1 Introduction

Three-level (3L) converters have several advantages over conventional two-level six-switch converters, such as reduction of harmonic component in the line current, suppression of voltage stress on switching devices, and increase of efficiency^[1-2]. The typical topology of 3L converters can be classified into three types: flying-capacitor (FC)^[3], cascade H-bridge (CHB)^[4], and neutral-point-clamped (NPC)^[5]. As the most prevailing 3L converter, the three-level neutral-point-clamped (3L-NPC) converter is applied in the field of railway traction, ship propulsion, mine lifting, wind power generation, and other medium-voltage high-power motor drive applications^[6-8].

When applied in the above mentioned fields, the output performance of 3L converter is significantly reduced during the process of pulse width modulation (PWM) on account of the switching frequency of switching devices. The harmonic distortion of the converter and the torque ripple of the motor are both increased^[9]. Thus, how to evaluate and improve the output waveform quality of 3L converter has been studied by researchers worldwide.

Several evaluation indexes, such as weighted total harmonic distortion (V_WTHD)^[10-11], RMS value of stator flux ripple^[12-13], and the amplitude of mean error current vector^[14-15], have been adopted to represent the output waveform quality of converters in recent years.

For V_WTHD, double Fourier series (DFS) method should be adopted to calculate the individual harmonic of the output waveform. The calculation is highly complicated^[10]. For RMS value of stator flux ripple, the d and q axis components of the stator flux ripple are calculated to synthesize RMS value of stator flux ripple^[12]. The computational burden is greatly reduced compared with that of V_WTHD. For the amplitude of mean error current vector, the distance between the centroid of the error current vector trajectory and the coordinate origin point is calculated on the basis of space vector modulation (SVM). Then the amplitude of mean error current vector can be obtained^[15]. The computational burden is further reduced compared with the above mentioned two methods.

All the above evaluation methods are based on the premise that DC-link voltage is constant. Only the high frequency error caused by the space vector synthesis is considered. However, for 3L-NPC converter, the neutral point is directly connected to the load by the clamping diode. The load current will charge and discharge upper and lower capacitors of the DC-link. If the charge is not equal to the discharge, there will be low-frequency oscillation in the neutral-point (NP), which leads to the ripple in the output of the 3L-NPC converter^[16]. At this time, if the traditional evaluation method is still used, the quantitative evaluation results of the output waveform will be inaccurate.

In order to solve the above issue, the NP voltage is regarded as a variable in this paper. The mathematical model of the NP voltage offset affected by the switching sequence was derived. On this basis, the amplitude and phase angle of basic voltage vectors were modified. Then an accurate evaluation method of the output waveform quality for 3L-NPC converter was established. The proposed method will be helpful in the design of switching sequence and the selection of modulation strategy.

2 SVM of 3L-NPC

Fig. 1 illustrates the topology of 3L-NPC converter. Each leg has 3 switching states (defined as S, S∈{0, 1, 2}, 2 represents S_x1 and S_x2 switch on, S_x3 and S_x4 switch off; 1 represents S_x2 and S_x3 switch on, S_x1 and S_x4 switch off; 0 represents S_x3 and S_x4 switch on, S_x1 and S_x2 switch off; x=A, B and C). So there are 3³ = 27 switching states for three phases. According to Eq. (1), the output voltage for each switch state can be transformed into the space vector in the space vector diagram.

$ {\mathit{\boldsymbol{V}}_j} = ({v_{{\rm{AO}}}} \cdot {{\rm{e}}^{j0}} + {v_{{\rm{BO}}}} \cdot {{\rm{e}}^{j\frac{{2\pi }}{3}}} + {v_{{\rm{CO}}}} \cdot {{\rm{e}}^{j\frac{{4\pi }}{3}}}) $

(1)

whereV_j, j=0, 1, 2, …, 18is the basic vector, as can be seen in Fig. 2. According to the amplitude, basic vectors can be classified into large vectorV₁ ~ V₆, medium vectorV₁₃ ~ V₁₈, small vectorV₇~V₁₂, and zero vectorV₀. Zero vector has three switching states, and each small vector has two switching states. If medium vector is taken as the center, Fig. 2 can be divided into sectors Ⅰ~Ⅵ, and each sector can be further partitioned into triangle 1~6.

