--- title: "Delta-Sigma DACs: How Oversampling Conversion Works" description: "How a delta-sigma DAC trades bit depth for speed, what noise shaping actually does, and how it really differs from an R-2R ladder in practice." url: "https://theaudiostuff.com/glossary/delta-sigma/" type: "website" author: "Jakub Charkiewicz" --- > Markdown rendering of https://theaudiostuff.com/glossary/delta-sigma/ > Canonical URL to cite: https://theaudiostuff.com/glossary/delta-sigma/ > Generated from the published page. Text and links only; see the HTML for layout. Breadcrumb: [Home](https://theaudiostuff.com/) > [Glossary](https://theaudiostuff.com/glossary/) > Delta-Sigma Audio glossary # Delta-Sigma DACs: How Oversampling Conversion Works A delta-sigma DAC converts audio using very few bits at a very high rate, then uses noise shaping to push the resulting quantisation error up out of the audible band where a filter removes it. It is the dominant architecture in modern audio because it sidesteps the component-matching problem that limits ladder DACs. ## The problem it solves To convert a digital sample to a voltage, you need a way to produce precise, distinct levels. The direct approach is a ladder of resistors, one per bit, each weighted exactly twice its neighbour. That works, and it runs into physics. A 16-bit ladder needs resistors matched to better than one part in 65,536; 24-bit needs one part in 16.7 million. Real resistors have tolerances, and they drift with temperature and age. Building and holding that precision is the central difficulty of ladder conversion, and it is expensive. [Delta-sigma](https://theaudiostuff.com/glossary/#delta-sigma) refuses the problem instead of solving it. ## Trading resolution for speed A delta-sigma converter uses very few levels, sometimes just one bit, but switches between them millions of times per second. The output is a dense pulse stream whose *average* over a short window represents the analogue value. If a single-bit output can only be at full positive or full negative, then a value halfway between is represented by spending half the time at each. That is the entire trick: [resolution](https://theaudiostuff.com/glossary/#resolution) in amplitude is traded for resolution in time, and time is something modern silicon is very good at. Because there are only one or two levels to produce, there is nothing to match. The linearity problem that dominates ladder design disappears. ## Noise shaping, which is the actual clever part Representing a smooth signal with a coarse pulse stream creates a large quantisation error. Left alone it would be audible noise. The modulator's feedback loop is arranged so that this error is not spread evenly across frequency. It is *shaped*: pushed up out of the audio band and concentrated at high frequencies, far above anything you can hear. A low-pass filter at the output then removes it. The result is a very low [noise floor](https://theaudiostuff.com/glossary/#noise-floor) where it matters and a great deal of ultrasonic noise where it does not. This is why delta-sigma [DACs](https://theaudiostuff.com/glossary/#dac) measure extremely well in the audio band, and also why their output filtering is not an afterthought: the filter is load-bearing, not a finishing touch. ## How it actually differs from R-2R The honest version, which is less dramatic than the debate suggests. **Measured performance.** Modern delta-sigma implementations generally achieve lower distortion and a lower noise floor than ladder designs, because they avoid the matching problem entirely. On measurements alone, this is not close. **Filtering.** Delta-sigma requires aggressive digital filtering and ultrasonic noise removal. Ladder DACs, especially non-oversampling ones, can use far gentler filtering. Filter behaviour, particularly [pre-ringing](https://theaudiostuff.com/glossary/#pre-ringing) on transients, is a real audible difference, and much of what listeners attribute to the conversion architecture is actually the filter in front of it. **Error character.** A ladder DAC's errors come from component mismatch, so they correlate with the signal level. A delta-sigma DAC's errors are shaped and largely pushed out of band. Whether one type of residual is preferable to the other is a listening question, not a measurement one. The practical conclusion: architecture is a weaker predictor of how a DAC sounds than its filter design, its analogue [output stage](https://theaudiostuff.com/glossary/#output-stage), and its power supply. Buy the converter, not the topology. ## Where you will meet it Nearly everywhere. Essentially every phone, laptop, streaming device and mainstream DAC chip uses delta-sigma conversion, because it is cheap, small, power-efficient and measures superbly. Ladder conversion survives as a deliberate, more expensive choice at the high end. ## Reviewed gear that shows this 1 review in the catalogue. Every score is anchored to the same published reference list. - [Fosi Audio ZD3 7.1/10 ยท $150](https://theaudiostuff.com/reviews/fosi-audio-zd3/) ## Keep reading - [Best DACs](https://theaudiostuff.com/guides/best-dacs/) - [Full audio glossary](https://theaudiostuff.com/glossary/) - [How we test and score](https://theaudiostuff.com/about/#methodology-title)