Delta-Sigma DAC: Low-Noise, High-Resolution Output
← Back to:Digital-to-Analog Converters (DACs)
Delta-sigma DACs win when the priority is ultra-low in-band noise and high resolution within a narrow bandwidth. The real success factor is proving the full chain—OSR/filter, clock, reference, driver, and reconstruction filter—with a repeatable test basis so shaped noise and tones stay out of the band.
What this page solves
Delta-sigma (ΔΣ) DACs are selected for one primary reason: very low in-band noise and high effective resolution when the signal bandwidth is moderate. This page clarifies when that advantage holds, what it costs (bandwidth and latency), and what must be designed around the DAC (clock quality, reconstruction filtering, and verification) so the expected noise floor is achieved on the bench.
Audio / Hi-Fi / Pro audio
Constraint: moderate bandwidth with demanding noise floor and dynamic range. Why ΔΣ: oversampling and noise shaping push quantization noise out of band, enabling an ultra-low in-band floor. Cost: digital filter group delay and clock purity become first-order design items.
Precision DC / low-frequency waveforms
Constraint: low-frequency resolution and repeatability matter more than wide bandwidth. Why ΔΣ: low in-band noise supports fine setpoints and small signal steps when measured under consistent bandwidth/weighting. Cost: reference/driver noise, drift, and filter stability can dominate if not budgeted and verified.
Slow biasing / trim / calibration loops
Constraint: slow updates, stable outputs, and low broadband noise are required. Why ΔΣ: ΔΣ architectures keep in-band noise low and can offer flexible digital filtering modes. Cost: output stage stability with load capacitance and reconstruction filters must be validated.
Ultra-wideband or extreme update rate
ΔΣ trades bandwidth for in-band noise; wideband synthesis often fits better with current-steering / RF DAC paths. Jump: CS-DAC / RF DAC pages
Ultra-low latency requirements
ΔΣ digital filtering introduces group delay; hard real-time phase alignment may require different architectures or sync strategies. Jump: Interfaces & Synchronization
Direct-RF / DUC-heavy transmit chains
Wideband RF output purity is dominated by high-speed switching behavior and clocking; ΔΣ is not the default choice there. Jump: RF DAC / Hybrid ΣΔ+CS
Scope boundary: This page goes deep on OSR/filters, noise shaping, idle tones, clock purity, reconstruction filtering, and verification. It does not provide a broad cross-architecture DAC comparison; related topics are referenced only as jump points.
Definition and key specs (use the same measurement basis)
A delta-sigma (ΔΣ) DAC converts digital samples to analog by generating a high-rate 1-bit or multi-bit stream using oversampling and noise shaping. The shaped out-of-band noise is then removed by the output chain (reconstruction filtering), resulting in a very low in-band noise floor.
Important: specs only compare cleanly under the same bandwidth, weighting, and filter mode. Many “better/worse” conclusions are caused by different FFT settings, different integration bandwidth, or different digital filter profiles.
Noise density (dBFS/Hz or nV/√Hz)
What it is: the noise floor expressed per unit bandwidth (a spectral density), typically measured over a region that avoids large tones.
Conditions to confirm: FFT length, window type, averaging, output amplitude, digital filter mode, and the frequency region used for the estimate.
Why it matters: this is the most direct indicator of how “quiet” the in-band output can be before integration over bandwidth.
Common misread: comparing noise density numbers from different filter modes or different FFT setups. A longer FFT (narrower bins) can visually lower the displayed floor without changing integrated noise.
Integrated noise (RMS over bandwidth)
What it is: the total noise power integrated over a stated bandwidth, reported as RMS voltage or RMS dBFS.
Conditions to confirm: integration limits (start/stop frequency), any weighting (A-weighting or none), and whether tones/spurs are excluded or included.
Why it matters: this is the number that determines real output resolution for a given band. Doubling bandwidth generally increases integrated noise unless the density is simultaneously lower.
Common misread: comparing “RMS noise” across different bandwidths. If bandwidth is not identical, the comparison is not meaningful without a density model.
Dynamic range / SNR / THD+N
What it is: a ratio between a reference signal level and noise (SNR/DR), sometimes including distortion (THD+N).
Conditions to confirm: signal amplitude, test tone frequency, analysis bandwidth, weighting, notch usage, and whether harmonics/spurs are counted as noise.