For SVM of 3L converter, the nearest three voltage vectors are commonly applied to compose the reference vector V_ref. The duty cycle of V_j is determined by the theory of volt-second balance. Take V_ref lying in triangle 1 of sector I as an example:

$ \left\{ {\begin{array}{*{20}{l}} {{\mathit{\boldsymbol{V}}_{{\rm{ ref }}}}{T_{\rm{s}}} = {\mathit{\boldsymbol{V}}_7}{T_1} + {\mathit{\boldsymbol{V}}_{13}}{T_2} + {\mathit{\boldsymbol{V}}_1}{T_0}}\\ {{T_{\rm{s}}} = {T_1} + {T_2} + {T_0}} \end{array}} \right. $

(2)

In each sample period T_s, the sequence of space voltage vectors used to compose V_ref could be arranged arbitrarily. More than one switching state corresponds to basic vector V₀ and V₇~V₁₂. Therefore, there are several switching sequences that can be applied to synthesize V_ref in each sample period. The continuous PWM (CPWM) and four kinds of discontinuous PWM (DPWM0~ DPWM3) strategies are presented in Refs. [16] and ^[17], respectively. Switching sequences of CPWM and DPWM0~DPWM3 are shown in Table 1. The output performance of the converter will be different for different strategies. It is necessary to establish the output current ripple model. Then the output performance can be analyzed quantitatively and accurately.

3 Error Model of Space Vector Synthesis

According to the theory of SVM, difference still exists between the actual output voltage vector and the reference vector at any arbitrary instant of the sample period. In order to analyze the influence of space vector synthesis process on the output performance of the converter, the above errors need to be analyzed and modeled.

3.1 Trajectory of Error Current Vector

The error voltage vector Δ V_j is considered as the difference between V_j and V_ref. As shown in Fig. 3, when V_ref lies in triangle 1 of sector Ⅰ, error voltage vectors generated by basic voltage vectors V₇, V₁, and V₁₃ are Δ V₇, Δ V₁, and Δ V₁₃, respectively.

The stator resistance can be neglected when 3L converter feeds the motor. The error current vector Δi_j will be generated by Δ V_j. The relationship between the error voltage vector and the error current vector can be derived as

$ \Delta {\mathit{\boldsymbol{V}}_j} = {L_\sigma }\frac{{{\rm{d}}\Delta {\mathit{\boldsymbol{i}}_j}}}{{{\rm{d}}t}} $

(3)

where L_σ is the inductance of the motor. According to Eq. (3), the error current vector Δi_j changes linearly with the error voltage vector. Take switching sequence 211↔210↔200 as an example. When t is between 0 and T₁/2, the switching state 211 corresponding to V₇ is applied to the converter. The error current vector Δi₇ starts from point O, and the terminal travels from point O to point A. When t=T₁/2, error current vector Δi₇=ΔV₇ ·T₁/2L_σ. Similarly, when switch states 210 and 200 apply for T₂/2 and T₀/2, respectively, the corresponding error current vectors are Δi₁₃=ΔV₁₃ ·T₂/L_σ and Δi₁=ΔV₁·T₀/L_σ, respectively. The trajectory of the terminal of error current vector Δi_j during [0, T_s/2] is triangle OAB, and the trajectory during (T_s/2, T_s] is triangle OCD.

The modulation index can be defined as

$ m = \sqrt 3 {V_{{\rm{ref}}}}/{V_{{\rm{dc}}}} $

(4)

From Fig. 3, the amplitudes of error voltage vectors ΔV₇, ΔV₁, and ΔV₁₃ are

$ \left\{ {\begin{array}{*{20}{l}} {\Delta {V_7} = \sqrt {(1 + 3{m^2})V_{{\rm{dc}}}^2 - 2\sqrt 3 m{V_{{\rm{dc}}}}{\rm{cos}}\theta } /3}\\ {\Delta {V_1} = \sqrt {(4 + 3{m^2})V_{{\rm{dc}}}^2 - 4\sqrt 3 m{V_{{\rm{dc}}}}{\rm{cos}}\theta } /3}\\ {\Delta {V_{13}} = \sqrt {(1 + 3{m^2})V_{{\rm{dc}}}^2 - 6m{V_{{\rm{dc}}}}{\rm{cos}}(\pi /6 - \theta )} /3} \end{array}} \right. $

