Improved Multilevel Hysteresis Current Regulation and Capacitor Voltage Balancing Schemes For Flying Capacitor Multilevel Inverter
Improved Multilevel Hysteresis Current Regulation and Capacitor Voltage Balancing Schemes For Flying Capacitor Multilevel Inverter
Improved Multilevel Hysteresis Current Regulation and Capacitor Voltage Balancing Schemes For Flying Capacitor Multilevel Inverter
2, MARCH 2008
TABLE I
SWITCHING SCHEME FOR FIVE-LEVEL FCMLI
Using (1), Table I lists the switch combinations used to syn- (3)
thesize five output voltage levels and the corresponding
states of the flying capacitors. Charging of a capacitor is indi- It is evident from (3) that the error can be reduced
cated by , the discharging by , while ‘NC’ indicates neither by increasing or decreasing , depending on the polarity of
charging nor discharging. The switch states given are for the out- . To implement the logic for this correct voltage level
going direction of the current waveform ( in Fig. 2). The states selection logic, two schemes have been discussed in the fol-
( and ) will reverse for the incoming current. It can be seen lowing subsections on the basis of the single-phase five-level
from Table I that the structure offers multiple switch combina- inverter of Fig. 2.
tions for and . As such redundancies are
available, one can choose a preferential switching state for these A. Multi-Offset-Band Multilevel Hysteresis Controller
output voltage levels that will help in maintaining the capacitor A five-level inverter can be current controlled by using the ex-
voltages. The operation and structure details of FCMLI can be tension of the three-level hysteresis control technique proposed
found in [5], [10]–[12], [14]–[18]. in [6] to five-level. As suggested in [6], for an -level inverter,
520 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 23, NO. 2, MARCH 2008
Fig. 4. (a) Hysteresis current control with fixed voltage applied at the band
crossings of the current error (A). (b) Inverter switched output voltage (V).
Fig. 3. Multi-offset-band five-level hysteresis current control.
TABLE II
bands are required in this scheme. By following the scheme SIMULATED FIVE-LEVEL FCMLI DEVICE PARAMETERS
of [6], a possible four-band arrangement for control-
ling a five-level inverter is shown in Fig. 3. It is evident from the
figure that as the current error touches the corresponding bound-
aries of , fixed output voltage levels are switched. The
switching takes place when: 1) the error reverses sign and then
crosses the boundary of a band and 2) the error does not reverse
sign but keeps increasing in magnitude and crosses the boundary
of the next higher band for positive error and lower band for
negative error. It can be followed that 0 V is switched at the definition, the voltage level 0 is switched at . However, before
lower limits of and upper limits of at , the output voltage level was and at , the error is
the upper limit of at the lower limit of moving away from the zero line in the negative direction. There-
at the upper limit of and at the lower limit of . fore, was first required to force the error in opposite
Its limitation when using this scheme for a five-level inverter direction. But due to the logical sequence of control, 0 voltage
can be seen by looking at the current error path from to . level is applied at , resulting in rapid increase in the positive
It is evident that a voltage level transition from to 0 V slope of the current error away from the zero-line. The error
occurs at , thereby, skipping the level . This results then touches the upper boundary of and due to more than re-
in bad quality inverter output voltage and large voltage stress quired voltage working on it, crosses and again reaches at the
across the devices at the switching instants. upper boundary of . This results in consecutive switching of
To get a further insight into operational performance of the and , respectively. It is also evident that in this
five-level hysteresis control of Fig. 3, a simulation study is per- process, the intermediate level is skipped as the current
formed using PSCAD/EMTDC software for a five-level inverter error travels from to . This operation is repeated in the con-
(of Fig. 2), supplying an RL-load of and secutive switching cycles and results in degraded voltage wave-
mH. The back emf voltage is taken as zero and the form. It should be noted that since the current error remains in
inverter devices are assumed nearly ideal. The dc-link voltage the allotted bands, the controlled current follows its reference.
