123 Main Street, New York, NY 10001

Submit Your BOM

Bio-Potential / Low-Frequency Front-End with RLD + Notch

Bio-Potential / Low-Frequency Front-End with RLD + Notch

← Back to:Active Filters & Signal Conditioning

This page shows how to build a bio-potential low-frequency front end that stays stable at microvolt levels: ultra-low-noise amplification plus 50/60 Hz rejection (notch + RLD) without sacrificing recovery.

The focus is practical and testable—impedance balance, AC CMRR, drift/noise separation, and production hooks—so the chain performs on real cables and real environments, not just on paper.

What this chain is and where it breaks

This is a microvolt, low-frequency bio-potential front end. Success depends on treating electrode offsets, 1/f + 0.1–10 Hz noise, and 50/60 Hz interference as one coupled system (signal path + RLD loop), not as isolated “low-noise parts.”

Scope and targets (boundary first)

  • Signal envelope: µV–mV; DC / 0.05 Hz up to ~250 Hz (application-dependent).
  • Three hard enemies: electrode DC offset/polarization, LF noise (1/f + 0.1–10 Hz), 50/60 Hz coupling.
  • Primary objectives: meet LF noise budget, achieve real-world hum rejection, and recover quickly after motion artifacts.
  • Out of scope: isolation/regulatory details, generic EMI theory, full ADC-drive stability deep dive (link-only placeholders below).
Pass criteria placeholders: • 0.1–10 Hz input-referred noise < X µVrms (derived from system SNR / resolution target) • 50/60 Hz residual at output < Y (defined under worst-case cable + electrode imbalance) • Large artifact recovery time < T (application latency tolerance)

Chain anatomy (two coupled systems)

Signal path: Input protection/bias → Ultra-low-noise BP/INA → (HP/DC servo optional) → 50/60 notch → Output/ADC
Common-mode control path: Sense common-mode → RLD amplifier/compensation → RLD electrode drive → common-mode reduction at inputs
Rule of thumb: the signal path sets noise/drift/headroom; the RLD loop sets real-world hum rejection and may introduce stability/recovery pitfalls.

Where it breaks (symptom → first check → fix direction)

  • Minutes of baseline drift: compare inputs shorted vs electrodes connected; drift that vanishes on short usually points to electrode polarization / bias network noise.
  • Strong 50/60 Hz that barely changes with RLD: verify input impedance balance and RLD loop effectiveness (open/close RLD and measure hum delta).
  • Cable touch/motion causes slow steps: suspect motion artifacts + capacitive common-mode injection; observe common-mode and differential simultaneously.
  • Long recovery after a large artifact: check notch Q/time-constants and any HP/DC servo behavior; confirm no stage is saturating.
Bring-up sequence (fast isolation): 1) Disable RLD → characterize noise + drift (shorted input baseline). 2) Enable RLD → measure hum reduction (same cable routing, same electrode setup). 3) Enable notch → validate depth + recovery using controlled injection (no “mystery” signals).
Bio-potential low-frequency chain overview with RLD loop Block diagram showing electrodes to protection/bias, ultra-low-noise amplifier, optional HP/DC servo, 50/60 Hz notch, ADC, and a separate RLD common-mode feedback loop with key metrics. Inputs / Electrodes +IN −IN RLD Protection / Bias AC CMRR Ultra-Low-Noise BP / INA en + 1/f HP / DC Servo (optional) Recovery 50 / 60 Hz Notch Depth & Q ADC / Output Headroom RLD Sense Common-Mode RLD Amp + Comp Loop PM Key metrics Noise (en, 1/f) AC CMRR (50/60) Notch recovery
System overview: the signal path defines noise/drift/headroom, while the RLD loop defines real-world hum rejection and stability/recovery behavior.
Cross-topic links (placeholders): Clamp & ESD Front-End · ADC Drive & Stability · Isolation / Compliance

Signal + interference model you must accept

Low-frequency bio-potential systems fail when the reality of electrodes, cables, and common-mode coupling is ignored. A compact model is enough to predict the dominant failure modes and to design tests that separate drift from noise.

Electrode model (why “slow steps” happen)

  • Equivalent: signal source + series resistance (Rs) + polarization impedance (Zp(f)).
  • Key fact: Zp(f) changes with frequency, contact pressure, and motion — producing baseline wander and slow recoveries.
  • Practical implication: large low-frequency steps are often artifacts, not random noise; treat them as a system disturbance with recovery constraints.
Quick check: compare PSD and time traces with electrodes connected vs input shorted. If slow behavior disappears on short, the dominant term is electrode/contact + bias network interaction.

50/60 Hz coupling (only three paths matter here)

  1. Capacitive pickup to electrodes/cables (mostly common-mode): large common-mode appears at both inputs before it becomes a differential problem.
  2. Shield/ground potential difference (common-mode → differential conversion): return paths and shield connection points decide how common-mode turns into measured hum.
  3. Input mismatch / impedance imbalance (AC CMRR collapse): small differences in series R/C, leakage, or bias paths can dominate real-world hum rejection.
Working rule: hum rejection is usually limited by impedance balance, not by a headline DC CMRR number. Measure common-mode and differential simultaneously to confirm the dominant path.

