Who it’s for

What this page delivers

• How to read en/in, 1/f corner, and integrated noise without mismatched conditions.
• How to compare parts using the same bandwidth / ENBW and avoid “single-number” traps.
• How to build a practical noise budget including source resistance and resistor network noise.
• Where distortion comes from in precision chains, and which constraints move THD/IMD in real circuits.
• Reusable low-noise topologies and a short set of measurements to validate results.

Out of scope (handled by sibling pages)

• Zero-drift / chopper artifacts (0.1–10 Hz focus, ripple, modulation side-effects) → use the Low-Offset / Zero-Drift page.
• TIA stability with sensor capacitance (Cpd, noise-gain compensation) → use the Low-Drift TIA page.
• ADC-driver / FDA details (sampling kickback, VOCM control, anti-alias filter design) → use the ADC Driver / FDA pages.

Diagram note: the “Out of scope” tags are boundary reminders only; details belong to the dedicated sibling pages.

The four noise items to interpret first

Comparison rules (avoid datasheet traps)

• Normalize frequency: compare en/in at the same frequency region (flat vs 1/f), not a mix of curve segments.
• Normalize bandwidth: compare integrated noise using the same integration limits and the same ENBW assumptions.
• Normalize gain setting: noise gain and closed-loop gain can change the effective bandwidth and stability.
• Normalize impedance: source impedance and feedback resistor values decide whether in and resistor noise dominate.

Practical steps to get RMS noise without heavy math

1. Define the band: use the real signal bandwidth set by filtering, sampling, or application constraints.
2. Split the spectrum: identify whether the band overlaps the 1/f region or is mostly in the flat region.
3. Account for filtering: convert “-3 dB bandwidth” to the correct ENBW effect of the filter shape.
4. Integrate consistently: compare parts using RMS noise over the same band and the same ENBW assumption.

Next key hook: source impedance decides whether en or in · Z dominates (covered in the next section).

Quick checks (engineering sanity)

• Confirm whether the target is low-frequency behavior or wideband noise before comparing “best numbers”.
• Do not use instrument bandwidth as the system band; define the band that the application actually uses.
• Noise curves without clear conditions (frequency, gain, filter) are not comparable across parts.
• In inverting stages, resistor noise and in-related terms often dominate even with a very low en op amp.
• Measurement setups can inject EMI that looks like noise; verify grounding and probing before blaming the op amp.

Diagram note: the shaded area represents a consistent integration band; comparisons should use the same bandwidth/ENBW assumptions.

Minimal source-impedance model (what matters for noise)

• R_S: sensor resistance, series protection/filter resistors, and any wiring resistance that sits in series with the input.
• C_S: cable/sensor capacitance, input capacitance, and layout parasitics that make impedance frequency-dependent.
• Equivalent source: use a Thevenin-style view so the input sees a clear Z(f) over the band of interest.

The decision rule (same units, fast comparison)

1. Define the band: use the real application bandwidth (filtering / sampling / required signal band).
2. Estimate |Z| in-band: at low frequency it may look like R_S; at higher frequency capacitance pulls |Z| down.
3. Convert current noise: treat in · |Z| as an equivalent input-voltage-noise term (same unit family as en).
4. Compare: if en is larger, voltage noise dominates; if in · |Z| is larger, current-noise conversion dominates.

Practical outcome: low-impedance sources usually benefit most from lower en; higher-impedance sources often require lower in and careful resistor choices.

BJT-input vs FET-input (trend + risk mapping)

Quick checks (avoid common misreads)

• Source impedance is not only the sensor: series protection/filter resistors can silently raise R_S.
• Capacitance makes Z frequency-dependent; evaluate |Z| in-band, not at DC only.
• If noise is higher than expected, verify whether the design drifted into current-noise dominance through larger resistors.
• Inverting structures often shift the “effective impedance” seen by current noise; resistor choices are part of the noise match.

Diagram note: evaluate the dominance using the in-band impedance Z(f), not a single DC resistance.

The minimal contributor set (input-referred view)

• Source resistance: thermal noise of the effective source R seen by the input.
• Op amp en: voltage noise density in the relevant (flat / 1/f) band.
• Op amp in · |Z|: current noise converted through the in-band impedance.
• Feedback/network resistors: thermal noise of R_IN/R_F and any series resistors.
• Next stage (referred): downstream noise divided by the gain ahead of it.

