Optimization of Higher Order Filters

for Grid Connected High Frequency Power

Converters

Vinod John Parikshith B. Channegowda Nilanjan Mukherjee

Anusyutha Krishnan Amith Karanth
Department of Electrical Engineering
IISc Bangalore

December 2009
Abstract

Use of inverters to interconnect to the grid has become widespread for a number of power
quality, regenerative motor drive and distributed generation applications. The design of
passive filters for such converters has not received as much attention as that for control
and operation of the semiconductor switches. Traditionally these filters have been first
order inductive chokes. Design of filters used in grid-connected inverter applications in-
volves a large number of constraints. The filter requirements are driven by tight tolerances
of standards such as IEEE 519-1992–IEEE Recommended Practices and Requirements
for Harmonic Control in Electrical Power Systems and IEEE 1547.2-2008–IEEE Appli-
cation Guide for IEEE Std. 1547-2003, IEEE Standard for Interconnecting Distributed
Resources with Electric Power Systems. Higher order LCL filters are essential to achieve
these regulatory standard requirements at compact size and weight. This objective of this
report is to evaluate design procedures for such higher order LCL filters and to provide
insight into methodologies for improved filter design.

The initial configuration of the third order LCL filter is decided by the frequency
response of the filter. The design equations are developed in per-unit basis so results
can be generalized for different voltage and power levels. The frequency response is
decided by IEEE specifications for high frequency current ripple at the point of common
coupling. The appropriate values of L and C are then designed and constructed. Power
loss in individual filter components is modelled by analytical equations and an iterative
process is used to arrive at the most efficient design. Different combinations of magnetic
materials (ferrite, amorphous, powder) and winding types (round wire, foil) are designed
and tested to determine the most efficient design. The harmonic spectrum, power loss and
temperature rise in individual filter components is predicted analytically and verified by
actual tests using a 3-phase 10 kVA grid connected PWM converter setup.

Experimental results of filtering characteristics show a good match with analysis in

the frequency range of interconnected inverter applications. The design process is stream-
lined for the above specified core and winding types. The output harmonic current spec-
trum is sampled and it is established that the harmonics are within the IEEE recommended
limits. The analytical equations predicting the power loss and temperature rise are ver-
ified by experimental results. Based on the findings, new LCL filter combinations are
formulated by varying the net L pu to achieve the highest efficiency while still meeting
the recommended IEEE specifications.Thus a design procedure which can enable an en-
gineer to design the most efficient and compact filter that can also meet the recommended
guidelines of harmonic filtering for grid-connected converter applications is established.
 Compared to the inductive filter use of an LCL filter can lead to resonance. It has been
shown that one can design effective resonance suppression by both active and passive
methods. Methods to evaluate the ability for suppression of resonance has been confirmed
with experimental results. Experimental results carried out in using closed loop grid
interactive control confirm the design procedure for resonance damping in LCL filters.
The ability to design smaller filters that meet high performance requirements leads to a
cost effective filtering solution for grid connected high frequency power converters.
 5

Acknowledgement

This project was graciously funded by the National Mission on Power Electronics of the
Government of India, NaMPET (www.nampet.org). Thanks to Prof. Krishna Vasudevan,
IIT Madras, Prof. Anil Kulkarni, IIT Bombay, Prof. V. Ramanarayanan, IISc Bangalore,
Dr. Praveen Vijayraghavan, IEC Bangalore, Mr. S.C. Dey, Department of Information
Technology and Mr. Suresh Babu VS, C-DAC Thiruvananthapuram for overseeing and
periodically reviewing the project.
 7

Preface

Background

High energy prices and environmental concerns are driving the search for alternative re-
newable energy sources. Simultaneously, rising cost and complexity in existing electricity
distribution systems, and the inability of current systems to serve remote areas reliably has
led to search for alternate distribution methods. One viable solution is use of renewable
energy sources directly at point of load, which is termed as Distributed Generation(DG).
Most renewable sources of energy, like wind, solar etc., are interfaced to the existing elec-
tric grid by a power converter. This eliminates the transmission and distribution losses
and improves reliability of the grid power supply. But use of power converters will also
introduce undesirable harmonics that can affect nearby loads at the point of common cou-
pling to the grid. Hence, all such converters have a filter to minimize these harmonics to
acceptable levels.

The present work is on design of such filters for high power (10’s to 100’s of kVA)
pulse width modulated high power voltage source converters for grid-connected converter
applications. The conventional method to interface these converters to grid is through a
simple first order low-pass filter, which is bulky, inefficient and cannot meet regulatory
requirements such as IEEE 512-1992 and IEEE 1547-2008. The goal of this report is
the design of efficient, compact higher order filters to attenuate the switching harmonics
at the point of interconnection to the grid to meet the requirement of DG standards of
interconnection.

Organization of the report

The filter design analysis is logically arranged into six self contained chapters featuring
filter component parameter selection, analysis and inductor design process. The concern
of grid interactive operation due to resonance in the filter is then addressed. The last
chapter reports the experimental results that were used to validate the design assump-
tions. Filter design is normally an iterative procedure. The last chapter demonstrates that
tradition design rules can lead to a bulky and inefficient filter.

Transfer Function Analysis This chapter takes a top-level system level approach to fil-
ter design. The factors which affect the initial selection of the LCL filter param-
eters include IEEE recommended limits on high frequency current ripple, closed
loop operation requirements of a grid connected filter, EMI filtering, power system
fault ride-through requirements etc.. The filter parameters obtained at the end of
the chapter satisfy all the hard constraints of a high power converter interfaced to
the grid. The subsequent chapters deal with the actual construction and efficiency
optimization of the filter.

Filter Component Construction This chapter is focused on the design and construction
of the individual components of the LCL filter. The design techniques to accurately
build an inductor of required inductance are discussed in detail. The familiar area
product approach for inductor design is modified and incorporated into new meth-
ods which are more accurate and material specific. The principles of construction
for three different magnetic materials -Ferrite, Amorphous and Powder is discussed.
Finally, capacitors and resistors suited for for high power filter applications are in-
troduced.

Simulation Using FEA Tools Modelling and simulation of the filter inductors using Fi-
nite Element tools are described here.The tools used are FEMM and MagNet. In-
ductance values, plots of flux lines and flux density in the inductor core are shown
in the results. Results of Finite Element Analysis of the filter inductors and mea-
surement of air gap and core flux density in the actual inductors are available in this
section.

Power Loss and Heating Effects In this chapter the filter parameters obtained from the
previous chapter are examined from the point of view of efficiency and temperature
rise. This efficiency and temperature optimization become highly significant as
recent trends suggest that more switching power converters at higher power ratings
are connected to the grid. This chapter derives the equations that describe the power
loss in inductors at high frequency operation. Finally, principles of heat transfer are
used to estimate the surface temperature of inductors. The entire design procedure
can be validated from the expected temperature rise of the inductor.
Grid Interactive Operation and Active Damping This chapter discuss the methods that
can be used for actively damping resonances in LCL filters. A state space based
control design is shown to have better performance. The performance of the active
damping controller is experimentally verified by operating the power converter in
grid interactive manner.

Results and Optimized LCL Filter Design This chapter reports the experimental re-
sults that are used to verify the filter design model. All aspects of the design process
are tested, with special emphasis placed on harmonic response and efficiency of the
constructed filter components. The designed and actual measurements are com-
pared to verify the validity of the design assumptions. The notable contribution of
this report is formulation of new LCL filter combinations by varying the net L pu to
achieve the highest efficiency while still meeting the recommended IEEE specifi-
cations.
Contents

Contents i

1 Transfer Function Analysis 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Starting assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Per unit system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.1 Base parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.2 DC voltage per unit . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.3 Voltage and current ripple per unit . . . . . . . . . . . . . . . . . 5
1.4 L filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 LC filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.1 Bandwidth consideration . . . . . . . . . . . . . . . . . . . . . . 11
1.5.2 Design procedure for an LC filter . . . . . . . . . . . . . . . . . 13
1.6 LCL filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6.1 Design procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.7 Resonance damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7.1 Quality factor and power dissipation . . . . . . . . . . . . . . . . 20
1.7.2 Design procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2 Filter Component Construction 33

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Area product approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.1 Design steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

i
ii Contents

2.3 Graphical iterative approach . . . . . . . . . . . . . . . . . . . . . . . . 37

2.3.1 Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.2 Disadvantages . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4 Fringing flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.1 Simple fringing model . . . . . . . . . . . . . . . . . . . . . . . 39
2.4.2 Bossche and Valchev model . . . . . . . . . . . . . . . . . . . . 40
2.4.3 Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 Fringing edge calculation using FEA . . . . . . . . . . . . . . . . . . . . 43
2.5.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.6 Inductor design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.6.1 Ferrites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.6.2 Amorphous material . . . . . . . . . . . . . . . . . . . . . . . . 48
2.6.3 Powder material . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.7 Capacitor selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.8 Power resistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.9 Design examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.9.1 Amorphous core example . . . . . . . . . . . . . . . . . . . . . 52
2.9.2 Powder core example . . . . . . . . . . . . . . . . . . . . . . . . 53
2.10 Measurements on inductors . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.10.1 Measurement of permeability of core material . . . . . . . . . . . 55
2.10.2 Measurement of airgap flux density . . . . . . . . . . . . . . . . 56
2.10.3 Measurement of core flux density . . . . . . . . . . . . . . . . . 56
2.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.12 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3 Simulation Using FEA Tools 61

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Ferrite core inductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3 Amorphous core inductor . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.4 Powdered core inductor . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.5 Comparison of FEMM and MagNet . . . . . . . . . . . . . . . . . . . . 73
3.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Contents iii

4 Power Loss and Heating Effects 75

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Core loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.1 Eddy current loss . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.2.2 Excess loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.2.3 Hysteresis loss . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.4 Total loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3 Copper loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4 Foil conductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.4.2 One dimensional H field . . . . . . . . . . . . . . . . . . . . . . 81
4.4.3 Power dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.4.4 AC resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5 Round conductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.5.1 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.5.2 Skin effect loss . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.5.3 Proximity effect loss . . . . . . . . . . . . . . . . . . . . . . . . 93
4.6 Thermal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.6.1 Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.6.2 Natural convection . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.6.3 Temperature estimation . . . . . . . . . . . . . . . . . . . . . . . 100
4.6.4 Design example . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5 Grid Interactive Operation and Active Damping 107

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.2 Active front end converter . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.3 Problem of LCL resonance . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.4 Active damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.4.1 Active damping based on traditional approach . . . . . . . . . . . 111
5.4.2 Active damping by means of state space method . . . . . . . . . 111
5.4.3 Filter modelling in state space . . . . . . . . . . . . . . . . . . . 111
iv Contents

5.4.4 Pole placement of the system . . . . . . . . . . . . . . . . . . . . 113

5.4.5 Per unitization . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.4.6 Physical realization of active damping . . . . . . . . . . . . . . . 114
5.4.7 Active damping loop realization . . . . . . . . . . . . . . . . . . 116
5.5 Control of the inverter in grid-interactive mode . . . . . . . . . . . . . . 117
5.5.1 Model for control design . . . . . . . . . . . . . . . . . . . . . . 118
5.5.2 Overview of control loop consisting of three states of system . . . 119
5.5.3 Current control strategy . . . . . . . . . . . . . . . . . . . . . . 120
5.5.4 Analysis of controller performance . . . . . . . . . . . . . . . . 121
5.5.5 Inclusion of innermost state-space based damping loop . . . . . . 123
5.5.6 Control in grid-parallel mode with LCL filter . . . . . . . . . . . 124
5.5.7 Sensorless operation . . . . . . . . . . . . . . . . . . . . . . . . 124
5.6 Experimental set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.7 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.7.1 Implementation stages . . . . . . . . . . . . . . . . . . . . . . . 128
5.7.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

6 Experimental Results and Optimized LCL Filter Design 137

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.2 LCL filter parameter ratings . . . . . . . . . . . . . . . . . . . . . . . . 137
6.3 Frequency response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.4 Harmonic analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.5 Power loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.6 Temperature rise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.7 Minimum power loss design . . . . . . . . . . . . . . . . . . . . . . . . 155
6.8 Loss profile for ferrite core inductors . . . . . . . . . . . . . . . . . . . . 157
6.9 Loss profile for amorphous core inductors . . . . . . . . . . . . . . . . . 162
6.10 Loss profile for powder core inductors . . . . . . . . . . . . . . . . . . . 166
6.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

7 Conclusions 171

A Transfer Function Tests 176

Contents v

A.1 Ferrite core inductor results . . . . . . . . . . . . . . . . . . . . . . . . . 176

A.2 Amorphous core inductor results . . . . . . . . . . . . . . . . . . . . . . 182
A.3 Powder iron core inductor results . . . . . . . . . . . . . . . . . . . . . . 187

B Temperature Rise Tests 193

B.1 Ferrite core inductor results . . . . . . . . . . . . . . . . . . . . . . . . . 194
B.2 Amorphous core inductor results . . . . . . . . . . . . . . . . . . . . . . 199
B.3 Powder iron core inductor results . . . . . . . . . . . . . . . . . . . . . . 199

C Test Set-up 201

C.1 Schematic and filter layout . . . . . . . . . . . . . . . . . . . . . . . . . 201
C.2 Pictures of filter components and test set-up . . . . . . . . . . . . . . . . 205

D Simulation of Inductors Using MagNet 213

D.1 Ferrite core inductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
D.2 Amorphous core inductors . . . . . . . . . . . . . . . . . . . . . . . . . 218
D.3 Powdered core inductor . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

E Electromagnetic Equations 225

E.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
E.2 Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
E.3 Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
E.4 Rectangular coordinate system . . . . . . . . . . . . . . . . . . . . . . . 229
E.5 Cylindrical coordinate system . . . . . . . . . . . . . . . . . . . . . . . 231
E.6 Bessel functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
E.7 Retarded potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
E.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Chapter 1

Transfer Function Analysis

1.1 Introduction

This chapter focuses on the design procedures to implement L, LC and LCL filters for grid
connected inverter applications. The design calculations are based on per-unit values, so
the results obtained are generalized for any application for power levels ranging from 10’s
of kW upto 100’s of kW. The procedure for passive damping unwanted resonance in the
third order filters is also discussed in detail.

1.2 Starting assumptions

There are certain simplifying assumptions that are made to analyse the frequency charac-
teristics of the grid connected low pass filter. The assumptions are made to keep the initial
design analysis simple. These constraints are subsequently relaxed later in the course of
the discussion for a more accurate analysis.

• All filter elements are considered ideal, i.e no winding resistance, inter-turn/inter-
winding capacitance in case of inductor, and no equivalent series resistance, para-
sitic inductance in case of capacitor.

• Grid is considered as an ideal voltage source, i.e zero impedance, and supplying
constant voltage/current at fundamental (50Hz) frequency. This is a reasonable as-
sumption since any impedance at the grid can be lumped with the output impedance
of the filter. We can see later that this assumption is also justifiable based on per

1
2 Transfer Function Analysis

unitized impedance calculations of grid interconnection.

• The filter design procedure is appropriate for grid connected PWM voltage-source
inverters or matrix converters. Current source inverters are not considered.

• The design procedure assumes only grid connected mode of operation. Stand alone
converter applications are only briefly discussed.

1.3 Per unit system

The per unit system is used to represent the voltage, current, kVA, frequency and other
electrical parameters. All the design equations are expressed in per unit basis of the
converter rating. The advantage of the per unit method is that we can generalize the
design procedure for a wide range of power levels and for different applications. This
also makes the design procedure compatible with the grid power system ratings where
most impedances are usually expressed in per unit basis.

1.3.1 Base parameters

The per unit system followed here is based on the volt ampere rating of the power con-
verter. The line to neutral output voltage VLN is the base voltage and the 3 phase KVA
rating KVA3φ is the base volt ampere. The fundamental frequency of 50Hz is the base
frequency.

Vbase = VLN (1.1)

KVA3φ (base) = 3φ Power rating (1.2)

Iactual KVA3φ (pu)
I pu = = (1.3)
Ibase Vpu
Vbase
Zbase = (1.4)
Ibase
Vpu
Z pu = (1.5)
I pu
Zbase
Lbase = (1.6)
2π fbase
Lactual
L pu = = Z pu (1.7)
Lbase
1.3 Per unit system 3

1
Cbase = (1.8)
Zbase × 2π fbase
Cactual 1
C pu = = (1.9)
Cbase Z pu
The per unit system can be easily extended to other parameters like dc bus voltage and
switching frequency.

Vdc
Vdc(pu) = (1.10)
Vbase

fsw
fsw(pu) = (1.11)
fbase

1.3.2 DC voltage per unit

A single leg of a three-phase inverter can be represented as shown in Fig 1.1. The in-
verter voltage and current are represented as vi , ii and the grid voltage and current are
represented as vg and ig .

L
L
Vdc
0
+ + i
i
ig
ig
Vi Vg
_ vi vg
_
_

Figure 1.1: Equivalent circuit of one leg of voltage source inverter

The DC bus voltage can be expressed in per unit of grid voltage depending on the
configuration of the inverter and whether reactive power compensation is required. We
can define the dc bus voltage Vdc in terms of pole voltage Vi . The pole voltage in turn
can be defined based on the base voltage Vg = Vbase . The assumptions are that the grid
voltage can have a maximum variation of ±10%, and the pole voltage will be reduced
by 5% because of dead band switching requirement. We are also taking into account the
voltage drop due to a series filter, which usually will not exceed 10% of the inverter pole
voltage.
4 Transfer Function Analysis

Topology 1

L
Vdc
0
+ +
ig
Vi Vg
_
_
_

Topology 2

L
Vdc
+ +
ig
Vi Vg
_
_
_

Figure 1.2: Different configurations of single phase inverters

For the single phase topologies shown in Fig 1.2 (Topology 1)

Vdc √
= Vi = Vbase × 2 × 1.1 × 1.05 × 1.1 (1.12)
2

Similarly for Topology 2,

√
Vdc = Vi = Vbase × 2 × 1.1 × 1.05 × 1.1 (1.13)

For the three phase topology shown in fig 1.3, the pole voltage Vi depends on the
modulation method. In case of sine-triangle modulation, the peak pole voltage amplitude
(in case of linear modulation) is

Vdc
Vi = ma (1.14)
2
1.3 Per unit system 5

where ma is the modulation index. So the maximum DC bus voltage will be when ma = 1.

Vdc √
= Vi = Vbase × 2 × 1.1 × 1.05 × 1.1 (1.15)
2

In case of space vector modulation used in 3φ 3 wire power converter, the maximum
magnitude of the voltage space vector in α -β coordinates is,
√
◦ 3
Vre f = Vdc cos 30 = Vdc (1.16)
2

In three phase basis, the pole voltage will be

2 Vdc √
Vre f = √ = Vi = Vbase × 2 × 1.1 × 1.05 × 1.1 (1.17)
3 3

Topology 3

+
i
g L
Vi
Vi +
Vdc 0 Vi + Vg
+ Vg _
Vg _
_

Figure 1.3: Three phase inverter configuration

1.3.3 Voltage and current ripple per unit

The per unit system is most useful to represent voltage and current ripple at switching
frequency in terms of the base parameters. From Fig 1.4 if δ i p−p is the peak to peak
ripple current in the inductor, then

δ i p−p
δ irms = √ (1.18)
2 3

δ irms
δ i pu = (1.19)
Ibase
6 Transfer Function Analysis

v
L

iL
δi
t

Figure 1.4: Voltage and current across filter inductor

The pole voltage is a combination of sinusoidal voltage at fundamental frequency

along with harmonic voltages at various higher frequencies, including switching fre-
quency. The rms of the harmonic voltages varies with the modulation method. We are
assuming that the modulation method used is sine-triangle modulation. This assump-
tion can be justified as this modulation method gives high harmonic voltages compared
to any other advanced modulation methods, and if the designed filter can pass filtering
criterion with this modulation method, it will satisfy the filtering requirements for any
other advanced modulation method. But this assumption will also give a bigger filter than
required if the modulation method is more sophisticated.

The grid voltage is assumed vary from -20% to +10%.

0.8Vg ≤ Vg ≤ 1.1Vg (1.20)

where Vg = Vbase =1 pu. Assuming a 10% drop in the series filter inductor, the variation
in pole voltage will be

Vi(pu) = 1.1Vg (1.21)

Since the control algorithm has to supply constant rated current at the inverter terminals
even with this variation in grid voltage, the corresponding range of modulation index ma
can be calculated.
√
Vi(pu) 2
ma = (1.22)
Vdc(pu)
2

Vi(pu) and Vdc(pu) can be substituted from Eqns (1.21) and (1.15). The range of modulation
index to supply rated current for grid voltage variation is given in Table 1.1
1.3 Per unit system 7

As described earlier, the inverter pole voltage is a combination of fundamental voltage

and harmonic voltages at various higher frequencies. By assuming that the most dominant
harmonic voltage is at switching frequency, we can write the rms value of the inverter pole
voltage as

2 2
Vi(rms) = Vi(50) +Vi(2 f sw) (1.23)

where Vi(50) is the rms value of the fundamental voltage at 50Hz and Vi( f sw) is the rms
value of the switching frequency harmonic voltage. From Fig 1.5, it is clear that the total
rms value of the inverter pole voltage is Vi(rms) = Vdc /2. The rms value of the fundamental
depends on the modulation index.
 2
1 V
2
Vi(50) = √ dc ma (1.24)
2 2

Hence we can find the switching frequency ripple voltage in terms of the rms pole voltage
and modulation index.

Vi(2 f sw) = Vi(rms)

2 2
−Vi(50) (1.25)

2
Vdc 1V2
Vi(2 f sw) = − dc m2a (1.26)
4 2 4
r
Vdc m2
Vi( f sw) = 1− a (1.27)
2 2
For the range of ma from Table 1.1, the range of Vi( f sw) is

Vdc Vdc Vdc

0.739 ≤ 0.791 ≤ 0.872 (1.28)
2 2 2
8 Transfer Function Analysis

Vref

Vtri

V pole

VDC
2
0

−VDC
2

Figure 1.5: Sine triangle pulse width modulation

Min Nominal Max

Vg(pu) 0.8 1 1.1
Vi(pu) = 1.1Vg(pu) 0.88 1.1 1.21
√
Vi(pu) 2
ma = 0.693 0.866 0.952
Vdc(pu)
2
r
m2
1− a 0.793 0.791 0.872
2
Table 1.1: Effect of amplitude modulation index on switching frequency ripple

1.4 L filter
sL

i i
i g

v v
i g
INVERTER GRID

Figure 1.6: L filter inserted between active front end and grid

The design of an L filter is based on the current ripple at switching frequency that is
1.4 L filter 9

present in the PWM output.

+ +
Vdc Vdc
2 2
+ _ + _

L ig L ig
vg vg
Vdc Vdc
2 _ 2 _

(a) (b)

Figure 1.7: Voltage across L in Ton and To f f

If we consider one single switching cycle of the inverter, from Fig 1.7(a), during Ton

δ i p−p Vdc
L = − vg (1.29)
Ton 2

and during To f f , Fig 1.7(b)

δ i p−p Vdc
L =− − vg (1.30)
To f f 2

where vg = Vm sin ω t, and Ton + To f f = Tsw . Since the modulation method is sine triangle
modulation, the duty ratio D is

Vm sin ω t vg
D = 0.5 + = 0.5 + (1.31)
Vdc Vdc

So from equations 1.29, 1.30 and 1.31, we get

δ i p−p
Ton = L (1.32)
Vdc (1 − D)

δ i p−p
To f f = L (1.33)
Vdc D
Adding the above two equations we get

Vdc × D × (1 − D)
Lactual = (1.34)
fsw × δ i p−p

Vdc × D × (1 − D)
Lactual = √ (1.35)
fsw × 2 3 × δ irms
Lactual Vdc × D × (1 − D) Ibase
L pu = = √ × × 2π fbase (1.36)
Lbase fsw × 2 3 × δ irms Vbase
10 Transfer Function Analysis

Vdc(pu) × D × (1 − D) × π
L pu = √ (1.37)
fsw(pu) × 3 × δ irms(pu)
Here D is the duty cycle of the switch such that the average voltage at fundamental fre-
quency is sinusoidal. The worst case current ripple occurs at 50% duty cycle, so the above
equation can be simplified.

π Vdc(pu)
L pu = √ × (1.38)
3 4 × fsw(pu) × δ i pu

This is the maximum current ripple for any switching cycle which will happen at every
zero crossing of fundamental voltage. But IEEE standards specify the current ripple limits
for multiple cycles of fundamental current, not for one switching cycle.

If we assume that the inverter is source of sinusoidal voltages at different harmonic

frequencies, we can find the current sourced by the switching frequency harmonic. At
switching frequencies, the grid is a short circuit. Hence the switching frequency current
will be
vi(sw)
δ irms = (1.39)
2π × fsw(pu)×L

This is the current ripple relevant for THD calculations.

1.5 LC filter
sL SWITCHGEAR sL
ig ig
ii ii

vi v vi
1 g 1 LOAD
INV sC GRID INV sC
i i
c c

(a) (b)

Figure 1.8: (a) LC filter inserted between active front end and grid; (b) LC filter inserted
between active front end and stand-alone load

The design of LC filter is more complicated compared to L filter since the placement
of the resonant frequency becomes an important factor which affects the closed loop
response. The allowable current ripple is once again the criteria for designing L. The
capacitor C is constrained by two factors.
1.5 LC filter 11

• The resonant frequency of the filter elements

• The bandwidth of the closed loop system

1.5.1 Bandwidth consideration

The capacitance of the LC filter is decided by the resonant frequency. The design decision
on selecting the resonant frequency depends on the bandwidth of the closed loop system.
This dependency is established keeping in mind that active control methods (which are
bandwidth dependent) can be used to implement loss-less resonant damping in higher or-
der filters. Since the bandwidth of the closed loop system is decided by the filter elements
and the control algorithm, it cannot be used straightaway in the design process. Here, we
estimate the maximum possible system bandwidth and use it in our design procedure. The
maximum possible bandwidth is certainly not achieved in practice, but this assumption is
reasonable for a first pass iteration. Figure 1.9 shows the closed loop system.

The output voltage of a grid connected power converter cannot be controlled since it
is decided by the grid conditions. The filter input current ii is usually sensed and given
back as feedback to close the control loop. But the grid current ig is the control variable
which is controlled by varying the inverter pole voltage. Hence, the transfer function
which decides the closed loop performance of the filter is the transfer function between
output current and input voltage of the filter for zero grid voltage.

Assuming the controller acts directly at the modulator without prior dynamics, there
are two delays in the closed loop system which limit the bandwidth.

1. The Inverter response delay. When the voltage command to the inverter is changed,
in the worst case, it takes Tsw /2 time for the voltage output of the inverter to change,
where Tsw is the switching time period.

2. Current sampling and computational delay. If the current sampling is sampled once
per pwm cycle this delay would be Tsw . Here we are assuming that the current is
sampled twice every cycle, on the rising half and falling half of the PWM switching
signal, so the delay in sensing is Tsw /2.

3. Phase delay due to the filter. This can be simplified to a constant phase delay of 90o
based the assumptions listed in section 1.2.

So the total system excluding the filter is essentially modelled as a pure delay e−std ,
where is td = Tsw . The resonant frequency is placed such that the closed loop system
12 Transfer Function Analysis

including the LC filter gives a phase margin of atleast 45◦ . The LC filter transfer function
which affects the closed loop system bandwidth is:

ig (s)  1
= (1.40)
vi (s) vg =0 sL


Since the LC filter transfer function has a constant phase of -90◦ for all frequencies, the
bandwidth of the system(excluding the filter) is limited at the frequency where its phase
is 45◦ . So the frequency at which the phase margin of the total system (LC filter + delay)
is 45◦ can be calculated.
π
45◦
ωbw = 180◦ (1.41)
td

Now the resonant frequency can be placed with reference to the bandwidth. If the
resonant frequency is within the bandwidth of the closed loop system, active damping
methods can be used to attenuate the filter resonance peaks. If the resonant frequency
is outside the bandwidth of the system, passive damping methods (i.e resistors) must be
used. Active damping means lower power loss at full load. Passive damping is essential in
grid connected applications, in case the inverter is switched off while still being connected
to the grid.

1
C pu = 2
(1.42)
fres(pu) × L pu

The transfer function of grid current ig to inverter voltage vi is same for L and LC
filters when parasitic grid impedances are neglected (Fig 1.8). Therefore, the size of
inductor does not change from L to LC filter. But Eq (1.40) will change if LC filter is
connected to a stand-alone load. Consider an LC filter connected between an inverter and
external load of R=1 pu resistance. Then the transimpedence transfer function will be

ig 1
= 2 (1.43)
vi s LCR + sL + R

Additionally, a grid connected LC filter can behave as an LCL filter because of the par-
asitic impedances of the grid. But this arrangement is not reliable since the parasitic
impedance of the grid is not under the control of the converter designer.
1.5 LC filter 13

Inverter response delay = Tsw

2
Line Impedence
+
V Filter
dc −
v (s)
i GRID

i (s)
g

d(s)

i (s)
g
Controller Sensing

Current samping delay = Tsamp

Figure 1.9: Closed loop system bandwidth assuming no delay in controller

vtri

vref
(a)
0

+Vdc
2
(b) _V
dc
2

(c)
Tsw

(d)
Tsw
2

Figure 1.10: (a) PWM sine triangle modulation; (b) Output pole voltage with respect to
ground; (c) Inverter output voltage update rate; (d) Current sampling rate

1.5.2 Design procedure for an LC filter

1. Selection of L pu based on switching cycle ripple current consideration.

Vdc(pu) × D × (1 − D) × π
L pu = √ (1.44)
fsw(pu) × 3 × δ irms(pu)
14 Transfer Function Analysis

2. Selection of C pu based on overall bandwidth and resonant frequency.

1
Cactual = (1.45)
ωres
2 ×L
actual

Cactual 1
C pu = = 2 Zbase × ωbase (1.46)
Cbase ωres × Lactual
1
C pu = Zbase × ωbase (1.47)
ωres
2 ×L ×L
pu base

1
C pu = 2
(1.48)
fres(pu) × L pu

Further aspects of LC filter design is not considered as this report. Here the focus is
on grid connected power converters, which requires the improved level of filtering offered
by LCL filters.