Low-frequency noise decomposition (what dominates depends on conditions)

  • Voltage noise (en) + 1/f: dominates when source impedance is moderate and bias networks are well-behaved.
  • Current noise × source impedance (in·Rs): dominates when electrode/source impedance is high or bias resistors are extremely large.
  • Bias network thermal noise: rises with very large resistors; trade noise against input bias current paths and recovery time constants.
  • Leakage/contamination (“pseudo-drift”): often looks like drift or random LF wander; guarding/cleanliness is a first-class design parameter.
Fast separation test (drift vs 0.1–10 Hz noise): • Record a long time trace (minutes), compute PSD for 0.1–10 Hz, and repeat with inputs shorted. • Drift-like behavior shows slow, non-stationary movement; noise shows stationary statistics with a stable PSD shape. • If shorted-input results are already poor, the dominant term is on-board (front end / bias / leakage).
Electrode equivalent and three 50/60 Hz coupling paths Diagram with electrode source models, differential amplifier input, capacitive pickup common-mode injection, shield/ground delta conversion, and impedance mismatch causing AC CMRR collapse, plus a small symptom-to-cause map. Electrode model V Rₛ Zₚ(f) V Rₛ Zₚ(f) Differential Front End AC CMRR OUT Environment 50 / 60 Hz field CM pickup Shield / Ground ΔV along return CM → DM ΔZ mismatch AC CMRR collapse Symptom → likely cause Hum unchanged with RLD RC added → hum worse Touch cable → LF steps
A practical model: most 50/60 Hz energy arrives as common-mode; real-world hum rejection is set by impedance balance and return/shield behavior.

Architecture choices inside the boundary

In low-frequency bio-potential systems, “best” is defined by stability under electrode offsets, impedance imbalance, hum coupling, and motion artifacts — not by a single noise number. Architecture selection must be driven by measurable boundary conditions (source impedance, lowest frequency requirement, headroom, and channel consistency).

Decision inputs (use these to avoid “spec-sheet selection”)

  • Source impedance: Rs range and imbalance class (ΔZ from R/C/leakage asymmetry).
  • Electrode DC offset + artifacts: worst-case baseline shift and step magnitude (saturation risk).
  • Lowest frequency requirement: DC / 0.05 Hz / 0.5 Hz class drives HP vs DC servo strategy.
  • Hum environment: 50/60 Hz common-mode level and whether RLD is available/allowed.
  • Production constraints: channel count, matching/repeatability, and calibration hook budget.
Selection rule: when AC CMRR is limited by impedance imbalance, the priority becomes symmetry + biasing + recovery control; architecture changes alone will not “fix” hum.

Architecture notes (best for / risk / check / fix)

INA (Instrumentation amplifier): Best for: simple single/low-channel-count chains; high input impedance; predictable signal path. Risk: real-world 50/60 Hz is often set by external imbalance; input RC/protection can quietly break symmetry. What to check: ΔZ balance at inputs (including leakage paths); hum delta with RLD on/off; saturation/recovery after steps. Typical fix: enforce symmetric input networks + guarding; manage DC via HP/DC servo to avoid rail-hits; stabilize RLD loop.
Chopper / Auto-zero INA: Best for: ultra-low drift targets and temperature-sensitive baseline stability. Risk: ripple/modulation products can appear as low-frequency beats; filtering/sampling choices can fold artifacts into the LF band. What to check: spectrum for ripple/sidebands; sensitivity to sampling window/averaging length; RLD-on noise change. Typical fix: place ripple suppression where it does not break input symmetry; verify with PSD/time-trace tests before committing.
Discrete differential / FDA-based front end: Best for: multi-channel systems, custom filtering, stronger drive, or differential downstream chains. Risk: complexity and matching drift; easier to create imbalance/leakage paths that dominate AC CMRR and recovery. What to check: thermal gradients + leakage symmetry; output common-mode headroom; step recovery under worst-case electrode offsets. Typical fix: strict symmetry + partitioning + guarding; add calibration hooks only when coefficients stay stable across time/temperature.

DC management choice (HP vs DC servo)

  • Use hard high-pass when very-low-frequency content is not required and large electrode offsets/artifacts would otherwise saturate the chain.
  • Use DC servo when the lowest-frequency content must be preserved but DC offset needs active cancellation to maintain headroom.
  • Always validate recovery using a controlled step (artifact-like disturbance) and measure time-to-within-ε rather than relying on steady-state specs.
Cross-topic placeholder: output loading/ADC drive stability is handled in the dedicated ADC-drive page; this section only ensures headroom and recovery are protected.
Front-end architecture comparison for low-frequency bio-potential chains Three-column comparison card: INA, Chopper INA, and Discrete differential front end, each with Best for, Risk, Check, and Fix, designed as a box diagram. Architecture comparison (inside this page boundary) INA Chopper INA Discrete Diff Diff amp block Chop + ripple Multi op-amp Best for Simple chain Best for Ultra-low drift Best for Multi-channel Risk ΔZ dominates Risk Ripple / beats Risk Complexity Check RLD on/off Check PSD edges Check Leakage Fix Symmetry Fix Ripple ctrl Fix Guarding
Use architecture as a system decision: impedance balance and recovery dominate hum performance, while drift and ripple management decide baseline stability.

Noise & drift budgeting for ultra-low frequency

Budgeting must produce three outputs: input-referred 0.1–10 Hz noise, in-band noise, and drift (µV/°C). The process must also prescribe what to measure so calculation and bench results converge.

Budget outputs (define what “good” means)

  • 0.1–10 Hz input-referred noise: baseline stability and low-frequency usability.
  • In-band noise (e.g., 0.05–250 Hz): overall SNR and usable resolution in the passband.
  • Drift (µV/°C): worst-case baseline movement across temperature and warm-up conditions.
Pass criteria placeholders: • 0.1–10 Hz < X µVrms • In-band < Z µVrms • Drift < D µV/°C • Recovery < T after a defined step

Contributors (what can dominate and when)

  • Front-end en + 1/f: dominates when source impedance is moderate and bias networks are clean/symmetric.
  • in × Rs: dominates when electrode/source impedance is high or bias resistors are extremely large.
  • Bias network thermal noise: increases with large resistors; trades directly against bias paths and recovery time constants.
  • RLD injected noise: can raise in-band noise or introduce “pseudo-LF” components if loop bandwidth/noise is not controlled.
Practical reminder: any element that improves hum (RLD) or prevents saturation (HP/servo) can also change the noise floor and recovery; budgeting must include these interactions.

