Bioinstrumentation — ECG, EEG, Biosensors — Engineering Reference

Bioinstrumentation — ECG, EEG, Biosensors

1. At a glance

Bioinstrumentation is the electronic measurement of physiological signals from living tissue and their conversion into displayable, recordable, or actionable data. The discipline brings together six subsystems — transducer / electrode, input protection and isolation, analog signal conditioning (instrumentation amplifier, filters), analog-to-digital conversion, digital signal processing (artefact rejection, feature extraction), and a regulated user interface (display, alarm, network) — and operates under a non-negotiable safety envelope set by IEC 60601 and equivalent international standards. Everything from a USD 30 fingertip pulse oximeter to a USD 250 000 PET-MRI scanner is composed of the same building blocks, scaled in noise floor, channel count, and bandwidth.

The standard physiological modalities are: ECG (electrocardiogram, heart, 1 mV, 0.05–150 Hz), EEG (electroencephalogram, brain, 10–100 µV, 0.5–100 Hz), EMG (electromyogram, muscle, 10 µV – 5 mV, 10–2000 Hz), EOG (electrooculogram, eye, 50–3500 µV, 0.1–30 Hz), GSR/EDA (galvanic skin response / electrodermal activity, 1–50 µS, DC–3 Hz), PPG (photoplethysmography, optical pulse, basis of SpO₂), respiration (impedance pneumography or strap), core temperature (NTC thermistor, IR tympanic, ingestible pill), arterial blood pressure (non-invasive oscillometric cuff, invasive arterial line, tonometry), blood glucose (electrochemical strip, fluorescent CGM), and motion / posture / activity (MEMS IMU). Outside the body sits the implantable / interventional set: pacemakers, ICDs, cochlear implants, deep-brain stimulators, microelectrode arrays (Utah, Michigan, Neuralink threads), ECoG grids, and a growing class of bioelectronic medicines (vagus nerve, sacral, hypoglossal stimulators).

Markets in 2026 break into four bands. Hospital monitoring (GE CareScape, Philips IntelliVue, Mindray, Dräger) is the regulated, high-margin core. Wearables and consumer health (Apple Watch ECG and SpO₂, Fitbit, Whoop, Oura, Garmin, Polar) brought millions of clinical-grade sensors into pockets. Point-of-care diagnostics (Abbott i-STAT, BioFire, Cepheid GeneXpert, Inflammatix, Roche cobas) decentralised lab medicine. Brain-computer interfaces (Neuralink, Synchron, Blackrock Neurotech, Paradromics, Precision Neuroscience, BrainGate consortium) commercialised intracortical recording in the early 2020s and are scaling channel counts past 10⁴ as of 2026. Continuous glucose monitoring (Dexcom G7, Abbott FreeStyle Libre 3, Medtronic Guardian 4) and closed-loop insulin delivery (Tandem Control-IQ, Medtronic 780G, Beta Bionics iLet) are the most clinically successful closed-loop bioelectronics deployed at consumer scale.

This note covers electrode physics, the canonical signal-conditioning chain, the dominant integrated bio-AFEs, the major modalities, and the safety / regulatory envelope. See [[Engineering/op-amps]] for in-amp internals, [[Engineering/signal-processing-dsp]] for digital filtering and FFT, and [[Engineering/semiconductor-devices]] for the ADC building blocks.

2. Why it matters

Physiological signals are tiny (microvolts), slow (DC to a few hundred hertz for biopotentials, very low frequency for some), and buried in environmental and physiological noise orders of magnitude larger. The dominant interferers are 50/60 Hz mains coupling, capacitive coupling from cabling and clothing, motion artefact from electrode movement, electrode polarisation drift, RF from cellular and Wi-Fi radios in the same enclosure, and physiological “noise” that is signal from the wrong organ (cardiac on EEG, respiration on ECG, blink on EOG). Op-amps, in-amps, ADCs, and DSP work together to extract diagnostic-grade information from this; getting any layer wrong yields false alarms, missed arrhythmias, mis-diagnosis, or — in closed-loop systems — incorrect therapy delivery.

The medical-device safety envelope is unforgiving. IEC 60601-1 classifies applied parts as type B (no patient connection), BF (patient connection, floating, not direct cardiac), or CF (direct cardiac, e.g. catheter electrode) and sets corresponding leakage-current and dielectric-strength limits. CF requires < 10 µA patient leakage under single-fault conditions; the only way to achieve this with mains-powered equipment is galvanic isolation between the patient side and everything else. IEC 62304 layers a software-lifecycle obligation on top: Class A (no injury possible), B (non-serious injury), or C (death or serious injury). Class C is where pacemakers, infusion pumps, ventilators, and closed-loop insulin pumps live and where the validation rigour starts to approach DO-178C-level avionics practice.

3. First principles

Physiological signal amplitudes and bandwidths

Modality	Typical amplitude	Bandwidth	Notes
ECG (12-lead diagnostic)	0.5–4 mV peak	0.05–150 Hz	IEC 60601-2-25 sets diagnostic mode 0.05 Hz; monitor mode 0.5 Hz
ECG (monitor / wearable)	0.5–2 mV peak	0.5–40 Hz	High-pass relaxed for less drift; loses ST-segment fidelity
EEG (clinical scalp)	10–100 µV	0.5–100 Hz	10-20 system electrode placement
EEG (intracranial / ECoG)	50 µV – 5 mV	0.5–500 Hz	Higher SNR than scalp; high-frequency bands accessible
Single-unit (microelectrode, action potential)	50–500 µV	300 Hz – 5 kHz	After 300 Hz HPF; spike sorting downstream
EMG (surface, sEMG)	50 µV – 5 mV	10–500 Hz	Wider for fast-twitch and gesture interfaces
EMG (intramuscular, iEMG)	100 µV – 10 mV	10–2000 Hz	Needle / fine-wire; clinical diagnosis
EOG	50–3500 µV	DC – 30 Hz	DC-coupled for slow eye position
GSR / EDA	1–50 µS conductance	DC – 3 Hz	Tonic + phasic (SCR)
PPG (SpO₂)	AC ~1% of DC	0.5–10 Hz	Pulsatile component is small fraction
Respiration (impedance)	0.5–5 Ω on 1 kΩ baseline	0.05–2 Hz	Tidal volume correlates with impedance change
Body temperature	30–42 °C	sub-Hz	NTC, RTD, IR, or ingestible
Blood glucose (CGM)	40–400 mg/dL	1/300 Hz (5-min sample)	Interstitial fluid lags blood by 5–15 min

These ranges are the design contract: the AFE must resolve the bottom of the signal range above noise floor while not clipping on the top of the range plus electrode-offset headroom.

Electrode–tissue interface

The metal–electrolyte interface is electrochemical, not purely resistive. A silver / silver-chloride (Ag/AgCl) electrode in contact with chloride-containing electrolyte (sweat, conductive gel, interstitial fluid) forms a reversible half-cell:

AgCl(s) + e⁻ ⇌ Ag(s) + Cl⁻(aq), E° ≈ −0.22 V vs SHE

Because it is reversible, current can flow with minimal polarisation overpotential — the defining property of a non-polarisable electrode and the reason Ag/AgCl dominates biopotential measurement. The interface model (Randles cell) is a charge-transfer resistance R_ct in parallel with a double-layer capacitance C_dl, in series with a solution resistance R_s. Typical magnitudes for a 1 cm² wet gel electrode: R_s = 100 Ω – 10 kΩ, R_ct = 10 kΩ – 1 MΩ, C_dl = 1–100 µF. At biopotential frequencies (DC – 100 Hz) the impedance is dominated by R_ct and C_dl and is moderately frequency-dependent.

