123 Main Street, New York, NY 10001

Submit Your BOM

Bridge / Strain Chains: PGA + Notch + Lock-In Conditioning

Bridge / Strain Chains: PGA + Notch + Lock-In Conditioning

← Back to:Active Filters & Signal Conditioning

Build a repeatable bridge/strain signal chain that stays stable under mains hum, cable pickup, and temperature drift. Use ratiometric excitation, staged gain + notch/LP, optional lock-in demod, and measurable pass/fail criteria to validate noise and drift end-to-end.

What this page solves: stable bridge measurements under noise + drift

Bridge / strain chains often fail in the field not because the sensor is “wrong,” but because the signal chain cannot hold a stable baseline when mains pickup, low-frequency noise, cable effects, and temperature drift stack together.

This page provides a reusable chain template (excitation → bridge → gain → mains rejection → bandwidth control → optional lock-in), plus a budgeting mindset and a bring-up checklist to turn symptoms into measurable fixes.

Common failure modes (field-visible)

  • “No-load” output drifts or wanders after power-up
  • Touching / moving a cable causes step-like reading jumps
  • Persistent 50/60 Hz ripple (and harmonics) rides on the signal
  • Temperature change causes offset to “fly away” or creep slowly
  • Changing cable length / connector alters zero or span
  • After a transient or overload, recovery is slow and looks like drift

What “stable” means (measurable KPIs)

Mains residue
Residual ripple at 50/60 Hz < X (set by budget)
Warm-up stability
Time-to-stable < X (set by production spec)
Temperature drift
Offset drift < X ppm/°C (set by system target)
Cable sensitivity
Touch/motion induced step < X (set by budget)

Deliverables (what this page gives)

  1. A reusable bridge/strain signal-chain template
  2. A budgeting mindset (noise, drift, headroom, mains residue)
  3. A bring-up and validation checklist (fast fault isolation)
Scope guardrail
Focus stays on bridge signal conditioning (excitation, cabling/sense, gain, mains rejection, bandwidth control, optional lock-in), plus calibration hooks and validation methods. Avoid expanding into sensor mechanics or converter architecture.
Bridge/Strain Signal Chain Map (template)
Bridge and strain signal chain map A block diagram showing excitation, bridge sensor, remote sense, INA/PGA gain, mains notch, low-pass/anti-alias filtering, optional lock-in demodulation, and ADC. Signal path Conditioning path Excitation sets power & ratio Bridge μV/V differential Sense / Remote cancels line-drop INA / PGA gain with headroom Notch rejects 50/60 Hz LP / AA limits noise/alias Lock-in (opt) coherent demod for very low SNR ADC digitize + sync Template intent: define responsibility per block, then budget noise/drift/headroom and validate with repeatable tests.

Bridge chain architecture: where errors are born

Stable bridge measurements come from fast fault isolation: identify where an error enters the chain, then verify it with the smallest set of measurements. A bridge chain typically contains both systematic errors (often calibratable) and environmental errors (must be controlled by structure, layout, and bandwidth).

The map below lists the dominant error sources with “symptom → clue → quick check,” so later sections can be approached as targeted fixes rather than trial-and-error tuning.

Two buckets (for fast decisions)

Systematic (often calibratable)
Excitation ratio error, gain/offset errors, line-drop bias, predictable temp coefficients.
Environmental (must be controlled)
Mains/magnetic pickup, rectification-type EMI, 1/f noise dominance, cable micro-motion, overload recovery.

How to use this map

  1. Start from the dominant symptom (ripple, drift, step jump, slow recovery).
  2. Apply the one-line clue to decide whether the injection point is upstream or downstream.
  3. Run the “quick check” using the smallest number of probe points.
  4. Then jump to the section that fixes that injection mechanism.

Anchor links (placeholders for later sections)

1) Excitation error / noise
Symptom: span changes with supply or time.
Clue: ratiometric behavior missing or reference not tracking excitation.
Quick check: log excitation + output together; verify ratio stays constant (target: X).
2) Bridge + cable / connector effects
Symptom: reading jumps when cable is touched/moved.
Clue: mechanical micro-motion changes contact resistance or shielding current path.
Quick check: strain-relieve cable; swap connector; compare 4-wire vs remote-sense behavior.
3) Common-mode pickup / ground loop
Symptom: hum changes with cable routing or nearby equipment.
Clue: shorting the bridge input reduces ripple only partially (pickup is downstream).
Quick check: battery-power test; probe CM at front-end inputs; look for loop current signatures.
4) Front-end offset & 1/f dominance
Symptom: slow wander that scales with gain and narrows with bandwidth.
Clue: noise floor improves with stronger LP, but drift-like behavior remains.
Quick check: measure output PSD at low frequency; confirm 1/f corner vs target band.
5) Mains coupling before notch (saturation)
Symptom: notch “exists,” but output still looks unstable or clipped.
Clue: upstream stages overload on hum, then recover slowly (appears as drift).
Quick check: observe pre-notch node; check headroom margin to rails under worst mains pickup.
6) Rectification-type EMI (RF → DC shift)
Symptom: offset shifts near radios/switchers; not a clean sine ripple.
Clue: adding small RC/CM filtering changes the DC reading noticeably.
Quick check: introduce controlled RF aggressor; verify offset shift < X after mitigation.
7) Bandwidth/alias fold-in
Symptom: low-frequency “mystery” noise worsens with certain sampling rates.
Clue: changing LP corner or sample timing changes apparent baseline noise.
Quick check: sweep LP corner and sampling frequency; confirm noise trend matches alias expectation.
8) Reference / coherence (lock-in use case)
Symptom: lock-in helps on bench but not in field.
Clue: demod reference is not coherent with excitation (phase/frequency mismatch).
Quick check: confirm reference source is derived from excitation; verify phase trim range covers X.
Error injection map (numbered points + symptom index)
Bridge chain error injection map A block diagram of the bridge chain with numbered injection points and a legend mapping common symptoms to those points. Excitation Bridge + Cable Sense/Remote INA/PGA → Notch → LP → ADC 1 2 3 4 5 6 7 8 Symptom index → likely injection points • Warm-up drift / slow wander 1, 4, 2 • Touch/move cable → step jump 2, 3, 6 • Strong 50/60 Hz ripple 3, 5, 7 • “Notch exists” but still unstable 5, 4 • Offset shifts near RF/switchers 6, 3 • Lock-in helps on bench, not field 8, 1

Excitation & sensing: ratiometric, remote sense, and line-drop control

A stable bridge chain starts with how the bridge is powered and sensed. If excitation and measurement do not form a coherent ratio chain, supply drift and cable drops turn into apparent strain. If the bridge-end excitation is not controlled, long cables and connectors create slow bias shifts that look like temperature drift.