Budget template (fields to fill before picking parts)

Inputs: Rs range · ΔZ class · expected DC offset · artifact step size Front end: en (white + 1/f) · in · input bias current class Biasing: bias R values · leakage control method · symmetry plan Bandwidth: flow (HP/servo) · notch target (50/60) · fhigh RLD: loop bandwidth class · noise constraint · stability margin target Outputs: 0.1–10 Hz noise · in-band noise · drift · recovery target
Measure-to-close loop: shorted-input noise → add equivalent Rs → enable RLD → enable notch/HP/servo, and verify each step moves the metric in the expected direction.
Noise and drift budget flow for ultra-low frequency bio-potential chains Left-to-right flow: source impedance and imbalance, input network, front-end noise, bandwidth shaping, and output metrics with pass criteria placeholders. Noise & drift budget flow (calculate + measure) Source & electrodes Rₛ + ΔZ Input network Bias + symmetry Front end en + in Bandwidth shaping HP + notch Output metrics 0.1–10 Hz noise In-band noise Drift µV/°C Pass criteria 0.1–10 Hz < X µVrms In-band < Z µVrms Drift < D µV/°C Recovery < T Measure-to-close loop Short input Add Rₛ Enable RLD Enable notch
Budgeting is a closed loop: define output metrics, sum dominant contributors, then validate each block using a controlled measurement sequence.

Input biasing, impedance balance, and why AC CMRR collapses

In low-frequency bio-potential inputs, 50/60 Hz interference usually arrives as common-mode voltage on the cable and body. The practical limit is often CM-to-DM conversion caused by impedance imbalance (ΔR, ΔC, leakage, and routing asymmetry), which can collapse AC CMRR even when the amplifier’s datasheet CMRR looks excellent.

Input biasing loop: noise, leakage sensitivity, and recovery

  • Very large bias resistors create a DC return path, but add thermal noise and increase sensitivity to PCB surface leakage.
  • Leakage and contamination can form unmodeled resistive paths that are both asymmetric and temperature/humidity dependent.
  • Bias time constants influence artifact recovery; slow returns can look like “drift” and worsen baseline settling.
Design intent: keep the bias network symmetrical, measurable, and robust to leakage (guard + cleaning), while meeting the required low-frequency bandwidth and recovery target.

Impedance imbalance → CM-to-DM conversion (the main 50/60 Hz leak mechanism)

  • Common-mode pickup: electric-field coupling injects a similar 50/60 Hz voltage into both input lines and the body.
  • Imbalance (ΔZ): if +IN and −IN do not see the same impedance to reference (R, C, leakage, routing), common-mode produces a residual differential voltage.
  • AC CMRR collapse: the system’s hum rejection is then limited by ΔZ rather than the amplifier’s intrinsic CMRR.
Practical check: if the 50/60 Hz component changes strongly with cable motion, humidity, or component tolerance, ΔZ-driven CM-to-DM conversion is likely the dominant mechanism.

Critical symmetry points (only what matters for this page)

  • Match the input networks: series resistors, bias resistors, and any input RC must be symmetric at the connector-to-front-end boundary.
  • Control parasitic capacitance: keep both inputs in similar routing environments; avoid one trace hugging ground while the other crosses splits/voids.
  • Guard + cleaning for high-Z nodes: guard rings around bias nodes and inputs reduce leakage sensitivity; cleaning reduces humidity-driven drift paths.
  • Protection paths must be symmetric: clamps/ESD devices and their return paths can introduce large asymmetry if only one side sees extra capacitance or leakage.
Field template (minimum): Rs range · ΔZ class · bias R values · input RC symmetry · leakage control method · artifact recovery target.
Impedance imbalance causing AC CMRR collapse via common-mode to differential conversion Two input paths with mismatched RC and leakage show how 50/60 Hz common-mode pickup converts into residual differential interference; side panel lists five common asymmetry sources. AC CMRR collapse: common-mode pickup + ΔZ → residual differential 50/60 Hz E-field pickup → Vcm on body/cable +IN path Rseries (match) Cbias to ref Rbias to ref Leakage / parasitic small but harmful −IN path Rseries (ΔR) Cbias (ΔC) to ref Rbias to ref Routing mismatch ΔC to ground/shield Differential front end Residual Vdm @ 50/60 Top 5 asymmetry sources ΔR (series / bias) ΔC (to ref / shield) Leakage / residue Routing mismatch Protection mismatch Symmetry + guard decide AC CMRR
When hum arrives as common-mode, even small ΔR/ΔC/leakage asymmetry can create a measurable differential residual and collapse practical AC CMRR.

50/60 Hz notch done right (depth, Q, tolerance, recovery)

The goal of a 50/60 Hz notch is not “infinite depth” — it is enough attenuation within budget without degrading recovery, group delay, or robustness to tolerance and temperature. A notch that looks perfect in frequency response can still fail in the field due to step response and mismatch-driven depth loss.

Notch objectives (define success with two constraints)

  • Depth target: residual 50/60 Hz at the output must meet the noise/interference budget under worst-case coupling.
  • Recovery target: artifact-like steps must settle within a defined time-to-within-ε (avoid long tails and ringing).
Trade-off: pushing depth via high Q often increases ringing and group delay; success requires balancing depth and recovery, not maximizing a single metric.

Practical implementations (only the two routes relevant here)

Twin-T / active notch: Best for: fixed 50/60 Hz suppression with a straightforward structure. Risk: depth is highly mismatch-sensitive; temperature/aging can erode depth; excessive Q creates long recovery tails. What to check: depth at the target frequency; center frequency shift; step response ringing and settling time. Typical fix: matched networks; add a trim/calibration hook; constrain Q to meet recovery target.
State-variable notch (SVF): Best for: controlled Q and tunable frequency; easier to compensate drift and tolerance with adjustment hooks. Risk: can be tuned to look great in frequency response while failing transient recovery if Q is pushed too high. What to check: frequency response + step response as a pair; sensitivity to parameter drift across temperature. Typical fix: treat recovery as a hard constraint; keep adjustment ranges bounded and verifiable.

