Table of Contents

MAX30100 pulse oximeter and heart-rate sensor

The MAX30100 integrates red and infrared LEDs, a photodetector, optical elements, and low-noise electronics for pulse-oximetry and heart-rate signal acquisition. This binding configures the sensor and reads its raw red and infrared FIFO samples. The ESP32 sample processes those samples and displays the estimated heart rate and blood oxygen saturation (SpO₂).

The sensor must be calibrated against a suitable trusted reference before its SpO₂ results can be relied upon. Without calibration, the displayed oxygen-saturation values are experimental estimates and cannot be fully trusted. The supplied LED currents and processing defaults produced stable results on only a small number of tested sensors and fingers; different modules, optical covers, mounting arrangements, and users may require different settings and calibration. This binding and sample as is are not suitable for medical use.

Documentation

Hardware required

  • ESP32 development board
  • MAX30100 breakout module
  • Jumper wires

ESP32 wiring

MAX30100 ESP32
VIN 3.3 V
GND GND
SDA GPIO21
SCL GPIO22
INT Not required by this sample

Use 3.3 V logic. Some breakout boards include voltage regulators and I2C pull-ups; check the specifications of your module before connecting it.

Usage

Configure the ESP32 I2C pins before creating the I2C device:

Configuration.SetPinFunction(21, DeviceFunction.I2C1_DATA);
Configuration.SetPinFunction(22, DeviceFunction.I2C1_CLOCK);

I2cConnectionSettings settings = new I2cConnectionSettings(1, Max30100.DefaultI2cAddress);
I2cDevice i2cDevice = new I2cDevice(settings);
Max30100 sensor = new Max30100(i2cDevice);
Max30100PulseOximeterOptions pulseOximeterOptions = new Max30100PulseOximeterOptions
{
    SampleRate = 50,
    WindowSeconds = 8,
};
Max30100PulseOximeterWorker pulseOximeter = new Max30100PulseOximeterWorker(
    pulseOximeterOptions,
    queueSeconds: 2);

sensor.Reset();
sensor.Configure(
    Max30100Mode.SpO2,
    Max30100SampleRate.SamplesPerSecond50,
    Max30100PulseWidth.Microseconds1600,
    Max30100LedCurrent.Milliamps27Point1,
    Max30100LedCurrent.Milliamps40Point2);

uint processedDroppedSamples = 0;

while (true)
{
    while (sensor.TryReadSample(out Max30100Reading reading))
    {
        if (sensor.DroppedSampleCount != processedDroppedSamples)
        {
            processedDroppedSamples = sensor.DroppedSampleCount;
            pulseOximeter.Reset();
        }

        pulseOximeter.AddSample(reading);
    }

    Thread.Sleep(2);
}

Always drain all available FIFO samples before sleeping. At the sample's 50 samples per second, the 16-sample FIFO contains 320 milliseconds of data. Writing every raw sample to the debugger can be slower than acquisition and cause overflow. The driver batches FIFO-status reads, recovers from a full FIFO, and exposes the cumulative number of discarded samples through DroppedSampleCount. The complete sample prepares its worker and diagnostic threads before enabling sensor sampling, gives the acquisition thread the highest priority, gives its diagnostic thread the lowest priority, and limits output to once every four seconds. The shortened acquisition loop above intentionally omits diagnostic output.

Max30100PulseOximeterWorker queues samples in a fixed ring buffer and performs filtering, averaging, and measurement calculations on a background thread. This keeps the I2C reading loop responsive while analysis is running. Its DroppedSampleCount reports samples discarded if the processing queue cannot keep up. Reset() is asynchronous, and Dispose() stops and joins the worker thread.

Use Max30100PulseOximeter directly when synchronous processing is preferred. Both processor types use fixed buffers and do not allocate per sample.

The sample rate passed to either processor must exactly match the sensor sample rate. A mismatch changes the filter frequencies, analysis-window duration, and calculated BPM. For example, configuring the sensor for 100 samples per second while constructing the processor for 50 samples per second makes the calculated pulse rate approximately half its actual value.

