2nd IEEE International Conference on Power and Energy (PECon 08), December 1-3, 2008, Johor Baharu, Malaysia

A Review on Controllers for PWM Inverters

S. M. Ayob, N. A. Azli and Z. Salam
Department of Energy Conversion
Faculty of Electrical Engineering, Universiti Teknologi Malaysia
81310 UTM Skudai, Johor Darul Takzim, MALAYSIA
Email: [email protected]

Abstract—This paper reviews the control methods that are inverter, an LC filter and a load. Under assumption that
used to regulate the output voltage of PWM inverters. the switching frequency is high enough (> 10KHz) over
Based on their structure, the controllers can be categorised the fundamental frequency (50Hz), the dynamics of the
under model-based controllers, non-model based and PWM inverter is usually ignored and by applying
mixed-mode based controllers. The fundamental concepts of
averaging technique, it can be represented as a DC
these controllers are described along with examples from
each category. Furthermore limitations in terms of practical voltage source, Va with a value of Vcont*KPWM ( KPWM =
implementation are also discussed. Vdc / Vtri). The dynamics of the system is now mainly
determined by the filter. The equivalent series resistor
KeywordsüReview study; PWM inverter; Controller (ESR) for inductor and capacitor are denoted as rL and rC
and Z is the unknown load representation. Choosing the
I. INTRODUCTION inductor current, iL and ouput voltage, Vout as the system
An inverter is the most crucial component of many DC variables, the state space model of the equivalent circuit
to AC conversion equipment such as the uninterruptable can be formed as in Figure 2.
power supply (UPS), induction motor drive, induction
heaters, etc. The main feature of a well-designed inverter
is its ability to provide a clean and stable AC voltage
regardless of the type of load that is connected to it. It
must also has the ability to recover from transients caused
by external disturbances as quickly as possible. However,
with the proliferations of power converters connected as
loads, the inverter is forced to provide non-linear currents.
These highly distorted currents may cause deterioration in
the quality of its output voltage. In addition, the high Figure 1. Typical single-phase inverter circuit
voltage and current switching of the inverter generates
harmonics which can be harmful to sensitive equipments.
Therefore, the challenge of designing an inverter is to
maintain high quality sinusoidal output voltage under any
load conditions. The key to achieve this end is the
appropriate design of its controller.
Over the past few decades, many control techniques
have been proposed and applied to inverter regulation
systems. Although they shared the same control
objectives, they may differ markedly in terms of concept, Figure 2. State-space inverter model block diagram
design and approach. Some are conceptually simple in
principle but may be difficult during implementation. On For PWM inverter, the complexity of the control design
the other hand, other controllers may require complex is caused by the system itself. The system is a nonlinear
understanding of its theory, but its implementation can be system in nature and there are nonlinearities that being
relatively simple. These are the factors that must be ignored during the modeling stage. While in real-time,
considered before selecting a specific control technique these nonlinearities oftenly exist. These nonlinearities can
for a particular application. be introduced by non-idealistic switching behavior,
computation control delay and dead-time. Since these
elements are ignored, the designed controller performance
II. INVERTER MODELLING may be affected.
Another problem of inverter regulation is the diversity
of load type connected to the inverter. As in Figure 3, it is
Like all power electronics converters, a PWM inverter apparently observed that the load current, io is fed back
is nonlinear by nature due to the existence of the solid- into the model. The current can be in various forms
state switches. However, its linear model representation depending on the load type connected to the inverter.
can be derived by applying the well-known averaging While the load is diversed, the derivation of generalised
method. Figure 1 shows a typical single-phase inverter inverter model to cover all load types is difficult and
system. The system consists of a full-bridge PWM impossible. Besides the typical resistive load, it is now