1.6 LCL filter

An LCL filter is preferred to an L filter in high power and/or low switching frequency
applications. This is because for the same (or lower) net inductance (i.e L1 + L2 ) we can
get better attenuation (60dB/decade) at switching frequency.

Figure 1.11: LCL filter inserted between active front end and grid

The design procedure for LCL filter cannot be treated as a progression from an LC fil-
ter, since there are more possible resonances (infact three) between the filter elements.
The three possible resonant frequencies based on the open or close positions of the
1.6 LCL filter 15

switches S1 and S2 in Fig 1.11 are:

1
ωL1C = √ (1.49)
L1C
1
ωL2C = √ (1.50)
L2C
1
ωL pC = p (1.51)
L pC
The actual poles of the filter can be obtained from the characteristic equation of the sys-
tem. The three poles of the system include a pair of complex conjugate poles due to ωL pC
and a pole at the origin.

L1 × L2
Lp = (1.52)
L1 + L2

The procedure for design of LCL filter as given in the current literature is as follows
[3]–[7].

• L1 is designed based on the current ripple.

• L2 is assumed to be a fraction of L1 , maybe greater than or lesser than L1 . This is

decided by the current ripple in inductor L2 .

• C is designed on the basis of the reactive power supplied by the capacitor at funda-
mental frequency.

This procedure has a few limitations.

• The upfront rule of thumb based selection of L and C makes it difficult to optimize
the overall filter design procedure.

• It is not possible to design the LCL filter on a per unit basis, where the per unit is
referenced from the VA of the power converter system.

• The resonant frequencies and their effect on system bandwidth is ignored in this
method.

• Even though the aim of filter design is to attenuate the switching frequency harmon-
ics, the basis of capacitor design is the reactive power of fundamental frequency.

• There is no simple way to compare L and LCL filters for the same application.
16 Transfer Function Analysis

1.6.1 Design procedure

In the proposed method the inverter plus filter is treated as a “black box,” so the only
input variables for the filter design are the KVA rating of the inverter and the switching
frequency output current ripple ig ( jωsw ). Let L be the total inductance of the filter,

L = L1 + L2 (1.53)

Let L1 and L2 be related as

L1 = aL L2 (1.54)

Next, the total system bandwidth (including filter) is estimated such that there is accept-
able phase margin in the system. The LCL filter transfer function which affects the closed
loop system bandwidth in grid connected mode of operation is

ig (s)  1
= 3 (1.55)
vi (s) vg =0 s L1 L2C + s(L1 + L2 )


The LCL filter transfer function has a constant phase of -90◦ below ωres and +90◦ above
ωres as can be seen from Eq (1.55). So the bandwidth of the closed loop system will be
same as that of the LC filter below ωres . The resonant frequency of interest is ωL pC , since
this is the resonant frequency of Eq (1.55).

1
ωres
2
= (1.56)
L pC

where

L1 × L2
Lp = (1.57)
L1 + L2

Substituting for L p in terms of L = L1 + L2

ig (s) 1
= (1.58)
vi (s) sL(1 + s2 L pC)

Converting all quantities to their per-unit equivalents, the resonant frequency is

1
ωres(pu)
2
= aL (1.59)
C pu × L pu
(aL + 1)2
1.6 LCL filter 17

150

100

Magnitude (dB)
50

−50

−100

−150
90

45
Phase (deg)

−45

−90
3 4 5
10 10 10
Frequency (rad/sec)

Figure 1.12: Frequency response of ig /vi for LCL filter

The capacitance in an LCL filter depends on the resonant frequency ωres and the ratio
in which we distribute the total inductance L1 + L2 . Assuming we have fixed ωres , the
ratio of L1 and L2 for minimum capacitance is given by

δ C pu
=0 (1.60)
δ aL

which simplifies to aL = 1. So, given a fixed output harmonics attenuation, the smallest
capacitance value of LCL filter is obtained when L1 = L2 . Since we know the dependence
of output current of filter ig on the inverter terminal voltage vi , we can again find the value
of aL which will give the minimum current ripple at the point of common coupling at any
frequency.

vi
ig = 3
(1.61)
s L1 L2C + s(L1 + L2 )

vi
ig = a L
(1.62)
s3 L2C + sL
(1 + aL )2
δ ig
=0 (1.63)
δ aL
18 Transfer Function Analysis

δ ig δ vi (1 + aL )2
 
= =0 (1.64)
δ aL δ aL s3 aL L2C + sL(1 + aL )2
[s3 aL L2C + sL(1 + aL )2 ][2vi (1 + aL )] = vi (1 + aL )2 [s3 L2C + sL2(1 + aL )] (1.65)

2aL = 1 + aL (1.66)

aL = 1 (1.67)

Eq (1.56) becomes

4
ωres
2
= (1.68)
L puC pu

To find L pu and C pu , Eq (1.55) is evaluated (in per unit) at switching frequency fsw .

 ig ( jωsw ) 
 
1
 vi ( jωsw )  = |− jω 3 L1 L2C + jωsw (L1 + L2 )| (1.69)
 
sw

ig ( jωsw ) is the switching ripple current at the point of common coupling to the grid at
switching frequency. This is guided by the recommendations of IEEE 519-1992 or IEEE
1547.2-2008 standard[1]–[2]. For example, the IEEE 519 recommended maximum cur-
rent distortion for a ISC /IL < 20 for current harmonics ≥ 35th is 0.3%. ISC refers to short
circuit current and IL is the nominal load current. This requirement of 0.3% refers to a
“weak” grid. The percentage of ripple current can be higher for a “stiff” grid. Since most
inverters can switch at higher frequencies exceeding 2 kHz using current IGBT technol-
ogy, the standard refers to harmonics ≥ 35th . vi ( jωsw ) is the inverter pole voltage ripple
at switching frequency, which is Vdc /2.

Eq (1.69) is solved by converting all parameters to per-unit and substituting Eq (1.68)

in Eq (1.69).

1 1
L pu = (1.70)
ωsw(pu)
 
2

i  
 g(pu)  
ωsw(pu)   1 − 2

 vi(pu)   ωres(pu) 


Then C pu will be calculated from Eq (1.68). This is different from most current literature
which focus on reactive current capability to decide value of C. The disadvantage of
the previous approach was that real and/or reactive power supplied to load/grid must be
known beforehand and must always have a fixed minimum. The proposed design allows
for deciding the value of C without setting the reactive power requirement. The reactive
power drawn by the filter can be subsequently verified to meet the system requirement.
1.7 Resonance damping 19

If the reactive current drawn is seen to be excessive, then the L and C can be traded off
by keeping a fixed resonant frequency or looking at the possibility of a higher power
converter switching frequency.

The biggest advantage of this method is that it simultaneously satisfies four constraints
of filter design for typical PWM inverter designs with fsw in the 10kHz range. The voltage
drop across the inductor at fundamental frequency will be less than 0.1 pu. And the
reactive current sourced by the capacitor at fundamental frequency will also be less than
0.1 pu. The switching frequency attenuation requirement and bandwidth requirements are
already met as discussed above.

Vbase KVAbase Ibase Zbase

V KVA A Ω
254 10 13.1 19.35

Table 1.2: Base values used for calculations

L LC LCL
L pu = L1(pu) + L2(pu) 2.352 2.352 0.105
C pu - 1.063e-3 0.095
δ ig(pu) 0.003 0.003 0.003
fres(pu) - 20 20
fsw(pu) 200 200 200

Table 1.3: Comparison of pu values of filter for same grid current ripple

1.7 Resonance damping

As described earlier, the resonant frequency of the LCL filter which affects the closed
loop response of the system for grid connected operation is ωL pC . The resonance effect
can cause instability in the output, especially if some harmonic voltage/current is near the
resonant frequency. The simplest type of damping is to put a resistance in series with the
inductors. But this also increases the losses in the filter at fundamental frequency. Thus
the important issue in implementing damping is to balance the trade-off between effective
damping- which is measured by the Q-factor of the circuit and power dissipation due to
damping elements. In this section we focus only on one passive damping method. The
damping circuit is shown in Fig 1.13.
20 Transfer Function Analysis

L1 L2
vc
ii ig
i1
Cd
C1 id
vi
vg

Rd

Figure 1.13: Passive damping scheme

1.7.1 Quality factor and power dissipation

The aim of damping is to reduce the Q-factor at the resonant frequency without affecting
the frequency response at other frequencies. Simultaneously, the total power dissipation
in the damping circuit is also an important parameter. Since the frequency response must
not change, the resonant frequency is kept same and the total filter component values are
unchanged ie. L = L1 + L2 and C = C1 +Cd is unchanged from an LCL filter without any
damping. Let

L1 = aL L2 (1.71)

Cd = aCC1 (1.72)
r
L
Rd = a R (1.73)
C
The transfer function which affects closed loop response is

ig (s)  1 + sCd Rd
= 4 (1.74)
vi (s) vg =0 s L1 L2C1Cd Rd + s L1 L2 (C1 +Cd ) + s2Cd Rd (L1 + L2 ) + s(L1 + L2 )
 3

Substituting

L1 + L2 = L (1.75)

L1 L2
= Lp (1.76)
L1 + L2
C1 +Cd = C (1.77)
C1Cd
= Cs (1.78)
C1 +Cd
1.7 Resonance damping 21

we can simplify Eq. (1.74) as


ig (s)  1
=    (1.79)
vi (s) vg =0 1 + sCs Rd
sL 1 + s2 L pC
1 + sCd Rd

The additional passive elements increases the order of the transfer function and it is dif-
ficult to analytically estimate the resonant frequency of the fourth order system of Eq
(1.74). Infact the resonant frequency now becomes a function of Rd which is difficult to
derive analytically. But since the variation of resonant frequency with damping elements
is not significant, the resonant frequency is assumed independent of variations in damping
circuit.
20

10
Magnitude (dB)

−10

−20

−30

−40
0

−45
Phase (deg)

−90

−135

−180
3 4 5
10 10 10
Frequency (rad/sec)

Figure 1.14: Frequency response of vc /vi . Here aR = 0.3, aC = aL = 1

The other transfer function of significance is

sL2 + s2 L2Cd Rd

vc (s) 
= (1.80)
vi (s) vg =0 s4 L1 L2C1Cd Rd + s3 L1 L2 (C1 +Cd ) + s2Cd Rd (L1 + L2 ) + s(L1 + L2 )

The Q-factor of Eq (1.80) can be reliably determined since the frequency response of this
transfer function has a constant magnitude at low frequencies. Hence this equation is
used to analyze the effect of variation of damping parameters aC and aR on Q-factor of
22 Transfer Function Analysis

the LCL filter circuit. Substituting s = jω in Eq (1.80).

vc ( jω ) jω L2 − ω 2 L2Cd Rd
= 4
vi ( jω ) ω L1 L2C1Cd Rd − jω 3 L1 L2 (C1 +Cd ) − ω 2Cd Rd (L1 + L2 ) + jω (L1 + L2 )
(1.81)

Dividing numerator and denominator by ω (L1 + L2 ) and using the condition that L1 = L2

vc ( jω ) j0.5 − ω 0.5Cd Rd
= 3 (1.82)
vi ( jω ) ω L pC1Cd Rd − jω 2 L pC − ωCd Rd + j1

where

L1 L2
Lp = (1.83)
L1 + L2

and

C = C1 +Cd (1.84)

This can be further simplified by substituting ωr = ωL pC = 1/

p
L pC

vc ( jω ) −0.5ωCd Rd + j0.5
= (1.85)
vi ( jω ) ω2
ωCd Rd (ω 2 L pC1 − 1) + j(1 − 2 )
ωr

By substituting

C1
L pC1 = L p (C1 +Cd ) (1.86)
C1 +Cd

we get

vc ( jω ) 0.5 + j0.5ωCd Rd
= (1.87)
vi ( jω ) ω2 ω 2 C1
  
1− + jωCd Rd 1 − 2
ωr2 ωr C1 +Cd

or in terms of aC

vc ( jω ) 0.5 + j0.5ωCd Rd
= (1.88)
vi ( jω ) ω2 ω2 1
  
1− + jωCd Rd 1 − 2
ωr2 ωr 1 + aC

The frequency response of Eq (1.80) is given in Fig 1.14. To find the Q-factor of this
1.7 Resonance damping 23

circuit, derive
 
 vc 
lim   = 0.5 (1.89)
ω →0 vi

 
 
 
 vc   0.5 + j0.5ω C R 
r d d
= (1.90)

aC 
 
 jωrCd Rd 1 + a 
 vi 
ω =ωr 
C
Dividing Eq (1.90) by Eq (1.89) and substituting for Cd in terms of C
 
 aC 
 1 + jωrCRd

1 + aC 

Q(aC ) =  (1.91)

aC2

 jωrCRd
 

 2
(1 + aC ) 

Q(aC ) is plotted in Fig. 1.15. From the figure we can see that there is no improvement
in the Q of the frequency response if aC is increased beyond 2. Therefore, we are setting
aC = 1 as the best choice, since it is practically easy to configure two capacitors of same
value.
18

16

14

12
Q−factor

10

0
0 2 4 6 8 10
ac
Figure 1.15: Q-factor vs aC

The Q-factor is also affected by the choice of Rd .

r
L
Rd = × aR (1.92)
C
24 Transfer Function Analysis

A very small value of aR makes Rd effectively a short circuit and does not provide any
damping. Similarly, a very large value of aR makes Rd effectively an open circuit which
does not proved any damping. It is seen that selection of aR = 1 is equivalent to making
Rd equal to the characteristic impedance of the LCL circuit. This gives the lowest Q for
the damping circuit. Since this fact is difficult to prove analytically, we can prove that by
plotting the frequency response of the LCL filter with damping for different kVA ratings
as shown in Fig 1.16. The system rating values for different kVA are given in Table 1.4.
20

a =0.3
R
15
a =0.5
R
aR=1
10 aR=2
aR=3
5
a =5
R
Magnitude (dB)

−5

−10

−15

−20

−25

−30
3 4 5
10 10 10
Frequency (rad/sec)

Figure 1.16: Frequency response vc /vi for different aR . System rating is 1kVA.

The net power dissipation in the damping circuit is another important factor which
will affect the damping parameters. The power dissipated in the damping circuit can be
calculated for the fundamental and switching frequency. From the Fig 1.17, the power
loss in the damping circuit for the fundamental frequency is given by

Pd(50) = Real[Vc Id∗] (1.93)

where

Vc = Vg = 1pu (1.94)

sCd
Id = Vc (1.95)
1 + sCd Rd
1.7 Resonance damping 25

Unit 1kVA 10kVA 100kVA

Vb V 254 254 254
Ib A 1.312 13.12 131.2
Zb Ω 193.6 19.36 1.936
L mH 64 6.458 0.6458
C µF 1.569 15.69 156.9
L1 = L2 mH 32 3.229 0.3229
C1 = Cd µF 0.7846 7.846 78.46
q
L
C Ω 202.8 20.28 2.02

Table 1.4: Filter circuit and damping circuit designed values for different KVA grid con-
nected inverter rating.
L1 L2
Vc

Ii I1 Ig
Cd
C1 Id
Vi
Vg
Rd

Figure 1.17: LCL circuit with C1 ,Cd , Rd damping at fundamental frequency

jω50Cd (1 − jω50Cd Rd )
Id = Vc (1.96)
1 + ω502 C 2 R2
d d

Vc2 ω50
2 C2 R
d d
Pd(50) = (1.97)
1 + ω50
2 C 2 R2
d d

Similarly, damping circuit for switching frequency is given in Fig 1.18. From Eq.
(1.88), we get

0.5 + j0.5ωswCd Rd
Vc = Vi  (1.98)
ωsw
2  ωsw
2
 
1
1− + jωswCd Rd 1 − 2
ωr2 ωr 1 + aC

ωswCd (ωswCd Rd + 1 j)
Id = Vc (1.99)
1 + ωsw
2 C 2 R2
d d

Pd(sw) = Real[VcId∗ ] (1.100)

26 Transfer Function Analysis

L1 L2
Vc

Ii I1 Ig
Cd
C1 Id
Vi

Rd

Figure 1.18: LCL circuit with C1 ,Cd , Rd damping at switching frequency

Representing Vc and Id as complex fractions,

a + jb
Vc = Vi (1.101)
c + jd

ωswCd
Id = Vc (x + jy) (1.102)
1 + ωsw2 C 2 R2
d d

a + jb ∗ ωswCd
 
∗ a + jb
Vc Id = Vi Vi (x + jy)∗ (1.103)
c + jd c + jd 1 + ωsw
2 C 2 R2
d d

a2 + b2 ωswCd
Pd(sw) = Real[Vc Id∗ ] = Vi2 x (1.104)
c + d 1 + ωsw
2 2 2 C 2 R2
d d
From Fig 1.19, total power loss in damping branch is almost linearly proportional to
aC . For highest efficiency, aC should be as low as possible. Hence aC = 1 is a good
compromise between Q-factor, shown in Fig 1.15, and Power dissipation in the damping
circuit, as shown in Fig 1.16. For this selection of aC , the power loss in the filter for
damping is less than 0.2% and the damping factor is close to 2. This is a reasonable
design for inverters with power levels that are in the order of 10kW.

1.7.2 Design procedure

The LCL filter design procedure is already discussed in the previous section. The extra
elements of damping circuit can be derived from the above discussion.

aC = 1 (1.105)

C
C1 = Cd = (1.106)
2
1.7 Resonance damping 27

Damping branch power loss (pu)

0.016
Pd50
PdSW
0.014 Pdtotal
0.012

Power loss [pu] 0.01

0.008

0.006

0.004

0.002

0
0 2 4 6 8 10
ac
Figure 1.19: Power dissipation in per unit for damping circuit at fundamental and switch-
ing frequency. Here aR = 1.

Similarly,

aR = 1 (1.107)

which means
r
L
Rd = (1.108)
C

The comparison with and without damping is shown in Fig 1.20 and Fig 1.21.
28 Transfer Function Analysis

150
No damping
C −C −R damping
100 1 d d
Magnitude (dB)
50

−50

−100

−150
180

90
Phase (deg)

−90

−180
3 4 5 6
10 10 10 10
Frequency (rad/sec)

Figure 1.20: Frequency response of ig /vi . Here aR =1, aC =aL =1, system rating 10kVA

150

100 No damping
C −C −R damping
1 d d
Magnitude (dB)

50

−50

−100
0

−45
Phase (deg)

−90

−135

−180
2 3 4 5 6
10 10 10 10 10
Frequency (rad/sec)

Figure 1.21: Frequency response of vg /vi . Here aR =1, aC =aL =1, system rating 10kVA
1.8 Summary 29

1.8 Summary

The system level design principles for grid connected low pass filters have been thor-
oughly examined. The per unit method gives the flexibility to adapt the design equations
for any power level. The relative merits between L, LC and LCL filter combinations
are discussed. The parameters of the LCL filter are derived from bandwidth constraints
and IEEE standard recommendations and are seen to be suitable for practical inverter de-
signs. The problem of resonance damping is considered and a low loss passive damping
structure is introduced.

1.9 References

IEEE Standards

1. “IEEE Application Guide for IEEE Std 1547, IEEE Standard for Interconnecting
Distributed Resources with Electric Power Systems,” IEEE 1547.2-2008

2. “IEEE Recommended Practices and Requirements for Harmonic Control in Elec-

trical Power Systems,” IEEE 519-1992

Transfer function analysis

3. M.Liserre, F.Blaabjerg and A. Dell’Aquila, “Step-by-step design procedure for a

grid-connected three-phase PWM voltage source converter,” Int. J. Electronics,vol.
91, no. 8, pp. 445-460, Aug 2004.

4. Y.Lang, D.Xu, et al., “A novel design method of LCL type utility interface for
three-phase voltage source rectifier,” IEEE 36th Conference on Power Electronics
Specialists, 2005.

5. B.Bolsens, K. De Brabendere et al., “Model-based generation of low distortion cur-

rents in grid-coupled pwm-inverters using an LCL output filter,” Power Electronics
Specialists Conference, 2004. PESC 04. 2004 IEEE 35th Annual

6. M.Liserre, F.Blaabjerg, S.Hansen, “Design and control of an LCL-filter-based three-

phase active rectifier,” Conference Record of the 2001 IEEE Industry Applications
Conference, 36th IAS Annual Meeting, 2001, v.1, pp. 299-307.
30 Transfer Function Analysis

7. Dahono, P. A., Purwadi, A. e Qamaruzzaman, “An LC Filter Design Method for

Single-Phase PWM Inverters,” Proceedings of 1995 International Conference on
Power Electronics and Drive Systems. PEDS 95, v. 2, pp. 571-576.

8. Parikshith B. C., V. John, “Higher Order Output Filter Design for Grid Connected
Power Converters,” National Power Systems Conference 2008, IIT Bombay, Mum-
bai, 16th - 18th December, 2008.

Resonance damping

9. T.Wang, Z.Ye et al., “Output filter design for a grid-interconnected three-phase

inverter,” Power Electronics Specialist Conference, 2003. PESC ’03. 2003 IEEE
34th Annual, pp. 779-784
1.9 References 31

Table 1.5: Significance of transfer functions

vc (s)
Voltage harmonic attenuation in stand-alone mode
vi (s)
increasing L1 /L2 is beneficial
vg (s) = vc (s) under open circuit conditions


ig (s) 
Inverter THD in grid connected mode
vi (s) vg short


ii (s) 
Inductor harmonic spectrum for inductor design
vi (s) vg short

ig (s)
Current filtering in grid connected mode
ii (s)
decreasing L1 /L2 is beneficial


ig (s) 
Grid admittance seen from filter when converter not switching
vg (s) vi open
ideally 0 meaning reject all disturbances from grid


ig (s) 
Admittance seen from grid ignoring controller interaction
vg (s) vi short
Chapter 2

Filter Component Construction

2.1 Introduction

This chapter is focused on the design and construction of the individual components of the
LCL filter. The design techniques to accurately build an inductor of required inductance
are discussed in detail. The familiar area product approach for inductor design is modified
and incorporated into new methods which are more accurate and material specific. The
principles of construction for three different magnetic materials -Ferrite, Amorphous and
Powder is discussed. Finally, capacitors and resistors suited for for high power filter
applications are introduced.

2.2 Area product approach

The product of core cross-section area and window area in an inductor (area product) is
a measure of the energy handling capability of the inductor. The area product equation is
a good starting point for design since it relates the electrical design inputs with material
and geometric constraints.

The minimum cross section area of the inductor winding (aw m2 ) is limited by the
rms current flowing in the winding Irms A, which depends on the temperature rating of
the insulation and the conductivity of the wire. The temperature limit is expressed in
terms of the current carrying capacity of the conductor Jm A/m2 .

Irms
aw = (2.1)
Jm

33
34 Filter Component Construction

Similarly the minimum cross section area of the inductor core (Ae m2 ) is limited by the
peak flux density of the core material expressed as Bm T. If N turns each carrying a peak
current of I p A create a peak flux of φm Wb, then

LI p = N φm (2.2)

where

φm = B m A e (2.3)

Therefore we get

LI p = NBm Ae (2.4)

A third constraint which affects the design is the amount of space available in the window
area (AW m2 ) for the winding. In order to accommodate the winding in the available
window space,

Naw < AW (2.5)

Converting this inequality to a equation

Naw
AW = (2.6)
ku

where kw is the window utilization factor. kw varies between 0 and 1 depending on several
factors like type of conductors, number of bobbins, insulation class, winding skill etc.
From Eq (2.6) and (2.4),

LI p Irms = ku JBm Ae AW (2.7)

The area product is defined as

LI p Irms
A p = Ae AW = , (2.8)
ku Bm Jm

or

Energy Stored
Ap = . (2.9)
Material Constraints
2.2 Area product approach 35

For inductors carrying AC, the stored energy is expressed as [3]

Vrms Irms
Energy Stored = (2.10)
f kf

where f is the frequency of the current waveform and k f is the form factor (k f =4.44 for
sinusoidal waves).

2.2.1 Design steps

The traditional design steps for inductor design is given below [2]. Transfer function
analysis along with ripple current limits and actual power rating of the converter is used
to arrive at L, I p and Irms (as already explained in Chapter 1). The flux density limit
of the magnetic material Bm can be taken from vendor datasheet. The current density
limit Jm is based on reasonable assumptions of current density to prevent overheating.
Window utilization factor ku depends on type of winding, number of bobbins used, type of
insulation and the winding skill of the manufacturer. At the end of the design process, the
temperature rise calculation is used to evaluate the effectiveness of the above assumptions.
If the final operating temperature is significantly different from initial approximation,
these assumptions need to be modified.

1. Compute

L I p Irms
Ae AW = (2.11)
ku Bm Jm

2. Select a core from core tables with area product equal to or greater than Ae AW .

3. For the selected core, find Ae and AW .

4. Compute

L Ip
N= (2.12)
Bm Ae

Select nearest whole number of N ∗ .

5. Compute

Irms
aw = (2.13)
J

Select nearest (greater) number of wire gauge and a∗w from wire table.
36 Filter Component Construction

6. Compute the required air gap in the core

µo N ∗ I p
lg = (2.14)
Bm

7. Check the assumptions:

• Core reluctance << Air gap reluctance; This condition ensures that the fi-
nal inductance does not vary with the tolerance of magnetic properties of the
manufactured core.

l
ℜc << ℜg ; << lg (2.15)
µr

• No fringing:
p
lg << Ae (2.16)

8. Recalculate

Irms
Jm∗ = (2.17)
a∗w

9. Recalculate

N ∗ a∗w
kw∗ = (2.18)
Aw

10. Compute from the geometry of the core, mean length per turn and the length of the
winding. From wire tables, find the resistance of winding at the operating temper-
ature.

2.2.2 Limitations

1. The design procedure is simple and completes in a single iteration. But for a given
core type, there is only one value of air gap and number of turns. Actually there are
several combinations of these parameters which will give the same inductance but
very distinct efficiency and performance characteristics.

2. This method does not take into account the fringing of the magnetic field at the air
gap. The fringing effect reduces the reluctance at the air gap, which means a higher
2.3 Graphical iterative approach 37

flux density in the core. Eq. (2.16) is an approximation and does not ensure that
the absence of fringing even if the condition is met.

3. The lack of a good reluctance model means that the core can saturate even if the
area product condition is met.

It is clear that even though the area product approach is conceptually correct, certain
modifications are necessary to ensure accurate modeling of the inductance. Additionally
a reasonably accurate fringing model is required to prevent saturation of the core.

2.3 Graphical iterative approach

In the design process of an inductor, there are two parameters that must be accurately
preserved-L and Bm and two parameters that can be adjusted-N and lg . So L and Bm are
basically functions of 2 variables.

L = f (N, lg ) (2.19)

Bm = g(N, lg ) (2.20)

We can define the functions f (N, lg ) and g(N, lg ) as

N2
L= (2.21)
ℜt

N Ip
Bm = (2.22)
Ae ℜt
where ℜt is the total reluctance of the flux path. Both L and Bm are restricted within
certain limits and the possible set of (N, lg ) which give this inductance and flux density
is plotted on a graph of lg vs N. These points are then fit using a second or third order
polynomial to generate two curves, one for L and second for Bm . The intersection of both
curves will give the possible (lg, N) for which the core will not saturate as well as the
required inductance is achieved. Additionally there will be also be several solutions in
the neighbourhood which satisfy the inductance and peak flux density requirements.
38 Filter Component Construction

2.3.1 Advantages

1. The number of possible solutions is larger which means there is greater flexibility
in the actual construction of the inductor.

2. Since the effect of fringing at the air gap is included in terms of ℜt , the built induc-
tor will have the inductance very close to the initial calculation.

2.3.2 Disadvantages

1. The material properties, especially the permeability must be linear in the operating
range.

2. The permeability should be independent of magnetic excitation.

Curve fitting of (lg,N,L) and (lg,N,Bm)

130
L
Bm
120
R =10 ´ R
gap core
110
3.4 mH < L < 3.5 mH 0.38 T < B < 0.4 T
m
100
Number of Turns

90

80

70

60

50

40

30
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Airgap [m]

Figure 2.1: Intersection of Bm and L curves

2.4 Fringing flux

When an air gap is introduced in the magnetic flux path, the flux spreads over an area
greater than the cross section of the magnetic path. The fringing of the magnetic flux at
2.4 Fringing flux 39

the air gap has two effects

• It increases the cross section area of the air gap

• It increases the length of the magnetic path at the air gap

This fringing at the air gap will reduce the theoretical reluctance at the air gap, and in-
troduce significant errors in the estimated value of the inductance. Hence there is a need
for a simple yet accurate air gap reluctance model to account for the fringing effect. The
fringing flux effect depends on the shape and geometry of the core at the gap, as well as
shape and location of winding and other objects such as clamps, brackets etc.

Fringing effect becomes more noticeable as the air gap increases and simultaneously
air gap reluctance becomes more difficult to estimate. The challenge is estimating this
new reluctance analytically to get a closed-form solution using the dimensions of the core
as the input.

