Noise in Active Filters: en/in, 1/f Corner, In-Band SNR
← Back to:Active Filters & Signal Conditioning
Active-chain noise is not a single number—it is a budget defined by the model (en/in/1/f), the source impedance, and the exact bandwidth (ENBW). This page shows how to convert an SNR target into an in-band RMS noise limit, pick the right IC class, and verify results without measurement traps.
Source impedance is the pivot: when in·Rs dominates, when en dominates
Source impedance sets the dominant input-referred noise term. The same amplifier can look “quiet” with a low-Rs source and become “noisy” with a high-Rs source because current noise scales as in·Rs, while resistor thermal noise rises as √(Rs). A practical selection line is the crossover Rs*, where in·Rs = en.
A) Working input-referred model (engineering form)
Use a compact density model to rank dominant terms before full integration:
e_total ≈ √( en² + (in·Rs)² + 4kT·Rs + e_Rnet² )
- Rs should represent the in-band effective source impedance.
- e_Rnet includes thermal noise from gain/feedback resistors that couple through topology-dependent noise gain.
B) The crossover pivot: Rs* = en / in
Define the decision point where current-noise and voltage-noise contributions match:
in·Rs = en → Rs* = en / in
- Rs ≪ Rs*: prioritize lower en and manage resistor thermal noise.
- Rs ≫ Rs*: prioritize lower in, then verify leakage/board cleanliness.
- Rs ≈ Rs*: both matter; final selection must use in-band integrated noise over the intended bandwidth.
C) High Rs warning: “looks like noise” but is often leakage
With high source impedance, bias/leakage effects can masquerade as low-frequency noise and inflate the apparent floor. Separate real random noise from non-random artifacts early.
Quick check
Compare PSD/RMS with input shorted vs known Rs source; repeat after cleaning or humidity change.
Fix
Guard rings, shorter high-Z traces, lower resistor values where possible, clean/coat, and reduce leakage paths.
Noise gain and topology shaping: why the same op-amp behaves differently in Sallen-Key vs MFB
Topology changes the noise transfer shape. Noise follows noise gain and injection paths, not the signal gain label. In active filters, resistor thermal noise and amplifier noise can be weighted very differently by the feedback network, which changes the in-band integrated noise even if the same amplifier is used.
A) The trap: noise gain ≠ signal gain
- Signal gain describes how the desired input maps to output.
- Noise gain describes how internal noise sources are amplified by the feedback network.
- Filter networks can create frequency-dependent weighting, so the correct comparison is in-band integrated noise, not a single density value.
B) Injection paths: where the noise enters
- Amplifier en: appears as an input-referred voltage source shaped by noise gain.
- Amplifier in: converts to voltage through Rs and feedback impedance.
- Resistor noise: each resistor injects thermal noise and is weighted by the topology’s transfer paths.
C) Practical design actions (noise-first ordering)
- Identify noise-sensitive resistors (high-impedance nodes and strong noise-gain paths).
- Set resistor value levels to manage thermal noise while maintaining drive, loading, and stability.
- Apply tolerance/matching where the topology converts mismatch into peaking or gain/Q drift.
- Then select the amplifier based on the dominant term (low en vs low in) and stability margin with the chosen network.
D) Scope guardrails (keeps this page narrow)
- Detailed topology equations and sizing steps belong to the dedicated Sallen-Key and MFB pages.
- This section focuses only on noise injection, noise gain shaping, and which resistors usually matter most.
1/f corner: how to translate it into in-band RMS noise (and when chopper helps)
Low-frequency systems often fail not because the midband noise is high, but because the integration window reaches into the 1/f region. The correct comparison is always in-band RMS noise over [fL, fH], not a single noise density number.
A) Engineering meaning of the 1/f corner
- Corner frequency (fc): where the 1/f noise density equals the white-noise density.
- Practical rule: if fL is well above fc, 1/f contribution is usually negligible; if fL is at or below fc, 1/f can dominate the RMS budget.
- Key requirement: always define fL (HPF/servo/window) and fH (effective bandwidth) before evaluating low-frequency noise.
B) Two-region model (usable budgeting form)
Use a two-region approximation that connects 1/f to the white-noise floor via the corner:
White region: e(f) ≈ e_w
1/f region: e(f) ≈ e_w · √(fc / f)
This form is intended for consistent budgeting and comparison, not device-physics modeling.