1-4244-2405-4/08/$20.00 ©2008 IEEE 589

 2nd IEEE International Conference on Power and Energy (PECon 08), December 1-3, 2008, Johor Baharu, Malaysia

common to find inverters connected to nonlinear loads dynamics. The gain margin or also referred to as the
such as rectifiers and triacs. system’s bandwidth should be wide enough to provide the
Thus, the controller designed for PWM regulation fastest dynamic response and robust operation [2].
purposes must have the following features; capable to However, for a power converter which is associated with
handle nonlinearity of the system and eliminates the load switching, the bandwidth is limited to lower than one
current effect. For this paper, the controllers can be tenth of the switching frequency. This limitation may
categorised under model-based, non-model based and cause difficulty in the design of a robust PI controller [3].
mix-mode model based controllers. They are unique and To improve the robustness of a PI controller, usually
totally different over each other. Several prominent multi-loop control structure consisting of voltage loop and
controllers from each category as illustarted in Figure 3 current loop is employed. In addition, unlike a single-loop
are reviewed and their advantages and limitations are control structure (usually output voltage as the feedback
discussed. variable), a multi-loop structure is capable of stabilising
the system when subjected to light load. In this control
structure, options can be made either to choose inductor
current or capacitor current as the feedback current loop.
If the inductor current is chosen, various decoupling
methods (feed-forward loop) can be applied, which
eliminates the load effect and hence enhance the controller
robustness.

Figure 4. Multi-loop with feedforward control scheme

Figure 3. Tree diagram showing the classification of controllers With the advance in microprocessor and Digital Signal
Processor (DSP), digital control is increasingly used in
various applications of power converters. It offers easier
implementation, more flexibility in modifying codes for
III. MODEL-BASED CONTROLLERS other applications, less sensitive to noise and environment
variation. Digital based linear PI controller can be
designed using either direct digital design or indirect
The controllers that are categorised under this type of
digital design. In direct digital design, the inverter system
control are typically requires a mathematical model of the
plant is transformed into z-domain and the controller
PWM inverter for their control parameters design. In
design is carried out directly in the z-domain. Whereas
addition, their performance and stability analysis can be
for indirect digital design, the controller is designed in
assessed by conventional control theories.
continuous time mode and then discretised using several
dicretised methods [4]. A study conducted in [4] has
A. Linear PI Controller shown that both methods revealed a small difference in
performance when the system is subjected to a small
Among the prominent model-based controller that has sampling period. In the study, the sampling frequency
been used to regulate an inverter system is the Linear used is 40kHz.
Proportional-Integral (PI) controller. The PI controller
application for controlling inverters can be found in [1, 2].
It provides a simple control theory and easy B. Deadbeat Controller
implementation. However, it is known to produce
Deadbeat controller is indeed the most attractive digital
excellent performance under limited operating range;
based control solution so far. By placing all close-loop
excellent for load variation within a close neighborhood of
poles (n), at z = 0, the controller is able to track back the
the operating point. When subjected to large load
reference at most at n sampling time. Its first application
variations and changes in operating point, its performance
on an inverter system with microprocessor based control
could not be longer guaranteed.
was proposed by Gokhale and Kawamura. In the work,
To a certain extent, the performance of the PI controller deadbeat control is derived based on the discrete model of
depends on the selection of the proportional and integral an inverter. Instead of offering a fast response
constants. Various tuning strategies can be found. By far, performance, the deadbeat controller however suffers
the most popular tuning method is the frequency response from high sensitivity to changes in system parameters.
analysis. It provides a graphical insight to the system
In practical, parameter values especially the inductance
stability and performance while tuning the PI controller.
and capacitance of an LC filter will change over a period
Using this method, the controller can be tuned using Bode
of time due to aging or environmental effect. Moreover,
plot and Nyquist chart to its specified phase, gain margin
during production, the values of these components are
and stability. In most power converters applications, the
normally slightly different from design. Therefore, even
phase margin is chosen to be around 45o to 60o. This is to
though the controller can stabilise the inverter system, its
ensure that there is less oscillation during transient
performance however degrades.