2.4.1 Simple fringing model

Fringing flux

lg

lg lg Ae d

Figure 2.2: Fringing effect approximation from [2]

This model is a modification of the fringing estimate given by [2]. This model was
chosen because of its simplicity and acceptable accuracy. The fringing at the air gap is
modeled as increase in area of the air gap cross section, and this increase is in terms of lg .
The air gap reluctance ℜg for an air gap of lg and core cross section area of Ae = f × d is
given by

lg
ℜg = (2.23)
µ0 [Ae + ( f + d)lg + lg2 ]
40 Filter Component Construction

Eq (2.23) was giving an error of 25% between the theoretical calculated inductance and
the actual measured value. The original equation was altered to reflect the actual in-
ductance that was measured. So Eq (2.23) was modified to include fringing flux at the
corners.

lg
ℜg = (2.24)
µ0 [Ae + 2( f + d)lg + π lg2 ]

In the case of EE type of core from Fig 2.3, there are three possible reluctances: reluc-

Rc Rc

Flux path Rc Rc

Rsg R cg R Rcg
sg
R sg Rsg
NI
Conductors

Figure 2.3: Magnetic circuit representation of EE core inductor

tance of the core ℜc , reluctance of the center leg of E core ℜcg and reluctance of side leg
of E core ℜsg . The total reluctance of the magnetic path will be

ℜsg ℜc
ℜt = ℜcg + + (2.25)
2 2

2.4.2 Bossche and Valchev model

The authors propose basic analytical approximations for fringing coefficients for several
basic cases of air gap configurations [6]-[7]. The total permeance of the air gap is a
summation of the air gap permeance and product of these fringing coefficients multiplied
by corresponding core dimensions.

ae
Λg = µ0 + µ0Cg F (2.26)
lg

Λg is the permeance of the air gap; ae is the cross section area of the core, Cg is the core
dimension (in m) corresponding to fringing coefficient F.
2.4 Fringing flux 41

dcu
d

b le

h
Nv Nh lg ae

x a
w
z f

Figure 2.4: Reference EE core for indicating dimensions

F F F
3 2 3

F F F F F F
3 1 1 1 1 3

F F F
3 2 3

Figure 2.5: Fringing coefficients F1 , F2 , and F3 at gaped inductor

The coefficients for the basic cases possible in an EE type core are:

1 1
!
2 q+ p (r − p)2 (r − 0.26p − 0.5q) q
F1 (p, q, r) = ln + + (2.27)
π 1
q+
1
r
3qr2 3r

1
0.44(r2 + q2 ) − 0.218pr + 0.67pq + 0.33qr + 0.7825p2 2

2
F2 (p, q, r) = ln (2.28)
π p2
 2 !
1 s
F3 (p, s) = cosh−1 3.4 + 1.3 (2.29)
π p
With reference to Fig 2.4, the variables p, q, r and s will be

lg
p= (2.30)
2
42 Filter Component Construction

q = dcu × Nh (2.31)
Nv
r = dcu × (2.32)
2
lg
s = a+ (2.33)
2
where dcu is the diameter of bare copper conductors, Nv is number of conductor layers in
vertical axis, and Nh is number of layers in horizontal axis.

The permeances of each leg of EE core are calculated separately using the fringing
coefficients.

2ae
Λcg = µ0 + µ0 [2(2 f )F2 + 2dF1 ] (2.34)
lg

ae
Λsg = µ0 + µ0 [3F3 f + F1 d] (2.35)
lg
µi µ0 ae
Λc = (2.36)
lg
where Λcg is the permeance of air gap of center leg, Λsg is permeance air gap of side leg,
Λc is permeance of core. The corresponding reluctances are

1 1 1
ℜcg = ; ℜsg = ; ℜc = (2.37)
Λcg Λsg Λc

The net reluctance of the flux path is

ℜsg ℜc
ℜt = ℜcg + + (2.38)
2 2

2.4.3 Comparison

The core measurements are given in Table 2.1. The analytical calculations from both the
fringing models are compared with the actual measured values of the inductor in Table
2.2. Based on the comparison, the simple fringing model has been used in the subsequent
design calculations.
2.5 Fringing edge calculation using FEA 43

ae 840 µ m2 w 34.6 mm Turns 120

le 354 mm lg 12 mm Nh 4
f 28 mm a 76 mm Nv 36
d 30 mm b 48 mm dcu 2.743 mm

Table 2.1: EPCOS Ferrite core UU 93/152/30 measurements used for fringing calcula-
tions

2.5 Fringing edge calculation using FEA

FEA studies were carried out to verify the accuracy of the analytical fringing effect mod-
els. The inductor was simulated for various air gap lengths to estimate how much the flux
lines fringe out in the air gap of the inductor for various air gap lengths. Air gap flux
density in the inductor was plotted for each case. Flux in the air gap was calculated by
integrating the flux density value. Fringing edge was taken as the distance at which the
flux falls to about 85% of maximum flux in the air gap (at the middle of the gap). These
simulations were carried out for ferrite and amorphous core inductors.

2.5.1 Analysis

Fig. 2.6 shows the plots of flux density in the air gap of the side limb of the inductors
plotted against distance across the air gap. Graphs are plotted for different air gaps varying
from 1mm to 24mm. Fig. 2.10 shows the plots of fringing distance and fringing factor
against air gap length. From the plot it is found that the fringing edge is almost equal to
the air gap length(lg) for large air gaps and lg /2 for smaller air gap designs.

2.5.2 Results

Unit Measured Simple model B & V model

L mH 3.439 3.064 4.145
ℜcg MH−1 2.845 2.184
ℜsg MH−1 3.557 2.574
Error % -10.9 20.5

Table 2.2: Comparison of accuracy of two fringing models

44 Filter Component Construction

Figure 2.6: Plot of Airgap flux density vs distance

Figure 2.7: Plot of Normalised airgap flux density vs distance

2.5 Fringing edge calculation using FEA 45

Figure 2.8: Plot of Airgap flux vs distance for one half of the sidelimb

Figure 2.9: Plot of Normalised airgap flux density vs distance for one half of the sidelimb
46 Filter Component Construction

Figure 2.10: Plot of fringing edge (D f ringe ) and fringing factor(k) vs Airgap length
2.6 Inductor design 47

2.6 Inductor design

Magnetic cores used in power electronic applications like transformers and inductors usu-
ally fall in four broad categories[9]. The first is bulk metal, like electrical steels which
are processed from furnace into ingots and then hot and cold rolled. Second is powdered
core materials where are manufactured from various types of iron powders mixed with
special binding agents and then die-pressed into toroids, EE cores and slugs. The third
is ferrite materials which are ceramics of iron oxide, alloyed with oxides or carbonate of
Mn, Zn, Ni, Mg, or Co. The most recent category is of metallic glasses where the bulk
metal is rapidly quenched from molten state to obtain a ‘glassy’ state without a regular
arrangement of metallic atoms in the material.

One of the design objectives is to derive most general procedures for inductor con-
struction. Theoretically, it should be possible to accurately design the inductor using just
the property of permeability of the core material. But practically, the design procedure
for Ferrite, Amorphous and Powdered material is different, mainly because vendors fol-
low different conventions and specify the material properties in many ways. Amorphous
and powder cores also have nonlinear permeability, ie the permeability varies with the
applied field, temperature, air gap etc. Hence the design procedure for different materials
is heavily affected by the available data from vendors, and it is not possible to define a
single generalized accurate design process for all materials.

2.6.1 Ferrites

Ferrites have the most stable (with temperature, flux density and air gap) permeability
of all the magnetic materials. Hence the magnetic circuit equations along with some
modifications for fringing effects at large air gap are sufficient to accurately determine
the inductance of ferrite cores. Ferrite materials also have very low core losses and are
well suited for high frequency operation upto hundreds of kHz range. The downside is
that since ferrite materials have low flux density (typically 0.3T-0.4T), the inductor will
bigger than using other core materials.

1. The area product equation is the starting point. Choose a core having A p greater
than calculated.

2. Use the Graphical Iterative method described in section(2.3) to decide the number
of turns and air gap, incorporating the fringing models discussed in section 3.4 in
the reluctance equations.
48 Filter Component Construction

3. Even though ferrite materials have very stable permeability, to compensate for ef-
fects of varying permeability and other manufacturing tolerance, the air gap should
be selected such that the reluctance of the total air gap is at least ten times the
reluctance of the core.

2.6.2 Amorphous material

Amorphous materials have a high flux density limit of upto 1.5T. The laminated structure
of the amorphous cores also reduces eddy current losses. However, the layered structure
of the C-cores vibrate at the switching frequency, which means in practical operation
Amorphous cores can be noisy especially if the switching frequency is within the human
range of hearing (upto 20 kHz). The noise is also directly proportional to the current
ripple at switching frequency. The noise can be minimized with vacuum impregnation,
reinforcement and by placing the cores in a damped enclosure.

Amorphous cores also have non linear permeability properties. Hence to accurately
design an inductor with amorphous cores, the published AL vs. H curves have to used.

1. Choose an amorphous core with area product greater than required for the specific
application.

2. Select an air gap from the AL curves published by the vendor (Fig 2.11). For this
AL , calculate the number of turns of copper winding
r
L
N= (2.39)
AL

where N is the number of turns of copper winding and L is the required inductance.
The unit of AL here is µ H/(turns)2

3. Ensure that the core is not saturated for this range of induction.

AL NI pk
Bm = (2.40)
Ae

where Bm is the peak flux density in the core for the peak current of I pk , Ae is the
cross section area of the core. If core is saturated, increase the air gap and select
new AL . If it is not possible to choose higher air gap, go to next larger core size.
2.6 Inductor design 49

Figure 2.11: AL vs H for AMCC 200 core. Source: Metglas Inc [12]

2.6.3 Powder material

Powder materials feature a distributed air gap and hence there is no need to include an
explicit air gap. But this distributed air gap also means that these materials have the lowest
permeability of all the core materials discussed. The absolute permeability ranges from
26µ H/m to 300µ H/m [14]. Hence a design decision should also specify the permeability
of the core. Powder materials are also sensitive to temperature variations because of the
binder materials used in the core, though some recent products are more resistant in this
regard.

1. Choose powder core size with area product greater than required for the specific
application.

2. Calculate the maximum number of turns that can be accommodated within the se-
lected core window.

Wa ku
Nmax = (2.41)
acu

where Nmax is the maximum number of turns that can be accommodated in the core
window of area Wa , acu is the bare copper conductor cross section area, ku is the
utilization factor which depends on the type of winding (round wire, foil), method
of winding (square lay, hexagonal lay), number of bobbins (single, two) and finally
50 Filter Component Construction

the winding skill.

3. Calculate the minimum permeability required for the specific application

Lle
µmin = 2
(2.42)
Ae Nmax

where L is the required inductance and le is the magnetic path length. Select a
permeability higher than µmin from the vendor datasheets [14]-[15].

4. AL is usually specified by the vendor for specific core shapes and sizes. Use this
information to calculate the actual required number of turns.
r
L
N= (2.43)
AL

2.7 Capacitor selection

Metallised Polypropylene capacitors are AC capacitors that are especially designed for
high frequency operation. These capacitors are constructed from polypropylene films
on which an extremely thin metal layer is vacuum deposited. The metal layer typically
consists of aluminium or zinc of thickness in range of 0.02µ m to 0.05µ m. Several such
layers are wound together in a tubular fashion to get higher capacitance.

Metallised film capacitors are characterized by small size, wide operating frequency
range, low losses, low to medium pulse handling capabilities, low parasitic impedances
and self-healing. In regular film-foil capacitors, if the electrode foils of opposite potential
are exposed to each other because of wearing away of the dielectric, the foils will short
and the capacitor will be destroyed. But in case of metallised polypropylene capacitors,
because of the extremely thin metal layer, the contact points at the fault area are vaporised
by the high energy density, and the insulation between foils is maintained. Due to the
above reasons, these capacitors are perfectly suited for grid connected filter operation.

For the LCL filter, the capacitors are connected in star and the voltage across each
capacitor will be the phase voltage. The star combination also ensures that the LCL filter
provides both common mode and differential mode attenuation.
2.8 Power resistors 51

Type AP4
Voltage rating (V) 415/440
Tolerance (%) ±5, ±10
tan δ 0.001 at 1kHz
Temperature range (◦ C) -40 to +85

Table 2.3: AC Capacitors specifications [19]

2.8 Power resistors

The resistors used in high power applications like grid connected inverters are termed
as power resistors. There are three types of power resistors, depending on the required
ohmic rating and current rating [20]-[21]. Individual power resistors are available for
upto 1 kW and 100 A. Wirewound resistors are designed for high ohmic value and low
current rating. Edgewound resistors which use metallic ribbon are designed for medium
ohmic value and high current rating. Grid and ribbon resistors are designed for low ohmic
value and high current rating. In the damping resistor for the proposed damping circuit
topology, parasitic inductance is not a major concern as the resulting corner frequency
is much higher than ωL pC . Hence, lower cost wire wound resistors are sufficient for this
application.

Voltage insulation (Vrms) 600

Tolerance (%) +10
Temperature rating
375
above ambient (◦ C)

Table 2.4: Power resistors specifications [20]-[21]

2.9 Design examples

The design procedure for Ferrite cores is already discussed in some detail in sections 3.3
and 3.4. This section will focus on the design examples using Amorphous and Powder
cores.
52 Filter Component Construction

Electrical Material
L (mH) 2.761 Jm (M A/m2) 3
Irms (A) 15.48 Bm (T) 1
I p (A) 26.19 ku 0.6
Vrms (V) 13.40 kf 4.44
VA rating 207.25 fb (Hz) 50

Table 2.5: Electrical and material constraints for amorphous inductor design example

2.9.1 Amorphous core example

The electrical and material constraints are detailed in Table 2.5.

1. The minimum area product is calculated.

VL IL
Ap = (2.44)
k f ku fb Bm Jm

2. We choose amorphous core AMCC 200 from Metglas Inc. The area product of the
selected core is 5.187×10−6m4 .

3. The air gap for the selected C core is selected such that variations in material mag-
netic properties do not affect the final inductance. The air gap reluctance is taken
to be 10 times the core reluctance. This gives the minimum air gap.

le
lg(min) = 10 (2.45)
µi

where lg(min) is the minimum selected total air gap, le is the magnetic path length
(from datasheet) and µi is the initial permeability. The initial permeability of amor-
phous material 2605SC is specified as 1500 [3].

4. For this air gap, use the AL curves to find number of turns and core flux density. If
the core flux density exceeds Bm increase the air gap and recalculate N and Bm .

5. The final settings are

lg = 4 mm (2.46)

AL = 0.467 µ H/(turns)2 (2.47)

2.9 Design examples 53

s
2.761 × 10−3
N= = 77 (2.48)
0.467 × 10−6
0.467 × 10−6 × 77 × 26.19
Bm = = 0.99T (2.49)
9.5 × 10−4

2.9.2 Powder core example

Electrical Material
L (mH) 0.276 Jm (M A/m2) 3
Irms (A) 15.48 Bm (T) 1.4
I p (A) 65.17 ku 0.674
Vrms (V) 1.34 kf 4.44
VA rating 20.73 fb (Hz) 50

Table 2.6: Electrical and material constraints for powder inductor design example

The electrical and material constraints are detailed in Table 2.6.

1. The minimum area product is calculated.

VL IL
Ap = (2.50)
k f ku fb Bm Jm

2. We choose BK 6320 (assembled unit 2) from Changsung corp. The area product of
the selected core is 1.44×10−6 m4 .

3. The maximum number of turns for this core size will be

12 × 10−4 × 0.674
Nmax = = 138 (2.51)
5.48 × 10−6

4. The absolute minimum permeability required for this inductor is

0.276 × 10−3 × 22.28 × 10−2

µmin = = 2.86 × 10−6 H/m (2.52)
12 × 10−4 × 1382

5. We choose MegaFlux powder core material of absolute permeability 40µ . AL for

54 Filter Component Construction

this core for the selected permeability is 270nH/(turns)2.

s
0.276 × 10−3
N= = 32 (2.53)
270 × 10−9

6. The flux density in the core will be

270 × 10−9 × 32 × 65.17

Bm = = 0.469T (2.54)
12 × 10−4

Material Cost Weight # Cost Weight

p.u (Rs) p.u (kg) of units per L (Rs) per L (kg)
Ferrite (UU93/152/30) 381 0.75 4 1524 3
Amorphous (367S) 1625 1.662 1 1625 1.662
Amorphous (630) 3220 3.67 1 3220 3.67
Powder (BK7320) 340 0.2735 8 2720 2.188

Table 2.7: Core material cost

Inductor type Core Copper Other Total Total

cost (Rs) cost (Rs) charges (Rs) cost (Rs) weight (kg)
Ferrite 1524 1046 1142 3712 4.28
Amorphous (367S) 1625 3733 2435 7793 4.61
Amorphous (630) 3220 2216 1826 7263 5.23
Powder -Foil 2720 2566 1643 6930 3.66
Powder -Round wire 2720 636 1643 5000 2.966

Table 2.8: Total cost of Inductors

Capacitor rating 6 µ F 8 µF 10 µ F 20 µ F
Cost (Rs) 72 96 95 180

Table 2.9: Cost of AC Capacitors 440V AC rating

2.10 Measurements on inductors 55

2.10 Measurements on inductors

2.10.1 Measurement of permeability of core material

An experiment was conducted to verify the permeability of the inductor core material to
see if it matched the manufacturer specified values. For this purpose a transformer was
made using the same core material as that of the inductor,without adding any airgap in the
fluxpath. For different values of ac current through the primary coil, the induced voltage
in the secondary coil were noted. The flux density inside the core was calculated using
transformer equation as below. Peak core flux density is specified in tesla (T).

E
B= (2.55)
4.44NA f

where E = induced emf in the secondary (rms) in volts

N = number of turns in the secondary winding

f = frequency of the input voltage in hertz

A = cross-sectional area of the core in sq.m

The magnetisation (H in Ampereturns/metre) of the coil can be calculated as

NI
H= (2.56)
l

where N = No. of turns in the coil

I = Current through the coil in amperes

l = Mean magnetic path length in metres

Relative permeability of the core material is obtained by using the equation

B
µr = (2.57)
H µ0

where µ0 is the permeability of air.

56 Filter Component Construction

Figure 2.12: Experimental setup for measuring core fluxdensity and permeability

Core material Relative permeability

Ferrite 3340
Amorphous (AMCC 630) 1200
Amorphous (AMCC 367S) 1800
Powdered core 45

Table 2.10: Measured relative permeability of the different core materials.

2.10.2 Measurement of airgap flux density

Flux density in the airgap of the ferrite inductor was measured using a gaussmeter to
validate analytical results and FEA calculations. Plot of the airgap flux density against
distance across the airgap is given in Fig. 2.13 Gaussmeter used was make: MAGNET-
PHYSIK, model : FH54.

2.10.3 Measurement of core flux density

The experiment to be conducted is same as that for obtaining permeability. Flux density
is calculated from the transformer equation. Refer to Table 2.11 for the results of the
experiment. Better match between analytical results and measurements were observed
when compared to the FEA.
2.11 Summary 57

Figure 2.13: Comparison between FEA and measurements of airgap fluxdensity in ferrite
core inductor.

Inductor type Fluxdensity (T)

Ferrite 0.252
Amorphous (AMCC 630) 0.688
Powdered core 0.259

Table 2.11: Core fluxdensity in the inductors.

2.11 Summary

The familiar area product approach for inductor design has been evaluated and modifica-
tions to improve the accuracy of the final constructed inductance are suggested. A new
approach for selection of air gap and number of turns in an inductor is proposed. The
effect of fringing of the magnetic flux at the air gap is investigated and simple equations
are suggested to model this effect. The design techniques for three magnetic materials
-Amorphous, Ferrite and Powder are discussed and elaborated using actual examples. Ta-
bles 2.7 and 2.8 summarize the size and weight of the inductors. It was seen that the
ferrite inductors were low cost and the powdered core had the lowest weight.
58 Filter Component Construction

2.12 References

Area Product method

1. N.Mohan,T.M.Undeland,W.P.Robbins, Power Electronics- Converters, Applications

and Design, 3rd ed., John Wiley and Sons, 2003, pp. 744-792.

2. G.S.Ramana Murthy, “Design of Transformers and Inductors at Power-Frequency–

A modified Area-Product method,” M.Sc(Engineering) Thesis, Indian Institute of
Science, March 1999.

3. Col.Wm.T. McLyman, “AC Inductor Design,” in Transformer and Inductor Design

Handbook, 3rd ed., Marcel Dekker, 2004, pp. 10-1–10-13

Fringing effect

4. T.G.Wilson, A.Balakrishnan, W.T.Joines, “Air-gap reluctance and inductance cal-

culations for magnetic circuits using a Schwarz-Christoffel transformation,” IEEE
Trans. on Power Electronics, vol. 12, no. 4, pp. 654-663, July 1997

5. A.F.Hoke, “An improved two-dimensional numerical modeling method for e-core

transformers,” A.B. Honor’s Thesis, Thayer School of Engg., Dartmouth Coll.,
Hanover, New Hampshire, June 2001

6. A. van den Bossche, V.Valchev, T.Filchev, “Improved approximation for fringing

permeances in gapped inductors,” Industry Applications Conference, 2002. 37th
IAS Annual Meeting Conference Record of the, 2002, vol.2, pp. 932- 938

7. A. van den Bossche, V.Valchev, Inductors and Transformers for Power Electronics,
1st ed., Taylor and Francis, 2005, pp. 333-342

Components

8. H.W. Beaty, Electrical Engineers Materials Reference Guide, McGrawHill Engi-

neering Reference Series, 1990

9. Col.Wm.T. McLyman, “Magnetic cores,” in Transformer and Inductor Design Hand-

book, 3rd ed., Marcel Dekker, 2004, pp. 3-1–3-48
2.12 References 59

10. H.Skarrie, “Design of Powder core inductors,” Ph.D. dissertation, Dept. of Ind.
Elec. Eng. and Auto., Lund Univ., Lund, Sweden, 2001

11. Ferrites and Accessories Catalog: EPCOS, (2007)

http://www.epcos.com

12. Amorphous PowerLite C cores: Metglas Inc,

http://metglas.com/products/page5_1_6.htm

13. MPP, SuperMSS,HiFLux : Arnold Magnetic Technologies,

http://www.arnoldmagnetics.com/products/powder/powder_catalogs.htm

14. MPP, HiFlux, Kool Mu : Magnetics Inc,

http://www.mag-inc.com/library.asp

15. MegaFlux power cores : Changsung Corp.

http://www.changsung.com

16. Standard Specification for Flat-Rolled, Grain-Oriented, Silicon-Iron, Electrical

Steel, Fully Processed Types, ASTM A876-03, 2003

17. Standard Specification for Fully Processed Magnetic Lamination Steel, ASTM
A840-06, 2006

18. Silicon Steels and their applications- Key to Steel;

http://www.key-to-steel.com/default.aspx?ID=CheckArticle&NM=101

19. AC Metallized Polypropylene Capacitors : Advance Capacitors Ltd,

http://www.advance-capacitors.com

20. Power Resistors : Mega Resistors,

http://www.megaresistors.com/en/index.php

21. Power Resistors : Powerohm Resistors,

http://www.powerohm.com/index.php
Chapter 3

Simulation Using FEA Tools

3.1 Introduction

Finite Element Method is a numerical technique used to solve partial differential equa-
tions. In this method, the problem domain is discretised into different cells or regions.
The field to be solved for is approximated using a polynomial in each of the cells. These
polynomials are solved for using numerical methods to obtain the field values in each
cell.

Simulations using Finite Element method were carried out to confirm the design of
inductors for the filter. The tools employed were

1. FEMM (Finite Element Method Magnetics) : FEMM is a freeware that makes use
of Finite Element Method to solve Maxwell’s equations. Simulation using FEMM
involves the following steps.

(a) Draw the inductor geometry

(b) Assign materials and boundary conditions
(c) Make coils and apply excitation
(d) Create Finite Element mesh
(e) Solve using static/time-harmonic solver
(f) View the solution results and fields

As a first approximation, the windings of the inductor were modelled as a single

sheet of copper carrying current. A refined model with different layers of copper

61
62 Simulation Using FEA Tools

was chosen in the next step. Further refinement in the winding geometry was done
by modelling the actual individual conductors in the winding. Flux and eddy current
can be solved for. Inductance value is calculated by FEMM and is shown as a
result under coil properties. Simulation using FEMM is possible only in 2D. Also
it cannot take into account hysteresis in magnetic materials.

2. MagNet :The steps followed in simulation are same as that in FEMM. These models
were simulated and solved in 2D as well as 3D. The optimum mesh size for FEM
was chosen by trial and error. Models were solved using static(for dc current)
and time-harmonic(for ac current)solvers. MagNet calculates the flux linkage and
energy and are displayed in the post processing bar. After solving the model, the
magnetic flux lines in the core can be viewed as a contour-plot. Flux density can be
plotted along any required contour. Inductance can be calculated by two methods.

(a) From flux linkage :

ψ
L= (3.1)
I

where ψ is the flux linkage in weber and I is the rms value of current through
the inductor in amperes
(b) From stored energy :

2W
L= (3.2)
I2

where W = Energy stored in the inductor in joules

Results of simulation using MagNet are given in the section()of appendix.

3.2 Ferrite core inductor

Table 3.2 shows the calculation done to estimate the dimensions for the winding geometry
of the inductor.
3.2 Ferrite core inductor 63

Core material Ferrite (EPCOS N87)

Core type UU core
Inductance 3mH
Relative permeability of core 2200
Number of turns in winding 120
Copper wire gauge 12SWG
Airgap length 12mm

Table 3.1: Inductor specifications for FEA.

Core Dimension Units

Window height (Wh ) 96 mm
Window breadth (Wb ) 37 mm
Air gap length (lg) 12 mm
Dielectric thickness (Dt ) 6 mm
Area of bare wire = Wh + lg − 2Dt 5.48 mm2
Available window height(H) = Wb − 2Dt 96 mm
Available window breadth(B) 25 mm
Available window area = HB 2400 mm2
Diameter of copper wire including insulation (D) 2.743 mm
Turns per layer (T) =H/D 35
Diameter of copper wire without insulation (Dc ) 2.642 mm
Layer height =T Dc 92.46 mm
Layer width = Dc 2.642 mm
Insulation thickness between two layers = D − Dc 0.101 mm
Number of layers (N) 4
Sweep distance for coil = DN 10.972 mm

Table 3.2: Calculations for determining copper winding geometry for ferrite core induc-
tor.
64 Simulation Using FEA Tools

Figure 3.1: Ferrite core inductor model with single layer winding.

Figure 3.2: Ferrite core inductor model with two layer winding.
3.2 Ferrite core inductor 65

Figure 3.3: Ferrite core inductor model with four layer winding.

Figure 3.4: Ferrite core inductor model with individual conductors.

66 Simulation Using FEA Tools

Figure 3.5: Finite element mesh created in FEMM.

Figure 3.6: Flux plot of ferrite core inductor.

3.2 Ferrite core inductor 67

Figure 3.7: Flux density plot of ferrite core inductor.

68 Simulation Using FEA Tools

3.3 Amorphous core inductor

Core material Amorphous Amorphous

Core type C core(AMCC630) C core(AMCC 367S)
Relative permeability of core 1500 1500
Inductance 4.975mH 4.975mH
Number of turns in winding 68 137
Conductor type Copper foil Copper foil
Foil thickness 5 mil 5 mil
Insulation thickness (mm) 0.127 0.2
Airgap length 1.6mm 4.6mm

Table 3.3: Amorphous core inductor specifications for FEA.

Figure 3.8: Amorphous core (core type: AMCC630)inductor model.

3.3 Amorphous core inductor 69

Figure 3.9: Flux in amorphous core (core type: AMCC630)inductor.

Figure 3.10: Flux density plot of amorphous core (core type: AMCC630)inductor.
70 Simulation Using FEA Tools

Figure 3.11: Amorphous core (core type: AMCC367S)inductor model.

Figure 3.12: Flux in amorphous core (core type: AMCC367S)inductor.

3.3 Amorphous core inductor 71

Figure 3.13: Flux density plot of amorphous core (core type: AMCC367S)inductor.
72 Simulation Using FEA Tools

3.4 Powdered core inductor

Core material Powdered Iron

Core type Block core (BK 7320)
Inductance (mH) 1.7
Relative permeability of core 40
Number of turns in winding 78
Conductor type Round conductor winding

Table 3.4: Powdered core inductor specifications for FEA.

Figure 3.14: Powder core inductor model.

3.5 Comparison of FEMM and MagNet 73

Figure 3.15: Flux in Powder core inductor.

Figure 3.16: Flux density plot of powder core inductor.

3.5 Comparison of FEMM and MagNet

FEMM is a suite of programs for solving low frequency electromagnetic problems on

two-dimensional planar and axisymmetric domains. It can address linear/nonlinear mag-
netostatic problems and linear/nonlinear time harmonic magnetic problems. FEMM is an
open-source software. Three-dimensional solutions cannot be done using this tool. Also
it cannot take into account hysteresis in magnetic materials. MagNet is a licensed tool.
It can handle both two-dimensional and three-dimensional problems. It was seen overall
that the analytical methods developed in chapter (2) is sufficient for the grid interactive
inverter filter design procedure.
74 Simulation Using FEA Tools

3.6 References

1. Ferrites and Accessories Catalog: EPCOS, (2007)

http://www.epcos.com

2. Amorphous PowerLite C cores: Metglas Inc,

http://metglas.com/products/page5_1_6.htm

3. MegaFlux power cores : Changsung Corp.

http://www.changsung.com

4. M.V.K. Chari and S.J. Salon, Numerical Methods in Electromagnetism,Elsevier

Inc.,2000

5. Robert Warren Erickson, Dragan Maksimovic, Fundamentals of power electronics,

Kluwer Academic Publishers Group ,2001

6. Infolytica Corporation, MagNet User guide,

http://www.infolytica.com

7. Foster Miller Inc., Finite Element Method Magnetics(FEMM) Documentation,

http://www.femm.info/wiki/HomePage
Chapter 4

Power Loss and Heating Effects

4.1 Introduction

The focus of the previous chapter was on the system level design, and hence the phys-
ical characteristics of the individual filter components was understated. This chapter is
focused on the power loss and efficiency of individual filter components, which has signif-
icant implication on the power converter efficiency and reliability. Losses in the inductive
part of the filter are more prominent compared to capacitive losses. Hence more atten-
tion is focused on inductor core and copper losses. Analytical equations predicting the
power loss in inductors are derived from the basic electromagnetic equations. The the-
oretical derivations in this chapter are necessarily brief and a more complete treatment
can be found in the references listed for each section. The theoretical background for this
chapter is covered in Appendix-A at the end of the report.

