First Design Study (8 Areas)

1. Analog front end (ECG input stage)

We designed an input stage that receives the three measuring electrodes (RA · LA · LL) and the reference electrode (RL) with a TI ADS1298 (8-channel 24-bit ΔΣ AFE). Each electrode has its own protection and filter, and right-leg drive (RLD) feedback is applied through RL to reduce common-mode noise.

Key decisions and figures

• Channel assignment: CH1 = LA − RA (lead I), CH2 = LL − RA (lead II), CH3 = LL − LA (lead III measured directly), CH7 = RL electrode voltage monitoring, CH8 = internal test signal (calib row)
• Reference voltage 2.4 V · PGA gain 6 · 500 SPS → full scale ±400 mV, 1 LSB 47.68 nV, 1 mV ≈ 20,973 codes
• Since an electrode DC offset of ±300 mV (the level required by IEC 60601-2-27) must be accepted, gain 6 is the upper limit. With gain 12 the full scale shrinks to ±200 mV and the requirement cannot be met
• Digital filter (sinc³) −3 dB bandwidth = 0.262 × 500 = 131 Hz
• Noise: calculated value including the input path (series resistors per electrode) approx. 5.9 µVpp
• System CMRR (60 Hz, including electrode imbalance 51 kΩ∥47 nF, worst-case sum): input stage alone 69.6 dB, with RLD approx. 89 dB
• What determines CMRR is not the amplifier itself but the absolute value of the input common-mode capacitance and the difference in trace parasitic capacitance. Therefore per-electrode wiring symmetry matters more than the component grade
• RLD DC loop gain target 10 → 60 Hz rejection 19.6 dB, phase margin approx. 78°. The rejection upper limit of this configuration is approx. 30–33 dB
• Electrode lead-off detection: DC current source 6 nA, response approx. 0.4 s. Only one pin per electrode opens its current source to avoid cancellation
• Calculated patient current under single fault (input pin shorted to supply) 25.1 µA — within the Type CF single-fault limit of 50 µA
• Defibrillation protection was not included. It was decided to state "disconnect the patient cable before defibrillation" in the instructions for use

Facts taken from the datasheet

• ADS1298 allowable supplies: AVDD 2.7–5.25 V, DVDD 1.65–3.6 V
• Input-referred noise 4 µVpp (150 Hz bandwidth, gain 6, value stated on the cover page), chip CMRR typ 115 dB, input bias typ 200 pA
• Internal test signal period = 2.048 MHz / 2²¹ = 0.977 Hz square wave, amplitude given as (VREFP − VREFN)/2.4 → 1 mV at a 2.4 V reference

Left as unverified: whether the internal test signal is ±1 mV or 1 mVpp, minimum CMRR, the RLD internal resistor value, recommendations for unused input pins — to be confirmed by the first measurements and comparison with the original datasheet.

2. Dual adaptive notch filter

We designed a two-stage notch filter that removes mains noise (60 Hz) and its 2nd harmonic (120 Hz) as an adaptive type that tracks the mains frequency by itself. It started from the specification requirement that "60 Hz noise is a major problem".

Why a fixed notch does not work

• A notch fixed at the nominal 60.000 Hz (Q = 60) drops to −14.1 dB attenuation if the mains deviates by only ±0.1 Hz. The required attenuation is −26 to −46 dB
• The domestic grid frequency maintenance range is known to be 60 ± 0.2 Hz (supporting clause unverified)
• In battery equipment, the larger source of deviation is the ADC clock → an external 2.048 MHz oscillator (±50 ppm) was assumed