590
 2nd IEEE International Conference on Power and Energy (PECon 08), December 1-3, 2008, Johor Baharu, Malaysia

Since then, various techniques devoted in enhancing the inaccuracy and uncertainty. Most of them are nonlinear in
deadbeat control performance have been reported. In [5], nature and its design approach is free from linear control
a multi-loop with feed-forward decoupling method is theory. Therefore they are relatively immune to
proposed. The employment of a multi-loop scheme with disturbances, robust and are capable of handling
feed-forward loop has simplified the deadbeat derivation nonlinearities in the system compared to model-based
and enhanced the controller robustness. However it controllers.
requires more than two sensors. To eliminate the effect of
parameter variations, authors in [6], proposed an on-line
parameter estimation scheme. The proposed scheme A. Neural Network
obviously overcomes the problem but suggests a complex One of the non-model based controllers that is capable
solution. of learning and comprehending a process adaptively is the
Neural Network (NN). It is a nonlinear mapping of input
and output; the map is a form of multi-layer connection
C. Sliding Mode Controller which contains neurons. The interconnection of the
Contrary to Linear PI and Deadbeat controllers, Sliding neurons between the layers depends on the weight of the
Mode controller (SMC) does not fully depend on the neurons which adaptively adjusts during training. The
model. Thus it exhibits more robust performance, highly application of this controller in an inverter control system
immune to any external disturbances and parameter can be found in [8].
changes.
In SMC, the system’s states variables are forced to
reach the switching plane and slide on the plane towards
its equilibrium point regardless of its initial states position.
This feature makes the controller very immune to external
disturbances. Moreover, its stability analysis is inherent
and can be assessed using Lypunov stability theorem.
Discrete Sliding mode control (DSMC) is a digital form of
SMC. Even though it has a similar structure, its control
law differs.
Although SMC is conceptually simple, the practical
realisation of the controller is difficult [1,3]. Some issues
such as guideline to obtain an optimised switching Figure 6. Simplest Neural Network
function are left unresolved [3]. If exist, they normally
involve very complex mathematics. Thus, most of Depending on the applications, the training of the
previous works related to SMC especially in power controller can be made either via online or offline mode.
converter control use heuristic approach to solve the In online training, the weights are adjusted adaptively.
switching function. However this training requires a fast processor since it
involves a lot of computations. Back propagation
In addition, practical SMC suffers from chattering algorithm is usually used for online training. To reduce
phenomenon. Chattering phenomenon is due to the computation burden, offline training can be an
imperfection of switching and computation delay. It alternative. Since the training is done offline using sample
usually appears as high frequency ripple at steady-state data obtained from experiment or simulation, the
and it is very severe due to the fact that it can excite non- calculation involved can be considered small, thus suitable
modeled high-frequency elements. Hence degrades the to be implemented in low-cost processors. However,
overall performance. Although there are several contrary to online training, the weight of the neurons using
approaches to alleviate the problem; the simplest is by off-line training is fixed. Moreover, a large number of
introducing a boundary layer around the switching plane, sample data must be obtained to guarantee excellent
it is however reported that a steady-state error appears [7]. performance.

B. Fuzzy Logic Controller

Fuzzy Logic Controllers (FLCs) are linguistic based
controllers. FLCs have been widely applied on inverters
through various control schemes. They can be used in
either direct control scheme [9], feed-forward
compensator along with other linear controllers [10] and
as a gain scheduler for a PI controller [11]. FLCs are
operated based on rules obtained through operators’
experience and knowledge of the systems’ processes.
Figure 5. Sliding mode control with boundary layer
Compared to SMC and NN controllers, FLCs require
simpler mathematics and exhibit excellent immunity to
IV. NON-MODEL BASED CONTROLLERS
disturbances. An FLC also offers a high degree of
freedom in tuning its control parameters. This makes it
Controllers under this category are obviously of great easily modified and can be tuned to approximate different
value for problems where the system is difficult to model kinds of linear or nonlinear controllers. Despite these
mathematically, due to system complexity, model advantages, FLC suffers from lack of standard design