4.2 Core loss

The relationship between H and B in any magnetic material is given by the magnetization
curve.The loop area of the magnetization curve represents the energy dissipated per unit
volume of the material over a complete magnetization period. Let us assume that a field,
slowly increasing with time, is applied by means of a magnetizing winding supplied with
a current i(t) to a magnetic circuit with path length lm . At any instant of time the supplied
voltage is balanced by the resistive voltage drop of the winding Rw i(t) and the induced

75
76 Power Loss and Heating Effects

i (t) φ(t)

u (t) N

Figure 4.1: Energy balance

B Saturation
Hp , Bp

Retentivity

Coercivity

−H H

Saturation

−Hp ,−Bp
−B

Figure 4.2: Magnetization curve of a magnetic material

emf d φ /dt.

dφ
u(t) = Rw i(t) + N (4.1)
dt

Starting from the demagnetized state, a certain final state with induction value B p is
reached after a time interval to . The corresponding supplied energy is E is partly dis-
sipated by Joule heating in the conductor and partly delivered to the magnetic circuit.
Z to
E= u(t)i(t)dt (4.2)
0
Z to Z to
2 dB
E= Rw i (t)dt + NAi(t) dt (4.3)
0 0 dt
4.2 Core loss 77

where N is the number of turns of the winding and A is the cross-sectional area of the
sample. Since

lm
i(t) = H(t) (4.4)
N

the energy delivered by the external system in order to bring the magnet of volume v = Alm
to the final state is
Z to Z Bp
dB
U =v H(t) dt = v HdB (4.5)
0 dt 0

The energy per unit volume to be supplied in order to reach the induction value B p is then
given by the area delimited by the BH curve and the ordinate (y) axis. If the integration in
Eq. 4.5 is carried out over a full cycle, the energy dissipated per unit volume is obtained
as the area of the hysteresis loop
I I I
W= HdB = µo HdH + µo HdM (4.6)

where Magnetization M is defined as the magnetic dipole moment per unit volume.

1 n∆v
M = lim ∑ mi
∆v→0 ∆v i=0
(4.7)

The integral HdB over one magnetization cycle gives the energy per unit volume trans-
formed into heat. This is termed as loss per cycle, whereas the term power loss is used to
denote the loss per unit time P = W f .The purely reactive term µo HdH, integrating the
energy exchanged between the supply system and the magnetic field averages out to zero.
So W is decided by the second term in the equation. In general the loss per cycle W is a
non-linear function of frequency, peak induction B and the harmonic content of the induc-
tion waveform. The loss per cycle increases non-linearly with f -it can be decomposed
into three different components– the frequency independent term Wh (hysteresis loss), the
classical loss Wc ∝ f , and the excess loss We ∝ f 1/2 . The loss decomposition can be
physically justified by the statistical theory of losses [11].

4.2.1 Eddy current loss

This can be directly calculated from Maxwell’s equations, assuming a perfectly homo-
geneous conducting material. The classical loss is present under all circumstances, to
which the other contributions are added when structural disorder and magnetic domains
78 Power Loss and Heating Effects

are present [11].

π 2σ d2 2 2
Pc = Bp f (4.8)
6

where d is the lamination thickness, σ is electrical conductivity, f is magnetization fre-

quency, B p is peak induction (sinusoidal). For a non-sinusoidal induction where B is
expressed as B(t) = ∑n Bn sin(2π f + φn ) where φn is the phase shift with respect to the
fundamental harmonic, the classical loss becomes,

π 2σ d2 f 2
Pc =
6 ∑ n2 B2n (4.9)
n

The minimum for this loss occurs for a triangular induction waveform, with dB/dt con-
stant in each magnetization half-cycle, then the coefficient π 2 /6 becomes 4/3. Classical
losses is independent of the magnetic property of the material and all materials behave
in the same way if the geometry and electric properties are kept constant. This simpli-
fication comes since dB/dt is assumed to be uniform throughout the thickness of the
material. But this condition only holds at low frequencies as the magnetic field produced
by eddy currents inhibit the applied field and tend to shield the interior of the core at
higher frequencies.

4.2.2 Excess loss

Excess losses occur since the eddy currents are concentrated in the vicinity of the moving
domain walls, causing losses higher, or excess than the classical terms. In case of a
lamination of thickness d with longitudinal magnetic domains of random width [11], the
Maxwell’s equation can be solved to find
!
48 1 2L ∼ 2L
Pe =
π3 ∑ n3 d
Pc = 1.63 Pc
d
(4.10)
odd n

where 2L is the average domain width, n is the harmonic order and Pc is the classical loss.
In case of highly optimized grain-oriented Si steel, 2L/d ∼
= 1. In the general case, the
excess loss can be approximately computed using this expression.
√
Pe = ke σ (B p f )3/2 (4.11)

where the parameter ke depends on the microscopic structure of the material.

4.2 Core loss 79

4.2.3 Hysteresis loss

Every atom has a small magnetic moment, and in Ferromagnetic materials the inter-
atomic forces tend to align these moments in the same direction over regions containing a
large number of atoms. These regions are called domains; the domain moments, however
vary in direction from domain to domain. When such materials are subjected to external
magnetic field, the domains which have moments in the direction of the field grow at the
expense of other domains. This is the process of magnetization of the material in the
direction of the applied external field.

At the microscopic level, the magnetization process proceeds through sudden jumps,
called Barkhausen jumps of the magnetic domain walls. Very intense and brief cur-
rent pulses of the order of 10−9 s [11] are generated close to the domain wall segments.
These spatially localized eddy currents induced by the domain-wall jump dissipate a fi-
nite amount of energy through the Joule effect. The sum of all the domain-wall jumps
will account for the observed hysteresis loss. With a higher rate of change dH/dt the
time interval will decrease, so number of Barkhausen jumps and the amount of energy
dissipation per unit time is proportional to the magnetization frequency. The expression
for the hysteresis loss in one magnetization cycle is

Ph = 4kh Bαp f (4.12)

The parameters kh and α depend on the structural properties of the material at the micro-
scopic level. No general rule exists for determining their values in different materials.

4.2.4 Total loss

A detailed evaluation of the core loss requires extensive knowledge of the microstructure
of the material along with the numerical implementation of mathematical models of hys-
teresis. The complexity of the problem coupled with the fact that the magnetic materials
chosen for filter design have very low core loss compared to copper winding loss suggests
that the core loss graphs published by vendors of magnetic material are sufficient to esti-
mate the core losses. This has been confirmed by experimental observation under steady
state operating conditions where temperature rise in the core was very less compared to
the winding.
80 Power Loss and Heating Effects

4.3 Copper loss

A voltage is induced in a conductor if it is subjected to time varying magnetic flux, ac-

cording to Faraday’s law. The inducing field may be due to its own current, which must be
time varying or due to time varying current carried by another adjacent conductor. In the
first case the phenomenon is called Skin effect and the second case is called Proximity
effect. The induced voltage gives rise to currents distributed throughout the body of the
conductor. These currents are called Eddy currents and they have three major effects.

• Heat because of ohmic losses

• Opposite magnetic reaction field

• Additional forces due to interaction of induced and inducing fields

The two eddy current effects discussed above will occur simultaneously in a conduc-
tor that carries an alternating current and is positioned in an external alternating field,
which is the exact situation of a conductor which is part of the winding of an inductor
or transformer. The effect of these eddy currents can be calculated by formulating elec-
tromagnetic equations, either in differential form or integral form. The differential form
of the Maxwell’s equations describe the electromagnetic field vectors- E, H, J, B at any
point in space. These differential equations can be solved by analytical or numerical
methods.

Analytical solutions are limited to linear equations, with specific geometries and sim-
ple excitation. Analytical methods normally use field equations since boundary condi-
tions are expressed in terms of magnetic and electric fields.The solutions are limited to
mostly one or two dimensional problems. One dimensional problems have closed form
solutions which give good insight into the problem. Numerical methods can handle com-
plicated geometries and both linear and non-linear equations. They however, require large
computation times. In most cases, numerical methods use magnetic vector potential in
conjunction with electric scalar potential. The integral form of these electromagnetic
equations are particularly suited for numerical methods.

4.4 Foil conductors

Foil conductors are well suited for applications which have both a high switching fre-
quency and high rated current. By proper selection of thickness of foil, it is possible to
4.4 Foil conductors 81

significantly reduce skin effect losses. Most high power inductor designs make use of foil
winding to minimize high frequency copper losses. The subsequent analysis of power
loss in foil winding is referred from [16], [17], [20] and [27].

4.4.1 Assumptions

1. The magnetic field distribution is solved for a winding portion. A winding portion is
a part of the winding which extends in either direction along the axis of the winding
height from a position of zero field intensity to the first positive or negative peak of
the magnetic field intensity.

2. The conductor foils are assumed to span the entire breadth of the core window.

3. Magnetic field in the winding space is assumed to be parallel to center leg of the
inductor. This is strictly accurate only in case of infinite solenoid windings. If the
foil winding is assumed to span the entire window height, then this assumption is
valid.

4. The winding layer is modelled as a finite portion of an infinite current sheet. This
gives the solution of field equation in rectangular coordinates.

5. The curvature of the foil conductors is neglected while calculating the radial field
distribution across the winding layer.

6. Almost all of the magnetic field intensity of any winding layer is assumed to exist
inside the region bounded by that layer and there is negligible magnetic field outside
this region.

4.4.2 One dimensional H field

Fig. 4.3 shows the typical cross section of inductor with foil windings. Ampere’s law can
be used to find the magnetic field intensity between conductor layers.

I pk
Ho = (4.13)
bwin

where I pk is the peak current flowing in each layer, and bwin is the width of the window.
The field equations of H and J can be solved in rectangular coordinates. Hence the
82 Power Loss and Heating Effects

Ho core

Hz (x=0) Hz(x=hcu)
bwin Jy (x)

mth layer
hcu

breadth Foil conductors

z depth
y

x
height

Figure 4.3: Inductor cross section with magnetic flux intensity and current density for mth
layer.

magnetic field intensity phasor can be represented as

H(x, y, z) = Hx (x, y, z)âx + Hy (x, y, z)ây + Hz (x, y, z)âz (4.14)

To simplify the analysis we will assume that the spatial magnetic field intensity phasor
is a function of x only and directed in z direction. Then the three-dimensional diffusion
equation becomes a one-dimensional equation.

H(x, y, z) = Hz (x)âz (4.15)

From the wave equation for time-harmonic fields we get,

∂ 2 Hz(x)
52 H − γ 2 H = − γ 2 Hz (x) = 0 (4.16)
∂x 2

where γ 2 ' jω µσ . Here, ω if frequency of the applied current in rad/s, µ is the absolute
permeability of the foil material (copper, µ same as air) and σ is the conductivity of the
foil winding. The general solution has the form

Hz (x) = Aeγ x + Be−γ x (4.17)

where A and B are determined by applying the boundary conditions of magnetic field
intensity at the surface of the current sheet, Hz (x = 0) and Hz (x = hcu ).
4.4 Foil conductors 83

Expressing in hyperbolic form using the boundary conditions,

1
Hz(x) = [Hz (x = hcu ) sinh γ x + Hz (x = 0) sinh γ (hcu − x)] (4.18)
sinh γ hcu

To prevent zeros from appearing in the denominator of some equations that are derived,
the variable x is changed to χ , such that the new variable does not become zero for any
boundary condition.
(
x if|Hz (x = hcu )| ≥ |Hz(x = 0)|
χ= (4.19)
hcu − x if|Hz (x = hcu )| < |Hz(x = 0)|

This definition causes χ = 0 to be always at the surface having the smaller of the two
boundary magnetic fields, and χ = hcu to be always at the surface with the larger field. If
we define the boundary condition ratio as

Hz (χ = 0)
p= (4.20)
Hz(χ = hcu )

From Fig. 4.3, the boundary condition ratio for the mth layer will be

(m − 1)Ho
p= (4.21)
mHo

Substituting Eq. (4.20) in Eq. (4.18)

Hz(χ = hcu )
Hz(χ ) = [sinh γχ + p · sinh γ (hcu − χ )] (4.22)
sinh γ hcu

4.4.3 Power dissipation

The current density phasor Jy (x) in terms of the magnetic field intensity phasor Hz (x) is
derived from Maxwell’s equations.

5 × H = σ E + jω εE (4.23)

where σ is the conductivity of the material carrying the alternating current of frequency
ω rad/sec, ε is the permittivity of the conducting material, which is almost same as free
space. Since for a good conductor σ  ω ε,

5 ×H ≈ σE = J (4.24)
84 Power Loss and Heating Effects

Since J is in y direction and H is in x direction and both are functions of x, taking the curl,

∂ Hz (x)
Jy (x) = − (4.25)
∂x

−γ Hz (χ = hcu )
Jy (χ ) = [cosh γχ − p · cosh γ (hcu − χ )] (4.26)
sinh γ hcu
The power dissipated per unit volume pd (t) is

|J|2
pd (t) = (4.27)
σ

The time-averaged power dissipated can be calculated from the above expression by in-
tegration. To simplify the calculation, the winding layer is assumed to be flat instead of
cylindrical, extending a distance equal to length of turn in the y direction.

hcu

bwin

lT
z
y
x
Figure 4.4: Current sheet approximation to find total power loss

Z T Z bwin Z lT Z hcu
1
Pd = pd (t) d χ dy dz dt (4.28)
T 0 0 0 0

where T is the period of the waveform. For sinusoidal waveforms, we can simplify the
above expression,
Z bwin Z lT Z hcu
1
Pd = pd (t) d χ dy dz (4.29)
2 0 0 0
4.4 Foil conductors 85

lT is the length of mean turn of coil. Since J does not vary with y or z, the integration
with respect to those variables becomes simple multiplication.
Z hcu
bwin lT
Pd = pd (t) d χ (4.30)
2 0

The power dissipated per square meter in the y − z plane is given by

Z h
Pd 1
[QJ ] = = Jy (χ ) · Jy∗ (χ ) d χ (4.31)
bwin lT σ 0

where Jy∗ (χ ) is the complex conjugate of Jy (χ ) and [QJ ] represents time-average. The
resulting expression is

|Hz(χ = hcu )|2 sinh ∆ cos ∆ + cosh ∆ sin ∆

 
2 sinh 2∆ + sin 2∆
[QJ ] = (1 + p ) − 4p
σδ cosh 2∆ − cos 2∆ cosh 2∆ − cos 2∆
(4.32)

where skin depth δ = 2/(ω σ µ), ∆ = hcu /δ is defined as the height of the winding layer
p

hcu normalized to skin depth δ , p is the boundary condition ratio, and |Hz(χ = hcu )|2 is
the square of the rms value of the larger magnetic field intensity at the two surfaces of the
current sheet.

Using the two following hyperbolic identities [20],

 
sinh 2a + sin 2a 1 sinh a + sin a sinh a − sin a
= + (4.33)
cosh 2a − cos 2a 2 cosh a − cos a cosh a + cos a

and

sinh 2a + sin 2a sinh a − sin a sinh a cos a + cosh a sin a

= +2 (4.34)
cosh 2a − cos 2a cosh a + cos a cosh 2a − cos 2a

we can simplify Eq. (4.32) as

|Hz (χ = hcu )|2 (1 − p)2 sinh ∆ + sin ∆ (1 + p)2 sinh ∆ − sin ∆

 
[QJ ] = + (4.35)
σδ 2 cosh ∆ − cos ∆ 2 cosh ∆ + cos ∆

Substituting for p from Eq. (4.21)

|Hz(χ = hcu )|2 1 sinh ∆ + sin ∆ 2 sinh ∆ − sin ∆

 
[QJ ] = + (2m − 1) (4.36)
σδ 2m2 cosh ∆ − cos ∆ cosh ∆ + cos ∆
86 Power Loss and Heating Effects

Replacing Hz (χ = hcu ) using Eq. (4.21) and Eq. (4.13)

2
I pk sinh ∆ + sin ∆ 2 sinh ∆ − sin ∆
 
1
[QJ ] = 2 + (2m − 1) (4.37)
bwin 2σ δ cosh ∆ − cos ∆ cosh ∆ + cos ∆

To find the average power dissipated per meter (in the y direction),

Pav = [QJ ] · bwin (4.38)

4.4.4 AC resistance

In electrical terms, the average power dissipated is also given in terms of resistance as

2
Pav = Rac Irms (4.39)

where Irms is the rms current in each foil conductor. Pav is also expressed as,

Rdc 2
Pav = Rac I (4.40)
Rdc rms

where Rdc is the dc resistance of the foil conductor per unit length

1
Rdc = (4.41)
σ bwin hcu

Hence we get

Rac 2 1
Pav = Irms (4.42)
Rdc σ bwin hcu

Equating Eq. (4.38) and Eq. (4.42),

∆ sinh ∆ + sin ∆ 2 sinh ∆ − sin ∆

 
Rac = Rdc + (2m − 1) (4.43)
2 cosh ∆ − cos ∆ cosh ∆ + cos ∆

This is the Dowell’s formula to calculate AC resistance of the mth layer of a foil winding
having a dc resistance of Rdc per unit length.
4.5 Round conductors 87

4.5 Round conductors

Round conductors are most widely used to construct inductors because of their low cost
and ease of use. But Rac in case round conductors is higher than foil conductors for
the same frequency, and increases much faster with number of turns and layers. The
following analysis of power loss in round conductors is referred from [1], [3], [15] and
[18].

4.5.1 Orthogonality

As discussed before, eddy current effects can be divided into skin effect and proximity ef-
fect losses. It is possible to separately calculate the losses due to skin effect and proximity
effect since the two currents are independent of each other. The conditions in which this
orthogonality is valid is detailed in [20] and [22]. A sufficient (but not necessary) con-
dition is that the conductor must have an axis of symmetry and the current distribution
due to skin effect current has odd symmetry about this axis and current distribution due
to proximity effect current has even symmetry about this axis (or vice verse), as shown in
Fig. 4.5 [22].
Symmetry axis Current: Even symmetry Current: Odd symmetry

H1 H2 −Ho Ho H H

Total loss in a layer Skin effect loss Proximity effet loss

Wire carrying current No net current
H2 −H1 H1 + H 2
Ho H
2 2

Figure 4.5: Orthogonality in Eddy current losses [22]

Let the current density vector be a function of x and z axis and directed along y axis
88 Power Loss and Heating Effects

(Fig. 4.4),

J = Jy (x, z) (4.44)

The average power dissipated per unit length for sinusoidal waveforms is already men-
tioned as

1
Z
Pd = J J ∗ dA (4.45)
2σ A

where A is the cross section area of the conductor. Separating the skin effect current and
proximity effect currents, Js and J p respectively,

1
Z
Pd = (Js + J p )(Js∗ + J p∗ ) dA (4.46)
2σ A

Since Js has even symmetry and J p has odd symmetry,

1
Z
Pd = (JsJs∗ + J p J p∗ ) dA (4.47)
2σ A

Pd = Pds + Pd p (4.48)

Hence the skin effect losses and proximity effect losses can be calculated separately. The
sum will give the total eddy current losses of the conductor.

4.5.2 Skin effect loss

Bφ

ρ
o

ρ
z
φ

Figure 4.6: Round conductor in cylindrical coordinates

The Bessel function solution can be used to find the current distribution in round
cylindrical conductors subjected to an alternating electric field E. All the field vectors are
expressed in cylindrical coordinates (ρ , φ , z). Consider a round conductor of radius ρo
4.5 Round conductors 89

carrying a time varying current of rms value Irms at a frequency ω rad/sec. For conductors,
γ 2 ' jω µσ . Maxwell’s equations are,

5 ×H = J = σE (4.49)

5 × E = − jω µH (4.50)

So

1
− 5 ×J = H (4.51)
jσ ω µ

Substituting Eq (4.51) in Eq (4.49)

1
− 5 ×(5 × J) = J (4.52)
jσ µω

52 J = jσ µω J (4.53)

52 J = γ 2 J (4.54)

where

γ 2 = jω µσ (4.55)

If the current density is z directed with no variation along z and φ , then we can expand
Eq. (4.54)

d 2 Jz 1 dJz
+ − γ 2 Jz = 0 (4.56)
dρ 2 ρ dρ

Multiplying throughout by ρ 2

d 2 Jz dJz
ρ2 +ρ − γ 2 ρ 2 Jz = 0 (4.57)
dρ 2 dρ

We can simplify the above equation as

γ 2 = jσ ω µ = jp (4.58)
p √
γ= j p (4.59)
2J
2d z dJz
ρ +ρ − ( jp ρ )2 Jz = 0
p
(4.60)
dρ 2 dρ
90 Power Loss and Heating Effects

The two independent solutions are [3]

jp ρ ) + BK0 ( jp ρ )
p p
Jz = AI0 ( (4.61)

where I0 (x) is the modified Bessel function of the first kind of order zero and K0 (x) is the
modified Bessel function of the second kind of order zero1 . Since ρ = 0 is a solution of
Eq. (4.61) but K0 (0) = ∞, the constant B must be zero to satisfy the solution at ρ = 0. So
the actual solution is

jp ρ )
p
Jz = AI0 ( (4.62)

The constant A is evaluated in terms of current density at the surface(ρ = ρ0 ), assuming

current density at surface to be σ E0 = J0 . Then the above equation becomes
√
I0 ( jp ρ )
Jz = J0 √ (4.63)
I0 ( jp ρ0 )
√ √
Or writing in terms of skin depth δ = 2/ p
√ √
I0 ( j 2ρ /δ )
Jz = J0 √ √ (4.64)
I0 ( j 2ρ0 /δ )

Separating the complex Bessel function into real and imaginary parts using the definition
that
p
ber(x) = Re[I0 (x j)] (4.65)
p
bei(x) = Im[I0 (x j)] (4.66)
p
I0 (x j) = ber0 (x) + jbei0 (x) (4.67)

Using the above definition, Eq (4.64) becomes

√ √
ber0 ( 2 ρ /δ ) + jbei0 ( 2 ρ /δ )
Jz = J0 √ √ (4.68)
ber0 ( 2 ρ0 /δ ) + jbei0 ( 2 ρ0 /δ )

Substituting ∆ = ρ /δ and ∆0 = ρ0 /δ , we can express the current density in a solid wire

of radius ρ0 in terms of instantaneous quantities,

J = Re[Je jω t ] = Re[Jze jω t ] (4.69)

1 Bessel functions are explained in some detail in Appendix A.6
4.5 Round conductors 91

J = |Jz|cos(ω t + 6 Jz ) (4.70)
" √ √ #1
ber20 ( 2 ∆) + bei20 ( 2 ∆)
2

J= √ √ J0 cos(ω t + θ ) (4.71)
ber20 ( 2 ∆0 ) + bei20 ( 2 ∆0 )
where θ is
√ √ √ √
ber0 ( 2 ∆0 )bei0 ( 2 ∆) − ber0 ( 2 ∆)bei0 ( 2 ∆0 )
θ = tan −1
√ √ √ √ (4.72)
ber0 ( 2 ∆0 )ber0 ( 2 ∆) + bei0 ( 2 ∆)bei0 ( 2 ∆0 )

Power dissipation

From Ampere’s law, the relation between current flowing in a round conductor I and
magnetic flux density at the surface of the conductor B is given by

µI
B= âφ (4.73)
2πρ 0

We can find the relation between current I and current density Jz (Eq 4.63) using Maxwell’s
equations.

5 × E = − jω B (4.74)

1
5 ×J = − jω B (4.75)
σ
Using the definition of Jz from Eq (4.63), we can evaluate Eq (4.75) at ρ = ρ0

∂ Jz
 
1
jω Bρ0 = (4.76)
σ ∂ρ ρ0

√
1 p I00 ( jp ρ0 )
jω Bρ0 = J0 jp √ (4.77)
σ I0 ( jpρ0 )
where I00 ( jρ ) represents the differential of I0 ( jρ ). Substituting the above equation in Eq.
(4.73)
√
2πρ 0 jp I00 (γρ 0 )
I= J0 (4.78)
µ jω σ I0 (γρ 0 )

Since

p = ω µσ (4.79)
92 Power Loss and Heating Effects

√
2πρ 0 I00 ( jp ρ0 )
I = √ J0 √ (4.80)
p j I0 ( jp ρ0 )
The average power dissipated per unit length of the wire is

1 |Jz|2 1 Jz Jz∗
dPav = 2πρ d ρ = 2πρ d ρ (4.81)
2 σ 2 σ

where Jz∗ is the conjugate of Jz . Conjugate of

p ∗  1 + j ∗ 1 − j p
j = √ = √ = −j j (4.82)
2 2

and

I0∗ ( jp ρ ) = I0 ( − jpρ ) = I0 (− j jp ρ )
p p p
(4.83)

So the conjugate of current density Jz∗ becomes

√
I0 (− j jp ρ )
Jz∗ = J0 √ (4.84)
I0 (− j jp ρ0 )
Z ρ0
Pav = dPav (4.85)
0
Z ρ0
π J02
jp ρ )I0 (− j jp ρ )ρ d ρ
p p
Pav = √ √ I0 ( (4.86)
σ I0 ( jp ρ0 )I0 (− j jp ρ0 ) 0

The result written in terms of ber and bei functions is

√ √ √ √
J02 πρ 0 ber0 ( pρ0 )bei00 ( pρ0 ) − ber00 ( pρ0 )bei0 ( pρ0 )
Pav = √ √ √ (4.87)
pσ ber20 ( pρ0 ) + bei20 ( pρ0 )

From Eq (4.78) we can get the rms current Irms

√ 0 √
2 2π 2 ρ02 2 I00 ( jp ρ0 )I0∗ ( jp ρ0 )
Irms = J0 √ √ (4.88)
p I0 ( jp ρ0 )I0∗( jp ρ0 )
0 0
where I0∗ (x) is the conjugate of I0 (x) , I0∗(x) is conjugate of I0 (x) and p = ω µσ .
√ √
2 2π 2 ρ02 2 (ber00 ( pρ0 ))2 + (bei00 ( pρ0 ))2
Irms = J0 √ √ (4.89)
p ber20 ( pρ0 ) + bei20 ( pρ0 )
4.5 Round conductors 93

AC resistance

If Rdc is the dc resistance per unit length, i.e

1
Rdc = (4.90)
σ πρ 02

then the ac resistance Rskin can be expressed in terms of dc resistance as

Rskin 2
Rdc × Irms = Pav (4.91)
Rdc

Pav
Rskin = 2
Rdc (4.92)
Irms Rdc
Pav
Rskin = 2
πρ 02 σ Rdc (4.93)
Irms
From Eqs. (4.93), (4.89) and (4.87) the AC resistance representing skin effect can be
calculated.
√ √ √ √ √
ρ0 p ber( pρ0 )bei0 ( pρ0 ) − ber0 ( pρ0 )bei( pρ0 )
Rskin = Rdc √ √ (4.94)
2 (ber0 ( pρ0 ))2 + (bei0 ( pρ0 ))2

where ρ0 is the radius of the cylindrical conductor.

4.5.3 Proximity effect loss

The current distribution in a round cylindrical conductor subjected to an external homo-

geneous magnetic field can be found from the magnetic vector potential. The magnetic
vector potential in terms of current density J is2

52 A − εµω 2 A = −µ J (4.95)

The net current density J includes conduction current density and displacement current
density. The term ω 2 ε A refers to the displacement current, while 52 A/µ refers to the
conduction current. We can ignore the displacement current since it is very insignificant
in conductors. So Eq. (4.95) becomes

52 A
− =J (4.96)
µ
2
This equation is derived in Appendix A.7
94 Power Loss and Heating Effects

The conduction current can be expressed in terms of electric field

J = σE (4.97)

where E, in terms of potentials A and Φ is

E = − 5 Φ − jω A (4.98)

From Eq. (4.96), (4.97) and (4.98)

52 A
= σ 5 Φ + jω σA (4.99)
µ

In the above case of the conductor subjected to an external magnetic field H0 , there is no
applied electric field, hence no source electric potential.

σ 5Φ = 0 (4.100)

So the magnetic vector potential equation reduces to

52 A
= jω σA (4.101)
µ

In cylindrical coordinates,

∂ Az 1 ∂ Az 1 ∂ 2 Az
+ + 2 = γ 2 Az (4.102)
∂ρ 2 ρ ∂ρ ρ ∂φ 2

After simplification [7], the vector potential inside the cylinder is

√
4µ0 H0 δ J1 ( j3/2 2∆)
Az = √ √ sin φ (4.103)
j3/2 2 F( j3/2 2∆0 )

where J1 (x) is the Bessel function of the first kind of order one, and F(x) is the regular
Coulomb wave function. The current density in terms of magnetic vector potential is

Jz = − jω σAz (4.104)
√ √
4µ0 H0 j3/2 2 J1 ( j3/2 2∆)
Jz = √ sin φ (4.105)
δ F( j3/2 2∆0 )
4.5 Round conductors 95

The eddy current losses per unit length of the cylinder is

Z ρ0 Z 2π
1
Pp = |Jz |2 ρ d ρ d φ (4.106)
2σ 0 0

Substituting Eq. (4.105) in Eq. (4.106)

√ √ √ √
2πγ ber2 ( pρ0 )ber0 ( pρ0 ) + bei2 ( pρ0 )bei0 ( pρ0 ) 2
Pp = − √ √ H0 (4.107)
σ (ber( pρ0 ))2 + (bei( pρ0 ))2

As in the case of skin effect resistance, the proximity effect ac resistance of the mth layer
of a multilayer round wire winding can be derived.
√  √ 0 √ √ 0 √
ρ0 p 2 ber2 ( pρ0 )ber ( pρ0 ) + bei2 ( pρ0 )bei ( pρ0 )

R pr(m) = Rdc(m) −2π (2m − 1) √ √
2 (ber( pρ0 ))2 + (bei( pρ0 ))2
(4.108)

where Rdc(m) is the dc resistance of the mth layer.