Tolerance and Q: why depth collapses, and why recovery gets worse

  • Depth is mismatch-limited: achievable notch depth is often set by component matching rather than amplifier bandwidth.
  • High Q increases time-domain cost: deeper/narrower notches tend to ring longer and can create long settling tails after artifacts.
  • Calibration hooks matter: trims or digitally-selectable parameters can restore center/depth, but must be validated for stability over time.
Verification focus: always test notch depth and step recovery together; passing only frequency response is not sufficient for field performance.
Notch trade-offs: depth vs recovery/group delay and tolerance-driven depth loss A conceptual plot shows notch aggressiveness increasing depth but worsening recovery/group delay; a bar mini-chart shows tolerance class reducing achievable notch depth. Notch trade-offs: depth, tolerance, and recovery Concept: deeper notch often costs recovery Notch aggressiveness / Q Metric Depth ↑ Recovery ↑ Tolerance → depth loss Tight Medium Loose Match networks / trim hooks Engineering targets Depth Recovery Validation Frequency + step response
A notch must satisfy both depth and recovery targets; mismatch reduces depth while aggressive Q tends to worsen ringing and settling.

RLD (Right-Leg Drive) loop: how it improves, how it bites

RLD reduces 50/60 Hz interference by actively driving the body/cable common-mode toward a quieter reference. It is an effective differentiator, but it is also a feedback loop that can inject noise, destabilize, or worsen recovery when its bandwidth, compensation, and coupling paths are not controlled.

What RLD improves (and what it cannot fix by itself)

  • Improves: reduces Vcm on the body/cable, which lowers the common-mode driving force of 50/60 Hz.
  • Requires: the input network must remain symmetric; otherwise reduced Vcm can still convert to residual Vdm via ΔZ.
  • Must be verified: effectiveness is proven by measuring Vcm reduction and output hum reduction with RLD on/off.
Acceptance test: RLD ON must reduce input Vcm and reduce output 50/60 Hz residual without increasing in-band noise beyond the budget.

Loop essentials: bandwidth, compensation, and output limiting

  • Plant uncertainty: body/electrode impedance varies with contact and motion; loop design must be robust, not optimized for a single condition.
  • Bandwidth control: excessive high-frequency loop gain increases noise injection and stability risk; use compensation to roll off HF gain.
  • Output limiting: series resistance and current limiting improve safety and stability; they also shape the loop dynamics and must be included in verification.
Failure prevention: treat “RLD makes the system noisier” and “RLD oscillates” as design failures that must be caught by on/off A/B tests and step/cable-touch stimuli.

How RLD bites: symptoms → likely causes → quick checks

RLD ON becomes noisier: loop bandwidth too high, noisy RLD amplifier, or asymmetrical coupling/back-injection paths.
Self-oscillation or ringing: insufficient phase margin under uncertain body impedance; compensation/series R not sized for worst case.
Large recovery after touching cable: loop drives large common-mode steps that convert to differential via ΔZ; recovery limited by bias time constants.
No improvement in hum: Vcm not actually reduced (wrong sense point, saturated driver, or open loop); or ΔZ conversion still dominates.
Minimal measurement set: measure input Vcm and output 50/60 Hz residual with RLD OFF/ON; then apply a controlled step/touch stimulus to evaluate recovery and stability.
RLD loop block diagram with stability knobs and failure modes A block diagram shows common-mode sense feeding an RLD amplifier with compensation and output limiting to drive an RLD electrode; side panels list stability knobs and common failure modes. RLD loop: reduce Vcm, but validate noise + stability Electrodes + cable +IN −IN Vcm pickup Front end Diff amp / INA CM sense point RLD loop RLD amplifier Comp / BW limit Series R + limit RLD electrode Drive body CM Stability knobs BW limit / compensation Series R / current limit Failure modes Oscillation / ringing Noise injection
RLD must be treated as a feedback system: verify Vcm reduction, ensure stability under impedance variation, and confirm it does not increase in-band noise.

High-pass / DC servo: removing drift without losing the science

Low-frequency bio-potential chains inevitably face a conflict: large electrode DC offsets and baseline wander can push the front end or ADC into saturation, yet meaningful information may extend close to DC. The correct choice is a system decision: use a high-pass when fast recovery and rail protection dominate, and use a DC servo when near-DC content must be preserved while still preventing saturation.

When a high-pass is the right tool

  • Rail risk is real: electrode DC offset or baseline wander pushes the chain close to saturation or causes clipping.
  • Near-DC content is not required: the minimum frequency to preserve is above the planned corner frequency.
  • Recovery must be predictable: step/artifact recovery time needs to be short and consistent.
Key linkage: the high-pass corner frequency and the artifact recovery time are tied. Raising the corner improves recovery and rail robustness, but removes more low-frequency content.

When a DC servo is the better answer

  • Lower-frequency content must remain: the chain needs to preserve signal down toward DC beyond what a practical high-pass would allow.
  • Saturation still must be prevented: large offsets/drift exist and cannot be ignored.
  • Validation is part of the design: servo behavior after artifacts and saturation must be measured (tail, overshoot, return-to-within-ε).
Engineering risk: servo loops can create long tails after motion-like steps if the time constant is too slow or the injection path is not controlled. Treat recovery as a hard requirement, not a afterthought.

Metrics that must be measurable (template fields)

  • Corner / effective cutoff: high-pass fc or servo equivalent loop corner.
  • Recovery time: time to return within ±ε after an artifact-like step or saturation event.
  • Post-saturation return: no rail lock; no long tail; no large opposite overshoot beyond budget.
  • Baseline residual: remaining baseline wander inside the preserved band stays within the system budget.
Pass criteria placeholders: step recovery < X s · no rail lock · residual baseline < Y · low-frequency content preserved down to fmin.
Decision tree: High-pass versus DC servo for drift removal A decision tree uses inputs such as DC offset, artifact severity, and minimum frequency requirement to select either a high-pass path or a DC servo path; a pass-criteria panel lists recovery and saturation targets. HP vs DC servo: choose by rail risk and lowest-frequency requirement Inputs DC offset / wander Lowest freq (f_min) Rail / headroom risk Need near-DC content? Saturation risk high? YES NO High-pass path Set f_c by f_min Fast recovery Check rail margin DC servo path Set τ by recovery Limiter / reset Validate tails YES NO Pass criteria (placeholders) Step recovery < X s No rail lock Preserve down to f_min
Choose HP when rail protection and predictable recovery dominate; choose DC servo when near-DC content must be preserved while still preventing saturation.