Measurement configuration

The sample uses:

sensor.Configure(
    Max30100Mode.SpO2,
    Max30100SampleRate.SamplesPerSecond50,
    Max30100PulseWidth.Microseconds1600,
    Max30100LedCurrent.Milliamps27Point1,
    Max30100LedCurrent.Milliamps40Point2);

Measurement mode

Max30100Mode.SpO2 enables both the red and infrared channels. Both channels are required for the experimental oxygen-saturation ratio. Heart-rate processing primarily uses infrared data. Max30100Mode.HeartRate uses only the infrared channel and is appropriate when oxygen saturation is not needed.

Sample rate

The ESP32 sample uses SamplesPerSecond50, which records a sample every 20 milliseconds and gives the 16-sample FIFO 320 milliseconds of capacity. This is sufficient for the supported pulse-rate range and is the recommended starting point when debugger output is enabled.

SamplesPerSecond100 records a sample every 10 milliseconds and provides finer beat-interval resolution, but it:

  • Doubles I2C traffic and signal-processing work.
  • Halves FIFO time capacity from 320 milliseconds to 160 milliseconds.
  • Increases the chance of FIFO overflow during debugging or lengthy processing.
  • Doubles the processor working-buffer memory for the same analysis-window duration.

Use 100 samples per second only when the application can guarantee sufficiently responsive acquisition. Always use the same rate when constructing Max30100PulseOximeter.

LED pulse width

Microseconds1600 selects a 1600-microsecond LED pulse and 16-bit ADC resolution. It is valid with the sample's 50-sample-per-second rate. Compared with Microseconds800 and its 15-bit resolution, it:

  • Uses a longer optical integration period and more LED energy per sample.
  • Provides twice the numerical ADC range and one additional bit of resolution.
  • Improves sensitivity when the pulsatile signal is weak.
  • Requires enough ADC headroom to accommodate the increased optical response.

The effective full-scale values are 32,767 at 800 microseconds and 65,535 at 1600 microseconds. As a practical starting point, keep stable finger-contact levels around 30% through 70% of full scale:

Pulse width Resolution Full scale Suggested stable range
800 microseconds 15 bit 32,767 Approximately 10,000–23,000
1600 microseconds 16 bit 65,535 Approximately 20,000–46,000

With a 1600-microsecond pulse, IR around 32,000 and red around 28,000 are near the middle of the ADC range and have adequate headroom. Quality 0 at those levels indicates that the filtered AC waveform did not contain enough consistent beat intervals; it does not indicate insufficient DC brightness.

LED currents

The sample drives the red LED at 27.1 mA and the infrared LED at 40.2 mA. The infrared current is one step above 37.0 mA to test whether additional infrared signal improves pulse-wave detection while the red current remains unchanged. Increasing current generally raises both the DC reflection and pulsatile AC signal, but it also increases power consumption, LED heating, and the risk of ADC clipping. It does not necessarily improve perfusion index because both AC and DC can rise together. The red current provides more signal than 24 mA while retaining more headroom than 37 mA, which can clip at 65,535 on some fingers and modules.

The chosen unequal currents compensate for different red and infrared optical response. Tune each channel independently:

  • Increase current when the raw level and pulsatile amplitude are too low.
  • Decrease current when readings flatten near their upper range or stop responding to finger-pressure changes.
  • Keep enough headroom for the pulse waveform and movement transients.
  • Prefer a stable, unclipped waveform over making both channel values numerically equal.

Changing pulse width changes the effective optical energy, so LED currents may need retuning after switching between 800 and 1600 microseconds.

