XXXII NATIONAL SYSTEMS CONFERENCE, NSC 2008, December 17-19, 2008

A FRACTAL APPROACH TO GENERATE SPACE VECTOR PWM FOR MULTILEVEL INVERTERS

AbstractThis paper aims to establish the inherent fractal structure in the space vector representation of multilevel inverter. The established fractal is used in the proposed method of generation of Space Vector PWM (SVPWM) for a 3-level inverter. The proposed method avoids look up tables for sector identification and switching vector determination associated with SVPWM implementation. The optimum switching sequence is also achieved for every sector without look up tables. The generation of SVPWM using inherent fractal structure has been explained for a 3-level inverter and simulation results are presented. The method can be extended to higher-level inverters. The implementation is carried out using dSPACE 1104 card.

INTRODUCTION

ULTILEVEL inverters serve as induction motor drives generating sinusoids of varying amplitudes and frequencies[1]. In comparison to a 2-level inverter, multilevel inverter generates sinusoids with lesser harmonic distortion [1,2]. Pulse Width Modulation techniques aid inverters in generating these sinusoids. Conventionally, sine-triangle comparison, and space vector concepts were used to generate PWM signals for multilevel inverters. In the generation of PWM using space vector, known as Space Vector PWM (SVPWM), the instantaneous reference space vector is approximated by switching the three nearest inverter voltage space vectors [3 - 9]. In the literature, SVPWM is considered as a better method of PWM generation [2 - 9]. However the complexity involved in the implementation of SVPWM for multilevel inverters increases with the levels of the inverter. The increase in complexity is attributed to the increase in the number of switching voltage space vectors with the inverter levels. The implementation of SVPWM involves (i) identifying the sector enclosing the reference space vector (ii) determining the switching voltage space vectors required to approximate the reference space vector (iii) determining the duration of each of these voltage space vectors and (iv) optimizing the switching sequence for each sector. Several methods for the implementation of SVPWM with reduced complexity have been proposed in the literature [4 - 11]. In [4 - 6], a method of implementing SVPWM using a coordinate transformation of the instantaneous reference space vector to another coordinate system is proposed. In [6], a method of implementing SVPWM using a mapping

technique is also proposed. In the conventional methods of SVPWM implementation, look up tables are necessary mainly to achieve an optimized switching sequence for every instantaneous reference space vector. Techniques utilizing the equivalence of SVPWM and Sine-Triangle PWM can also generate SVPWM signals without using look up tables [10,11]. In the present work, the inherent fractal structure in the switching voltage space vector representation of multilevel inverter is brought out and the same is used to generate SVPWM signals for multilevel inverters. The implementation of SVPWM using the inherent fractal structure reduces the complexity in the various stages of implementing SVPWM. The proposed method of SVPWM implementation reduces the complexity in sector identification [12]. In the proposed method, the switching vectors are also determined simultaneously with sector identification without look up tables. The determination of the duration of operation of these switching voltage space vectors is done by mapping the switching vector located at a vertex of the identified sector to the actual zero vector [6]. The proposed method also determines the optimum switching sequence in every sector without look up tables. This paper explains the generation of SVPWM using the inherent fractal structure for a 3-level inverter. The simulation is performed for a 3-level inverter configuration. The method can be applied to any multilevel inverter configurations. The implementation for a 3-level inverter is carried out using DS1104 card and the experimental results are also presented. II. FRACTAL IN THE SWITCHING SPACE VECTOR REPRESENTATION OF 3-LEVEL INVERTER In SVPWM, the reference space vector is obtained from the three instantaneous reference phase voltages. The triangle formed by the three switching voltage space vectors located nearest to the reference space vector is a sector. Fig.1 shows the switching space vector locations of a 2-level inverter. The 2-level inverter has six triangular regions, sector, and are numbered I-VI. These six regions result from the eight switching vectors corresponding to a 2-level inverter.

