METHOD FOR CONTROLLING HARMONICS AND RESONANCES IN AN INVERTER

FIELD OF THE INVENTION

The invention relates to the field of high power inverters. In particular, the invention relates to a method and a program element for controlling an inverter, a computer-readable medium, a controller of an inverter and an inverter.

BACKGROUND OF THE INVENTION

Electrical inverters may be used for transforming an input voltage into an output AC voltage. For example, the AC voltage from an electrical grid may be transformed into a variable AC voltage supplied to an electrical drive or another AC voltage to be supplied to another electrical grid.

For generating the usual multi-phase output voltage, the inverter comprises a plurality of semiconductor switches, for example thyristors or IGCTs, which may be controlled by an electronic controller of the inverter.

One possibility of controlling an inverter is direct torque control (DTC). In DTC the torque and the flux of the electrical drive may be controlled by estimating the actual torque and the actual flux form measured voltages and currents that are output from the inverter, and selecting a switching state for the switches of the inverter in such a way that the actual flux and actual torque move towards a reference flux and a reference torque, when the switching state is applied to the inverter switches.

However, one of the drawbacks of DTC may be the fact that the average switching frequency of the inverter cannot be directly controlled. Since the switching frequency is usually proportionally related to the switching losses of the inverter, which may be a major part of the overall losses of the electrical drive, any reduction of the switching frequency may have a significant impact on the operational cost of the drive and increase the overall system robustness and reliability. Such a reduction has been shown to be possible through the Model Predictive Direct Torque Control (MPDTC) method, which, for example, is described in EP 1 670 135 Al. In MPDTC, possible sequences of switching states are determined, which may be applied to the inverter switches in the future. These switching sequences may be constrained by the actual switching state of the inverter and the inverter topology. Since two different switching states may result in the same output phase voltages, i.e. the same voltage vector, more general, possible voltage vector sequences may be considered that comprise a sequence of voltage vectors. For each possible switching sequences or voltage vector sequence, the switching losses are estimated and the first switching state of the switching sequence is applied to the motor.

For achieving a more drive friendly voltage supply, it is possible to place a harmonic filter, for example an LC filter, between the inverter and the electrical drive. Such a filter may smooth out the effects of irregular switching actions of DTC, which may lower the harmonic distortion of the motor currents.

However, in this case, the introduction of the harmonic filter may render drive quantities, like flux and torque, not directly controllable. Specifically, it is no longer possible to directly and rapidly manipulate the stator flux by the application of a specific voltage vector, since what is applied to the motor terminals is the voltage of the capacitor of the LC filter, which features much slower dynamics, and is not immediately affected by the applied voltage vector. This implies that the control objective of producing the desired torque by the suitable and rapid positioning of the stator flux vector is no longer attainable.

To overcome this issue with DTC, one solution is to modify the control problem and target the control of certain inverter (instead of motor) variables. Namely, the notions of the inverter flux and inverter torque are introduced; the first being the integral over time of the inverter voltage, and the second expressing the interaction of the inverter flux and the inverter currents. These two variables are different from the actual corresponding motor flux and torque (especially during transients) but their average values at steady state are the same. Those virtual notions are the electric equivalent of the motor torque and flux though they do not correspond to physical quantities. The advantage of introducing and working with such quantities is that they can be directly and quickly manipulated by the application of the proper voltage vector. These features imply that the DTC problem can be recast as an inverter control problem, where the objective is to keep the inverter flux and torque within certain bounds. The physical properties of the system then assure that the motor will also reach the appropriate steady state conditions.

DESCRIPTION OF THE INVENTION

An additional issue that arises, when a harmonic filter is introduced, may concern the frequency decoupling that must be maintained to avoid the emergence of unwanted harmonics in the current and torque of the drive. Specifically, since the resonance frequency of the harmonic filter might interfere with the switching frequency, it is necessary to set the latter such as to preserve the frequency decoupling. Therefore, the simple minimization of the switching frequency may not longer be used as a way to indirectly affect the switching losses as was the case with the above mentioned MPDTC algorithms.

Further issues that may require attention may be the presence of electrical and mechanical resonances that alter the inverter performance.

