\b[Pulse oximetry] estimates arterial <oxygen saturation> noninvasively from wavelength-dependent optical signals that vary with the arterial pulse. A typical sensor combines red and infrared <light-emitting diodes> with a <photodetector>. The pulsatile component helps distinguish changing arterial blood contribution from relatively steady tissue and blood contributions; it is not a direct measurement of <arterial oxygen partial pressure>.

A primary instrument-design note implements 660 nm and 940 nm illumination, a shared <photodetector> and alternating channel acquisition. It separates pulsatile and steady components before estimating saturation.<Pulse Oximeter Fundamentals and Design by Santiago Lopez>[Sections 2.2 and 3.1.2, printed pp. 3–5 and 7, optical channels and acquisition]{ref} These wavelengths and electronics describe that implementation rather than every possible sensor.

Writing pulsatile and steady signal magnitudes as AC and DC, a common normalized optical quantity is
$$
R=\frac{\mathrm{AC}_{\rm red}/\mathrm{DC}_{\rm red}}
{\mathrm{AC}_{\rm IR}/\mathrm{DC}_{\rm IR}}.
$$
For original channel values AC$_{\rm red}=0.01$, DC$_{\rm red}=1.0$, AC$_{\rm IR}=0.02$ and DC$_{\rm IR}=1.0$, this ratio is 0.5. Halving both red-channel magnitudes leaves the ratio unchanged. This illustrates intensity normalization; it does not determine a saturation without a validated relation between $R$ and saturation.

That relation is experimentally checked against an independent reference. The FDA's technical guidance describes comparing oximeter readings with simultaneous <co-oximetry> of arterial samples during controlled inspired-oxygen changes in volunteers, and reporting an accuracy statistic.<Pulse Oximeters: Premarket Notification Submissions by the FDA>[Clinical Performance, SpO2 Accuracy: reference arterial sampling, inspired mixtures and paired readings]{ref} Such comparisons test the whole sensor–algorithm measurement chain, rather than only the arithmetic of a two-channel ratio.

One summary is the <root mean square> difference
$$
A_{\rm rms}=\sqrt{\frac1n\sum_{i=1}^n(S_{p,i}-S_{a,i})^2}.
$$
For invented paired errors $-2,0,+2$ percentage points, $A_{\rm rms}=\sqrt{8/3}\approx1.63$ percentage points even though mean error is zero. A small mean bias does not guarantee small individual errors.

Motion, weak pulsatile signals and additional absorbing <hemoglobin> species can invalidate assumptions behind a simple estimate. <Pulse oximetry> does not measure <pH>, carbon-dioxide tension or <hemoglobin> concentration. A stable displayed percentage therefore cannot establish every aspect of ventilation, oxygen content or gas exchange.


 Pulse oximetry (source code)