See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/3219588 Reduced Switching-Frequency-Modulation Algorithm for High-Power Multilevel Inverters Article in IEEE Transactions on Industrial Electronics · November 2007 DOI: 10.1109/TIE.2007.905968 · Source: IEEE Xplore CITATIONS READS 167 150 3 authors, including: Samir Kouro Jaime Rodriguez Universidad Técnica Federico Santa María University of A Coruña 217 PUBLICATIONS 17,745 CITATIONS 155 PUBLICATIONS 3,416 CITATIONS SEE PROFILE Some of the authors of this publication are also working on these related projects: Partial Power Converters for Two-Stage PV Systems View project Inspection of transmission lines View project All content following this page was uploaded by Samir Kouro on 05 February 2013. The user has requested enhancement of the downloaded file. SEE PROFILE A Reduced Switching Frequency Modulation Algorithm for High Power Multilevel Inverters J. Rodrı́guez, S. Kouro, J. Rebolledo and J. Pontt Universidad Técnica Federico Santa Marı́a Department of Electronics Engineering Av. España 1680, Valparaı́so, CHILE Email: samir.kouro@ieee.org Multilevel modulation Abstract— Multilevel inverters have emerged as the state of the art power conversion systems for high power medium voltage applications. Many topologies and modulation methods are commercially available. This paper presents a new adaptive duty cycle modulation algorithm, that reduces the switching frequency and consequently the switching loses. This can be important for high power applications, where high frequency modulation methods like PWM are not suitable. Results are shown for a nine level asymmetric cascaded inverter. Output voltage waveforms obtained for references with variable frequencies and amplitudes show a similar switching pattern, with a reduced and near constant number of commutations per cycle, regardless the reference frequency and amplitude. The proposed modulation is compared to a Multiple Carrier PWM method, achieving the same fundamental reference tracking performance with a reduced number of commutations. I. I NTRODUCTION Multilevel inverters are considered today the most suitable power converters for high voltage capability and high power quality demanding applications [1]–[4]. Voltage operation above classic semiconductor limits, lower common mode voltages, near sinusoidal outputs together with small dv/dt’s, are some of the characteristics that have made this power converters popular for industry and research, specially for medium-voltage applications. There are several topologies available, being the Neutral Point Clamped [5], Flying Capacitor [6] and Cascaded Hbridge inverter [7] the most studied and used. In recent years many variations and combinations of these topologies have been reported, one of them is the cascaded H-bridge fed by non equal DC sources, known as asymmetric multilevel inverter [8], [9]. The main characteristic of this topology is the reduction, or even elimination of redundant output levels to maximize the output power quality of the inverter using less semiconductors. Many modulation algorithms have been adapted or created for multilevel inverters [1]–[3]. Most of them are based on multiple carrier Pulse Width Modulation (PWM) techniques: Phase Disposition (PD), Phase Opposition Disposition (POD), Alternative Phase Opposition Disposition (APOD) and Phase Shifted (PS) carrier PWM have been studied [10]. Space Vector Modulation (SVM) is also extended for the multilevel case [11]. Selective Harmonic Elimination [12] and Space Vector Control [13] are used for lower switching frequency applications. These modulation methods are summarized in Fig. 1. In this paper, a new modulation method is presented based on time domain duty cycle calculation between the two nearest output voltage levels to the reference. An adaptive modulation 0-7803-9033-4/05/$20.00 ©2005 IEEE. Fundamental Switching Frequency Space Vector Control High Switching Frequency PWM Selective Harmonic Elimination Space Vector PWM Phase Shifted PWM Phase Disposition PWM a1 a2 a3 Symmetric Disposition Fig. 1. Opposition Disposition Alternate Opposition Disposition Principal modulation methods for multilevel inverters. time calculation based on the reference voltage slope is derived, and used to achieve a reduction in the number of