IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO.

2, MARCH 2004 491

A Complete Solution to the Harmonic

Elimination Problem
John N. Chiasson, Senior Member, IEEE, Leon M. Tolbert, Senior Member, IEEE,
Keith J. McKenzie, Student Member, IEEE, and Zhong Du, Student Member, IEEE

Abstract—The problem of eliminating harmonics in a switching

converter is considered. That is, given a desired fundamental
output voltage, the problem is to find the switching times (angles)
that produce the fundamental while not generating specifically
chosen harmonics. In contrast to the well known work of Patel
and Hoft and others, here all possible solutions to the problem
are found. This is done by first converting the transcendental
equations that specify the harmonic elimination problem into
an equivalent set of polynomial equations. Then, using the
mathematical theory of resultants, all solutions to this equivalent
problem can be found. In particular, it is shown that there are new
solutions that have not been previously reported in the literature.
The complete solutions for both unipolar and bipolar switching
patterns to eliminate the fifth and seventh harmonics are given. Fig. 1. Bipolar switching scheme.
Finally, the unipolar case is again considered where the fifth,
seventh, 11th, and 13th harmonics are eliminated along with
corroborative experimental results. is obtained. Specifically, in [2]–[4] the harmonic elimination
problem was formulated as a set of transcendental equations
Index Terms—Bipolar, harmonic elimination, switching con-
verter, unipoplar.
that must be solved to determine the times (angles) in an elec-
trical cycle for turning the switches on and off in a full bridge
inverter so as to produce a desired fundamental amplitude
I. INTRODUCTION while eliminating, for example, the fifth and seventh har-
monics. These transcendental equations are then solved using
T HE PROBLEM of eliminating harmonics in switching
converters has been the focus of research for many years.
If the switching losses in an inverter are not a concern (i.e.,
iterative numerical techniques to compute the switching angles.
(See Figs. 8–34 of [1] for a plot of these angles as a percent
switching on the order of a few kHz is acceptable), then the of the fundamental or Fig. 2.) Here a method is presented that
sine-triangle PWM method and its variants are very effective not only obtains these solutions, but also another (different)
for controlling the inverter [1]. This is because the generated set of the switching angles, and this other set of switching
harmonics are beyond the bandwidth of the system being angles actually generates a smaller harmonic distortion due
actuated and therefore these harmonics do not dissipate power. to the eleventh and thirteenth harmonics. The unipolar case is
On the other hand, for systems where high switching efficiency also considered (including the case where the fifth, seventh,
is of utmost importance, it is desirable to keep the switching eleventh, and thirteenth harmonics are eliminated) along with
frequency much lower. In this case, another approach is to corroborative experimental results.
choose the switching times (angles) such that a desired funda- The paper is organized as follows. In Section II, the solution
mental output is generated and specifically chosen harmonics method is illustrated for the bipolar case with the problem
of the fundamental are suppressed [1]–[5]. This is referred to formulated as achieving the fundamental while not generating
as harmonic elimination or programmed harmonic elimination the fifth and seventh harmonics. In Section III, it is then shown
as the switching angles are chosen (programmed) to eliminate how the method can be used in the case of a unipolar PWM
specific harmonics. switching scheme, again formulating the problem so as to
In this work, it is shown how the complete solution (i.e., achieve the fundamental while not generating the fifth and
all possible solutions) to the problem considered in [2]–[5] seventh harmonics. Section IV then formulates and solves the
unipolar case using five switching angles in which the funda-
mental is achieved and the fifth, seventh, 11th, and 13th are not
Manuscript received December 11, 2002; revised August 4, 2003. This paper generated. Experimental results are presented in Section V, and
was presented at the American Power Electronics Conference (APEC), Miami,
FL, February 2003. This work was supported in part by the National Science a summary of the results is presented in Section VI.
Foundation through Grant NSF ECS-0093884 and the Oak Ridge National Lab-
oratory under UT/Battelle Contract 4000007596. Recommended by Associate
Editor J. R. Rodriguez.
II. BIPOLAR CASE
The authors are with the Electrical and Computer Engineering Department, In this work, a standard H-bridge is used wherein choosing
The University of Tennessee, Knoxville, TN 37996-2100 USA (e-mail: chi-
asson@utk.edu; tolbert@utk.edu; kmc18@utk.edu; zdu1@utk.edu). the switching angles , , for the bipolar case results in an
Digital Object Identifier 10.1109/TPEL.2003.823207 output waveform of the form shown in Fig. 1. (In this figure, the
0885-8993/04$20.00 © 2004 IEEE
492 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

Fig. 2. Bipolar switching angles versus m.