The sample processor first discards two seconds while finger pressure and optical levels stabilize, then uses an eight-second window of uninterrupted finger data. It restarts collection after an abrupt red or infrared level change because that indicates finger motion. It normally requires two consistent pulse candidates before HasValidHeartRate can become true. A candidate with at least four consistent beat intervals and 80% quality can be accepted immediately. The normal heart-rate miss limit is three: the last confirmed rate is held through two invalid analysis windows and cleared on the third. When the optical signal remains coherent, the stable-optical miss limit is six: the rate is held through five invalid windows and cleared on the sixth. Finger removal is debounced across five samples. Reset the processor after any FIFO overflow because missing samples invalidate the waveform timing:

  • HeartRate fuses adaptive peak intervals with normalized autocorrelation of the filtered infrared waveform. Local peaks must pass RMS-relative height and prominence thresholds. A refractory period derived from the maximum configured BPM retains only the strongest competing peak. Median interval filtering rejects outliers and compensates for isolated missed beats.
  • OxygenSaturation uses an uncalibrated ratio-of-ratios approximation.
  • PerfusionIndex is the infrared AC signal as a percentage of its DC level.
  • SignalQuality combines beat-interval consistency, the amount of interval evidence, and the proportion of raw intervals accepted. Two matching intervals can no longer report 100% quality, and windows containing several rejected intervals are penalized. At least four accepted intervals are needed to reach the full consistency score. Heart-rate readings below 45% are rejected.
  • HasValidHeartRate reports infrared pulse validity independently of the red-channel SpO₂ calculation.
  • IsHeartRateHeld indicates that the last confirmed heart rate is being retained through a weak or disagreeing analysis window. With the normal miss limit of three, the value is held through two consecutive misses and discarded on the third. The coherent-optical exception uses its separately configured limit. Signal resets still discard the value immediately.
  • HasValidOxygenSaturation reports whether the experimental ratio-of-ratios result is available.
  • UsedWeakSignalOxygenSaturation identifies SpO₂ values accepted through repeated weak-signal stability checks.
  • CandidateHeartRate exposes the current unconfirmed pulse candidate for diagnostics.
  • SignalDiagnostic reports the processor's reason for not publishing a heart rate, so applications do not need to duplicate processing thresholds.
  • AnalysisWindowProgress distinguishes expected zero diagnostics while the stable-sample window is still filling from an analyzed window that contains no usable pulse.
  • InfraredAcRms reports the detrended infrared pulse amplitude in ADC counts.
  • RedPerfusionIndex reports the red AC signal relative to its DC level.
  • ChannelCorrelation reports whether the detrended red and infrared pulse shapes agree. Experimental SpO₂ requires correlation of at least 0.80 and at least 0.10% perfusion in both channels.
  • RatioOfRatios reports (red AC / red DC) / (infrared AC / infrared DC). The demonstration converts it with 110 - 25R, but accurate conversion coefficients depend on the sensor module, optics, finger placement, and calibration against reference measurements.
  • OxygenSaturationCalibrationIntercept and OxygenSaturationCalibrationSlope expose the active coefficients in SpO₂ = intercept + slope × ratio.
  • BeatDetectionThreshold reports the minimum positive peak height used to detect beats.
  • BeatDetectionProminenceThreshold reports the required rise from the preceding local valley.
  • AutocorrelationHeartRate independently reports the strongest prominent periodic waveform in the supported heart-rate range.
  • AutocorrelationQuality reports periodic peak strength, prominence over its shoulders, and whether the window contains enough cycles.
  • HeartRateEstimatorsDisagree identifies windows rejected because valid interval and periodic estimates differ by more than 12 BPM.
  • UsedAutocorrelationFallback identifies candidates obtained from strong periodicity when threshold-crossing intervals were unusable.
  • DetectedBeatPeaks reports how many prominence-qualified peaks remained after refractory-period suppression.
  • RawBeatIntervals reports how many peaks produced an interval in the configured BPM range.
  • DetectedBeatIntervals reports how many raw intervals passed consistency filtering. At least four are required for a candidate and full interval evidence.
  • FingerDetected indicates whether both optical channels exceed the contact threshold.
  • IsSettling indicates that initial finger-placement samples are still being discarded.

Movement, changing finger pressure, ambient light, clipping near the selected pulse width's full scale, or FIFO overflow can invalidate the result. Adjust the red and infrared LED currents if either channel regularly clips or has an inadequate signal.

