IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO.

3, MARCH 2010 527

A Discrete Matlab–Simulink Flickermeter Model for

Power Quality Studies
Leonard W. White, Senior Member, IEEE, and Subhashish Bhattacharya, Member, IEEE

Abstract—A discrete model of the IEC 61000-4-15 flickermeter

is developed for use with the Matlab–Simulink simulation software
package. The discrete model is a faithful reproduction of a flick-
ermeter defined by the IEC Standard; it duplicates the functional
blocks described by the standard and operates in the same way as
the standardized instrument to assure complete fidelity of results.
The model is intended to be used within larger simulations that
contain discrete model blocks, either alone or in combination
with the Simulink SimPowerSystems blockset, to provide direct
quantification of flicker. The model is fully developed, and all
model parameters are provided for use with 60-Hz systems. A
discrete calibrator is described that duplicates the test waveforms
described by the IEC Standard.
Index Terms—Lighting, measurement, modeling, power quality,
standards.

I. I NTRODUCTION
Fig. 1. Sensitivity of flicker is a function of frequency and ΔV /V . The curve

E LECTRICAL flicker can be thought of as a rapidly chang-

ing system voltage that causes perceivable visual effects
upon lighting systems. In general, the variations in voltage
shows locations where the perception level is unity for both 50- and 60-Hz
systems.

over time fall well within the “safe bounds” of the Computer model for use with 60-Hz simulations that meets all the relevant
Business Equipment Manufacturing Association curve [1] and requirements of the Standard and operates in exactly the same
typically do not cause direct damage to or disruption of electri- way as the model described by the Standard.
cal utilization equipment. The problems associated with flicker It is recognized that there are changes that could be made
are generally due to the rapidly changing light levels that result to the design presented in the Standard that would provide
when systems with significant levels of electrical flicker are various operational enhancements, for example, improved re-
used to supply lighting loads. sponse time. No efforts have been made to improve the response
A standard, i.e., IEC 61000-4-15 [2], has been created to times of the various filters used in the flickermeter or to make
describe the characteristics of an instrument to measure the other “improvements,” as the goal is to duplicate the instrument
level of flicker in terms of the base unit of flicker perceptibility. described by the Standard with complete fidelity and then to
Throughout this paper, IEC 61000-4-15 will be referred to as verify that the resulting model complies with the performance
the “Standard.” requirements of the Standard in all respects.
The basic instrument defined by the Standard is a physi-
cal analog device intended to be connected to an ac power
system to measure the level of flicker; such physical instru- II. IEC 6100-4-15 S TANDARD
ments are commercially available from several manufacturers. The Standard defines an instrument that quantifies the fol-
A Simulink model for a continuous instrument was described lowing into a single numeric output that is called perceived
in [3], but this model will not directly interface with discrete flicker: 1) the percent voltage change; 2) the frequency of the
model blocks in general and with the blocks of the SimPower- voltage change; 3) the physics of the light source; and 4) the
Systems blockset in particular, which are extremely useful in way the human brain recognizes the light output. The changes
the modeling and simulation of flicker mitigation schemes. It in voltage and frequency are relatively easy to quantify, but the
is the purpose of this paper to describe a discrete Simulink human response to the flicker is extremely nonlinear.
The human response to flicker was empirically determined
Manuscript received July 16, 2008; revised December 30, 2008. First by exposing a group of individuals to a light source driven by
published January 12, 2010; current version published February 10, 2010.
The Associate Editor coordinating the review process for this paper was Dr. varying voltage and frequency and recording the level at which
Antonios Tsourdos. 50% of the test subjects indicated that they perceived flicker.
The authors are with the Future Renewable Electric Energy Delivery and This information has been included in the Standard as a set
Management (FREEDM) System Center, North Carolina State University,
Raleigh, NC 27695 USA (e-mail: lwwhite@ncsu.edu; sbhatta4@ncsu.edu). of data for sinusoidal and rectangular voltage fluctuations for
Digital Object Identifier 10.1109/TIM.2009.2023121 both 50 and 60-Hz systems. A graph showing the relationship
0018-9456/$26.00 © 2010 IEEE
528 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 3, MARCH 2010

Fig. 2. Functional block diagram of the IEC 61000-4-15 flickermeter.

