IEEE MELECON 2002, May 7-9,2002, Cairo, EGYVT.

Calibration and Error Correction Algorithms

for Smart Pressure Sensors
M. Mozek, D. Vrtacnik, D. Resnik, U. Aljancic, M. Cvar, !3. Amon
Faculty of Electrical Engineering, University of Ljubljana,
Trzaska c. 25, 1000 Ljubljana, Slovenia
Tel. +386 1 4768380, Fax. +386 1 4264630
E-mail: matej [email protected]

Abstract point. In case of smart pressure sensor its output is a

floating-point number in SI units ready for further
The essential properties of smart sensor error computation. The engine itself is reversible: A
correction algorithms and calibration procedures
floating-point input value can be transformed to an
according to IEEE1451.2 standard are presented. Using integer representing raw contrcll input of an actuator,
a modem design microcontroller it is possible to
so its formulation is also valid for actuators. The
implement STIM in a single chip design, but due to
correction engine defined by the standard uses a
complexity of error correction and calibration
multivariate polynomial with unlimited degree of
algorithms it is preferable to distribute these tasks
polynomial. In order to limit the degree of used
between several processors available in a network. A
polynomial, segmentation is used; different sets of
description of a simple calibration algorithm intended
coefficients may be used between specified limits of
for implementation in modular STIM design will be
the input values. The formula below describes the
given. Other calibration and error ' correction
general formulation of correction engine:
approaches were tested on designed and fabricated
smart sensor network. Their implementation is
discussed and the results of the most promising
bivariate spline approximation algorithm will be 1 s 130 @
presented. Where Xk variables represent diita from a set of inputs
(either transducer-side or corrected), D(k) are integer
Keywords smart pressure sensor, calibration polynomial degrees, C,,, are floating point coefficients
procedure, error correction algorithm and Hk are offsets stored to calibration TEDS.
Mathematical formulation of (1.1) can also be
1. INTRODUCTION presented as two-dimensional calibration in Fig. 2.1.
Measurement plane is divided into several cuboid
Smart sensors represent today's peak in sensor shaped cells, bounded with segment borders HL.
applications for industry automation. , Standard Standard specifications leave a method of calibration
IEEE1451.2 [l] defines 8n interface between to user choice. However when choosing the correction
transducers and microprocessors useful in industrial engine algorithm one must consider several factors
' automation. The standard defines various features of a
which impact the choice for c,alibration or correction
smart sensor device from physical interface, triggering,
engine algorithm:
interrupt signaling and hot swap capability. Perhaps the
most important feature of a STIM is a TEDS - Amount of non-volatile memory available in
(Transducer Electronic Data Sheet), which is a digital STIM for storing calibration TEDS entries C,,,
presentation of a STIM description, stored in a large and D(k).
portion of non-volatile memory located on a STIM
(Smart Transducer Interface Module) side. It consists - Availability and precision of floating point
of eight parts, but calibration procedures and operations in system.
algorithms presented in this paper will be mainly - Minimum of polynomial degrees D(k), which
related to a part of TEDS, named calibration TEDS. produce a fit within desired sensor uncertainty.
The main feature of calibration TEDS is a correction
engine. Correction is the application of a specified 2. CALIBRATION PROCEDURES
mathematical function upon transducer data from one
of more STIM channels. Correction engine Most general approach to calculation of calibration
simultaneously performs correction and conversion to coefficients Ci,,, is to use tensor product spline
and from physical unit values presented in floating approximation [3, pp. 2671. For sensor error correction