Motion rejection treats a one-sample red or infrared DC change above 4% as abrupt movement. A minimum threshold of 750 ADC counts prevents ordinary low-level noise from restarting collection. These are empirical processing thresholds rather than MAX30100 register or datasheet values. Both can be configured when constructing either processor:

Max30100PulseOximeterOptions options = new Max30100PulseOximeterOptions
{
    SampleRate = 100,
    WindowSeconds = 8,
    AbruptChangeRelativePercent = 5,
    MinimumAbruptChangeCounts = 1000,
    MinimumHeartRate = 40,
    MaximumHeartRate = 180,
    MinimumHeartRateQuality = 45,
    MinimumAutocorrelationQuality = 55,
    AnalysisIntervalSeconds = 2,
};

Max30100PulseOximeterWorker pulseOximeter = new Max30100PulseOximeterWorker(
    options,
    queueSeconds: 2,
    publishRate: 10);

Max30100PulseOximeterOptions initializes every property to the recommended default. Only properties that need tuning have to be assigned. new Max30100PulseOximeter() and new Max30100PulseOximeterWorker() use the same default options automatically.

Increasing either value makes motion detection less sensitive and avoids unnecessary restarts, but allows larger movement artifacts into the analysis window. Decreasing either value rejects smaller disturbances but can repeatedly restart collection because of normal optical noise. A value of zero disables that threshold component; setting both to zero treats every nonzero sample change as abrupt movement.

HeartRateCandidateTolerance defaults to 12 BPM. Estimates within the tolerance are smoothed into the established candidate. A larger difference is tracked as an alternate candidate and must repeat before replacement. Lower values resist sudden jumps but follow real heart-rate changes more slowly; higher values follow changes faster but can merge noise or harmonic estimates into the current rate.

HeartRateCandidateMissLimit defaults to three analysis windows. A confirmed value is marked as held for up to missLimit - 1 consecutive invalid windows and is cleared when the limit is reached. At the default two-second analysis interval, a limit of three holds through approximately four seconds of weak analysis and clears on the analysis around six seconds. Signal resets such as finger removal, abrupt motion, or dropped samples still clear immediately.

The remaining high-value processing thresholds are configurable through either processor constructor:

  • FingerDetectionThreshold sets the raw ADC contact threshold for both channels. Increase it to reject ambient-light and no-finger baselines, or decrease it for modules with lower optical output.
  • MinimumHeartRate and MaximumHeartRate define the accepted BPM range and the interval/autocorrelation search range.
  • HeartRateEstimatorAgreementTolerance sets how closely interval and autocorrelation BPM estimates must agree.
  • MinimumHeartRateQuality and MinimumAutocorrelationQuality reject weak interval and periodic estimates. Higher values are stricter.
  • MinimumBeatIntervalCount controls the amount of consistent interval evidence required. It can be set from 2 through 20.
  • StrongCandidateQuality controls when a candidate with enough intervals can be accepted immediately instead of waiting for another analysis window.
  • MinimumHeartRatePerfusionIndex rejects infrared pulse signals whose AC amplitude is too small relative to their DC level.
  • MinimumOxygenSaturationPerfusionIndex, MinimumOxygenSaturationCorrelation, MinimumOxygenSaturationRatio, and MaximumOxygenSaturationRatio gate the experimental SpO₂ estimate.

These are empirical defaults, not medical limits or MAX30100 register values. Loosening them can produce readings sooner but increases false or unstable results. Tightening them improves rejection but can make readings unavailable on weak signals.

When both channel perfusion indexes are below the normal 0.10% threshold but remain at least 0.04%, the processor can use adaptive weak-signal validation. It requires correlation of at least 0.90 and a ratio that changes by no more than 0.10 for three consecutive analysis windows. This allows stable low-amplitude signals without treating a single weak window as valid. The thresholds are configured with MinimumWeakSignalOxygenSaturationPerfusionIndex, MinimumWeakSignalOxygenSaturationCorrelation, WeakSignalOxygenSaturationRatioTolerance, and WeakSignalOxygenSaturationConfirmationCount. Weak-signal values remain experimental and are marked by UsedWeakSignalOxygenSaturation.