Bandwidth and filtering (why more BW means more RMS noise)

• Noise density → RMS: RMS noise increases as the integration band widens.
• Use the system band: define the band from the application filter/sampling limits, not test-instrument bandwidth.
• Match ENBW: comparisons must use the same effective-noise-bandwidth assumption for filtering.

Budget comparisons only make sense when every term is evaluated in the same band.

Multi-stage allocation (where gain placement helps)

• Front-end gain reduces referred noise from later stages, because downstream terms divide by earlier gain.
• Gain changes bandwidth and margins: verify noise band and stability after gain placement changes.
• Allocate targets: split an overall noise goal into per-stage budgets before selecting parts.

Quick checks (budget sanity)

• If source thermal noise is close to the target, changing op amps will not move the system floor much.
• Inverting designs often have resistor-network noise as a major contributor; budget the resistors early.
• If measured RMS noise is higher than expected, re-check the effective bandwidth (filtering, noise-gain, oscillation edge).
• Keep one reference point: all terms should be compared as input-referred in the same band.

Diagram note: the stack is an input-referred view; evaluate each term in the same bandwidth before changing parts.

Root causes (what actually creates distortion)

• Output-stage nonlinearity: rising load current, low-Ω loads, or capacitive drive push the output devices into nonlinear regions.
• Input-stage transconductance change: large input or common-mode movement alters gm and raises harmonic and intermodulation products.
• Insufficient loop gain: loop gain falls with frequency; less corrective action means a higher THD floor.
• Common-mode interaction: dynamic CMRR/PSRR and supply bounce can translate into spurs that look like “distortion”.

What moves THD/SFDR the most (in practice)

• Output headroom: THD often rises quickly as swing approaches the rails or output current limits.
• Load type (R and C): resistive loading raises current demand; capacitive loading can add phase shift and stress the output stage.
• Frequency: higher frequency reduces loop gain, raising the distortion floor in-band.
• Closed-loop gain: gain and noise-gain choices affect loop-gain margin and the frequency where distortion starts to climb.
• Supply integrity: dynamic PSRR matters under fast load steps and ripple; poor decoupling can appear as spurs.

Typical “low-noise ≠ low-distortion” failures

Fast validation plan (minimum measurements)

• Check THD at two swings: a small swing and the real operating swing (headroom sensitivity).
• Check THD at two loads: light load vs the real R/C load (output-stage stress).
• Check THD/IMD at two frequencies within band: low vs high (loop-gain roll-off sensitivity).
• If spurs appear, verify supply decoupling and ground return before assuming “intrinsic” distortion.

Diagram note: distortion rises when loop-gain margin falls in-band or when output headroom/load stress pushes the output stage nonlinear.

Template cards (best use + watch-outs)

Diagram note: labels indicate typical dominant contributors; the actual limiter should be confirmed with the input-referred budget.

Resistor and network floors (value choices matter)

• Thermal noise is unavoidable: higher resistance raises the resistor-noise floor; “high-R to save power” often costs noise.
• Inverting networks are sensitive: R_IN/R_F are direct contributors; large values can dominate the input-referred budget.
• Series protection/filter resistors: small “helper” resistors can silently raise the effective source resistance seen by the input.

Layout and return paths (where “extra noise” is injected)

• Keep the sensitive loop small: input node and feedback loop area should be minimal and tightly routed.
• Feedback network placement: resistors should sit close to the op-amp pins to avoid picking up fields and return-current noise.
• Return-path discipline: high-current returns (loads, switching rails, digital edges) must not cross the analog sensitive region.
• Decoupling as a loop: the capacitor-to-pin-to-ground loop must be short and local, not “somewhere on the rail”.
• Leakage/guarding (when high-Z nodes exist): contamination and humidity can create slow “noise-looking” movement.

EMI that looks like noise (principle-level, not device-specific)

• Rectification effects: RF pickup can mix with nonlinear junctions and create apparent low-frequency drift or raised 1/f-like behavior.
• Most common entry points: input traces, feedback loop, cable shields, and ground-return discontinuities.
• First checks: probe method, grounding/shielding, and return routing before assuming the amplifier itself is the limit.

Minimal debug order (5 steps)

1. Confirm the effective bandwidth/ENBW used for the measurement.
2. Check whether resistor values (source + network) set the thermal-noise floor.
3. Inspect the sensitive input and feedback-loop area and routing.
4. Verify return paths: ensure high-current returns do not cross sensitive analog zones.
5. Investigate EMI-looking noise: grounding, shielding, probe method, and cable entry.

Diagram note: minimize the sensitive loop area, keep feedback resistors close, and route high-current returns away from the sensitive zone.