commutations. The multilevel modulation algorithm proposed in this paper is applicable to any multilevel inverter topology, with an arbitrary number of levels, either for single-phase or three-phase applications. In this work, it is tested on a nine level, three phase, asymmetric H-bridge cascaded inverter. Results are compared, under same conditions, to Multicarrier PD-PWM . II. T HE I NVERTER A generalized power circuit of an asymmetric H-bridge cascaded inverter is shown in Fig. 2. The inverter is composed by the series connection of m monophasic H-bridges fed by non equal DC sources (vk , with k = 1, . . . , m). The use of asymmetric input voltages can reduce, or when properly chosen, eliminate redundant output levels, maximizing the number of different levels generated by the inverter. Therefore this topology can achieve the same output voltage quality with less number of semiconductors, space and costs than the symmetric fed topology. When cascading three level inverters like H-bridges (output levels: −vdc , 0 and +vdc ), the optimal asymmetry is obtained by using voltage sources scaled proportional to the power of three. Applying this criteria to the voltage sources illustrated in Fig. 2, optimal design leads to ⎤ ⎡ ⎤ ⎡ 30 v1 ⎥ ⎢ v2 ⎥ ⎢ 31 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ v3 ⎥ ⎢ 32 (1) ⎥=⎢ ⎥ · vdc . ⎢ ⎥ ⎢ .. ⎥ ⎢ .. ⎦ ⎣ . ⎦ ⎣ . 867 vm 3(m−1) v1 b va1 v2 c vb 1 va2 vc 1 vb 2 v o* vo 5vdc Voltage Levels a vc 2 4v dc 3vdc 2v dc tcom vdc Tm 2Tm Time 3Tm 4T m (a) vo* vo 5vdc vam vcm vbm Voltage Levels vm n Fig. 2. Asymmetric cascaded H-bridge multilevel inverter. 4vdc 3vdc vdc Tm Then the individual output voltages per H-bridge are Voltage Levels (2) vam ∈ {−3(m−1) vdc , 0, +3(m−1) vdc }, for phase a, which is analog for phases b and c of the inverter. The total phase voltages generated by the inverter can be expressed as vaj ; j=1 vbn = m vbj ; Time 3T m 4T m vo* vo 5vdc va2 ∈ {−3vdc , 0, +3vdc }, .. . m 2T m (b) va1 ∈ {−vdc , 0, +vdc }, van = tcom 2vdc vcn = j=1 m vcj . 4vdc 3vdc 2vdc tcom vdc Tm (3) 2Tm Time 3Tm 4 4Tm (c) j=1 Replacing in (3) the different combinations of individual outputs generated by each H-bridge given in (2), a total of 3m different levels per phase can be generated to the load. Fig. 3. Duty Cycle Modulation: (a) Left-justified timer schedule when dvo∗ /dt > 0; (b) Left-justified timer schedule when dvo∗ /dt < 0; (c) Rightjustified timer schedule when dvo∗ /dt < 0. This can be expressed mathematically by III. M ODULATION STRATEGY vo∗ (2Tm ) = v̄o (∆T3 ) 3Tm 1 = vo (t)dt (4) Tm 2Tm 2Tm +tcom 3Tm 1 = dt + 3vdc dt . 2vdc Tm 2Tm 2Tm +tcom A. Basic principle The proposed modulation algorithm is based on a combination of left-justified timer scheduling and right-justified timer scheduling duty cycle modulation [14], together with an adaptive commutation time calculation derived from the voltage reference slope. The aim of the proposed method is to change the modulation time frame to reduce the switching frequency. The left-justified timer scheduling operating principle is shown in Fig. 3(a). Consider, for example, the third modulation period ∆T3 =[2Tm , 3Tm ] in Fig. 3(a) (shadowed area). To achieve fundamental frequency tracking, the mean output voltage during this period, v̄o (∆T3 ), has to be equal to the sampled reference at the beginning of the interval, vo∗ (2Tm ). Then, the commutation time, tcom is computed by solving (4). This task can be simplified seeing the modulation of vo∗ (2Tm ) as a duty cycle between the two nearest output voltage levels. Then tcom can be easily computed by 868 tcom = v ∗ (2Tm ) 1− o − f loor vdc vo∗ (2T ) vdc · Tm . (5) Once tcom is calculated by (5), the nearest lower voltage level to the reference, vo∗ (2Tm ) vdc · vdc , Voltage [pu] f loor 1 (6) is generated from t = 2Tm up to t = 2Tm + tcom , then the nearest upper voltage level, vo∗ (2Tm ) vdc · vdc , tcom = vo∗ (2Tm ) − f loor vdc vo∗ (2T ) vd c · Tm . (8) Once tcom is determined by (8) first the nearest upper level is generated and is switched to the nearest lower level at tcom . The slope sign detection and timer schedule justification is selected according to the following criteria: vo∗ (t) > vo∗ (t − Tm ), =⇒ use left-justified timer schedule else, if (9) =⇒ use right-justified timer schedule Voltage [pu] 1 0.5 0 Vo V * −0.5 −1 o 0 0.02 0.04 0.06 Time[s] 0.08 0.1 0.12 (a) Voltage [pu] 1 0.5 0 Vo Vo* −0.5 −1 0 0.02 0.04 0.06 0.08 Time[s] 0.1 0.12 (b) Fig. 4. Results for a 81 level asymmetric inverter with Duty Cycle Modulation: (a) Left-justified timer schedule; (b) left&right-justified timer schedule with voltage reference slope sign detection. 