AC resistance

The final analytical expression for the ac resistance of the mth layer of a multilayer round
wire winding is
√  √ √ √ √
ρ0 p ber( pρ0 )bei0 ( pρ0 ) − ber0 ( pρ0 )bei( pρ0 )
Rac(m) = Rdc(m) √ √
2 (ber0 ( pρ0 ))2 + (bei0 ( pρ0 ))2
√ 0 √ √ 0 √ (4.109)
2 ber2 ( pρ0 )ber ( pρ0 ) + bei2 ( pρ0 )bei ( pρ0 )

−2π (2m − 1) √ √
(ber( pρ0 ))2 + (bei( pρ0 ))2

The Rac /Rdc ratio is calculated for a Ferrite core inductor with round wire winding
and an Amorphous core inductor with foil winding. The relevant parameters used for
calculation is detailed in Table 4.1. Figures 4.7–4.8 illustrate the Rac /Rdc curves for
foil and round wire winding. We can conclude from both the figures that the proximity
effect loss component is the dominant loss at frequencies close to the switching frequency.
Another observation is that the proximity effect loss becomes dominant at much lower
frequencies in round wire winding compared to foil winding.
96 Power Loss and Heating Effects

Round wire Foil

Dia / Thickness (mm) 2.643 0.127
Skin depth 50Hz (mm) 9.348 9.348
Skin depth 10kHz (mm) 0.661 0.661
Porosity factor 0.854 *
Turns 120 137
Layers 4 137

Table 4.1: Parameter list for Rac /Rdc calculation

Rac/Rdc for Foil conductors

10
Rac/Rdc
Skin Effect
Proximity effect
8

6
Rac/Rdc

0
1 10 100 1000 10000 100000 1e+06
Frequency [Hz]

Figure 4.7: Resistance variation as a function of frequency indicating skin effect, prox-
imity effect and overall Rac /Rdc for foil winding.
4.5 Round conductors 97

Rac/Rdc for Round conductors

10
Rac/Rdc
Skin Effect
Proximity effect
8

6
Rac/Rdc

0
1 10 100 1000 10000 100000 1e+06
Frequency [Hz]

Figure 4.8: Resistance variation as a function of frequency indicating skin effect, prox-
imity effect and overall Rac /Rdc for round wire winding.
98 Power Loss and Heating Effects

4.6 Thermal analysis

Heat is a form of energy that can be transferred from one system to another as a result
of temperature difference. Heat transfer is the science that determines the rate of this
energy transfer. The basic requirement for heat transfer is the presence of a temperature
difference. Transfer of energy is always from a body of higher temperature to a body of
lower temperature, and the transfer stops when both reach the same temperature. There
are three mechanisms of heat transfer -conduction, convection and radiation. Conduction
occurs in a substance when energy is transferred from more energetic particles to adja-
cent less energetic ones. Conduction takes place because to lattice vibrations and free
flow of electrons in solids and collision and diffusion in liquids and gases. Convection is
energy transfer between a solid surface and adjacent liquid or gas that is in motion, and
involves conduction along with fluid motion. Radiation is energy transfer in form of elec-
tromagnetic energy. It is the fastest type of heat transfer and does not require a material
medium.

The principles of heat transfer are used to estimate the surface temperature of the filter
components, especially the inductor. The power loss in the inductor, including the core
loss and copper loss get converted to heat. By knowing the heat transfer rate and the
ambient temperature, we can find out how hot the inductor will become at rated condi-
tions. This is the final step of the inductor design where the entire design procedure is
validated on the basis of expected temperature rise. The crucial temperature constraint
is the temperature rating of the insulation. The insulation used is Nomex which is rated
for 200◦ C, but filter components are designed to operate at temperature of 100◦ C, at an
ambient temperature of 45◦ C.

There are three modes of heat transfer mentioned in the previous paragraph, but in this
specific case, only two are possible. Heat is transfered from the surface of the inductor by
natural convection and/or radiation. In order to solve this heat transfer problem, certain
assumptions are made to obtain the simplest model which still yields reasonable results.

1. The inductor consists of a core material (which is usually a bad conductor of heat)
surrounded by copper winding (good conductor) and insulation (bad conductor).
Also from the previous analysis we have concluded that copper losses far exceed the
core losses. Still, from the thermal analysis point of view, the inductor is considered
to be a uniform body with uniform temperature. This also rules out conduction as
a mode of heat transfer from the interior to the surface, which has to be accounted
for later.
4.6 Thermal analysis 99

2. The surface temperature of the inductor is directly proportional to the power loss
and inversely proportional to the surface area available. The total surface area is the
sum of the surface area of the core and the surface area of the outermost winding
layer. There is some overlap between the two surfaces which will reduce the net
surface area, but this is ignored.

3. The temperature of air immediately surrounding the inductor is assumed to be

45◦ C, while the temperature at a point far away (or room temperature) is assumed
to be 25◦ C.

4. The inductor is designed to be cooled with natural convection currents. The con-
vection heat transfer coefficient hconv is not a property of the fluid. It is an experi-
mentally determined parameter which depends on surface geometry, nature of fluid
motion, properties of the fluid, bulk fluid velocity and other variables which affect
convection. This means that hconv cannot be accurately determined by analytical
methods. Some analytical solutions exist for natural convection, but they are for
specific simple geometries with further simplifying assumptions. The analytical
expressions for a natural convection over a simple vertical plate is used to approx-
imate the heat transfer in the inductor. The accuracy of this assumption is verified
by experimental measurements of temperature.

4.6.1 Radiation

Thermal radiation is a form of radiation emitted by bodies because of their temperature.

Unlike conduction and convection, transfer of heat through radiation does not require an
intervening medium. The maximum rate of radiation that can be emitted by a surface of
area As at a temperature of Ts K is given by the Stefan-Boltzmann’s law

Prad = σ As Ts4 (4.110)

where σ =5.67×10−8 W/m2 K4 is the Stefan-Boltzmann constant.

When a surface of emissivity ε and surface area As at thermodynamic temperature Ts

is completely enclosed by an much larger surface at a thermodynamic temperature Tsurr
separated by a gas (like air) that does not intervene with radiation, the net rate of radiation
heat transfer between these two surfaces is given by

Prad = εσ As (Ts4 − Tsurr

4
) (4.111)
100 Power Loss and Heating Effects

4.6.2 Natural convection

Convection is a form of heat transfer from a solid surface to an adjacent liquid or gas
in motion, and it involves the combined effects of conduction and fluid motion. Bulk
fluid motion enhances the heat transfer between the solid surface and the fluid, but it also
complicates the determination of heat transfer rates. The rate of convection heat transfer
is observed to be proportional to the temperature difference, and is expressed by Newton’s
law of cooling as

Pconv = hconv As (Ts − T∞ ) (4.112)

where hconv is the convection heat transfer coefficient in W/m2 ◦ C, As is the surface area
through which convection heat transfer takes place, Ts is the surface temperature and T∞ is
the temperature of the fluid sufficiently far away from the surface. As mentioned before,
hconv is not a property of the fluid, but is an experimentally determined factor based on a
variety of hard-to-estimate factors.

When the fluid motion occurs only because of the density difference between heated
‘lighter’ air and cooler ‘heavier’ air, it is termed as Natural convection. Although the
mechanism of natural convection is well understood, the complexities of fluid motion
make it very difficult to obtain simple analytical equations for heat transfer. However,
analytical solutions exist for some simple geometries and we are using the analytical
equations for natural convection over a simple vertical plate of length L and temperature
Ts .

4.6.3 Temperature estimation

The quantity of interest is the final steady state surface temperature or Ts . Since the
convection heat transfer coefficient hconv also depends on Ts , to initiate the calculation,
the surface temperature is assumed. This initial surface temperature is used to find hconv .
Since the net heat transfer rate is know, using the calculated hconv , the surface temperature
can be calculated. This process is repeated until the surface temperature converges the
actual value.

0
1. Assume an initial surface temperature Ts .

2. For radiation, Tsurr is assumed to be 25◦ C and for natural convection, T∞ is assumed
to be 45◦ C. Also surface emissivity for radiation is assumed ε =0.6.
4.6 Thermal analysis 101

3. Calculate the film temperature

0
T + T∞
Tf = s (4.113)
2

4. For the film temperature T f , the properties of air at 1 atm pressure are defined. k is
the thermal conductivity of air (W/m K), υ is the kinematic viscosity (m2 /s) and Pr
is the Prantl number.

5. The length of the vertical surface is know as the characteristic length Lc . In case of
the inductor, this will be equal to the height of the inductor.

6. Calculate the Rayleigh number RaL and Nusselt number Nu.

0
gβ (Ts − T∞ )L3c
RaL = Pr (4.114)
υ2
 2
 1

 6

 0.387RaL 
0.825 +
Nu =   (4.115)
  9 ! 278 
0.492 16
 
1+
 
Pr

where g is the acceleration due to gravity(9.8m/s2 ) and β is the volume expansion

coefficient =1/Ts.

7. The convection heat transfer coefficient is given by this equation.

Nu k
hconv = (4.116)
Lc

8. The net heat transfer rate is equal to the total power dissipated.

Pconv + Prad = Ploss (4.117)

Pconv + Prad = hconv As (Ts − T∞ ) + εσ As (Ts4 − Tsurr

4
) (4.118)
00
9. Solving this fourth order polynomial equation gives Ts . Steps 3-8 are repeated till
the surface temperature converges to one number, which will be the actual surface
temperature of the inductor for a power loss of Ploss .
102 Power Loss and Heating Effects

4.6.4 Design example

The example considered is Ferrite core inductor with a single bobbin round wire winding.
The inductor shape is defined in Fig. 3.4 (Chapter 3). The power loss in this inductor for
a dc current of 14 A was measured to be 37 W. The total surface area, including core and
copper surface area is 0.062 m2 . The characteristic length Lc is equal to the height of the
inductor. Following the steps specified in the previous section,

1. Assume surface temperature is 85◦ C.

2. T f =65◦ C. For this T f , the properties of air at 1 atm pressure are

• k=0.028881 W/m K
• υ =1.995×10−5
• Pr=0.7177

3. RaL =7.349×106 , Nu=28.64, hconv =5.67 W/m2 K .

4. Substituting hconv in Eq. (4.118) and solving for Ts , we get Ts =91◦ C.

5. Using this new Ts , we once again recalculate the convection coefficient. The new
hconv =5.643.

6. This time, solving Eq. (4.118) with the new hconv gives the same Ts =91◦ C. Hence,
this is the final surface temperature (the experimentally measured temperature was
88◦ C).

4.7 Summary

The various sources of power loss in an inductor are discussed in detail. Particular atten-
tion is given to winding copper losses in the inductor. The equations describing the copper
loss at various frequencies are derived for both foil winding and round wire winding. The
principles of heat transfer are used to estimate the surface temperature of the inductor.
The steps involved in thermal estimation are explained using a real-world example.
4.8 References 103

4.8 References

Electromagnetics and Mathematics

1. W.R.Smythe, Static and Dynamic Electricity (2nd ed.), McGraw Hill Book Com-
pany, 1950

2. W.H.Hayt Jr, J.A.Buck, Engineering Electromagnetics (6th ed.), Tata McGraw-

Hill, 2001

3. S.Ramo, J.R.Whinnery, T. van Duzer, Fields and Waves in Communication Elec-

tronics (3rd ed.), John Wiley and Sons, 1994

4. C.A.Balanis, Advanced Engineering Electromagnetics (1st ed.), John Wiley and

Sons, 1989

5. A.Jeffrey, Advanced Engineering Mathematics (1st ed.), Academic Press Elsevier

India, 2002

6. M.Abramowitz, I.E.Stegun, Handbook of Mathematical Functions, Dover, 1970

7. J.Lammeraner, M.Stafl, Eddy Currents, Iliffe Books London, 1966

Inductor total power loss

8. M.Bartoli, A.Reatti, M.K.Kazimierczuk, “Minimum copper and core losses power

inductor design,” in Conference Record of the 1996 IEEE Industry Applications
Conference, Thirty-First IAS Annual Meeting,1996, vol.3, pp. 1369-76

9. M.Sippola, R.E.Sepponen, “Accurate prediction of high frequency power trans-

former losses and temperature rise,” IEEE Trans. on Power Electronics, vol. 17,
no. 5, pp. 835-847, Sept 2002

10. H.Skarrie, “Design of Powder core inductors,” Ph.D. dissertation, Dept. of Ind.
Elec. Eng. and Auto., Lund Univ., Lund, Sweden, 2001

Inductor Core loss

11. G.Bertotti, I.Mayergoyz, The Science of Hysteresis, vol. 1,3, Elesevier Inc, 2005
104 Power Loss and Heating Effects

12. A.V.Bossche, V.C.Valchev, Inductors and Transformers for Power Electronics, 1st
ed, CRC Press, 2005

13. W.Roshen, “Ferrite core loss for power magnetic components design,” IEEE Trans.
on Magnetics, vol. 27, no. 6, pp. 4407-4415, Nov 1991

14. W.Roshen, “A practical, accurate and very general core loss model for non-sinusoidal
waveforms,” IEEE Trans. on Power Electronics, vol. 22, no. 1, pp. 30-40, Jan 2007

Inductor Copper loss

15. W.G.Hurley, E.Gath, J.G.Breslin, “Optimizing the AC resistance of multilayer trans-

former windings with arbitrary current waveforms,” IEEE Trans. on Power Elec-
tronics, vol. 15, no. 2, pp. 369-376, March 2000

16. N.H.Kutkut, D.W.Novotny, D.M.Divan et al, “Analysis of winding losses in high

frequency foil wound inductors,” in Conference Record of the 1995 IEEE Industry
Applications Conference Thirtieth IAS Annual Meeting, 1995, vol.1, pp. 859-867

17. W.G.Odendaal, J.A.Ferreira, W.A.Cronje, “Combined numeric and analytic meth-

ods for foil winding design,” in 25th Annual IEEE Power Electronics Specialists
Conference, 1994, vol.2, pp. 843-849

18. M.Bartoli, A.Reatti, M.K.Kazimierczuk. (1995, Dec). Modeling winding losses in

high frequency power inductors. J. Circuits, Systems and Computers, vol.5, no.4,
pp. 607-626

19. A.Reatti, M.K.Kazimierczuk, “Comparison of various methods for calculating the

AC resistance of inductors,” IEEE Trans. on Magnetics, vol. 38, no. 3, pp. 1515-
1518, May 2002

20. J.A.Ferreira, “Improved analytical modeling of conductive losses in magnetic com-

ponents,” IEEE Trans. on Power Electronics, vol. 9, no. 1, pp. 127-131, Jan 1994

21. Xi Nan, C.R.Sullivan, “An improved calculation of proximity-effect loss in high-

frequency windings of round conductors,” 2003 IEEE 34th Annual Power Electron-
ics Specialists Conference, 2003, vol. 2, pp.853-860

22. Xi Nan, C.R.Sullivan, “Simplified high-accuracy calculation of eddy-current loss

in round-wire windings,” 2004 IEEE 35th Annual Power Electronics Specialists
Conference, 2004, vol. 2, pp. 873-879
4.8 References 105

23. P.Wallmeier, “Improved analytical modeling of conductive losses in gapped high-

frequency inductors,” IEEE Trans. on Industry Applications, vol. 37, no. 4, pp.
1045-1054, July/Aug 2001

24. V.A.Niemala, G.R.Skutt, A.M.Urling et al, “Calculating the short circuit impedances
of a multiwinding transformer from its geometry,” 20th Annual IEEE Power Elec-
tronics Specialists Conference - PESC, 1989, vol.2, pp. 607-617

25. C.E.Hawkes, T.G.Wilson, R.C.Wong, “Magnetic-field intensity and current-density

distributions in transformer windings,” 20th Annual IEEE Power Electronics Spe-
cialists Conference - PESC, 1989, vol. 2, pp. 1021-1030

26. M.P.Perry, “On calculating losses in current carrying conductors in an external al-
ternating magnetic field,” IEEE Trans. on Magnetics, vol. 17, no. 5, pp. 2486-2488,
Sept 1981

27. J.P.Vandelac, P.D.Ziogas, “A novel approach for minimizing high-frequency trans-

former copper losses,” IEEE Trans. on Power Electronics, vol. 3, no. 3, pp. 266-
277, July 1988

28. E.E.Kriezis, T.D.Tsibourkis, S.M.Panas, J.A.Tegopoulos, “Eddy Currents: Theory

and Applications,” Proc. of the IEEE, vol. 10, no. 10, Oct 1992

Thermal analysis

29. Y.A.Cengel, Heat and Mass Transfer - A practical approach (3rd ed.) , Tata Mc-
Graw Hill, 2007

30. A.V.Bossche, V.Valchev, J.Melkebeek, “Thermal modeling of E-type magnetic com-

ponents,” Proceedings of the 2002 28th Annual Conference of the IEEE Industrial
Electronics Society, 2002, vol. 2, pp. 1312-1317
Chapter 5

Grid Interactive Operation and Active

Damping

5.1 Introduction

The previous chapters looked at the design of the LCL filter and its components. In this
chapter we will study the grid interactive operation of converters with LCL filters. The
basic operation of such converters have been widely studied [1,2]. The focus here is on
aspects of damping that arise during grid interactive operation of the power converter.
Active rectifiers and active front end converters(AFEC) have been used in drives as well
as distributed generation system and are now becoming more and more popular because
of their ability to control the line side power factor and load voltage at the same time.
These type of converters are connected between load and the grid or utility in order to
supply fine quality of power to the load.

5.2 Active front end converter

The converter consists of a three-phase bridge, a high capacitance on the dc side and a
three-phase filter in the line side. The voltage at the midpoint of a leg or the pole voltage
Vi is pulse width modulated (PWM) in nature. The pole voltage consists of a fundamen-
tal component (at line frequency) besides harmonic components around the switching
frequency of the converter. Being at high frequencies, these harmonic components are
well filtered by the high inductances (L) or some higher order line filter (LCL). Hence the

107
108 Grid Interactive Operation and Active Damping

Figure 5.1: Grid connected operation with the simple L filter

current is near sinusoidal. The fundamental component of Vi controls the flow of real and
reactive power.

It is well known that the active power flows from the leading voltage to the lagging
voltage and the reactive power flows from the higher voltage to the lower voltage. There-
fore, controlling the phase and magnitude of the converter voltage fundamental compo-
nent with respect to the grid voltage can control both active and reactive power. As the
grid voltage leads the converter pole voltage, real power flows from the ac side to the dc
side, while the reactive power flows from the converter to the grid based on the difference
in magnitude of the grid and inverter fundamental voltages.

Apart from control of real and reactive power flow, an FEC should also have a fast
dynamic response. Operation of FEC with the first order L filter is well reported in the
literature but operation of this type active rectifier with LCL filter has now started drawing
attention [2,12].
5.3 Problem of LCL resonance 109

Figure 5.2: Grid connected operation with third order LCL filter

Figure 5.3: PWM waveform applied to the filter

5.3 Problem of LCL resonance

Normally grid impedance reflected back to the converter side is low and in particular, the
resistive component is very less so, if the resonance is excited, the oscillation of that can
continue for a long time, which can make the entire system vulnerable. Actually when
the PWM converter is switched on, the filter (LCL) encounter a sudden pulse at the input
as a result filter starts to oscillate in its cutoff frequency. Here is an attempt to show how
the resonance is excited by PWM converter (AFEC) itself. For that LCL filter is modeled
inside the FPGA based controller and fed from a very narrow single pulse from the same
controller. After exciting from the pulse, filter starts to oscillate at resonant frequency.
110 Grid Interactive Operation and Active Damping

Figure 5.4: Unit impulse excitation applied to filter

Pulse is being generated in FPGA by means of switch de-bouncing logic.

In the grid-connected operation with LCL filter damping is a significant part of design
if we want to utilize the full advantages of higher order filter and modern high perfor-
mance digital controllers. There are two ways to damp the resonance, passive damping
and active damping. The passive approach can damp resonance in all condition but it has
a loss penalty that needs to be traded off with the amount of damping as seen in chapter
(1). The active damping approach can act only when the power converter is switching. It
is seen to be desirable to combine both active and passive approaches in a hybrid manner
so that some minimum level of damping is always present which can be enhanced when
the power converter is switching in a lossless manner.

5.4 Active damping

In passive damping, damping is provided by physical elements like resistors. But this
process is associated with losses. To reduce losses and improve the performance induc-
tors, capacitors are provided along with resistors in passive damping networks as seen in
chapter (1).

In active damping, damping is being provided by means of control algorithm, this

process is not a lossy process so this process is much more attractive. But there is a
limitation of active damping, such as this control technique depends on the switching of
power converter, so this is effective only when power converter is switching. In addition,
5.4 Active damping 111

the switching frequency of the power converter is limited hence the control bandwidth of
the active damping is also limited. There are broadly two methods of active damping can
be thought one is based on traditional PI-controller, and the other is based on generalized
state-space approach. In this chapter we will focus on a method of active damping based
on state-space for arbitrary pole placement.

5.4.1 Active damping based on traditional approach

The traditional approach is based on different current control strategy such as conven-
tional PI-controller based [3] (in rotating frame) combined with lead compensator or a
resonant controller as a main compensator in α − β domain. In these approaches band-
width of the system or settling time cannot be arbitrarily fixed as these based upon main
current controller bandwidth. In other words placement of the closed loop poles is deter-
mined by the current controller design.

5.4.2 Active damping by means of state space method

This approach is more generalized than traditional PI-controller based method because
of flexibility. This method gives us the freedom of arbitrary pole placements or in other
words bandwidth can be independently fixed without depending upon the current con-
troller bandwidth. More over the energy required for damping can be optimized by means
of state-space based method. So, state-space based method offers good stability margin
and robustness to parameter uncertainty in the grid impedance at the same time imple-
mentation is simpler in case of state-space based method. For a stable under-damped
system, its poles should be on left half of s-plane. In our present case providing damping
is equivalent to shifting the closed loop poles in the left half of s-plane. So all we have to
do is to shift the closed loop poles in the left half of s-plane by means of a suitable gain
matrix. Shifting of the poles can be determined by required settling time of the closed
loop system. Along with the damping, the transient response can also be improved for
the different states.

5.4.3 Filter modelling in state space

There are two inputs to the system in Fig. 5.3: PWM output of the power converter
and grid voltage. First one is the active input that can be controlled and second one is the
112 Grid Interactive Operation and Active Damping

uncontrolled disturbance input. The states of the system are elected to be the two inductor
currents and one capacitor voltage.

dVc 1 −1
        
dt 0 C C Vc 0 0
diL1 −1
= 0 0   iL1  +  L11  Uinv +  0  U1 (5.1)
        
 dt L1
diL2 1 −1
dt L2 0 0 iL2 0 L2

In more compact form,

ẋLCL = ALCL xLCL + B1 uinv + B2Ug (5.2)

y = CLCL xLCL (5.3)

where
1 −1
 
0 C C
−1
ALCL =  0 0  (5.4)
 
L1
1
L2 0 0

 
0
B1 = 
 1 
(5.5)
L1 
0

 
0
B2 =  0  (5.6)
 
−1
L2

!
1 0 0
CLCL = (5.7)
0 1 0

Now position of the poles are on imaginary axis hence the system is oscillatory and
5.4 Active damping 113

Figure 5.5: Typical scheme for active damping

Figure 5.6: Pole placement to LHS of s-plane

highly sensitive to outside disturbances.

5.4.4 Pole placement of the system

Typical scheme (in brief) for active damping control can be visualized by Fig.5.5. PWM
delay and the digital controller delays correspond to phase errors in the requirements of
the active damping control.

As mentioned earlier that poles of the closed loop system is on imaginary axis for
un-damped system. Now the poles can be moved to the left half of the s-plane properly
114 Grid Interactive Operation and Active Damping

choosing the BW or settling time of the entire system.

Let the imaginary axis pole is to be shifted to ξω r ± jωr 1 − ξ 2 . Where ξ is the

p

damping factor of the system. We need to select this damping factor for designing this
system. ξ is taken as 0.6 in this case to provide sufficient damping. For assigning above
pole in the system we use control law,

Uinv = −KxLCL (5.8)

or in another form,

Uinv = −K1Vc − K2 iL1 − K3 iL2 (5.9)

The task is to find the gain matrix for the system. The system is perunitized for this
purpose.

5.4.5 Per unitization

For 10kVA inverter and 440V grid voltage we can per unitize the system as below. After
perunitization, L1 = L2 = 3mH and C = 16 mF become L1 = L1 = 0.05pu and C = 0.09pu
Now the required pole placement is at −12.25 ± j16.32. Hence the required gain matrix
is
 
K= −0.0005 1.5 −1.5 (5.10)

The complete system model can be written in state-space as

ẋLCL = (ALCL − B1 K)xLCL + B1 r (5.11)

Where r is reference input, which is determined by the overall control This gives the
damping loop description, which is based on statespace based method.

5.4.6 Physical realization of active damping

The concepts of active damping can be realized from the equation 5.11. After splitting
the state-space form we get,

dVc
C = iL1 − iL2 (5.12)
dt
5.4 Active damping 115

Figure 5.7: Active damping by weightage capacitive current feedback

Figure 5.8: Approximate circuit representation of active damping

diL1
L1 = −(1 + K1 )Vc − K2 iL1 − K3 iL2 +Uinv (5.13)
dt
di2
L2 = Vc −Ug (5.14)
dt
So if we try to synthesise the circuit form of the above equation then it can be shown to be
as in Fig. 5.8. Rv is the series and R p is the parallel virtual resistance and depends on gain
matrix parameters k2 and k3 from equation (5.11). From circuit representation it is clear
that these two resistances are providing the damping to the LCL resonance even though
these resistances do not exit physically. These occur just because of control action and
can be used to damp the resonance. These are treated as virtual resistances, [5]. It can be
116 Grid Interactive Operation and Active Damping

Figure 5.9: Comparison of different damping factors in active damping

seen from the following transfer functions

iL1 1 + s2 L2C
= (5.15)
Uinv sL1 L2C(s2 + s LK + w2r )
1

and

iL2 1
= (5.16)
Uinv sL1 L2C(s + s Lk + w2r )
2
1

k
that the damping factor D = 2L1 wr is proportional to k. Where k = k1 = −k2 .

5.4.7 Active damping loop realization

Fig. 5.10 shows the general practical approach of active damping loop. It consists of
three feedbacks with two inductor currents and one capacitor voltage. It is shown that
there is no need of feeding backs the capacitor voltage for arbitrary pole placements in
the previous section.
5.5 Control of the inverter in grid-interactive mode 117

Figure 5.10: State-space based active damping loop)

Figure 5.11: Comparison of virtual resistance based damping and actual resistance based
damping

5.5 Control of the inverter in grid-interactive mode

For the control of the inverter in grid-interactive mode, dq-based control strategy [1] is
adopted. The main difference between the control of LCL filter based system with that
of an L filter is the addition of the active damping loop. The number of sensors are also
more in case of LCL filter. The control scheme consists of inner ac current controllers,
dc voltage controller and the damping controller. The at last state-space based damping
loop is in between the inner ac current controller and the PWM modulator.
118 Grid Interactive Operation and Active Damping

5.5.1 Model for control design

For control design, the grid is modelled as an ideal sinusoidal three phase voltage source
without line impedances, although, in reality there are line impedances and distortions
like line harmonics and unbalances. The space notation is used. Three phase values are
transformed into the dq-reference frame that rotates synchronously with the line voltage
space vector. From control point of view it is advantageous to control dc values since PI
controller can achieve reference tracking without steady state errors.

Modeling of the LCL filter in the dq-reference frame without frequency dependence
of the inductances is performed here. The differential equations are written in space

Figure 5.12: Active rectifier with LCL filter

Figure 5.13: Grid connection with LCL filter

vector domain:

diL1
L1 = Vc −Uinv (5.17)
dt

diL2
L2 = Ug −Vc (5.18)
dt
5.5 Control of the inverter in grid-interactive mode 119

dVc
C = iL2 − iL1 (5.19)
dt
Here for simplicity of the analysis the parasitic resistance of the inductors is neglected
and at the same time ESR of the capacitance is also being neglected. After transforming
to dq-domain we get the differential equations:

diL1q
L1 = Vcq −Uinvq − wL1 iL1d (5.20)
dt

diL1d
L1 = Vcd −Uinvd + wL1 iL1q (5.21)
dt
diL2q
L2 = Ugq −Vcq − wL2 iL2d (5.22)
dt
diL2d
L2 = Ugd −Vcd + wL2 iL2d (5.23)
dt
dVcq
C = iL2q − iL1q (5.24)
dt
dVcd
C = iL2d − iL1d (5.25)
dt
If we include the dynamics of the DC-bus voltage of the power converter we get,

dVdc 3 iL2qUgq
Cdc = idc − iload = − iload (5.26)
dt 2 Vdc

The control will contain simple decoupling terms in order to decouple the d and q axis
current dynamics. No perfect dynamic decoupling can be achieved due to delays in the
loop and filter resonance.

5.5.2 Overview of control loop consisting of three states of system

Figure 5.14: Conventional three loop control strategy for LCL filter

A traditional approach to control design is to assign a controller for each state as

shown in Fig. 5.14. For this type of filter the capacitor voltage can be indirectly controlled
and so there is no need of capacitor voltage controller for LCL filter. The iL2 − iL1 = ic
120 Grid Interactive Operation and Active Damping

and the Vc = C1 iC dt, hence if we can control iL1 and iL2 separately then that itself controls
the ic followed by the Vc . The control loop may be reduced to following fashion as shown
in Fig. 5.15. Here the output of the line side current controller becomes the reference of

Figure 5.15: Two loop control strategy for LCL filter

converter side current. The line side current and converter side current are almost equal
in magnitude and phase in fundamental, as capacitor size is limited in LCL filter because
of the low reactive power burden. Hence, further more simplification is possible.

The converter side current controller can also be omitted and only line side current
controller is fair enough to control the current. The output of line side current controller
will become inverter input reference. Single grid current loop controller is not sufficient

Figure 5.16: Single loop control strategy for LCL filter

for stability of the overall system. The resonance of the filter can make the system un-
stable as here we are only concentrating on the fundamental current where LCL filter has
significant amount of resonance frequency super imposed over the fundamental. So we
need to consider the resonance carefully. Higher-level control [10] loops are required to
provide fast dynamic compensation for the system disturbances and improve stability.

5.5.3 Current control strategy

Conventional PI controller [7] in the innermost loops is not selected from point of view
of speed and complexity. In order to control the resonance in the filter inner most current
loop should be very fast, ideally instantaneous. For this type of filters inner most current
loop iL1 can be designed by a proportional controller. This controller or this inner loop
makes the system dynamic response very good by limiting the unwanted resonance in
5.5 Control of the inverter in grid-interactive mode 121

Figure 5.17: Modified two loop control strategy for LCL filter

the system or in other words it shifts the closed loop poles in LHS of s-plane. The total
current control loop structure becomes as shown in Fig. 5.18 where outer loop is line side
current and inner loop is the converter side current and converter current feedback is used
for stability [7] for damping oscillation.

Figure 5.18: Two-loop control strategy for LCL filter

We can see the transfer function of the closed loop current controller with the inner
loop, it is a 4th order system. In the outer loop PI-controller used as usual and proportional
controller is used in inner loop. The closed loop transfer function is given below, is used
to analyse the stability.
i∗L2 (K p Kc KPW M )s + K p Kc KPW M /Tc
= (5.27)
iL2 (L1 L2C)s4 + (Kc KPW M L2C)s3 + (L1 + L2 )s2 + (K p Kc KPW M + Kc KPW M )s + K p Kc KPW M /Tc

5.5.4 Analysis of controller performance

A Analysis of the outer loop: The outer loop is a PI-controller, where the value of K p
should be quite high value to track the reference but at the same time K p should not
be as large as wish, which can be seen from the following analysis. As it grows
bigger, the poles will shift towards the right side of s-plane.