Issues with the torque and current harmonics may arise in particularly in the case below 10% and 20% of the maximal speed of the drive. Such distortions are characterized by the presence of a 6 ^th harmonic in the drive and inverter torques and 5 ^th and 7 ^th harmonics in the inverter currents and phase voltages. The 5 ^th and 7 ^th harmonics are generated by the switching and then transferred to the inverter current. Finally the distortion appear as 6 ^th harmonic on the inverter and motor torque.

Another, yet related issue may be the following: The mechanical load is generally connected via a rotational shaft with the drive. In medium voltage drive applications the inertia of the mechanical load is often very large and the shaft is fairly long and stiff. Such applications include large compressor trains, which are typically found in the oil and gas industry. In the very high power range, the electrical machine acts only as a starter and helper motor, while the majority of the power is provided by a gas turbine that is also connected to the shaft. In such applications the mechanical system formed by the load, gas turbine, electrical machine and shaft exhibits distinctive torsional resonant modes. The natural frequencies of these torsional modes range from a few Hz to up to several hundred Hz. The amplification factor (the so called Q factor) is often 40 and more.

The inverter of medium voltage drives typically employs a low switching frequency giving rise to pronounced torque harmonics. If such a torque harmonic coincides with a natural frequency of the mechanical system large mechanical torsional vibrations can occur. This issue of mechanical resonances may be directly related to the former problem with electrical resonances, in the sense that the same damping principles can be employed. The main difference is that the mechanical resonances typically have a lower natural frequency than the electrical phenomena.

It is an object of the invention to provide an MPDTC controlled inverter with low switching losses and low harmonic distortion.

This object is achieved by the subject-matter of the independent claims. Further exemplary embodiments are evident from the dependent claims and the following description.

A first aspect of the invention relates to a method for controlling the harmonics and resonances in an inverter. The inverter may be a three-phase, three-level inverter for driving an electrical motor. The electrical system may be a high or medium voltage system. The method may be part of a on the Model Predictive Direct Torque Control (MPDTC) algorithm, in particular for the control of a three-phase induction motor comprising a three- level dc-link inverter with an output LC filter. In such a way, the method may be adapted for controlling an inverter for an electrical system.

According to an embodiment of the invention, inverter torque predictions of the MPDTC algorithm are filtered and added to the compensation term of the reference torque. This compensation enables to reduce the harmonic distortion at low speed while preserving the achieved losses reduction in all the operating range. For example, harmonic distortion that is present in the torque at low speeds (10% and 20%) may be handled in this way. According to an embodiment of the invention, predicted possible voltage vector sequences of the MPDTC algorithm are discarded, when the predicted harmonic distortion is above a predefined value. Already at the MPDTC control decision level, the voltage vector sequences that produce minimal harmonics may be chosen and the other voltage vector sequences may be discarded.

Summarized, the embodiments as described in the above and in the following for damping harmonics are based on using the MPDTC predictions to extract information about the frequency content that is about to be introduced into the electrical system by the actions of the control algorithm. These embodiments may not have to replace standard state-of-the art active damping, but may act as additional compensation to handle the harmonic distortion that the standard compensation cannot address. Moreover, these embodiments may not interfere with the performance of MPDTC in terms of switching losses reductions.

According to an embodiment of the invention, the method comprises the step of (a) determining of possible voltage vector sequences that may be generated by the inverter by switching switches of the inverter and that may be supplied to the electrical system. In MPDTC, with a discrete-time algorithm possible time sequences of switching states may be determined, which may be applied to the inverter switches in the future. Since two different switching states may result in the same output phase voltages, i.e. the same voltage vector, more general, possible voltage vector sequences may be considered that comprise a time sequence of voltage vectors. In a three-phase system a voltage vector may comprise three voltage values. Usually, a voltage vector sequence is determined over a time horizon of 2 or three steps (i.e. time instants).

According to an embodiment of the invention, the method comprises the step of (b) determining of candidate sequences from the possible voltage vector sequences by estimating system response data for each voltage vector sequence and by keeping voltage vector sequences with admissible system response data. The system response data may comprise the inverter torque, the inverter flux and neutral point potentials. In particular, trajectories of these values may be estimated over the horizon. In general, a system response data may be estimated by estimating a trajectory of a system response value from a possible voltage vector sequence, for example a torque or flux trajectory. A possible voltage vector sequence may admissible, if the corresponding estimated trajectory is lying within bounds or the estimated trajectory approaches the bounds that are based on a reference value. The system response value may comprises a predicted torque, a predicted flux, and/or a predicted neutral point potential of the inverter and the reference value comprises a reference torque, a reference flux and/or a reference neutral point potential.

