See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224062963

A New Calibration Method for Current and Voltage Sensors Used in Power
Quality Measurements

Conference Paper in Conference Record - IEEE Instrumentation and Measurement Technology Conference · May 2006
DOI: 10.1109/IMTC.2006.328579 · Source: IEEE Xplore

CITATIONS READS
7 7,041

3 authors, including:

Pedro M. Ramos Nuno B Brás

University of Lisbon Institute of Telecommunications
200 PUBLICATIONS 2,562 CITATIONS 15 PUBLICATIONS 75 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Pedro M. Ramos on 20 May 2014.

The user has requested enhancement of the downloaded file.

IMTC 2006 – Instrumentation and Measurement
Technology Conference
Sorrento, Italy 24-27 April 2006

A New Calibration Method for Current and Voltage

Sensors Used in Power Quality Measurements
Pedro M. Ramos, Nuno B. Brás and A. Cruz Serra
Instituto de Telecomunicações, Department of Electrical and Computer Engineering,
Instituto Superior Técnico, Technical University of Lisbon
Av. Rovisco Pais 1, 1049-001 Lisbon, Portugal
Phone: +351-218418485, Fax:+351-218418472, Email: [email protected]

Abstract – Sensor calibration plays an important role in instrument with a hole for the primary winding and are optimized to
development and accuracy. This paper presents a method to decrease total harmonic distortion (THD).
calibrate current and voltage sensors specially dedicated to power
quality measurements which can significantly reduce the distortions II. THE HALL EFFECT SENSORS
introduced by the sensors.
The first approach to sensor calibration is based on the response of
The calibration method was developed using a closed loop
sensors to DC values imposed by a calibrator. The second method is
an AC calibration, where a 50 Hz power signal is applied to the compensated Hall Effect current transducer (LA 25-NP [3])
sensors also through a calibrator. The calibration coefficients are and a closed loop compensated Hall Effect voltage transducer
obtained by minimizing the output distortion. The minimized output (LV 25-P [4]) both from LEM.
distortion is assessed by the signal to noise and distortion ratio According to the manufacturers specifications, the voltage
(SINAD) at the output of the sensor correction module. sensor has an overall accuracy of 0.9 %, linearity error below
0.2 %, nominal input voltages up to 500 V RMS, draws a
Keywords – power quality, sensor calibration, sine-fitting. nominal primary current of 10 mA and outputs a nominal
current of 25 mA. The output current is fed to a 120 Ω
I. INTRODUCTION resistor and the voltage drop in this resistor is sampled by a
data acquisition board. The current sensor as an overall
Nonlinear loads, power utilities deregulation and accuracy of 0.5 %, linearity error below 0.2 % and is
distributed generation are among the biggest contributors to configured to operate up to 8 A of nominal current with an
power quality (PQ) disturbances [1]. The need for power output nominal current of 24 mA. The sensor output is also
quality assessment has become consumer driven as industry fed to a 120 Ω resistor for current to voltage conversion.
and individual consumers are becoming increasingly aware of The sensors are assembled in a sensor box, whose basic
the importance of power quality. schematic is presented in Fig. 1. This setup is used to acquire
Inexpensive, reliable, flexible and accurate power quality simultaneously the current and the voltage supplied to a
instrumentation is based on digital signal processors with specific load.
powerful algorithms for the detection, classification and Voltage Current
measurement of the variety of PQ disturbances. Within these Sensor Sensor
instruments, two very important components are the analog to Output Output
digital converters (ADC) and the voltage/current sensors.
Sensor calibration is very important since an incorrect
calibration can mask some PQ events making them
undetectable by the algorithms. The easiest way to implement Load
Supply
sensor calibration is by applying a correction polynomial to
the acquired samples. This polynomial is also responsible for
the required scaling and basically amounts to inverting the
transfer functions of the sensors and the ADC.
In this paper, voltage and current close loop Hall effect
sensors are calibrated using different methods to estimate the Fig. 1 – Power network sensor box schematic containing current and
best coefficients for a polynomial correction function of the voltage sensors. The sensor outputs are voltages suitable for direct
measured ADC samples. connection to data acquisition boards.
Different compensation techniques were developed for
closed loop Hall Effect transducers to improve linearity [2]. The sensor box can be used to monitor the current and
Such techniques can be combined with the proposed voltage provided to a load or just to monitor the power grid
calibration method to improve overall sensor linearity. voltage in a specific access point. In this configuration no
However, the methods developed in [2] are limited to sensors load is connected.

