【正文】 As pointed out in [4], a DTC scheme can be used to control, besides theelectromagnetic torque of course, the directaxis current or reactive power instead of the stator flux linkage. In the following these schemes are not considered, as such all theconsidered schemes are of the type direct torque and flux control (DTFC). In [5] anexcellent overview of DTC techniques is given, but the focus is on DTC for inductionmachines.In this section an attempt to summarize the different known implementations ofDTC for PMSMs is given. The schemes are divided according to voltage vectorselection, but are also different in terms of (initial) stator flux estimation and the use ofposition sensors. Some of the discussed schemes namely require the rotor position ,thus losing the advantage of inherent motionsensorless control.Switchingtable DTC1) Basic Switchingtable DTC: A classical DTC scheme has a hysteresis parator for the stator flux linkage and a quantisizer for the torque. A typical scheme is shown in , the quantities and denote reference values and the optional encoder is shown as a dashed line. The instantaneous error for the stator fluxlinkage thus has two possible values (1 and ?1), whereas the instantaneous torque error has three (?1, 0 and 1). Furthermore the plane is divided in six sections. The errors and , together with the section number containing the stator flux vector serve as input for a switching table. The output of the switching table is one of the eight possible voltage vectors. Such a scheme is implemented in [6] for SPMSMs, with the same switching table as used in [1] for induction machines. Furthermore a first order filter is proposed as quasiintegrator to solve the problem of initial flux estimation. As the steadystate output of a first order filter is independent of the initial conditions the quasiintegrator will indeed yield good results, but not at start up of the drive.Switchingtable DTC is also implemented in [3], but no zero voltage vectors areused to control the motor. This essentially reduces the quantisizer for the torque error toa normal hysteresis parator. The flux estimation is based on (3) and the initial flux position is assumed to be known. The method is applicable to IPMSMs and SPMSMs.In [7] and [8] this scheme with reduced switching table is applied for IPMSMs and the initial rotor position is known from a low resolution encoder. It is shown that by varying the stator flux linkage reference either maximum torque per ampere (MTPA) or field weakening operation of the drive is possible. Recent papers have further reported on the use of these referencefluxgenerating methods. In [9] a maximum torque per flux (MTPF) scheme is discussed, based on switching table DTC. A method to optimize efficiency under switching table DTC of PMSMs is given in [10], where the stator flux linkage is selected to yield maximu efficiency. In all of these referencefluxgenerating methods offline calculations are needed to determine the lookup tables for the reference stator flux.2) Extended Number of Voltage Vectors: One of the main drawbacks of DTC is theripple in torque and stator flux linkage. This ripple can be reduced by using more,different voltage vectors. When motoring in basic DTC there is only a limited numberof voltage vectors available per sector, the switching table chooses the most it is very unlikely that both the radial and tangential ponents of the vector are aligned with the desired ponents. With adding more voltage vectors and/or adding sectors, a closer match for both ponents can be achieved. In [11] a DTC scheme is proposed which allows, by means of space vector modulation (SVM), to use 24 voltage vector directions at three amplitude levels. With the quantization of torque and flux error and the availability of 72 voltage vectors a different switching table is constructed. As a result a lower torque ripple is achieved. In [11] SVM is used to generate more, different voltage vectors during the entire operation of the drive. It is however also possible to use a hybrid algorithm, making more voltage vectors available during certain operating conditions. In [12] a method is proposed to ensure a fast torque response during startup of IPMSMs. At startup SVM is used to generate the optimal voltage vector, . the voltage vector allowing the fastest rise in torque. However the calculations are dependent on the rotor position, thus the initial rotor position has to be known from an encoder and during the torque development duration the rotor position is assumed to be constant. Once the torque reference value is reached, a regular switching table (containing only voltage vectors) is used and there is no further need for the encoder.Multilevel converters make more voltage vectors available to control flux andtorque, hence reducing ripple and achieving a less variable switching frequency. As adisadvantage more power switches are needed, thus increasing system cost andplexity as well as switching losses. In [13] such a DTC is proposed for induction machines, but application to PMSMs is not reported in Switching Frequency DTCTo further eliminate torque and stator flux ripples and to obtain a fixed switchingfrequency, it is possible to use a model of the PMSM to calculate the most appropriatevoltage vector during the next switching interval. This most appropriate voltage vector can then be realised by SVM. Furthermore the use of SVM permits avoiding some other disadvantages of switchingtable DTC, such as violating polarity consistency rules, high sampling frequency for digital implementation of the parators and distortions due to sector changes. However there are several ways of calculating the most appropriate voltage vector and the required motor parameters and putational plexity have to be taken into account when paring the different schemes.Voltage Vector Calculation1) SVMDTC with Closed Loop Torque Control: Atypical scheme of this type is shown in