Typical triggers (what pushes the loop unstable)

• Direct C-load drive: output connected to large capacitors (filters, hold caps, isolation caps).
• Long lines/cables: distributed capacitance plus inductance and reflections.
• Remote capacitors: “far” capacitors behave differently than local ones and can add phase shift.

Recognize edge-of-stability (noise and distortion symptoms)

• Raised “noise floor”: unexpected broadband lift or narrow peaks/spurs on a spectrum view.
• THD/IMD degradation: distortion worsens without obvious noise-spec changes.
• Step response changes with load: overshoot/ringing that varies strongly with C-load is a warning sign.

Common fixes (isolate the load with minimal trade-offs)

Fast verification (minimum tests)

• Step response: check overshoot and ringing with a small-signal step.
• C-load sweep: compare response and spectrum across multiple capacitance values.
• Long-line check: verify behavior with representative cable length and termination assumptions.
• After any fix: re-check noise band and distortion at the real operating swing and load.

Diagram note: isolate the capacitive load first, then re-verify noise band and distortion under real operating conditions.

What the buffer must guarantee

• Low output noise: reference noise maps directly into system accuracy and stability.
• Fast settling: return to a tight error band after load steps, start-up, or reference drift.
• Low bias/leakage interactions: avoid errors through high-value networks and leakage paths.
• Dynamic load immunity: tolerate pulse-like charge demand without ringing or long recovery.

Common failure modes (symptom → likely cause)

• Noise looks higher than expected: bandwidth is wider than assumed, or the load/network adds extra noise.
• Slow start-up or long recovery: large output capacitance, current limiting, or overload recovery limits.
• Glitches under activity: pulse loading creates short charge demand and forces a transient error.
• Spurs near operation rate: repetitive loading modulates the reference node and appears as discrete tones.

Kickback modeled as a charge pulse load

A practical way to reason about kickback is to treat it as a short charge pulse drawn from the reference node. The buffer must supply transient current, keep the glitch small, and recover quickly without entering ringing or edge-of-stability.

Key checks (design + selection checklist)

• Noise → accuracy mapping: confirm how reference noise converts into output error for the system.
• Settling to an error band: verify settling time to the allowable reference error, not only “stable”.
• Start-up + overload recovery: verify behavior after power-up and after forced disturbances.
• Static + dynamic loads: include divider/reference-pin loads and pulse-like charge demand.

Minimum verification plan

• Measure output noise with the real load and the real bandwidth.
• Apply load steps and verify settling to the allowed error band.
• Emulate a charge pulse load and record glitch height and recovery time.
• Test start-up and overload recovery by forcing a temporary disturbance on the node.

Diagram note: treat kickback as a charge pulse load; verify glitch height and recovery under representative operating conditions.

What makes distortion worse (priority order)

• Swing and headroom: distortion rises quickly when operation approaches output limits or rails.
• Load stress: low impedance, capacitive loads, and cable effects increase output-stage nonlinearity.
• Frequency and loop gain: less loop gain at higher frequency raises the in-band distortion floor.
• Closed-loop gain: gain choices set bandwidth and loop-gain margin for the operating band.

IMD vs THD (why two-tone reveals problems earlier)

• IMD is more sensitive: it highlights nonlinear mixing under realistic signal combinations.
• Two-tone validation: verify IMD under real swing, real load, and real bandwidth rather than at an easy lab condition.
• Watch for spurs: discrete products near the band can dominate perceived performance even when THD looks fine.

Biasing and coupling (principles that prevent low-frequency surprises)

• Bias currents and source impedance: can create offsets and low-frequency movement if networks are high-value.
• Coupling capacitors: form a low-frequency corner with input impedance and can cause start-up transients.
• Supply margin: limited headroom shifts the operating point and can raise distortion earlier than expected.

Minimum verification plan (keep it realistic)

• Measure THD at small swing and at real operating swing (headroom sensitivity).
• Run a two-tone IMD test under real load and real gain settings.
• Sweep within-band frequency (low vs high) to expose loop-gain roll-off effects.
• Sweep load (light vs real) to capture output-stage stress behavior.

Diagram note: validate IMD/THD under real swing, real load, and real bandwidth; headroom and load stress dominate practical distortion.

A) Inquiry fields (ask with conditions so numbers are comparable)

B) Risk mapping (field → risk → minimum verification)

C) Quick shortlist rules (3 rules + buckets with part numbers)

Diagram note: keep each decision node tied to test conditions; shortlist is only valid if noise and distortion were compared under matching swing, load, and bandwidth.