Two electrodes in a differential pair never have identical half-cell potentials: the resulting DC offset is typically 100–300 mV and drifts on a timescale of seconds to minutes. This offset is the largest DC component in the signal path and sets the headroom requirement for the in-amp — any AFE that DC-couples electrodes to a high-gain in-amp will saturate. The solutions are AC coupling (high-pass at 0.05–0.5 Hz), DC servo (integrator feeds the offset back into the in-amp reference), or dedicated bio-AFEs with high-input-range first stages that swallow the offset.

Tissue impedance seen between two skin electrodes is dominated by the stratum corneum (10–100 kΩ·cm² dry, falling 10× with hydration or abrasion) at low frequencies, transitioning to bulk-tissue impedance (10–1000 Ω·cm) above ~1 kHz. The dominant capacitive term across the stratum corneum (~50 nF/cm²) gives biopotential measurement its characteristic high source impedance at low frequencies, which is why bias current of the in-amp matters: I_B = 1 nA into 100 kΩ skin generates 100 µV offset — half the EEG signal range.

Common-mode rejection and the body antenna

A patient connected to two electrodes sits in a 50 Hz (Europe / Asia) or 60 Hz (Americas) electric field from the building wiring. The body acts as an antenna; the common-mode voltage induced is typically 1–10 V_pp at mains frequency, sometimes worse near power supplies, motors, or fluorescent ballasts. The wanted differential signal is microvolts to millivolts. Rejection of the common-mode interference by ≥ 10⁴–10⁶ is mandatory.

The instrumentation amplifier’s intrinsic CMRR of 100–120 dB at DC accomplishes most of this, but only when (i) the source impedances at the two inputs are matched (mismatch converts common-mode to differential at a rate set by the impedance imbalance), and (ii) a driven right leg (DRL) circuit actively cancels common-mode by feeding back the inverted sum of the two inputs through a current-limited buffer to a third reference electrode on the patient. DRL adds 20–30 dB of effective CMRR at mains frequency and is standard on every clinical ECG and EEG front-end.

Galvanic isolation

Patient-side electronics must be galvanically isolated from mains-referenced electronics — the leakage path from a mains transformer winding capacitive coupling through the AFE and out the patient electrode is the safety hazard IEC 60601 was written to eliminate. Isolation technologies in current use:

• Optical (LED-photodiode + linear receiver, e.g. HCPL-7800, IL300, ACPL-C87) — analog or digital; 3–7 kV isolation; the historic baseline.
• Capacitive (silicon-dioxide barrier with on-chip transmitter / receiver — TI ISO77xx, ISO12xx, AMC131M03) — modern monolithic, 5–10 kV reinforced isolation, digital or sigma-delta-modulated analog.
• Magnetic / iCoupler (transformer-coupled, ADI’s iCoupler line: ADuM4xxx, ADuM1xx) — 5–7 kV; well-suited to combined isolated power + signal.
• Galvanic isolation power is the other half: an isolated DC-DC converter (transformer-based, often 1:1 with ~1 W output) feeds the patient-side rail. Reinforced isolation parts: TI UCC25800-Q1, SN6505A, Murata NMV modules, Recom RxxPxx modules.

Defibrillation protection

ECG and biopotential inputs in a hospital must survive a defibrillator shock applied to the patient — up to 5 kV, 50–360 J delivered through the same chest the electrodes are on. Standard front-ends protect with: (i) gas-discharge tube or spark gap across the input pair (clamps at ~70 V), (ii) series resistor (1–10 kΩ) limiting current into (iii) silicon clamp diodes or TVS array to the local rails, (iv) optional fuse (one-shot, in implant or ambulatory designs). The AFE must recover within 5 s of the shock per IEC 60601-2-27 — chopper-stabilised front-ends with their long auto-zero recovery ([[Engineering/op-amps]] section 10c) are unsuitable here.

Stochastic noise floor

Thermal (Johnson) noise from source impedance, shot noise from bias current, and 1/f flicker noise from the input stage compose the AFE noise floor. For a 10 kΩ source at body temperature (310 K), Johnson noise density is √(4·k·T·R) = √(4·1.38e-23·310·10e3) = 13 nV/√Hz; integrated over a 100 Hz ECG bandwidth (noise BW ≈ 1.57·100 = 157 Hz) → 163 nV RMS. A good biopotential AFE adds ~50 nV/√Hz of input-referred amplifier noise (input-referred RMS ~625 nV over 157 Hz), so the source-thermal and amplifier contributions add in quadrature to roughly 650 nV RMS — under a microvolt, easily below the 10 µV EEG signal range and far below the 1 mV ECG range.

4. Signal conditioning chain (canonical)

The standard biopotential front-end, source-to-display:

1. Electrode + lead. Ag/AgCl with hydrogel for clinical, dry textile / carbon-polymer for wearables, capacitive non-contact for through-clothing; gold or platinum-iridium for implants. Lead wires shielded and twisted; carbon-impregnated for low triboelectric noise.
2. Input protection. Spark gap or gas-discharge tube for defib survival; TVS / Zener clamps to the patient-side rails; current-limit series resistor (1–10 kΩ); optional input fuse for ambulatory.
3. Instrumentation amplifier. Three-op-amp monolithic in-amp — TI INA826, INA333 (zero-drift), ADI AD8429 (low-noise), AD620 (classic) — set for moderate gain (10–100) at this stage; CMRR ≥ 100 dB at DC, ≥ 80 dB at 60 Hz.
4. Driven right leg. Inverting buffer summing both inputs and driving a current-limited (typ. 1 µA max) third electrode; raises effective CMRR by 20–30 dB at mains frequency.
5. DC offset rejection. Either AC coupling at the in-amp reference (0.05–0.5 Hz high-pass), or DC servo (integrator output drives the in-amp’s REF pin to cancel offset), or dedicated bio-AFE first stage with ±300 mV input range.
6. Anti-alias filter. Active 2nd–4th order Bessel low-pass at ~½ to ⅓ of f_s (Bessel for linear phase preservation); on Σ-Δ AFEs the on-chip decimator handles this and an external 1st-order RC suffices.
7. ADC. 24-bit sigma-delta is the dominant choice for biopotentials — TI ADS1292 (2-ch ECG), ADS1298 / ADS1299 (8-ch ECG / EEG), ADS131M0x, ADI AD7124-8 (24-bit, 8-ch, integrated PGA), Maxim MAX30001 (single-channel biopotential + BioZ + PPG); 12–16 bit SAR for less-demanding modalities; oversampling 12-bit MCU ADCs for hobby-grade.
8. Digital filtering. 50/60 Hz notch (IIR with adaptive frequency tracking), low-pass for display bandwidth, high-pass to baseline. See [[Engineering/signal-processing-dsp]].
9. Feature extraction. Pan–Tompkins (1985) R-peak detection for ECG; band-power (delta / theta / alpha / beta / gamma) for EEG; envelope + RMS for EMG.
10. Isolation barrier. Optical, capacitive, or magnetic — between the patient-side digital domain and the host-side processor (display, network, storage).
11. Communication. Bluetooth Low Energy 5.x (most wearables), Wi-Fi (hospital monitors with HL7 / IHE PCD-01 backhaul), USB (lab equipment), MICS/MedRadio 401–406 MHz (implants), proprietary 2.4 GHz for some hearing aids.
12. Display, alarm, and storage. IEC 60601-1-8 governs alarm priority and acoustic / visual indication; HL7 FHIR for EHR integration; DICOM-Waveform for diagnostic ECG storage.