is 72 V and flying-capacitor voltages (Fig. 2) are supposed to It is the voltage waveform, which is degraded. However, as is
be balanced at their corresponding reference values using the evident from Fig. 4, the error is bounded within a smaller band
schemes discussed later in Section IV. The output current of the ( or ) in the region when switching the voltage levels 0
inverter ( , Fig. 2) is controlled using the hysteresis scheme and , while due to the control actions of this scheme,
(Fig. 3) to follow a sinusoidal reference having peak-to-peak the error is bounded within a larger band ( or ) in the re-
values of A. Corresponding to Fig. 3, the hysteresis band gion when it is required to output one of the two extreme voltage
sizes are taken to be A and levels . This results in variable tracking performance
A. These values are taken for simplicity by following the con- in a single cycle of the current waveform itself.
siderations presented in Section III-C. Fig. 4 shows the simu- It should be noted that for the simulation studies, the built-in
lation results under this case, where it is evident that this con- IGBT and diode models in the PSCAD/EMTDC software’s
trol results in a poor quality voltage waveform. To justify this master library with the parameters listed in Table II have been
observation, let us first focus on the current error trajectory in used. These parameters are inherent with the device in the soft-
Fig. 4(a). As suggested earlier, between the points and , ware’s library. PSCAD is a graphical front-end to EMTDC for
the error variation outputs the voltages 0 and and as creating models and analyzing results. For building a simulation
the error moves away from , it touches the upper boundary of model of the FCMLI as in Fig. 2, such power semiconductor de-
(Fig. 3) at . By definition, at , the voltage level vice models are accordingly combined with the built-in models
is switched. It causes the current error to reverse its direction, of wires, capacitors, voltage sources, resistances, inductances,
which then reaches the lower boundary of at . Again, by etc. As PSCAD/EMTDC allows the user to develop his own
SHUKLA et al.: IMPROVED MULTILEVEL HYSTERESIS CURRENT REGULATION 521
model in FORTRAN [20], such models have been developed rent error trajectory analysis can be performed in Fig. 6 to justify
for the proposed control schemes implementation. Thus, all the the better waveforms using the control scheme of Fig. 5. A com-
power circuits are implemented using built-in devices models, parison of Fig. 6 with Fig. 4 shows that in the new scheme the
while the proposed control tasks have been achieved using user switching always occurs between adjacent levels and no level
defined FORTRAN programs. is skipped. It should be noted that, in Fig. 4, the controller acts
To overcome the drawbacks of the five-level control of Fig. 3, as desired when switching between , 0 and and
in this paper, a multi-offset-band hysteresis control is proposed. degrades when higher voltage levels are needed to be
The band placement and functioning of the proposed scheme for switched. This indicates that fixed voltage level switching as in
a five-level inverter is shown in Fig. 5. In this scheme, the cur- [6] works fine for the three-level inverter and needs modifica-
rent error is required to be bounded mainly between the bands tion (as in Fig. 5) for higher-level inverters.
and , which are displaced by a small offset . Further, Observing Figs. 3–6, a limitation when using the multi-offset-
two additional offsets of the same width are placed out of band scheme is that this switching process introduces a positive
and to provide a reliable and robust control of the in- or negative dc offset error into the average output current, de-
verter. In general, a total number of offsets are required pending on the polarity of the active output voltage. However,
for an -level inverter in both the positive and negative current this error can be corrected by adding a compensation factor of
error regions. It differs from the method of Fig. 3 in the decision half the hysteresis band offset magnitude to the phase current
logic of the output voltage levels at the crossing points of the [4], [6].
current error and the corresponding boundaries of the hysteresis
bands and also in the total number of bands required. In the B. Time-Based Multilevel Hysteresis Controller
proposed approach, the switched voltage at the band crossing An alternative technique was proposed in [4] to use only one
points of the current error is not fixed but depends on the pre- hysteresis band to detect an out of bounds current error. Digital
vious voltage level, i.e., just before the crossing point. If the cur- logic is used to select the “correct” voltage level in response.
rent error crosses the positive boundary of a band with positive Upon detecting the current error exceeding the upper (or lower)
slope, next lower (than the previous) voltage level is switched hysteresis limit, the inverter output is switched down (or up) one
(e.g., at in Fig. 5). Similarly, if the error crosses the negative voltage level so as to return the error back to zero, as before. But
boundary of a band with negative slope, next higher (than the if the new inverter switched state is inadequate to reverse the
previous) voltage level is switched (e.g., at in Fig. 5). The error back to zero, output is switched further down (or up) until
advantage of the proposed method of Fig. 5 over that of Fig. 3 the current error direction reverses. A possible current error tra-
is evident in the manner that by using the proposed logic, output jectory and inverter switched output for a five-level inverter are
voltage quality is improved and the current follows its reference shown in Fig. 7. Referring to Fig. 7, the objective of this method
with minimum change in voltage levels needed. It should also is to force the current error in a manner so that it remains within
be noted that in this scheme, the number of offset bands is de- band . It is evident that the inverter output is switched one level
cided by the number of steps needed to switch the voltage from up or down as the current error touches the boundary of . If
one extreme ( or ) to another extreme ( this changed output is insufficient to force the error back towards
or , respectively) as the error travels from positive (neg- zero (as at ), next higher or lower voltage level is switched
ative) to negative (positive) region. This can be further under- at the next crossing point of the error and the band limit (as at
stood by following the error trajectory from to in Fig. 5. ). From this figure, it is obvious that the technique does not
Another simulation study is performed using the control of create the steady-state tracking error of the multi-offset-band
Fig. 5 with the same inverter parameters as considered earlier approach (of Figs. 3 and 5). To improve the performance and
and hysteresis band sizes are taken equivalently as robustness of this technique, a current error slope detection algo-
A and A. Fig. 6 shows the results. Similar cur- rithm was used in [8] to switch the voltage levels. An outer band
522 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 23, NO. 2, MARCH 2008
Fig. 9. (a) Current error variation across the hysteresis bands (A). (b) Inverter
output voltage (V). (c) Time-difference between consecutive switching (ms).