This section defines the practical rules for ratiometric measurement, the 4-wire vs 6-wire boundary, excitation mode trade-offs, and a verification method for self-heating and thermal settling.

Ratiometric rule: cancel supply/reference drift by design

  • Goal: keep the measurement proportional to bridge excitation, not absolute voltage.
  • Requirement: ADC reference (or measurement reference) must track excitation or be derived from the same source.
  • Failure signature: reading changes with supply even when the bridge load is unchanged.
Quick check
Log Vexc,bridge and output code together; verify the ratio stays within X (budget-defined) over supply and time.

4-wire vs 6-wire: when remote sense is worth it

  • 4-wire risk: line drop changes bridge-end excitation → span/zero shifts.
  • 6-wire benefit: regulation targets bridge-end voltage via sense leads.
  • Hidden culprit: connector contact resistance (micro-motion) can create step-like errors.
Boundary conditions
Remote sense is justified when ΔVline/Vexc > X, temperature swing is large, or connector repeatability is not guaranteed.

Excitation mode: constant voltage vs constant current

Constant voltage
Bridge output scales with Vexc. Line drop and supply drift must be controlled (ratiometric + sense).
Constant current
Output ties directly to ΔR under Iexc. Self-heating and R(T) behavior must be verified against drift targets.
Selection rule
Choose the mode that minimizes the dominant error path (line-drop, supply drift, or R(T)/self-heating) under worst-case conditions.

Self-heating: power → temperature rise → offset/span drift

  • Symptom: slow creep after excitation changes; looks like “warm-up drift.”
  • Cause chain: excitation power changes bridge temperature → resistance shifts → output drifts.
  • Control: limit power, manage duty cycle, define a measurable settle window.
Verification method
Apply an excitation step, record output vs time, and define “stable” as staying within ±X for Y seconds (budget-defined).

Bring-up micro-checklist (excitation & sense)

Bridge-end excitation
Measure VFORCE vs VSENSE; enforce |VFORCE−VSENSE| < X.
Cable disturbance
Touch/motion test: induced step < X; if not, prioritize connector and shield current path.
Warm-up settle
Time-to-stable < X; if longer, reduce excitation power or define a controlled preheat window.
Remote sense diagram (4-wire vs 6-wire) with RL and Rc
Remote sense for bridge excitation A block diagram comparing 4-wire excitation and 6-wire remote sense, showing FORCE and SENSE loops with line resistance RL and contact resistance Rc, and a checklist for when 6-wire is worth it. Bridge excitation control at the bridge-end (not at the driver pins) 4-wire (FORCE only) 6-wire (FORCE + SENSE) AFE driver control AFE remote sense Bridge sensor Bridge sensor FORCE+ FORCE− RL Rc RL Rc Bridge-end Vexc depends on (RL + Rc) in 4-wire FORCE+ FORCE− RL Rc RL Rc SENSE+ SENSE− 6-wire regulates bridge-end Vexc using SENSE feedback 6-wire worth it when long cable / high RL large temp swing connector Rc varies tight drift budget Verification points VFORCE vs VSENSE touch/motion step warm-up settle

Front-end gain: INA/PGA partitioning and overload recovery

Gain selection for bridge signals is not only about noise. It is primarily about preventing overload from mains pickup and transients, because saturation and slow recovery often masquerade as drift. A correct gain plan keeps the signal resolvable while preserving headroom for interference and common-mode excursions.

The key is to budget Signal, Interference (50/60 Hz + transients), and Headroom on the same scale at the critical nodes, then decide whether high front-end gain or staged gain is the safer structure.

Three quantities that must coexist

Signal
Minimum/maximum bridge differential level, including offset and worst load.
Interference
50/60 Hz pickup, harmonics, and plug/unplug transients at the input.
Headroom
Allowed input CM range and output swing margin under supply limits.
Guardband rule
Ensure worst-case (Signal + Interference + bias) stays below X% of headroom at the key nodes (budget-defined).

High front-end gain vs staged gain

High front-end gain
Better noise floor, but higher risk of overload from mains pickup and transients before rejection happens.
Staged gain
Preserves early headroom, enables rejection first, then raises gain; noise budget must still close.
Decision trigger
If interference is uncertain or large, prefer staged gain so suppression occurs before any stage can saturate.

Common-mode and swing: small differential does not mean “easy”

  • Bridge differential can be microvolts, while input common-mode may sit near the excitation midpoint.
  • CMRR collapses with impedance mismatch and layout asymmetry, turning CM pickup into differential error.
  • Common-mode steps can induce long recovery even if DC range looks valid on paper.
Quick check
Probe both inputs; estimate CM and differential simultaneously and verify CM remains within the allowed window under worst routing.

Overload & recovery: “fake drift” mechanism

  • Trigger: mains pickup or plug/unplug transient saturates an upstream stage.
  • Result: recovery is slow (filter discharge, input stage recovery), resembling warm-up drift.
  • Key test: observe a pre-rejection node; if it clips, later notch/LP cannot “undo” the overload.
Pass criteria
After a defined disturbance, the output returns within ±X of baseline within Y seconds (budget-defined).

Mux / multi-channel: settling and crosstalk traps

  • Channel switching forces input networks and filters to re-settle; first sample is often invalid.
  • Crosstalk increases when source impedance is high and switching edges inject charge.
  • Scheduling control (wait + discard) is often the cheapest stability upgrade.
Quick check
Compare single-channel steady-state vs round-robin mode; if instability appears only in round-robin, settling time is insufficient.
Gain partition and headroom bars (high gain vs staged gain)
Gain partitioning and headroom comparison Two side-by-side bar diagrams comparing high front-end gain and staged gain. Each shows signal, interference, and headroom on the same scale to visualize overload risk. High front-end gain Staged gain Same scale per column: compare Signal, Interference, and Headroom Signal Interference Headroom Overload risk Interference approaches headroom before rejection Signal Interference Headroom Structure benefit Rejection before high gain reduces clipping Use the same scale per node: verify no stage clips under worst-case interference before notch/LP take effect.