5. Practical math + worked examples

Driven right leg (DRL) — the math

DRL is a single op-amp (sometimes two) that takes the common-mode component of the two biopotential inputs — typically the midpoint of the in-amp’s first-stage outputs, or an explicit (V₁ + V₂)/2 from two summing resistors — inverts it with high gain (typically 100×, set by R_f/R_in), and drives the result through a current-limiting resistor R_o (typically 100 kΩ – 1 MΩ) into a third reference electrode on the patient. The closed-loop gain from common-mode source to patient body is the in-amp’s CMRR multiplied by the DRL loop gain, minus the body-electrode impedance contribution.

The current-limit resistor R_o serves two purposes: (i) safety — under single-fault conditions the maximum injectable current is V_rail/R_o, kept below the IEC 60601 leakage limit (10 µA for type CF); (ii) stability — the DRL loop must remain stable in the presence of the patient’s body capacitance to ground (typ. 100 pF – 1 nF coupled to mains earth through dielectric of clothing, chair, bed). The standard design constraint is R_o · C_body well below the in-amp’s small-signal bandwidth, so the loop crosses unity gain with > 45° phase margin. Modern implementations: the AD8232 integrates a “right-leg drive” amplifier on-chip with selectable R_o.

Filtering bandwidth choices — ECG diagnostic vs monitor

Per IEC 60601-2-25 (diagnostic ECG), the high-pass corner is 0.05 Hz — relaxed enough to preserve the slow ST-segment shape that is diagnostic for myocardial ischaemia. Per IEC 60601-2-27 (monitor ECG), the corner is 0.5 Hz — adequate for rhythm analysis, much faster recovery from baseline wander but loses ST-segment fidelity (ST elevation / depression measured against the PR baseline becomes filter-shape-dependent). A modern monitor that claims diagnostic-grade ST analysis must switch to 0.05 Hz dynamically when the user enables “diagnostic mode” — and then accepts the longer recovery from motion artefact.

The low-pass corner per −2-25 is 150 Hz (preserves the high-frequency content of the QRS complex relevant to bundle-branch and conduction-block diagnosis); −2-27 monitor is 40 Hz (rejects EMG / muscle tremor at the cost of QRS detail).

Example A — ECG front-end noise budget (TI ADS1298)

Target: 1 mV peak QRS, SNR ≥ 60 dB → noise floor < 1 µV RMS over the ECG diagnostic bandwidth (0.05–150 Hz).

The ADS1298 specifies input-referred noise of 3 µV_pp in a 150 Hz bandwidth at gain 6, which converts to ≈ 0.5 µV RMS (peak-to-peak / 6.6 for Gaussian). Electrode and source thermal noise: assume 50 kΩ skin impedance (gelled disposable Ag/AgCl) at 310 K → e_n,source = √(4·k·T·R) = 28 nV/√Hz. Over noise bandwidth 1.57·150 ≈ 235 Hz: 28e-9 · √235 = 0.43 µV RMS.

Total (in quadrature): √(0.5² + 0.43²) = 0.66 µV RMS — below the 1 µV target with margin. SNR = 20·log(1 mV / 0.66 µV) = 63.6 dB. Specification met.

If the source impedance climbs to 500 kΩ (poor skin prep), source thermal noise rises to 88 nV/√Hz · √235 = 1.35 µV RMS — now total noise is 1.45 µV RMS, SNR = 56.8 dB, and the design fails. Skin prep (alcohol wipe, gentle abrasion) is part of the design.

Example B — EEG common-mode immunity

A 50 Hz mains common-mode voltage of 1 V_pp couples onto the patient. Signal of interest is a 100 µV alpha rhythm (8–13 Hz). Required CMRR at 50 Hz to keep the residual mains < 1 µV (1/100 of signal):

CMRR_required = 20·log(1 V / 1 µV) = 120 dB

A standalone in-amp like AD8429 delivers 90 dB CMRR at 50 Hz (degraded from 130 dB at DC). The shortfall of 30 dB is exactly what a DRL circuit provides — the active common-mode cancellation adds 20–30 dB at mains frequency. Real EEG systems also place a 50/60 Hz IIR notch in the digital path as a final safety net. With DRL (+25 dB) plus the in-amp’s 90 dB = 115 dB analog CMRR, plus the digital notch (typ. 30–40 dB rejection at the notch centre, well below 5 dB ripple in passband), the residual is well under 1 µV. The digital notch alone is insufficient — the in-amp would saturate or its dynamic range would be consumed by the 1 V mains before the ADC ever saw it.

Example C — Pulse oximetry SpO₂ ratio computation

Pulse oximetry uses two LEDs — red 660 nm and infrared 940 nm — alternately illuminating perfused tissue (fingertip, earlobe). Beer-Lambert through tissue:

I(λ) = I₀(λ) · exp(−α(λ) · d)

where α(λ) is the wavelength-dependent absorption coefficient and d is path length. The AC component (pulsatile, from arterial blood expansion) carries oxygenation; the DC component (steady, from skin, bone, venous and capillary blood) normalises out path length and tissue variability.

R = (AC_red / DC_red) / (AC_IR / DC_IR)

The empirical Masimo / Nellcor calibration is approximately SpO₂ ≈ 110 − 25·R, with vendor-specific curves stored in the sensor or instrument. At R = 0.5 → SpO₂ ≈ 97% (healthy room air). At R = 0.85 → SpO₂ ≈ 89% (hypoxic). At R = 2.0 → SpO₂ ≈ 60% (severe; instrument typically saturates the displayed value at 70%).

The AC fraction is ~1% of DC, so the photodiode must resolve ~14 bits of dynamic range to deliver 1% SpO₂ resolution. Modern integrated PPG AFEs (TI AFE4404, AFE44I0, Maxim MAX30101, MAX86150) include LED drivers, photodiode TIA, ambient-light cancellation, and 22-bit ADC in one package. Motion artefact is the dominant clinical limitation — shake corrupts the AC envelope; Masimo SET (1995) introduced adaptive filtering using a reference noise model that revolutionised motion-tolerant pulse oximetry and remains the clinical gold standard.

Example D — Photodiode TIA for PPG / SpO₂

A typical reflective-mode PPG photodiode (Vishay VEMD8080, OSRAM SFH 2440, broadband 400–1100 nm with peak at 900 nm) produces a DC photocurrent of ~10 µA under 10 mA LED forward current at the wrist, with an AC pulsatile component of ~100 nA (1% modulation) at the heart-rate frequency band 0.5–4 Hz.

Two-stage approach is standard: (i) TIA stage with R_f ≈ 100 kΩ → V_out,DC = 10 µA × 100 kΩ = 1.0 V, V_out,AC = 10 mV pulsatile; (ii) AC-coupled second stage with high-pass at 0.3 Hz and gain 100 → AC swing of 1 V is comfortably digitisable by a 14-bit ADC.

Alternative architecture used by modern integrated AFEs (TI AFE4404, Maxim MAX30101, ADI ADPD188): time-domain multiplexing — the LED is pulsed at 100–1000 Hz with the photodiode current integrated only during the ON pulse, and an “ambient” sample is taken during the OFF pulse and subtracted. The technique rejects ambient (sunlight, fluorescent) light by 60+ dB and is the reason wrist PPG works in daylight.

Stability of the TIA in a wearable: photodiode junction capacitance is 30–100 pF; combined with the 100 kΩ feedback gives the same feedback-loop pole problem analysed for op-amps in [[Engineering/op-amps]] worked-example 2. Standard fix: 0.5–2 pF feedback capacitor; bandwidth is bounded to ~kHz, well above the PPG signal but well below the LED pulse modulation if used.