B Analysis of the inner loop: The inner loop basically improves the stability of the
system and increases the robustness. In other words more important role of the
122 Grid Interactive Operation and Active Damping

i∗L2
Figure 5.19: Root locus of iL2 with K p = 0.5

i∗L2
Figure 5.20: Root locus of iL2 with K p = 1

inner loop is to damp the resonance peak but at the same time very high of Kc can
make system unstable also. So, the value of Kc has to be limited and we cannot
depend the value of Kc to damp the oscillation.
5.5 Control of the inverter in grid-interactive mode 123

i∗L2
Figure 5.21: Root locus of iL2 with K p = 2

i∗L2
Figure 5.22: Bode-plot of iL2 with different values of Kc

5.5.5 Inclusion of innermost state-space based damping loop

As mentioned earlier that in the converter current loops the value of Kc is limited from
the point of view of stability. So, when high damped system is desired this method of
damping described in the previous section is not preferable.

The state-space based method offer more flexibility of choosing the controller param-
124 Grid Interactive Operation and Active Damping

eter and at the same time robustness. The total current control loop with the damping loop
is shown in Fig. 5.23. Here the value of damping can be decided as given by equation
(5.16)

Figure 5.23: Current control with State-space based damping loop

5.5.6 Control in grid-parallel mode with LCL filter

In grid connected mode, load is connected across the DC-bus which is to be supplied
from the grid with good power quality (FEC mode)[16]. So, naturally the load voltage
has to be maintaining constant. Here DC-link voltage controller is must for this kind of
operation.

5.5.7 Sensorless operation

For the control of LCL filter based system several loops cascaded. Naturally while im-
plementation in practice, it needs quite a few sensors. These sensors, specially the LEM
current sensors are costly. So, in the LCL filter based system minimizing the number of
sensors is desirable. The way out is to estimate the corresponding quantities like voltages
or currents. There are two ways to eliminate sensors, one way is to run a parallel pro-
cess in the controller and then calculate quantities and use for control, the second way to
design the reduced order observer to estimate the states.
5.6 Experimental set-up 125

Figure 5.24: Vector control in grid parallel mode

5.6 Experimental set-up

Experimental setup consists of 10KVA power converter, LCL filter interfaced with grid,
a diode-bridge rectifier, FPGA based controller and a PC programming for FPGA. Fig.
5.25 shows the main components of the experimental setup.

1. Diode Bridge Rectifier

This is required for pre-charging the DC-bus of the power converter. This is fed
from a 3-phase auto-transformer connected to grid. The autotransformer ended by
a large capacitance at the output to make a pure DC. Autotransformer is adjusted
to change the DC-link voltage appropriately. In the grid parallel mode of operation
this precharging circuit is automatically disconnected as we boost up the DC-link
voltage above the pre-charging voltage by the DC-link voltage control.

2. Pre-charging Auto-Transformer
An auto transformer is placed before the grid and it is used to feeding the inverter
DC-link voltage. This facility is used to vary the grid voltage also (we can operate
the system at any grid voltage). In a practical power converter appropriately sized
precharging resistor can be used.
126 Grid Interactive Operation and Active Damping

Figure 5.25: Complete Hardware Set-Up

3. IGBT based Inverter

The power devices are used in the inverter are IGBTs . There are three legs in
the inverter with two IGBTs (one module) in each leg. IGBTs are mounted on
heat sink and are connected to the DC bus voltage via DC bus bar. The additional
components of the IGBT inverter are protection and delay card, gate drive card,
front panel annunciation card and voltage and current sensing cards.

4. FPGA Controller
The digital platform consists of FPGA device and other devices interfaced to FPGA.
The devices interfaced include configuration device (EEPROM), ADC and DAC;
dedicated I/O pins are also provided. The FPGA has logic elements arranged in
rows and columns. Each logic elements has certain hardware resources, which will
be utilized to realize the user logic. The vertical and horizontal interconnects of
varying speeds provide signal interconnects to implement the custom logic. The
choice of an FPGA device for a given application is based on the size required (no.
of logic elements), clock speed and number of I/O pins. ALTERA EP1CQ240C8 is
found to be suitable for the given platform. The resources available in this device
are listed in Table 5.1. The main components of the FPGA controller board is as
follows:
5.6 Experimental set-up 127

• Configuration Device: The configuration device is an EEPROM (EPCS41N)

which is used to a PC through a parallel port or USB using Byte blaster II or
USB blaster cable. The digital design for the implementation of the proposed
scheme is done using Quartus-II tool (Alteras design tool for FPGA) and the
output file after compilation is downloaded to EEPROM through Byte blaster
II or USB blaster cable.
• ALTERA FPGA device data

Figure 5.26: Block diagram of the FPGA Board

• Analog to Digital converter (ADC): ADC on the board, AD7864AS-1 of ana-

log devices, is used to convert the analog input signals from the system to
digital signals which are used for further processing. This MQFP packaged,
12 bit, 44-pins simultaneous ADC has 4 channels with a conversion time of
1.6 µ s per channel. There are four such ADCs on the board and hence can
take up to 16-analog input.
• Digital to Analog converter (DAC): DAC on the board, DAC-7625U, is used
to output the digital variables in the controller in analog form. The DAC is
TTL devices working with +5V and -5V power supply. This is a 12-bit, 28
pins DAC of TEXAS has 4 channels with conversion time of 10 µ s.
128 Grid Interactive Operation and Active Damping

Part Number EP1C12Q240C8

Manufacturer Altera
Number of pins 240
Number of I/O pins 173
Total internal memory bits 2,39,616
Package PQFP
Number of logic elements 12,000
Number of PLLs 2
Maximum clock frequency using PLL 275MHz

Table 5.1: ALTERA FPGA device data

• Digital I/Os: Dedicated digital I/Os are necessary to interface to ADC, DAC
etc which are present on the board. Apart from that 56 I/O pins are provided
for the user to interface application specific hardware.

5.7 Experimental results

5.7.1 Implementation stages

The implementation can be done divided into two stages

(a) The first stage involved estimation and calculation of different quantities such as sine,
cosine tables, PWM switching patterns, etc.
(b) The second stage involved controller design and realization of the overall system in
the FPGA controller.

The experiments carried out focussed on the active damping performance of the in-
verter with LCL filter. The ac voltage was limited to less than 200V due to limitations of
the voltage sensor card. The results of the experiments are given in Figs 5.27 - 5.36.
The controller was tested to check the ability of the system to function as a PWM rectifier.
Figs. 5.27 and 5.28 show the transient ability of the grid connected inverter to regulate
DC bus voltage in response to a change in reference command. Fig. 5.29 shown that
there can be a significant current at the filter resonant frequency without any damping.
The level of damping can be improved based on the state feedback gains. Appropriate
5.7 Experimental results 129

gains can be set based on the desired level of damping as shown in Figs. 5.30 and 5.31.
The transient response of the power converter ac current command shows that resonate
oscillations are suppressed even under transient as shown in Fig. 5.32. It can also be
seen that the active damping acts rapidly in a couple of milliseconds after enabling of the
damping controller in Fig. 5.34. The rapid response of the active damping loop indicates
that it can be designed such that it does not interfere with the current control and voltage
control loop that are used in active front end converters.

5.7.2 Summary

Its is seen that damping is an important consideration in higher order filter design. A
simple state feedback gain based damping controller has been analysed. The performance
of such a controller is seen to adequate for a wide range of operating conditions.

Figure 5.27: DC-Bus control test (voltage rise)

130 Grid Interactive Operation and Active Damping

Figure 5.28: DC-Bus control test (voltage falling)

Figure 5.29: Distorted current from grid (full of resonance)

5.7 Experimental results 131

Figure 5.30: Less distorted current grid (state weightage K=10)

Figure 5.31: Smooth current from grid (state weightage 25) and its FFT
132 Grid Interactive Operation and Active Damping

Figure 5.32: Grid side current dynamics when sudden change in load.

Figure 5.33: Grid side current dynamics when sudden change in load
5.7 Experimental results 133

Figure 5.34: Line side current when active damping is being enabled mid-way

Figure 5.35: Active damping loop enabled mid-way with BW=1.2kHz

134 Grid Interactive Operation and Active Damping

Figure 5.36: Distortion in utility voltage and smoothing out by active damping
5.8 References 135

5.8 References

1. V. Blasko and V. Kaura, A new mathematical model and control of a three-phase

AC-DC voltage source converter, IEEE Trans. Power Electron., vol. 12, pp. 116123,
Jan. 1997.

2. M. Liserre, F. Blaabjerg, and S. Hansen, Design and control of an LCL-filter based

active rectifier, in Proc. IAS01, Sept./Oct. 2001, pp. 299307.

3. V. Blasko and V. Kaura, A novel control to actively damp resonance in input lc filter
of a three-phase voltage source converter, IEEE Trans. Ind. Applicat., vol. 33, pp.
542550, Mar.Apr. 1997.

4. R. W. Erickson and D. Maksimovic, Fundamentals of Power Electronics. Boston,

MA: Kluwer, 2001.

5. M. Liserre, A. DellAquila, and F. Blaabjerg, Stability improvements of an LCL-

filter based three-phase active rectifier, in Proc. PESC02, June 2002, pp. 11951201.

6. P. A. Dahono, A control method to damp oscillation in the input LC filter of AC-DC

PWM converters, in Proc. PESC02, June 2002, pp. 16301635.

7. E. Twining and D. G. Holmes, Grid current regulation of a three-phase voltage

source inverter with an LCL input filter, in Proc. PESC02, June 2002, pp. 11891194.

8. N. Abdel-Rahim and J. E. Quaicoe, Modeling and analysis of a feedback control

strategy for three-phase voltage-source utility interface systems, in Proc. 29th IAS
Annu. Meeting, 1994, pp. 895902.

9. M. Lindgren and J. Svensson, Control of a voltage-source converter connected to

the grid through an LCL-filter-application to active filtering, in Proc. Power Elec-
tron. Spec. Conf. (PESC98), Fukuoka, Japan, 1998.

10. Poh Chiang Loh. Analysis of Multiloop Control Strategies for LC/CL/LCL-Filtered
Voltage-Source and Current-Source Inverters, IEEE Trans on Industry Applica-
tions, 2005, 2(41):644-654.

11. R. Teodorescu, F.Blaabjerg, U. Borup and M.Liserre. A New Control Structure

for Grid-Connected LCL PV Inverters with Zero Steady- State Error and Selective
Harmonic Compensation, APEC, 2004, vol.1: 580-586
136 Grid Interactive Operation and Active Damping

12. Hamid R. Karshenas and Hadi Saghafi. Basic Critia in Designing LCL Filters for
Grid Connected Converters, IEEE ISIE, 2006: 1996-2000

13. F.A. Magueed and J. Svensson, Control of VSC connected to the grid through LCL-
filter to achieve balanced currents, in Proc. IEEE Industry Applications Society
Annual Meeting 2005, vol. 1, pp. 572-8.

14. M. Liserre, A. DellAquila, and F. Blaabjerg, Genetic algorithm-based design of the

active damping for an LCL-filter three-phase active rectifier, IEEE Transactions on
Power Electronics, vol. 19, no. 1, pp. 76-86, 2004.

15. M. Prodanovic and T.C. Green, Control and filter design of three-phase inverters
for high power quality grid connection, IEEE Transactions on Power Electronics,
vol. 18, no. 1, pp. 373-80, 2003.

16. Vector control of three-phase AC/DC front-end converter - J S SIVA PRASAD,

TUSHAR BHAVSAR, RAJESH GHOSH and G NARAYANAN. Sadhana Vol. 33,
Part 5, October 2008, pp. 591613.

17. Operation of a three phase - Phase Locked Loop system under distorted utility
conditions - Vikram Kaura, Vladimir Blasko. IEEE Transactions 1996.
Chapter 6

Experimental Results and Optimized

LCL Filter Design

6.1 Introduction

This chapter discusses some of the experimental results which were used to verify the
filter design model derived in the previous chapters. All aspects of the design process
were tested and verified by actual experiments to confirm the design assumptions. The
frequency response characteristics of the LCL filter configuration is obtained from a net-
work analyzer. Harmonics are sampled to ensure the output current conforms to the rec-
ommended IEEE current harmonic limits. Special attention is given to verify the power
loss and thermal models. Based on the percent of match between the assumed model and
actual experimental data, new predictions are made to find the most efficient LCL filter
combination which still gives the required harmonic filtering.

6.2 LCL filter parameter ratings

R Y B
L1 (mH) 3.385 3.374 3.349
L2 (mH) 3.439 3.407 3.369
Damping branch C1 =8µ F Cd =8µ F Rd =25 Ω

Table 6.1: LCL filter values for Ferrite core inductors

137
138 Experimental Results and Optimized LCL Filter Design

R Y B
L1 (mH) 5.434 5.292 5.374
L2 (mH) 5.323 5.266 5.323
Damping branch C1 =6µ F Cd =6µ F Rd =25 Ω

Table 6.2: LCL filter values for Amorphous core inductors

R Y B
L1 (mH) 1.737 1.784 1.737
L2 (mH) 1.772 1.772 1.757
Damping branch C1 =10µ F Cd =10µ F Rd =10 Ω

Table 6.3: LCL filter values for Powder core inductors

6.3 Frequency response

The impedence frequency response of the individual L and C components was measured
to evaluate the differential mode parasitic impedances of each filter component. The
differential mode impedence model of all inductors was found to be

1
ZL (s) = (sL + R)|| (6.1)
sC

and the differential mode impedence model of the AC capacitors was

1
ZC (s) = + R + sL (6.2)
sC

It was observed that parasitics of the individual L and C were insignificant at the fre-
quency range of operation of the LCL filter. All the filter components showed reasonable
expected operation in the frequency range of operation (Figures 6.1-6.5).

The frequency response of the LCL filter was measured using an analog network an-
alyzer manufactured by AP Instruments. The network analyzer has a frequency range
from 0.01 Hz to 15 MHz, with a maximum output of 1.77V. Current measurements were
made with Textronix TCP300 AC/DC current probe and amplifier which has a bandwidth
of 120 MHz. All the transfer functions of the LCL filter as detailed in chapter 1 were
measured for different combination of L and C with each measurement in the frequency
range of 10 Hz to 1 MHz with atleast 1000 data points, each point averaged 40 times.
Figures 6.6-6.11 show the actual output of the network analyzer for the transfer func-
6.3 Frequency response 139

tions of ig /vi (vg =0) and vg /vi (ig =0) compared with the simulated frequency response.
The figures show the effect of damping with the Q-factor reducing considerably at the
resonant frequency. Significant deviations in magnitude and phase can be observed in the
frequency response characteristics from the expected ideal characteristics at frequencies
beyond 100kHz. These deviations are caused by the parasitic impedances of the individ-
ual filter components which are dominant at such high frequencies.

120

100
Magnitude [dB]

80

60

40

20
4 5
10 10
Frequency [Hz]

100

50
Angle [deg]

−50

−100
4 5
10 10
Frequency [Hz]

Figure 6.1: Differential mode impedence response 5 kHz to 500 kHz; Ferrite core induc-
tor
140 Experimental Results and Optimized LCL Filter Design

120

100

Magnitude [dB]
80

60

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [Hz]

100

50
Angle [deg]

−50

−100
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [Hz]

Figure 6.2: Differential mode impedence response 10 Hz to 1 MHz; Amorphous core

367S inductor

120

100
Magnitude [dB]

80

60

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [Hz]

100

50
Angle [deg]

−50

−100
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [Hz]

Figure 6.3: Differential mode impedence response 10 Hz to 1 MHz; Amorphous core 630
inductor
6.3 Frequency response 141

120

100

Magnitude [dB]
80

60

40

20

−20
1 2 3 4 5 6 7
10 10 10 10 10 10 10
Frequency [Hz]

100

50
Angle [deg]

−50

−100
1 2 3 4 5 6 7
10 10 10 10 10 10 10
Frequency [Hz]

Figure 6.4: Differential mode impedence response 10 Hz to 1 MHz; Powder core inductor,
foil winding

120

100
Magnitude [dB]

80

60

40

20

−20
1 2 3 4 5 6 7
10 10 10 10 10 10 10
Frequency [Hz]

100

50
Angle [deg]

−50

−100
1 2 3 4 5 6 7
10 10 10 10 10 10 10
Frequency [Hz]

Figure 6.5: Differential mode impedence response 10 Hz to 1 MHz; Powder core inductor,
round wire winding
142 Experimental Results and Optimized LCL Filter Design

50
Sim
Act
0
Magnitude [dB]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure 6.6: LCL filter response ig /vi (vg =0); Ferrite core inductor, no damping

50
Sim
Act
0
Magnitude [dB]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure 6.7: LCL filter response ig /vi (vg =0); Ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω
6.3 Frequency response 143

50
Sim
0 Act

Magnitude [dB]
−50

−100

−150

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure 6.8: LCL filter response ig /vi (vg =0); Amorphous core inductor, no damping

50
Sim
0 Act
Magnitude [dB]

−50

−100

−150

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure 6.9: LCL filter response ig /vi (ig =0); Amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω
144 Experimental Results and Optimized LCL Filter Design

50
Sim
Act
0
Magnitude [dB]
−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure 6.10: LCL filter response ig /vi (vg =0); Powder core inductor, no damping

50
Sim
Act
0
Magnitude [dB]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure 6.11: LCL filter response ig /vi (ig =0); Powder core inductor, C1 =Cd =10µ F,
Rd =10Ω
6.4 Harmonic analysis 145

6.4 Harmonic analysis

IEEE 519-1992 ‘Recommended Practices and Requirements for Harmonic Control in

Electrical Power Systems’ defines distortion limits for both current and voltage in order
to minimize interference between electrical equipment (Table 6.5). It is a system standard
applied at the point of common coupling of all linear and nonlinear loads, and assumes
steady state operation. Most utilities insist that current harmonic limits should be met
at the output terminals of the nonlinear equipment. Hence to measure the effectiveness
of the LCL filter, it is important to measure the output current harmonics at inductor L2
which is connected to the point of common coupling to grid.

Open loop tests were conducted with the 3 phase inverter switching at 10 kHz with
the modulation index adjusted to get full rated current at full dc bus voltage of 600V. This
test enabled us to test the ability of the filter to attenuate the high frequency current ripple
under the worst possible ripple conditions of low modulation index and open loop sine
triangle modulation.

The equation for calculating current Total Harmonic Distortion THD is

q
I22 + I33 + I42 + . . .
IT HD = × 100% (6.3)
I1

The equation for calculating current Total Demand Distortion TDD is

q
I22 + I33 + I42 + . . .
IT DD = × 100% (6.4)
IL

Filter type Modulation Index Inverter side Grid side

Inom (A) Isw (A) Inom (A) Isw (A)
Ferrite 0.165 14.58 1.39 13.71 0.01
Amorphous 0.294 14.34 0.91 14.73 0.02
Powder 0.135 13.72 2.64 14.39 0.05

Table 6.4: Inverter settings for harmonic measurement; Vdc =600V, fsw =10 kHz
146 Experimental Results and Optimized LCL Filter Design

Maximum Harmonic Current Distortion in Percent of IL

Individual Harmonic Order (Odd Harmonics)
ISC /IL <11 11≤h<17 17≤h<23 23≤h<35 35≤h TDD
<20 4.0 2.0 1.5 0.6 0.3 5.0

Table 6.5: Current distortion limits for general distribution systems IEEE 519-1992

L1 =L2 C1 =Cd Rd fres Inom Imax Isw /Imax TDD

mH µF Ω kHz A A % %
3.4 8 25 1 13.71 14.58 0.08 7.7
5.4 6 25 1 15.01 15.45 0.09 1.5
1.7 10 10 1.18 13.97 15.45 0.32 10

Table 6.6: Measured output current harmonics and TDD

Current waveform Inverter side Current Harmonics Inverter side
30 16
14
20
12
10
Current [A]

Current [A]

10
0 8
6
−10
4
−20
2
−30 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)
Current waveform Grid side Current Harmonics Grid side
30 16
14
20
12
10
Current [A]

Current [A]

10
0 8
6
−10
4
−20
2
−30 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)
Current waveform Damping branch (Cd−Rd) Current harmonics Damping branch (Cd−Rd)
0.6 0.6

0.4 0.5

0.2 0.4
Current [A]

Current [A]

0 0.3

−0.2 0.2

−0.4 0.1

−0.6 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)

Figure 6.12: Current waveform and harmonic spectrum; Ferrite core inductor
6.4 Harmonic analysis 147

Current waveform Inverter side Current Harmonics Inverter side

30 16
14
20
12
10
Current [A]

Current [A]
10
0 8
6
−10
4
−20
2
−30 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)
Current waveform Grid side Current Harmonics Grid side
30 16
14
20
12
10
Current [A]

Current [A]

10
0 8
6
−10
4
−20
2
−30 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)
Current waveform Damping branch (Cd−Rd) Current harmonics Damping branch (Cd−Rd)
0.6 0.6

0.4 0.5

0.2 0.4
Current [A]

Current [A]

0 0.3

−0.2 0.2

−0.4 0.1

−0.6 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)

Figure 6.13: Current waveform and harmonic spectrum; Amorphous core inductor
148 Experimental Results and Optimized LCL Filter Design

Current waveform Inverter side Current Harmonics Inverter side

30 16
14
20
12
10
Current [A]

Current [A]
10
0 8
6
−10
4
−20
2
−30 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)
Current waveform Grid side Current Harmonics Grid side
30 16
14
20
12
10
Current [A]

Current [A]

10
0 8
6
−10
4
−20
2
−30 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)
Current waveform Damping branch (Cd−Rd) Current harmonics Damping branch (Cd−Rd)
2 1
1.5
0.8
1
Current [A]

Current [A]

0.5 0.6
0
−0.5 0.4

−1
0.2
−1.5
−2 0
0 0.005 0.01 0.015 0.02 0 50 100 150 200 250
Time [s] Harmonic order (0=dc,1=50Hz...)

Figure 6.14: Current waveform and harmonic spectrum; Powder core inductor
6.5 Power loss 149

6.5 Power loss

The efficiency of the LCL filter was tested under short circuit conditions (Fig. 6.15). The
modulation method used was sine triangle modulation under open loop conditions. Ad-
ditionally the dc bus mid-point was connected to the three phase capacitor star point. The
combination of the low modulation index (to get rated current) and high dc bus voltage
gave the worst case current ripple. Hence the losses in this section represent the highest
possible losses for the LCL filter.

Tables 6.7-6.10 show the comparison between measured and calculated power loss.
The “Designed” column shows the predicted current harmonics and power loss for the
dominant harmonics of fundamental and switching frequency. The expected core loss for
both fundamental and switching harmonic current is also shown as a single number. The
“Actual” column shows the actual measured current harmonics and measured power loss.
Harmonics were calculated from the current waveform sampled by a digital oscilloscope.
Power measurements were made using a three phase six channel Yokogawa WT1600
digital power analyser. The “Expected” column shows the expected power loss -both
copper and core, for the actual current harmonics, which were calculated by using the
measured current harmonics in the power loss equations (Chapter 2). The last row shows
the percent error between the measured (or actual) power loss and expected power loss.
3 phase 10 kVA inverter

+ LCL Filter setup

r r
L1 L2
y
L1
y
Vdc 0 Vi L2
b b y
ii L1 L2
Cd ig
r
b C
y
C1 Cdr
C1 Cdb 1
y r
Rd Rd
_ Rbd

Switchgear Switchgear

Figure 6.15: Power loss test setup

150 Experimental Results and Optimized LCL Filter Design

Designed Actual Expected

A W A A W
DC 0 0 5.445 5.445 2.938
Copper Loss f1 14.58 17.53 13.886 13.886 19.09
SW 1.138 4.336 1.155 1.155 5.363
Core Loss (W) * 0.052 * * *
Total Loss (W) * 21.925 38.6 * 27.391
Error (%) * * * * -29

Table 6.7: Comparison between designed efficiency and actual measured power
loss;Ferrite core inductor-round wire winding

Designed Actual Expected

A W A A W
DC 0 0 0.264 0.264 *
Copper Loss f1 15.45 12.80 13.73 13.73 12.32
SW 0.683 0.097 0.865 0.865 0.191
Core Loss (W) * 3.857 * * 6.379
Total Loss (W) * 16.75 24 * 18.89
Error (%) * * * * -21.3

Table 6.8: Comparison between designed efficiency and actual measured power loss;
Amorphous core inductor-foil winding

Designed Actual Expected

A W A A W
DC 0 0 * * *
Copper Loss f1 15.45 12.18 14.29 14.29 12.71
SW 2.115 0.494 2.741 2.741 1.012
Core Loss (W) * 7.645 * * 12.924
Total Loss (W) * 20.31 35.7 * 26.64
Error (%) * * * * -25.3

Table 6.9: Comparison between designed efficiency and actual measured power loss;
Powder core inductor-foil winding
6.6 Temperature rise 151

Designed Actual Expected

A W A A W
DC 0 0 * * *
Copper Loss f1 15.45 24.03 14.149 14.149 22.95
SW 2.115 8.19 2.517 2.517 13.21
Core Loss (W) * 4.766 * * 12.924
Total Loss (W) * 36.98 35.3 * 47.04
Error (%) * * * * 33.25

Table 6.10: Comparison between designed efficiency and actual measured power loss;
Powder core inductor-round wire winding

6.6 Temperature rise

The thermal model from Chapter 2 was verified through DC temperature tests. All the
inductors were connected in series with a adjustable DC current source. Initially current
was set at the rated current of the inductor. The temperature of individual inductors was
measured using K-type thermocouples embedded inside the winding of the inductor. In
most cases, two thermocouples were used per inductor, one embedded close to the first
turn (“inner”) and the second at the last turn (“outer”). Concurrently, the electrical power
loss in each inductor was accurately measured. When the inductor reached thermal sta-
bility the current setting was reduced to a new lower value. Again the temperature was
tracked till it became constant. This ensured that precise steady state temperature read-
ing was available for different power levels. The experiment was repeated for decreasing
current levels –14 A, 10 A, 7.5 A, 5 A and 2.5 A.
152 Experimental Results and Optimized LCL Filter Design

Measured Expected
Inductor type Power loss Ambient Inductor Ambient Inductor
W ◦C ◦C ◦C ◦C

Ferrite 38.8 30 88 25 91
Amorphous AMCC 367S 13.52 30 57 25 67
Amorphous AMCC 630 9.01 30 50 25 59
Powder foil winding 13.13 29 70 25 69
Powder round wire winding 13.32 29 70 25 72

Table 6.11: Theoretical temperature prediction and actual steady state temperature read-
ings

DC Temperature test
90
Inner
80 Outer
Ambient
Temperature [deg C]

70

60

50

40

30

20
0 1 2 3 4 5 6 7 8 9
Time [hr]
Figure 6.16: DC temperature test; Ferrite core inductor with round wire winding
6.6 Temperature rise 153

DC Temperature test
60
Inner
55 Outer
Temperature [deg C] Ambient
50
45
40
35
30
25
20
0 1 2 3 4 5 6 7 8
Time [hr]
Figure 6.17: DC temperature test; Amorphous core AMCC367S inductor with foil wind-
ing

DC Temperature test
60
Inner
55 Outer
Ambient
Temperature [deg C]

50
45
40
35
30
25
20
0 1 2 3 4 5 6 7 8
Time [hr]
Figure 6.18: DC temperature test; Amorphous core AMCC630 inductor with foil winding
154 Experimental Results and Optimized LCL Filter Design

DC Temperature test
Inductor
70 Ambient
Temperature [deg C]

60

50

40

30

20
0 1 2 3 4 5 6 7 8
Time [hr]
Figure 6.19: DC temperature test; Powder core inductor with foil winding

DC Temperature test
Inductor Bob 1
70 Inductor Bob 2
Ambient
Temperature [deg C]

60

50

40

30

20
0 1 2 3 4 5 6 7 8
Time [hr]
Figure 6.20: DC temperature test; Powder core inductor with round wire winding, two
bobbin design
6.7 Minimum power loss design 155

6.7 Minimum power loss design

As discussed earlier, losses in the inductor are quite significant and any efficiency and
thermal optimization of the LCL filter will have to focus on the inductors L1 and L2
to make a noticeable difference in the efficiency of the overall filter. The cost of the
passive filter components is another area where significant gains can be made by reducing
the size and weight of the individual filter components. At the same time, the IEEE
recommended limits for high frequency current ripple must also be met if the filter is to
be used for a grid connected power converter. For a 10kVA system with a base voltage
of Vbase =239.6V operating at a switching frequency of 10kHz with the dc bus voltage at
861V, the minimum L pu = L1(pu) + L2(pu) is given by Eq. (1.70) (Chapter 1).

1 1
L pu = (6.5)
ωsw(pu)
 
2

i  
 g(pu)  
ωsw(pu)   1 − 2

 vi(pu)   ωres(pu) 


1 1
L pu =    = 0.015
2
(6.6)
0.003  200
200 1 − 
0.898  202 
The resonant frequency is set at 1kHz and voltage harmonic at switching frequency is
assumed to be one-fourth of dc bus voltage.

Hence a minimum L pu of 0.015pu is sufficient to meet IEEE recommendations for

harmonics≥35. The next step is to test if this is also the most efficient rating. The power
loss in each individual component of the LCL filter should be examined to derive the
most efficient filter configuration. The inverter-side inductor L1 is subjected to both the
fundamental load current and the switching frequency ripple current. The switching fre-
quency ripple is attenuated sufficiently at the output of the filter, so that the power loss
in the grid-side inductor will be almost exclusively because of the fundamental current.
The power loss in the damping branch depends not only on the fundmental and switching
frequency voltage ripple, but also the damping resistor Rd and the ratio of Cd /C1 .