Noise budgeting for bridges: 1/f, source impedance, bandwidth, and alias

Bridge chains operate at low bandwidth, where 1/f noise, source impedance, and interference folding can dominate the output. A usable budget must be executable: list contributors, refer them to a single node, integrate over the target bandwidth, and compare against a numeric requirement.

This section provides a worksheet-style framework: contributors → RMS over BW → dominant term → design action, plus a practical alias check for mains/harmonics that show up as low-frequency wander.

Step 0: define the budget interface (target, bandwidth, and reference node)

Choose one reporting point
Use a single node for the entire worksheet: EIN (equivalent input noise) or EOUT (output-referred). Avoid mixed units.
Target bandwidth
Set BWtarget = X (application-defined). Bridge chains are usually low-frequency; 1/f often dominates.
Sampling / update context
Record Fs, decimation/averaging, and notch/LP locations. These settings decide whether mains content can fold into baseband.

Step 1: list contributors (en, in×Rs, bridge Johnson, reference/excitation)

  • Bridge Johnson: thermal noise from the bridge equivalent source impedance Rs.
  • Amp en: voltage noise density referred to the input.
  • Amp in×Rs: current noise converted to voltage by source impedance (often dominant at low frequency / high R).
  • Reference/excitation noise: enters through the ratio chain; if not perfectly canceled, it becomes a real contributor.
Quick check
Estimate in×Rs at the operating band; if it exceeds en, source impedance is the lever.

1/f vs drift: separate “random low-frequency” from “slow change”

Engineering boundary
1/f raises RMS as bandwidth moves lower. Drift appears as a slow monotonic or step-like change (thermal settling, overload recovery, contact resistance). Both look similar unless tested with controlled bandwidth and time windows.
Two-test split
Run two bandwidth settings: if RMS scales with √BW as expected, noise dominates. If not, drift/settling dominates (treat as a stability problem).

Step 2: integrate over BW (good-enough math, budget-grade)

White-noise items
Convert density to RMS with: Vrms ≈ N × √BW. Use BWtarget and refer all terms to the same node.
1/f items (practical handling)
Use the 1/f corner as a decision boundary. If BW sits mostly below the corner, treat low-frequency noise as dominant and validate with measured PSD + numeric integration (preferred in production-grade budgets).
RMS combine
Combine uncorrelated contributors by RMS: Vtotal = √(ΣVi2), then compare to Noisetarget = X.

Alias: when mains/harmonics reappear as low-frequency wander

  • Mechanism: strong 50/60 Hz and harmonics fold into baseband after sampling/decimation if not suppressed first.
  • Symptom: slow ripple or “breathing” in the reading, often misread as drift.
  • Risk amplifier: any upstream nonlinearity (overload/rectification) creates low-frequency content that cannot be filtered back out.
Quick check
Change update rate or decimation and observe the “low-frequency” component; if it shifts significantly, folding is likely. Enforce pre-sampling suppression so no stage clips before notch/LP act.

One-page workflow (budget → dominant term → action)

Contributor Density To EIN Action lever
Bridge Johnson X nV/√Hz direct reduce BW / raise excitation within drift limits
Amp en X nV/√Hz / gain choose lower en / move gain earlier if headroom allows
Amp in×Rs X nA/√Hz ×Rs lower Rs / select lower in / limit leakage paths
Ref / Exc noise X µV/√Hz ratio path tighten ratiometric chain / filter ref / enforce sense regulation
Pass criteria
Within BWtarget, total RMS noise stays below X (EIN or EOUT, consistent), and no upstream node clips under worst-case mains pickup.
Noise worksheet + spectrum sketch (contributors, BW, 50/60, notch, LP)
Bridge noise budgeting worksheet and spectrum sketch Left panel shows a contributor table for bridge Johnson noise, amplifier en, amplifier in times source resistance, and reference or excitation noise. Right panel shows a simplified spectrum with 1/f rise, mains peaks, notch depth, and low-pass bandwidth shaping. Budget = list contributors → refer to one node → integrate over BW → compare to target Noise contributors (example fields) Contributor Density Path Bridge Johnson X nV/√Hz direct Amp e_n X nV/√Hz input Amp i_n×R_s X nV/√Hz R_s Ref / Exc X µV/√Hz ratio Dominant term → action lever reduce R_s / reduce BW / improve ref Spectrum sketch (not to scale) frequency → amplitude 1/f rise 50/60 H Notch depth BW target (LP) after LP

50/60 Hz suppression: notch placement, depth, and temperature stability

Mains pickup is not only a ripple problem; it is a headroom and linearity problem. If 50/60 Hz enters a high-gain stage before suppression, saturation and rectification can create low-frequency artifacts that no “perfect” digital notch can remove.

This section defines where to place notch filtering (analog vs digital), how to specify depth using measurable criteria, and how to ensure stability over temperature and component tolerance.

Notch placement: analog first for headroom, digital later for precision

Analog notch (front)
Reduces mains before high gain, preventing overload and rectification artifacts. Also protects settle time and recovery behavior.
Digital notch (back)
Achieves deeper and more accurate attenuation when upstream stages remain strictly linear.
Decision trigger
If mains pickup can push any upstream node near clipping, enforce analog suppression or limiting before high gain.

Depth requirement: specify in terms of residual ripple and headroom

Define measurable fields
  • Vmains,in = X (measured worst-case at the relevant node)
  • Vripple,out < X (allowed residual ripple in the output band)
  • Headroom margin > X% (no upstream saturation under worst routing/shield conditions)
Engineering mapping
Required attenuation is set by Vmains,in vs Vripple,out and by the need to keep pre-notch nodes linear. Express notch depth in dB with placeholders: Depth > X dB.

Twin-T / active notch stability: center drift and tolerance collapse depth

  • Deep notches require matching; tolerance and tempco shift f0 and reduce effective depth.
  • Field symptom: hum returns over temperature even though the bench case looked clean.
  • Best practice: use stable parts (matched networks, low-tempco), include trim where needed, and keep front-end linear.
Pass criteria
Across temperature, center drift stays within ±X Hz and depth remains > X dB (measurement-defined).

50 vs 60 Hz regions: selection strategy without expanding into DSP

Selectable notch
Use a configuration to target 50 or 60 (and optional harmonics). Keep analog protection if headroom risk exists.
When notch is not enough
If the signal is narrowband and SNR is extremely low, consider coherent extraction methods (covered in the lock-in section).