According to an embodiment of the invention, the method comprises the step of (c) determining a cost value for each candidate sequence, wherein the cost value is based on predicted switching losses of the inverter when switched with the candidate sequence.

According to an embodiment of the invention, the method comprises the step of (d) applying a first voltage vector of a candidate sequence with the lowest cost value to the inverter. Not the whole sequence but only the first element may be applied to the inverter. According to an embodiment of the invention, the method comprises the steps of (e) extracting frequency information from predicted data, the predicted data comprising data of at least one of the possible voltage vector sequence and/or system response data. For example, as already mentioned the frequency information may comprise data about the 6 ^th harmonic distortion of the inverter torque. The at least one of the possible voltage vector sequences may be in particular all possible voltage vector sequences, at least one and / or all candidate sequences and / or the candidate sequence with the lowest cost value.

According to an embodiment of the invention, the method comprises the steps of (f) damping harmonic distortion of the electrical system by reintroducing the extracted frequency information into a control loop of the inverter. As already mentioned, the extracted frequency information may be used to discard specific voltage vector sequences and/or to alter reference values that are used for determining, if a voltage vector sequence is admissible or not.

According to an embodiment of the invention, in step (e) a harmonic distortion of a phase voltage is determined by extracting frequency information from voltage values associated with a voltage vector sequence. The voltage values may comprise voltage values of past sampling times and of future sampling times. In particular, phase voltage differences are calculated and filtered to calculate a value, which contains frequency information about a frequency band or range of a sequence of phase voltage differences. In this case, step (f) is executed by discarding the voltage vector sequence, when the harmonic distortion (i.e. the value) of the phase voltage leaves predefined bounds.

According to an embodiment of the invention, in step (e) a harmonic distortion of a predicted torque is determined, in step (f) a reference torque is modified with the harmonic distortion of the torque, and in step (b) admissible system response data is determined with bounds based on the reference torque. For example, a correction value may be added to the original reference for obtaining a modified reference torque. The reference value may be determined by applying a filter to the torque trajectory corresponding to the voltage vector sequence which first element is applied to the inverter.

According to an embodiment of the invention, in step (e) a harmonic distortion is determined from the predicted data by applying a digital filter to the predicted data for extracting frequency information. Formulas for a digital filter are given with respect to the description of Fig. 2. The digital filter may be a high-pass or band-pass filter, which may be adjusted by coefficients in the formulas.

According to an embodiment of the invention, in step (e) a sliding discrete Fourier transformation may be applied to the predicted data for extracting frequency information. In particular with a Goertzel algorithm a specific frequency component of the predicted data may be extracted very efficiently.

According to an embodiment of the invention, the frequency information being extracted in step (e) at least contains the frequency information of the harmonic distortion which is to be damped.

A further aspect of the invention relates to program element or computer program for controlling an inverter, which when being executed by at least one processor is adapted for executing the steps of the method as described in the above and in the following.

A further aspect of the invention relates to a computer-readable medium, in which such a program element is stored. A computer-readable medium may be a floppy disk, a hard disk, an USB (Universal Serial Bus) storage device, a RAM (Random Access Memory), a ROM (Read Only memory), a FLASH and an EPROM (Erasable Programmable Read Only Memory). A computer readable medium may also be a data communication network, e.g. the Internet, which allows downloading a program code.

A further aspect of the invention relates to a controller for controlling an inverter. The controller is adapted for executing the method as described in the above and in the following. For example, the controller may be an FPGA.

A further aspect of the invention relates to an inverter for supplying a load with an AC voltage. The inverter may comprise an inverter circuit with switches, the inverter circuit being adapted for generating an AC output voltage for at least one phase, a filter circuit interconnected between the inverter circuit and the load, and a controller for controlling the switches. The controller is adapted for executing the method as described in the above and in the following, thus being adapted for compensating harmonic distortion of the AC voltage. It has to be understood that features of the method as described in the above and in the following may be features of the program element, the controller and the inverter as described in the above and in the following and vice versa.