0-7803-9360-0/06/$20.00 ©2006 IEEE 2283

 III. CALIBRATION MEASUREMENT SETUP The purpose of the calibration procedure is to find a
function F (mi ) , to be implemented in the correction module,
To calibrate the sensor box the experimental setup
that minimizes the error between the correction function
includes a PC controlled (IEEE 488.2) Wavetek 9100
calibrator. The PC is also equipped with a National output ci = F (mi ) and the sensor input signal at instant ti .
Instruments 6013 16-bit data acquisition board (DAQ) to A schematic representation of the correction procedure
acquire the sensor outputs. The voltage calibration applied to the input signal is presented in Fig. 4.
architecture is shown in Fig. 2 while in Fig. 3 the equivalent
circuit for current measurement is presented. Input Signal x (t )

Sensor

Sampling
Frequency
fS DAQ

mi Digitized Samples

Correction Correction
Coefficients
ak
Function

ci Corrected sample values

Fig. 4 – Schematic description of a sensor function correction.

The correction function is a polynomial function described

Fig. 2 – Voltage sensor calibration setup. by

M
F ( x) = ∑ ak xk (1)
k =0

where M is the order of the polynomial function.

This allows good results in the calibration process and is
also a simple computational expression which can be easily
implemented in DSP based instrumentation.

IV. DC CALIBRATION METHOD

The DC calibration method consists of controlling the

calibrator to apply a set of DC voltages to the voltage sensor
Fig. 3 – Current sensor calibration setup. and determining the ak coefficients that minimize the
distance between the corrected values ci and the imposed DC
A previously developed software platform based on
voltages in the calibrator. The same technique is also applied
LabVIEW and XML that allows easy acquisition, control of
the calibrator and process structured data coming from to the current sensor where the calibrator now sets the DC
several acquisitions was used [5]. current in the sensor box (Fig. 3).
For the voltage sensor, voltages between –400 V and
An analog input signal x(t ) is applied to each sensor. The
400 V were applied with the calibrator for a total of 800
sensor output is then digitized by the DAQ originating the different test voltages. This voltage range corresponds to
measured values mi with i = 1 N where N is the number about 52 % of the DAQ range in its ±5 V range.
of acquired samples at relative time instants ti = (i − 1) f S For each calibrator voltage, 10000 samples were acquired.
After the acquisition, the average value was determined
( f S is the DAQ sampling rate).
together with standard deviation of the acquired samples.
The digitized record of samples mi is then fed to a A weighted polynomial least-squares fit was applied varying
correction module that should output the value of x (t ) at the order of the polynomial fit from M = 1 up to M = 9 .
instant ti . This way, the sensor and DAQ transfer functions This process was also performed for the current within the
are inverted in the correction module to obtain an instant –3 A to 3 A range which accounts for 22 % of the DAC ±5 V
estimative of the sensor input signal. range.

2284
 The validity of the DC calibration procedure is assessed V. AC CALIBRATION METHOD
with a 50 Hz 230 V RMS voltage applied to the voltage
sensor. 10 sets of acquisitions each with 10000 samples were To take into account the dynamic behavior of sensors and
executed. The average power spectrum was obtained with the the fact that the main input signal is a 50 Hz sinewave, a new
DFT and the SINAD of the corrected values was determined. calibration method was developed and implemented. Instead
It should be noted that not all of the ADC range was used, so of using DC values in the calibration step, a sine wave of
the determined SINAD should not be used to determine the 50 Hz was applied by the calibrator to the sensors. For the
effective number of bits of the acquisition channel. However, voltage sensor a 230 V RMS was used while 3 A RMS were
the different values of SINAD obtained with the different used for the current sensor.
order of the polynomial correction function can be used to In this situation, since the input signal is a 50 Hz sine, the
compare the calibration procedures. corrected values ci should correspond to the digitized
For the current sensor a 50 Hz, 3 A RMS current was samples of the input signal at the ADC sampling rate.
used, and the average spectrum was also determined to However, there is some information missing that is very
estimate the SINAD and compare the different order important to compare the input signal x(t ) with the corrected
correction polynomial using the DC calibration results.
Fig. 5 shows the SINAD as a function of the polynomial values ci . This information is the exact value of the
order ( M ). It can be seen that, for the voltage sensor, there is normalized signal frequency (although the calibrator
a steep decrease in the SINAD value for polynomial order frequency is known with very good accuracy the same does
above 2. For the current sensor, the SINAD also shows some not apply to the DAQ sampling frequency) and the input
variation but not as relevant as the change in the voltage signal phase. In fact, since the acquisition of samples mi is
sensor. not triggered by the actual signal x(t ) , the initial phase of the
60
samples mi is a random variable with a uniform distribution.
In short, to compare the record ci with the samples of x(t )
58
sampled at f S , the exact signal frequency of x(t ) and the
56
phase corresponding to the first sample must be determined.
SINAD [dB]