Why it matters: it captures “how much clean signal” is available before the floor (and distortion products) become limiting. For ΔΣ DACs, the underlying band definition and filter mode often dominate the final number.
Common misread: mixing weighted and unweighted results, or comparing THD+N when one setup excludes idle tones/spurs and another includes them.
Passband ripple / stopband attenuation / group delay
What they are: digital-filter (and sometimes analog-chain) traits that shape amplitude and phase in-band, and determine how much out-of-band shaped noise is rejected.
Conditions to confirm: which filter profile is active (fast/slow, linear-phase/minimum-phase), and whether the stated group delay is constant or varies across the passband.
Why they matter: these parameters set reconstruction-filter requirements, multi-channel phase alignment, and closed-loop stability margins.
Common misread: focusing only on “noise numbers” while ignoring group delay and stopband. In practice, stopband and delay determine whether the low floor remains usable in the target system.
Architecture inside a ΔΣ DAC
A delta-sigma DAC achieves its low in-band noise by converting samples into a high-rate stream using a feedback modulator. The architecture choices that most affect real performance are quantizer resolution (1-bit vs multi-bit) and loop order (how aggressively noise is shaped).
1-bit modulator path
- Linearity advantage: a 1-bit element avoids multi-level weight mismatch, so element matching is not the dominant error source.
- Noise cost: larger quantization steps typically require higher OSR (and/or stronger filtering) to reach the same in-band floor.
- System implication: internal rate, clock quality, and out-of-band suppression become more demanding to preserve the low in-band noise.
Multi-bit modulator path
- Noise advantage: smaller quantization steps reduce quantization noise so the same target can often be met with lower OSR or gentler filtering.
- Engineering trade: multi-bit DAC elements introduce mismatch and code-dependent behavior that can create spurs if unmanaged.
- What to confirm: whether DEM or calibration features exist and how they change the spectrum and verification requirements (details belong to matching/calibration pages).
Loop order is not “free performance.” Higher order can shape noise more steeply, but it also tightens stability margins and increases the risk of idle tones and limit cycles. Architecture choices must be paired with a verification plan (DC, small-signal, and configuration sweeps).
OSR and interpolation filtering
Oversampling ratio (OSR) and interpolation filtering are the main knobs that turn a ΔΣ DAC into a practical low-noise source. OSR pushes images away and can reduce in-band noise, while interpolation filter profiles define the usable passband, reject out-of-band content, and set the group delay seen by the system.
What higher OSR changes
- In-band noise: the target noise floor becomes easier to reach when the system can keep shaped noise out of the passband.
- Images: interpolation moves images farther from the passband so reconstruction filtering can be gentler.
- Analog burden: analog filters can focus on suppressing shaped out-of-band noise instead of aggressively policing images close to band.
What it costs
- Processing / switching burden: higher internal rates stress clocking and increase sensitivity to coupling and power integrity.
- Group delay: stronger stopband and linear-phase profiles often increase delay, which is a hard constraint in closed-loop or phase-aligned systems.
- Measurement basis: different filter modes change what “noise” means unless bandwidth, weighting, and setup are kept consistent.
The three interpolation filter knobs: passband ripple (amplitude flatness), stopband attenuation (leakage control), and group delay (timing/phase constraint).
Noise budgeting (turn “noise” into a budget)
A ΔΣ DAC can deliver an excellent in-band noise floor, but real systems often measure worse than the datasheet because non-core noise sources become dominant. A practical budget assigns ownership to each noise class and ties it to a repeatable measurement method; otherwise the “budget” cannot be verified.
Budget basis: define target bandwidth, weighting, digital filter mode, output amplitude, and load. Integrated noise only compares cleanly when the basis matches.
1) Quantization noise (shaped out-of-band)
- Ownership: ΔΣ modulator + digital filter profile.
- Coupling risk: if stopband rejection or reconstruction filtering is insufficient, shaped noise can leak back into what is integrated as “in-band”.
- Verification method: FFT + fixed integration bandwidth with identical window/averaging and identical filter mode across comparisons.
2) Reference noise (gain modulation)
- Ownership: reference source, reference filtering, and routing.
- Coupling risk: reference noise directly modulates output amplitude and can dominate low-frequency performance.
- Verification method: low-frequency noise test (e.g., 0.1–10 Hz or 1–100 Hz) plus a controlled A/B change in reference filtering.