702

A03

A02

A03 A16
A01

A14 A15

A02 A12 A13 A11 A112 A11 A01

II III
A04 A00

I VI V
A04

IV

A17 A110

A00

A18
A06

A05

Fig.1 Switching voltage space vectors of 2-level inverter

A05

A111

A06

The six switching vectors are located at A01, A02, A03, A04, A05, and A06. The remaining two vectors are the zero vectors located at A00. The 2-level inverter has two redundant zero vectors. Fig.2 shows the switching voltage space vectors of a 3-level inverter. The 3-level inverters have 12 new space vector locations in addition to the space vector locations of a 3-level inverter. These 12 new space vector locations are named A1i, in Fig.2, (where i from 1 to 12). It may be noted from Fig.2 that these additional space vector locations are at the midpoints of each side of the six triangular regions, called sectors, of the 2-level inverter. The 12 new space vector locations will increase the total number of sectors of 3-level inverter to 24. Thus the sectors of a 3-level inverter can be seen to have evolved from the sectors of 2-level inverter by dividing each triangular region of the 2-level inverter into four smaller triangular regions. This can be done by determining the midpoints of each side of the sectors of the 2- level inverter. Fig.3 shows the generation of the sectors of a 3-level inverter from the sectors of an equivalent 2-level inverter. The six sectors of a 2-level inverter are shown in Fig. 3(a). In Fig.3 (b) the midpoints of each sector of the 2-level inverter is obtained. The midpoints are shown as solid dots in Fig. 3(b). The joining of the three midpoints, Fig, 3.(c),
A03 A16
A00

Fig.2 Switching voltage space vectors of 3-level inverter

will result in the switching voltage space vector representation of 3-level inverter. According to fractal theory [13] the procedure that divides the triangle into four smaller triangles by determining the midpoints of each side of the original triangle is called Sierpinski iteration. The structure, which is the outcome of Sierpinski iteration, is the Sierpinski triangle. III. GENERATING SVPWM USING THE INHERENT FRACTAL
STRUCTURE

The generation of SVPWM involves three stages (i) sector identification and switching vector determination, (ii) determination of duration of the switching vectors and (iii) optimization of switching sequence for every sector. Sector identification and switching vector determination Sector identification involves determining the sector enclosing the instantaneous reference space vector. The (, ) components of the instantaneous reference space vector can be obtained from the three instantaneous A14 A02 A03 A14 A02 reference phase voltages as
A15 A12 A00 A19 A11 A111 A112 A13 A01

A03

A02

A16 A17

A15 A00 A110

A12 A11 A112

A13 A01

A04

A01

A04

A17

A04 A18

A18
A05 A06

A11 A05 A111

(a)

06 (b) Fig.3. Generating the sectors of 3-level inverter from the sectors of 2-level inverter

A05

A110

A06

703

V =

3 Va 2 3 V = (Vb Vc ) 2

(1)

In the proposed method, sector identification begins with determining the region enclosing the reference space vector from the six regions of an equivalent 2-level inverter. This is done by repeatedly comparing the three instantaneous reference phase voltages and determining the region as in [9]. The region identified for the reference vector A00R is region I. The inverter states corresponding to the switching vectors located at A00, A01 and A02 are (000, 111, 222), (200) and (220) respectively. The midpoints corresponding to each side of the region I of 2-level inverter are determined. The midpoints are located at A11, A12 and A13, refer Fig.4 (b). The average of the and coordinates of the vectors at A00 and A01 will result in the coordinates of A11. The averaging applied to the inverter states will result in the inverter states corresponding to the switching vector located at A11 as (100, 211). Similarly, the coordinates and switching states of A12 are obtained as the average of the coordinates and inverter states of A00 and A02. The inverter states corresponding to the vector located at A12 are (110, 221).The coordinates and the inverter states of the location A13 are obtained as the average of the coordinates and switching states of A01 and A02. The inverter states of the vector located at A13 are obtained as (210). Thus the new voltage space vector locations and the inverter states corresponding to these new voltage space vectors are obtained simultaneously. The three new vectors located at the midpoints will result in four new triangles, which are the sectors of 3-level inverter. The centroids of each of these four new sectors are now determined. Since each of these four triangles are equilateral triangles the coordinates of the centroid of each triangle can be obtained as the average of the coordinates of the three vertices.

The distance of the reference space vector from each of these centroids is determined. The triangle with the centroid closest to the instantaneous reference space vector tip is the sector to be identified. For the instantaneous reference space vector A00R, the sector identified is A12A13A02. The sector A12A13A02 encloses the reference space vector, in Fig.4(c), and the space vectors located at the vertices of the sector are the switching vectors that will approximate the instantaneous reference space vector. Thus in the proposed method, together with sector identification, the switching vectors are also determined without look up tables.