between the normalized voltage change, which is denoted as

ΔV /V , and the frequency of the changes, which is denoted
as f , for 50-Hz systems is included in [3]; the 50-Hz graph,
as well as a similar graph for 60-Hz systems, is produced in
Fig. 1 for comparison purposes. Both of these graphs have been
normalized around the point of maximum sensitivity to flicker,
which is approximately 8.8 Hz. It should be noted that, while
the curves are similar in shape and have the same peak value
because of the normalization process, the actual ΔV /V magni-
Fig. 3. Block diagram of the flickermeter calibrator.
tude associated with the 50-Hz peak is approximately 80% less
than the corresponding 60-Hz value. In general, an observer is
more sensitive to voltage variations at 50 Hz than at 60 Hz. In (1), the sinusoidal carrier is Ac cos(ωc t + φ), with magni-
The functional diagram of the IEC flickermeter is directly tude Ac , angular frequency ωc , and phase shift φ. The modu-
taken from the Standard and is shown in Fig. 2. The develop- lating waveform is kV (t), which can be in any time-varying
ment of the discrete model in this paper retains the same block format. The term “k” is generally known as the modulation
designations and descriptions as those used in the Standard. factor, but the Standard slightly differently defines the overall
expression, with the net results being that k = ΔV /2V ; the
phase shift φ is zero.
III. D ISCRETE M ODEL The final configuration of Block 1 is shown in Fig. 3. Note
that the modulating waveform is inverted to obtain the exact
A. Block 1—Input and Calibrator
form described by (1), as the output of the Modulating Signal
Block 1 provides two independent functions: 1) an input volt- block cannot be phase shifted.
age adaptor and 2) a signal generator for calibration verifica-
tion. Both functions are important to a physical instrument, but
only the calibrator is important in a model situation since there B. Block 2—Demodulator With Squaring Multiplier
are no concerns about overvoltage, saturation, nonlinearity, etc. Block 2 provides a demodulation function by squaring the
The only conditioning at the input is a scaling function, which is output provided by Block 1. In the frequency domain, the
accomplished by dividing the input voltage by the base voltage. squaring operation separates the carrier from the modulating
The calibrator is composed of two separate sources: a base function, which contains the information of interest, i.e., the
source that generates the carrier sine wave and a programmable amplitude and the frequency of the disturbance that is the root
source that can generate the modulation as either a sine wave or cause of the flicker. The demodulation function is achieved
a rectangular waveform. Both sine and rectangular modulating using a Simulink math block with the sample time set to match
wave inputs are described by the Standard, and both are used in the model sample time. Block 2 is shown, along with other parts
the final validation of the model. of the model, in Fig. 6.
The calibrator produces an amplitude-modulated waveform, The sample time designated as Ts in the model (T in
as described in [4] and shown as follows: the equations to conserve space) is selected as 0.001 s. The
choice is one of convenience, as [5] indicates that 0.01 s will
v(t) = Ac [1 + kV (t)] cos(ωc t + φ). (1) provide acceptable results in a physical version of a discrete
WHITE AND BHATTACHARYA: DISCRETE MATLAB–SIMULINK FLICKERMETER MODEL FOR POWER QUALITY STUDIES 529