Example E — Sigma-delta ADC oversampling SNR for biopotentials

A 24-bit ΣΔ ADC like the ADS1298 does not actually deliver 24 bits of noise-free resolution; the noise-free bit count at 500 Hz output rate, PGA gain = 6 is about 19.5 ENOB (effective number of bits) per the datasheet. That is 2¹⁹·⁵ ≈ 750 000 distinct codes over the ±2.4 V / 6 = ±400 mV input range — LSB ≈ 1.1 µV. Good enough to resolve a 10 µV EEG signal with ~9 levels of digital headroom.

Doubling the output decimation rate (i.e. halving f_out) gives +0.5 ENOB per the ΣΔ modulator order — the modulator OSR is fixed but the decimation filter trades bandwidth for noise. A wearable that needs only 250 Hz output (well above 100 Hz physiological bandwidth) gains another 0.5 ENOB ≈ 18 % lower noise vs the 500 Hz setting — useful margin for low-amplitude signals.

6. Bioelectrodes

Surface (skin) — non-invasive

• Disposable hydrogel Ag/AgCl — pre-gelled adhesive patch, the workhorse for clinical ECG, EEG, EMG, ICU/ED monitoring. Vendors: 3M, Ambu, Vermed, ConMed. Useful life 24–72 h; gel dries out beyond.
• Reusable cup electrode (EEG) — gold-plated cup with conductive paste (Ten20, EC2); applied with collodion or tape; long-recording EEG (sleep, epilepsy monitoring).
• Dry electrode — textile (silver-coated fabric, conductive yarn), stainless-steel pins on a headset, conductive elastomer. Higher impedance (100 kΩ – 1 MΩ) than wet, but no gel mess and acceptable for wearable ECG (Apple Watch ECG, AliveCor KardiaMobile, Whoop strap PPG/ECG, Polar H10 chest strap).
• Capacitive / non-contact — through-clothing measurement using a guarded high-impedance buffer; SNR lower but mechanically convenient (automotive seat ECG, smart-textile applications). Reference: Plessey EPIC sensors.

Subcutaneous / intramuscular — minimally invasive

• Needle EMG — concentric or monopolar steel needle inserted into a muscle belly; clinical neuromuscular diagnosis. Tightly localised pickup.
• Fine-wire EMG — insulated wire with bared tip inserted via hypodermic; long-recording motion studies, gait analysis.
• Tunnelled subcutaneous loop (Medtronic Reveal LINQ, Abbott Confirm Rx, BIOTRONIK BIOMONITOR) — implanted 2–3 cm under chest skin for years-long cardiac arrhythmia monitoring.

Microelectrode arrays (intracortical)

• Utah Array (Blackrock Neurotech) — 96 silicon shanks 1.5 mm long, 400 µm pitch, 4×4 mm footprint, sputtered platinum or iridium oxide tips. The legacy and clinical workhorse — used in BrainGate, Neuralink (pre-thread design), Synchron’s early work, dozens of academic groups.
• Utah Slanted Electrode Array (USEA) — 100-channel array with graduated shank lengths, for peripheral nerve.
• Michigan probe (NeuroNexus) — silicon shank with multiple in-line recording sites along its length; high spatial resolution along a single axis.
• Neuropixels (IMEC) — single-shank CMOS probe with 384 simultaneously recordable channels out of 960 sites, sub-20 µm pitch; the modern systems-neuroscience workhorse.
• Neuralink threads — flexible polymer thread, 16 electrodes per thread × 64 threads = 1024 channels in the N1 implant; surgical robot insertion.
• ECoG (subdural) grid / strip — platinum disc electrodes 2–10 mm spacing on silicone substrate; placed under dura for clinical epilepsy localisation. Vendors: Ad-Tech Medical, PMT, Integra.
• Stereo-EEG depth electrodes — multi-contact lead inserted through skull on a frame; pre-surgical seizure localisation.

Implant materials and coatings

Long-term implant electrodes must resist corrosion in saline body fluid and maintain low impedance over years. Standard materials: platinum-iridium (90/10 alloy, the cochlear-implant / DBS standard), stainless steel (legacy pacemaker leads), iridium oxide (IrOx) sputtered film (very high charge-injection capacity), PEDOT:PSS electroplated polymer (lowers impedance by 10–100× vs bare metal). Encapsulation: parylene-C vapour-deposited polymer, silicone elastomer, polyimide (Neuralink threads). The biocompatibility lifetime is set by the chronic foreign-body response — a glial encapsulation forms around any implant over weeks, raising impedance and degrading SNR. Reducing this is an active research area (PEDOT coatings, sub-cellular cross-section threads, anti-inflammatory drug elution).

7. Specialised modalities

ECG (electrocardiography)

Origin: Willem Einthoven 1903 (string galvanometer), Nobel 1924. 12-lead diagnostic (10 electrodes, 12 derived leads — three limb leads I/II/III, three augmented aVR/aVL/aVF, six precordial V1–V6) remains the clinical standard for arrhythmia, myocardial infarction, ischaemia, and chamber-enlargement diagnosis. Monitor-grade 3- or 5-lead is the ICU and ambulance default. Single-lead wearable ECG (Apple Watch Series 4+, Withings ScanWatch, AliveCor KardiaMobile, Fitbit Sense) gained FDA clearance in 2018 for atrial-fibrillation detection — now a multi-tens-of-millions-of-units-per-year category.

The AI ECG era opened in 2019 with Mayo Clinic’s convolutional-net algorithm detecting asymptomatic LV dysfunction from a normal-looking 12-lead trace; AliveCor’s KardiaAI and Apple’s irregular-rhythm notification followed. FDA clearance under the Software as a Medical Device pathway is now routine.

EEG (electroencephalography)

Origin: Hans Berger 1929 (first human EEG). Clinical EEG uses the 10–20 system (Jasper 1958) of 21 electrode positions named relative to anatomical landmarks (nasion, inion, preauricular points). High-density research EEG uses 64, 128, or 256 channels. Five canonical bands: delta (< 4 Hz, deep sleep), theta (4–8 Hz, drowsiness, memory), alpha (8–13 Hz, relaxed wakefulness with eyes closed, occipital), beta (13–30 Hz, alert / active), gamma (30–100 Hz, cognition / binding).

Clinical applications: seizure detection (long-term video-EEG), sleep staging (PSG, scored per AASM), intraoperative monitoring (anaesthesia depth, BIS index, Sedline). Consumer / research-grade headsets: Muse, Neurosity Crown, Emotiv EPOC X, OpenBCI Cyton. Magnetoencephalography (MEG) is the magnetic counterpart — SQUID sensors at LHe temperatures, now being supplanted by room-temperature optically pumped magnetometers (OPM) that allow wearable MEG.

Pulse oximetry and PPG

Origin: Takuo Aoyagi 1972; commercial Nellcor 1983; Masimo SET 1995. Finger and forehead are clinical standards; earlobe, wrist (Apple Watch SpO₂), and forehead-reflectance variants extend the technique. Photoplethysmography (PPG, single wavelength) is sufficient for heart rate and rhythm; SpO₂ requires the red / IR pair. Recent extensions: continuous blood-pressure estimation from PPG morphology (Aktiia, BIOSENCY — accuracy versus invasive arterial line remains the clinical question), and respiratory rate from PPG baseline modulation.