C) Set the integration limits: fL and fH
fL (lower limit)
- Input high-pass or AC coupling cutoff
- DC servo / offset-servo effective high-pass
- Observation window (effective lowest frequency in the measurement)
fH (upper limit)
- Effective passband bandwidth after filtering
- Sampling chain constraints (anti-alias bandwidth)
- Verification bandwidth limits (FFT/measurement settings)
D) Practical RMS translation recipe (no calculus required)
- Take the white-noise density e_w from the flat region of the datasheet curve.
- Take the corner frequency fc (or estimate it from the PSD plot).
- Define [fL, fH] from the real signal chain (HPF/servo/window and effective bandwidth).
- Estimate white contribution with ENBW (or BW as a first pass): Vw_rms ≈ e_w · √(ENBW).
- Decide whether 1/f must be included:
- If fL ≥ 10·fc, 1/f is often negligible.
- If fL ≤ fc, 1/f can raise RMS materially and must be budgeted.
- Combine by RMS: Vtotal_rms ≈ √(Vw_rms² + V1f_rms²), then compare to the noise allowed by the target SNR.
E) Chopper / auto-zero: when it helps, when it backfires (verification-first)
Helps when
The RMS budget is dominated by 1/f and the observation window includes low frequencies.
Backfires when
Ripple/tones appear, spurs fold into band, or noise becomes non-stationary and degrades statistical stability.
Verification
Check PSD for tones, check time-domain for ripple/steps, and accept only if in-band RMS and spurs meet the system limit.
In-band integrated noise: bandwidth, filter order, and “where to integrate”
In-band RMS noise depends on the integration area. Two filters with the same cutoff frequency can produce different RMS noise because their equivalent noise bandwidth (ENBW) is not the same. Always integrate the power transfer shape, not a label.
A) What ENBW means (practical definition)
ENBW converts a real filter shape into an equivalent “brick-wall” bandwidth that preserves the same integrated noise power.
Vn_rms ≈ e_w · √(ENBW)
Use ENBW whenever the filter shape is non-flat or when comparing different orders/topologies.
B) When flat-BW approximation is OK vs not OK
BW approximation is OK
- Rough early budgeting
- Passband is nearly flat
- No strong peaking / high-Q behavior
ENBW is required
- Higher-order / steep roll-off filters
- Noticeable passband shaping or peaking
- Comparing topologies (noise transfer differs)
C) Filter order: noise benefit vs practical cost (noise view only)
- Higher order often reduces ENBW, which reduces in-band RMS noise.
- However, higher order can increase group delay and implementation burden, which increases verification and stability risk.
- Choose the minimum order that meets the noise target while respecting latency and robustness constraints.
D) Where to integrate (prevents measurement traps)
- Integrate over the effective passband defined by the chain, not by a single “fc” label.
- Keep the same [fL, fH] across options when comparing filters or amplifiers.
- Accept results only when spurs/tones are checked separately from RMS (a lower RMS with larger spurs is not a win).
From noise to SNR: convert a target SNR into an allowable noise budget
A usable noise budget starts from the target SNR and the signal RMS level, then back-solves the allowable in-band RMS noise. The result must be expressed in two consistent views—input-referred and output-referred—so gain does not silently break the math.
A) Unified SNR definition (RMS over the same band)
SNR(dB) = 20·log10( Vrms_signal / Vrms_noise )
- Vrms_signal and Vrms_noise must be defined over the same [fL, fH] and measurement window.
- Vrms_noise is in-band RMS (use ENBW when the filter shape is not brick-wall).
B) Back-solve allowable noise from a target SNR
Convert the target SNR into an allowable RMS noise limit directly:
Vrms_noise_allow = Vrms_signal / 10^(SNR/20)
Sanity checks
- +20 dB SNR → allowable noise ÷ 10
- +6 dB SNR → allowable noise ÷ ~2
C) Input-referred vs output-referred (avoid gain mistakes)
Input-referred view
Each stage noise is referred back to the input for direct comparison to the sensor RMS level and source impedance.
Output-referred view
Each stage noise is expressed at the output node to align with full-scale swing, headroom, and ADC-interface verification.
Conversion rules
- To refer noise to the input: divide by gain-to-that-node.
- To refer noise to the output: multiply by downstream gain.
- Budget comparisons are valid only if the same [fL, fH] is used throughout.
D) Stage allocation logic (not equal shares)
- Reserve a placeholder for fixed downstream noise (e.g., ADC interface / post-processing) so it cannot be “forgotten”.
- Allocate the majority of the remaining noise budget to the front-end (closest to the sensor), because it sets the reference.
- Use filter/PGA budgets to finish and protect, not to rescue a noisy front-end.
- Keep guardband for temperature, tolerance, and operating range (worst-case stability of the budget).