Verification checklist (notch as a deliverable)

Center accuracy
|f0 − 50/60| < X (define test method and sweep window).
Depth
Depth > X dB at nominal, and stays > Y dB over temperature.
Residual ripple
In-band residual ripple < X under worst coupling, and no upstream node clips before notch action.
Notch before/after spectrum (depth, center drift, residual ripple spec)
Notch filtering before and after: depth, drift, residual ripple A simplified spectrum comparison showing mains peak before and after notch, highlighting notch depth, center frequency drift band, and a residual ripple specification line. Also includes a warning that clipping before suppression invalidates later filtering. Analog protection preserves linearity; digital precision adds depth when upstream is clean Before notch noise + 1/f 50/60 H Warning if any stage clips here, later filtering cannot restore linear data After notch in-band floor residual spec (X) 50/60 depth drift ±X Hz Recommended split front: protect headroom back: add precision depth

Lock-in / PSD for strain: when narrowband demod beats brute-force filtering

When bridge measurements are limited by low-frequency noise (1/f, mains pickup, slow environmental interference), stacking filters often trades bandwidth for latency but still misses the noise floor. Lock-in moves the signal to a controlled modulation frequency and extracts it coherently, shrinking effective noise bandwidth to what the post-demod low-pass allows.

This section defines value boundaries, coherence requirements, I/Q robustness, modulation frequency constraints, and practical implementation options (analog switching vs digital multiply) without expanding into general DSP theory.

When lock-in is worth it (value boundary)

  • Signal can be modulated: AC excitation (sin/cos, square, polarity switching) is feasible.
  • Noise dominates at low frequency: 1/f and mains pickup overwhelm the strain band of interest.
  • Coherent reference exists: reference is derived from the same excitation source (shared time/phase domain).
  • Output intent is narrowband: magnitude/phase at fmod is the deliverable, not a wideband waveform.
Decision trigger
If required SNR cannot be reached without unacceptable latency using LP/notch alone, lock-in becomes the preferred architecture lever.

Coherence + I/Q: prevent phase error from becoming amplitude error

Coherent reference
Reference must come from excitation. Any free-running reference breaks coherent gain and turns lock-in into ordinary filtering.
Why I/Q
Single-phase demod output scales with cos(φ). I/Q supports magnitude √(I²+Q²) and a diagnostic phase output for trim.
Pass criteria
Coherence is verified (reference locked to excitation), and phase trim keeps magnitude error below X (application-defined).

Modulation frequency selection (dynamic limits)

Hard constraints
  • Move above the 1/f-dominated region (noise floor is flatter).
  • Avoid 50/60 Hz and harmonics (and known mechanical/vibration lines).
  • Stay within the linear, phase-controlled bandwidth of the analog front end.
Engineering placeholder
Choose fmod = X…Y where PSD shows a clean window and analog gain remains linear with sufficient headroom.

Implementation split: analog switching vs digital multiply (no DSP deep-dive)

Analog demod (switch/multiplier)
Reduces dynamic-range pressure early. Risk points: injection, reference feedthrough, and rectification if any stage overloads.
Digital demod (after ADC)
Adds consistent depth and easy I/Q. Risk point: if analog clips or mixes before ADC, digital demod cannot recover linear information.
Non-negotiable rule
Maintain linear headroom at every node before suppression. Otherwise, “extra low-frequency drift” can be created by rectification/mixing.

LP time constant: noise floor vs response time

  • Post-demod low-pass sets effective noise bandwidth (smaller BW → lower noise).
  • Time constant τ sets response time (larger τ → slower updates).
  • Choose τ from the required update/settling spec: τ = X (placeholder).
Quick check
Inject a known modulation amplitude step and verify 90% settling time < X while RMS output noise meets the target.

PSD usage here: pick fmod and diagnose residual artifacts

Before design
Use PSD to find a clean window away from 1/f rise, mains lines, and known mechanical bands; place fmod there.
During debug
If “slow wander” persists after demod, PSD can reveal sidebands, harmonic leakage, or reference feedthrough at predictable offsets.
Lock-in I/Q block map (coherent reference, phase trim, LP time constant)
Lock-in I/Q demod block diagram for bridge strain Signal chain shows coherent excitation reference feeding cosine and sine multipliers, low-pass filters with time constant tau, I and Q outputs, and magnitude computation. Side tags highlight reference coherence, phase trim, and LP time constant. Coherent demod: move strain to fmod → multiply by ref → low-pass → I/Q → magnitude Excite sin / cos Bridge INA / PGA linear × cos I path × sin Q path LP (τ) bandwidth LP (τ) bandwidth I Q Magnitude |IQ| Reference Phase trim Reference coherence same source / locked phase Phase alignment I/Q magnitude is robust LP time constant (τ) noise BW vs response time

Temperature drift: separating true strain from thermal effects

Temperature-related drift is often misdiagnosed as “noise.” In bridge chains, thermal effects split into three actionable buckets: zero shift, sensitivity shift, and self-heating. Each bucket has distinct symptoms, test hooks, and countermeasures.

This section provides a measurable causal map plus a temperature-sweep workflow that enforces a stabilization criterion before data is used for attribution.

Classify first: zero shift vs sensitivity shift vs self-heating

Zero shift
Input offset drift, thermoelectric EMF, lead/contact resistance changes → looks like an additive error.
Sensitivity shift
Excitation drift, bridge arm tempco, gain network tempco → looks like a multiplicative error.
Self-heating
Excitation power → temperature rise → slow time-constant drift; often mistaken for “low-frequency noise.”

Zero shift: isolate offset drift, thermoelectric EMF, and lead/contact effects

Quick checks (measurable)
  • Zero-input test: short the input / simulate zero load; drift that remains points to front-end offset paths.
  • Thermal-gradient test: introduce small connector/cable temperature differences; reversible shifts indicate thermoelectric EMF.
  • Cable/connector swap: change cable length or touch the connector; drift sensitivity indicates lead/contact resistance paths.
Pass criteria
In a controlled zero-input condition, drift stays below X over the specified temperature range (test-defined).

Sensitivity shift: detect proportional errors (excitation + gain tempco)

  • Proportional symptom: error scales with applied strain amplitude (multiplicative behavior).
  • Excitation check: small programmed excitation change produces proportional output change beyond expectation → excitation path dominates.
  • Gain mapping: switching gain states shifts the temperature slope → gain network tempco is implicated.
Output split
Record both zero and two load points per temperature step to separate offset(T) from gain(T).