Fig. 3. Normalized error (a =a ) + (a =a ) for Bipolar PWM due to

angle corresponds to the time , etc and corre- the 11th and 13th harmonics.
sponds to the fundamental period .) The Fourier series expan-
sion of this output voltage waveform is to transform the conditions (2) into the equivalent conditions

(1)

Given a desired fundamental voltage , the problem here is to

determine the switching angles , , so that
(4)

where and . Equation (4) is

(2) a set of three polynomial equations in the three unknowns , ,
. Further, the solutions must satisfy .
Such a transformation to polynomial equations was also used
where . This is a system of 3 transcendental in [5] where the polynomials were then solved using iterative
equations in the unknowns , , . One approach to solving numberical techniques. In contrast, it is shown here how the
this set of nonlinear transcendental (2) is to use an iterative polynomial equations can be solved directly for all solutions.
technique such as the Newton-Raphson method [3], [4]. Such
a method results in the solution in Figs. 8–34 in [1] (or Fig. 2).
Here, a methodology for finding all the solutions to (2) is pre- A. Elimination Using Resultants
sented, and our method not only gives the solutions reported in
[1], [3], [4], but also a new set of solutions which are found to In order to explain how one computes the zero sets of poly-
generate a lower harmonic distortion due to the 11th and 13th nomial systems, a brief discussion of the procedure of solving
harmonics (see Fig. 3). such systems is now given. A systematic procedure to do this is
To use the method, the conditions (2) are first converted to an known as elimination theory and uses the notion of resultants
equivalent polynomial system. Specifically, one defines [6]–[9]. Briefly, one considers and as poly-
, , and uses the trigonometric nomials in whose coefficients are polynomials in . Then,
identities for example, letting and have degrees 3 and
2, respectively in , they may be written in the form

(3) (5)
CHIASSON et al.: COMPLETE SOLUTION TO THE HARMONIC ELIMINATION PROBLEM 493

The Sylvester matrix, where

, is defined by

(6)
The resultant polynomial is then defined by

(7)
Fig. 4. Unipolar PWM switching scheme.
and is the result of solving and
simultaneously for , i.e., eliminating . See the Appendix The problem is to determine the switching angles , , such
for a brief explanation of this fact.
that
B. Solving the Bipolar Equations
Following the procedure just outlined [10], the resultant
methodology is used to solve for all possible switching angles.
That is, is used to eliminate from and (11)
in (4) to get the two polynomials equations ,
in two unknowns which must be solved Converting (11) to polynomial equations
simultaneously. This is reduced to one polynomial in one
unknown by computing the resultant polynomial of
the polynomial pair (see [7] and [8]
for background on resultants) to get

(8)

where is a polynomial of 9th degree (see the Appendix). (12)

As the parameter is incremented in steps of 0.01, the roots of
are found and used to back solve for and . The set as in the bipolar example, the resultant methodology as pre-
of all three tuples ( , , ) which satisfy sented in [10] was again used to solve for all possible switching
then give angles. That is, is used to eliminate
from and in (12) to get the pair of polynomial equations
, that must be solved simultane-
(9) ously. As in the bipolar case, this is done by computing resul-
as the set of all possible solutions to (2) for the particular value of tant polynomial of the pair
. This computation was done as was incremented between to get
0 and 1 resulting in the switching angles versus as given in (13)
Fig. 2. As the figure shows, only at high values of
do the two sets of solutions merge into one. where is a polynomial of ninth degree (see the Ap-
To compare the two sets of solutions, the normal- pendix).
ized magnitude of their 11th and 13th harmonics (i.e., As the parameter is incremented in steps of 0.01, the roots
where is the harmonic) is of are found and used to back solve for and . The
plotted in Fig. 3. As this figure shows, the new set of solutions set of all three tuples ( , , ) which satisfy
generates less harmonic distortion due to the 11th and 13th then give
harmonics.