NIRS / fNIRS

Near-infrared spectroscopy uses 700–900 nm light to probe haemoglobin oxygenation in deeper tissue. Cerebral oximetry (INVOS, FORE-SIGHT, NIRO) monitors regional brain oxygenation during cardiac surgery and ICU stays. Functional NIRS (fNIRS) measures cortical haemodynamics during cognitive tasks — a portable, motion-tolerant alternative to fMRI, used in BCI research, infant neuroscience, and behavioural studies. Vendors: NIRx, Artinis, Kernel Flow.

Bioimpedance

A small AC current (typ. 50 kHz, ~100 µA) injected between two electrodes and the voltage measured between two others yields tissue impedance. Applications: body composition (Tanita, InBody, withings scales — fat / muscle / water), respiration rate (impedance pneumography on ECG monitors), lung-fluid status (CardioMEMS HF), stroke volume (impedance cardiography), EDA (skin-conductance for affective state).

Glucose

Origin: Clark 1962 (oxygen electrode), enzymatic test strip Yellow Springs 1969, commercial home meter LifeScan 1986. SMBG (self-monitoring blood glucose) test strips use glucose oxidase or glucose dehydrogenase on a screen-printed carbon electrode (Roche Accu-Chek, Abbott FreeStyle, LifeScan OneTouch, Ascensia Contour). CGM (continuous glucose monitoring) uses a subcutaneous filament with the same enzymatic chemistry, sampling interstitial-fluid glucose every 1–5 minutes for 10–14 days per sensor (Dexcom G7, Abbott FreeStyle Libre 3, Medtronic Guardian 4, Senseonics Eversense). Closed-loop / automated insulin delivery (Tandem t:slim X2 with Control-IQ, Medtronic 780G, Beta Bionics iLet, Insulet Omnipod 5) couples a CGM to an insulin pump with a model-predictive control or PID algorithm — the most clinically successful closed-loop bioelectronic system in routine use.

Non-invasive optical / RF glucose is a long-standing unsolved problem; hundreds of startups have failed to clear FDA. As of 2026, all approved CGMs remain minimally invasive (subcutaneous filament).

BCI and neural interfaces

Modern field leaders: Neuralink (1024-channel implant, threads, surgical robot, primate / early human trials), Synchron (Stentrode endovascular electrode, no craniotomy, FDA Breakthrough Device, human implants ongoing), Blackrock Neurotech (Utah-array foundation, clinical BrainGate consortium), Paradromics (high-channel-count microwire bundle), Precision Neuroscience (Layer 7 thin-film cortical array, sub-cranial, reversible), Onward (spinal cord stimulation for motor recovery). Clinical results 2024–2026 include human typing / cursor control at conversational speeds, speech decoding from cortical activity, and motor restoration after spinal injury.

Pacemakers, ICDs, and CIEDs

Pacemaker (cardiac implantable electronic device, CIED): senses intrinsic cardiac activity and delivers low-energy pacing pulses (~5 V, 0.5 ms, 60–500 µJ) when the heart rate drops below programmed thresholds. Single-chamber (one lead), dual-chamber (atrial + ventricular), CRT (biventricular for heart failure resynchronisation). Leadless variants: Medtronic Micra, Abbott Aveir — capsule directly attached to right-ventricular endocardium.

ICD (implantable cardioverter-defibrillator): adds high-energy shock delivery (up to 40 J) for ventricular tachycardia / fibrillation. Subcutaneous variants (Boston Scientific S-ICD, EMBLEM) avoid intracardiac leads.

Vendors: Medtronic, Abbott (formerly St. Jude), Boston Scientific, BIOTRONIK, MicroPort. Telemetry: legacy 175 kHz inductive; modern Bluetooth Low Energy and proprietary 401–406 MHz MedRadio band.

Hearing aids and cochlear implants

Hearing aid — microphone + DSP (compression, noise reduction, feedback cancellation, wind-noise suppression) + miniature receiver (speaker). Modern devices run on Bluetooth LE Audio (Auracast) for direct streaming. Vendors: Phonak (Sonova), Oticon (Demant), Widex, Signia, Starkey, ReSound.

Cochlear implant — bypasses damaged hair cells with a 12–24 channel electrode array inserted in the scala tympani driving the auditory nerve directly. External processor maps acoustic spectrum to electrode-channel current pulses (CIS, ACE strategies). Vendors: Cochlear (Australia), Advanced Bionics (Sonova), MED-EL (Austria).

Blood pressure measurement

Three modalities are clinically deployed:

• Oscillometric NIBP (cuff) — pneumatic cuff, pressure transducer, inflate above systolic then slowly deflate while measuring cuff-pressure oscillations. Peak oscillation amplitude indicates MAP; systolic and diastolic are derived by vendor-specific ratios (typically systolic = MAP at oscillation 50–55% of peak, diastolic at 70–80%). The dominant modality for ICU and home BP; vendors include Welch Allyn, GE, Suntech, Omron. AAMI / ISO 81060-2 validation against intra-arterial reference is mandatory for clearance.
• Auscultatory — Korotkoff sounds on a manual cuff with a stethoscope; the gold-standard reference and the modality the AHA / ESH guidelines anchor on. Now largely a teaching method; oscillometric has replaced it clinically due to operator independence.
• Invasive arterial line — catheter in radial / femoral artery connected to a fluid-filled pressure transducer (typ. Edwards TruWave, ICU Medical Transpac); silicon-strain-gauge bridge, 4 wires, mV-scale output amplified by a dedicated in-amp; calibrated to atmospheric reference at the phlebostatic axis. Real-time beat-to-beat waveform; gold standard in OR and ICU.
• Continuous non-invasive (cNIBP) — finger-cuff volume-clamp (Edwards ClearSight, Finapres), pulse-wave-velocity-based (Caretaker, Aktiia), PPG-morphology AI (multiple research / wearable). Continuous arterial-line-like trace without the catheter; accuracy vs invasive remains the open clinical question.

Body temperature

• NTC thermistor — the workhorse, sub-Ω resolution at body-temperature range with simple Wheatstone-bridge or ratiometric ADC sense.
• RTD (Pt100, Pt1000) — higher accuracy and stability; used in laboratory and reference instruments.
• IR tympanic — uncooled microbolometer or thermopile (Texas Instruments TMP006/007, Melexis MLX90614) — non-contact ear-canal IR measurement; rapid but operator-dependent.
• Forehead temporal artery — IR thermometer (Exergen); peer-reviewed accuracy vs core lower than tympanic.
• Ingestible pill — radio-pill (Equivital, BodyCap eCelsius, HQ-Inc CorTemp); core temperature from gut for thermal physiology research and athletic monitoring.
• Skin-patch continuous — adhesive sensor with NTC + Bluetooth (TempTraq, VivaLNK CTM); 24/7 continuous core-temperature trend.

Other implantables

Deep-brain stimulation (DBS) — Medtronic Activa, Abbott Infinity, Boston Scientific Vercise; subthalamic / pallidal / VIM stimulation for Parkinson’s, essential tremor, dystonia, OCD. Vagus nerve stimulation (VNS) — LivaNova Sentiva for epilepsy and depression. Sacral neuromodulation — Medtronic InterStim, Axonics for urinary urge incontinence. Hypoglossal nerve stimulation — Inspire Medical for obstructive sleep apnoea. Spinal cord stimulation — Abbott, Medtronic, Boston Scientific, Nevro for chronic pain.