points shown in the figure (i.e., , etc.,), voltage state transi-
tion takes place as mentioned earlier. A total number of
bands required for an -level inverter in this scheme can be jus-
this scheme, e.g., between and in Fig. 9(a). In Fig. 9(c), the
tified by following the current error trajectory in Fig. 8 and the
time differences between two consecutive switching are plotted.
discussions presented in the earlier presented schemes. It is ev-
This value is checked each time before a next voltage level is
ident from Fig. 8 that this proposed scheme does not need to
applied to have a time-based control. It should be noted that the
measure the derivative of the current error and therefore, does
tuning of along with and should be properly done to
not suffer from noise amplification problem as in [7], [8]. It is
have a good harmonic spectrum of the controlled current and
also clear that it can efficiently work under varying load condi-
voltage, while also taking into consideration the maximum al-
tions as well. The design considerations of and are the
lowable switching frequency (discussed in the next subsection).
same as presented latter in Section III-C.
The time-based control applied in the controlling the current for
To get further insight into the developed hysteresis control
this scheme can also be applied to the other scheme discussed
scheme of Fig. 8 and exemplify its working, simulation studies
earlier corresponding to Figs. 3 and 5.
are performed on a five-level inverter with the current refer-
ence and inverter and load parameters being the same as con- C. Hysteresis Band Size Considerations
sidered in the previous subsection with hysteresis band sizes of
A and A. The value and (delay in the To achieve accurate steady-state reference tracking, precise
time-based control) is taken to be 300 s. This value of is pur- offset tuning of the separation of the hysteresis bands is required
posely taken to be almost equal to the minimum time interval [8]. The size of the hysteresis bands ( in Figs. 3 and 5
between two consecutive switching decisions under the given and in Figs. 7 and 8) is largely determined by the maximum
system conditions to have a better viewing of the controller per- permitted level of current distortion. Generally, under this con-
formance. Note that, for this set of results, the switching fre- dition, a good performance with this scheme can be achieved
quency has been considerably reduced by taking corresponding by simply having the offset band sizes ( , Figs. 3, 5, 7, and
larger hysteresis band sizes, to clearly illustrate the switching 8) equal to the half of the size of the main bands. However, for
process within a fundamental cycle. The simulated waveforms systems requiring more accuracy, the size of should be as
are shown in Fig. 9. The current error variation across the hys- small as possible, as the controlled current and output voltage
teresis bands can be followed from the discussions presented waveform may degrade for a larger , depending on the load
earlier corresponding to Fig. 8. It is evident that at , the error type. The minimum possible size of the bands is mainly deter-
touches the upper boundary of and voltage level 0 is switched mined by the maximum allowable switching frequency of the
at the output of inverter to force the error in the opposite direc- power devices. For example, let us consider the control scheme
tion. However, at , when the error crosses the lower boundary of Fig. 5 and assume that the current error is contained within the
of , the next higher voltage level is not switched as the time main bands using the logic described earlier. At any
interval between the instants and is less than s. point of time, (2) can be rearranged by neglecting the voltage
Therefore, the error crosses at and is forced back in oppo- across the load resistance as
site direction at , i.e., at from the lower boundary of ,
where voltage level is switched. In this way, the cur-
rent is controlled to follow its reference by using the four bands
for a five-level inverter and a five-level output voltage wave-
form is obtained [Fig. 9(b)] for a sinusoidal reference current.