If technically possible but not explicitly mentioned, also combinations of embodiments of the invention described in the above and in the following may be embodiments of method, the program element, the controller and the inverter.

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

BRIEF DESCRIPTION OF THE DRAWINGS

The subject matter of the invention will be explained in more detail in the following text with reference to exemplary embodiments which are illustrated in the attached drawings. Fig.1 schematically shows an inverter according to an embodiment of the invention.

Fig. 2 shows a flow diagram with an MPDTC algorithm according to an embodiment of the invention.

The reference symbols used in the drawings, and their meanings, are listed in summary form in the list of reference symbols. In principle, identical parts are provided with the same reference symbols in the figures.

DETAILED DESCRIPTION OF EXEMPLARY EMBODIMENTS

Fig. 1 shows an inverter 10 with an inverter unit 12 with three inverter legs 14. The inverter legs 14 are connected parallel to a DC link 16 and are adapted to transform a DC voltage from the DC link 16 into a phase 20 of the variable output voltage of the inverter 10. Each inverter leg 14 is adapted to connect the output phase 20 with the positive or negative voltage in the DC link or the neutral voltage in the neutral point 18.

In such a way a three-phase, three level output voltage is provided to the drive 24 which is connected over a LC filter 22 with the inverter unit 12.

The inverter 10 comprises further a controller 26 which is adapted to control the switches 28 in the inverter legs 14 and to measure the currents and voltages in the output phases 20 and the neutral point potential in the neutral point 18.

In the controller 26, the MPDTC control algorithm is implemented, for example on a FPGA. For example, the inverter 10 may be an ACS6000 or ACS1000 designed by ABB. Fig. 2 shows a flow diagram with an MPDTC algorithm that may be implemented in the controller 10.

Considering the dynamical model of the drive 24, inverter unit 12 and LC filter 22, the state vector may be denoted by x = fi _inv , v _c , W _inv , i _stat , v _n J, which comprises the inverter current i _inv , the capacitor voltage v _c , inverter flux W _inv , stator current i _stat and neutral point potential v„. The output vector may be denoted by y = [T _inv , Ψί„ _ν , v„] comprising inverter torque T _inv , inverter flux W _inv and neutral point potential v„ respectively. The data of the state vector x and of the output vector y are the system response data. In the controller the discrete time version of the dynamical model is implemented and the signals i _inv , v _c , T _inv , Ψίην, istat, v _n are represented as discrete function depending on discrete time steps k, i.e. iinv(k), v _c (k), T _inv (k), W _inv (k), i _sta t(k), v„(k\ where k is an integer number

Given the current state x(k), the last voltage vector u(k - 1), the bounds on the controlled variables, and using the discrete-time model of the DTC drive, the controller 26 computes at time-instant k the voltage vector u(k) according to the following procedure.

In step S10, the last voltage vector u(k - 1) that was applied to the switches 28 of the inverter unit 12 is used for calculating the possible voltage vector sequences U'(k) = [ u'(k), u'(k + 1), ... , u'(k + N _H - 1)J, where i e 1 and I is the index set of the possible voltage vector sequences. The possible voltage vector sequences are determined by taking into account the constraints on the switch transitions induced by the inverter topology. For example, all possible voltage vector sequences U'(k) over a horizon N _H of N _H = 2 steps are determined.

In step SI 2, the actual inverter flux Ψ and the actual inverter torque T are determined from an inverter model and the actual voltages and currents measured in the inverter 10. In step S14, for the possible voltage vector sequences U'(k), the system response is calculated. In particular, the open-loop inverter torque T _inv , flux W _inv and neutral point potential v„ trajectories Y'(k) are calculated starting from x(k) over the horizon N _H , for = 2 given by Y(k) = [y ^l (k), y ^l (k + 1), y ^l (k + 2)J.

In step SI 6, the candidate sequences U'(k) with i e l _c cl are determined. The candidate sequences are those possible voltage vector sequences that have output trajectories T(k) that are either feasible at the end of the horizon or pointing in the proper direction for all time-steps within the horizon. Feasibility means that the trajectory of the controlled variable ¥ _inv , T _inv , v„ lies within its corresponding bounds at time-step k to k+Nn- Pointing in the proper direction means that the trajectory of the controlled variable W _inv , T _inv , v„ is not necessarily feasible, but the degree of the violation is decreasing for all time-steps within the prediction horizon, which means for the time steps k to k+Nn- In other words, the trajectory is pointing towards the bounds.