54
To achieve this objective, a four-parameter sine-fitting [6] is
used to estimate the phase ( φ ) and the frequency ( f ) of the
52
corrected values ci . Afterwards a virtual version of the input
Voltage
50 signal is now available
Current
48
xˆ (t ) = ARMS 2 cos ( 2πf t + φ ) (2)
46
1 2 3 4 5 6 7 8 9
Polynomial Order (M) where ARMS is calibrator imposed value (230 V for the
Fig. 5 – SINAD for the current and voltage as a function of the voltage and 3 A for the current). The polynomial correction
polynomial correction order for DC calibration. The average power function can be assessed by comparing the corrected samples
spectrum was obtained from 10 sets of 10000 samples. The SINAD ci with the sampled record of (2) at the sampling rate f S
was determined from the average power spectrum.

It is clear that the SINAD is not improved when using  i −1  f 

higher order polynomial correction functions. Since in terms
xˆi = xˆ  t =  = Aef 2 cos  2π (i − 1) + φ  . (3)
 fS   fS 
of SINAD, the linear adjustment (polynomial order M = 1
corresponds to a linear correction) is equivalent to not
correcting the harmonic amplitudes, there are no advantages To compare the records ci and xˆi , the sum of the squared
of using higher orders polynomials. errors can be used
The unimproved SINAD is mostly due to the fact that
calibration was performed at DC which doesn’t really take N
ε = ∑ ( xˆi − ci ) .
2
(4)
into account the distortions that the sensors introduce at i =1
50 Hz. Also, the DC calibration method is not directly
intended to improve the SINAD value. It merely adjusts the This cost function amounts to quantify the distance
DC output values to the expected DC input values. However, between the corrected record ci and the input sine signal.
since the input signal is now a 50 Hz sine signal (either
The correction procedure determines the correction function
voltage or current), it introduces dynamic behavior to the
coefficients ak that minimizes (4), which is the same as
system (sensor and ADC) that the DC calibration can not take
into account. maximizing the SINAD of ci .

2285
 The optimization problem that leads to the best set of A. SINAD Analysis
polynomial coefficients can be represented by
In this section the average power spectrums of 10 sets of
 N
2
acquisitions before and after calibration are presented. The
ε ( a0 ,..., aM ) = min ∑ ( xˆi − ci )  (5) results were obtained with 99001 samples per set (adjusted to
ak
 i=1  reduce spectral leakage) acquired at 50 kS/s. In Fig. 6 the
normalized average power spectrum of the acquired samples
where ci = F ( mi ) . ci is shown. Cleary present are several signal harmonics,
The proposed optimization algorithm can be defined with namely the second, third and many more. The SINAD value
the pseudocode is 60.1 dB.
Algorithm
0

Input:

Average Power Spectrum [dBc]

-20
Æ Acquired sensor samples ( mi ) and
-40
timestamps, ( ti )
Æ Initial estimation of polynomial -60

coefficients ( ak ) for function F ( x )

-80

Output:
-100
Æ Coefficients ( ak ) that minimize the cost
function ε -120

-140
Body: 0 100 200 300 400 500 600 700 800 900 1000
Frequency [Hz]
SET iter to 0
Fig. 6 – Voltage sensor normalized average spectrum, obtained
REPEAT with 10 sets of 99001 samples each, acquired at 50 kS/s
INCREMENT iter for a 50 Hz signal before correction.

