to a finite set of bits, looks like this:

A few things are worth knowing about why an ADC's two core parameters, sample rate and resolution, each throw away information in a different, deliberate way:

An ADC's resolution, expressed in bits, sets exactly how many discrete voltage levels it can distinguish, and every value between those levels gets rounded to the nearest one — an unavoidable error called quantization error that exists no matter how perfectly the rest of the converter is built. An 8-bit ADC divides its input voltage range into 256 discrete levels; a 12-bit ADC divides the same range into 4,096 levels, and a 16-bit ADC into 65,536 — each additional bit of resolution doubles the number of distinguishable levels, squeezing the quantization error in half. This rounding happens even in a theoretically perfect converter with zero other sources of error, which is why resolution alone sets a hard floor on how precisely an ADC can represent its input, regardless of how good the rest of its design is.

Sample rate, how many snapshots per second the ADC takes, has its own separate hard limit, set by the Nyquist-Shannon sampling theorem: a signal must be sampled at more than twice its highest frequency component, or the digitized result becomes permanently, unrecoverably corrupted by a distortion called aliasing. When an input signal contains frequency content above half the sample rate, that content doesn't simply disappear — it folds back down and masquerades as a lower, false frequency that's indistinguishable from a genuine low-frequency signal once digitized, and no amount of digital processing after the fact can separate the real signal from the alias it got mixed into. This is why real ADC systems place an anti-aliasing filter, an analog low-pass filter, before the converter itself, deliberately removing frequency content above the Nyquist limit before sampling ever happens — catching the problem before digitization rather than trying to fix it afterward, when it's already too late.

Different ADC architectures trade speed, resolution, and die area against each other in genuinely different ways, which is why a chip's analog front end often contains several different ADC types chosen for different jobs rather than one universal converter. A successive-approximation-register, or SAR, ADC narrows in on the input value one bit at a time through a binary search, offering a solid balance of speed, resolution, and power efficiency that makes it the default choice for general-purpose, moderate-speed conversion. A sigma-delta ADC instead oversamples the input at a much higher rate than its final output rate and uses digital filtering to trade that excess sample rate for extra resolution, making it the preferred choice for high-precision, lower-speed applications like audio and precision sensor readout. A flash ADC compares the input simultaneously against many reference levels in parallel, trading a large number of comparators and significant die area for extremely high speed, which is why it's reserved for applications needing very fast conversion, like high-speed data acquisition and radio-frequency receivers, where the area cost is worth the speed.

An ADC's real-world accuracy depends on more than its advertised resolution number alone, because non-ideal behavior in the actual silicon — nonlinearity, noise, and reference voltage instability — can erode precision below what the bit count alone would suggest. Differential and integral nonlinearity describe how far an individual ADC's actual step sizes and overall transfer curve deviate from perfectly even, ideal spacing; effective number of bits is a measurement-derived figure that captures how much real usable precision an ADC delivers once noise and nonlinearity are accounted for, which is often meaningfully lower than its nominal resolution suggests. This is exactly why a precise voltage reference, often a bandgap reference, matters so much to overall ADC accuracy — if the reference voltage that defines the top of the conversion range drifts or carries noise, every single converted value inherits that same error, no matter how clean the rest of the ADC's internal circuitry is.

Read an ADC as two deliberate, separate approximations stacked together, not a seamless analog-to-digital translation: sample rate approximates a continuous signal in time, bound by the Nyquist limit and the risk of aliasing, while resolution approximates it in amplitude, bound by quantization error — and the architecture chosen to make those approximations, SAR, sigma-delta, or flash, is a direct answer to whichever of speed, precision, or area a given application actually needs most.