0-7803-7527-O/02/317.0002002 IEEE. 240

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on December 15,2023 at 11:01:32 UTC from IEEE Xplore. Restrictions apply.
 the optimal algorithm is spline interpolation. Simpler three calibration inputs (e.g. for pressure sensor
methods are also applicable such as polynomial applications: temperature readout, pressure readout and
approximation or a lookup table. The latter is the ADC preamplifier gain). Generally the number of
fastest, but consumes enormous amount of memory, variables has to be considered unlimited, hence tensor
often unavailable by STIh4. Another simple method product approach to spline interpolation is used. The
applicable for small microconuollers is polynomial basic concept of tensor product splines is to break a
approximation. When noise is present, the polynomial multidimensional problem into series of univariate
interpolation rather fits noise than signal, hence fit spline interpolations, described in detail in [4], [7] and
polynomial degrees rise rapidly. Approximation using [SI. Algorithms for univariate and multivariate (tensor
high degree polynomials also tends to produce severe product) splines are given in [4] and [7]. Another
oscillations. Because of all those inconveniences, algorithm has also been tested for purposes of
piecewise polynomials are normally used instead of calibration, performed solely by STIM: It is a
polynomial approximation functions. Set of combination of piecewise polynomial interpolation and
coefficients C,J,p,given in (1.1) can be also determined a look up table. Input parameter of calibration is
differently: One can solve a system of linear equations desired sensor uncertainty, and the amount of memory
formed by all calibration points. This algorithm is available for calibration TEDS storage (i.e. maximum
simple and fast, but tends to produce errors even in number of calibration points). Calibration points are
calibration points - hence it is unusable. Why? The taken as dense as possible. After sweeping entire
system to be solved is based on a Vandermonde temperature and pressure range a calibration algorithm
matrix. Such systems are generally ill conditioned. is started. At a constant temperature first calibration
Upper calibration options should be considered, when point is considered the offset point. Algorithm moves
the calibration is to be performed solely by STIM. two sampled points forward and performs polynomial
interpolation of midpoint. If the resulting midpoint
interpolation error is within specified uncertainty
interval, algorithm moves to the next point and
calculates error for new interval. The tested point is
assumed to be within sensor uncertainty and cannot be
considered a calibration point. Eventually the
difference isn’t in range of uncertainty interval and
hence the next calibration point is determined. After
scanning entire pressure range at a single temperature,
the calibration points are counted and if there are too
many to fit into dedicated portion of TEDS, algorithm
is restarted and the degree of interpolating polynomial
is increased. The polynomial degree is usually limited
Fig. 2.1: Multivariate polynomial segmentation to quadratic or cubic. The sensor’s response should not
representation exhibit severe temperature dependency, hence a look
up table approach is used for temperature dependence
Another approach in determining a set of coefficients calibration.
C,J,p, is to use spline approximation theory. The basic
problem of a spline theory is how to set the net of
knots. Most techniques require equally spaced samples 3. CALIBRATION SETUP
or at least a linear transformation of coordinates, which STIM calibration setup consists of a STIM connected
is often time consuming, since a dense net of
to NCAP, which is connected to network. The latter
calibration points has to be measured first. Next equal connects several NCAPs, dedicated computers and
calibration points have to be selected from dense net, various gateways (as is the case with Ethernet).
complying with an orthogonal system selected. Calibration can be performed in three ways:
Obtaining a dense net of measurements is very time
consuming, hence reduced set of measurements is 1. Calibration procedure is performed exclusively by
normally used. Determining the minimum set of input STIM.
measurements is often transducer specific.
2. Calibration procedure is performed by NCAP and
Transducer’s response is often depending on secondary
STIM.
variables, so calibration procedures have to support
multidimensional problems. Standard spline algorithms 3. Calibration procedure is performed by a remote
are available [4] for single or two dimension problems. placed server, local NCAP and a STIM connected
The calibration of example sensor requires at least to it.

24 1

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on December 15,2023 at 11:01:32 UTC from IEEE Xplore. Restrictions apply.
 First option is rare, due to computation demanding n
correction engine, which has to be implemented in
STIM. Calibration algorithm for this option will be
described in detail later. Calibration procedure of a
second setup is controlled by NCAP, which collects
the data sent by STIM. The NCAP also ‘controls a
reference sensor and a digitally controlled pressure
source. Desired pressure is directed from NCAP to
pressure source. When an arbitrary pressure is set,
NCAP triggers both STIM channels, samples the
values and stores them for correction engine
processing.
The second option is the most frequent (see Fig. 3.2),
since NCAP normally supports hardware floating point
operations (e.g. Agilent Technologies BFOOT-66501
is based on NETtARM processors). U L
The last option differs from the second option only in
NCAP role. The samples are transmitted via network Fig. 3.2: Calibration setup of modular pressure
to remote server, which performs correction sensor
computation. The third option is often unnecessary,
since NCAPs perform their task well and remote 3.2 Using the correction model for
servers often aren’t available. In presented example we configuration dependent measurements
used second option during calibration process. The
NCAP was implemented in local PC. Its firmware was The correction model can be i&o used to alter the
programmed as an application extension of Labview 5 . computations and presentation depending on
measurement configuration options detectable by
STIM (e.g. A four-channel voltage measuring STIM
3.1 Calibration setup of a modular STIM could be also used for phase order determinatioi: in
The calibration process of a modular STIM consists of three phase systems). Another use of the correction
two subsequent steps. First step is to perform factory model is to derive secondary measurements from the
calibration on a signal-conditioning module. Second primary measurements available from individual
step is a calibration described in previous section. channels. For example if two independent pressure
Factory calibration can be achieved in two ways: transducers were assigned to chimels 1 and 2, then a
automatic and semi-automatic. virtual channel (channel 3), could be assigned to
In automatic mode a D/A converter is used to adjust represent differential pressure. Channel three is virtual.
the full-scale pressure value to ADC readout between In equation (1.1) the coefficients would have to be set
OxFFFO and OxFFFF. As you can see, the ADC to HI=H~=C”CII=O, Clo=l and Col=-l. This
headroom is of designer choize. In the tested STIM a technique can be used for cakulating any derived
10-bit D/A converter and a 16-bit A D converter were quantity representable as a multivariate polynomial
used, yielding less than 0.2% of headroom. The value expression of primary values (eg. power is calculated
sampled on a D/A converter is stored into STIM using data from current and voltage measuring
internal data for further processing. channel). Another consideration is related to use of
Other option is to adjust the reference level via a splines. STIM’s output is not necessarily a directly
resistor divider, where remotely controlled resistor sampled value, it can be a mathematical transformation
laser trimmer is used. The factory calibration setup of a measured value. Splines have an obvious
comprises of a main computer, a laser trimmer and a advantage over simpler calibration algorithms:
signal-conditioning module of a modular STIM. Once Numerous mathematical transformations can be
again, the computer controls the laser, which sets the performed easily on splines. Algorithms for spline
full-scale readout of ADC to be between desired levels. integration and derivation have been programmed. The
After performing full-scale calibration, an offset derivatives of a spline are continuous functions. (e.g. if
voltage is calibrated by the signal - conditioning a temperature or pressure gradient are to be measured).
module (by executing one of module calibration Algorithms for basic mathematical transformations are
commands), before it is considered valid. Factory also provided in [4]. In case of presented modular
calibration is generally cheaper, but gives a fixed STIM with 1 bar relative pressure sensor the
sensor range and an order of decade less efficient calibration was calculated using bivariate cubic spline
headroom. interpolation. The auto range option of a sensor was