8. Safety and regulatory

Core medical electrical safety standard — IEC 60601-1

The 3rd edition with Amendment 2 (2020) is current. Defines:

• Type B applied part — no patient connection (e.g. blood-pressure monitor with cuff on arm); ≤ 100 µA patient leakage normal, ≤ 500 µA single fault.
• Type BF — patient connection, floating (isolated); ≤ 100 µA normal, ≤ 500 µA single fault.
• Type CF — direct cardiac contact (catheter electrode); ≤ 10 µA normal, ≤ 50 µA single fault. Requires reinforced isolation.
• Means of Patient Protection (MOPP) — 1500 V_AC dielectric withstand per layer; two layers for reinforced.
• Means of Operator Protection (MOOP) — 1500 V_AC; one layer typical.

Collaterals add modality-specific requirements: IEC 60601-2-25 (electrocardiographs, diagnostic), −2-27 (electrocardiographic monitoring), −2-26 (EEG), −2-47 (ambulatory ECG / Holter), −2-37 (ultrasound), −2-31 (cardiac pacemakers / external), −2-49 (multifunction monitors). IEC 60601-1-2 governs EMC (immunity and emissions); −1-8 governs alarms; −1-6 governs usability.

Quality management — ISO 13485 / 21 CFR 820

ISO 13485:2016 is the international medical-device QMS standard. 21 CFR Part 820 (FDA Quality System Regulation, QSR) is the US equivalent, harmonising with ISO 13485 under the 2024 final rule (effective 2026). Both mandate design controls, design history file (DHF), device master record (DMR), CAPA, supplier controls, complaint handling.

Regulatory pathways

Region	Low risk	Moderate risk	High risk
US (FDA)	Class I (general controls, 510(k) exempt)	Class II (510(k) substantial-equivalence clearance)	Class III (PMA, premarket approval)
EU (MDR 2017/745)	Class I (self-declaration)	Class IIa / IIb (notified body audit)	Class III (notified body + clinical evidence + scrutiny)
Japan (PMDA)	Class I (general)	Class II (controlled)	Class III / IV (specially controlled)
China (NMPA)	Class I	Class II	Class III

De novo (FDA) is the path for novel low-to-moderate-risk devices without a predicate; Breakthrough Device designation accelerates review for high-impact devices. IVDR 2017/746 (EU in vitro diagnostics regulation) applies to lab assays including glucose meters and home tests; mandatory since 2022 with phased transition through 2027.

Software, risk, and cybersecurity

• IEC 62304:2006+AMD1:2015 classifies medical software A / B / C by harm potential; defines lifecycle artefacts (SWRS, architecture, unit tests, V&V, traceability).
• ISO 14971:2019 — risk management lifecycle; FMEA / fault-tree style hazard analysis is the practical workhorse.
• IEC 62366-1 — usability engineering for medical devices; use-error vs use-related risk.
• FDA premarket cybersecurity guidance (2023) — mandates a Software Bill of Materials (SBOM), threat model, vulnerability disclosure plan, and post-market patch capability. IEC 81001-5-1 is the international counterpart.
• HIPAA (US) / GDPR (EU) for health-data privacy on the cloud-connected side.

MRI compatibility

ASTM F2503 defines three categories: MR Safe (no metallic components, no risk in any field), MR Conditional (safe under defined conditions — field strength, gradient slew, SAR limits), MR Unsafe. Implants and external accessories must be labelled accordingly. Pacemakers historically excluded patients from MRI; modern “MR-conditional” pacemakers (Medtronic Advisa MRI, Abbott Assurity MRI) allow scanning under restricted protocols.

Other standards

• AAMI EC57 — testing and reporting of ECG arrhythmia detection algorithms (sensitivity, PPV vs MIT-BIH database).
• AAMI SP10 / ISO 81060 — non-invasive blood-pressure validation.
• ISO 80601-2-61 — pulse oximetry accuracy (typ. ±2% A_RMS across 70–100% SpO₂).
• ANSI/AAMI/IEC 80369 series — small-bore connectors (Luer, enteral, neuraxial — designed to prevent mis-connection).

Standards reference matrix

Standard	Domain	Scope
IEC 60601-1	Safety, general	Electrical safety, mechanical, leakage, isolation, MOPP/MOOP
IEC 60601-1-2	EMC	Immunity (radiated/conducted) and emissions for medical
IEC 60601-1-6	Usability	Use-error and use-related risk
IEC 60601-1-8	Alarms	Audible/visual alarm priority, message content
IEC 60601-1-9	Environment	Environmental design (energy, materials, EoL)
IEC 60601-2-25	ECG diagnostic	12-lead, 0.05 Hz HPF, 150 Hz LPF, ST-segment fidelity
IEC 60601-2-26	EEG	Multichannel, common-mode rejection, defib recovery
IEC 60601-2-27	ECG monitor	Continuous monitoring; pace-pulse detection
IEC 60601-2-47	Ambulatory ECG	Holter; storage and retrieval
IEC 60601-2-31	External pacemaker	Output energy, sensing
IEC 60601-2-34	Invasive BP	Catheter pressure transducer interface
IEC 60601-2-49	Multi-parameter	Combined monitor architectures
IEC 62304	Software	Lifecycle, classification A/B/C
IEC 62366-1	Usability eng	Process for use-error mitigation
ISO 13485	QMS	Quality management for medical devices
ISO 14971	Risk	Risk management lifecycle
ISO 10993	Biocompatibility	Series for materials in contact with body
ISO 80601-2-61	Pulse oximetry	Accuracy specification (A_RMS)
AAMI EC57	ECG algorithms	Sensitivity / PPV / specificity reporting
AAMI/IEC 80369	Connectors	Non-interchangeable small-bore connectors
21 CFR 820	US QMS	FDA Quality System Regulation
21 CFR 803	US reporting	Medical Device Reporting (MDR)
EU MDR 2017/745	EU medical	Device classes, CE marking, notified body
EU IVDR 2017/746	EU diagnostics	In vitro diagnostic regulation