Self-heating: power → temperature rise → slow drift (time-constant signature)

Signature
Output changes follow a slow exponential-like response after excitation changes, start-up, or duty-cycle changes.
Quick check
Change excitation amplitude or duty. If drift magnitude and time constant change accordingly, self-heating dominates.
Stability criterion (placeholder)
Consider the system settled when |d(out)/dt| < X for T seconds (X and T are budget-defined).

Temperature sweep workflow (production-grade discipline)

  1. Define temperature points: T1…Tn (range and spacing are application-defined).
  2. At each point, wait until the stability criterion is met (|d(out)/dt| < X for T seconds).
  3. Record: zero + two load points + excitation/REF values (for ratiometric cross-check).
  4. Split results into offset(T) and gain(T); flag a self-heating signature if time constants are dominant.
Pass criteria
Data used for drift attribution is only accepted after stabilization; otherwise slopes and “drift” conclusions are invalid.
Drift attribution flow (ratiometric, self-heating, offset vs sensitivity)
Temperature drift attribution flowchart for bridge strain chains Flowchart starts with observed monotonic drift versus temperature, then branches into ratiometric coherence check, self-heating signature check, and offset or sensitivity drift checks. Each branch includes a quick measurement action and ends in labeled conclusions. Observe → classify → quick checks → conclude (zero shift / sensitivity shift / self-heating) Observation Output drifts with temperature Ratiometric OK? REF/EXC coherence Self-heating? duty / power test Offset vs Gain? zero + two loads Quick check log REF + EXC same reference path Quick check change duty watch τ signature Quick check zero-input test gain mapping Zero shift offset / EMF / leads additive error Self-heating power → τ drift time-constant Sensitivity shift exc / gain tempco multiplicative Temperature sweep rule accept data only after |d(out)/dt| < X for T seconds

Anti-alias LP & settling: bandwidth vs response time trade-off

In bridge/strain chains, the low-pass stage is not only “noise smoothing.” It defines whether out-of-band interference becomes a slow in-band artifact (alias-like behavior) and whether the measurement remains usable under real operating events (gain steps, mux switching, start-up recovery).

This section provides a bridge-specific decision flow for fc selection, order vs group delay, and settling verification without expanding into filter topology catalogs.

fc selection: start from update/loop/UX, not from components

  1. Define the required update cadence (display/logging/control-loop input).
  2. Define the maximum allowed latency (human perception or loop stability margin).
  3. Pick fc to meet the latency target while keeping out-of-band interference from folding into the band.
Engineering placeholder
Choose fc = X × update_rate (X is budget-defined). Verify that interference residual is below the allowed ripple at the output.

Filter order: stopband suppression vs group delay and step behavior

What higher order buys
Stronger attenuation of 50/60 harmonics, switching artifacts, and EMI-derived content that can otherwise appear as slow wander.
What higher order costs
More group delay and slower settling. In bridge chains, this can look like drift during transients or range changes.
Bridge-specific guidance
When out-of-band interference is far above the signal, prioritize attenuation (order/notch/lock-in) instead of forcing fc extremely low.

Settling is event-driven: define and verify the “switching cases”

Common bridge cases
  • Gain steps (PGA): output takes time to converge due to internal state and analog settling.
  • Mux channel switching: residual charge and common-mode differences show as a long tail.
  • Start-up / recovery: saturation and filter state can create a “slow return” that mimics drift.
Unified spec (placeholder)
Define settle to X% within T for each event. X and T must be derived from the measurement budget and update-rate needs.

Anti-alias viewpoint: avoid “slow drift” created by out-of-band content

  • Out-of-band interference can be transformed into low-frequency wander after sampling and filtering.
  • In bridge systems, that wander is frequently misinterpreted as temperature drift or sensor instability.
  • LP design must be validated under worst-case interference amplitude and real switching events.
Pass criteria
With worst-case interference present, residual output ripple stays below X while step settling stays within T.

Verification checklist (bridge-ready)

  • Step response: settle to X% < T.
  • Gain step: settle to X% < T with no long tail.
  • Mux step: cross-channel residue < X after T.
  • Recovery: after overload/start-up, output returns within X in T.
Note
X and T are placeholders. Set them from the system noise/latency budget and product requirements.
Step response vs bandwidth (settle-to-X% comparison)
Step response versus low-pass bandwidth for bridge settling Chart compares fast, medium, and slow low-pass bandwidth step responses and shows a settle-to-X-percent criterion. Includes axes, a step input icon, and minimal labels for readability on mobile. Bandwidth trade-off: fast response vs tighter smoothing (settle-to-X% is the test) Step input Time Output Settle to X% Fast BW Mid BW Slow BW

Calibration hooks: zero/span, shunt cal, auto-zero, and production repeatability

A bridge front end that “looks stable” on a bench can still fail in production if calibration is not repeatable, versioned, and traceable. This section defines practical calibration hooks that map to the two dominant error types: zero (additive) and span (multiplicative), plus shunt calibration for field health checks and auto-zero timing for predictable offset control.

The deliverable is a production-ready method matrix and a minimal data schema to ensure consistency across lots, temperature points, and firmware builds.

Zero & span: decide if 1–2 points are enough

  • Zero targets additive errors (offset-like behavior).
  • Span targets multiplicative errors (gain-like behavior).
  • 1–2 points are sufficient when curvature/nonlinearity is negligible and measurement uncertainty is well below the target.
Pass criteria
After calibration, residual zero and span errors remain within X across the specified temperature and time window.

Shunt calibration: field health check and consistency monitor

  • Inject a repeatable “known-direction” response using a bridge shunt resistor.
  • Detect gain-path changes, wiring/contact degradation, and abnormal front-end behavior via trend monitoring.
  • Use as a Go/No-Go + drift tracker; do not treat it as a high-accuracy absolute calibration method.
When to run
Startup and periodic intervals (period placeholder), and after service events (cable replacement, connector reseat).

Auto-zero: timing strategy without corrupting measurements

Before measurement
Clears offset prior to acquisition; must avoid injecting discontinuities into the data stream.
After measurement
Suited for background correction; must ensure true signal is not misclassified as offset during changes.
Bridge-safe rule
Trigger auto-zero at controlled moments (startup, after stabilization, or scheduled idle windows) rather than during active transients.