When beat detection temporarily fails but infrared perfusion remains above the heart-rate threshold and red/infrared correlation remains valid, StableOpticalSignalHeartRateMissLimit replaces the normal miss limit. Its default limit is six, so the last confirmed rate is held through five consecutive invalid analysis windows and cleared on the sixth. The normal HeartRateCandidateMissLimit defaults to three, holding through two invalid windows and clearing on the third. This keeps SpO₂ available through short waveform-shape changes while exposing the BPM as held. Finger removal, abrupt motion, FIFO loss, or loss of optical coherence still resets the signal immediately or uses the shorter normal miss limit.

Filter and beat-detection heuristics are also configurable:

  • BaselineTrackingRate controls how quickly the baseline follows slow optical-level changes. Increasing it rejects faster drift but can remove part of a slow pulse waveform.
  • SignalSmoothingRate controls how quickly the smoothed AC signal follows the baseline-removed input. Increasing it preserves faster detail and noise; decreasing it produces stronger smoothing.
  • FilterSettlingSeconds controls how much of the start of each filtered window is excluded while filter state converges.
  • BeatDetectionThresholdFraction sets the minimum positive peak height as a fraction of infrared AC RMS. Increasing it rejects small fluctuations but may miss weak beats.
  • MinimumBeatDetectionThreshold prevents near-zero RMS windows from producing peaks from numerical or ADC noise.
  • PeakDetectionProminenceFraction sets the required rise from the preceding local valley as a fraction of infrared AC RMS. Increasing it rejects secondary waveform details but can reject shallow pulses.
  • BeatRefractoryPeriodFraction sets the refractory period relative to the shortest interval allowed by MaximumHeartRate. Competing peaks inside this period are reduced to the strongest peak.
  • MinimumAcRms is the minimum infrared amplitude for heart rate and minimum red amplitude for SpO₂.
  • MissedBeatIntervalFactor identifies unusually long intervals that are divided by two to compensate for one missed crossing.
  • BeatIntervalToleranceFraction controls how far an interval may differ from the median before it is rejected.

Filter rates are normalized by the configured sample rate internally, so the defaults retain comparable behavior at supported sensor sample rates. These parameters describe processing behavior rather than exact analog cutoff frequencies.

Contact and processing timing can be tuned independently:

  • FingerLossSignalResetSamples clears partial waveform analysis after consecutive no-contact samples.
  • FingerRemovalDebounceSamples confirms complete removal and must be greater than the signal-reset count.
  • InitialFingerSettlingSeconds discards pressure and optical transients after initial contact.
  • MotionRecoverySettlingSeconds discards samples after abrupt motion or temporary contact loss.
  • AnalysisIntervalSeconds controls how often a full analysis runs after the window is filled. Shorter intervals improve responsiveness but increase processing load.

Candidate confirmation and smoothing are configurable:

  • HeartRateConfirmationCount controls how many matching normal candidates are needed.
  • AlternateHeartRateConfirmationCount controls how many matching candidates are needed to replace an established rate after a large change.
  • HeartRateCandidateSmoothingFactor, HeartRateSmoothingFactor, and OxygenSaturationSmoothingFactor are the weights of the newest values. A value of one disables smoothing; smaller values respond more slowly.

Estimator fusion can be adjusted with MinimumAutocorrelationChannelCorrelation, EstimatorDisagreementQualityFactor, and AutocorrelationFallbackQualityFactor. The correlation threshold gates periodic fallback. The quality factors reduce confidence when estimators disagree or when only autocorrelation supplies a candidate.

IntervalDeviationQualityPenalty controls how strongly beat-interval variation reduces quality. The interval-evidence score uses MinimumBeatIntervalCount, so changing the required interval count also changes the amount of evidence needed for a full score.

Autocorrelation behavior is configurable:

  • AutocorrelationTargetSampleRate controls decimation before autocorrelation. Lower rates reduce processing work but provide coarser lag resolution.
  • AutocorrelationHarmonicThreshold controls how strong the half-lag peak must be before it replaces a likely two-beat period.
  • AutocorrelationShoulderFraction controls how far from a peak its prominence is measured.
  • AutocorrelationProminenceForFullQuality controls the prominence needed for full prominence evidence.
  • AutocorrelationCyclesForFullQuality controls how many waveform cycles are needed for full cycle evidence.