Structure and figures

• No 180 Hz stage: calculated 180 Hz residual in diagnostic mode 0.27–1.1 µVpp — smaller than the AFE noise of 4 µVpp
• Method: adaptive IIR notch (frequency tracking). Battery-floating equipment has no mains reference signal, so this is simpler than reference-signal cancellation methods
• One frequency-estimation filter set (4 Hz bandwidth) generates the coefficients, and two output notch stages per channel use them. Because the three leads share one set of coefficients, the relation II = I + III is preserved after filtering
• The 120 Hz stage coefficient is computed directly from the 60 Hz coefficient (a₂ = 2 − a²). Tracking range is bounded to 58–62 Hz
• Bandwidth: diagnostic mode 60 Hz 1 Hz · 120 Hz 2 Hz (Q 60), monitor mode 2 Hz · 4 Hz (Q 30)
• Calculated post-QRS ringing: diagnostic mode typical QRS peak ≤ 10.6 µV, narrow QRS worst case ≤ 28.6 µV, monitor mode after 40 Hz low-pass 8.3 µV. Internal target ≤ 25 µV
• Phase: phase ≤ 2.2° and group delay ≤ 0.22 ms in the main ECG band 0.67–40 Hz. The real-time path uses a causal notch; zero-phase processing is an optional feature for offline analysis
• Stability: if the coefficients are bounded within range, the poles are always inside the unit circle (Jury test). Design convergence budget ≤ 2.6 s
• On-screen states: stabilizing → adapting → locked (shown as e.g. "AC 60.02 Hz") → hold → out of range

3. External EKG simulator

An external standalone, battery-operated simulator for analyzing calib and the notch filter with comparison waveforms. The 1 V p-p output is divided 1000:1 at the main unit's input stage to feed 1 mV p-p, and switched with the input stage's sim ↔ signal switch.

Key decisions and figures

• RA · LA · LL 3-terminal independent outputs + common terminal (COM). The condition V_RA + V_LA + V_LL = 0 uniquely determines the terminal potentials
• Lead ratio I : II : III = 0.6 : 1.0 : 0.4 (electrical axis approx. +50°). Since all three leads differ, swapped electrodes can be identified on screen
• Making all three leads 1 mV simultaneously is mathematically impossible (III = II − I) → calibration alternates between two modes, "I · II simultaneously 1 mV" and "II · III simultaneously 1 mV"
• 60 Hz noise injection: battery equipment does not pick up 60 Hz by itself, so it is generated intentionally with DDS. Resolution 1.16 µHz, adjustment step 0.01 Hz. 120 Hz is generated from the same phase accumulator so its phase is locked to the fundamental
• Test modes: frequency sweep 59.5 ↔ 60.5 Hz, step 60.0 → 60.1 Hz, amplitude modulation 0.1–1 Hz
• Waveforms: Gaussian-sum P · QRS · T model (20–300 bpm), 1 mV calibration square wave (period 0.5 / 1 / 2 s), 3 mV × 100 ms impulse, baseline wander, EMG noise (20–150 Hz, 0–100 µVrms)
• DAC 1 LSB referred to the main unit input is 38 nV — on the same level as the ADS1298 gain-6 LSB (approx. 47.7 nV)
• Electrode contact emulation 51 kΩ∥47 nF (same values as the IEC 60601-2-25/-27 test network) placed after the divider, switchable per electrode, giving an imbalance CMRR test configuration
• The selector switch is 4-pole including RL, break-before-make — so there is never a moment when the patient and simulator are connected simultaneously
• Amplitude accuracy: before calibration RSS ±0.25 %, after calibration ±0.05 %. Frequency: crystal ±20 ppm → ±0.0012 Hz at 60 Hz
• Estimated runtime: display on approx. 133 h, display off approx. 400 h (assuming AA effective capacity approx. 2000 mAh)
• Primary cells instead of rechargeable, no USB port — because if the simulator is connected to earth, the main unit's floating applied part becomes earth-referenced

What this simulator cannot verify

• All of SpO2 · PPG (it does not generate optical signals) — oxygen saturation number accuracy cannot be claimed with this equipment
• Real common-mode rejection (common mode at the several-V level), electrode DC offset ±300 mV, pacemaker pulses, defibrillation, leakage current · dielectric strength

4. SpO2 · pulse measurement section

A measurement section that receives red · infrared pulse waves (PPG) with a clip-type transmissive finger probe and a TI AFE4490, producing screen rows 5 (SpO2) and 6 (Pulse). In the first bundle this area's document was not completed, so it was filled in on 2026-10-11 based on the TI datasheet (SBAS602H) and development guide (SLAU480C).