242

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on December 15,2023 at 11:01:32 UTC from IEEE Xplore. Restrictions apply.
 achieved by adding an additional calibration dimension calibration procedures and error correction algorithms
to sensor, defined by gain of the preamplifier in signal- has been presented. A simpler calibration algorithm
conditioning module. The correction calculation is has been proposed and was implemented into STIM,
carried out by the NCAP for sake of data throughput but shows significant discrepancies in error correction
rate. Fig. 4.3 presents the FSO error plot versus ADC due to its simplicity and single precision floating point
readouts of both channels. Results after performing calculation. A calibration algorithm, which was
bivariate least square spline approximation are programmed in STIM, is presented. For accurate
presented in Fig. 4.4. analysis a tensor product spline approximation has also
been implemented in NCAP. Its code is available on
4 MEASUREMENTS AND RESULTS http://paris.fe.uni-Ij .si/matej/ Web site under Programs
item. Described example of smart pressure sensor is
The calibration procedure for a pressure has been very illustrative for STIM design. Future work will be
implemented into STIM. After applying cubic spline focused on optimization of the described prototype and
calibration algorithm upon raw sensor data an error its integration in multi-drop communication protocols
graph, shown in Fig. 4.3 has been plot, which also (e.g. IEEE145 1.3). Various signal-conditioning
determines total sensor uncertainty. The resolution of modules will have to be designed in order to achieve
pressure sampling ADC was 10 bit and the resolution higher bit resolutions and/or conversion rates.
of temperature sampling ADC was 8bit.
References
IEEE Std. 1451.2 D3.05 - Aug 1997 “IEEE
standard for a smart transducer inregace for
sensors and actuators - Transducer to
microprocessor communication protocols and
transducer electronic data sheet (TEDS)
formats” Institute of Electrical and Electronics
Engineers, September 1997
Lee Bradford, Qiang Li “Choosing Designs of
Calibration Transducer Electronic Data Sheets”
Hewlett Packard tech report HPL-98-166,
September 1998
Press et.al. “Numerical recipes in C‘,second
Fig. 4.3: Error plot of a bivariate spline edition, Cambridge University Press, 1992.
approximation C. de Boor “Spline toolbox for use with
MATLAB, Mathworks Inc., January 1998
C. de Boor “A practical guide to splines”,
Springer Verlag, 1978
Moore J. “An IEEE1451 Standard Transducer
Interface Chip”, IEEE International Solid State
Circuits Conference. pp. 225-227, 1998.
Schumaker Lany L. “Spline finctions: basic
theory”, Krieger Publishing company, 1993
Gerald Farin “Curves and =?$aces for Computer
Aided Geometric Design”, third edition,
Academic press Inc., 1993

Fig. 4.4: Bivariate spline approximation of

measured data

5 CONCLUSION
A modular derivative of a STIM has been designed.
Sensor calibration setup and calibration has been tested
on described calibration setup. A brief overview of

243

Authorized licensed use limited to: Universita degli Studi di Bologna. Downloaded on December 15,2023 at 11:01:32 UTC from IEEE Xplore. Restrictions apply.