III. UNIPOLAR CASE (14)

The Fourier expansion of the unipolar waveform given in as the set of all possible solutions to (11) for the particular value
Fig. 4 is of . The parameter is then varied between 0 and 1, and these
switching angles are plotted versus in Fig. 5. Fig. 6 is a plot
of magnitude of the distortion (i.e., )
due to the 11th and 13th harmonics. As seen in the figure, there
are two sets of solutions for and that the two
(10) sets of solutions produce approximately the same distortion.
494 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

Fig. 5. Unipolar switching angles versus m. Fig. 6. Normalized error (a =a ) + (a =a ) for Unipolar PWM due
to the 11th and 13th harmonics.
IV. UNIPOLAR PWM WITH FIVE SWITCHING ANGLES
In the bipolar scheme,
the RMS voltage
is constant because
and therefore the THD is constant and is
only being shifted in the frequency spectrum. However, the
unipolar PWM scheme can also produce zero voltage and
therefore inherently has lower harmonic content than the
bipolar scheme. Consequently, this scheme is now considered
for the case where five switching angles are used. The Fourier
expansion of a unipolar waveform with switching angles ,
, , , leads to the conditions

(15) Fig. 7. Unipolar switching angles versus m with five switching angles.

Here, is the modulation index and the angles

must satisfy (see Fig. 9 for a typical
waveform). Let if the coefficient of is 1 and
if it is (16)
and letting , , ,
where .
, the conditions become
Remark: It is interesting to note that the set of polynomials in
(16) are the same equations as that of a multilevel inverter with
five dc sources and a fundamental frequency staircase output
waveform [10]. The difference between the two solutions is in
the region where the must lie. In the multilevel case, the con-
ditions are .
Following a procedure similar to that given in Sections II
and III, one systematically solves these equations by elimination
theory. This was done, and the complete set of switching angle
solutions are plotted versus in Fig. 7. Each set of solutions
( , , , , ) is labeled vertically in Fig. 7. Note that for
there are two sets of solutions; for
CHIASSON et al.: COMPLETE SOLUTION TO THE HARMONIC ELIMINATION PROBLEM 495

Fig. 9. Voltage waveform with m = 0:7 and f = 42 Hz.

Fig. 8. THD versus m for each set of switching angles.

there is only one solution set; for there

are three sets of solutions; and finally, for ,
there are again two sets of solutions.
The corresponding total harmonic distortion (THD) was com-
puted out to the 31st according to

(17)
and is plotted versus in Fig. 8 for each of the solution sets
shown in Fig. 7. As this figure shows, one can choose a partic-
ular solution for the switching angles such that the THD is 32%
or less for .
It is important to point out that if one had used an iterative
method such as Newton-Raphson, then the third solution set that
exists for would not have been found, and
this is the solution set that results in the lowest THD for this
range of modulation indices. The reason the Newton-Raphson Fig. 10. FFT of the voltage waveform of Fig. 9 with m = 0:7 and f = 42 Hz.
method would not have found this solution set is simply due to
the way it is implemented. One starts with an initial guess for
the angles at . Then this solution is used as the initial V. EXPERIMENTAL RESULTS
guess for the solution when is incremented by to its An inverter was used to perform experiments to validate the
next value and so on. At , the only possible solutions predicted results, that is, the elimination of the fifth, seventh,
are , , , , 11th, and 13th harmonics in the output of a three phase inverter.
or , , , , . A real-time computing platform [11] was used to interface the
As Fig. 7 shows, if the first solution set is used as the starting logic signals from the computer to the gate driver board of the
point in the Newton-Raphson scheme for , then as inverter. The switching algorithm is implemented as a lookup
is incremented, one would obtain a set of solutions valid for table in SIMULINK which is then converted to code. The soft-
. If the second set of solutions is used as the ware provides icons to interface the SIMULINK model to the dig-
starting point, then a set of solutions valid for ital I/O board and converts the code into executables. The
would be obtained. Neither of these sets results in the minimum computational time step size was 32 m. The induction motor
THD for . Consequently, the method proposed used in the experiments had the following name plate data:
here that finds the complete solution set allows one to be sure
that the solution with the lowest THD is used. In the interesting Rated hp hp
work [13], a homotopy approach was used for the bipolar case
Rated Current A
only. Though it appears to be able to find all solutions in the
bipolar case, it is not clear that it would be able to do so in the Rated Speed rpm
unipolar case (e.g., find the third set in Figs. 7 and 8). Rated Voltage V RMS line to line Hz
496 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

Fig. 11. Current waveform in phase a of the test (induction) motor with m = Fig. 13. Voltage waveform of phase a with m = 0:5 and f = 30 Hz.
0:7 and f = 42 Hz.