CALCULATE ci = F ( mi ) The corrected normalized average power spectrum is

shown in Fig. 7 for the ninth order polynomial correction
CALL SINEFIT with ti , ci RETURNING φ and f function. These results were obtained from the uncorrected
samples of Fig. 6. The first few harmonics with higher
 f  amplitudes of Fig. 6 now have much reducer amplitudes
CALCULATE xˆi = Aef 2 cos  2 π (i − 1) + φ
 fS  demonstrating the usefulness of the proposed correction
method. The SINAD value after correction is 74.9 dB, which
N

CALCULATE ε = ∑ ( xˆi − ci )
2 is an improvement of 14.8 dB (the equivalent of an increase
i=1 in the number of effective bits of 2.46).

SOLVE min (ε ) RETURNING new coefficients ak 0

ak
Average Power Spectrum [dBc]

-20
UNTIL (∆ε<εresol OR ε<εmin OR iter>MaxIter)
-40

In this algorithm, εmin is a fixed threshold for the cost -60

function. When ε<εmin the algorithm stops because it has
achieved the desired cost function value. εresol is the -80

desired resolution for the cost function. When the algorithm

-100
fails to improve above εresol, the algorithm stops because it
has converged. The algorithm stops when the number of -120
iterations exceeds MaxIter.
Convergence depends on the initial ak correction -140
0 100 200 300 400 500 600 700 800 900 1000
coefficients. Our experience shows that, by choosing the DC Frequency [Hz]
calibration coefficients as the initial estimates, convergence is
assured. Fig. 7 – Corrected voltage sensor normalized average spectrum,
obtained with 10 sets of 99001 samples each, acquired at 50 kS/s for
a 50 Hz signal after correction with 9th order polynomial function.
 Also note that, the higher harmonics have their amplitude To quantify the improvements, the SINAD as a function of
unchanged. This can be explained by the analysis of the the polynomial order was calculated and is shown in Fig. 10,
natural powers of a sine signal [7] for both current and voltage sensors calibration.

2n 1  2n  1 n −1  2n  78
 cos ( x ) =   + 2 n −1 ∑   cos 2 ( n − k ) x  , (6)
22 n  n 2 k =0  k 
74

2 n +1 1 n
 2n + 1 
 cos ( x ) = ∑  cos ( 2n + 1 − 2k ) x  . (7)

SINAD [dB]
70
4n k =0  k 

This means that for example the 9th power can introduce 66

frequency corrections at f , 3 f , 5 f , 7 f and 9 f . Since

Voltage
the power of the correction polynomial is limited to 9, only 62

harmonics up the 9th can be corrected. Current

In Fig. 8, the normalized average power spectrum for the 58

current is shown while the corrected spectrum is shown in 1 2 3 4 5 6 7 8 9

Fig. 9. The SINAD improves from 59.7 dB up to 68.4 dB. Polynomial Order (M)

Fig. 10 – SINAD value of the corrected sample values as a function of

0 the polynomial degree of calibration.
Average Power Spectrum [dBc]

-20
Contrarily to what happened in the DC calibration, there is
-40 a considerable improvement in terms of SINAD. The
difference between the improvements obtained in current and
-60 voltage can be consequence of the different DAQ input
voltage range used in each case (42% for voltage and 30% for
-80
current) or different sensor distortion at 50 Hz.
-100
In Table I the correction coefficients for the voltage and
current sensor obtained with the AC calibration method are
-120 presented.

-140 Table I. Correction coefficients for the voltage and current

0 100 200 300 400 500 600 700 800 900 1000 sensor obtained the AC calibration method.
Frequency [Hz]

Polynomial Voltage Current

Fig. 8 – Voltage sensor normalized average spectrum, obtained
with 10 sets of 99001 samples each, acquired at 50 kS/s
Coefficient sensor Sensor
for a 50 Hz signal before correction.
a0 0.25939 V -0.00523 A
0
a1 154.94905 2.77857 A/V
Average Power Spectrum [dBc]

-20
a2 -0.13760 V-1 -0.00386 A/V2
-40

a3 -0.36503 V-2 -0.01199 A/V3

-60

-80
a4 -0.00385 V-3 0.00092 A/V4

-100 a5 0.20956 V-4 0.01285 A/V5

-120 a6 0.00158 V-5 -0.00065 A/V6

-140
0 100 200 300 400 500 600 700 800 900 1000
a7 -0.05362 V-6 -0.00618 A/V7
Frequency [Hz]
a8 -0.00013 V-7 0.00015 A/V8
Fig. 9 – Corrected current sensor normalized average spectrum,
obtained with 10 sets of 99001 samples each, acquired at 50 kS/s for -8 9
a9 0.00486 V 0.00105 A/V
a 50 Hz signal after correction with 9th order polynomial function.