(5)

According to Eq. (2), the duty cycle of voltage vectors V₁, V₇, and V₁₃ in T_s are

$ \left\{ {\begin{array}{*{20}{l}} {{T_0} = {T_{\rm{s}}}[2m{\rm{sin}}(\pi /3 - \theta ) - 1]}\\ {{T_1} = 2{T_{\rm{s}}}[1 - m{\rm{sin}}(\pi /3 + \theta )]}\\ {{T_2} = 2m{T_{\rm{s}}}{\rm{sin}}\theta } \end{array}} \right. $

(6)

As can be deduced from Fig. 4(a), angles δ₁, δ₂, and δ₃ between Δi₇, Δi₁, Δi₁₃, and d-axis are:

$ \left\{ {\begin{array}{*{20}{l}} {{\delta _1} = \frac{\pi }{2} - {\rm{arccos}}\frac{{V_{{\rm{ref}}}^2 + \Delta V_7^2 - V_{{\rm{dc}}}^2/9}}{{2{V_{{\rm{ref}}}} \cdot \Delta {V_7}}}}\\ {{\delta _2} = {\rm{arccos}}\frac{{V_{{\rm{ref}}}^2 + \Delta V_{13}^2 - V_{{\rm{dc}}}^2/3}}{{2{V_{{\rm{ref}}}} \cdot \Delta {V_{13}}}} - \frac{\pi }{2}}\\ {{\delta _3} = {\rm{arccos}}\frac{{V_{{\rm{ref}}}^2 + \Delta V_1^2 - 4V_{{\rm{dc}}}^2/9}}{{2{V_{{\rm{ref}}}} \cdot \Delta {V_1}}} - \frac{\pi }{2}} \end{array}} \right. $

(7)

Error current vectors Δ i₇, Δi₁, and Δi₁₃ are then decoupled into d-axis and q-axis components, as illustrated in Fig. 4(a).

$ \left\{ {\begin{array}{*{20}{l}} {\Delta {i_{7d}} = - \frac{{\Delta {V_7}{T_1}}}{{2{L_\sigma }}}{\rm{cos}}{\delta _1},\Delta {i_{7q}} = \frac{{\Delta {V_7}{T_1}}}{{2{L_\sigma }}}{\rm{sin}}{\delta _1}}\\ {\Delta {i_{13d}} = \frac{{\Delta {V_{13}}{T_2}}}{{2{L_\sigma }}}{\rm{cos}}{\delta _2},\Delta {i_{13q}} = - \frac{{\Delta {V_{13}}{T_2}}}{{2{L_\sigma }}}{\rm{sin}}{\delta _2}}\\ {\Delta {i_{1d}} = - \frac{{\Delta {V_1}{T_0}}}{{2{L_\sigma }}}{\rm{cos}}{\delta _3},\Delta {i_{1q}} = - \frac{{\Delta {V_1}{T_0}}}{{2{L_\sigma }}}{\rm{sin}}{\delta _3}} \end{array}} \right. $

(8)

By substituting Eqs. (5)-(7) into Eq. (8), it can be deduced that the expressions of d-axis and q-axis components of error current vector are both functions of m and θ.

3.2 Expression of RMS Value of Output Current Ripple

RMS value of output current ripple in one sample period is usually considered as the index to evaluate the quality of output current waveform.

$ {I_{{\rm{ ripple }}}} = \sqrt {\frac{1}{{{T_{\rm{s}}}}}\int_0^{{T_{\rm{s}}}} | \varDelta \mathit{\boldsymbol{i}}{|^2}{\rm{d}}t} $

(9)

As can be seen from Fig. 3, trajectories of error current vector during the first and the second half of each sample period are symmetrical with respect to the origin O. Therefore, only the RMS value of the output current ripple during (0, T_s/2] needs to be calculated. Thus, Eq. (9) can be transformed into

$ {I_{{\rm{ ripple }}}} = \sqrt {\frac{2}{{{T_{\rm{s}}}}}\int_0^{{T_{\rm{s}}}/2} {(\Delta {i_d}^2 + \Delta {i_{dq}}^2)} {\rm{d}}t} $