The worker's publishRate constructor argument controls how many state snapshots are requested per second. Lower values reduce locking and copying; higher values make raw values and progress more responsive. The actual interval is an integer number of sensor samples and can therefore be slightly slower than the requested rate.

The processor removes residual linear drift from each filtered analysis window before calculating perfusion and detecting beats. This prevents a gradual pressure or temperature trend from inflating perfusion and producing misleading low-rate peaks.

The processor defaults to a six-second analysis window, and the ESP32 sample uses an eight-second window. The additional interval evidence makes it less likely that a short noise burst or harmonic will replace an established BPM candidate. Candidates based on only three intervals are rejected. A candidate more than 12 BPM from the existing candidate must also repeat before it replaces that candidate.

The interval and autocorrelation estimates must normally agree within 12 BPM. If interval detection is unusable but autocorrelation has at least 55% confidence and red/infrared correlation is at least 0.75, the periodic estimate can be used at reduced quality and still requires normal multi-window candidate confirmation. When both estimators are individually valid but disagree, that analysis window is rejected instead of selecting one arbitrarily.

After a heart rate has been confirmed, up to two weak or disagreeing windows retain the last value and expose IsHeartRateHeld. This prevents display flicker while making stale data explicit. A third consecutive miss clears the result.

Autocorrelation is evaluated on a waveform decimated to approximately 25 samples per second. Pulse frequencies remain well below that rate, while the reduced work prevents signal analysis from starving FIFO acquisition on constrained devices. Parabolic peak interpolation recovers finer BPM resolution than the decimated lag spacing, and half-rate harmonic correction prefers a credible one-beat period when the strongest peak spans two beats. The worker defaults to a two-second processing queue. A rising processing dropped-sample count means the processor still cannot keep up; those losses reset analysis because waveform timing is no longer continuous.

ReadTemperature() returns a UnitsNet Temperature containing the MAX30100 internal die temperature. It is not a body, skin, or ambient-temperature measurement:

UnitsNet.Temperature dieTemperature = sensor.ReadTemperature();
Debug.WriteLine($"Sensor die temperature: {dieTemperature.DegreesCelsius:F2} °C");

Typical output from the sample

This is typically what you can get from the sample. Collecting can take a bit of time but if you are stuck in this, move your finger or use another one. Hold without moving the finger on the sensor.

MAX30100 part ID: 0x11, revision: 0x05
Place a finger on the sensor.
Place a finger on the sensor.
Collecting a stable signal. IR: 43482, red: 45127, perfusion: 0.00%, red perfusion: 0.00%, correlation: 0.00, ratio: 0.00, quality: 0%, candidate: 0.0 BPM, periodic: 0.0 BPM/0%, AC RMS: 0.0, threshold: 0.0, prominence: 0.0, peaks: 0, intervals: 0/0, diagnostic: filling window (29%).
Collecting a stable signal. IR: 42277, red: 43792, perfusion: 0.00%, red perfusion: 0.00%, correlation: 0.00, ratio: 0.00, quality: 0%, candidate: 0.0 BPM, periodic: 0.0 BPM/0%, AC RMS: 0.0, threshold: 0.0, prominence: 0.0, peaks: 0, intervals: 0/0, diagnostic: filling window (80%).
Pulse: 84.2 BPM, SpO2: 97.3%, perfusion: 1.42%, red perfusion: 0.72%, correlation: 0.99, ratio: 0.51, periodic: 0.0 BPM/0%, quality: 70%.
Pulse: 84.2 BPM (held), SpO2: 97.3%, perfusion: 1.34%, red perfusion: 0.71%, correlation: 0.99, ratio: 0.53, periodic: 0.0 BPM/0%, quality: 0%.
Pulse: 84.0 BPM, SpO2: 97.2%, perfusion: 0.54%, red perfusion: 0.28%, correlation: 0.99, ratio: 0.52, periodic: 84.9 BPM/56%, quality: 63%.
Pulse: 85.0 BPM, SpO2: 97.3%, perfusion: 0.51%, red perfusion: 0.26%, correlation: 1.00, ratio: 0.51, periodic: 87.6 BPM/68%, quality: 71%.
Pulse: 87.8 BPM, SpO2: 97.1%, perfusion: 0.62%, red perfusion: 0.33%, correlation: 1.00, ratio: 0.54, periodic: 83.5 BPM/52%, quality: 60%.
Pulse: 89.0 BPM, SpO2: 97.0%, perfusion: 0.66%, red perfusion: 0.34%, correlation: 1.00, ratio: 0.52, periodic: 78.8 BPM/53%, quality: 52%.
Pulse: 85.9 BPM, SpO2: 97.1%, perfusion: 0.53%, red perfusion: 0.27%, correlation: 1.00, ratio: 0.51, periodic: 82.9 BPM/42%, quality: 97%.