2287
B. Frequency Response Analysis -40

Spurious Signal Amplitude [dBc]

-42
To assess the frequency response of the voltage sensor, a
different setup is used. Two Wavetek 9100 are connected in -44

series as shown in Fig. 11. With this setup it is possible to

add a spurious frequency to the larger signal whose -46

frequencies and amplitudes can be independently changed.

-48
The setup is also useful to test real power quality
measurement situations, where a small amplitude spurious -50

harmonic affects the power signal. 100 1000

Spurious Signal Frequency [Hz]
10000

The small amplitude spurious signal can also be directly

measured by a third differential channel of the data Fig. 13 – Frequency response of the spurious signal with 2.3 V.
acquisition board. The power signal has 230 V at 50 Hz.

Clearly, the bandwidth for the spurious signal is much

smaller than the bandwidth of the power signal (Fig. 12). This
may be caused by the presence of the larger power signal.

VI. CONCLUSIONS

A method for AC calibration of power quality sensors has

been presented. It can improve the measured SINAD by
14.8 dB for the voltage sensor and 8.7 dB for the current
sensor, using a ninth order polynomial correction function.
The sensors bandwidth has been assessed with a setup that
allows the addition of a variable amplitude and frequency
Fig. 11 – Setup to measure the voltage sensor response to smaller spurious signal.
amplitude spurious voltages with independent control of both signals.
Further improvements require a more precise knowledge
of the calibrator actual imposed signal to the sensors. In fact,
Preliminary results of the assessment of the voltage sensor the calibrators own THD is affecting the results since the
include: (i) frequency response of the sensor to the large calibrating procedure is also trying to correct the input signal
power signal and (ii) frequency response of the sensor to the THD. However, with the spectral composition of the input
smaller spurious signal. signal, spectral correction techniques must be used.
In Fig. 12 the response of the voltage sensor to the large
signal frequency variations is shown. Up to 30 kHz (which is ACKNOWLEDGMENTS
the frequency limit of the calibrator), the frequency response
of the sensor introduces only a maximum loss of 2 dB.
Work sponsored by the Portuguese national research
48
project reference POSC/EEA-ESE/57708/2004 entitled “Fast
and accurate power quality measurements using analog to
Power Signal Amplitude [dBV]

digital converters and digital signal processing techniques”.

47

REFERENCES
46
[1] Roger C. Dugan, Mark F. McGranaghan, Surya Santoso, H. Wayne
Beaty, Electrical Power Systems Quality, 2nd ed., McGraw-Hill, 2002.
45
[2]
“A Linearization Method for Comercial Hall Efect Current
L. Cristaldi, A. Ferrero, M. Lazzaroni and R. Ottoboni,
Transducers”, IEEE Trans. on Instrum. Meas, Vol. 50, No. 5, pp.
44
1149-1153, Oct. 2001.
100 1000 10000 100000 [3] Datasheet, “Current Transducer LA 25-NP”, LEM Corporation,
Power Signal Frequency [Hz] available at http://www.lem.com.
[4] Datasheet, “Voltage Transducer LV 25-P”, LEM Corporation, available
Fig. 12 – Frequency response of the power signal amplitude without at http://www.lem.com.
spurious component. [5] Nuno B. Brás, Pedro M. Ramos and A. Cruz Serra, “Flexible PC
measurement system based in Extensible Mark-up Language files”,
The frequency response of the sensor to the spurious 14th IMEKO Symposium on New Technologies in Measurement and
Instrumentation, Gdynia, Poland, vol. II, pp. 404-407, Sept. 2005.
signal frequency was determined with a 230 V, 50 Hz power [6] Standard for digitizing waveform records, IEEE Std. 1057-1994,
signal and a 2.3 V spurious frequency signal. In Fig. 13 the December 1994.
measured spurious signal amplitude is shown as a function of [7] W. H. Beyer, CRC Standard Mathematical Tables, 28th ed. Boca
the spurious frequency. Raton, FL: CRC Press, 1987.

2288

View publication stats