3) Clock phase noise / jitter (spectral spreading)
- Ownership: clock source, distribution, and jitter-cleaning choices.
- Coupling risk: jitter can spread tones and raise skirts; it can also alter how noise is measured depending on bandwidth and stimulus conditions.
- Verification method: clock A/B comparison or controlled phase-noise / jitter-stress (injection) to confirm sensitivity and dominance.
4) Output driver / load noise (noise gain + stability)
- Ownership: output buffer, reconstruction filter, and load network.
- Coupling risk: noise gain and stability under capacitive loads can raise the floor or create spurs/ringing that mask ΔΣ benefits.
- Verification method: load-step / capacitive-load sweep and driver A/B validation while holding the FFT basis constant.
Rule: every noise line item must have a test. If a budget entry cannot be verified by a defined measurement setup, it cannot be used to justify a design decision.
Idle tones and limit cycles (the most common ΔΣ failure)
Idle tones are deterministic spurs that can appear when a ΔΣ DAC operates near DC or with very small signals. They are not random noise: a repeating internal pattern can lock into a periodic limit cycle, creating discrete spectral lines that show up as FFT spikes, audible “whine”, or low-frequency ripple.
Trigger conditions
- Small input: near-DC or very low-level tones.
- Repeating pattern: fixed code sequences that do not sufficiently decorrelate.
- Internal periodicity: loop dynamics settle into a stable repeating cycle.
How it appears
- FFT: discrete spurs (often stable in frequency and amplitude under fixed settings).
- Audio: sharp tonal components that stand out from broadband noise.
- Time domain: ripple on “DC” outputs or small waveforms, sometimes persistent after a step.
Fix strategy must match the mechanism: averaging more does not remove a deterministic spur. Mitigation must break the repeating pattern or change the modulator behavior, then confirm with sweeps.
Mitigation layers (practical order)
- System-level: add dither / controlled noise to decorrelate patterns; use small jitter or noise injection when appropriate.
- Device-level: confirm modulator modes, multi-bit options, and DEM features that alter tone behavior (details belong to matching/calibration topics).
- Verification-level: run DC + small-tone sweeps across OSR/filter modes and compare with identical analysis basis.
Clocking and phase noise
A ΔΣ DAC is not “jitter-insensitive.” The impact often shows up differently than in Nyquist DACs: clock phase noise and threshold uncertainty can raise skirts, lift the integrated noise floor, or create spurs. Because ΔΣ architectures can achieve a very low in-band floor, the clock chain frequently becomes the limiting system bottleneck.
What to watch (observables)
- Skirt: near-tone noise rises around a carrier and is sensitive to phase noise.
- Noise floor: integrated in-band noise increases even when harmonic distortion does not change dramatically.
- Spurs: discrete lines can appear from PLL behavior, threshold coupling, or supply-induced modulation.
Hidden couplings that break “clean clocks”
- Supply noise → PLL/cleaner: power rail noise can translate into phase noise and spurs if isolation is weak.
- Ground bounce → clock threshold: return-path disturbance changes the effective switching threshold and behaves like jitter.
Verification rule: hold the analysis basis constant (FFT length, window, averaging, integration bandwidth, filter mode, output level, load) and change only one clock-chain variable at a time.
Design actions (high leverage)
- Clock-tree isolation: keep the clock distribution and its returns away from high di/dt digital switching regions.
- Power-to-clock containment: isolate PLL/cleaner rails; prevent shared impedance with driver and digital logic.
- Threshold integrity: control clock amplitude, termination, and return paths so ground bounce cannot modulate the input threshold.
- Cleaner when needed: use jitter cleaning when phase noise or spurs are proven dominant by A/B measurements.
Output driver and reconstruction filter co-design
The final analog chain determines the delivered output. Reconstruction filtering must suppress out-of-band shaped noise and images while preserving in-band flatness and acceptable delay. The output driver must remain stable under the filter network and load; marginal stability can appear as ringing, worse THD, or a higher noise floor.
Engineering rule (do it in this order)
- Set required out-of-band attenuation: define how many dB of shaped-noise and image suppression are needed for the system.
- Select a topology: choose passive/active LPF order and cutoff to meet stopband needs while controlling ripple and delay.
- Verify stability: validate phase margin behavior using load sweeps (including capacitive loads, cables, and instrument input capacitance).
Failure signatures (fast diagnosis)
- THD worsens after filtering: driver interaction or reduced stability margin under the filter network.