flickermeter. In most simulations, the overall sample time will The evaluation of (5) gives
be much less than the sample time required for the present
flickermeter. 1
FGLP (s) =  2   
s s
ωlp +a ωlp +1
C. Block 3—Weighing Filters
1 1
Block 3 contains three sets of filters: the first two strip the de- ·  2    ·  
2    (6)
s s s s
modulated signal of the high- and low-frequency components, ωlp +b ωlp +1 ωlp + c ωlp + 1
and the third provides an output that corresponds to human
sensitivity to flicker. The first filter is defined as a first-order which is the standard form of the transfer function for the filter
high-pass filter with a 3-dB cutoff frequency, which is denoted in the s domain.
as fco , of 0.05 Hz. The purpose of this filter is to remove The values of the constants are the following:
the dc component of the waveform being evaluated. The dc is
removed from the subject waveform because although it will a ≈ 0.5176
alter the computed value of the flicker, it has no effect on the
flicker visually perceived. The transfer function of such a filter b ≈ 1.4142
is given in the following equation, where the numeric value of c ≈ 1.9319. (7)
ωhp = 2πfco :
Again, the Tustin transform is used to discretize the continu-
Vo (s) s ous transfer function (6). Each of the three sections has the same
= FHP (s) = . (2)
Vi (s) s + ωhp general form, which is presented in (8), shown at the bottom of
the page. In (8), the variable “x” is used to represent either a, b,
To model the filters as discrete devices, it is necessary to or c, as defined in (7).
express transfer function (2) in the z domain. There are several The final version of the z-domain version of the 42-Hz low-
discretization methods that can be applied to this problem. The pass filter is composed of three of the functions shown in (8)
method of choice will be the bilinear transformation, which is multiplied together with the appropriate values of the constants.
also known as the Tustin transformation or the trapezoid ap- In the Simulink model, the discrete filter block is again used for
proximation. [6] The transformation equation is the following: each of the three sections cascaded in series.
  The filters described by (4) and the three sections of (8)
2 z−1
s≈ . (3) remove the low and high frequencies from the input waveform.
T z+1 The next filter in Block 3, which is the weighing filter, responds
to the frequency content of the input waveform with an output
After substituting (3) into (2) and simplifying, the final
that matches the human sensitivity response portrayed in Fig. 1.
discrete transfer function for the 0.05-Hz high-pass filter is
The transfer function of this filter is defined by the Standard
shown as
along with constants for use at both 50 and 60 Hz. The transfer
1 − z −1 function of the weighing filter in the s domain is shown as
FHP (z) =     . (4)
T ωhp T ωhp
2 +1 + 2 − 1 z −1 kω1 s 1 + s/ω2
FLP (s) = · . (9)
s2 + 2λs + ω12 (1 + s/ω3 )(1 + s/ω4 )
In the Simulink model, the filter is modeled by the use of a
discrete filter block. The Tustin transform is again used on the two sections of
The second filter in Block 3 is a sixth-order Butterworth (9) with the results given in (10), shown at the bottom of the
low-pass filter. The generalized standard form of the transfer next page.
function for this filter in the s domain is given by The product F1 (z)F2 (z) is implemented in Simulink by the
use of two cascaded discrete filter blocks. The constants used in
Vgo (s) 1 this filter, along with the constants from the previous filters, are
= FGLP (s) = n (5)
Vgi (s)   2
2 
presented in Table I.
s − 2s cos 2k+1−1 π +1
k=1
2n Similar to the input adaptor shown by the Standard for
Block 1, the range selector at the output end of Block 3 is
where the value of the cutoff frequency, which is denoted as needed for a physical instrument but not for a Simulink model.
ωlp , is given by ωlp = 2πflp , and the value of n = 6 [7]. An output from Block 3, i.e., a weighted voltage output, is

2 2 2 2 −1 2 2 −2
ωlp T + 2ωlp T z + ωlp T z
      (8)
2 T 2 + 2xω T + 4 + 2ω 2 T 2 − 8 z −1 + ω 2 T 2 + 2xω T + 4 z −2
ωlp lp lp lp lp
530 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 3, MARCH 2010

TABLE I as the response times of the various filters described by the

SYMBOL DEFINITIONS AND CONSTANT VALUES
Standard are quite long.

D. Block 4—Squaring and Smoothing

The Standard indicates that the processing of Block 4 is
provided “to simulate the nonlinear eye–brain perception” and
to provide “a sliding mean averaging of the signal to simulate
the storage effect in the brain.” The square function is accom-
plished, as before in Block 1, by using a Simulink math block.
The Standard defines the sliding mean operator as having the
transfer function of a first-order low-pass resistance filter with
a time constant of 300 ms. The transfer function in the s domain
of such a filter is shown as

Vo (s) 1
= FLP (s) = (11)
Vi (s) s + Rlp Clp

where the values of Rlp Clp = τlp = 0.300 s.

Using the Tustin transformation, the transfer function in the
z domain is shown as

1 + z −1
FLP (z) =     . (12)
2τlp 2τlp
T +1 + 1− T z −1

The low-pass filter is implemented in Simulink by the use of a

discrete filter block.

E. Output Measurement
The Standard does not detail a method to be used to extract
a reading from the flickermeter; however, the overall accuracy
must be within ±5% of unity when the waveforms described in
the normalized flickermeter response tables are applied at the
input. As the magnitude at the output of Block 3 is proportional
to the perceived flicker level, the reading must be in the form of
a relative magnitude.
The waveform used as the calibration point was selected
from the “normalized flickermeter response for sinusoidal volt-
age fluctuations,” i.e., Table 1 in the Standard, with the fre-
quency being 8.8 Hz and ΔV /V being equal to 0.323. The
scaled output of Block 4 when the input to the flickermeter
is the calibration waveform is shown in Fig. 4 for a total
simulation time of 25 s.
It is observed in Fig. 4 that there is a significant transient in
shown by the Standard but is not used in this paper. This output, the output from Block 4 due to the action of the two long-time-
after appropriate scaling, would provide an instantaneous value constant filters. During the calibration phase of the work, the
of perceived flicker. Note that “instantaneous” in this case ef- Simulink model was arranged to delay the measurement of the
fectively means “without the introduction of deliberate delay,” output magnitude until the transient had decayed. The actual