Oxygen-saturation calibration

The default conversion is:

SpO₂ = 110 - 25 × RatioOfRatios

This is a generic approximation, not a factory calibration for a particular MAX30100 module. Optical cover material, sensor spacing, LED characteristics, current, pulse width, finger placement, skin properties, temperature, and mechanical pressure can all change the relationship.

CalibrateOxygenSaturation performs an ordinary least-squares fit of:

reference SpO₂ = intercept + slope × RatioOfRatios

using paired ratio and reference measurements:

double[] ratios = new double[]
{
    0.52,
    0.61,
    0.73,
    0.88,
    1.02
};

double[] referenceOxygenSaturations = new double[]
{
    99.0,
    97.0,
    94.0,
    90.0,
    86.0
};

pulseOximeter.CalibrateOxygenSaturation(ratios, referenceOxygenSaturations);

Debug.WriteLine(
    $"Calibration intercept: {pulseOximeter.OxygenSaturationCalibrationIntercept:F4}, "
    + $"slope: {pulseOximeter.OxygenSaturationCalibrationSlope:F4}");

The arrays are examples of the required shape only; they are not calibration data and must not be copied into an application.

For Max30100PulseOximeterWorker, calibration is applied asynchronously on the processing thread. Read the coefficient properties after the worker has processed the request, or store the coefficients printed after calibration and restore them at startup:

pulseOximeter.SetOxygenSaturationCalibration(112.35, -27.42);

Use this procedure to collect calibration data:

  1. Keep the exact production configuration unchanged: sensor module, optical cover, mounting, sample rate, pulse width, LED currents, and analysis settings.
  2. Use a calibrated optical simulator or collect measurements under qualified clinical supervision with a suitable reference instrument. Never intentionally reduce a person's oxygen level to create calibration points.
  3. Record reference SpO₂ and RatioOfRatios simultaneously. Do not pair readings taken at different times.
  4. Accept a point only after the analysis window is full and readings are stable. As a starting criterion, require valid heart rate, signal quality of at least 80%, channel correlation of at least 0.90, and at least 0.10% perfusion in both channels.
  5. Average several consecutive, independent analysis results at each reference level. Do not treat duplicated sample output between processor updates as separate observations.
  6. Collect at least five distinct reference levels distributed across the intended measurement range. Many measurements confined to 95–100% do not provide enough ratio spread to determine a reliable slope.
  7. Remove points affected by motion, clipping, FIFO loss, changing pressure, or failed validity checks before fitting.
  8. Fit the coefficients with CalibrateOxygenSaturation, then save the reported intercept and slope in application configuration.
  9. Validate the saved coefficients on separate measurements that were not used for fitting. Report error across the range, not only at one convenient saturation level.
  10. Repeat calibration whenever the module, enclosure, optical path, LED settings, or mounting changes.

The fitted slope must be negative, the ratios must not all be equal, and at least two paired points are required. In practice, two-point calibration is highly sensitive to noise and is not recommended. Least-squares fitting reduces random error but does not make the MAX30100 implementation medically validated.

The MAX30100 does not directly return beats per minute or oxygen saturation. The estimates from this binding are intended for experimentation and demonstration only. They are not calibrated, validated, or suitable for medical diagnosis.

Edit this page