E) Verification hooks (turn a budget into measurable checks)
Node checks
- Measure in-band RMS at the front-end output and post-filter output with the defined [fL, fH].
- Compare input short vs known Rs / real sensor to separate intrinsic noise from source-related terms.
- Track tones/spurs separately from RMS (RMS alone can hide discrete artifacts).
Pass criteria
Each measured node RMS must remain below its allocated budget (including guardband), and no discrete spur may violate the system limit.
Practical design levers: what actually reduces noise (and what just moves it)
Noise improvement must be evaluated as in-band RMS under the same [fL, fH]. Some levers truly reduce RMS (reduce sources or reduce ENBW); others mainly redistribute noise or trade noise for bandwidth, headroom, distortion, stability, or power.
A) Lower Rs and resistor value level
Mechanism
Reduces in·Rs conversion and √(4kTR) thermal noise.
Side effects
Higher loading and current, possible stability and distortion changes due to heavier drive.
Verify
Compare in-band RMS with shorted input vs known Rs; confirm the reduction matches the budget expectation.
B) Select low en vs low in using Rs*
Mechanism
For Rs ≪ Rs*, voltage noise dominates; for Rs ≫ Rs*, current noise dominates.
Side effects
Low-en choices can increase current/power; low-in choices can become leakage-sensitive in high-Z layouts.
Verify
Confirm dominance with a PSD/RMS test across source impedance and temperature; reject solutions that improve RMS but create tones.
C) Control ENBW with filter shape/order
Mechanism
Reduces the integration area, reducing in-band RMS noise for the same noise density floor.
Side effects
Higher order typically increases group delay and implementation burden; verify robustness before committing.
Verify
Compare in-band RMS using the same [fL, fH]; do not compare by fc alone.
D) Gain split and PGA usage
Mechanism
Earlier gain can suppress downstream noise when referred to input; PGA aligns ranges across sensors and operating points.
Side effects
Higher front-end gain risks saturation, reduces headroom, and can increase distortion or transient settling time.
Verify
Validate large-signal headroom and post-switch settling; confirm that the input-referred noise budget improves across gain states.
E) Differential chain (noise view only)
Mechanism
Controls common-mode behavior and reduces coupling that appears as in-band noise or spurs in the measurement band.
Side effects
Higher part count and power; layout symmetry becomes a primary constraint for repeatability.
Verify
Compare single-ended vs differential configurations under identical bandwidth and loading; check both RMS and discrete artifacts.
F) Common traps: “looks quieter” but does not improve real SNR
- Over-filtering: RMS drops but response becomes too slow or delay becomes unacceptable.
- Servo-induced peaking: low-frequency “stability” can hide a peak that increases in-band RMS or creates spurs.
- Blind gain increase: reduces downstream-referred noise but causes saturation or distortion, reducing effective SNR.
Fast decision checklist
Confirm RMS improves under the same band, confirm spurs do not increase, and confirm headroom/latency constraints remain satisfied.
Layout & real-world “noise imposters”: leakage, microphonics, ground currents, EMI coupling
Many “noise” complaints are not intrinsic amplifier noise. Slow drift, touch sensitivity, and line-frequency hum often come from leakage, microphonics, ground-return voltage drops, or coupling into high-impedance nodes. The correct goal is still in-band RMS under the same [fL, fH] plus a check of discrete spurs (50/60 Hz and switching-related).
A) Fast symptom triage (RMS vs spurs vs drift)
RMS increases
True in-band noise may be higher, or the measurement band is contaminated by coupling. Confirm with a spur check.
Spurs increase
Line frequency (50/60 Hz) or switching tones indicate ground currents, supply ripple, or field coupling—not random noise.
Slow drift / jumps
Often leakage/contamination, thermoelectric offsets, or microphonics. Do not label it as “1/f noise” before isolating the cause.
B) Leakage & contamination (humidity, flux residue, surface paths)
- Signature: touch/hand proximity causes slow changes; humidity makes it worse; shorted input looks clean but high-Z source drifts.
- Why it happens: surface resistance drops, creating unintended DC/low-frequency paths at high-impedance nodes.
- Quick checks: compare shorted input vs known Rs; dry/heat vs humid conditions; clean vs unclean PCB (same setup).
- Most effective actions: guard ring around sensitive nodes, shortest input routing, controlled solder mask strategy, proper cleaning, and conformal coating when required.
C) Microphonics (cable motion, triboelectric effects, vibration)
- Signature: output changes correlate with tapping, vibration, or cable motion; behavior is setup-dependent and often repeatable by movement.
