What Is a DAC in Audio? How It Works and What Matters

If you searched for “what is a dac in audio explained”, the short answer is that a digital to analog converter takes the stream of numbers that make up a digital recording and turns it into a continuously varying voltage an amplifier can drive into headphones or speakers. The more useful question is how much of a product’s sound the converter chip actually controls, and which numbers on the box tell you anything.

One fact reframes the topic: every device that plays sound already contains a DAC, because there is no other way to get music out of a file. An “external DAC” does not add a missing stage. It replaces one that was already there, and whether that changes anything depends on what was wrong with the original.

What a DAC actually does

A digital recording is a list of samples: measurements of the signal’s amplitude taken at a fixed rate, each stored as a binary number with a fixed number of bits. CD audio uses 44,100 samples per second at 16 bits. A DAC reads those numbers in order, at exactly the rate they were recorded, and produces a voltage that follows them.

A DAC is not an amplifier: conversion and headphone drive are separate jobs, as DAC vs amp: which one you actually need explains.

Two rules govern how faithfully that happens. The first is the sampling theorem. Texas Instruments’ oversampling data converters seminar states it plainly: “the sampling frequency must be at least twice the information bandwidth of the input signal.” At 44.1 kHz, everything below 22.05 kHz is captured completely, and the intuition that samples form a coarse staircase is wrong. Xiph.Org’s explainer on high sample rates notes that the analog signal “can be reconstructed losslessly, smoothly, and with the exact timing of the original.”

The second rule concerns bit depth. The TI seminar gives the ideal-converter formula for a full-scale sine wave: SNR = 6.02N + 1.76 dB, where N is the number of bits. That puts 16-bit at roughly 98 dB and 24-bit at roughly 146 dB on paper. The 16-bit figure is conservative: with shaped dither, which moves quantisation noise toward frequencies where hearing is least sensitive, Xiph reports the effective dynamic range of 16-bit audio “reaches 120 dB in practice.” No consumer chain approaches the 24-bit paper figure; analog noise in the output stage sets the floor first.

Inside a modern audio DAC: oversampling and sigma-delta

Almost no audio DAC sold today converts 24 bits directly with a resistor ladder. The dominant architecture is sigma-delta (also written delta-sigma), and it explains most spec-sheet vocabulary.

The digital input is first oversampled: an interpolation filter raises the sample rate to many times the original. TI describes the payoff: “Oversampling increases the bandwidth considerably, hence it spreads the quantization noise over a wider bandwidth. The result is lower inband quantization noise.” The oversampled stream then feeds a modulator that outputs a very fast, very low resolution bit stream (one bit in the classic form, a few bits in most current parts), using noise shaping to push the remaining quantisation noise into ultrasonic frequencies. A simple analog low-pass filter on the output removes that noise along with the images of the audio band above half the oversampled rate.

The interpolation filter is where the “filter options” on many products come from. ESS Technology’s ES9038PRO datasheet lists “7 ready-to-use preset filters with linear/minimum phase and low-delay.” Linear phase filters ring symmetrically before and after a transient; minimum phase filters ring only after it. Whether the difference is audible is contested; that both are digital-domain choices is not.

The last stage is analog. Many high-end chips output a current rather than a voltage, so the product designer adds a current-to-voltage stage, an output buffer, and in a combined unit a headphone amplifier, none of which the chip vendor designs.

Bits, sample rates, and what the headline numbers buy you

The ES9038PRO datasheet claims “up to 140dB DNR (mono mode)” and “-122dB THD+N,” figures ESS announced with the SABRE PRO series on 6 January 2016. The same part handles “up to 32-bit 768kHz PCM, DSD256 via DoP and native DSD1024.”

The 140 dB figure is chip-level, in mono mode, meaning several converter channels paralleled onto one output. A finished stereo product with a real power supply and output stage measures lower, and two products built on the identical chip can differ by tens of decibels. The chip sets a ceiling; the implementation decides how far below it the product lands, and a listing that names the chip without measuring the finished unit is describing the ceiling.

The sample rate figures are capability, not benefit. Xiph argues that ultrasonic content in 96 kHz or 192 kHz files is at best inaudible and at worst harmful, because “any nonlinearity will shift some of the ultrasonic content down into the audible range as an uncontrolled spray of intermodulation distortion products.” A DAC that accepts 768 kHz does not produce better sound from a 44.1 kHz file. It is format compatibility, which matters only if you own files in those formats.