Protection & safety-adjacent input front-end (without killing noise)

This page only covers the minimum necessary protection for low-frequency bio-potential inputs. The core rule is simple: protection must not break symmetry or introduce leakage and capacitance that collapse AC CMRR and drift performance. Treat protection as part of the analog network that must be matched and verified.

The three parameters that matter most

  • Leakage: creates unintended bias paths that look like drift and can be humidity/temperature dependent.
  • Capacitance (Cj / parasitics): adds frequency-dependent imbalance and converts common-mode into differential hum.
  • Mismatch: even small ΔR/ΔC/return-path asymmetry can dominate practical 50/60 Hz rejection.
Selection intent: use low-leakage, low-capacitance elements with predictable behavior, and place them symmetrically with symmetric return paths.

Series R and input RC: protect without collapsing AC CMRR

  • Series R: limits current and works with clamps, but adds thermal noise and must be matched across +IN/−IN.
  • Input RC: can help shape fast transients, but asymmetry directly creates ΔZ and accelerates CM-to-DM conversion at 50/60 Hz.
  • Return paths: clamp return routing must be symmetric; “one-side-to-ground” shortcuts usually backfire as imbalance.
Rule of thumb: any element added to one input must be mirrored on the other input, including the return path and its parasitic capacitance to reference/shield.

RLD interaction (only the key coupling points)

  • Back-injection: clamp capacitance and leakage can couple RLD-driven common-mode changes back into the inputs.
  • Asymmetric protection paths: can convert RLD action into differential noise or longer recovery after touch-like steps.
Verification hint: compare hum and in-band noise with RLD ON/OFF before and after adding protection; unexpected degradation indicates coupling or mismatch.
Minimum symmetric protection set for bio-potential inputs A mirrored protection network shows symmetric series resistors, low-capacitance clamps, and bias return paths feeding a differential front end; a panel highlights sensitive parameters: leakage, capacitance, mismatch. Minimum protection set: symmetric by design Electrodes +IN −IN RLD Symmetric input network Rseries Low-C clamp to return Bias return high-Z Rseries Low-C clamp to return Bias return high-Z Mirror placement + return paths Clamp return symmetric routing Front end INA / diff amp Sensitive parameters Leakage Cj / capacitance Mismatch
Protection should be treated as part of the input impedance network: mirror elements and return paths, and prioritize low leakage, low capacitance, and low mismatch.

Layout & cabling rules for microvolt LF chains (only what matters)

Microvolt low-frequency bio-potential chains fail in repeatable ways: practical 50/60 Hz rejection collapses when symmetry breaks, drift grows when high-impedance nodes are contaminated, and RLD can back-inject noise when routing crosses zones. The rules below are a focused checklist: only what materially moves hum, drift, and touch sensitivity.

Symmetry first High-Z discipline No cross-zone lines Cable mechanics matter

A) Symmetry & return paths (AC CMRR survival)

1) Action: Route +IN and −IN as a mirrored pair with the same environment (length, reference plane, proximity).
Why: ΔZ/ΔC converts common-mode hum into differential error and collapses practical 50/60 Hz rejection.
Quick check: Overlay +IN/−IN routes by layer; verify identical plane reference and no asymmetric neighbors.
Pass criteria: parasitic asymmetry stays within the system budget (ΔC/ΔR placeholder).
Reviewer: MIRROR
2) Action: Keep high-impedance input nodes fully inside the Input High-Z zone; do not cross functional partitions.
Why: crossing zones changes parasitics and return paths, increasing CM→DM conversion and touch sensitivity.
Quick check: mark High-Z nets and confirm they never cross the zone boundary on any layer.
Pass criteria: all High-Z nodes reside in one zone with continuous reference and controlled adjacency.
Reviewer: ZONE
3) Action: Define a single, reviewable shield/return connection point (and keep it consistent across channels).
Why: ambiguous or multi-point shield bonds can create ground-potential coupling that looks like hum or drift.
Quick check: trace the shield path from connector to reference/chassis and confirm one intended bond location.
Pass criteria: shield bond is fixed, documented, and does not create unintended loops near the inputs.
Reviewer: SHIELD

B) Guarding & cleanliness (drift control)

4) Action: Guard all high-impedance nodes with a consistent guard reference; avoid guard breaks and gaps.
Why: high-Z nodes are leakage-dominated; guarding reduces surface leakage and humidity sensitivity.
Quick check: ensure the guard ring fully encloses the node and is not cut by vias, slots, or unrelated nets.
Pass criteria: critical high-Z areas achieve full guard coverage with no exposed contamination paths.
Reviewer: GUARD
5) Action: Treat cleaning and coating as design inputs: keep flux residues away from high-Z zones and validate humidity behavior.
Why: residues and absorbent coatings can create time-varying leakage that presents as slow drift and mismatch.
Quick check: compare drift/hum before and after cleaning; repeat after a humidity soak (A/B test).
Pass criteria: drift and 50/60 Hz residual do not worsen beyond the budget after humidity exposure.
Reviewer: CLEAN

C) RLD isolation (prevent back-injection)

