Pulse Oximeter

Red/infrared PPG ratio and SpO2 calibration. Educational only, not medical advice.

Switch to 2D

Presets

Playback

Patient physiology

True blood state, pulse, perfusion, skin and motion.

Sensor and calibration

Scattering path ratio, device calibration line, noise and averaging.

Camera

Quality

Layers

Model checks

The science

What the clip does

A fingertip pulse oximeter has two LEDs, red (about 660 nm) and infrared (about 940 nm), on one side of the finger and a photodiode on the other. The LEDs flash in turn, with a dark slot to measure room light. Oxygenated haemoglobin (HbO2) absorbs infrared more than red; deoxygenated haemoglobin (Hb) absorbs red much more than infrared. Only the arterial blood pulses with the heart, so the small beat-by-beat (AC) part of each colour carries the arterial information and the steady (DC) part is skin, tissue, bone and venous blood.

Beer-Lambert and the ratio of ratios

Light through an absorber falls as I = I0 exp(-eps c d). When the arterial blood thickness changes by a small delta d the log-intensity changes by eps_lambda(S) c delta d, so AC/DC of each colour is proportional to that colour's extinction. Dividing the red ratio by the infrared ratio cancels the unknown blood thickness: R = (AC_red/DC_red) / (AC_IR/DC_IR) = k eps_red(S) / eps_IR(S) with eps(S) = S eps_HbO2 + (1 - S) eps_Hb. The tabulated molar extinctions used here (cm^-1 per mol/L, typical literature values, quoted from memory) are Hb 3227 and HbO2 320 at 660 nm, Hb 693 and HbO2 1214 at 940 nm. Solving for the saturation gives the closed form S = (eps_Hb,r - R' eps_Hb,ir) / ((eps_Hb,r - eps_HbO2,r) + R' (eps_HbO2,ir - eps_Hb,ir)) with R' = R/k.

Why an empirical calibration

Skin and blood scatter light, so the two colours do not travel the same path and the pure Beer-Lambert curve is wrong. This page lumps that into one path ratio k (about 1.5 reproduces a typical device). Real devices are calibrated on healthy volunteers breathing reduced oxygen, and the result is close to a straight line, shown here as SpO2 = A - B R with the illustrative values 110 and 25 (R about 0.5 gives 97 %, R = 1 gives 85 %). The sliders let you move A, B and k to see how a wrong calibration or different tissue shifts the reading. Below about 70 % there is no ethical calibration data, so real devices are not accurate there.

The estimator on this page

The pulse rate comes from the autocorrelation of the infrared signal refined by a Fourier scan. The AC amplitude of each colour is the Hann-windowed lock-in amplitude of the first three harmonics at that rate, and the DC is the window mean. The shown SpO2 is the average of the last eight updates, four per second.

What goes wrong

Carbon monoxide: carboxyhaemoglobin absorbs red like HbO2 and almost nothing in the infrared, so the device reads close to HbO2 + COHb and over-states the oxygen actually carried. Motion moves venous blood whose ratio looks like about 70 % saturation, pulling readings toward low values. Weak pulses (cold fingers), strong melanin, nail polish, ambient light and amplifier noise lower the signal-to-noise ratio and add scatter.

Simplifications and illustrative constants

Educational model: this is not a medical device and not medical advice. The pulse is a fixed shape (systolic peak plus a dicrotic bump), the finger is a single homogeneous path with one lumped scattering factor k, the static layer absorbances in the stack diagram are invented illustrative numbers, the COHb extinction at 940 nm is approximate, and skin-colour bias reported in clinical studies is not modelled (here pigment only reduces light and increases noise). Photons drawn on the finger are only an illustration of absorption, not a Monte Carlo transport calculation.

What to try
  1. Start with the healthy preset: the device reads 97-98 % and R is near 0.5. Lower the true saturation slider to 85 % and watch R climb to about 1.0.
  2. Open the calibration graph and drag k: the physics curve (cyan) moves away from the yellow device line, so the reading is off by several points even in a noise-free signal.
  3. Choose the carbon monoxide preset: the true HbO2 fraction is 64 %, but the device still shows about 77 %.
  4. Switch on motion artefact at 0.7: the reading wanders toward 70 % because moving venous blood has its own ratio.
  5. Choose the cold fingers preset: the perfusion index drops to 0.5 % and the reading scatter grows; widen the averaging window to calm it.
Loading Blender model...
Red and infrared PPG (AC divided by DC)
Calibration: SpO2 against ratio of ratios R
Reading and true saturation over the last minute
Reading error across the saturation range
Made with 🟠 Blender · Three.js