B. Determination of duration of each switching vector The determination of duration of switching vector is done by mapping the sector identified to enclose the reference space vector to a sector of the 2-level inverter [6]. Choosing a vector located at one of the vertices of the identified sector to correspond to the actual zero vectors does the mapping. The vector chosen is the one, which has the maximum value for the sum of all the redundant states. The vector chosen to correspond to the zero vectors is subtracted from the remaining two vectors located at the other vertices of the sector and also from the instantaneous reference space vector. This will reduce the identified sector to a sector of the 2-level inverter. For the sector identified to enclose the reference space vector A00R, the voltage space vector located at A12 is chosen to correspond to the zero vector. When the vector located at A12 is chosen to correspond to zero vector, the sector will be equivalent to region I of the 2-level inverter as shown in Fig.5. Similar mapping can be applied to every identified sector and can be mapped onto one of the six regions of the 2-level inverter. The conventional equations, which determine the duration of operation of switching voltage vectors for each of the six regions of the 2-level inverter, can be used after such mapping is applied.

020

A03

A02

220

020

A03
110 221

A02 A12

220

020

A03
110 221

A02

R
A04
022

R
A11
100 211

A13

R A
12
100 211

220

A13

210

210

A00
000 111 222

200

A01

A04
022

A00
000 111 222

200

A01

A04
022

A00
000 111 222

A11

200

A01

201

002

A05

(a)

A06

202

002

A05

102

(b)

A06

202

002

A05

(c)

A06

202

Fig.4. Sector identification by the proposed method

704

A02
R

220

211 221. Thus it involves only one switching per phase for a sampling interval. IV. EXPERIMENTAL AND SIMULATION RESULTS

110 221

A12

A13

210

A00

000 111 222

100 211

A11

200

A01

Fig.5. Mapping to determine the duration of switching vectors

C. Optimization of the switching sequence of the sector. In order to determine the duration of operation of the switching vectors of the identified sector, one of the vectors is selected to coincide with the actual zero vector. As stated earlier, the vector selected to coincide with the zero vector is the vector which has maximum value for the sum of all the switching states including the redundant states. In the present work, a strategy is adopted which will result in an optimum switching for every sector. Optimum switching implies that there occurs only one switching as the inverter moves from one switching vector to another. In the adopted strategy, the last two redundant states are selected for the vector, which is selected to coincide with the actual zero vector. For the remaining two vectors of the sector, the last redundant state is selected. This selection can result in an optimum switching. The proposed method of obtaining an optimum switching sequence using the adopted strategy of choosing a virtual zero vector will result in an optimum switching without using any look up tables. For the instantaneous reference space vector A00R, enclosed by the sector A12A13A02, formed by the vectors located at the vertices of the A12A13A02, the switching states corresponding to the vectors located at the A12, A13 and A02 are (110, 221), ( 100, 211) and (210) respectively. The sum of the switching states for the voltage vector located at A12 is 1+1+0+2+2+1=7. The sum of the switching states for the vectors located at A13 and A02 are 5 and 3 respectively. The switching voltage space vector located at A12 is thus selected to coincide with the zero vector since it has maximum value for sum of switching states. This is true for any virtual zero vector since the virtual zero vector should be chosen as the one closest to the actual zero vector and hence has the maximum redundant states. According to the adopted strategy, two of the redundant states are obtained for this vector. For the remaining two vectors, the last redundant states are selected. Thus the vectors located at A13 and A02 have the states selected as (211) and (210) respectively. So the optimum switching sequence for this sector is 110 210
705

The proposed method has been simulated for a 3-level inverter. The proposed method has also been implemented, for a 3-level inverter, on TMS 320F240 platform using a dSPACE 1104 card. Fig.6 shows all the sectors of a 3-level inverter generated through the proposed method. The 24 sectors of the 3-level inverter has been generated. Fig.7 and Fig.8 shows the PWM signals for the three phases for two consecutive sampling instants. The PWM signals are shown when the tip of the reference space vector is located in different sectors. The figure shows that an optimum switching achieved since it involves only one switching as the inverter moves from one voltage space vector to another. Fig.9 shows the simulated pole voltage for the 3level inverter corresponding to a 2-level operation. The figure shows that the inverter switches between the voltages 150V and 300V. The method is implemented using DS1104 card and the levels of the inverter have been generated using a DAC card. The three levels of the inverter is represented at the output of the DAC with three voltage levels. Fig.10 shows the result for 2-level operation of the inverter. The output switches between the levels 2.5V and 5V. Fig.11 shows the simulated phase voltage for a 3-level inverter. The result is obtained by simulating the proposed method for an induction motor model. Fig.12 shows the simulated pole voltage for a 3-level inverter corresponding to 3-level operation of the inverter. The figure shows that the inverter switches between the levels of 0V, 150V and 300V. The output obtained from the DAC has three voltage levels as shown in Fig.13. Fig.14 shows the phase voltage for three level inverter corresponding to 3-level operation