Clocks, jitter, and why USB mode matters

A DAC must output each sample at precisely the right moment. Sound On Sound’s guide to digital problems defines the failure: jitter is “very short-term timing variations between one sampling moment and the next.” At 48 kHz each sample should land 20.83 microseconds after the last, and a sample produced slightly early or late “is distorting the true waveform.” The same article calls jitter “rarely a practical problem these days,” because modern converters isolate the incoming clock from their own reconstruction clock; the ES9038PRO datasheet lists a “Time Domain Jitter Eliminator” for exactly that.

Over USB the fix is architectural: in asynchronous mode the DAC, not the computer, owns the clock. Benchmark’s DAC2 application note puts it in one sentence: “A clock generated inside the DAC2 is used to transfer the data.” USB Audio Class 2.0 supports asynchronous operation, but the USB class and the software driver are separate compatibility questions.

USB Audio Class 2.0 and the usbaudio2.sys driver

Microsoft has shipped the in-box USB Audio Class 2.0 driver, usbaudio2.sys, since Windows 10 version 1703; later Windows 10 releases and Windows 11 include it. Microsoft’s driver documentation lists PCM support from 8 to 32 bits and asynchronous output with explicit feedback. Compatible devices can use it without a separate vendor installation. A matching vendor driver may take precedence if one is already installed or supplied through Windows Update.

These three names describe different parts of the audio path:

• The class driver connects Windows to compatible USB audio hardware. Seeing usbaudio2.sys identifies that driver; it does not identify the playback application’s audio mode.
• WASAPI is the Windows application interface. In shared mode, the audio engine mixes application streams; exclusive mode gives one application direct access to the endpoint. Either mode still needs a device driver underneath.
• ASIO is an application-facing driver interface often used by recording software. A manufacturer’s native ASIO support needs its ASIO driver. Steinberg’s documentation also describes a universal driver for devices without native ASIO, so an ASIO option alone does not establish which hardware driver is active.

Install vendor software when the device’s instructions require it for compatibility or native ASIO. Ordinary playback through a supported class-compliant device does not automatically need that extra layer; follow the requirements for the particular device and application.

If the available formats stop at 48 kHz when the DAC advertises higher rates, check its USB mode and the selected playback endpoint before replacing hardware. A UAC1 compatibility mode can reduce available formats, but 48 kHz is not a universal UAC1 ceiling: FiiO documents up to 24-bit/96 kHz in the K3’s USB Audio 1.0 mode. Its instructions require reconnecting or restarting the unit after switching modes. Compare the formats offered with the device manual; a single 48 kHz setting is not proof of fallback.

What the chip does not decide

The parts of a DAC product that change what you hear at a desk are mostly not the converter:

• Output stage noise and drive. Whether the jack hisses with sensitive in-ear monitors, and whether it swings enough voltage for 300 ohm headphones, is set by the amplifier stage. The arithmetic is in matching a DAC and amp to your headphones.
• Output impedance. A high value shifts the tonal balance of low-impedance headphones. It is a product property, never a chip property.
• Power supply. Bus power from a noisy laptop port causes whine that tracks screen activity. Moving conversion outside the case cures it, and that is a placement benefit, not a conversion-quality one.

The same logic holds for the converter inside an audio interface feeding powered monitors: the room and speakers are the weak link, which is why the sister guide on studio monitors versus hi-fi speakers spends its time on placement rather than chip names.

When a separate DAC is worth buying

Buy one when the built-in converter has an audible defect at the level you actually listen (hiss, interference, dropouts, an output that cannot drive your headphones) or when you need an input the current device lacks, such as optical from a television. Do not buy one because a sheet quotes 140 dB or 768 kHz; those describe a chip in a test fixture, not the box on your desk. The decision framework is in do you need a DAC and amp, and if the answer is yes, combo unit versus separates covers which form to buy.

Related across the network

• Best Filament for Beginners: What to Buy First — 3dfilamentguide.com
• Filament Density: Convert Spool Weight to Length — 3dfilamentguide.com
• How to Dry Filament: Temperatures by Material — 3dfilamentguide.com
• PLA vs PETG vs ABS: Which Filament to Use — 3dfilamentguide.com
• Adversarial Attack Libraries: ART, Foolbox, torchattacks — adversarialml.dev