- Noise rises: noise gain or out-of-band leakage dominates the integrated measurement basis.
- Ringing/overshoot: capacitive load or cable/instrument input interacts with the driver loop.
Verification checklist: keep FFT basis fixed, then sweep load (R/C/cable), amplitude (small to large), and LPF configuration (cutoff/order) while logging ringing, THD, and noise rise together.
Layout, grounding, and supplies
ΔΣ DAC performance is often limited by board-level noise rather than by the core converter. The lowest in-band noise requires return-path control, rail prioritization, and thermal awareness. The most sensitive blocks are the reference domain, the clock domain, and the output driver/filter domain; their return currents must not share uncontrolled impedance with fast digital I/O.
Partition priorities
- REF domain first: isolate reference routing and its decoupling loop from switching currents.
- CLK domain second: protect clock threshold integrity with clean returns and controlled termination.
- OUT driver + LPF: keep high dv/dt and load currents local to the driver/filter return loop.
- IO last: route digital I/O so its return current cannot cross REF/CLK areas.
Return-path rule
A “good ground” is a predictable return path. If digital or driver return currents share impedance with the reference or clock threshold region, ground bounce can modulate the threshold or gain and appear as spurs, skirts, or a higher noise floor.
Rail priorities (different goals)
- Reference rail: rail noise becomes gain modulation; prioritize isolation and low-impedance decoupling.
- Clock / PLL rail: rail noise becomes phase noise or spurs; prioritize isolation from driver/IO rails.
- Driver rail: transient currents create return disturbance; prioritize local loops and shared-impedance avoidance.
Thermal and drift
Low-frequency and precision outputs are sensitive to temperature gradients. Keep heat sources (driver, regulators, PLL) away from the reference and output path, and treat thermal gradients as a slow noise source to be verified.
Production test and verification plan
A repeatable test plan prevents false conclusions. The same DAC can report different noise and dynamic range values if FFT basis, bandwidth, weighting, or filter mode changes. A production-ready SOP defines conditions, analysis settings, and stimulus and load for each test category.
Reporting requirement: sample rate, record length, window, averaging, integration bandwidth, weighting, filter mode/OSR, output level, load, and temperature.
1) Noise floor (integrated)
- FFT settings: record length • window • averaging.
- Basis: integration BW • weighting • filter mode/OSR.
- Stimulus: silence / fixed code state (record it).
- Handling: keep out-of-band integration rules consistent across runs.
- Conditions: load and temperature fixed or explicitly swept.
2) Tones / spurs (sweeps)
- Stimulus: DC + small tone + mid-level tone.
- Sweep: frequency sweep and amplitude sweep.
- Mode compare: OSR / filter mode A/B.
- Clock A/B: source/cleaner comparison when skirts or spurs dominate.
- Logging: spur height, skirt level, and floor together.
3) Linearity / step (time-domain)
- Large step: full-scale step response and settling behavior.
- Major carry zones: step around key code boundaries (record conditions).
- Load sweep: R/C/cable and instrument input capacitance.
- Temperature: at least two points to expose drift and stability margins.
- Outputs: ringing • THD change • noise rise correlation.
Applications (delta-sigma–strong fit only)
Delta-sigma DACs shine when the priority is very low in-band noise and the system can accept narrower bandwidth and filter / group delay. Each application pattern below is written in a fixed structure: Constraints → Key options → Verification.
Hi-Fi / Pro audio
dominant risk: jitterConstraints
Near-tone skirts and discrete spurs are audible and can dominate perceived “black background.” Filter ripple and group delay must match the audio chain requirements.
Key options
- Digital filter modes: passband ripple / stopband / group delay.
- Clock guidance: allowed clock range and jitter sensitivity notes.
- Output chain: driver + LPF stability under real loads.
Verification
- Hold FFT basis fixed; do clock A/B and log skirt + floor + spurs together.
- Compare filter modes with identical bandwidth/weighting conditions.
Representative parts (examples)
ESS ES9038PRO · Cirrus Logic CS43198 · TI PCM1609A · TI PCM175x-Q1
Precision DC / low-frequency waveform
dominant risk: driftConstraints
The real limiter is often 0.1–10 Hz noise, reference stability, and thermal gradients rather than bandwidth. Shaped out-of-band noise must not leak through the analog chain.