- Quick checks: fix the cable and connector (strain relief) and compare to free cable; shorten or swap the cable; isolate the fixture mechanically.
- Most effective actions: secure wiring near the input, use low-microphonic cable types, avoid floating high-Z nodes, and add an appropriate input RC only if the bandwidth budget allows.
D) Ground-return currents & supply ripple (50/60 Hz and switching pollution)
- Signature: 50/60 Hz hum or switching tones rise with load changes, digital activity, or DC/DC operating modes.
- Quick checks: change load/digital activity and observe spur amplitude; compare shield termination options (single-point vs alternate) without changing the band.
- Most effective actions: keep high di/dt return loops away from sensitive nodes, preserve return-path continuity, place decoupling/returns intentionally, and route differential pairs symmetrically.
E) Quick decision table (first checks only)
| Symptom | Likely imposter | First check |
|---|---|---|
| Slow drift after touch / humidity sensitivity | Leakage / contamination | Shorted input vs known Rs; clean/dry vs humid compare |
| Output changes when cable moves / taps | Microphonics | Fix cable & strain relief; shorten/swap cable and repeat |
| Strong 50/60 Hz spur | Ground return / shield path | Change load/digital activity; compare shield termination options |
| Switching tones / sidebands rise with power modes | Supply ripple / coupling | Hold band constant; vary power mode/load and track spur amplitude |
Measurement & verification: how to measure en/in and in-band RMS without fooling yourself
Noise measurements must separate random in-band RMS from discrete spurs. Use a repeatable band definition [fL, fH], keep settings constant when comparing options, and rely on two core fixtures: shorted input and known Rs.
A) Choose the metric: FFT vs time-domain RMS
FFT (spectrum)
Best for locating 50/60 Hz, switching tones, and sidebands that inflate “noise” but are not random.
Time-domain RMS
Best for direct comparison to the noise budget. Always compute RMS over the defined band and measurement window.
B) Band discipline (avoid alias and “settings drift”)
- Define [fL, fH] once and keep it identical across comparisons.
- Use anti-aliasing (a simple analog limit is acceptable) so out-of-band energy does not fold into the band.
- Keep processing consistent (window/averaging/integration method). Change only one variable at a time.
C) Fixture 1: shorted input (approach intrinsic noise)
- Purpose: minimize source-related terms so the baseline reflects intrinsic noise and internal networks.
- Rule: short near the input node with the shortest possible loop to avoid pickup.
- Output: record spectrum shape and in-band RMS as the reference baseline for budgeting.
D) Fixture 2: known Rs source (expose in·Rs and 4kTR)
- Purpose: reveal source-related contributions by testing multiple Rs values (low/mid/high).
- Decision: if RMS rises strongly with Rs, current-noise conversion dominates; if not, voltage noise or downstream noise dominates.
- Flag: if results depend on touch/humidity/motion, treat it as coupling (layout/fixture) before interpreting en/in.
E) Pass criteria (three checks, all required)
Spectrum shape
White region and low-frequency rise behave as expected, without unexplained peaks that indicate coupling or instability.
In-band RMS
Vrms_meas ≤ Vrms_budget (with guardband) under the defined [fL, fH].
50/60 Hz spur
Spur amplitude ≤ spur_limit (spur_limit comes from the system budget; treat “RMS pass but spur fail” as a fail).
Engineering Checklist: noise budgeting + layout review + production hooks
This checklist closes the loop from design to verification to production consistency. Keep one noise metric throughout: in-band RMS over the defined [fL, fH], plus discrete spur limits (50/60 Hz and switching-related).
A) Budget (minimum set)
Fill-in template (use one consistent band)
| Item | Required input | Rule | Pass criterion |
|---|---|---|---|
| Signal level | Vrms_signal = ____ V | Define at the same node (input-ref or output-ref) | Node definition is consistent across stages |
| Band | fL = ____ Hz, fH = ____ Hz | Use the same [fL, fH] for design + test | Measurement uses identical band + settings |
| ENBW | ENBW = ____ Hz (or “flat BW approx”) | Same fc can produce different ENBW by filter shape | Chosen approximation is documented |
| Target SNR | SNR_target = ____ dB | SNR(dB)=20·log10(Vrms_signal/Vrms_noise) | Budget uses Vrms, not Vpp |
| Allowable noise | Vrms_allow = ____ V | Vrms_allow = Vrms_signal / 10^(SNR/20) | Measured Vrms ≤ Vrms_allow (with guardband) |
| Source impedance | Rs = ____ Ω (note if f-dependent) | Check dominance: en vs in·Rs vs √(4kTRs) | Dominant term is identified + validated |
| Op-amp noise | en(f), in(f), 1/f corner + conditions | Use the datasheet test conditions as the baseline | Worst-case deltas (temp, gain, Rs) are covered |
Guardband rule (simple)
Allocate an explicit margin: Vrms_allow_used = Vrms_allow × (0.7–0.85) or reserve +2 to +3 dB in the noise budget
when humidity, cable motion, or power-mode changes are possible.