Production repeatability: minimal data fields for traceability

  • Calibration version: coefficients + algorithm version id.
  • Temperature point: recorded T and stability status (criterion met or not).
  • Key results: zero, span, shunt response (magnitude; phase if using I/Q).
  • Drift deltas: difference vs last known-good baseline.
  • Firmware/build id: ensures consistent reproduction across updates.
Pass criteria
Coefficients and outcomes are traceable and repeatable across lots; outliers can be attributed by data, not guesswork.
Calibration / self-test matrix (method × catches × when to run)
Calibration and self-test matrix for bridge strain chains Matrix table shows methods (Zero, Span, Shunt cal, Auto-zero) versus what each catches (offset, gain, wiring trend) and when to run (startup, periodic, after temp settle, after service). Uses icons and check marks with minimal text. Methods mapped to failure coverage and run timing (for production repeatability) Method What it catches When to run Zero Offset Additive path Startup After settle Span Gain Multiply path After settle Service Shunt cal Gain path Trend check Periodic Startup Auto-zero Offset Timing safe Startup Idle window Record cal version + temperature + stability status + outcomes to keep production repeatable

Engineering checklist: layout, grounding, cabling, and validation tests

This section converts the full bridge/strain chain into a copy-paste checklist. Every item is written as Action + Pass criteria (threshold placeholders are defined by the system error budget).

A) Layout review checklist (bridge inputs, leakage, symmetry, and return paths)

A1 · Input entry + protection (protect without “killing” precision)
  • Action: Place the first series limiter (R/RC) immediately after the connector; place TVS/clamps at the entry; force the clamp return to a defined analog return node (no long shared digital return).
    Pass: After EFT/ESD event, recovery time < [T_REC]; post-event zero shift < [X_ZERO].
  • Action: Model clamp current during worst-case transients; verify series element limits clamp current below [I_CLAMP_MAX].
    Pass: No long-tail drift after transients; no repeated saturation of the front-end.
A2 · Leakage control (guard, cleanliness, humidity robustness)
  • Action: Add guard rings around high-impedance nodes (driven to the same potential); keep flux residues away from bridge inputs; define a cleaning/conformal-coating rule for high-ohmic front-ends.
    Pass: Touch/near-hand test causes step < [X_STEP]; humidity soak drift < [X_DRIFT].
  • Action: Avoid solder mask “pools” near input pins; avoid long exposed high-impedance traces.
    Pass: Input bias/leakage induced offset error < [X_IBIAS] with worst-case source resistance.
A3 · Differential symmetry + continuous return (reduce hidden loop pickup)
  • Action: Route the bridge differential pair tightly coupled and symmetric (length/impedance); avoid ground splits under the pair; keep a continuous return reference.
    Pass: Common-mode step injection produces differential spike < [X_CM2DM]; settling < [T_CM_REC].
  • Action: Define star-point intent explicitly (where analog meets power/digital); verify no “accidental star” through long copper.
    Pass: 50/60 Hz pickup meets residual ripple target < [X_MAINS].
A4 · Digital edge isolation (avoid rectification into “fake drift”)
  • Action: Keep SPI/I²C/clock edges away from bridge nodes; ensure local decoupling loops are minimal; separate fast return currents from the analog reference.
    Pass: Communication activity changes RMS noise by < [X_RMS_DELTA]; no new low-frequency spurs after enabling digital traffic.

B) Cabling checklist (shielding, twisting, connectors, strain relief)

B1 · Shield termination rule (define the boundary condition)
  • Action: Specify shield termination strategy (single-end / both-ends / RC to chassis) and document which node is “quiet reference.”
    Pass: Touch/move cable step < [X_TOUCH]; mains ripple meets < [X_MAINS].
B2 · Twisting + loop area (the fastest win)
  • Action: Twist signal pair and excitation pair; keep FORCE/SENSE as paired runs; avoid large “out-and-back” loops.
    Pass: External 50/60 Hz magnetic field test injects < [X_HUM] at the ADC code.
B3 · Connector contact stability + strain relief (prevent intermittent Rc)
  • Action: Add mechanical strain relief; define an insertion/removal procedure; avoid stress transfer to solder joints.
    Pass: Tap/pull test produces no step > [X_TAP]; repeated re-mate shift < [X_REMAP].

C) Validation tests (repeatable setups with pass/fail thresholds)

C1 · Mains injection (verify notch/LP, avoid overload)
  • Action: Inject 50/60 Hz at the cable/bridge input using a controlled coupling method (amplitude [V_INJ]).
  • Pass: Residual ripple after notch/LP < [X_RESID]; no front-end saturation observed.
C2 · Thermal sweep (separate true strain vs thermal effects)
  • Action: Sweep temperature in steps; define soak completion by stability threshold [X_STAB] over [T_SOAK].
  • Pass: Zero drift < [X_ZERO_T]; span drift < [X_SPAN_T]; repeatability < [X_REPEAT].
C3 · Touch/move cable (detect leakage + shield/return issues)
  • Action: Standardize a “touch/move” pattern and record step size + recovery time.
  • Pass: Step < [X_TOUCH]; recovery < [T_TOUCH].
C4 · Overload recovery (prevent “fake drift” tails)
  • Action: Apply a controlled overload pulse (amplitude/time [V_OL]/[T_OL]).
  • Pass: Recovery < [T_REC]; no long-tail baseline shift > [X_TAIL].
C5 · Common-mode step (CMRR in the real build)
  • Action: Inject a CM step and observe DM output spike + settling.
  • Pass: CM→DM spike < [X_CM2DM]; settle < [T_CM_REC].
Validation Test Cards — Bench, Thermal, EMI/Cable Three test categories with compact pass/fail placeholders for bridge and strain measurement chains. Validation Test Cards (copy-paste) Each card: setup → stimulus → observe → pass (< X, < T) Bench Thermal EMI / Cable Mains inject Pass: residual < [X] CM step Pass: spike < [X] Overload recovery Pass: < [T_REC] Step sweep Pass: drift < [X] Soak stability Pass: stable in [T] Repeatability Pass: < [X_REPEAT] Touch cable Pass: step < [X] Shield rule Pass: ripple < [X] EFT/ESD recover Pass: < [T_REC]

Applications & IC selection notes for bridge chains (with example part numbers)

The goal is a “chain recipe” and a vendor-facing selection schema. Part numbers below are starting points for datasheet lookup and lab evaluation; selection must be verified under worst-case conditions and guardbands.