Key decisions and figures

• A Nellcor-compatible DB9 analog probe is set as the design default for the probe connector. Actual probe model undecided
• LED current maximum 75 mA (1 code 0.293 mA). Firmware limit 50 mA until the probe rating is confirmed
• Sample timing: PRF 500 Hz, duty 25 % per LED (TI published example value), ADC reset extended to 6 clocks as recommended in the datasheet
• Of domestic lighting flicker (120 Hz and its multiples), only the 25th multiple (3000 Hz) folds to 0 Hz at 500 Hz sampling — the rest fall outside the pulse-wave band (0.5–5 Hz)
• Signal processing: 500 → 100 SPS, AC band 0.5–5 Hz, R = (AC/DC)red ÷ (AC/DC)infrared, median of the last 8 beats. The gain within one channel cancels out in R, so gain error does not enter directly
• No published motion-compensation algorithm could be confirmed, so instead of trying to correct it, quality is judged and the number is hidden ("---" shown, waveform gray)
• Inside the same patient isolation island as the ECG, the same Type F applied-part group → designed to Type CF limits
• 24-hour load-side energy of the SpO2 section: 0.68 Wh at the default 10 mA, worst case 3.08 Wh at the firmware limit of 50 mA

Facts taken from the datasheet

• 2-pin back-to-back LEDs driven by an H-bridge, LED1 = infrared · LED2 = red (datasheet figure annotation)
• TIA feedback resistor 10 kΩ – 1 MΩ (full-scale input current 50 µA – 0.5 µA), ADC 22-bit · full scale ±1.2 V · 1 LSB 0.572 µV
• 9 built-in probe diagnostics on chip (LED · photodetector open/short, etc.), taking 8 ms
• Output pins do not go high-impedance when the chip is deselected → the SPI bus is not shared with the ECG; a dedicated isolator is used
• Confirmed LED wavelengths (public manual): red ~660 nm, infrared ~900 nm or 920 nm. The wavelengths of the probe to be used are unverified

Understanding of accuracy — the number is "Reference value · not calibrated"

• Manufacturers do not publish the calibration curve that converts R into SpO2 %, and it can only be made through controlled hypoxia clinical studies under ISO 80601-2-61 → impossible for an in-house build
• Example: the textbook approximation (110 − 25·R) and another published quadratic formula differ by 4.9 %p at R = 1.0. There is no way to determine which one fits this probe
• Only the 95–100 % range can have its offset aligned by comparison with a commercially certified oximeter; below 90 % there is no way to verify
• The two-wavelength method mistakes carboxyhemoglobin for oxygen and overestimates (commercial equipment has the same limitation)
• Therefore the screen shows waveforms · trends, and always attaches "Reference value · not calibrated" next to the number

5. Limb leads and calib waveform

A summary of how the three limb leads are measured and calculated, and how the gain is monitored in real time with the calib (calibration) row on screen.

Lead equations and cross-check

• I = LA − RA, II = LL − RA, III = LL − LA, Einthoven relation II = I + III
• If III is calculated as II − I, the relation becomes an identity and cannot serve as a check. If all three leads are measured directly, the residual r = II − I − III can reveal channel gain mismatch and wiring errors in real time (internal criteria: 2 s window RMS ratio ≤ 1 % normal, > 3 % abnormal)
• The notch filter must use the same coefficients for all three leads for the relation to hold after filtering
• Electrode swap table: LA↔RA inverts all of lead I, LA↔LL inverts lead III, RA↔LL inverts lead II — the firmware only shows a "Check electrodes" warning (since this cannot be distinguished from dextrocardia)

calib standard values (reference standards)

Calibration method

• The internal test signal passes only through the PGA · ADC, while the external 1 mV (simulator ÷ 1000) passes through the entire path from the connector. Worst-case error budget for the external 1 mV approx. ±0.25 %
• 2-stage calibration: at shipment · inspection, per-channel gain correction factors are obtained with the external 1 mV, and the actual amplitude of the internal test signal is registered at the same time → during operation, the internal test signal is shown in the calib row for continuous comparison
• High-pass filtering makes the square wave droop (88 % droop at the end of a half period at 0.67 Hz). So the amplitude is measured as the step height at the edge
• Criteria (internal): amplitude deviation ≤ 2 % normal, 2–5 % check, > 5 % error
• Screen pixels may not be square. In a calculation for a typical 7" 800×480 panel the vertical gain error was −6.1 % → the pixel pitch is measured with a ruler and stored

6. Electrodes and skin contact

After comparing electrode materials · gels · skin impedance · standard requirements, we decided to use disposable Ag/AgCl electrodes.