2kω1 T − 2kω1 z −2
F1 (z) =
(ω12 T 2 + 4λT + 4) + (2ω12 T 2 − 8) z −1 + (ω12 T 2 − 4λT + 4) z −2
ω3 ω4 (T 2 ω2 + 2T ) + 2T 2 ω2 z −1 + (T 2 ω2 − 2T )z −2
F2 (z) = · 2 (10)
ω2 [T ω3 ω4 + 2T (ω3 + ω4 ) + 4] + (2T 2 ω3 ω4 − 8)z −1 + [T 2 ω3 ω4 − 2T (ω3 + ω4 ) + 4] z −2
WHITE AND BHATTACHARYA: DISCRETE MATLAB–SIMULINK FLICKERMETER MODEL FOR POWER QUALITY STUDIES 531

20–25-s evaluation interval. A factor, other than simplicity, for

the choice of a peak reading instead of, for example, an RMS
reading is that the time required to acquire the peak value is
minimal, thus reducing the overhead that the final flickermeter
inserts into the larger simulation of which the flickermeter is
only a part.
The scaling factor is found by applying the calibrating wave-
form, determining the peak of the output with a scaling factor
of unity, and then determining the value of the scaling factor
that would make the peak value equal to unity.
The final calibrated discrete flickermeter model, which is
ready for use, is shown in Fig. 6.

F. Model Validation
After the determination of the scaling factor, the model
was validated by using the calibrator to generate each of the
waveforms described in the normalized flickermeter response
tables from the Standard. Under ideal conditions, each of the
values in the table should produce a perceived level of flicker
of unity. During the validation phase, the reset of the maximum
Fig. 4. Complete output from Block 4 with the calibration waveform applied block was not used so that the maximum value over the entire
to the input of the flickermeter model. 5-s calibration interval would be captured. The actual values
were measured, and a percent error was computed; the result is
presented in Fig. 7. The validation of the model for rectangular
data [2, Table II] was similar. All results, for both sinusoidal and
rectangular waveforms, were within ±5% of unity as required
by the Standard, with the majority of values observed to be
within 1.5% of unity.

G. Model Application
The model may be applied by either using the flickermeter, as
shown in Fig. 6, within a larger simulation or using the model
as a stand-alone simulation, importing previously recorded data
via the Simulink “From File” block.
As a demonstration of the second input method, a data set
has been imported and ran through the model. The flickermeter
output, along with the input waveform, is shown in Fig. 8.
The input data set has been extracted from an 8-h recording of
several variables taken from metering at an operating electric
arc furnace (EAF). The specific data imported into the model
are voltage data from a 120-V RMS circuit located on the same
supply source as the EAF. The time interval selected shows first
a period of EAF operation and then an interval of nonoperation,
Fig. 5. Enlarged output waveform showing subharmonic ripple. followed by a final interval of operation. The data presented in
Fig. 8 represent a time interval of about 9.6 min, around 34 500
time of the delay is 20 s, with the calibration measurement cycles of the voltage waveform at 60 Hz. The model input, in
being conducted over the 20–25-s interval. Matlab .mat format, is approximately 156.6 MB.
A closer look at the Block-3 output, shown in Fig. 5, in- The lower plot in Fig. 8 shows the input voltage waveform
dicates that the waveform contains a small amount of subhar- over the time of interest; only the portion of the waveform
monic ripple. The origin of the ripple is the cross products from above zero is shown. The upper plot shows the perceived level
(1) that fall within the region of inclusion of the 0.05- and 42-Hz of flicker as computed by the model. The center section of the
filters. The total ripple is small, i.e., less than 1% of the lower waveform, where the EAF is not operating indicates a
perceived flicker over the region of interest. This level of ripple level of perceived flicker that is essentially zero.
is not large enough to seriously impinge on the overall ±5% The usefulness of the model becomes apparent by visu-
accuracy of the instrument; therefore, the output is measured by ally comparing the input waveform to the corresponding level
determining the peak value of the output waveform during the of perceived flicker. It is difficult to determine the level of
532 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 3, MARCH 2010

Fig. 6. Final diagram of the discrete flickermeter.