The bounds of the controlled variables W _inv , T _inv , v„ depend on the corresponding reference values _re f, T _re f, , which are supplied to the controller 10, for example by a speed measurement of the drive 10, or which may be preset, for example, for a drive with constant speed. The controlled variable v„ should be controlled to be zero (v„ = 0) or within a predefined range around zero.

The condition applied in step S 16 needs to hold component wise, i.e. for all three controlled variables W _im> , T _inv , v„. As an example, consider the following situation: the inverter torque T _inv is feasible, the inverter flux W _inv points in the proper direction, and the neutral point potential v„ is feasible.

In step S I 8, for the candidate sequences, the output trajectories Y'(k) are extrapolated in excess of the horizon. For the given example of NH = 2, Y'(k) are extrapolated from time- instant k+2 on linearly using the samples at k+1 and k+2. Alternatively the extrapolation could also be non-linear. The number of the extrapolation time-steps is derived when the first of the three output variables W _inv , T _inv , v„ leaves the feasible region in between of the corresponding upper and lower bound. In this way for each candidate sequence U'(k), the time step k at which at least one of controlled variables is not feasible any longer. From the time step k, the number of extrapolation time-steps n i e l _c before the next predicted switching can be determined.

In step S20, for each candidate sequence the cost are calculated that approximates the average switching losses by the number of switch transitions weighted with respect to each semiconductor switch and the current through it over the number of time-steps in each of the i candidate sequence can be applied before switching again. The number of time-stepsij can be interpreted as a time-varying horizon. Next, the sequence U'fk) with the minimum cost is chosen.

In step S22, the first voltage vector u'(k) of the chosen sequence lf(k) is applied to the inverter switches. At the next time-instant the algorithm starts again. To compensate harmonic distortions, for example, the above mentioned 6 ^th harmonic in the torque and/or the low-frequency mechanical resonances, the control algorithm may be supplemented in the following ways. The following embodiments are described with respect to the damping of the 6 ^th harmonic; however they may be also applied to damping of low-frequency mechanical resonances or in general to all kinds of harmonic distortions.

The first embodiment directly tackles the 6 ^th harmonic in the torque. The predictions of the inverter torque T _inv (k) (T(k) in Fig. 2) provide possible time-evolutions that the system can follow, depending on the selection of the optimal vector u(k) by MPDTC.

In step S24, the predicted torque sequence T _inv (k), k=0, 1, 2, is filtered through a high pass filter with cutoff frequency at approximately 30 Hz to retain the 6 ^th and higher order harmonics

N

t _v6 (k)=∑ a _t T _inv (k) + b r _inv6 (k - l - i) where a bj are the filter coefficients which define the frequency band of the filter, , i.e. the range in which a frequency is not or nearly not damped by the filter. N is the length of the filtered signal. For a high pass filter, the filter coefficients may be chosen such that the upper bound of the frequency band is higher than the frequencies the discrete time algorithm can resolve. The value or signal _jnv6 (k) contains the information about the undesired harmonics that will be produced in the torque T _Jnv (k) by the chosen input vector u(k). Depending on the voltage vector sequence U(k) that the MPDTC algorithm selects as the best option to minimize the losses, the corresponding signal _jnv6 (k) , describing the harmonic content that the controller's actions will be introducing to the torque, is added to the reference of the inverter torque _re f(k), to produce the new or altered reference torque:

T' _re k) = T _re k) + f _inv6 (k)

Introducing the filter may also improve the harmonic content at higher speeds, but in some cases may decrease the performances in terms of losses, due to the introduction of excessive oscillations in the hysteresis bounds.

Alternatively, a band-pass filter may be considered. For example, the frequency band of the filter may be set to comprise the 6 ^th harmonic of the inverter torque, i.e. the frequency of the drive. The result in such a case is the elimination of the harmonics problem at low speed, but for this case the filter may need to be tuned separately for each specific speed, which may be impractical for a drive with variable speed.