Material comparison and decision

• Ag/AgCl is non-polarizable (standard potential +0.222 V, potential fixed by Cl⁻) — electrode-pair DC offset is typically a few mV or less; standard limit ≤ 100 mV (AAMI EC12)
• Polarizable electrodes such as stainless steel · nickel have offsets drifting from tens to hundreds of mV, eating into the AFE input range (±400 mV at gain 6)
• Decision: disposable pre-gelled Ag/AgCl snap type, solid hydrogel, products marked as AAMI EC12 compliant. Gels without Cl⁻ (ultrasound gel, etc.) must not be substituted

Skin impedance and noise (around 10 Hz)

• System CMRR is determined not by the amplifier but by electrode imbalance and input capacitance (example calculation at 50 Hz, assuming 51 kΩ∥47 nF imbalance · 100 pF input capacitance → −57.9 dB)
• Thermal noise of electrode resistance (150 Hz bandwidth): 5 kΩ 0.75 µVpp, 50 kΩ 2.4 µVpp, 500 kΩ 7.5 µVpp — measured noise is larger than this
• Motion artifact is mainly caused by skin potential changing by several mV, so it cannot be removed by filtering; light abrasion is most effective
• Lead-off detection current set to the minimum of 6 nA (6 mV offset at 1 MΩ)

AAMI EC12 disposable electrode requirements (summary)

• AC impedance average ≤ 2 kΩ · DC offset ≤ 100 mV · offset instability + internal noise ≤ 150 µVpp · polarization potential after defibrillation overload ≤ 100 mV · offset change at 200 nA bias current ≤ 100 mV

Operation

• Skin preparation: clip hair → clean with soap · dry → light abrasion → attach from the edges → strain relief for lead wires → measure after 1–5 minutes of stabilization
• Two placements: standard limb placement (wrists · ankles) and Mason-Likar torso placement (monitoring type). The two placements differ in QRS axis · amplitude, so the placement method is shown on screen and they are not compared directly with each other
• The IEC color scheme (RA red · LA yellow · LL green · RL black) and AHA (RA white · LA black · LL red · RL green) use the same colors for different electrodes, so one is used consistently — IEC scheme recommended
• Biocompatibility (ISO 10993 -1 · -5 · -10 · -23): for electrodes, the manufacturer data of approved products is used instead; among in-house parts touching the skin, no stainless steel · nickel-plated parts are used

7. 6-waveform screen and comparative analysis screen

We designed a main screen that continuously scrolls the 6 waveforms defined by the specification (calib · lead I · lead II · lead III · SpO2 · pulse), and a comparative analysis screen showing the notch filter before · after simultaneously. The first study is based on a 7" 800×480 panel.

Main screen

• For each row, a title and readout are shown as text in the right panel (so waveforms do not cover the text). Pulse rate number · heart blinking in time with the pulse · oxygen saturation xx % ("Reference value · not calibrated")
• Scroll: 65 Hz × 2 px = 130 px/s = 25.04 mm/s (+0.15 % vs. 25 mm/s). 65 Hz has the side effect that even if panel drive noise mixes in, it falls outside the 60 Hz notch tracking range
• When reducing 500 SPS to approx. 3.85 samples per column, a min · max envelope is drawn instead of simple decimation so that R-wave peaks are not lost
• Heart sync: ECG R peak first → if ECG is invalid for 3 s or more, PPG upstroke → returns when ECG is valid for 5 s continuously. Icon color is the source waveform color; red is not used, to avoid overlap with the alarm color. If ECG heart rate and PPG pulse rate differ by 10 bpm or more for over 10 s, "HR≠PR" is shown
• All rows are drawn with the same fixed delay (initially 300 ms) so the arrival time difference between ECG and pulse wave appears as-is on screen. Total screen latency approx. 340 ms

