Accuracy, Precision and Noise of Gas Sensors – Definitions

Introduction

Accu­ra­cy“ is one of the most important qua­li­ty cri­te­ria of a gas sen­sor. But some­ti­mes „pre­cis­i­on“ rather than „accu­ra­cy“ is requi­red for sen­sors. And unfort­u­na­te­ly, in prac­ti­ce, high mea­su­re­ment accu­ra­cy often has to be bought at the pri­ce of redu­ced pre­cis­i­on and vice versa.

Data sheets often fail to decla­re what is actual­ly meant by accu­ra­cy and pre­cis­i­on. For exam­p­le, „signal reso­lu­ti­on“ is often sim­ply declared as accu­ra­cy, which is defi­ni­te­ly not cor­rect in most cases.

And to make mat­ters even more com­pli­ca­ted: the mea­su­red value dis­play­ed on the sen­sor rea­ding is limi­t­ed by the „signal reso­lu­ti­on“ and „sen­si­ti­vi­ty“ of the sensor.

On the other hand, the abo­ve terms are defi­ned slight­ly dif­fer­ent­ly in dif­fe­rent are­as of application.

This artic­le aims to avo­id misun­derstan­dings: In the fol­lo­wing, the terms are defi­ned. In a fol­lo­wing artic­le examp­les are given to show in which appli­ca­ti­ons „accu­ra­cy“ is requi­red and whe­re „pre­cis­i­on“ is needed.

High pre­cis­i­on, low accuracy:

HighPrecision_lowAccuracy

High accu­ra­cy, low precision:

Accuracy precision
signal noise and probability

Detail of the pro­ba­bi­li­ty dis­tri­bu­ti­on χ of a noi­sy signal around the mean value xm in rela­ti­on to the real con­cen­tra­ti­on xr with the mea­su­re­ment error e:

 

Gaussian signal probability distribution

The signal out­put from the sen­sor defi­ning the sen­si­ti­vi­ty S, also con­ta­ins sen­sor-indu­ced noi­se!

Sensitivity

Internal Digital Resolution

Inter­nal Digi­tal Reso­lu­ti­on is spe­ci­fied as the smal­lest detec­ta­ble incre­men­tal chan­ge of the sen­sor signal. This is limi­t­ed by the ana­log-to-digi­tal con­ver­ter (ADC) bit reso­lu­ti­on and the sen­si­ti­vi­ty S.

For exam­p­le if the sen­sor abo­ve with a sen­si­ti­vi­ty S = 1E‑4 V/ppm is sam­pled with an 16-bit ADC, a inter­nal signal reso­lu­ti­on of 6.5 digit/ppm will result, or 0.15 ppm/digit resolution.

 

gas concentration signal resolution

Displayed Reading Resolution

The dis­play reso­lu­ti­on is defi­ned by the full-sca­le-ran­ge (FSR) sca­ling and the dis­play­ed smal­lest unit of phy­si­cal chan­ge dis­play­ed. A high reso­lu­ti­on means small chan­ges of gases can be displayed.

For exam­p­le, if the digi­tal out­put of the dis­play is sca­led to show the gas con­cen­tra­ti­on in % (ins­tead of ppm) and the dis­play shows two deci­mal places, the dis­play reso­lu­ti­on is Δc = 0.01% gas con­cen­tra­ti­on, which is much lar­ger than the inter­nal signal reso­lu­ti­on of the exam­p­le above.

 

2-digit resolution

Reso­lu­ti­on is not the same as the pre­cis­i­on of a sensor! 

Nor­mal­ly, the reso­lu­ti­on of a digi­tal dis­play is set lower (less pre­cise) than the noi­se level of the mea­su­re­ment signal so that the value dis­play­ed does not chan­ge constantly.

In con­trast, when mea­su­ring values are trans­mit­ted digi­tal­ly, a reso­lu­ti­on finer than the noi­se level is often sel­ec­ted. This is becau­se it allows the recei­ver to pro­cess the noi­sy signal opti­mal­ly (noi­se suppression).

Detectability

Improving the Dectection Limit by Averaging

The LOD can be impro­ved by aver­aging the noi­sy mea­su­re­ment signal. In a Gaus­si­an noi­se signal, the LOD impro­ves by the squa­re-root of the aver­a­ged data points.

Exam­p­le:

Aver­aging the signal over 4 con­se­cu­ti­ve mea­su­re­ments will impro­ve noi­se by a fac­tor 2 as well as the lower limit of detectivity.

The limit of impro­ving the detec­ti­vi­ty is given by sys­te­ma­tic errors such as ther­mal drift and is very well dis­play­ed in a Allan-Vari­ance dia­gram [3].

 

Exam­p­le:

Same signal as abo­ve, but with moving aver­a­ged of N=16 samples aver­a­ged, resul­ting in a 4 times smal­ler noi­se figu­re and bet­ter detec­tion limit.

Detectability of averaged signal

References

[1] DIN 55350–13: „Con­cepts in qua­li­ty and sta­tis­tics; con­cepts rela­ting to the accu­ra­cy of methods of deter­mi­na­ti­on and of results of determination“

[2] ISO 5725–1 : „Accu­ra­cy (true­ness and pre­cis­i­on) of mea­su­re­ment methods and results – Part 1: Gene­ral prin­ci­ples and definitions.“

[3] Hae­ri, H., Beal, C.E., and Jerath, K. (2021). Near­op­ti­mal moving avera­ge esti­ma­ti­on at cha­rac­te­ristic times­ca­les: An allan vari­ance approach. IEEE Con­trol Sys­tems Let­ters, 5(5), 1531–1536. doi:10.1109/LCSYS. 2020.3040111.