(10)

From Figs. 4(b) and 4(c), Δi_d and Δi_q can be derived as

$ \Delta {i_d} = \left\{ \begin{array}{l} \Delta {i_{{\rm{7d}}}}\frac{t}{{{T_1}}},0 \le t < {T_1}/2\\ \Delta {i_{{\rm{7d}}}} + \Delta {i_{13d}}\frac{t}{{{T_2}}},{T_1}/2 \le t < ({T_1} + {T_2})/2\\ \Delta {i_{{\rm{7d}}}} + \Delta {i_{13d}} + \Delta {i_{1d}}\frac{t}{{{T_0}}},({T_1} + {T_2})/2 \le t \le {T_{\rm{s}}}/2 \end{array} \right. $

(11)

$ \Delta {i_q} = \left\{ \begin{array}{l} \Delta {i_{{\rm{7}}q}}\frac{t}{{{T_1}}},0 \le t < {T_1}/2\\ \Delta {i_{{\rm{7}}q}} + \Delta {i_{13q}}\frac{t}{{{T_2}}},{T_1}/2 \le t < ({T_1} + {T_2})/2\\ \Delta {i_{{\rm{7}}q}} + \Delta {i_{13q}} + \Delta {i_{1q}}\frac{t}{{{T_0}}},({T_1} + {T_2})/2 \le t \le {T_{\rm{s}}}/2 \end{array} \right. $

(12)

By substituting Eq. (8) into Eqs. (11) and (12), and then substituting Eqs. (11) and (12) into Eq. (10), the RMS value of output current ripple can be obtained. It is also the function of modulation index and phase angle of reference vector. According to Eq. (10), the output waveform quality under different switching sequence can be calculated. Moreover, the arithmetic mean value of RMS value of output current ripple in one fundamental period can be easily calculated to compare the output performance of various modulation methods.

4 Error Model of NP Voltage Ripple

From Fig. 1, the neutral-point O is directly connected to the load through the clamping diode. When CPWM or DPWM0-DPWM3 is adopted, the unbalanced charge and discharge of upper and lower capacitor caused by the NP current i_O will lead to low-frequency oscillation in the NP voltage^[16]. However, the error model established in Section 3 is based on the premise that basic vectors are not changed. Only the high-frequency error caused by space vector synthesis is taken into account, which is not suitable for 3L-NPC converter. It is necessary to analyze the low-frequency error caused by the NP voltage ripple so as to evaluate the output waveform quality of 3L-NPC converter accurately. Then a unified error model considering both high-frequency error and low-frequency error needs to be established.

4.1 Calculation of NP Voltage Offset

The three-phase load current can be defined as

$ \left\{ {\begin{array}{*{20}{l}} {{i_{\rm{A}}} = \sqrt 2 I{\rm{cos}} (\theta - \varphi )}\\ {{i_{\rm{B}}} = \sqrt 2 I{\rm{cos}} (\theta - \varphi - 2\pi /3)}\\ {{i_{\rm{C}}} = \sqrt 2 I{\rm{cos}} (\theta - \varphi + 2\pi /3)} \end{array}} \right. $

(13)

where I is the RMS value of load current, φ is the power factor angle. With the change of modulation index, the reference vector rotates through different triangles in each fundamental period. The space vector diagram can be divided into three regions with regard to the modulation index. Take sector I as an example (as shown in Fig. 5), when 0≤m < 0.5 (region ①), the reference vector V_ref passes through triangle 3 and 4; when 0.5≤m < 0.577 (region ②), the reference vector V_ref passes through triangle 3, 2, 5, and 4; when 0.577≤m < 1 (region ③), the reference vector V_ref passes through triangle 1, 2, 5, and 6.

By taking triangle 1 of region ③ as an example, switching states 211, 210, and 200 will be applied to the converter one by one if DPWM1 is adopted. The phase angle θ of reference vector can be considered as a constant in each sample period. The offset of the NP voltage Δv can be obtained according to Eq. (14).