0 −0.5 0 0.005 0.01 0.015 0.02 Time [s] 0.025 0.03 0.035 0.04 (a) (7) is generated to complete the duty cycle from t = 2Tm + tcom to t = 3Tm . The qualitative example explained before and illustrated in Fig. 3(a) looks adequate for positive voltage reference slopes. However the negative slope case, shown in Fig. 3(b), reveals that the commutation sequence (first the lower level and then the higher level) still achieves fundamental reference tracking, but introduces higher harmonic distortion due to higher dv/dt’s since a two level difference appears in the switching pattern. It clearly follows that a right-justified timer schedule would be more appropriate for the negative slope case as shown in Fig. 3(c). In this case tcom is computed by o vo −1 1 Voltage[pu] ceil v * 0.5 v * o 0.5 v o 0 −0.5 −1 0 0.025 0.05 0.075 0.1 Time[s] 0.125 0.15 0.175 0.2 (b) Fig. 5. A 9 level asymmetric H-bridge inverter with left&right-justified timer schedule modulation: (a) Output voltage for a 50[Hz] reference; (b) Output voltage for a 10[Hz] reference. Figure 4(a) and (b) show results for a 81 level inverter (4 asymmetric H-bridges per phase) with left-justified timer schedule and the combined left&right-justified timer schedule respectively. It is clear that the voltage reference slope sign detection reduces dv/dt’s by alternating both methods adequately. This algorithms ensures mean voltage tracking with zero error in each modulation interval. However, since the modulation period is fixed (Tm = constant), the selected value for implementation has to be sufficiently small to allow good performance for the highest voltage reference frequencies. When using this small Tm for a low frequency reference, unnecessary additional commutations will be performed with this modulation algorithm (this occurs with all PWM methods).This effect is shown in Fig. 5(a) and (b), where the same algorithm (same Tm ) is applied to a 50[Hz] and 10[Hz] voltage references respectively. It can be seen that the considered modulation period of 1[ms] produces a minimum necessary commutations for the fast reference, while to many commutations are performed to follow the 10[Hz] reference. For High power applications those extra switching losses can be avoided introducing a variable modulation period, which should be determined according to the reference frequency. B. Adapting the modulation period It is possible to reduce the number of commutations by using appropriately an estimation of the voltage reference slope. Assuming for a certain instant t = t(k) that the slope mk stays fixed over one modulation period Tmk , and that in addition the difference between two adjacent voltage levels is always vdc , a modulation period that produces only one commutation can be calculated by vdc , (10) Tmk = mk This holds only if no big changes occur in the reference slope, which is common in high power applications. The estimated slope is calculated using a least squares algorithm and some past samples of the voltage reference. For 869 n samples and a fixed sampling period Ts = t(i) − t(i − 1), the slope is given by n mk = k t(i) · v ∗ (i) − i=k−n n· k t(i) · i=k−n k 2 t (i) − i=k−n k k 2) Use the last n samples to compute the reference slope mk according to (11), and compare to mmin . 3) Calculate the modulation period Tmk using (10). 4) Determine the nearest voltage levels to the reference as given in (6) and (7). 5) Compute commutation time tcom : a) using (5), if mk > 0 (left timer scheduling). b) using (8), if mk < 0 (right timer scheduling). 6) Tmk , tcom and the nearest voltage levels are sent to a FPGA to schedule the commutation and deliver the gating signals necessary to generate the desired voltage levels at the corresponding times. 7) Save the last n voltage reference samples before the modulation period Tmk is finished. 