Example material numbers (noise-relevant)
Verify package, voltage rating, value suffix, and availability before locking BOM.
- Chopper / ultra-low 1/f: ADA4522-2, OPA188, LTC2057
- JFET / very low bias (high-Z sources): OPA140, ADA4625-1, OPA2140
- Low en (bipolar, watch in·Rs): ADA4898-2, OPA1612, LT1028
- Thin-film resistors (low excess noise): Vishay TNPW0603 / TNPW0805 series, Susumu RG1608 / RG2012 series
- Foil (extreme stability, low noise): Vishay VHP202Z / VSM series
- C0G/NP0 caps (filter/ENBW stability): Murata GRM series (C0G), TDK C0G series (value-specific suffix varies)
B) Layout review (noise-critical)
P0 (most likely to cause drift/jumps/hum)
- High-Z node leakage control: shortest routing, no exposed copper near input, solder mask strategy documented.
- Cleaning / residue: flux removal process defined; contamination risk points identified (connectors, guards, test pads).
- Guard ring present: ring surrounds the sensitive node; driven/biased correctly; no gaps at vias.
- Return-path continuity: high di/dt loops do not cross the analog reference region; no split-plane surprise paths.
- Decoupling loop area: cap placement + return via strategy keeps loop minimal at the amplifier and reference points.
P1 (spur/pickup and mismatch drivers)
- Differential symmetry: length/impedance/return symmetry maintained; common-mode paths are controlled.
- Partitioning: switching nodes, clocks, and high-slew traces are kept away from input and feedback networks.
- Cable/shield entry: connector placement supports stable shield termination experiments (single-point options).
- R-network placement: noise-critical resistors are compact, temperature-matched, and away from heat sources.
P2 (mechanical/environment drivers)
- Strain relief: cable and connector are mechanically secured; motion does not modulate the input node.
- Coating decision: conformal coating or potting is defined for humidity/condensation environments.
- Thermal gradients: sensitive networks avoid localized heat; airflow/heat sources are considered for drift.
Example material numbers (layout & assembly)
- Conformal coating (humidity leakage control): HumiSeal 1B31, HumiSeal 1B73, MG Chemicals 422B
- Low-microphonic coax examples: RG-316 (PTFE coax), RG-174 (general) — select per bandwidth/shielding need
- Shield termination hardware: Keystone 5015 (test point), Keystone 5016 (test point) for repeatable ground probing
C) Production hooks (consistency without over-testing)
Two fixtures (must be repeatable)
- Shorted input: baseline intrinsic noise check (minimize pickup).
- Fixed Rs source: expose source-related contributions and detect leakage sensitivity.
Three required pass criteria
1) In-band RMS
Vrms_meas ≤ Vrms_allow_used under the defined [fL, fH] and fixed settings.
2) Spur limit
50/60 Hz spur amplitude ≤ spur_limit (system budget). Treat “RMS pass but spur fail” as a fail.
3) Spectrum sanity
No unexplained peaks/sidebands; shape matches the expected white + low-frequency rise behavior.
Minimum record fields (keep it lightweight)
- SN / lot / date
- Fixture: shorted or Rs=____ Ω
- Band: fL=____, fH=____, ENBW=____
- Results: Vrms_meas, 50/60 Hz spur_amp, shape pass/fail
- Retest points: room + one cold/hot checkpoint (defined by the product plan)
Example material numbers (fixture & switching)
- Analog switch (fixture select): ADG1419, ADG1409, TMUX1108
- Reed relay (very low leakage switching): Coto 9007 series, Pickering 100 series (select coil/contacts per design)
- Precision source resistors: Vishay Dale RN55 / RN60 series, Vishay TNPW series (value/package specific)
Notes on part numbers
Example material numbers are provided to speed up datasheet lookup and fixture planning. Final selection must be driven by the budget (Rs, ENBW, SNR), operating conditions (temperature, supply, gain), and verification results under the fixed [fL, fH].
Request a Quote
FAQs (Noise in Active Signal Chains)
Each answer is structured for fast debugging: Likely cause → Quick check → Fix → Pass criteria. Keep the same [fL,fH] / ENBW definition when comparing results.