A) Application recipes (chain “formulas”)

Recipe 1 · Long-cable weigh scale / strain (remote sense + staged gain + mains control)
Remote sense INA/PGA (staged) Analog notch LP / AA 24–32b ΣΔ ADC
  • Focus: line-drop drift (RL/Rc), CM pickup, 50/60 overload and recovery tails.
  • Pass: residual mains ripple < [X_MAINS]; touch-cable step < [X_TOUCH]; overload recovery < [T_REC].
Example BOM part numbers (IC)
  • Sensor-ready ADC/AFE: TI ADS1262 / ADS1263; TI ADS124S08; ADI AD7124-8; ADI AD7190
  • Instrumentation amp: TI INA828; TI INA333; ADI AD8421
  • Precision reference (non-ratiometric cases): TI REF50xx (e.g., REF5050); ADI ADR4525 / ADR4550 family
  • Precision switch / routing / shunt-cal control: TI TMUX1109
Recipe 2 · Micro-strain at very low frequency (AC excitation + lock-in demod + slow LP)
AC excite Synchronous demod I/Q Slow LP Magnitude
  • Focus: reference coherence, phase trim, demod LP time constant vs response.
  • Pass: phase-induced amplitude error < [X_PHASE_ERR]; demod noise floor < [X_NF].
Example BOM part numbers (IC)
  • Analog lock-in demod core: ADI AD630 (synchronous demod / lock-in building block)
  • Front-end INA: TI INA828; ADI AD8421; TI INA333
  • ADC for digitized lock-in path: ADI AD7124-8; TI ADS1262 / ADS124S08 (when demod runs in DSP/MCU)
Recipe 3 · Multi-channel bridge arrays (mux + per-channel settling + traceability)
Mux Per-ch settle Auto-zero Shunt cal Logging
  • Focus: charge injection, leakage, channel-to-channel memory, settle-to-[X%] timing definition.
  • Pass: channel residue < [X_RES]; settle-to-[X%] < [T_SETTLE].
Example BOM part numbers (IC)
  • Precision mux / routing: TI TMUX1109 (low leakage / low charge injection class)
  • Programmable gain (fast settle): ADI AD8250 (digitally programmable gains)
  • Integrated AFE ADC options: TI ADS1262 / ADS124S08; ADI AD7124-8 / AD7190

B) IC selection schema (what to ask vendors + quick verification)

Spec field Why it matters in bridge chains Quick verification Example parts
1/f + offset drift Dominates “slow wandering” at low bandwidth; often confused with real strain. Shorted-input + worst-case source-R; temp steps with stability rule [X_STAB]/[T_SOAK]. INA828, INA333, AD8421; AD7124-8 / ADS1262 class AFEs
CMRR (DC/AC) + CM step recovery Long cables and mains fields turn CM into DM via imbalance; looks like hum + drift tails. Inject CM step; measure DM spike < [X_CM2DM] and settle < [T_CM_REC]. INA828, AD8421; (layout/cabling dominates the final result)
Overload recovery Saturation from 50/60 or cable events causes long recovery tails (“fake thermal drift”). Controlled overload pulse [V_OL]/[T_OL]; recovery < [T_REC]; tail < [X_TAIL]. ADS1262 / ADS124S08 / AD7124-8 class AFEs; INA828 class INAs
Gain programmability + settle Multi-range and multi-channel systems fail if settle-to-[X%] is not guaranteed. Step gain/mux; measure residue < [X_RES] within [T_SETTLE] at worst case. AD8250; TMUX1109; ADS1262/AD7124-8 with internal PGA
Reference / excitation noise Non-ratiometric paths need a low-noise VREF; ratiometric paths suppress common source noise. Compare ratiometric vs fixed-ref build; verify low-frequency noise contribution < [X_REF]. REF5050 (REF50xx), ADR4525/ADR4550 family
Lock-in demod capability (optional) Narrowband PSD extracts micro-strain under strong low-frequency noise and drift. Verify reference coherence and phase trim; magnitude error < [X_PHASE_ERR]. AD630 (analog demod); AD7124-8 / ADS1262 + digital demod
Bridge Chain Selection Field Map Spec-to-impact-to-test mapping with example part numbers for bridge and strain signal chains. Selection Field Map (Spec → Why → Quick Verification) Example part numbers are evaluation starting points, not universal recommendations. Spec Why it matters Quick check 1/f + drift Zero stability Temp steps CMRR Long cables CM step Recovery No tail drift OL pulse Settle Mux/range Step test VREF Noise floor A/B build Example parts ADS1262 AD7124-8 AD7190 INA828 INA333 AD630 REF50xx

Request a Quote

Accepted Formats

pdf, csv, xls, xlsx, zip

Attachment

Drag & drop files here or use the button below.

FAQs (Bridge / Strain Chains) — troubleshooting + pass/fail criteria

Each answer is fixed to 4 lines: Likely cause / Quick check / Fix / Pass criteria. Threshold placeholders ([X_*], [T_*], [F_*], [TEMP_*]) must be set by the system error budget.