The first embodiment is based on the idea of predicting the harmonics that will be introduced in the torque T _inv (k) by the selected switching signal u(k) and compensate the reference signal T _ref to cancel out such harmonics.

According to a second embodiment, already at the MPDTC control decision level, the voltage vector sequences lf(k) that produce minimal harmonics are chosen. The second embodiment addresses the problem at its source, by constraining the voltage vector sequences lf(k) to avoid the 5th and 7th harmonic in the switching phase voltages.

This is done in the MPDTC algorithm by introducing the harmonic content of the input voltages as three (one per phase) additional variables v 's7(k), j = 1, 2, 3 that are controlled together with the inverter flux Ψ and torque T. The variables v ¹ 57(k) contain frequency information of a possible voltage vector sequence U'(k).

Specifically, in step SI 6, using the history of the converter voltages u'(k), u(k - 1), ... , u (k - p) during (a few) past sampling times p and the future possible sequences of voltage vectors U'(k) over the prediction horizon that the MPDTC considers as options, the phase voltage differences v'fk) = ii(k)- ii(k), i, j = 1, 2, 3 are filtered across a Nth order filter to capture the 5th and 7th harmonic signals of the phase voltages as follows

where, as before, a bj are the filter coefficients which define the frequency band of the filter. Once obtained, the variables v ¹ ₅₇ (k), j = 1, 2, 3 become part of the control problem, and the MPDTC algorithm tries to keep them within pre-specified bounds (in the same way that inverter torque T and flux Ψ are also kept within their respective bounds), thus minimizing the effect of these harmonics on the system's behavior.

Summarized, in the second embodiment, a phase voltage difference sequence v'(k) is calculated from a possible voltage vector sequence lf(k) and a frequency information value v^f j is extracted from the phase voltage difference sequence. The harmonic distortion of the electrical system is damped by keeping those voltage vector sequences lf(k) which frequency information value v ¹ ₅₇ (k) is within predefined bounds.

Similar to the first embodiment, in the second embodiment a high pass filter or a band pass filter may be used.

Instead of using the high- or band pass filters that are applied in the first and second embodiment, the spectral information may also be obtained by using a Sliding Discrete Fourier Transformation (SDFT). The SDFT's main advantage may be its computational efficiency and simple implementation. In the case of the setting considered here we are interested in obtaining the harmonic content for the torque, current or phase voltage differences. Consider a discrete-time signal x (inverter torque, current or phase voltage difference) at time-step k that we wish to analyze over a certain length in time using a fixed-length window with N samples (N is usually the length of the filtered signal).

x(k) = {x(k - N + l), x (k - N + 2), ... , x(k - l), x(k) } In particular, x(k) may be the inverter torque T(k) or the phase voltage difference v'(k).

Assume that the sampling interval is T _s . NT _S could be, for example, equal to the length of one fundamental period. Performing an N-DFT operation on x(k) yields the Fourier transformed discrete-time signal

X(k) = {Xo(k), X _} (k), ... , X _N . ₂ (k), X (k)}

Here, we are interested only in certain specific harmonics, say in the n-th value ofX(k), ie X„(k).

Instead of computing the whole spectrum X(k) using an FFT, the Goertzel algorithm can be used to compute isolated X _n (k). We also observe that at two successive time-steps k-1 and k, the windowed sequences x(k-l) and x(k) substantially contain the same elements. In such a sliding window scenario the information computed at the previous time-step k-1 can be used to drastically reduce the computational effort at time-step k.

Specifically, assume that at k-1 X _n (k-1) was computed. Then, it can be shown that the spectrum of the shifted time-sequence at time-step k is given by

X„(k) = fX„(k-l) - x(k-N) + x(k) ] d ^{2n n/N}

This implies that regardless of the window length N, the SDFT requires a constant number of operations to compute a successive DFT, namely two real additions and a complex multiplication. Note that this computation assumes that the DFT of the previous time-step is available.

While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive; the invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art and practising the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or controller or other unit may fulfil the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting the scope.

LIST OF REFERENCE SYMBOLS

10 inverter

12 inverter unit

14 inverter leg

16 DC link

18 neutral point

20 output phase

22 LC filter

24 drive

26 controller

28 switch