6) Action: Route the RLD drive as a separate zone signal; do not run it parallel to +IN/−IN for long distances.
Why: RLD is an active driver; proximity and parallelism couple energy into inputs and can raise hum/noise.
Quick check: review minimum spacing and parallel run length between RLD and input pair on all layers.
Pass criteria: RLD never enters the Input High-Z zone and does not shadow the input pair.
Reviewer: RLD-ZONE
7) Action: Keep the RLD return path controlled; prevent RLD currents from crossing the input reference region.
Why: return-path crossing modulates the input reference and turns common-mode motion into differential error.
Quick check: identify the intended RLD current loop and confirm it closes inside the RLD zone.
Pass criteria: RLD loop is local; no forced return through high-Z input reference areas.
Reviewer: LOOP
8) Action: Place RLD compensation/limit components near the driver and keep the high-Z sensing node short and guarded.
Why: long RLD sense/comp routes leak driver energy and amplify parasitic coupling back into the inputs.
Quick check: confirm RLD RC/limiter are inside the RLD driver zone and the sensing path is short.
Pass criteria: RLD network stays local with minimal exposed parasitic coupling surfaces.
Reviewer: LOCAL

D) Connectors & cabling actions (touch and motion robustness)

9) Action: Use twisted differential input pair with a shield; keep the shield connection point fixed and documented.
Why: twisting reduces electric-field pickup and keeps coupling symmetric; shield point control prevents ground-coupled hum.
Quick check: verify cable spec and shield termination; compare hum with alternative terminations during debug.
Pass criteria: touch/motion does not shift 50/60 Hz residual beyond budget (placeholder).
Reviewer: TWIST
10) Action: Add strain relief and connector fixation; prevent micro-motion at the termination and at the board entry.
Why: micro-motion changes parasitics and contact conditions, producing step-like artifacts and long recovery tails.
Quick check: bend/tug test the cable while monitoring baseline and hum; compare with a fully fixed harness.
Pass criteria: mechanical disturbance does not trigger artifacts beyond the defined threshold (placeholder).
Reviewer: STRAIN
PCB zoning map for microvolt low-frequency bio-potential chains A PCB top-view diagram partitions the design into Input High-Z zone, Notch and Servo zone, RLD Driver zone, and ADC Interface zone. Red crossed lines mark forbidden cross-zone routes. A do/don’t panel summarizes symmetry, guarding, and strain relief actions. Partition the PCB: keep High-Z, RLD, and ADC interfaces in separate zones Input High-Z zone Connector Guard + Bias Symmetry RC Notch & Servo 50/60 Notch HP / DC Servo Buffer / Output RLD Driver zone Sense Comp Limiter ADC Interface ADC Digital IO Ref / Clock NO Do not cross Do / Don’t Do: mirror routing Do: guard High-Z Do: fixed shield bond Do: strain relief Don’t: cross zones Don’t: parallel RLD Don’t: dirty High-Z
Partition the design into four zones. Keep High-Z inputs isolated, keep RLD local, and keep ADC/digital interfaces out of the High-Z region. Avoid cross-zone routing.

Engineering checklist: bring-up, verification, and production hooks

This section turns the bio-potential low-frequency chain into a repeatable workflow: a bring-up sequence that isolates variables, verification tests with consistent measurement recipes, and minimal production hooks for unit-to-unit consistency. Thresholds are intentionally placeholders (X/Y/Z) and must be set by the system noise and recovery budget.

Bring-up order How to measure Pass criteria Production hooks

A) Bring-up sequence (one variable at a time)

Step 1 — RLD OFF, Notch BYPASS: establish a baseline before adding loop dynamics.
Setup: short/fixture-ready input (or quiet electrode simulator), RLD disabled, notch bypassed.
Measure: DC offset range, rail headroom, wideband noise floor snapshot.
Pass criteria: no rail saturation; baseline RMS noise within budget (placeholder).
Step 2 — 0.1–10 Hz noise recipe: verify low-frequency noise before enabling rejection features.
Setup: keep RLD OFF and notch bypassed to avoid masking fundamental noise issues.
Measure: record a long enough time series (e.g., 2–5 minutes as a starting point); compute band-limited RMS after consistent detrend + band-pass.
Pass criteria: 0.1–10 Hz < X µVrms (X from the noise budget).
Step 3 — RLD ON (A/B compare): validate mains reduction without raising in-band noise.
Setup: same wiring/fixture; only toggle RLD state (OFF → ON).
Measure: compare 50/60 Hz tone and nearbands in the spectrum; check for broadband lift when RLD is enabled.
Pass criteria: Δ(50/60) ≥ Y dB and noise increase ≤ Z (placeholders).
Step 4 — Enable Notch: validate depth, collateral distortion, and interaction with RLD.
Setup: notch ON; keep all other settings unchanged from Step 3.
Measure: notch depth at 50/60; verify no excessive ringing or baseline deformation.
Pass criteria: depth ≥ A dB, and recovery tail ≤ B s (placeholders).
Step 5 — Artifact / step recovery: test touch/motion-like disturbances and measure time-to-recover.
Setup: apply a repeatable disturbance (fixture step / controlled cable perturbation); log time series.
Measure: peak-to-settle time; confirm no sustained oscillation after notch + RLD.
Pass criteria: settle to within C% of baseline within D seconds (placeholders).
Step 6 — Quick temperature / tolerance sweep: confirm the design does not “look perfect at room” only.
Setup: repeat Step 3–4 at two temperatures or a light sweep; re-run depth and recovery checks.
Measure: notch center shift, depth degradation, recovery tail growth, and RLD stability margin indicators.
Pass criteria: center shift ≤ E, depth ≥ F, recovery ≤ G across the sweep (placeholders).

Rule of thumb: if Step 2 fails, do not proceed to loop features (RLD/notch). Fix baseline noise and leakage first.