Fig. 14. FFT of the voltage waveform of Fig. 13 with m = 0:5 and f =
Fig. 12. FFT of the current in phase a of the test motor with m = 0:7 and 30 Hz.
f = 42 Hz.
voltage THD computed using (17) was 29.7% based on the FFT
Two sets of experiments were performed to compare with the data in Fig. 10 which compares well with the predicted value of
computational results given in Figs. 7 and 8. 31.5% given in Fig. 8. The total current THD was found to be
12.6% using the FFT data in Fig. 12.
A. First Experimental Set
In this first experiment, the modulation index was set as B. Second Experimental Set
and the frequency . Fig. 9 shows the measured In the second experiment, the modulation index was set as
voltage waveform from phase of the inverter output. A (nor- and the frequency . Fig. 13 shows the
malized) fast fourier transform (FFT) of this waveform is plotted measured output voltage waveform from phase of the inverter,
in Fig. 10. As predicted, the fifth, seventh, eleventh, and thir- and its corresponding FFT is plotted in Fig. 14. As predicted, the
teenth harmonics are quite small consistent with their predicted fifth, seventh, eleventh, and thirteenth harmonics are essentially
value of zero. As can be seen from the harmonic specturm in zero consistent with their predicted value of zero. Application
Fig. 10, the lowest substantial nontriplen harmonics are the sev- of the voltages to the test motor resulted in the current waveform
enteenth and nineteenth. As the interest here is a three phase given in Fig. 15, and the FFT of this waveform is presented in
drive, the triplen harmonics in the phase voltages will cancel Fig. 16. The total voltage THD computed using (17) was 43.3%
in the line-line voltages. Application of the voltages to the test based on the FFT data in Fig. 14 which compares favorably with
motor resulted in a current waveform for phase as given in the predicted value of 39% given in Fig. 8. The total current
Fig. 11 with its corresponding FFT plotted in Fig. 12. The total THD was found to be 17.6% using the FFT data in Fig. 16.
CHIASSON et al.: COMPLETE SOLUTION TO THE HARMONIC ELIMINATION PROBLEM 497

Consider and as polynomials in whose

coefficients are polynomials in . There is always a polynomial
(called the resultant polynomial) such that

So if then , that is, if

( , ) is a common zero of the pair ,
then the first coordinate is a zero of . To see how
one obtains , let

Next, see if polynomials of the form

Fig. 15. Current waveform in phase a of the test (induction) motor with m =
0:5 and f = 30 Hz.
can be found such that

(18)

Equating powers of , this equation may be rewritten in ma-

trix form as

The matrix on the left-hand side is called the Sylvester matrix

Fig. 16. FFT of the current in phase a of the test motor with m = 0:5 and and is denoted here by . The inverse of has the
f = 30 Hz. form

VI. CONCLUSION
The complete solution to the harmonic elimination problem
can be found using the theory of resultants from elimination where is the adjoint matrix and is a 5 5 poly-
theory. The solution is complete in the sense that any and all nomial matrix in . Solving for , gives
solutions were found. Experimental work was presented to cor-
roborate the developed technique.

APPENDIX I

RESULTANTS [7]–[9], [12]

Given two polynomials and how does one Choosing guarantees that ,
find their common zeros? That is, the values ( , ) such that , , , are polynomials in . That
is, the resultant polynomial defined by is
the polynomial required for (18).
498 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