9. Edge cases and gotchas

• Electrode polarisation drift. Even matched Ag/AgCl drifts on a seconds-to-minutes timescale; AC coupling at 0.05 Hz is the standard, but slow recovery after motion artefact or defib can take seconds.
• Motion artefact dominates wearables and exercise ECG. Reflection-mode PPG is more motion-sensitive than transmission. Accelerometer reference signal subtraction (Masimo SET, Apple’s optical heart sensor) is the production answer.
• Triboelectric noise from cable flexing. Standard solution: low-noise carbon-impregnated shield, twist-bonded leads, strain relief at the connector.
• RF interference. Hospitals are dense RF environments — cellular, Wi-Fi, BLE, Zigbee, RFID readers, MRI gradients. Faraday-shielded enclosures, ferrite chokes on cables, isolation across the patient barrier.
• Defibrillator pulse artefact. Even with protection, the AFE saturates; recovery time is specified by IEC 60601-2-27 (≤ 5 s, ECG monitor return to ≤ 10% baseline deviation).
• Pacemaker pulse blanking. Cardiac monitors must detect and blank pacemaker spikes from rhythm analysis; modern monitors auto-classify paced vs intrinsic beats.
• Stim artefact in BCI. Recording during cortical or peripheral stimulation requires blanking circuits and artefact-template subtraction; otherwise the stim pulse saturates the recording front-end.
• Wireless coexistence. A BLE 2.4 GHz radio in a hearing aid or wearable shares spectrum with Wi-Fi, microwave ovens, and adjacent BLE devices. Coexistence engineering and adaptive frequency hopping mitigate.
• Battery vs mains. Implants use primary (Li/I₂ for pacemakers, Li-CFx for ICDs) or secondary (Li-ion for newer high-power devices, wirelessly recharged via inductive link). Hospital mains equipment requires the full isolation stack.
• Sterilisation compatibility. Ethylene oxide (EtO, low-T, electronics-friendly but slow), gamma (high-energy, can damage CMOS and polymers), e-beam (similar), autoclave (high-T steam, kills most consumer electronics). Implant electronics are typically EtO or e-beam sterilised with material qualification.
• Coating aging. PEDOT:PSS impedance reduction degrades over months in vivo; parylene-C and silicone encapsulation can develop pinholes; long-term-implant reliability is an active failure-mode area.
• Patient comfort vs adhesion. Gel electrodes dry out, conductive textiles itch, dry electrodes have higher impedance and worse SNR — the sensor that delivers a clinical-grade signal in a research clinic may be unworn at home.
• Paediatric vs adult. Lower body mass, smaller electrode area, thinner skin → tighter leakage-current limits, lower stimulation amplitudes, different default filter settings.
• Privacy and connected health. HIPAA in the US, GDPR in Europe, equivalents elsewhere; data minimisation, pseudonymisation, breach-notification timelines.
• AI / ML validation. Algorithms claiming diagnostic indication require FDA validation; “wellness” / “general wellness” claims under FDA’s 2019 guidance face a lower bar but cannot make diagnostic claims.
• Predicate dependence. A 510(k) clearance requires “substantial equivalence” to a predicate device — meaning a novel sensor or algorithm without a credible predicate must go through De Novo or PMA, adding 12–36 months to time-to-market and meaningful clinical-evidence cost.
• Connector mis-mating. Patient cables historically used DIN 42802 (touch-proof, 1.5 mm or 0.7 mm) jacks; in 2026 most clinical equipment still uses these but vendors have moved to proprietary connectors for IP-protection on consumables. Mis-mating an EEG cable into an ECG port can deliver pacing currents into a sub-millivolt amplifier — input protection per section 4 is the only defence.
• Common-mode → differential conversion. Source impedance imbalance is the single most-underappreciated failure mode of an apparently-CMRR-adequate AFE. A 1 kΩ mismatch between two 100 kΩ electrodes converts 1% of common-mode voltage to differential — turning a 1 V mains pickup into a 10 mV signal-band corruption, several times the ECG amplitude.
• Algorithm overfitting on PhysioNet datasets. The MIT-BIH, AHA, and CinC challenge datasets are heavily used for ECG algorithm development, but they over-represent certain demographics and recording conditions; out-of-distribution performance on diverse populations or different ECG hardware can be 30–50% worse than the published benchmark.

Engineering judgement — design tradeoffs

Decision	Cheap / fast path	Premium / clinical path
Electrode	Dry textile, conductive elastomer	Disposable gel Ag/AgCl
In-amp	INA826 (mid-range)	INA333 / AD8429 (precision / low-noise)
ADC	Σ-Δ in MCU (12-bit, oversampled)	ADS129x / MAX30003 dedicated bio-AFE
Isolation	Skip (battery-powered single-patient)	Reinforced (capacitive ISO77xx + isolated DC-DC)
RTOS	Bare metal, polled	Safety-certified (SafeRTOS, Zephyr LTS)
Connectivity	BLE only	BLE + LTE + Wi-Fi
Regulatory	”Wellness” claim, no FDA	510(k) + IEC 60601 testing
Clinical	Internal usability + bench	Multi-site clinical study
BoM ($, 1 ku)	$5–30	$200–2000
Time to market	6–12 months	18–60 months

The middle path — clinical-grade signal quality with consumer aesthetics and a 510(k) clearance — is where most current wearable medical successes (KardiaMobile, Apple Watch ECG, Withings ScanWatch) have landed, and it’s much harder to execute than either pure endpoint.

10. Tools and software

Schematic, PCB, simulation

• Schematic / PCB: KiCad (open source, increasingly used in startups), Altium Designer (commercial standard), OrCAD / Allegro (Cadence), Eagle (legacy). Flex / flex-rigid stackups for wearables: Altium with flex support, Cadence Flex DRC.
• Analog simulation: LTspice (free, the workhorse for op-amp / front-end design — manufacturers publish SPICE models for every bio-AFE block), TINA-TI (TI-flavoured), QucsStudio. See [[Engineering/op-amps]].
• System-level: MATLAB / Simulink with Biomedical Toolbox, LabVIEW Biomedical Toolkit (NI), Python (NumPy / SciPy + NeuroKit2 for biopotential pipelines).

DSP and signal processing

• MNE-Python — EEG / MEG analysis, the academic standard.
• NeuroKit2 — high-level ECG / EEG / EMG / EDA processing in Python.
• BioSPPy — biosignal processing toolkit in Python.
• EEGLAB / FieldTrip / Brainstorm — MATLAB EEG/MEG toolboxes.
• PhysioNet WFDB — open clinical biosignal databases (MIT-BIH, MIMIC) + reference processing tools.
• OpenBCI — open-source EEG / EMG / ECG hardware + software (GUI, Brainflow library).

Embedded development

• TI MSP430 / MSPM0 / TM4C — low-power MCU lines paired with TI bio-AFEs (ADS129x, AFE4404).
• Nordic Semi nRF52 / nRF54 — BLE-integrated MCU, dominant in wearables; Nordic SoftDevice BLE stack, Zephyr RTOS.
• STMicro STM32WB / STM32WBA — BLE; STM32H7 for higher-end with DSP / ML accelerators.
• Ambiq Apollo4 / Apollo5 — sub-threshold ultra-low-power MCU for always-on wearables.
• NXP MCXN / Kinetis — automotive / medical-leaning MCU + DSP.

Test instrumentation and reference designs

• Patient simulators: Fluke ProSim 8 (multi-parameter), Whaleteq AECG100 (12-lead ECG), Fluke MPS450 (pulse-ox / NIBP / ECG / temperature / respiration), Symbio SmartSimulator (pacemaker / ICD test).
• Reference designs: TI ADS129x EVM, AFE4404 EVM; ADI EVAL-AD8232 (single-lead ECG front-end), EVAL-AD8233; Maxim MAX30001 / MAX30003 / MAX30101 EVKs; Microchip MIC4-EVAL glucose meter dev kit.
• Bench instrumentation: low-noise spectrum analyser (R&S FSV / Keysight EXA) for AFE noise characterisation; precision sourcemeter (Keithley 2400) for bias-current and offset measurements; vector network analyser for isolation barrier characterisation; electrostatic discharge gun (NoiseKen ESS, EM Test) for IEC 61000-4-2 compliance.
• EMC test labs: full IEC 60601-1-2 compliance testing typically requires a certified lab (TÜV, UL, Element); pre-compliance scans on bench (Tekbox TBOH01 for conducted, GTEM cells for radiated) shorten the iteration cycle.

Software for medical-device development

• Requirements / traceability: Polarion, IBM DOORS, Jama Connect — used to maintain SWRS / SyRS / HWRS traceability matrices required by IEC 62304 and FDA design-controls.
• Test management: Tricentis qTest, Polarion, Helix ALM, custom MATLAB / Python frameworks driving patient simulators.
• Static analysis: Coverity, PC-Lint Plus, Helix QAC, MISRA-C checkers — common in Class B/C medical firmware.
• Coding standards: MISRA C:2012 (with AMD 1, 2, 3, 4), Barr Group Embedded C Standard.
• Trusted hash / sign-off: file integrity hashes (SHA-256 + timestamp) on every release artefact; design history file (DHF) maintained per device.