The losses in an inductor depend on the type of core material and type of winding
but there are certain trends that are common for all types of inductors. The copper losses
at fundamental frequency directly depend on the number of turns of copper, so it will
increase with higher L pu . The copper losses at switching frequency is more sensitive
to skin effect and proximity effect and will not change linearly with L pu . In particular,
losses in round wire windings are affected by the number of layers of winding. The
156 Experimental Results and Optimized LCL Filter Design

core loss at fundamental frequency for the magnetic materials used in high frequency
operation is insignificant and can be usually ignored. But core loss at switching frequency
is dependent on the flux density due to switching frequency current ripple and can be quite
prominent in certain materials. High current ripple because of low L pu will translate to
higher core loss at switching frequency. Additionally, the total C pu has to be increased
to maintain the same resonant frequency while decreasing L pu . The power loss in the
damping circuit will increase linearly with C pu .

The essence of the previous discussion is that as the L pu is varied (with the minimum
at 0.015 pu), the total power loss of the LCL filter will follow an approximate inverted bell
shaped curve with high loss at low L pu because of the higher current ripple and high loss
again at high L pu because of fundamental current. But there is an minimum point in this
curve which will give the highest efficiency and the lowest total loss. At the same time,
as this L pu is greater than the minimum required, it will satisfy the IEEE requirements for
filtering. The L pu at which this minimum loss occurs will not be affected by the losses in
the damping circuit, since this loss varies linearly. The subsequent figures investigate this
optimum L pu for inductors designed with different core materials and different windings.
All the data points are viable designs with the least possible number of turns for each value
of inductance and all designs ensure that flux density in the core is within the saturation
limits. The total capacitance C pu is adjusted to keep the resonant frequency at 1kHz in
every case.

The minimum loss designs in Figures 6.21–6.42 are optimized considering only the
LCL filter efficiency. But if we consider the entire power converter, the LCL filter is
only one part of the entire converter hardware, and choice of most efficient L pu can have
implications for the overall efficiency of the power converter. The LCL filter is designed
for high power voltage source converters switching at a minimum of 10kHz. IGBTs are
the most suitable switching devices for such applications. The switching loss in IGBTs is
approximately unaffected by the high frequency current ripple assuming same turn on and
turn off loss. The conduction loss depends on the on-state resistance RDS which actually
varies with the current, which means it will be affected by the current ripple. But since
RDS of most common IGBT devices is less the dc resistance of the total inductance of the
LCL filter, the minimum loss design is not expected to significantly increase the losses in
the IGBT devices.
6.8 Loss profile for ferrite core inductors 157

6.8 Loss profile for ferrite core inductors

Figures 6.21–6.28 show the power loss and temperature rise for the various components of
the LCL filter for increasing L pu . The inductors used are Ferrite core inductors with round
wire winding. Figures 6.21–6.24 detail the various losses and estimated temperature of
the inverter-side inductor L1 . From Fig. 6.21, we can observe the effect of proximity
effect on the copper loss at switching frequency. As L pu changes from 0.04pu to 0.05pu,
the number of layers is incremented by one, hence there is a noticeable rise in the copper
loss. Subsequently, as the current ripple decreases (because of higher L pu ) the switching
frequency copper loss reduces till the number of layers is again increased at 0.11pu. But
the fundamental copper loss is independent of decrease in current ripple. Core losses do
not affect the total loss after inductance is increased beyond 0.04pu. Fig. 6.23 shows the
total loss and the minimum loss point occurring at 0.04pu. Fig. 6.24 shows the operating
temperature which mirrors the total loss in shape. It can be seen that upto a 10oC reduction
in temperature is possible at the optimum design point, corresponding to an approximate
increase in component life by a factor of two.

Figures 6.25–6.26 show the losses in grid-side inductor L2 , where the switching cur-
rent ripple is sufficiently attenuated and therefore does not contribute to power loss. The
loss in the damping branch C1 -Cd -Rd is shown in Fig 6.27. The resonant frequency is
kept constant, and hence as L pu is increased, C pu is simultaneously reduced, which also
reduces the damping losses. The total LCL filter losses are shown in Fig 6.28. We can
observe that the minimum power loss point for L1 and the entire LCL filter are essentially
same.
158 Experimental Results and Optimized LCL Filter Design

Copper loss at 50Hz and switching frequency in L1

50Hz Loss
20 SW Freq Loss

15
Power loss [W]

10

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
L1+L2 inductance [pu]

Figure 6.21: Estimated copper loss in L1 for different pu ratings of L1 +L2 ; Ferrite core
inductor with round wire winding

Core loss at 50Hz and switching frequency in L1

0.04 4
50Hz Loss
SW Freq Loss
0.035 3.5

0.03 3
SW Freq Loss [W]
50Hz Loss [W]

0.025 2.5

0.02 2

0.015 1.5

0.01 1

0.005 0.5

0 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
L1+L2 inductance [pu]

Figure 6.22: Estimated core loss in L1 for different pu ratings of L1 +L2 ; Ferrite core
inductor with round wire winding
6.8 Loss profile for ferrite core inductors 159

Power loss in L1
25

20
Power loss [W]

15

10

5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
L1+L2 inductance [pu]

Figure 6.23: Estimated total power loss in L1 for different pu ratings of L1 +L2 ; Ferrite
core inductor with round wire winding

Operating temperature of L1
80

75

70
Temperature [deg C]

65

60

55

50

45

40
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
L1+L2 inductance [pu]

Figure 6.24: Estimated operating temperature of L1 for different pu ratings of L1 +L2 ;

Ferrite core inductor with round wire winding
160 Experimental Results and Optimized LCL Filter Design

Copper loss at 50Hz in L2

20

15
Power Loss [W]

10

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
L1+L2 inductance [pu]

Figure 6.25: Estimated copper loss in L2 for different pu ratings of L1 +L2 ; Ferrite core
inductor with round wire winding

Core loss at 50Hz in L2

0.03

0.025

0.02
Power Loss [W]

0.015

0.01

0.005

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
L1+L2 inductance [pu]

Figure 6.26: Estimated core loss in L2 for different pu ratings of L1 +L2 ; Ferrite core
inductor with round wire winding
6.8 Loss profile for ferrite core inductors 161

Damping branch power loss

90 3.5
50Hz Loss
80 SW Freq Loss
3
70
2.5

SW Freq Loss [W]

60
Power Loss [W]

50 2

40 1.5
30
1
20
0.5
10

0 0
1 0.8 0.6 0.4 0.2 0
C1+Cd capacitance [pu]

Figure 6.27: Estimated damping circuit loss for different pu ratings of C1 + Cd ; Ferrite
core inductor with round wire winding

Total power loss in LCL filter

120

110

100

90
Power Loss [W]

80

70

60

50

40

30
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
L1+L2 inductance [pu]

Figure 6.28: Estimated total power loss in LCL filter for different pu ratings of L1 +L2 ;
Ferrite core inductor with round wire winding
162 Experimental Results and Optimized LCL Filter Design

6.9 Loss profile for amorphous core inductors

Figures 6.29–6.35 show the power loss and temperature rise for varying L pu for the LCL
filter made of Amorphous core inductors with foil winding. Figures 6.29–6.32 detail the
power loss and temperature of inverter-side inductor L1 . Fig 6.29 indicates that copper
loss at switching frequency decreases consistently with L pu with no upward bumps in the
curve. Amorphous core materials have higher core losses compared to Ferrite materials
and it can be observed from Fig. 6.30 that at low L pu , the switching frequency core loss is
the dominant loss. The total power loss curve is constant from 0.04pu to 0.1pu since any
decrease in switching frequency core loss is offset by increase in fundamental frequency
copper loss.

The power loss in the damping circuit is not shown since it is same as Fig 6.27.
The loss in the damping circuit depends on the switching frequency and base voltage
rating, and is unaffected by choice of inductors. As in the case of the ferrite inductors,
the minimum power loss point depends strongly on the loss curve of the inverter-side
inductor L1 .

Copper Loss at 50Hz and switching frequency

18 3
50Hz Loss
16 SW Freq Loss
2.5
14

12 2 SW Freq Loss [W]

50Hz Loss [W]

10
1.5
8

6 1

4
0.5
2

0 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
L1+L2 inductance [pu]

Figure 6.29: Estimated copper loss in L1 for different pu ratings of L1 +L2 ; Amorphous
core inductor with foil winding
6.9 Loss profile for amorphous core inductors 163

Core Loss at 50Hz and switching frequency

0.12
50Hz Loss
SW Freq Loss
0.1 20

SW Freq Loss [W]

0.08
50Hz Loss [W]

15

0.06
10
0.04

5
0.02

0 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
L1+L2 inductance [pu]

Figure 6.30: Estimated core loss in L1 for different pu ratings of L1 +L2 ; Amorphous core
inductor with foil winding

Total power loss in L1

30

25
Total power loss [W]

20

15

10
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
L1+L2 inductance [pu]

Figure 6.31: Estimated total power loss in L1 for different pu ratings of L1 +L2 ; Amor-
phous core inductor with foil winding
164 Experimental Results and Optimized LCL Filter Design

Operating temperature of L1
120

110
Temperature [deg C]

100

90

80

70
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
L1+L2 inductance [pu]

Figure 6.32: Estimated operating temperature of L1 for different pu ratings of L1 +L2 ;

Amorphous core inductor with foil winding

Copper loss at 50Hz in L2

16

14

12
Power Loss [W]

10

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
L1+L2 inductance [pu]

Figure 6.33: Estimated copper loss in L2 for different pu ratings of L1 +L2 ; Amorphous
core inductor with foil winding
6.9 Loss profile for amorphous core inductors 165

Core loss at 50Hz in L2

0.12

0.1

0.08
Power Loss [W]

0.06

0.04

0.02

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
L1+L2 inductance [pu]

Figure 6.34: Estimated core loss in L2 for different pu ratings of L1 +L2 ; Amorphous core
inductor with foil winding

Total power loss in LCL filter

120

100
Power Loss [W]

80

60

40

20
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
L1+L2 inductance [pu]

Figure 6.35: Estimated total power loss in LCL filter for different pu ratings of L1 +L2 ;
Amorphous core inductor with foil winding
166 Experimental Results and Optimized LCL Filter Design

6.10 Loss profile for powder core inductors

Figures 6.36–6.42 show the power loss and temperature rise for varying L pu for the LCL
filter made of Powder core inductors with round wire winding. Figures 6.36–6.39 show
the power loss and operating temperature of L1 . The switching frequency copper loss
curve in Fig. 6.36 is similar to Fig. 6.21, since the same proximity effect is dominant
in this case. Powder core materials are temperature sensitive and maximum operating
temperature is around 200◦ C, hence the L pu designs below 0.04pu are not feasible. The
minimum L pu for powder core inductors is around 0.06pu which does not change even
with the addition of losses in damping branch and grid-side inductor L1 .

Copper loss at 50Hz and switching frequency

35
50Hz Loss
SW Freq Loss
30
Power Loss [W]

25

20

15

10

5
0 0.02 0.04 0.06 0.08 0.1
L1+L2 inductance [pu]

Figure 6.36: Estimated copper loss in L1 for different pu ratings of L1 +L2 ; Powder core
inductor with round wire, two bobbin design
6.10 Loss profile for powder core inductors 167

Core loss at 50Hz and switching frequency

60
50Hz Loss
0.14 SW Freq Loss
50
0.12

SW Freq Loss [W]

0.1 40
50Hz Loss [W]

0.08
30
0.06
20
0.04
10
0.02

0 0
0 0.02 0.04 0.06 0.08 0.1
L1+L2 inductance [pu]

Figure 6.37: Estimated core loss in L1 for different pu ratings of L1 +L2 ; Powder core
inductor with round wire, two bobbin design

Total Power loss in L1

90

80
Total Power loss [W]

70

60

50

40

30
0 0.02 0.04 0.06 0.08 0.1
L1+L2 inductance [pu]

Figure 6.38: Estimated total power loss in L1 for different pu ratings of L1 +L2 ; Powder
core inductor with round wire, two bobbin design
168 Experimental Results and Optimized LCL Filter Design

Operating Temperature of L1
220

200
Temperature [deg C]

180

160

140

120
0 0.02 0.04 0.06 0.08 0.1
L1+L2 inductance [pu]

Figure 6.39: Estimated operating temperature of L1 for different pu ratings of L1 +L2 ;

Powder core inductor with round wire, two bobbin design

Copper loss at 50Hz in L2

35

30

25
Power Loss [W]

20

15

10

5
0 0.02 0.04 0.06 0.08 0.1
L1+L2 inductance [pu]

Figure 6.40: Estimated copper loss in L2 for different pu ratings of L1 +L2 ; Powder core
inductor with round wire, two bobbin design
6.10 Loss profile for powder core inductors 169

Core loss at 50Hz in L2

0.14

0.12

0.1
Power Loss [W]

0.08

0.06

0.04

0.02

0
0 0.02 0.04 0.06 0.08 0.1
L1+L2 inductance [pu]

Figure 6.41: Estimated core loss in L2 for different pu ratings of L1 +L2 ; Powder core
inductor with round wire, two bobbin design

Total power loss in LCL filter

200

180

160
Power Loss [W]

140

120

100

80

60
0 0.02 0.04 0.06 0.08 0.1
L1+L2 inductance [pu]

Figure 6.42: Estimated total loss in LCL filter for different pu ratings of L1 +L2 ; Powder
core inductor with round wire, two bobbin design
170 Experimental Results and Optimized LCL Filter Design

6.11 Summary

Experimental results of filtering characteristics show a good match with analysis in the
frequency range of interconnected inverter applications. The high frequency harmonic
spectrum of the output current was well within the IEEE specifications for the rating of
the power converter. The analytical equations predicting the power loss in inductors were
verified through short circuit tests using a 3φ 10kVA power converter. The steady state
temperature rise in individual inductors was measured and compared with the expected
temperature rise. Loss curves for core loss and copper loss for different per unit rating
of total inductance were fomulated. Simultaneously, the total capacitance per unit was
adjusted to maintain the same resonant frequency. Power loss in the damping circuit was
calculated for different per unit ratings. The total LCL filter loss per phase was plotted.
These loss curves were used to find the most efficient LCL filter design for three different
core material -ferrite, amorphous, powder and two different winding types -round and
foil.

A traditional rule of thumb approach to LCL filter design would use L1 and L2 in the
range of 10%. Such a filter is a feasible design but would have higher losses than the
proposed optimized design. It has been shown that it is possible to select lower values of
L1 , L2 and C that can lead to cost effective designs of smaller size, and that would have
lower overall filter power loss.
Chapter 7

Conclusions

The present research work originated from a project to investigate the optimal size and
rating of low pass filters for grid connected power converters. As part of this project, an
extensive literature survey was conducted to ascertain the current state of art in the area of
filter design for grid connected power converters. There were several deficiencies in the
present approach that were identified -some issues include use of arbitrary “thumb-rules”
for design, design procedures that resulted in over-rated designs and design assumption
which would result in bulky and as well as lossy designs.

The approach followed in this report tries to overcome some of the deficiencies of
the previous approaches. The third order LCL filter was found to offer a good balance
between harmonic filtering as well as additional complexity in control. A system level
approach is used to obtain the most relevant transfer functions for design. The IEEE stan-
dard recommendations for high frequency current ripple were used as a major constraint
early in the design to ensure all subsequent optimizations were still compliant with the
IEEE limits.

The inductors of the LCL filter were identified as the component with the most po-
tential for improvement. Attention was given to the power loss in an inductor, and all the
major sources of loss -copper loss, core loss were thoroughly investigated and analytical
equations derived. Thermal analysis of inductors ensured that the steady state operating
conditions of the entire filter was within normal bounds.

The current methods for inductor construction were tested and deficiencies in the
present methods were identified. New methods to easily and accurately design inductors
for three different core materials -Amorphous, Powder and Ferrite were formulated. The
effectiveness of foil winding versus round wire winding was also investigated by incor-

171
172 Conclusions

porating both the winding types in the design.

All the design assumptions were thoroughly tested by actual construction and test-
ing. Frequency response was measured using a network analyzer. Harmonic spectrum
of output was sampled and verified to be within IEEE norms. Power loss in individual
inductors was measured by short circuit testing in combination with a 3 phase 10kVA
power converter. The steady state temperature rise in individual inductors was measured
and compared favourably with the expected temperature rise. Using these results, the
most efficient LCL filter design with least temperature rise but which still meets IEEE
harmonic standards was found for ferrite, amorphous and powder core materials.

Future research possibilities

The system level transfer function analysis currently does not include several grid depen-
dent parameters like low voltage ride-through requirements, EMI filtering, and dynamic
response requirements. These additional constraints can be included to generate guide-
lines for the filter packaging and design process.

Presently the analytical equations used for power loss estimation have a limited range
of accuracy. Similarly, thermal analysis can be enhanced by including fluid modelling
methods for natural convection to accurately estimate the operating temperature of the
inductors and can be extended to forced cooled designs.

The power loss of individual components in this report is tested in stand-alone con-
verter mode under short circuit conditions. These conditions give the worst case current
ripple conditions because of the low modulation index of the power converter. The ef-
ficiency of the filter can be tested in actual grid connected mode to test the variation in
the losses. This test will result in lower losses and improved thermal characteristics of
the filter. Additionally, advanced PWM methods with non-conventional sequences can be
implemented to further reduce the high frequency ripple current losses.

The minimum loss L pu designs have other implications on grid connected power con-
verter which can be investigate to further optimize the LCL filter and the power converter
including the IGBTs and the DC bus components.
List of AppendixFigures

A.1 LCL filter response ig /ii (vg =0); ferrite core inductor, no damping . . . . 177

A.2 LCL filter response ig /ii (vg =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω177

A.3 LCL filter response vg /ii (ig =0); ferrite core inductor, no damping . . . . 178

A.4 LCL filter response vg /ii (ig =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω178

A.5 LCL filter response vg /vi (ig =0); ferrite core inductor, no damping . . . . 179

A.6 LCL filter response vg /vi (ig =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω179

A.7 LCL filter response vi /ii (vg =0); ferrite core inductor, no damping . . . . 180

A.8 LCL filter response vi /ii (vg =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω180

A.9 LCL filter response vi /ii (ig =0); ferrite core inductor, no damping . . . . . 181

A.10 LCL filter response vi /ii (ig =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω181

A.11 LCL filter response ig /ii (vg =0); amorphous core inductor, no damping . . 182

A.12 LCL filter response ig /ii (vg =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

A.13 LCL filter response vg /ii (ig =0); amorphous core inductor, no damping . . 183

A.14 LCL filter response vg /ii (ig =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

A.15 LCL filter response vg /vi (ig =0); amorphous core inductor, no damping . 184

A.16 LCL filter response vg /vi (ig =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

A.17 LCL filter response vi /ii (vg =0); amorphous core inductor, no damping . . 185

173
174 LIST OF APPENDIXFIGURES

A.18 LCL filter response vi /ii (vg =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

A.19 LCL filter response vi /ii (ig =0); amorphous core inductor, no damping . . 186

A.20 LCL filter response vi /ii (ig =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

A.21 LCL filter response ig /ii (vg =0); powder core inductor, no damping . . . . 187

A.22 LCL filter response ig /ii (vg =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

A.23 LCL filter response vg /ii (ig =0); powder core inductor, no damping . . . . 188

A.24 LCL filter response vg /ii (ig =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

A.25 LCL filter response vg /vi (ig =0); powder core inductor, no damping . . . 189

A.26 LCL filter response vg /vi (ig =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

A.27 LCL filter response vi /ii (vg =0); powder core inductor, no damping . . . . 190

A.28 LCL filter response vi /ii (vg =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

A.29 LCL filter response vi /ii (ig =0); powder core inductor, no damping . . . . 191

A.30 LCL filter response vi /ii (ig =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

D.1 Dimensions of ferrite core(in mm) . . . . . . . . . . . . . . . . . . . . . 213

D.2 Inductor model with single layer winding . . . . . . . . . . . . . . . . . 214

D.3 Inductor model with double layer winding . . . . . . . . . . . . . . . . . 214

D.4 Inductor model with four layer winding . . . . . . . . . . . . . . . . . . 215

D.5 Inductor model with individual conductors . . . . . . . . . . . . . . . . . 215

D.6 3D Inductor model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

D.7 Plot of Fluxdensity in the ferrite inductor with steel support . . . . . . . . 216
Appendix A 175

D.8 Plot of flux density along the airgap of ferrite inductor with steel strap
around the core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

D.9 Amorphous core inductor model(Core type: AMCC630) . . . . . . . . . 218

D.10 Flux Plot for Amorphous inductor with AMCC630 core . . . . . . . . . . 218

D.11 Fluxdensity plot for Amorphous inductor with AMCC630 core . . . . . . 219

D.12 Airgap flux density plot for Amorphous inductor with AMCC630 core . . 219

D.13 Amorphous core inductor model(Core type: AMCC367S) . . . . . . . . . 220

D.14 Flux Plot for Amorphous inductor with AMCC367S core . . . . . . . . . 220

D.15 Fluxdensity plot for Amorphous inductor with AMCC367S core . . . . . 221

D.16 Airgap flux density plot for Amorphous inductor with AMCC367S core . 221

D.17 Powder core inductor model . . . . . . . . . . . . . . . . . . . . . . . . 222

D.18 Flux plot for powder core inductor . . . . . . . . . . . . . . . . . . . . . 223

D.19 Fluxdensity plot for powder core inductor . . . . . . . . . . . . . . . . . 223

Appendix A

Transfer Function Tests

System used for Frequency Response Analysis:

AP Instruments Inc., Frequency Response Analyzer - Model No.: 200 ISA

Frequency Response Analyzer settings (basic) under test-conditions:

Sweep Frequency Range : 10Hz - 100kHz
Averaging : 100
Points : 1000
V Level : 1.77V
DC Source : 0.0V
Bandwidth : 10Hz

A.1 Ferrite core inductor results

176
Appendix A 177

Frequency response of LCL filter i /i R−phase v =0

60 g i g
Sim
40 Exp
20

Magnitude [dB]
0

−20

−40

−60

−80
1 2 3 4 5 6
10 10 10 10 10 10

50 Frequency [rad/s]
Sim
Exp
0
Phase [deg]

−50

−100

−150

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.1: LCL filter response ig /ii (vg =0); ferrite core inductor, no damping

Frequency response of LCL filter ig/ii R−phase vg=0

20
Sim
Exp
0
Magnitude [dB]

−20

−40

−60

−80
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

50
Sim
Exp
0
Phase [deg]

−50

−100

−150

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure A.2: LCL filter response ig /ii (vg =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω
178 Appendix A

Frequency response of LCL filter vg/ii R−phase vg=0

60
Sim
Exp
Magnitude [dB] 40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

100
Sim
Exp
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.3: LCL filter response vg /ii (ig =0); ferrite core inductor, no damping

Frequency response of LCL filter vg/ii R−phase ig=0

80
Sim
Exp
60
Magnitude [dB]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

100
Sim
Exp
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure A.4: LCL filter response vg /ii (ig =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω
Appendix A 179

Frequency response of LCL filter v /v R−phase i =0

60 g i g
Sim
40 Exp
20

Magnitude [dB]
0

−20

−40

−60

−80
1 2 3 4 5 6
10 10 10 10 10 10

100 Frequency [rad/s]

Sim
50 Exp

0
Phase [deg]

−50

−100

−150

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.5: LCL filter response vg /vi (ig =0); ferrite core inductor, no damping

Frequency response of LCL filter vg/vi R−phase ig=0

20
Sim
Exp
0
Magnitude [dB]

−20

−40

−60

−80
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

0 Sim
Exp
−50
Phase [deg]

−100

−150

−200
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure A.6: LCL filter response vg /vi (ig =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω
180 Appendix A

Frequency response of LCL filter vi/ii R−phase ig=0

80
Sim
Exp
60
Magnitude [dB]
40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

100
Sim
Exp
50
Phase [deg]

−50

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.7: LCL filter response vi /ii (vg =0); ferrite core inductor, no damping

Frequency response of LCL filter vi/ii R−phase vg=0

80
Sim
Exp
60
Magnitude [dB]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

100
Sim
80 Exp

60
Phase [deg]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure A.8: LCL filter response vi /ii (vg =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω
Appendix A 181

i i g

80
Sim
60 Exp

Magnitude [dB]
40

20

−20

−40
1 2 3 4 5 6
10 10 10 10 10 10

100 Frequency [rad/s]

Sim
Exp
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.9: LCL filter response vi /ii (ig =0); ferrite core inductor, no damping

Frequency response of LCL filter vi/ii R−phase ig=0

70
Sim
60 Exp
Magnitude [dB]

50

40

30

20

10
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

100
Sim
Exp
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10
Frequency [rad/s]

Figure A.10: LCL filter response vi /ii (ig =0); ferrite core inductor, C1 =Cd =8µ F, Rd =25Ω
182 Appendix A

A.2 Amorphous core inductor results

Frequency response of LCL filter i /i R−phase v =0

g i g
100
Sim
Act
50
Magnitude [dB]

−50

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.11: LCL filter response ig /ii (vg =0); amorphous core inductor, no damping

20
Sim
0 Exp
Magnitude [dB]

−20

−40

−60

−80

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]
200
Sim
Exp
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.12: LCL filter response ig /ii (vg =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω
Appendix A 183

Frequency response of LCL filter v /i R−phase i =0

g i g
80
Sim
Act
60

Magnitude [dB]
40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

−85
Sim
Act
−90
Phase [deg]

−95

−100

−105

−110
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.13: LCL filter response vg /ii (ig =0); amorphous core inductor, no damping

Frequency response of LCL filter vg/ii R−phase ig=0

80
Sim
Exp
60
Magnitude [dB]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

−70
Sim
Exp
−80
Phase [deg]

−90

−100

−110
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.14: LCL filter response vg /ii (ig =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω
184 Appendix A

Frequency response of LCL filter vg/vi R−phase ig=0

40
Sim
20 Act
Magnitude [dB]
0

−20

−40

−60

−80

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]
200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.15: LCL filter response vg /vi (ig =0); amorphous core inductor, no damping

20
Sim
0 Exp
Magnitude [dB]

−20

−40

−60

−80

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]
200
Sim
Exp
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.16: LCL filter response vg /vi (ig =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω
Appendix A 185

Frequency response of LCL filter vi/ii R−phase vg=0

100
Sim
80 Act

60

Magnitude [dB]
40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

100
Sim
Frequency [rad/s] Act
50
Phase [deg]

−50

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.17: LCL filter response vi /ii (vg =0); amorphous core inductor, no damping

Frequency response of LCL filter vi/ii R−phase vg=0

80
Sim
Exp
60
Magnitude [dB]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

100
Sim
80 Exp

60
Phase [deg]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Figure A.18: LCL filter response vi /ii (vg =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω
186 Appendix A

Frequency response of LCL filter v /i R−phase i =0

i i g
80
Sim
Act
60
Magnitude [dB]
40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

100
Sim
Act
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.19: LCL filter response vi /ii (ig =0); amorphous core inductor, no damping

Frequency response of LCL filter vi/ii R−phase ig=0

80
Sim
70 Exp
60
Magnitude [dB]

50

40

30

20

10
1 2 3 4 5 6
10 10 10 10 10 10

100 Frequency [rad/s]

Sim
Exp
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.20: LCL filter response vi /ii (ig =0); amorphous core inductor, C1 =Cd =6µ F,
Rd =25Ω
Appendix A 187

A.3 Powder iron core inductor results

Frequency response of LCL filter ig/ii R−phase vg=0

40
Sim
20 Act

Magnitude [dB]
0

−20

−40

−60

−80

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]
200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.21: LCL filter response ig /ii (vg =0); powder core inductor, no damping

Frequency response of LCL filter ig/ii R−phase vg=0

20
Sim
Exp
0
Magnitude [dB]

−20

−40

−60

−80
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

200
Sim
Exp
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.22: LCL filter response ig /ii (vg =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω
188 Appendix A

Frequency response of LCL filter v /i R−phase i =0

g i g
60
Sim
Act
40
Magnitude [dB]
20

−20

−40
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

−80
Sim
−85 Act

−90
Phase [deg]

−95

−100

−105

−110
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.23: LCL filter response vg /ii (ig =0); powder core inductor, no damping

Frequency response of LCL filter vg/ii R−phase ig=0

60
Sim
Exp
40
Magnitude [dB]

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

−60
Sim
Exp
−70
Phase [deg]

−80

−90

−100

−110
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.24: LCL filter response vg /ii (ig =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω
Appendix A 189

Frequency response of LCL filter v /v R−phase i =0

g i g
50
Sim
Act

Magnitude [dB]
0

−50

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

200
Sim
Act
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.25: LCL filter response vg /vi (ig =0); powder core inductor, no damping

Frequency response of LCL filter vg/vi R−phase ig=0

20
Sim
Exp
0
Magnitude [dB]

−20

−40

−60

−80
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

200
Sim
Exp
100
Phase [deg]

−100

−200
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.26: LCL filter response vg /vi (ig =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω
190 Appendix A

i i g

80
Sim
60 Act
Magnitude [dB]
40

20

−20

−40
1 2 3 4 5 6
10 10 10 10 10 10

100 Frequency [rad/s]

Sim
Act
50
Phase [deg]

−50

−100
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.27: LCL filter response vi /ii (vg =0); powder core inductor, no damping

Frequency response of LCL filter vi/ii R−phase vg=0

80
Sim
Exp
60
Magnitude [dB]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

100
Sim
80 Exp

60
Phase [deg]

40

20

−20
1 2 3 4 5 6
10 10 10 10 10 10

Figure A.28: LCL filter response vi /ii (vg =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω
Appendix A 191

i i g

80
Sim
60 Act

Magnitude [dB]
40

20

−20

−40
1 2 3 4 5 6
10 10 10 10 10 10

100 Frequency [rad/s]

Sim
Act
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.29: LCL filter response vi /ii (ig =0); powder core inductor, no damping

Frequency response of LCL filter vi/ii R−phase ig=0

70
Sim
60 Exp
Magnitude [dB]

50

40

30

20

10

0
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]
100
Sim
Exp
50
Phase [deg]

−50

−100

−150
1 2 3 4 5 6
10 10 10 10 10 10

Frequency [rad/s]

Figure A.30: LCL filter response vi /ii (ig =0); powder core inductor, C1 =Cd =10µ F,
Rd =10Ω
Appendix B

Temperature Rise Tests

Systems used for Temperature rise tests:

DC Power Supply: Agilent Technologies - Model No.: N5768A

Specifications: 80V/19A, 1520W

Thermal Imager: Fluke Corporation - Model: Fluke-T120 INT/9

Specifications:
Temperature Range : −100C to +1270C
Display Range : −150C to +3600C
RTC Range : −500C to +4600C
Detector Columns : 128
Detector Rows : 96

Data Logger: Yokogawa - Model No.: MX100

Specifications:
Logging Type : Mainly PC Measurement
Total max. no. of connect-able channels : 1200 (20 units*6 modules)
Display Monitor System : through MX100 Software or API
Operating Temp. Range : 00C to +500C
Power Supply Frequency : 50Hz
Standard Interface : Ethernet
Measurement Interval : 10ms to 60s
Application Software : Windows 2000/XP/Vista

193
194 Appendix B

Thermocouple Type : K-type (Chromel-Alumel)

Specifications:
Temperature Range : −2000C to +13500C
Sensitivity : approximately 41µ V/0C

B.1 Ferrite core inductor results

Figure B.1: DC temperature test; ferrite core inductor with round wire winding; all three
phases
Appendix B 195

Figure B.2: DC temperature test; thermal image of ferrite core inductor - A1 with round
wire winding
196 Appendix B

Figure B.3: DC temperature test; thermal image of ferrite core inductor - B1 with round
wire winding
Appendix B 197

Figure B.4: DC temperature test; thermal image of ferrite core inductor - C1 with round
wire winding
198 Appendix B

Figure B.5: DC temperature test; thermal image of ferrite core inductor - A1, B1, C1 with
round wire winding
Appendix B 199

B.2 Amorphous core inductor results

Figure B.6: DC temperature test; amorphous core AMCC367S and AMCC630 inductor
with foil winding; all three phases

B.3 Powder iron core inductor results

200 Appendix B

Figure B.7: DC temperature test; powder core inductor with foil winding and round wire
winding - two bobbin design; all three phases
Appendix C

Test Set-up

C.1 Schematic and filter layout

The configuration of the test bed set-up is common for all three LCL filter types. The
common platform facilitates appropriate physical placement of the individual filter com-
ponents simultaneously satisfying the performance constraints (enables easy and imme-
diate swapping of LCL filters connected between the Inverter and Grid for damping case
and no damping case).