It is also evident that the time-based control is also operative in (4)
524 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 23, NO. 2, MARCH 2008
(5)
Fig. 13. Control block diagram for one phase leg of a five-level FCMLI.
Fig. 16. Experimental results showing the three flying-capacitor voltages and current error without the time-based capacitor voltage control [14 V/div, 0.15 A/div
(vertical axis), 5.0 ms/div (horizontal axis)].
Fig. 17. Experimental results showing the three flying-capacitor voltages and current error with the time-based capacitor voltage control and t = 1:6 ms [14
V/div, 0.15 A/div (vertical axis), 5.0 ms/div (horizontal axis)].
which improves the controlled current quality as compared with studies, i.e., 72 V and mH, respectively.
Fig. 10(a). This is because in the present case, the “loading ef- The presence of back emf would serve to create more variation
fect” of the capacitors on inverter performance is much less and in the switching frequency but without affecting the nature
the load current is mainly influenced by only the compo- of the current error trajectory. Therefore, for simplicity, back
nent of load. emf voltage source has not been used. The current reference,
It should be noted that the controller of Fig. 13 works on hysteresis band sizes and flying capacitors values are also same
the principle of utilizing the power devices with almost same as considered in the simulation studies presented in Section IV.
switching frequency under all loading conditions. Therefore, Figs. 16–19 show the experimental results of performances
switching frequency limitation is not an issue as can be set at of the flying-capacitor voltage control (Fig. 13) and hysteresis
any suitable value by taking into consideration the flying-capac- current control (Fig. 8). Figs. 16 shows the three flying-capac-
itor values, current rating of inverter and hysteresis band size. It itor voltages and corresponding current error when the capac-
is also to be noted that for sufficiently large capacitances and/or itor voltages are controlled using the method corresponding to
small size of hysteresis bands, the control of Fig. 13 may work the simulation results shown in Figs. 10 and 12(a). As expected
suitably enough without needing the time-based control, i.e., the from the conclusions made above, although the capacitor volt-
time counter block. ages are kept regulated around their corresponding reference
values, they suffer from large fluctuations. This is further justi-
V. EXPERIMENTAL RESULTS fied in Fig. 16(a), where between instants and varies
A prototype of a single-phase five-level IGBT-based FCMLI by a large amount due to large time gap between these instants.
has been built in the laboratory. The experimental investiga- Similar observation can be carried out in Fig. 16(b) between
tions are carried out to validate the proposed flying-capacitor and and in Fig. 16(c) between and . The corresponding
voltage balancing scheme (Fig. 13), while the inverter load output voltage of the inverter is shown in Fig. 19(a), which
current is regulated using the hysteresis control of Fig. 8. The is clearly of poor quality due to large capacitor voltage vari-
flying-capacitor voltages and the load current ations. As the capacitor voltages directly add (or subtract)-up
are sensed using Hall effect voltage and current sensors. with dc-link voltage to generate the intermediate output voltage
These voltage and current signals are acquired by a PC (P-1V, levels, it is evident in Fig. 19(a) that between and , the
2.4 GHz) through analog-to-digital converter (ADC) channels output voltage correspondingly degrades due to large capac-
of a standard data acquisition card (NIDAQmx PCI-6259). itor voltage variation. The results shown in Figs. 17 and 19(b)
Based on these quantities, a program written in Borland C++ are with the control of Fig. 13 with ms. This value
is implemented for the control tasks of Figs. 8 and 13. The of is approximately taken to be four times of the minimum
corresponding switching decision signals are generated at the time duration between two consecutive switching in the case of
digital-output port of the DAQ and are passed to the IGBT Fig. 16(a), i.e., without time-based control (0.4 ms). Similarly,
driver circuits after introducing a lockout delay of 7.5 s the result of 18 and 19(c) are with ms. It is evident
using blanking circuits. The dc-link voltage and inverter load from these figures that the waveforms of capacitor voltages and
parameters are same as considered earlier in the simulation hence output voltage of the inverter improve by reducing . In
528 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 23, NO. 2, MARCH 2008
Fig. 18. Experimental results showing the three flying-capacitor voltages and current error with the time-based capacitor voltage control and t = 0:4 ms [14
V/div, 0.15 A/div (vertical axis), 5.0 ms/div (horizontal axis)].
Fig. 19. Experimental results showing inverter output voltage and current error trajectories (a) without time-based capacitor voltage control, (b) with time-based
control and t = 1:6 ms, and (c) with time-based control and t = 0:4 ms [18 V/div, 0.15 A/div (vertical axis), 2.5 ms/div (horizontal axis)].