591
 2nd IEEE International Conference on Power and Energy (PECon 08), December 1-3, 2008, Johor Baharu, Malaysia

procedure; the design is basically heuristic in nature. The signal disturbances and a conventional FLC for large
difficulty with heuristic approach has to do with the large signal disturbances. There exists a “break point” in which
number of rules and tuning parameters that has to be the transition between PI and FLC occurs. This method is
considered to obtain acceptable performance. known as the PI- FLC.
Consequently, long design time cycle remains the The main advantage of this method is that, unlike FLC,
challenge for FLC. Moreover its stability analysis is still no heuristic design approach is required for the design of
an unresolved issue [3]. the PI-FLC for small signal disturbances. For this stage,
the controller is designed totally based on the linear PI
control theory. This will result in a guaranteed stability
V. MIX-MODE BASE CONTROLLERS and excellent performance. While when subjected to
It is inherent that both model-based and non-model large-load disturbance, its performance could not be
based controllers have their respective limitations. Non- guaranteed; but it is improved by tuning the rules of the
model-based controllers are obviously more robust and PI-FLC [14].
immune to disturbance. The only limitation of this
controller type is that their design are done heuristically
hence optimised performance are difficult to achieve.
While on model-based side, their performance and
stability are inherent and can be assessed. However, the
performance and stability can only be guaranteed for a
limited range of operation. Mix-mode based controllers
are those that have been combined to overcome each
others limitations.

A. Sliding Mode Fuzzy Controller

FLCs as described earlier, suffer from lack of design
guideline. To solve the problem, several authors have Vertical scale: output voltage 25V/div, output current 5A/div, Time
proposed an approach of designing an FLC based on the scale 2ms/div
sliding mode theory (SM-FC). In this technique, the SMC Figure 8. Performance of discrete linear PI controller when subjected
is first designed with guaranteed performance and stability to small load disturbance
properties. Then the SMC is “fuzzified”, producing an
FLC-type performance while maintaining the stability of
an SMC. An example of an SM-FLC applied to a PWM
inverter can be found in [12]. It is also reported that by
using this scheme, the chattering problem is eliminated. A
further study revealed that, the reason for the elimination
can be attributed to the fact that the fuzzification forms a
boundary layer around the SMC’s switching plane [13].
However, although the chattering is eliminated, the
formation of this layer forces the system to produce a
steady-state error which is impossible to be removed. In
addition this technique still suffers from the difficulty to
find an optimised switching function.
Vertical scale: output voltage 25V/div, output current 5A/div, Time
scale 2ms/div
Figure 9. Performance of PI-FLC when subjected to small load
disturbance

Figure 7. Fuzzification of switching plane in SMFC

B. PI-Fuzzy Logic Controller

A combination of PI control and the FLC can be

another attractive approach in designing an improved
fuzzy controller. Contrary to the SM-FLC, this method Vertical scale: output voltage 25V/div, output current 5A/div, Time
requires simpler mathematical formulation. Moreover scale 2ms/div
control engineers are more familiar with PI design Figure 10. Performance of discrete linear PI controller when subjected
techniques compared to sliding mode. The idea is to to sudden large load disturbance
design an FLC using the PI approach to cater for small

592
 2nd IEEE International Conference on Power and Energy (PECon 08), December 1-3, 2008, Johor Baharu, Malaysia