B) Key verification tests (repeatable measurement recipes)

Test 1 — 0.1–10 Hz noise: measure what the budget is written in.
  • Duration: use a long record; start with 120–300 s and extend if the result is not stable run-to-run.
  • Processing: apply a consistent detrend method, then band-pass to 0.1–10 Hz; compute RMS (and optionally peak-to-peak).
  • Trap: avoid “over-smoothing” that hides 1/f behavior; keep processing identical across boards and test stations.
  • Pass: 0.1–10 Hz < X µVrms.
Test 2 — 50/60 Hz rejection (RLD A/B): verify improvement without side effects.
  • Method: keep everything identical; only toggle RLD OFF vs ON and compare the spectrum.
  • Look for: reduced mains tone and nearbands, plus no broad in-band noise lift.
  • Pass: Δ(50/60) ≥ Y dB and noise increase ≤ Z.
Test 3 — Notch depth & drift: confirm depth, center stability, and recovery tail.
  • Depth: compare 50/60 tone before vs after notch; record depth margin against the budget.
  • Drift: repeat depth/center checks at two temperatures (or a short sweep) and across units.
  • Recovery: apply a step-like artifact and quantify time-to-return to baseline.
  • Pass: depth ≥ A dB, center shift ≤ B, recovery tail ≤ C s (placeholders).

C) Production hooks (minimal, high leverage) + example part numbers

Hooks below are “design-for-test” anchors: they enable fast A/B isolation (RLD on/off, notch on/off, input short/open, known common-mode injection) and allow storing per-unit trim coefficients. Part numbers are starting points; selection must be driven by leakage, charge injection, on-resistance, voltage range, and package constraints of the chain.

Hook 1 — Input short / zero baseline: short the differential inputs to a known state for offset/noise baselining.
Example part numbers:
  • Reed relays (low leakage, great for zeroing): Coto 9007-05-00 / 9007-24-00 (coil voltage variant); Pickering 101-1-A-12/2D; StandexMeder SIL05-1A72-71L.
  • Precision analog switch (solid-state option): TI TMUX1136 (2-channel 2:1 precision switch).
Pass goal: with inputs shorted, measured 0.1–10 Hz noise matches the baseline budget and is repeatable unit-to-unit.
Hook 2 — Known common-mode injection: inject a controlled common-mode stimulus to validate AC CMRR and RLD benefit.
Example part numbers:
  • Analog mux / differential mux for routing stimulus: Analog Devices ADG1208 (8:1 mux) / ADG1209 (4:1 differential mux).
  • Precision switch for clean A/B routing: TI TMUX1136.
Pass goal: injected CM produces bounded DM error; enabling RLD improves the 50/60 line item by ≥ Y dB without broadband penalty.
Hook 3 — Notch trim and coefficient retention: allow per-unit notch centering or depth trim and store coefficients.
Example part numbers:
  • Nonvolatile digital potentiometer (trim element): Analog Devices AD5272 (I²C, NVM).
  • I²C EEPROM (store coefficients, serial, and test metadata): Microchip 24LC02B.
Pass goal: after power-cycle and temperature change, notch center and depth remain within the specified drift window.
Practical selection notes (keep the chain honest)
  • Reed relay vs solid-state switch: reed relays often win for leakage and zeroing; solid-state wins for speed and integration but must be checked for leakage and charge injection in microvolt regimes.
  • Fixture consistency: define the fixture cable (twist/shield/strain relief) and keep it identical across stations; a “better” board cannot overcome an inconsistent harness.
  • Firmware reproducibility: lock the exact processing steps (detrend/filter/RMS) used for 0.1–10 Hz tests and store the version tag alongside the coefficients.
Verification flow strip for bring-up and production hooks A six-step horizontal verification flow for a microvolt low-frequency bio-potential chain. Each step shows toggles, what to measure, and a placeholder pass criterion. A hooks panel highlights input short, common-mode injection, and nonvolatile coefficient storage. Verification flow (Step 1 → Step 6): toggle one feature at a time Step 1 RLD OFF Notch BYP Baseline Pass: OK Step 2 Noise 0.1–10 Hz Record Pass: < X Step 3 RLD ON A/B compare Mains Pass: ≥ Y Step 4 Notch ON Depth + tail Depth Pass: A/B Step 5 Artifact Recovery Settle Pass: ≤ C Step 6 — Temp / tolerance quick sweep Pass: shift ≤ E, depth ≥ F, tail ≤ G Production hooks Input short CM injection NVM coeff
Use the flow as a station script: Step 1–2 validate baseline noise first; Step 3–4 quantify hum reduction and notch behavior; Step 5 confirms recovery; Step 6 catches temperature/tolerance sensitivity early.

Request a Quote

Accepted Formats

pdf, csv, xls, xlsx, zip

Attachment

Drag & drop files here or use the button below.

FAQs (10–12) — short, actionable, and testable

Each answer uses a fixed 4-line structure to close long-tail bring-up and verification questions without expanding page scope.