Biopotential AFE chip families (current production, 2026)

Family	Vendor	Channels	ADC	Notes
ADS1292 / ADS1292R	TI	2 + resp	24-bit Σ-Δ	Wearable ECG / BioZ; battery-friendly
ADS1298 / ADS1298R	TI	8 + resp	24-bit Σ-Δ	12-lead diagnostic ECG; respiration
ADS1299	TI	8	24-bit Σ-Δ	EEG / EMG specific; lower-noise variant
ADS131M0x	TI	2 / 4 / 6 / 8	24-bit Σ-Δ	Industrial + medical multi-purpose
AD7124-8	ADI	8 (mux)	24-bit Σ-Δ	Generic precision, integrated PGA
AD8232 / AD8233	ADI	1	external	Single-lead ECG analog front-end
ADAS1000	ADI	5 (lead-derive 12)	19-bit	Hospital ECG
MAX30001 / MAX30003	Analog Devices (Maxim)	1 ECG + BioZ	18-bit	Wearable single-lead ECG with pace detect
MAX30101 / MAX86150	Analog Devices (Maxim)	PPG + (MAX86150 adds ECG)	18-bit	Wrist PPG / SpO₂ / HR; MAX86150 fingertip combo
AFE4404 / AFE4900	TI	PPG + ECG (4900)	22-bit	Optical-and-electrical combo for wearables
ADPD4xxx / ADPD188BI	ADI	PPG + ambient	14–20-bit	LED-driver + photodiode integrated

Wireless connectivity chips

• Nordic nRF52840 / nRF5340 / nRF54L15 — BLE 5.x; most wearables.
• STMicro BlueNRG-2 / -LP — BLE; medical and IoT.
• TI CC2640R2F / CC2652 — BLE / Thread / Zigbee multi-protocol.
• Silicon Labs EFR32BG24 — BLE with on-chip AI/ML.

Reference textbooks for tooling and design

The Webster “Medical Instrumentation” textbook (5th ed 2020), Bronzino “Biomedical Engineering Handbook” (4th ed 2015), and the Northrop “Analog Electronic Circuits to Biomedical Instrumentation” (2nd ed 2011) are the standard engineering references — each has worked AFE designs at the schematic level and discusses tool choices.

11. Cross-references

• [[Engineering/op-amps]] — instrumentation-amplifier internals, photodiode TIA design, chopper-stabilised precision amplifiers, the in-amp + DRL pattern that anchors every biopotential AFE.
• [[Engineering/semiconductor-devices]] — diodes (input clamps), MOSFET (switch, ESD), Σ-Δ ADC internals; physical layer behind every AFE.
• [[Engineering/signal-processing-dsp]] — anti-alias, notch, decimation, FIR / IIR, FFT and wavelet methods used in ECG R-peak detection, EEG band-power, vibration analysis on bioelectrical data.
• [[Engineering/microcontrollers]] — BLE-SoC and low-power MCU host platforms for wearables.
• [[Engineering/pcb-design]] — flex PCB stackups, guard rings, isolation barrier layout, EMC partitioning.
• [[Engineering/biomechanics]] — companion note (same batch); joint kinematics and force plates use the same MEMS-IMU / load-cell + in-amp stack.
• [[Engineering/mems]] — MEMS accelerometer / gyro / barometer / microphone — the motion and pressure transducers in every wearable.
• [[Engineering/microfluidics]] — lab-on-chip and point-of-care assay platforms; share electrode patterning and AFE strategies.
• [[Engineering/photonics]] — LED drivers, photodiodes, and optical filters for PPG / SpO₂ / NIRS.
• [[Robotics/sensors-pose-motion]] — wearable IMU pipelines.
• planned [[Languages/Tier3/healthcare-clinical]] — HL7 FHIR, DICOM-Waveform, IEEE 11073 PHD, AAMI EC57 reporting formats.

12. Citations

• Webster, J. G. (ed.) (2020). Medical Instrumentation: Application and Design (5th ed.). Wiley. The canonical textbook; chapters on biopotential electrodes, amplifiers, ECG, EEG, EMG, blood pressure, respiration, and clinical safety underpin most of this note.
• Bronzino, J. D. & Peterson, D. R. (eds.) (2015). The Biomedical Engineering Handbook (4th ed., 3 vols.). CRC Press. The reference handbook of biomedical engineering; section coverage spans physiology, biomechanics, biomaterials, biomedical signal analysis, and clinical equipment.
• Northrop, R. B. (2011). Analysis and Application of Analog Electronic Circuits to Biomedical Instrumentation (2nd ed.). CRC Press. Schematic-level analysis of bio-AFEs, in-amps, isolation, and noise.
• Carr, J. J. & Brown, J. M. (2000). Introduction to Biomedical Equipment Technology (4th ed.). Prentice Hall. Clinical engineering / biomed-tech perspective; the standard text for hospital BMET training.
• Plonsey, R. & Barr, R. C. (2007). Bioelectricity: A Quantitative Approach (3rd ed.). Springer. The theory of action potentials, volume conduction, and bioelectrical sources behind every biopotential measurement.
• Einthoven, W. (1903). Ein neues Galvanometer. Annalen der Physik. The string-galvanometer paper that founded clinical ECG; Nobel Prize in Physiology or Medicine 1924.
• Berger, H. (1929). Über das Elektrenkephalogramm des Menschen. Archiv für Psychiatrie und Nervenkrankheiten 87, 527–570. The original human EEG paper.
• Severinghaus, J. W. & Honda, Y. (1986). “History of blood-gas analysis. VII. Pulse oximetry.” Journal of Clinical Monitoring 3(2), 135–138. The historical reference for the development of pulse oximetry, including Aoyagi’s 1972 work.
• Pan, J. & Tompkins, W. J. (1985). “A real-time QRS detection algorithm.” IEEE Transactions on Biomedical Engineering 32(3), 230–236. The standard QRS-detection algorithm — still the baseline against which new methods are compared.
• Olansen, J. B. & Rosow, E. (2002). Virtual Bio-Instrumentation: Biomedical, Clinical, and Healthcare Applications in LabVIEW. Prentice Hall.
• IEC 60601-1:2005+AMD1:2012+AMD2:2020. Medical electrical equipment — Part 1: General requirements for basic safety and essential performance. International Electrotechnical Commission. The umbrella safety standard.
• IEC 60601-2-25:2011+AMD1:2020. Medical electrical equipment — Part 2-25: Particular requirements for the basic safety and essential performance of electrocardiographs.
• IEC 60601-2-26:2012. Medical electrical equipment — Part 2-26: Particular requirements for the basic safety and essential performance of electroencephalographs.
• IEC 60601-2-27:2011+AMD1:2020. Particular requirements … for electrocardiographic monitoring equipment.
• IEC 62304:2006+AMD1:2015. Medical device software — Software life cycle processes.
• ISO 13485:2016. Medical devices — Quality management systems — Requirements for regulatory purposes.
• ISO 14971:2019. Medical devices — Application of risk management to medical devices.
• 21 CFR Part 820 (US FDA). Quality System Regulation (current, with 2024 harmonisation rule to ISO 13485).
• FDA (2023). Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions — Guidance for Industry and Food and Drug Administration Staff.
• AAMI EC57:2012/(R)2020. Testing and reporting performance results of cardiac rhythm and ST segment measurement algorithms.
• ASTM F2503-23. Standard Practice for Marking Medical Devices and Other Items for Safety in the Magnetic Resonance Environment.
• Texas Instruments. ADS1298 Low-Power, 8-Channel, 24-Bit Analog Front-End for Biopotential Measurements (datasheet, current revision). The reference biopotential AFE datasheet.
• Maxim / Analog Devices. MAX30003 Ultra-Low Power, Single-Channel Integrated Biopotential AFE (datasheet, current revision).
• Analog Devices. AD8232 Single-Lead, Heart Rate Monitor Front End (datasheet, current revision).