8) Begin algorithm again. v ∗ (i) i=k−n 2 . (11) t(i) i=k−n Figure 6 illustrates this operating principle. For example, at t = t2 the slope m2 is computed according to (11), then the modulation period Tm2 is calculated using (10), finally the left timer schedule modulation is performed since m2 > 0. Note that the adaptive modulation time will adjust itself to the reference signal. C. Modulation period saturation There is however a drawback at this point with this strategy, from (10) it follows that for mk ’s close to zero, too large modulation periods are obtained. Therefore Tmk has to be saturated using a certain criteria; when the estimated slope of the reference is lower than a specified limit mmin , the last modulation time will be used. When the estimated slope is higher than mmin the calculated modulation time will be used. This can be summarized as follows: if mk < mmin , =⇒ use Tm(k−1) else, (12) =⇒ use Tmk The design of mmin will depend on the highest frequency available in the reference bandwidth. D. Algorithm summary IV. R ESULTS Figure 7(a) and (b) show output voltage results for a 9 level asymmetric H-bridge inverter controlled with the proposed slope sign detection and adaptive modulation time calculation for a 10[Hz] and 50[Hz] reference respectively. The algorithm was tested using a least squares estimation with n = 5 samples and a sample period Ts = 10[µs]. It can be seen that both waveforms are very similar in shape, due to the adaptation of the algorithm to both reference frequencies. The results confirm that the algorithm works properly, achieving fundamental frequency control and similar waveforms, same number of commutations per cycle, independently of the voltage reference frequency. Note that this modulations techniques introduces a zeroorder-hold effect, because of its discrete time nature [14]. Figure 8 shows the algorithm timer. Each saw tooth represents the execution of the algorithm, showing clearly how the modulation periods change according to the slope of the reference. Note that the modulation period is saturated The following is a simplified overview of the necessary steps to perform the proposed modulation algorithm: 1) Continuously sample the voltage reference at DSP clock time Ts . Voltage[pu] 1 4vdc vo* 0.5 v o 0 −0.5 −1 0 m 1 2vdc Tm1 t1 t2 Fig. 6. 0.04 1 v o* vo Tm2 0.08 0.1 (a) Tm3 2 0.06 Time [s] Voltage [pu] m 3vdc vdc 0.02 m3 vo* 0.5 v o 0 −0.5 −1 0 0.02 0.04 0.06 0.08 0.1 Time [s] t3 Time t4 Adapting modulation time operating principle. (b) Fig. 7. Results for a 9 level asymmetric H-bridge inverter with the adaptive modulation period algorithm: (a) Output voltage for a 10[Hz] reference; (b) Output voltage for a 50[Hz] reference. 870 1 Voltage [pu] Modulation timer [s] 0.01 0.008 0.006 0.004 0.002 0 0.5 0 v * o −0.5 v o −1 0 0.025 0.05 0.075 0.1 Time [s] 0.125 0.15 0.175 0 0.2 0.02 0.04 0.06 0.08 0.14 0.16 0.18 0.2 0.14 0.16 0.18 0.2 (a) (a) 1 Voltage [pu] 0.002 Modulation timer [s] 0.1 0.12 Time [s] 0.001 0.5 0 v * o −0.5 vo −1 0 0 0.005 0.01 0.015 0.02 Time [s] 0.025 0.03 0.035 0 0.04 0.02 0.04 0.06 0.08 0.1 0.12 Time [s] (b) (b) Fig. 8. Adaptive modulation periods: (a) For a 10[Hz] reference; (b) For a 50[Hz] reference. Fig. 9. Response to step change in amplitude and frequency: (a) Proposed algorithm; (b) Multiple carrier Phase Disposition PWM. 0.06 Voltage [pu] when mmin is reached, and that two to three sample periods have he same time length until the slope restriction is deactivated. 0.02 0 V. C OMPARISON VI. C ONCLUSION In this paper a new duty cycle modulation technique, with adaptive modulation periods has been presented. Reduced commutations and similar output voltage quality, independently of the reference frequency, are the main features achieved. The algorithm is by nature developed in discrete time, and therefore its DSP implementation is straightforward. The algorithm can be easily extended for any multilevel inverter topology, either for one or three-phase applications. 