APPENDIX II
RESULTANT POLYNOMIALS AND

REFERENCES
[1] N. Mohan, T. M. Undeland, and W. P. Robbins, Power Electronics: Con-
verters, Applications, and Design, 2nd ed. New York: Wiley, 1995.
[2] P. N. Enjeti, P. D. Ziogas, and J. F. Lindsay, “Programmed PWM tech-
niques to eliminate harmonics: a critical evaluation,” IEEE Trans. Ind.
Applicat., vol. 26, pp. 302–316, Mar./Apr. 1990.
[3] H. S. Patel and R. G. Hoft, “Generalized harmonic elimination and
voltage control in thryristor inverters: part I—harmonic elimination,”
IEEE Trans. Ind. Applicat., vol. 9, pp. 310–317, May/June 1973.
[4] , “Generalized harmonic elimination and voltage control in
thryristor inverters: part II—voltage control technique,” IEEE Trans.
Ind. Applicat., vol. 10, pp. 666–673, Sept./Oct. 1974.
[5] J. Sun and I. Grotstollen, “Pulsewidth modulation based on real-time
solution of algebraic harmonic elimination equations,” in Proc. 20th Int.
Conf. Ind. Electron., Contr. Instrum. IECON, vol. 1, 1994, pp. 79–84.
[6] C. Chen, Linear Systems Theory and Design, 3rd ed. London, U.K.:
Oxford Press, 1999.
[7] D. Cox, J. Little, and D. O’Shea, Ideals, Varieties, and Algorithms: An
Introduction to Computational Algebraic Geometry and Commutative
Algebra, 2nd ed. New York: Springer-Verlag, 1996.
[8] J. Joachim von zur Gathen and Jürgen Gerhard, Modem Computer Al-
gebra. Cambridge, U.K.: Cambridge Univ. Press, 1999.
[9] T. Kailath, Linear Systems. Englewood Cliffs, NJ: Prentice-Hall, 1980.
[10] J. Chiasson, L. M. Tolbert, K. McKenzie, and Z. Du, “Eliminating har-
monics in a multilevel inverter using resultant theory,” in Proc. IEEE
Power Electron. Spec. Conf., Cairns, Australia, June 2002, pp. 503–508.
[11] RTLab, Opal-RT Technologies (2001). [Online]. Available:
http://www.opal-rt.com/
[12] D. Cox, J. Little, and D. O’Shea, Using Algebraic Geometry. New
York: Springer-Verlag, 1998.
[13] , “Sequential homotopy-based computation of multiple solutions
for selected harmonic elimination in PWM inverters,” IEEE Trans. Cir-
cuits Syst., vol. 46, pp. 586–593, May 1999.

John N. Chiasson (S’82–M’84–SM’03) received the B.S. degree in mathe-

matics from the University of Arizona, Tucson, the M.S. degree in electrical
engineering from Washington State University, Pullman, and the Ph.D. degree
in controls from the University of Minnesota, Minneapolis.
He has been with Boeing Aerospace, Control Data, and ABB Daimler-Benz
Transportation. Since 1999, he has been on the faculty of Electrical and Com-
puter Engineering, University of Tennessee, where his interests include the con-
trol of ac drives, multilevel converters, and hybrid electric vehicles.
CHIASSON et al.: COMPLETE SOLUTION TO THE HARMONIC ELIMINATION PROBLEM 499

Leon M. Tolbert (S’89–M’91–SM’98) received the B.E.E., M.S., and Ph.D. Keith J. McKenzie (S’01) received the B.S. degree in electrical engineering
degrees from the Georgia Institute of Technology, Atlanta, all in electrical en- from The University of Tennessee, Knoxville, in 2001 where he is currently
gineering. pursuing the M.S. degree.
He joined the Engineering Division, Lockheed Martin Energy Systems, in
1991 and worked on several electrical distribution projects at the three U.S. De-
partment of Energy plants in Oak Ridge, TN. In 1997, he became a Research
Engineer in the Power Electronics and Electric Machinery Research Center, Oak
Ridge National Laboratory (ORNL). In 1999, he was appointed as an Assistant
Professor in the Department of Electrical and Computer Engineering, University
of Tennessee, Knoxville. He is an Adjunct Participant at ORNL and conducts
joint research at the National Transportation Research Center (NTRC). He does
research in the areas of electric power conversion for distributed energy sources, Zhong Du (S’01) received the B.E. and M.E. degrees from Tsinghua Univer-
motor drives, multilevel converters, hybrid electric vehicles, and application of sity, Bejing, China, in 1996 and 1999, respectively, and is currently pursuing
SiC power electronics. the Ph.D. degree in electrical and computer engineering at The University of
Dr. Tolbert received the National Science Foundation CAREER Award and Tennessee, Knoxville.
the 2001 IEEE Industry Applications Society Outstanding Young Member He has worked in the area of computer networks, both in academia as well
Award. He is an Associate Editor of the IEEE POWER ELECTRONICS LETTERS as in industry. His research interests include power electronics and computer
and a registered Professional Engineer in the state of Tennessee. networks.