Key elements of the Test set-up schematic:

Variac/Auto transformer can be adjusted ranging from 0V to 230V
Rectifier is a 3-phase diode bridge rectifier ranging uptil 600V
Load is a 3-phase star connected resistive load rated for 5A per phase
Switches: S1, S2 and S3 are circuit breakers rated for 25A

Switch-S1 Switch-S2 Switch-S3 Mode

Closed Short Open Standalone inverter
Closed Open Open Standalone inverter
Closed Load Open Standalone inverter
Closed Grid Open Not connected
Open Grid Closed Grid parallel

Table C.1: Test set-up schematic operating configurations

201
Appendix C
202

Figure C.1: Test Set-Up Schematic

Appendix C 203

Figure C.2: Top View of Filter-Connection Layout - with damping

204 Appendix C

Figure C.3: Top View of Filter-Connection Layout - without damping

Appendix C 205

C.2 Pictures of filter components and test set-up

Figure C.4: Front View of the Ferrite core Inductor

Figure C.5: Side View of the Ferrite core Inductor

206 Appendix C

Figure C.6: Front View of the Amorphous core Inductor - AMCC367S

Figure C.7: Side View of the Amorphous core Inductor - AMCC630

Appendix C 207

Figure C.8: Front View of the Powder core Inductor - two bobbin design

Figure C.9: Side View of the Powder core Inductor - two bobbin design
208 Appendix C

Figure C.10: Front View of the Powder core Inductor

Figure C.11: Side View of the Powder core Inductor

Appendix C 209

Figure C.12: Front View of the Inverter

210 Appendix C

Figure C.13: Side View of the Inverter

Appendix C 211

Figure C.14: Top View of the Ferrite Filter Set-Up

212 Appendix C

Figure C.15: Complete Physical Test Set-Up; for LCL Ferrite Core Inductor
Appendix D

Simulation of Inductors Using MagNet

D.1 Ferrite core inductor

This section contains the results of simulations of the filter inductor using the FEA tool
called MagNet. The inductor models and plots of fluxlines and flux density are given
along with the inductance values.

Figure D.1: Dimensions of ferrite core(in mm)

213
214 Appendix D

Figure D.2: Inductor model with single layer winding

Figure D.3: Inductor model with double layer winding

Inductor model Mesh size(mm) Inductance (mH)

Single current sheet 1 1.911
Two layer winding 3 1.91
Four layer winding 3 1.914
Individual Conductors 0.5 1.9

Table D.1: Inductance values from simulation of Ferrite inductor (at 14.58A)
Appendix D 215

Figure D.4: Inductor model with four layer winding

Figure D.5: Inductor model with individual conductors

216 Appendix D

Figure D.6: 3D Inductor model

Figure D.7: Plot of Fluxdensity in the ferrite inductor with steel support
Appendix D 217

Figure D.8: Plot of flux density along the airgap of ferrite inductor with steel strap around
the core
218 Appendix D

D.2 Amorphous core inductors

Figure D.9: Amorphous core inductor model(Core type: AMCC630)

Figure D.10: Flux Plot for Amorphous inductor with AMCC630 core

Inductor model Mesh size(mm) Inductance (mH)

Amorphous core (AMCC 630) 5 5.73
Amorphous core (AMCC 367S) 5 3.57

Table D.2: Inductance values from simulation of Amorphous core inductor (at 14.58A)
Appendix D 219

Figure D.11: Fluxdensity plot for Amorphous inductor with AMCC630 core

Figure D.12: Airgap flux density plot for Amorphous inductor with AMCC630 core
220 Appendix D

Figure D.13: Amorphous core inductor model(Core type: AMCC367S)

Figure D.14: Flux Plot for Amorphous inductor with AMCC367S core
Appendix D 221

Figure D.15: Fluxdensity plot for Amorphous inductor with AMCC367S core

Figure D.16: Airgap flux density plot for Amorphous inductor with AMCC367S core
222 Appendix D

D.3 Powdered core inductor

Figure D.17: Powder core inductor model

Inductor model Mesh size(mm) Inductance (mH)

Powdered core (BK7320) 5 1.7

Table D.3: Inductance values from simulation of Powdered core inductor (at 14.58A)
Appendix D 223

Figure D.18: Flux plot for powder core inductor

Figure D.19: Fluxdensity plot for powder core inductor

Appendix E

Electromagnetic Equations

E.1 Introduction

This appendix gives the theoretical background for the electromagnetic equations used
in chapter 4. The vector equations and other derivations are referenced from established
texts on electromagnetics [1]–[6].

E.2 Maxwell’s Equations

The differential form of Maxwell’s equations are used to describe and relate field vectors,
current densities, and charge densities at any point in space at any time. These equa-
tions are valid only if field vectors are single-valued, bounded, continuous functions of
position and time and exhibit continuous derivatives. But most practical field problems
involve systems containing more than one kind of material. In case there exist abrupt
changes in charges and current densities, the variation of the field vectors are related to
the discontinuous distribution of charges and currents by boundary conditions. So a com-
plete description of field vectors at any point requires both the Maxwell’s equations and
the associated boundary conditions. In differential form, Maxwell’s equations are written
as

∂B
5×E = − (E.1)
∂t

∂D
5×H = J+ (E.2)
∂t

225
226 Appendix E

5·D = q (E.3)

5·B = 0 (E.4)

All these field quantities - E, H, D, B, J are assumed to be time-varying, and each

is a function of space-time coordinates, i.e E = E(x, y, z;t). However, in many practical
systems involving electromagnetic waves the time variations are of cosinusoidal form
and are referred as time-harmonic. Such time variations are represented by e jω t and the
instantaneous electromagnetic field vectors are related to their complex forms in a very
simple manner.

E(x, y, z;t) = Re E(x, y, z)e jω t

 
(E.5)

H(x, y, z;t) = Re H(x, y, z)e jω t

 
(E.6)

D(x, y, z;t) = Re D(x, y, z)e jω t

 
(E.7)

B(x, y, z;t) = Re B(x, y, z)e jω t

 
(E.8)

J(x, y, z;t) = Re J(x, y, z)e jω t

 
(E.9)

E, H, D, B, J represent instantaneous field vectors while E, H, B, D, J represent

the corresponding complex spatial forms which are only a function of position. Here we
have chosen to represent the instantaneous quantities by the real part of the product of the
corresponding complex spatial quantities with e jω t . Another option would be to represent
the instantaneous quantities by the imaginary product of the products. The magnitudes of
√
the instantaneous fields represent peak values that are related to the RMS values by 2.

The Maxwell’s equations in differential form can be written in terms of the complex
field vectors by a simple substitution.

• Replace the instantaneous field vectors by corresponding spatial forms

• Replace ∂ /∂ t by jω .

5 × E = − jω B (E.10)

5 × H = J + jω D (E.11)

5·D = q (E.12)

5·B = 0 (E.13)
Appendix E 227

E.3 Wave Equation

The first two Maxwell’s equations (Eq (E.1) and (E.2)) are first order, coupled equations;
i.e both unknown fields E, H appear in each equation. To uncouple these equations, we
have to increase the order of the differential equations to second order. Taking curl on
both sides of each equation,

∂
5 × 5 × E = −µ (5 × H) (E.14)
∂t

∂
5×5×H = 5×J+ε (5 × E) (E.15)
∂t
Substituting Eq (E.1) and Eq (E.2) and using the vector identity

5 × 5 × F = 5(5 · F) − 52 F (E.16)

we get

∂J ∂ 2E
5(5 · E) − 52 E = −µ − µε 2 (E.17)
∂t ∂t

∂ 2H
5(5 · H) − 5 H = 5 × J − µε 2
2
(E.18)
∂t
Substituting Eq (E.3) and Eq (E.4) in the above equation

q
5 · D = ε 5 ·E = q ⇒ 5 · E = (E.19)
ε

5 · B = µ 5 ·H = 0 (E.20)

and using the constitutive parameter σ

J = σE (E.21)

we get

1 ∂E ∂ 2E
52 E = 5 ·q + µσ + µε 2 (E.22)
ε ∂t ∂t

∂H ∂ 2H
52 H = µσ + µε 2 (E.23)
∂t ∂t
228 Appendix E

Equations (E.22) and (E.23) are referred to as vector wave equations for E and H. For
source-free regions, q=0.

∂E ∂ 2E
5 E = µσ
2
+ µε 2 (E.24)
∂t ∂t

∂H ∂ 2H
52 H = µσ + µε 2 (E.25)
∂t ∂t
For lossless media, σ = 0,

∂ 2E
52 E = µε (E.26)
∂ t2

∂ 2H
52 H = µε (E.27)
∂ t2
For time-harmonic fields, the wave equations (for source-free media) are

52 E = jω µσ E − ω 2 µε E = γ 2 E (E.28)

52 H = jω µσ H − ω 2 µε H = γ 2 H (E.29)

where

γ 2 = jω µσ − ω 2 µε (E.30)

γ = α + jβ (E.31)

• γ = propagation constant

• α = attenuation constant

• β = phase constant (or wave number)

If we allow positive and negative values of σ

(
±(α + jβ ) for +σ
γ = ± jω µ(σ + jω ε) =
p
(E.32)
±(α − jβ ) for -σ

For source-free and lossless media,

52 E = −ω 2 µε E = −β 2 E (E.33)

52 H = −ω 2 µε H = −β 2 H (E.34)
Appendix E 229

Equations of the form of (E.33) and (E.34) are known as homogeneous vector Helmholtz
equations.

The time variations of most practical problems are time-harmonic. Fourier series can
be used to express time variations of other forms in terms of a number of time-harmonic
terms. For many cases, the vector wave equations reduce to a number of scalar Helmholtz
equations, and general solutions can be constructed once solutions to each of the scalar
Helmholtz equations are found.

One method that can be used to solve the scalar Helmholtz equation is known as
separation of variables. This leads to solutions which are products of three functions
(for three-dimensional problems), each function depending upon one coordinate variable
only. Such solutions can be added to form a series which represent very general functions.
Also, single-product solutions of the wave equation represent modes which can propagate
individually.

E.4 Rectangular coordinate system

In rectangular coordinate system, the vector wave equations are reduced to three scalar
wave Helmholtz equations. Assuming source free (q = 0) but lossy medium (σ 6= 0), both
E and H must satisfy Eqns (E.28) and (E.29). We can consider the solution to E and write
the solution to H by inspection.

In rectangular coordinates, the general solution for E(x, y, z) can be written as

E(x, y, z) = âx Ex (x, y, z) + âyEy (x, y, z) + âzEz (x, y, z) (E.35)

52 E − γ 2 E = 52 (âx Ex + ây Ey + âz Ez ) − γ 2 (âx Ex + ây Ey + âz Ez ) = 0 (E.36)

which reduces to three scalar wave equations of

52 Ex (x, y, z) − γ 2Ex (x, y, z) = 0 (E.37)

52 Ey (x, y, z) − γ 2Ey (x, y, z) = 0 (E.38)

52 Ez (x, y, z) − γ 2Ez (x, y, z) = 0 (E.39)

where

γ 2 = jω µ(σ + jω ε) (E.40)
230 Appendix E

Equations (E.37), (E.38) and (E.39) are of same form; once a solution of any of them
is obtained, the solution to the others can be written by inspection. Choosing Ex , in
expanded form

∂ 2 Ex ∂ 2 Ex ∂ 2 Ex
5 2 Ex − γ 2 Ex = + + − γ 2 Ex = 0 (E.41)
∂ x2 ∂ y2 ∂ z2

Using method of separation of variables, we can assume a solution for Ex (x, y, z)

Ex (x, y, z) = f (x)g(y)h(z) (E.42)

where the x, y, z variations of Ex are separable. Substituting Eq (E.42) in Eq (E.41),

∂2 f ∂ 2g ∂ 2h
gh + f h + f g − γ 2 f gh = 0 (E.43)
∂x 2 ∂y 2 ∂z 2

Since f (x), g(y) and h(z) are each a function of only one variable, we can replace partial
by ordinary derivatives. Also dividing by f gh we get

1 d 2 f 1 d 2g 1 d 2h
+ + = γ2 (E.44)
f dx2 g dy2 h dz2

Each of the terms on the left hand side is a function of only a single independent variable;
hence the sum of these terms can equal γ 2 only if each term is a constant. So Eq (E.44)
separates into three equation of the form,

1 d2 f
= γx2 (E.45)
f dx2

1 d 2g
= γy2 (E.46)
g dy2
1 d 2h
2
= γz2 (E.47)
h dz
In addition

γx2 + γy2 + γz2 = γ 2 (E.48)

Eq (E.48) is known as the constraint equation. The solution to Eq (E.45) can take different
forms. In terms of exponentials f (x) has solution

f1 (x) = A1 e−γx x + B1 e+γx x (E.49)

Appendix E 231

or writing in terms of hyperbolic functions

f2 (x) = C1 cosh(γx x) + D1 sinh(γx x) (E.50)

g(y) and h(z) can be expressed in exactly the same form, with different constants and
roots of the solution.

g1 (y) = A2 e−γy y + B2 e+γy y (E.51)

g2 (y) = C2 cosh(γy y) + D2 sinh(γy y) (E.52)

h1 (z) = A3 e−γz z + B3 e+γz z (E.53)

h2 (z) = C3 cosh(γz z) + D3 sinh(γzz) (E.54)

The appropriate form of f , g and h chosen to represent the solution of Ex is decided by the
geometry of the problem. A similar procedure can be used to derive the other components
of E i.e Ey and Ez . The instantaneous electric and magnetic field components can be
obtained by multiplying the factor e jω t and taking the real part.

E.5 Cylindrical coordinate system

Z
z
φ
A
ρ

Y
z
φ
X ρ B

Figure E.1: Cylindrical coordinate system and unit vectors

If the geometry of the system is in cylindrical configuration then the boundary-value

problem for E and H should be solved using cylindrical coordinates. As in rectangu-
lar coordinates, we can consider the solution of E and H will have the same form. In
232 Appendix E

cylindrical coordinates a general solution to the vector wave equation is given by

E(ρ , φ , z) = âρ Eρ (ρ , φ , z) + âφ Eφ (ρ , φ , z) + âzEz (ρ , φ , z) (E.55)

where ρ (rho), φ (phi) and z are the cylindrical coordinates as shown in fig E.1. From Eq
(E.28)

52 (âρ Eρ + âφ Eφ + âz Ez ) = γ 2 (âρ Eρ + âφ Eφ + âz Ez ) (E.56)

which does not reduce to three simple scalar wave equations, similar to Eqns (E.37)-
(E.39) because

52 (âρ Eρ ) 6= âρ 52 Eρ (E.57)

52 (âφ Eφ ) 6= âφ 52 Eφ (E.58)

However,

52 (âz Ez ) = âz 52 Ez (E.59)

If we consider two different points (points A and B on fig E.1) we can see that direction
of âρ and âφ are different while âz still has same direction. This means the unit vectors
âρ and âφ cannot be treated as constants but as functions of (ρ , φ , z). So only one of the
three scalar equations reduces to

5 2 Ez − γ 2 Ez = 0 (E.60)

In the following discussion, γ is assumed to be real, to simplify calculation. Using the

vector identity

52 E = 5(5 · E) − 5 × 5 × E (E.61)

Substituting for 52 E,

5(5 · E) − 5 × 5 × E = γ 2 E (E.62)

Expanding the individual terms we get three scalar partial differential equations,

2 ∂ Eφ
 
Eρ
2
5 Eρ + − 2 − 2 = γ 2 Eρ (E.63)
ρ ρ ∂φ
Appendix E 233

2 ∂ Eφ
 
Eφ
2
5 Eφ + − 2 + 2 = γ 2 Eφ (E.64)
ρ ρ ∂φ
5 2 Ez = γ 2 Ez (E.65)

Eqns (E.63) and (E.64) are coupled (each contain more than one electric field component)
second-order partial differential equations, which are the most difficult to solve. However
Eq (E.65) is an uncoupled second-order partial differential equation. In each of the above
equations, 52 ψ (ρ , φ , z) is the Laplacian of a scalar that in cylindrical coordinates takes
the form

1 ∂ ∂ψ 1 ∂ 2ψ ∂ 2ψ
 
5 ψ (ρ , φ , z) =
2
ρ + 2 + 2 (E.66)
ρ ∂ρ ∂ρ ρ ∂φ 2 ∂z

∂ 2 ψ 1 ∂ψ 1 ∂ 2ψ ∂ 2ψ
52 ψ (ρ , φ , z) = + + + 2 (E.67)
∂ρ 2 ρ ∂ρ ρ 2 ∂φ 2 ∂z
In expanded form Eq (E.65) can be written as

∂ 2 ψ 1 ∂ψ 1 ∂ 2ψ ∂ 2ψ
+ + + 2 = γ 2ψ (E.68)
∂ρ 2 ρ ∂ρ ρ ∂φ
2 2 ∂z

where ψ (ρ , φ , z) is a scalar function representing a field or a vector potential component.

Using the method of separation of variables,

ψ (ρ , φ , z) = f (ρ )g(φ )h(z) (E.69)

Substituting it in above equation,

∂2 f 1∂f 1 ∂ 2g ∂ 2h
gh + gh + f h + f g = γ 2 f gh (E.70)
∂ρ 2 ρ ∂ρ ρ 2 ∂φ 2 ∂ z2

Dividing both sides by f gh and replacing partial by ordinary derivatives,

1 d 2 f 1 1 d f 1 1 d 2g 1 d 2h
+ + + = γ2 (E.71)
f dρ 2 f ρ dρ g ρ dφ
2 2 h dz 2

The last term on left hand side is only a function of z, so in the same way as rectangular
coordinates,

d 2h
= γz2 h (E.72)
dz2
234 Appendix E

where γz2 is a constant. Substituting in above equation and multiplying both sides by ρ 2 ,

ρ 2 d2 f ρ d f 1 d2g
+ + + (γz2 − γ 2 )ρ 2 = 0 (E.73)
f dρ 2 f dρ g dφ 2

Now the third term on the left hand side is only a function of φ , so it can be set equal to a
constant −m2 .

d 2g
= −m2 g (E.74)
dφ 2

Let

γz2 − γ 2 = γ p2 (E.75)

Using the two substitutions, and multiplying both sides of the Eq (E.73) by f ,

d2 f df
ρ2 +ρ + [(γ pρ )2 − m2 ] f = 0 (E.76)
dρ 2 dρ

Eq (E.76) is the classic Bessel differential equation with real arguments, and Eq (E.75) is
the constraint equation for the wave equation in cylindrical coordinates.

Solutions to Eq (E.72), Eq (E.74) and Eq (E.76) take the form

f1 (ρ ) = A1 Jm (γ p ρ ) + B1Ym (γ p ρ ) (E.77)

or

(1) (2)
f2 (ρ ) = C1 Hm (γ pρ ) + D1 Hm (γ p ρ ) (E.78)

and

g1 (φ ) = A2 e− jmφ + B2 e+ jmφ (E.79)

or

g2 (φ ) = C2 cos(mφ ) + D2 sin(mφ ) (E.80)

and

h1 (z) = A3 e− jγz z + B3 e+ jγz z (E.81)

Appendix E 235

or

h2 (z) = C3 cos(γz z) + D3 sin(γz z) (E.82)

Jm (γ pρ ) and Ym (γ p ρ ) represent the Bessel functions of first and second kind respectively;
(1) (2)
Hm (γ p ρ ) and Hm (γ pρ ) represent Hankel functions of the first and second kind respec-
tively. Although Eqns (E.77) to (E.82) are valid solutions for f (ρ ), g(φ ) and h(z), the
most appropriate form depends upon the problem in question. Bessel functions are used
to represent standing waves while Hankel functions are used to represent traveling waves.
Exponentials represent travelling waves while Trigonometric functions represent periodic
waves.

E.6 Bessel functions

The standard form of Bessel’s equation can be written as

d 2y dy
x2 2
+ x + (x2 − ν 2 )y = 0 (E.83)
dx dx

where ν ≥ 0 is a real number. Another useful form is obtained by changing the variable
x = uλ and replacing u by x.

d 2y dy
x2 2
+ x + (x2 λ 2 − ν 2 )y = 0 (E.84)
dx dx

When ν is not an integer we can write the solution as

y(x) = A1 Jν (x) + B1 J−ν (x) for x 6= 0 (E.85)

or

y(x) = A1 Jν (λ x) + B1 J−ν (λ x) for x 6= 0 (E.86)

where
∞
(−1)m (x/2)2m+ν
Jν (x) = ∑ m!(m + ν )! (E.87)
m=0

∞
(−1)m (x/2)2m−ν
J−ν (x) = ∑ m!(m − ν )! (E.88)
m=0
236 Appendix E

m! = Γ(m + 1) (E.89)

If ν is an integer, then the two functions Jν (x) and J−ν (x) become linearly dependent i.e
if ν = n where n = 1, 2..

Jn (x) = (−1)n Jn (x) for n = 1, 2, .. (E.90)

As the combination of two dependent solutions of a differential equation is itself a solu-

tion, the second solution of Bessel’s function is

Jν (x)cos(νπ ) − J−ν (x)

Yν (x) = (E.91)
sin(νπ )

with

Yn (x) = lim Yν (x) (E.92)

ν →n

The function Y0 (x) is also called Neumann or Weber function of order zero and denoted
by N0 (x). For integral values of ν , Yν (x) becomes infinite at x = 0, so it cannot be present
in any problem for which x = 0 is included in the region over which the solution applies.
When ν = n is an integer,

y(x) = A2 Jn (x) + B2Yn (x) ν = n = 0 or integer (E.93)

So for all ν , the general solution of Bessel’s equation in the standard form is

y(x) = C1 Jν (x) +C2Yν (x) (E.94)

or

y(x) = C1 Jν (λ x) +C2Yν (λ x) (E.95)

Jν (x)is the Bessel’s function of the first kind of order ν , Yν (x) is the Bessel’s function of
the second kind of order ν and Γ(x) is the gamma function.

Replacing the independent variable x in Bessel’s equation by jx changes the differen-

tial equation to

x2 y00 + xy0 − (x2 + ν 2 )y = 0 (E.96)

which is called Bessel’s modified equation of order ν . This equation has two linearly
Appendix E 237

independent complex solutions Jν (ix) and Yν (ix). Since they are not convenient to use,
they are scaled and combined to give two real linearly independent solutions denoted by
Iν (x) and Kν (x). These are modified Bessel functions of the first and second kinds of
order ν .
∞
x2m+ν
Iν (x) = ∑ 2m+ν m!Γ(m + ν + 1)
(E.97)
m=0 2

Provided ν is not an integer, the general solution of Bessel’s modified equation can be
written as

y(x) = C1 Iν (x) +C2 I−ν (x) ν 6= 0 or integer (E.98)

Usually Kν (x) is used in place of I−ν (x)

π
 
I−ν (x) − Iν (x)
Kν (x) = (E.99)
2 sin νπ

In case ν is an integer, the function Kn is defined as

π
 
I−ν (x) − Iν (x)
Kn (x) = lim (E.100)
ν →n 2 sin νπ

The general solution of Bessel’s modified equation can be written in the form

y(x) = C1 Iν (x) +C2 Kν (x) (E.101)

with no restriction placed on ν . When the Bessel’s modified equation is written in the
form

x2 y00 + xy0 − (λ 2 x2 + ν 2 )y = 0 (E.102)

its general solution is given by

y(x) = C1 Iν (λ x) +C2 Kν (λ x) (E.103)

The modified Bessel functions are related to the regular Bessel functions as

Iν (x) = j−ν Jν ( jx) = jν J−ν ( jx) = jν Jν (− jx) (E.104)

I−ν (x) = jν J−ν ( jx) (E.105)

238 Appendix E

Hankel functions are defined as

(1)
Hν (x) = Jν (x) + jYν (x) (E.106)

(2)
Hν (x) = Jν (x) − jYν (x) (E.107)
(1) (2)
where Hν (x) is the Hankel function of the first kind of order ν , and Hν (x) is the
Hankel function of the second kind of order ν . Since both functions contain Yν (x), both
are singular at x = 0. The Henkel functions are related to the modified Bessel functions
as

π ν +1 (1) π
j Hν ( jx) = (− j)ν +1 Hν (− jx)
(2)
Kν (x) = (E.108)
2 2

If the argument is complex (i.e xe3π i/4 ), we get Kelvin functions.

berν (x) + jbeiν (x) = Jν (xe3π i/4 ) (E.109)

The modified Bessel function of order ν and argument x may be defined as the integral
function given by
Z π
1
Iν (x) = ex cos θ cos νθ d θ (E.110)
2π −π

To obtain the derivatives of Iν (x), we differentiate both sides of Eq. (E.110) with respect
to x to obtain
Z π
dIν (x) 1
= cos θ ex cos θ cos νθ d θ (E.111)
dx 2π −π

dIν (x) 1 1 π x cos θ 1 π x cos θ

 Z Z 
= e cos(ν + 1)θ d θ + e cos(ν − 1)θ d θ (E.112)
dx 2 2π −π 2π −π
dIν (x) 1
= [Iν +1 (x) + Iν −1 (x)] (E.113)
dx 2
Eq. (E.113) is valid for ν ≥ 1. For ν = 0,

dI0 (x)
= I1 (x) (E.114)
dx

Similarly,

d p
I0 ( j ω x) = j ω I1 ( j ω x)
p p
(E.115)
dx
Appendix E 239

E.7 Retarded potentials

The potential functions A and Φ for time-varying fields are called the retarded potentials.
The magnetic vector potential A is defined in terms of the magnetic field density B

B = 5×A (E.116)

Substituting this in the Maxwell’s equation Eq. (E.1)

∂
5×E = − (5 × A) (E.117)
∂t

∂A
5 × (E + )=0 (E.118)
∂t
Eq. (E.118) says that the curl of some vector is zero, which means that the vector can be
derived as the gradient of some scalar. If we assume the electric potential Φ as the scalar
function

∂A
E+ = −5Φ (E.119)
∂t

∂A
E = −5Φ− (E.120)
∂t
Substituting Eq. (E.120) in Eq. (E.3) we get

∂ q
− 52 Φ − (5 · A) = (E.121)
∂t ε

Replacing H by B in Eq. (E.2) and substituting Eq. (E.116) and (E.120) we get

∂Φ ∂ 2A
  
5 × 5 × A = µ J − µε 5 + 2 (E.122)
∂t ∂t

or

∂Φ ∂ 2A
   
5(5 · A) − 5 A = µ J − µε 5
2
+ 2 (E.123)
∂t ∂t

Eq. (E.121) and Eq. (E.123) can be further simplified if we define A. To completely
define a vector, we have to specify both its curl and divergence and its value at any one
point. The curl of A is already defined, so if we choose the divergence as

∂Φ
5 · A = −µε (E.124)
∂t
240 Appendix E

we can simplify Eq. (E.121) and Eq. (E.123).

∂ 2Φ q
52 Φ − µε =− (E.125)
∂t 2 ε

∂ 2A
52 A − µε = −µ J (E.126)
∂ t2
The potentials A and Φ are now defined in terms of sources J and q and can be used to
derive the electric and magnetic fields.

For static fields the derivatives with respect to time will become zero and Eq. (E.125)
and Eq. (E.126) reduce to

q
52 Φ = − (E.127)
ε

52 A = − µ J (E.128)

For fields that are time-harmonic, Eq. (E.125) and Eq. (E.126) will become

q
52 Φ − εµω 2 Φ = − (E.129)
ε

52 A − εµω 2 A = −µ J (E.130)

E.8 References

Electromagnetics and Mathematics

1. W.R.Smythe, Static and Dynamic Electricity (2nd ed.), McGraw Hill Book Com-
pany, 1950

2. W.H.Hayt Jr, J.A.Buck, Engineering Electromagnetics (6th ed.), Tata McGraw-

Hill, 2001

3. S.Ramo, J.R.Whinnery, T. van Duzer, Fields and Waves in Communication Elec-

tronics (3rd ed.), John Wiley and Sons, 1994

4. C.A.Balanis, Advanced Engineering Electromagnetics (1st ed.), John Wiley and

Sons, 1989
Appendix E 241

5. A.Jeffrey, Advanced Engineering Mathematics (1st ed.), Academic Press Elsevier

India, 2002

6. M.Abramowitz, I.E.Stegun, Handbook of Mathematical Functions, Dover, 1970

242 Appendix E

————————————————————————