$ \begin{array}{l} \Delta v\left( t \right) = \\ {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \left\{ {\begin{array}{*{20}{l}} { - \frac{{\sqrt 2 I{\rm{cos}} (\theta - \varphi )}}{{2C}}t,0 \le t < {T_1}/2}\\ { - \frac{{\sqrt 2 I}}{{2C}}[{\rm{cos}}(\theta - \varphi ){T_1}/2 - }\\ {{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\rm{cos}}(\theta - \varphi - 2\pi /3)t],\quad {T_1}/2 \le t < {T_1} + {T_2}/2}\\ { - \frac{{\sqrt 2 I}}{{2C}}[{\rm{cos}}(\theta - \varphi ){T_1}/2 - }\\ {{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\rm{cos}}(\theta - \varphi - 2\pi /3){T_2}/2],{T_1} + {T_2}/2 \le t < {T_s}/2} \end{array}} \right. \end{array} $

(14)

When capacitance C, sample period T_s, load current I, and load power factor φ are not changed, Δv can be regarded as a linear function of t, where C is capacitance of C₁ and C₂. Fig. 6 shows the trend of the offset Δv against the phase angle θ.

Take the reference vector V_ref located at θ=0 as an example. In the first half of sample period, when t=0 (k-th instant), v_C1 is equal to v_C2, where v_C1 and v_C2 are voltages of C₁ and C₂, respectively, and NP voltage offset Δv=0; when t is between 0 and T₁/2, the switching state 211 is adopted, C₁ is discharged and C₂ is charged by the NP current i_o, and NP voltage offset Δv is increased; when t=T₁/2 ((k+1)-th instant), Δv=- $\sqrt 2 $Icos(θ-φ)T₁/4C; when t∈[T₁/2, T₁/2+T₂/2), the switching state 210 is adopted and Δv is still increased; when t=T₁/2+T₂/2 ((k+2)-th instant), Δv=- $\sqrt 2 $ Icos(θ-φ)T₁/4C+ $\sqrt 2 $ Icos(θ-φ-2π/3)T₂/4C, when t∈[T₁/2+T₂/2, T_s/2), the switching state 200 is adopted, NP current i_o=0, Δv remains unchanged; when t=T₁/2 ((k+3)-th instant), Δv is the same as that of (k+2)-th instant. The expression similar to Eq. (14) can be obtained when V_ref is in other regions.

4.2 Modification of Amplitude and Phase Angle of Reference Vector

When NP voltage ripple occurs, the output phase voltage is also shifted. The output voltage of each phase at (k+1)-th instant can be determined by the output voltage at k-th instant, the switching state, and the offset of the NP voltage, which yields

$ \left\{ {\begin{array}{*{20}{l}} {v_{{\rm{AO}}}^{k + 1} = v_{{\rm{AO}}}^k + (S - 1)\Delta v}\\ {v_{{\rm{BO}}}^{k + 1} = v_{{\rm{BO}}}^k + (S - 1)\Delta v}\\ {v_{{\rm{CO}}}^{k + 1} = v_{{\rm{CO}}}^k + (S - 1)\Delta v} \end{array}} \right. $

(15)

By substituting Eq.(15) into Eq.(1), the amplitude and phase of V_j can be determined. Define NP voltage ripple coefficient λ as

$ \lambda = \frac{{\varDelta v}}{{{V_{{\rm{dc}}}}/2}} $

(16)

From Fig. 7, when NP voltage is shifted (assuming λ= 0.25), the phase angle of V₇ is unchanged, while the amplitude is changed, and V₇ is changed to V′₇. The amplitude and phase angle of V₁₃ are both changed, and V₁₃ is changed to V′₁₃. Consequently, during the reference vector synthesis, error voltage vector Δ V₇ and Δ V₁₃ are changed into Δ V′₇ and Δ V′₁₃, respectively. The trajectory of error current vector is also changed, and the calculation results of the error model are inaccurate.

Therefore, the amplitude and phase angle modification model, i.e., the error model of NP voltage ripple, needs to be added to the error model of space vector synthesis. The high-frequency error caused by space vector synthesis and the low-frequency error caused by NP voltage ripple are both considered. Then the unified model can be established. The output waveform quality of 3L-NPC converter can be evaluated accurately. Fig. 8 shows the algorithm flow of the unified error model.