Why does the baseline drift for minutes after power-up even with a “low-drift” front end?
Likely cause: high-Z leakage/absorption and bias-network settling dominate early time; thermal gradients and servo/chopper startup can add long tails.
Quick check: short inputs (relay/switch) and log drift slope vs time; compare “shorted” vs “electrodes/cable” to separate board leakage from electrode effects.
Fix: clean + guard high-Z nodes, reduce exposed high-Z area, define warm-up and optional input-short during warm-up; reset/bypass servo on startup if present.
Pass criteria: after Y minutes, drift slope X µV/min and repeatable across power cycles (placeholders).
Why does touching/moving the electrode cable create large slow jumps (not just spikes)?
Likely cause: cable triboelectric effects and changing parasitics create common-mode shifts that convert to differential via impedance imbalance; high-Z nodes amplify slow “recharge” behavior.
Quick check: repeat the motion with (a) cable strain-relieved and fixed, (b) alternative cable/harness, and (c) RLD OFF vs ON; log step amplitude and settling time.
Fix: use twisted pair + shield low-tribo cable, add strain relief, keep shield termination consistent, and enforce mirrored input RC/geometry to prevent CM→DM conversion.
Pass criteria: standardized cable motion causes baseline step X µV and settles < Y s (placeholders).
RLD is enabled but 50/60 Hz barely improves—what is the first impedance check?
Likely cause: +IN/−IN impedance mismatch (RC, bias path, protection symmetry, electrode contact) collapses AC CMRR, so RLD cannot “pull down” the residual differential hum.
Quick check: measure/verify matched input RC values and placement symmetry; swap left/right electrodes/cables and see if the residual follows the swap.
Fix: make the entire input network mirror-symmetric (R, C, layout, return), and ensure RLD senses the true common-mode node (not a mismatched point).
Pass criteria: with RLD ON, mains tone improves by ≥ Y dB vs RLD OFF with in-band noise increase ≤ Z (placeholders).
Why does CMRR look great in the datasheet but collapse on my board at 50/60 Hz?
Likely cause: board-level AC CMRR is set by impedance matching, parasitic capacitance asymmetry, return paths, and contamination—not the IC’s typical DC CMRR spec.
Quick check: compare hum with inputs shorted vs real cable/electrodes; verify ΔR/ΔC symmetry and confirm a single, intentional shield bond location.
Fix: mirror routing and component placement, guard/clean high-Z areas, avoid cross-zone traces, and keep cable/shield termination consistent and documented.
Pass criteria: board-level 50/60 residual stays within budget with representative cable; channel-to-channel residual difference < X dB (placeholders).
Notch depth looks great on the analyzer, but in real signals it’s shallow—why?
Likely cause: the interference is not exactly at the notch center (frequency drift/sidebands), or hum has already become differential due to imbalance, making “notch-only” suppression ineffective.
Quick check: log the actual tone frequency in the spectrum, sweep/estimate notch center, and compare notch ON vs BYPASS while checking for front-end overload/saturation.
Fix: fix AC CMRR first (symmetry), add a notch centering/trim hook if needed, and avoid extreme Q if center uncertainty is high.
Pass criteria: attenuation at the actual tone frequency ≥ A dB across temperature with center shift ≤ B (placeholders).
Why does a deep notch cause long recovery after motion artifacts?
Likely cause: high notch Q increases group delay and ringing; artifacts can drive internal nodes into nonlinear regions, stretching the tail (especially when combined with HP/DC servo).
Quick check: inject a repeatable step-like artifact and measure settle time with notch BYPASS vs notch ON; compare different Q (or component sets).
Fix: reduce Q/steepness, add an overload bypass/limiter strategy, and retune servo time constants to avoid long post-artifact tails.
Pass criteria: return within C% of baseline in < D seconds after the standardized artifact (placeholders).
Why does adding an input RC reduce noise but worsen hum pickup?
Likely cause: RC mismatch and asymmetric parasitics convert common-mode hum into differential error; the RC may also shift the effective impedance balance at 50/60 Hz.
Quick check: measure actual R/C values, swap to tightly matched components, and compare mains residual before/after while keeping wiring identical.
Fix: enforce matched RC (same package, symmetric placement), minimize parasitic asymmetry, and move filtering to a more controlled differential stage when possible.
Pass criteria: noise reduction achieved with mains residual degradation < X dB (placeholder) and no new beat/sideband artifacts.
How do I separate 0.1–10 Hz noise from true offset drift in a quick bench test?
Likely cause: drift is a trend/step (time constant), while 0.1–10 Hz noise is a stationary statistic; inconsistent processing makes drift look like noise (or the opposite).
Quick check: compute 0.1–10 Hz RMS after a consistent detrend; repeat with two record lengths—noise RMS should be stable, drift slope/step should not vanish.
Fix: if drift dominates, attack leakage/cleanliness/bias return paths; if noise dominates, optimize en/1/f, bandwidth, and source impedance; keep processing fixed across tests.
Pass criteria: run-to-run noise metric variation < X% and drift slope < Y µV/min over the defined window (placeholders).
Why do chopper/auto-zero front ends show “ripple” or strange low-frequency beats?
Likely cause: residual chopping ripple and mixing/aliasing with sampling, notch, or interference creates beat products that land in the low-frequency band.
Quick check: inspect spectrum for chopping-related lines/sidebands; change sample rate or a filter corner—true beats typically shift predictably.
Fix: retune post-filtering and sampling strategy, avoid placing notch/filters where they amplify beat perception, and select chopper settings that keep ripple outside the critical band.
Pass criteria: ripple/beat components contribute < X µVrms to 0.1–10 Hz and no sustained low-frequency oscillation is observed (placeholders).
Why does electrode DC offset push the output into saturation even with high CMRR?
Likely cause: high CMRR does not guarantee sufficient input common-mode range or headroom; electrode DC plus bias return paths can shift operating points into rails.
Quick check: measure input common-mode vs device ICMR and output headroom; short inputs to confirm saturation is driven by DC offset rather than noise.
Fix: add HP or DC servo to remove DC while preserving band of interest, provide a correct bias return path, and set output common-mode for margin.
Pass criteria: no rail saturation under worst-case DC offset; overload recovery < X s with defined stimulus (placeholders).
Why does RLD sometimes increase noise or cause oscillation in certain subjects/cables?
Likely cause: RLD loop stability depends on variable body/electrode impedance; coupling from the RLD drive into the high-Z inputs can create noise injection or oscillation.
Quick check: monitor RLD output for oscillation, compare cable lengths/types, and A/B test RLD OFF vs ON while tracking in-band noise and mains tone.
Fix: reduce RLD loop gain/bandwidth, add compensation and output limiting (series R/RC), and keep RLD routing isolated from input pairs.
Pass criteria: stable for the defined impedance/cable set; RLD ON increases in-band noise by < X and improves mains by ≥ Y dB (placeholders).
What’s the simplest production hook to verify hum rejection without a human subject?
Likely cause: hum rejection is not verifiable in production without a repeatable input state (short/open) and a controlled common-mode stimulus for A/B comparison.
Quick check: implement a station script: Input short → RLD OFF/ON spectrum compare → Notch OFF/ON depth check; store results per unit.
Fix: add input-short hardware (e.g., Coto 9007-05-00 reed relay or TI TMUX1136), CM routing (ADI ADG1209/ADG1208), and NVM retention (Microchip 24LC02B; optional trim via ADI AD5272).
Pass criteria: station A/B shows Δ(50/60) ≥ Y dB and notch depth ≥ A dB; coefficients persist after power cycle (placeholders).
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.