An analogue-to-digital converter, usually abbreviated ADC, is an electronic circuit that converts an analogue signal into a digital representation consisting of numerical samples. Analogue signals can vary continuously in time and amplitude, such as microphone voltage, temperature sensor output, radio-frequency signals or the voltage produced by a scientific instrument. Digital processors cannot directly operate on an ideal continuous-time quantity, so an ADC measures the analogue input at specified instants and assigns each measurement a digital code.
Two conceptually distinct processes are involved: sampling discretises the signal in time, while quantisation discretises its amplitude. The resulting digital sequence can then be stored, transmitted or processed by CPUs, DSPs, microcontrollers, FPGAs and other digital systems. ADCs are fundamental to digital audio, cameras, mobile devices, industrial automation, medical instrumentation, software-defined radio, scientific measurement and control systems. Their quality determines how faithfully physical information is represented before subsequent digital processing takes place.
Sampling frequency is one of the central ADC parameters. It specifies how frequently the analogue input is measured, normally expressed in samples per second or hertz. The Nyquist-Shannon sampling theorem establishes that a band-limited signal can be reconstructed from ideal samples when the sampling frequency is greater than twice its highest frequency component. If significant frequencies above half the sampling frequency reach the ADC, they can be aliased into lower frequencies and become indistinguishable from genuine lower-frequency content.
Because aliasing cannot generally be removed after sampling, practical ADC systems use analogue anti-aliasing filters before conversion. Oversampling can make filtering easier by moving unwanted spectral components farther from the signal band. The required sampling rate depends on the application: digital audio may use rates such as 44.1 or 48 kHz for conventional audio, while instrumentation, communications and scientific systems may use rates ranging from very low frequencies to many gigasamples per second.
ADC resolution determines how finely the input amplitude range is divided into digital codes. An ideal N-bit ADC has 2^N possible output codes. With an ideal unipolar converter spanning a voltage range from zero to a reference voltage, the nominal quantisation interval is approximately the full-scale range divided by 2^N. Quantisation introduces a difference between the actual analogue value and the discrete code selected to represent it.
Under appropriate assumptions, quantisation error can be modelled approximately as noise, although real converters exhibit additional error mechanisms. Resolution should not be confused with accuracy. A 16-bit ADC can distinguish many nominal levels but may have offset error, gain error, nonlinearity, noise, reference instability and other imperfections that reduce its effective performance. Effective number of bits, often abbreviated ENOB, is a practical measure that reflects the converter's usable resolution after noise and distortion are considered.
Several ADC architectures exist because different applications require different combinations of speed, resolution, power consumption and cost. Flash ADCs use many comparators operating in parallel and can achieve extremely high conversion rates, but their hardware complexity grows rapidly with resolution. Successive-approximation-register ADCs, or SAR ADCs, use a comparator, internal DAC and successive decision process to determine the input code and provide an effective balance of speed, resolution and power for many applications.
Pipeline ADCs divide conversion into stages and can provide high throughput with some latency, making them common in high-speed data acquisition and communications. Delta-sigma ADCs use oversampling and noise shaping and can provide high effective resolution at comparatively lower bandwidths, making them common in precision measurement and audio. Dual-slope and integrating ADCs are slower but can provide strong rejection of certain noise components and are widely used in some digital measurement instruments.
ADC performance depends on considerably more than nominal bit depth and sampling rate. Important specifications include signal-to-noise ratio, total harmonic distortion, signal-to-noise-and-distortion ratio, integral nonlinearity, differential nonlinearity, offset error, gain error, aperture uncertainty and conversion latency. Integral nonlinearity measures deviation of the transfer characteristic from an ideal relationship, while differential nonlinearity measures variation in individual code widths. Missing codes can occur when differential nonlinearity becomes sufficiently severe.
Aperture uncertainty concerns uncertainty in the precise instant at which the analogue input is effectively sampled and becomes particularly important at high input frequencies. Clock jitter can therefore degrade the effective signal-to-noise ratio of high-frequency inputs even when the converter has many bits. Reference-voltage accuracy and stability are also crucial because the reference establishes the scale used to translate analogue input values into digital codes.
Before the ADC, analogue signal conditioning is normally required. Sensors may produce signals that are too small, too large, noisy or incorrectly centred for direct conversion, so amplifiers, attenuators, filters, instrumentation amplifiers, level shifters and other circuits may be used. An anti-aliasing filter limits the analogue bandwidth before sampling. The ADC input may also require a suitable source impedance and acquisition time, especially for SAR architectures with switched-capacitor inputs.
In some systems, a sample-and-hold circuit captures the input voltage and holds it sufficiently constant during conversion. Differential inputs can improve immunity to common-mode interference and can support signals centred around a specified common-mode voltage. Power-supply quality, grounding, PCB layout and electromagnetic interference can affect converter performance because small analogue errors can become measurable digital errors. The ADC is consequently part of an entire signal-acquisition chain rather than an independent component whose specification alone determines system accuracy.
ADCs are used whenever a digital system needs information from the physical analogue world. A microphone and audio interface use an ADC to convert electrical representations of sound into digital samples. A camera's image sensor produces electrical measurements that are converted into digital pixel values by ADC circuitry, often integrated closely with the sensor architecture. Medical instruments use ADCs to digitise physiological signals, while industrial controllers use them to measure temperatures, pressures, currents and voltages.
Radio receivers use high-speed ADCs to digitise portions of received radio signals so that DSP can perform filtering, demodulation and decoding. Data loggers repeatedly sample sensors and store the resulting codes for later analysis. In each case, the ADC determines how much information about the analogue source is preserved in the digital representation and how much is lost through bandwidth limits, noise, quantisation and converter imperfections.
A useful distinction is between sampling, quantisation and encoding. Sampling determines when the analogue waveform is measured; quantisation determines which discrete amplitude level represents each measurement; encoding represents that selected level as a digital code. An ADC therefore does not merely "measure voltage and make it binary" as a single indivisible action. The conversion process is governed by references, comparator decisions, internal DACs or other circuit mechanisms depending on architecture.
Once digitised, information can be subjected to DSP, compression, storage, transmission and analysis without requiring continuous analogue circuitry for every subsequent operation. However, digitisation cannot recover information that was never captured within the chosen bandwidth and dynamic range, and aliasing can permanently alter frequency information if sampling is inadequate. Modern ADCs range from tiny low-power converters embedded in microcontrollers to extremely fast, high-resolution devices used in scientific and communications equipment. Their fundamental purpose remains to establish a controlled numerical representation of a continuously varying analogue input.