5 Experimental Analysis and Verification

As shown in Fig. 9, a rapid control prototype dSPACE DS1007 was adopted as the control circuit, and IGBT module Infineon F3L75R07W2E3_B11 was used as the power circuit to compose 3L-NPC converter. Parameters of the experimental prototype are listed in Table 2.

Based on the above experimental prototype, the experimental results of the output current are sampled. Instead of total harmonic distortion (I_THD), total demand distortion of output current (I_TDD) is introduced to describe the output performance of 3L-NPC. I_TDD varies slightly throughout the modulation range compared with I_THD, which is beneficial for the comparison of the performance of different modulation strategies. Define I_TDD as^[18]

$ {I_{{\rm{TDD}}}} = \frac{{\sqrt {0.5\sum\limits_{h = 2}^\infty {I_h^2} } }}{{{I_{{\rm{nom}}}}}} $

(17)

where I_h represents the h-th harmonic of load current and I_nom represents normal load current. Then the curve of I_TDD versus the modulation index can be obtained by the Fourier analysis. On this basis, the RMS value of the output current ripple calculated by the traditional error model and the unified error model are compared and analyzed.

Fig. 10 shows the experimental results of output line voltage v_AB, output current i_A, and neutral-point voltage v_C1-v_C2 for DPWM3 when m=0.8, DC-link capacitance C=1 000 μF, and C=180 μF, respectively.

As can be seen, the neutral-point voltage ripple increases with the reduction of DC-link capacitor, and I_TDD increases consequently.

Fig. 11 shows the variation of I_TDD under CPWM and DPWM0~DPWM3 when the DC-link capacitance C=1 000 μF and C=180 μF, respectively. From Fig. 11, the decrease of the DC-link capacitance will lead to the increase of I_TDD. From Eq. (14), the smaller the C is, the larger the NP voltage offset Δv is. Thus, the output waveform quality of the converter decreases visibly.

Fig. 12 shows the calculation results I_ripple of the traditional error model. It can be deduced from Section 3 that the NP voltage offset Δv is not introduced into the traditional error model. The influence of NP voltage ripple on the output performance cannot be measured. Comparison of Fig. 11 and Fig. 12 shows that the calculation results of the traditional error model coincide with the curve of I_TDD when the DC-link capacitance is relatively high. When the DC-link capacitance is low, NP voltage offset Δv increases. There is a large deviation between the calculation results of I_ripple and I_TDD, which indicates that they are inaccurate.

Fig. 13 illustrates the calculation results I_ripple of the unified error model presented in this paper.

When the modulation index m=0.3, the RMS value of output current ripple I_ripple of DPWM3 (C=180 μF) is 0.1 A and the amplitude of output current is 1.2 A. Thus, according to the unified error model, the ripple component accounts for 8.3% of the output current, while for the experimental results shown in Fig. 10, the value of I_TDD is 9.05%. The output current is obviously affected by the error during reference vector synthesis and the neutral-point voltage ripple. So the calculation results are in accordance with the experimental results.

Comparison of Fig. 11 and Fig. 13 shows that the calculation results of the unified error model are consistent with the curve of I_TDD under different DC-link capacitance conditions. It shows that the influence of NP voltage ripple on the output waveform quality can be represented by the error model of NP voltage ripple. The disadvantage of the traditional error model, which only evaluates the high-frequency error caused by space vector synthesis and does not consider the low-frequency error introduced by NP voltage ripple, can thus be modified.

It is worth noting that the value of L is only a proportional coefficient in Eq. (3). The comparison of RMS value of output current ripple for different modulation strategies is not affected by the variation of L. Even if L is changed to L^′, the results in Fig. 12 only need to be multiplied by L/L′, and the comparative relationship between I_ripples remains the same.

6 Conclusions

An accurate evaluation method of the output waveform quality for 3L-NPC converter is presented in this paper. RMS value of output current ripple was used as the index. The influence of the NP voltage ripple on the output waveform quality was fully considered. The mathematical model of NP voltage offset was established and used to modify the amplitude and phase angle of the basic vector. The unified error model considering both the high frequency error caused by space vector synthesis and the low frequency error introduced by NP voltage ripple simultaneously was established. The accurate evaluation of the output waveform quality of 3L-NPC converter was realized.