Key options
- Reference chain: noise / PSRR / thermal drift guidance.
- Filter mode / OSR: document passband/stopband/group delay conditions.
- Layout & thermal: keep heat sources away from REF and output path.
Verification
- Measure low-frequency noise and drift under realistic thermal/airflow conditions.
- Confirm OOB suppression with spectrum plots using a fixed integration basis.
Representative parts (examples)
ST RHRDAC1612 · TI DAC1221
Bias / setpoint DAC
dominant risk: stabilityConstraints
The delivered value must remain stable during power-up, load changes, and long-cable or capacitive loads. Driver interaction can raise noise and create ringing.
Key options
- Output form: voltage/current/differential and load-drive guidance.
- Update strategy: soft start and step management to reduce disturbance.
- Driver + LPF: stability notes and recommended operating region.
Verification
- Load sweep (R/C/cable) + step response; log ringing + noise rise + THD change.
- Power-up sequence test with the real load attached.
Representative parts (examples)
TI DAC1221 · Analog Devices / Maxim DS3911
Instrumentation stimulus
dominant risk: tonesConstraints
Discrete spurs, idle tones, and mode-dependent artifacts can corrupt the DUT and mask real behavior. The stimulus must remain clean across operating points.
Key options
- Idle-tone mitigation: dither / config modes (must be documented).
- Stopband needs: prevent OOB leakage into the measurement band.
- Clock chain: verify skirt/spur sensitivity with A/B tests.
Verification
- DC + small-tone sweep (frequency and amplitude) across filter/OSR modes.
- Record spur height and locations using a fixed FFT basis.
Representative parts (examples)
ESS ES9038PRO · Cirrus Logic CS43198
Not a fit: ultra-wideband, ultra-low latency, or direct-RF output requirements generally belong to other DAC architectures.
IC selection logic (fields → risks → proof)
A correct selection process starts from requirements and forces each datasheet field to map to a specific risk and a repeatable verification test. This avoids “spec chasing” and prevents mismatched measurement setups from producing misleading comparisons.
Required datasheet fields (must be asked and recorded)
OSR / digital filter options
passband ripple · stopband attenuation · group delay
Idle-tone mitigation
dither · DEM · configurable modulator modes
Clock requirements
allowed clock range · jitter/phase-noise guidance · sensitivity notes
Output type & load drive
voltage / current / differential · recommended load · stability guidance
Reference requirements
noise · PSRR · thermal drift notes
Noise spec conditions
bandwidth · weighting · temperature · filter mode
Risk mapping (field → risk → proof)
| Datasheet field | Risk | Verification test (evidence) |
|---|---|---|
| Dither / idle-tone features | Idle-tone spurs in DC and tiny signals | DC + small-tone FFT sweep; mode A/B; fixed analysis basis |
| Group delay (filter modes) | Control-loop phase / latency limits | Step / latency measurement; alignment check for multi-channel |
| Stopband attenuation | Out-of-band residue leaks into the system band | Spectrum with defined BW; verify LPF chain attenuation |
| Clock range & guidance | Skirts / floor / spurs dominated by clock chain | Clock A/B (source/cleaner); log skirt + floor + spurs |
| Output drive / stability notes | Ringing, THD degradation, noise rise under real loads | Load sweep (R/C/cable/instrument C) + step response correlation |
| Noise spec conditions | Uncomparable DR/noise results across setups | SOP report: fs, record length, window, avg, BW, weighting, mode, load, temp |
Vendor inquiry template (copy-paste)
Application: Bandwidth (signal BW / integration BW / weighting): Noise target (density or integrated RMS) + conditions: Allowed delay / group delay limit: Output type (voltage/current/diff) + load (R/C/cable) + stability constraints: Clock plan (source / cleaner / frequency range) + jitter/PN expectation: Required OSR / filter modes (passband ripple / stopband / group delay): Idle-tone concerns (DC + small-signal use cases) and mitigation needs: Temperature range + drift expectations (0.1–10 Hz / airflow sensitivity): Verification method (FFT basis: fs, length, window, avg; sweeps: amp/freq; load sweep; temp points):
Request a Quote
FAQ (Delta-Sigma DAC)
Each answer uses a fixed, data-oriented structure: When → Why → Check → Fix. Keep the same measurement basis when comparing results: fs, record length, window, averaging, integration BW, weighting, filter/OSR mode, output level, load, temperature.