Comparative analysis screen (notch before − after)

• A pre-filter · B post-filter (always at the same gain) · C difference waveform e = before − after · D beat-synchronous average and spectrum (0–250 Hz)
• If the difference contains only pure 60 Hz it is normal; if QRS shape is mixed in, it is direct evidence that the notch is cutting into the ECG
• Leakage ratio quantified with the beat-synchronous average (64 beats): < 2 % normal, 2–5 % caution, ≥ 5 % leakage (internal criteria, not based on a standard)
• A simulator beat rate that is not an integer ratio of 60 Hz (e.g. 73 bpm) is recommended — with an integer ratio, averaging does not cancel the 60 Hz
• Displayed values: 60 Hz input · output amplitude, attenuation dB, notch center frequency (0.01 Hz units), frequency variation during QRS

Colors and text (light background)

• Text contrast ratio target 7 : 1 or higher (WCAG enhanced criterion), no gray text
• Minimum text size em 28 px (17.6′ at a 700 mm viewing distance), large heart rate · SpO2 numbers 64 px — based on ISO 9241-303 recommendations
• The first study had no alarm function, so it assumed "not for unattended monitoring"

8. Patient safety · isolation · power

Because this device touches the human body, isolation of patient-contact parts and leakage current were made the starting point of the design. Even with battery operation, the structure separately isolates a patient area within the board.

Leakage current limits (IEC 60601-1 Table 3 · Table 4, Type CF)

Why an internal isolation barrier is needed even with battery operation (calculation)

• In the test applying 110 % of mains voltage to the applied part, keeping within the CF limit of 50 µA requires the capacitance between the entire circuit and the enclosure foil to be at most 548 pF (242 V) / 502 pF (264 V). The screen metal plate · battery · ground plane alone easily reach several hundred pF
• If the backlight boost voltage (roughly 10–30 V) reaches an electrode line through an insulation short, 30 V / 200 kΩ = 150 µA > 50 µA → non-compliant. Therefore the highest voltage in the patient area is confined to 5 V or less
• Total barrier capacitance target ≤ 50 pF. Mains ↔ patient 2 MOPP dielectric strength 4000 Vrms for 1 minute, creepage 8 mm · clearance 5 mm
• Two resistors in series per electrode, so that even if one is shorted the remaining resistance is 100 kΩ or more (5 V / 50 µA)

Applied-part classification and usage restrictions

• ECG electrodes and SpO2 probe are grouped as one Type F applied-part group and designed as Type CF (not defibrillation-proof). The defibrillation protection required by IEC 60601-2-27 is not included, and this is disclosed
• Stated usage restrictions: disconnect all electrodes and probes before using a defibrillator · AED; no use in hospital · emergency · surgical · intensive care environments; no simultaneous use with electrosurgical units; do not bring into MRI rooms; heart rate display cannot be trusted for pacemaker users
• When a sim signal is input, the switch position is read in hardware and "SIMULATION" is forcibly shown on the entire screen

Risk management and testing

• ISO 14971 risk list of 16 items — excessive leakage current, mains coupling while charging, defibrillation, battery thermal runaway, incorrect heart rate · SpO2 display, frozen screen, mistaking the sim signal, etc. Of these, defibrillation · incorrect SpO2 display · absence of alarms cannot be eliminated by design and are reduced only by usage restriction notices
• IEC 62304 software safety class: Class B in the first study, assuming no alarms · no diagnosis
• 8 in-house test items: patient auxiliary current · patient leakage · total patient leakage · mains voltage on applied part · touch current · dielectric strength 1500 Vrms 1 minute · creepage/clearance · probe temperature. Tests involving lethal voltages are recommended to be commissioned to an accredited test laboratory
• Periodic in-use testing per IEC 62353 every quarter, and after drops · water ingress

First study power estimate

• Li-ion 1S2P certified pack 24.1 Wh, estimated load approx. 2.7 W (battery side) → runtime approx. 7.6 h. The backlight is likely the largest consumer
• Charging: mechanical exclusion by default — the charging jack and patient cable connector are covered by a single cover so they cannot be plugged in at the same time