[10] B.-R. Lin and C. Hua, “Uninterruptible Power Supply with Fuzzy
Logic Approach”, International Conference on Industrial
Electronics, Control and Instrumentation, Vol. 2, pp1123-1128,
November 1993.
[11] E. D. Bolat, K. Erkan and S. Postalcioglu, “Using Current Mode
Fuzzy Gain Scheduling of PI Controller for UPS Inverter”, The
2005 International Conference on Computer As A Tool, pp1505-
1508, November 2005.
[12] E.-C. Chang, T.-J. Liang, J.-F. Chen and R. -L. Lin, “A Sliding-
Mode Controller Based on Fuzzy Logic for PWM Inverters”, The
2004 IEEE Asia-Pacific Conference on Circuits and Systems,
pp965-968, December 2004.
[13] F. Song and S. M. Smith, “A Comparison of Sliding Mode Fuzzy
Controller and Fuzzy Sliding Mode Controller”, The 19th
International Conference of the North American Fuzzy
Vertical scale: output voltage 25V/div, output current 5A/div, Time
Information Processing Society, pp480-484, July 2000.
scale 2ms/div
[14] A. G. Perry, Y.-F. Liu and P. C. Sen, “A New Design Method for
Figure 11. Performance of Sugeno type PI-FC when subjected to large PI-like Fuzzy Logic Controllers for DC-to-DC Converters”, The
load disturbance 35th Annual IEEE Power Electronics Specialists Conference,
Vol.5, pp3751-3757, June 2004.
VI. CONCLUSION
This paper provides the general concept of several
controllers that are popularly applied to PWM inverter
regulation. Based on the study, the controllers can be
categorised as model-based, non-model and mix-mode
based. From the discussion, it is agreed that each
controller from the model-based and non-model based
category, has its respective limitations. In order to
alleviate the problem, a combination of controllers (mix-
mode based) can be an alternative and it is seen as the
recent trend in PWM inverter controls. In this type of
controller, while retaining the good side of a particular
controller, another controller is used to cater the first
controller limitations. Hence an improved overall
controller is obtained.

REFERENCES

[1] P. Sanchis, A.Ursaea, E.Gubia and L. Marroyo, “Boost DC-AC

Inverter: A New Control Strategy”, IEEE Transaction on Power
Electronics, Vol. 20, No. 2, pp343-353, March 2005.
[2] N. Mohan, “Electric Drive: An Integrative Approach”, John
Wiley, 2001.
[3] T.G. Habetler and R.G. Harley, “Power Electronic Converter and
System Control”, Proceeding of the IEEE, Vol. 89, No.6, pp913-
925, June 2001.
[4] Y. Duan and H. Jin, “Digital Controller Design for Switchmode
Power Converters”, The 14th Annual Applied Power Electronics
Conference and Exposition, Vol. 2, pp. 967-973, March 1999.
[5] S-L.Jung, H-S. Huang, M-Y. Chang and Y-Y. Tzou, “DSP-Based
Multiple Loop Strategy for Single-Phase Inverters Used in AC
Power Sources”, The 28th Annual IEEE Power Electronics
Specialist Conference, Vol.1, pp706-712, June 1997.
[6] P. Mattavelli, “An Improved Deadbeat Control for UPS Using
Disturbance Observers”, IEEE Transactions on Industrial
Electronics, Vol. 52, No. 1, pp206-212, February 2005.
[7] V.S.C. Raviraj and P.C. Sen, “Comparative Study of Proportional-
Integral, Sliding Mode and Fuzzy Logic Controllers for Power
converters”, IEEE Transaction on Industry Applications, Vol. 33,
No. 2, pp 518-524, Mac/April 1997.
[8] X. Sun, M. H. L.Chow, F.H.F. Leung, D. X. Y. Wang and Y. -S.
Lee, “Analogue Implementation of A Neural Network Controller
for UPS Inverter Applications”, IEEE Transaction on Power
Electronics, Vol.17, No.3, pp305-313, May 2002.
[9] N.A.Azli and W.S.Ning, “Application of Fuzzy Logic in an
Optimal PWM Based Control Scheme for a Multilevel Inverter”,
The 5th International Conference on Power Electronics and Drive
Systems, Vol.2, pp 1280-1285, November 2003.

593