Why does the reading drift for the first few minutes after power-up?
Likely cause: Thermal settling (AFE/reference) and/or bridge self-heating before equilibrium; auto-zero executed too early.
Quick check: Log zero-load output vs time; log excitation V/I vs time; verify auto-zero timing relative to stabilization.
Fix: Add warm-up/soak gating; reduce excitation or duty-cycle it; schedule auto-zero after stability is reached.
Pass criteria: |drift slope| < [X_ZERO_PER_MIN] after [T_WARMUP], and stable within [X_STAB] over [T_STAB].
Why does touching/moving the cable change the measured strain/weight?
Likely cause: Shield/return ambiguity, leakage paths, triboelectric cable effects, or CM pickup converting to DM via imbalance.
Quick check: Standardized touch test; measure input CM with a scope; check insulation resistance and connector contact stability.
Fix: Define shield termination (single-end or RC to chassis); twist pairs; add strain relief; improve symmetry and leakage control (guard/cleanliness).
Pass criteria: Touch step < [X_TOUCH] and recovery < [T_TOUCH]; CM→DM spike < [X_CM2DM].
My output shows a strong 50/60 Hz ripple—why isn’t the notch working?
Likely cause: Notch placed after a saturating stage, wrong center frequency, insufficient depth, or front-end rectification before the notch.
Quick check: Probe 50/60 amplitude before/after the notch; verify no clipping at PGA/INA outputs; confirm mains frequency [F_MAINS].
Fix: Move analog notch earlier (before high gain) or increase headroom; retune to [F_MAINS]; add a post-ADC digital notch for additional depth.
Pass criteria: Residual ripple at ADC < [X_MAINS_RESID] with no observable saturation during worst-case interference.
Notch is deep on the bench but shallow on the board—what changed?
Likely cause: In-circuit loading/parasitics, return-path differences, CM pickup, or component tolerance/temperature drift shifting the notch center.
Quick check: Inject a known 50/60 test tone in-circuit; measure notch response at the same node used on bench; verify component values and grounding continuity.
Fix: Use tighter-tolerance NP0/C0G parts; buffer the notch node if loading is present; move the network away from digital edges; fix return path symmetry.
Pass criteria: In-system notch depth ≥ [D_NOTCH_DB] at [F_MAINS], and center drift ≤ [F_DRIFT_MAX].
Why does higher PGA gain make the system “more sensitive to hum”?
Likely cause: Reduced headroom causes overload/recovery tails; hum enters as CM and converts to DM via imbalance, then gets amplified more.
Quick check: Check for saturation at PGA/INA output during hum peaks; compare hum level at input vs output across gain settings.
Fix: Use staged gain (less gain before notch/LP); add notch earlier; improve cable symmetry/shield rule; increase headroom or reduce input CM stress.
Pass criteria: No saturation at worst-case gain; hum contribution at ADC < [X_HUM_OUT] across all gain steps.
How do I tell true thermal drift from self-heating of the bridge?
Likely cause: Self-heating tracks excitation power and has a predictable time constant; true ambient drift tracks environment temperature changes and gradients.
Quick check: Step excitation on/off (or duty-cycle change) and observe the output time constant; log bridge resistance vs time/temperature.
Fix: Reduce excitation or duty-cycle it; improve thermal coupling to a stable mass; use ratiometric measurement and temperature compensation when needed.
Pass criteria: After excitation step, output settles within [T_SELFHEAT]; residual drift across [TEMP_MIN..TEMP_MAX] < [X_THERMAL].
4-wire looks OK short cable, but long cable fails—when is 6-wire remote sense worth it?
Likely cause: Line drop and contact resistance change the actual bridge excitation at the sensor; error grows with cable length and temperature.
Quick check: Measure excitation voltage at the bridge vs at the source under load; wiggle/re-mate connectors and observe ∆V at the bridge.
Fix: Use 6-wire remote sense when |∆V_bridge/V_bridge| exceeds the error budget; improve connectors; reduce excitation current if feasible.
Pass criteria: |∆V_bridge/V_bridge| < [X_LINE_DROP], and long-cable zero/span error < [X_LONG].
Why does the reading jump after gain switching or channel multiplexing?
Likely cause: Charge injection and memory effects (RC/filter state), insufficient settle time, or digital filter latency after switching.
Quick check: Capture post-switch step response; compare with an increased settle delay; check channel residue by reading a “dummy” known-zero path.
Fix: Add break-before-make and discharge paths; define per-channel settle time; reset/flush digital filter state on switches; sequence auto-zero after settle.
Pass criteria: Settles to [X_PCT]% within [T_SETTLE], and channel residue < [X_RES].
RC input filter reduces noise but worsens settling—what’s the right compromise?
Likely cause: RC time constant dominates step settling; large R also amplifies bias/leakage error and can increase CM→DM via imbalance.
Quick check: Compute τ = R_eq·C and compare to required response time; run step tests and record noise vs settle trade-off.
Fix: Use smaller R with adequate input protection; move most filtering to a later low-impedance stage; split into two mild stages instead of one heavy RC.
Pass criteria: RMS noise < [X_RMS] while settling to [X_PCT]% within [T_SETTLE].
Lock-in demod helps a lot in the lab but not in the field—what coherence condition is missing?
Likely cause: Reference is not phase-coherent with excitation, or phase drifts in the field (different clocks, routing, or EMI coupling at the modulation frequency).
Quick check: Verify reference is derived from the same excitation source; log phase vs time; verify I/Q balance and modulation frequency separation from interference.
Fix: Share clock/reference; add phase trim; choose modulation frequency away from field interferers; enforce synchronous sampling and stable routing.
Pass criteria: Coherence > [X_COH], phase drift < [X_PHASE] over [T_FIELD], SNR gain ≥ [X_SNR_GAIN].
Why does shunt calibration look good but real load changes are off?
Likely cause: Shunt injects an electrical imbalance that does not perfectly represent mechanical strain; shunt resistor tolerance/tempco or placement creates mismatch.
Quick check: Compare shunt vs known weights at multiple points; measure shunt resistance vs temperature; verify shunt location relative to bridge and sense points.
Fix: Use real-load multi-point span calibration for accuracy; keep shunt as a consistency check; use precision shunt resistor and stable placement.
Pass criteria: Load error < [X_LOAD_ERR] across range, and shunt repeatability < [X_SHUNT_REPEAT] over [TEMP_MIN..TEMP_MAX].
Why does offset look stable at room temp but fails across temperature?
Likely cause: Thermal EMF from junctions/gradients, input offset drift, gain network tempco, cable/contact tempco, or excitation/reference temp dependence.
Quick check: Perform a stepped thermal sweep with soak rule [X_STAB]/[T_SOAK]; log zero and span separately; reverse excitation polarity to expose thermal EMF sensitivity.
Fix: Reduce thermal gradients (layout/mechanics), use ratiometric excitation where applicable, add temperature compensation (table/fit), and control leakage/guarding over temperature.
Pass criteria: Zero drift < [X_ZERO_T] and span drift < [X_SPAN_T] over [TEMP_MIN..TEMP_MAX], with repeatability < [X_REPEAT].
icnavigator

ICNavigator helps engineers find verified ICs, alternatives, and fast quotes.

Explore
Categories
  • Power Management
  • MCU
  • Automotive ICs
  • Sensors
  • Interface & Transceivers
  • Analog Front-End
Get in Touch
  • Email: hello@icnavigator.com
  • WeChat/WhatsApp: +86-xxxx

© 2025 ICNavigator. All rights reserved.