Fig.6. Switching voltage space vector locations for a 3-level inverter

Fig.7 PWM signals for the 3 phases for 2 consecutive sampling instants for a particular sector

Fig.10 Pole voltage obtained through a DAC output

Fig.8 PWM signals for the 3 phases for 2 consecutive sampling instants for another sector

Fig.11. Phase voltage for the 2-level operation simulated with a motor model

Fig.9. Pole voltage for the 2-level operation

Fig.12. Pole voltage for the 3-level operation

706

REFERENCES
[1] Jose Rodriguez, Jih-Sheng Lai and Fang Zheng Peng," Multilevel Inverters: A Survey of Topologies, Controls and Applications ", IEEE Transactions on Industrial Electronics, vol.49, No.4, August 2002, pp 724-738. [2] Joachim Holtz, Pulse width Modulation- A Survey,IEEE Transactions on Industrial Electronics, vol.39,No.5, December 1992,pp 410-420. [3] Keliang Zhou and Danwei Wang, Relationship Between Space-Vector Modulation and Three-Phase Carrier-Based PWM: A Comprehensive Analysis, IEEE Transactions on Industrial Electronics, vol.49, No.1, February 2002, pp 186196. [4] B.P. McGrath, Donald.G.Holmes, Thomas Lipo, Optimized Space Vector switching sequences for Multilevel inverters,IEEE Transactions on Power Electronics, vol.18,No.6, November 2003, pp 1293-1301. [5] Yo-Han Lee,Bum-Seok Suh and Dong-Seok Hyun, A Novel PWM Scheme for a Three-Level Voltage Source Inverter with GTO Thyristors,IEEE Transactions on Industrial Applications, vol.32,No.2, March/April 1996, pp 260-268. [6] A.K.Gupta and A.M Khambadkone A General Space Vector PWM Algorithm for Multilevel Inverters, Including Operation in Overmodulation Range, IEEE Transactions on Power Electronics, vol.22,No.2, March 2007,pp 517-526. [7] N.Celanovic and Dushan Boroyevich, A Fast Space-Vector Modulation Algorithm for Multilevel Three-Phase Converters,IEEE Transactions on Industry Applications, vol.37,No.2, March/April 2001,pp 637-641. [8] P.F Seixas, M.A. Severo Mendes, P.Donoso Garcia, A.M.N. Lima, A Space Vector PWM Method for Three-Level Voltage Source Inverters, Proc. APEC, vol.1, 2000,pp 549-555. [9] V.T Somasekhar and K.Gopakumar, Three-level inverter configuration cascading two-level inverters, Proc. IEE- Electr. Power Appl, vol.150, No.3, May 2003, pp 245-254. [10] R.S.Kanchan, M.R.Baiju, K.K.Mohapatra, P.P.Ouseph and K.Gopakumar, Space vector PWM signal generation for Multilevel Inverters using only sampled amplitudes of reference phase voltages ,Proc. IEE, vol.152,No.2,March 2005,pp 297309. [11] Baiju M.R, Gopakumar. K, Somasekhar V.T, Mohapatra K.K and Umanand L, A space vector based PWM method using only the instantaneous amplitudes of reference phase voltages for three-level inverters, EPE J., 2003, 13,(2), pp. 35-45. [12] A.Gopinath, M.R.Baiju, Space vector PWM for Multilevel inverters A Fractal Approach, Proc, PEDS 2007, pp 842849. [13] Heinz- Otto Peitgen, Hartmut Jurgens, Dietmar Soupe, Chaos and Fractals- New Frontiers of Science, Second Edition, Springer,2004

Fig.13. Pole voltage obtained through a DAC output

Fig.14. Phase voltage for the 3-level operation simulated with a motor model

CONCLUSION This paper establishes the inherent fractal structure in the space vector representation of multilevel inverters and the same is used to generate SVPWM for a 3-level inverter. This paper explains the method of generating SVPWM for a 3-level inverter avoiding look up tables. The generation of PWM using the inherent fractal structure reduces the complexity involved in implementing SVPWM for higher-level inverters. The simulation is done for a 3-level inverter using SIMULINK and the results are also presented.

707