Fig. 7. Instrument percent error for sinusoidal voltage fluctuations.

perceived flicker, even as an approximation, from only the

appearance of the voltage waveform. Although it is easily
observed that the perceived flicker level is low during the EAF-
off interval, when the voltage waveform is smooth, it is not so
easy to estimate the flicker level at other points on the curve
and to correlate the observed variations to a particular level of
flicker. Fig. 8. Demonstration of model operation with a prerecorded data set. The
It is noted that, during a simulation using either a model lower plot is the model input; the upper plot is the computed level of perceived
input or a data file input, the 25-s time interval used during the flicker.
calibration phase is not useful. During actual operation of the
model, the “Maximum” function block is periodically reset in taken from the larger model; however, for analysis of field
synchronization with the line frequency. acquired data, the input can first be imported into the Matlab
workspace and then brought into Simulink.

IV. C ONCLUSION R EFERENCES

The discrete flickermeter model described has provided [1] IEEE Recommended Practice for Power and Grounding Electronic Equip-
a simple way to compare the efficacy of flicker mitigation ment, IEEE Std. 1100-19990, 1999.
[2] Electromagnetic Compatibility (EMC)—Part 4: Testing and Measurement
schemes within larger discrete Simulink models with a recog- Techniques—Section 15: Flicker—Functional and Design Specifications,
nized standard. The input to the model is intended to be directly IEC Std. 61000-4-15, 2003. CEI.
WHITE AND BHATTACHARYA: DISCRETE MATLAB–SIMULINK FLICKERMETER MODEL FOR POWER QUALITY STUDIES 533

[3] A. Bertola, G. C. Lazaroiu, M. Roscia, and D. Zaninelli, “A Matlab- Subhashish Bhattacharya (M’85) received the B.E.
Simulink flickermeter model for power quality studies,” in Proc. 11th Int. (Hons.) degree in electrical engineering from the
Conf. Harmonics Quality Power, 2004, pp. 734–738. University of Roorkee (IIT Roorkee), Roorkee,
[4] H. Black, Modulation Theory. New York: Van Nostrand, 1953. India, in 1986, the M.E. degree in electrical engi-
[5] G. Neri, G. D. Cain, T. Salmon, and A. Yardim, “A microprocessor-based neering from the Indian Institute of Science (IISc),
digital flickermeter,” IEEE Trans. Instrum. Meas., vol. 40, no. 6, pp. 1008– Bangalore, India, in 1988, and the Ph.D. degree
1014, Dec. 1991. in electrical engineering from the University of
[6] T. Adam, S. Dadvandipout, and J. Futas, “Influence of discretization Wisconsin–Madison, in August 2003.
method on the digital control system performance,” Acta Montan. Slovaca, He was with York International Corporation,
vol. 8, no. 4, pp. 197–200, 2003. where he worked on the commercialization of his
[7] S. Butterworth, “On the theory of filter amplifiers,” Experimental Wireless active filter Ph.D. research for air-conditioner chiller
and the Wireless Engineer, pp. 536–541, Oct. 1930. application. From December 1998 to August 2005, he was with the FACTS
and Power Quality Division, Siemens Power Transmission & Distribution.
Since August 2005, he has been an Assistant Professor with the Department of
Leonard W. White (SM’90) was born in Electrical and Computer Engineering, North Carolina State University, Raleigh,
Henderson, NC, in 1945. He received the B.S. where he is also a Faculty Member of the Future Renewable Electric Energy
degree in engineering operations and the M.S. Delivery and Management (FREEDM) System Center. He has been involved in
degree in electrical and computer engineering from several FACTS projects, including the NYPA 2 × 100 MVA Convertible Static
North Carolina State University, Raleigh, in 1969 Compensator, Korea 40 MVA UPFC, and AEP 150 MVA STATCOM projects.
and 1986, respectively. He is currently working His research interests are FACTS, utility applications of power electronics such
toward the Ph.D. degree in electrical engineering as custom power and power quality issues, active filters, high-power converters,
with the Future Renewable Electric Energy Delivery and converter control techniques.
and Management (FREEDM) System Center, North
Carolina State University.
He is one of the founding partners and a former
Senior Principal of Stanford White, Inc., a midsized engineering firm spe-
cializing in engineering services for the construction industry. His area of
specialization is power quality.
Mr. White is a Registered Professional Engineer in eight states and a
Registered Communications Distribution Designer. He serves on the NFPA-99
hospital electrical systems committee.