0 200 400 600 800 1000 1200 Frequency [Hz] 1400 1600 1800 2000 (a) 0.06 Voltage [pu] Figure 9(a) and (b) present a comparison between the proposed method and Multiple Carrier Phase Disposition PWM operating in a 9 level Asymmetric Inverter. Both methods are tested under same conditions, showing the dynamic behavior when a simultaneous step is applied in amplitude and frequency at t = 0.1[s]. Both methods achieve fundamental frequency control, however the carrier frequency for the PD-PWM was designed to follow correctly the high frequency component of the reference, and therefore it produces a large number of commutations until t = 0.1[s]. The proposed method adapts the modulation period, obtaining almost a constant number of commutations per cycle. The steady state spectrum for both methods for a 10[Hz] reference are plotted in Fig. 10(a) and (b). As expected, the harmonic content for PD-PWM is concentrated around multiples of the carrier frequency 1[kHz), while the proposed algorithm produces a dispersed low frequency harmonic content. This is a drawback considering the filtering nature of most loads. However this work is focused for high power applications where reducing commutation losses is required. If this issue becomes relevant the use of an output filter is recommended. THDv = 12.2 0.04 THDv = 15.7 0.04 0.02 0 0 200 400 600 800 1000 1200 Frequency [Hz] 1400 1600 1800 2000 (b) Fig. 10. Steady state output voltage spectrum: (a) Proposed Algorithm; (b) Multiple carrier Phase Disposition PWM. Compared to high switching frequency modulations methods (like carrier based PWM), less commutations are obtained while achieving same reference tracking. This makes the algorithm suitable for high power applications. Compared to low frequency modulation methods, the proposed algorithm ensures fundamental reference tracking, and also can be applied to inverters with reduced number of levels and achieves high quality outputs even for low modulation indexes (not achieved with Space Vector Control). In addition it is easy to implement and not limited to pure sinusoidal references (necessary for Selective Harmonic Elimination). ACKNOWLEDGMENT The authors gratefully acknowledge financial support provided by the Chilean National Fund of Scientific and Technological Development (Fondecyt), under grant No. 1040183, and by the General Direction of Research (DGIP) of the Universidad Técnica Federico Santa Marı́a. 871 R EFERENCES [1] J. S. Lai and F. Z. Peng, “Multilevel converters–A new breed of power converters,” IEEE Trans. Ind. Applicat., vol. 32, no. 3, pp. 509–517, May/June 1996. [2] R. Teodorescu, F. Blaabjerg, J. K. Pedersen, E. Cengelci, S. Sulistijo, B. Woo, and P. Enjeti, “Multilevel converters – A survey,” in European Power Electronics Conference, 1999, Lausanne, Switzerland. [3] J. Rodrı́guez, J. S. Lai, and F. Z. Peng, “Multilevel inverters: A survey of topologies, controls and applications,” IEEE Trans. Ind. Electron., vol. 49, no. 4, pp. 724–738, 2002. [4] L. M. Tolbert, F. Z. Peng, and T. G. Habetler, “Multilevel converters for large electric drives,” IEEE Trans. Ind. Applicat., vol. 35, no. 1, pp. 36–44, January/February 1999. [5] A. Nabae and H. Akagi, “A new neutral-point clamped pwm inverter,” IEEE Trans. Ind. Applicat., vol. 17, no. 5, pp. 518–523, September 1981. [6] T. Meynard and H. Foch, “Multi-level choppers for high voltage applications,” Eur. Power Electron. Journal, vol. 2, no. 1, pp. 45–50, March 1992. [7] M. Marchesoni, M. Mazzucchelli, and S. Tenconi, “A non conventional power converter for plasma stabilization,” in Power Electronics Specialist Conference, 1988, pp. 122–129. [8] O. M. Mueller and J. N. Park, “Quasi-linear IGBT inverter topologies,” in APEC’94, February 1994, pp. 253–259. [9] S. Mariethoz and A. Rufer, “Design and control of asymmetrical multilevelinverters,” in IECON’02, November 2002, pp. 840–845, Sevilla, Spain. [10] B. P. McGrath and D. G. Holmes, “Multicarrier PWM strategies for multilevel inverters,” IEEE Trans. Ind. Electron., vol. 49, no. 4, pp. 858–867, August 2002. [11] N. Celanovic and D. Boroyevich, “A fast space-vector modulation algorithm for multilevel three-phase converters,” IEEE Trans. Ind. Applicat., vol. 37, no. 2, pp. 637–641, Mar./Apr. 2001. [12] L. Li, D. Czarkowski, Y. Liu, and P. Pillay, “Multilevel selective harmonic elimination pwm technique in series-connected voltage inverters,” pp. 1454–1461, October 1998. [13] J. Rodrı́guez, L. Morán, P. Correa, and C. Silva, “A vector control technique for medium-voltage inverters,” IEEE Trans. Ind. Electron., vol. 49, no. 4, pp. 882–888, August 2002. [14] K. Corzine and J. Baker, “Multilevel voltage-source duty-cycle modulation: Analysis and implementation,” IEEE Trans. Ind. Electron., vol. 49, no. 5, pp. 1009–1016, October 2002. 872 View publication stats