Analysis and Modeling of

Magnetic Coupling
Bryce Hesterman
Advanced Energy Industries

Tuesday, 10 April 2007

Discovery Learning Center
University Of Colorado, Boulder, Colorado

Denver Chapter, IEEE Power Electronics Society

www.denverpels.org
 SLIDE # 1

Analysis and Modeling of Magnetic Coupling

Denver Chapter of IEEE PELS

Discovery Learning Center
University of Colorado
April 10, 2007
Bryce Hesterman
Member of Technical Staff I
Advanced Energy Industries, Inc.
Fort Collins, Colorado
 SLIDE # 2

Presentation Outline
Introduction

Modeling magnetic coupling with electric circuit equations

Measuring electric circuit model parameters

Equivalent circuits for transformers and coupled inductors

Magnetic circuit modeling overview

Tips for creating magnetic circuit models

Deriving electric model parameters from magnetic model parameters

Matrix theory requirements for coupling stability

Examples
 SLIDE # 3

Motivation For This Presentation

Magnetic coupling often seems to be mysterious and hard to quantify

I had the good fortune of having a mentor, Dr. James H. Spreen, who
taught me how to analyze magnetic coupling

Goal: help make magnetic coupling less mysterious by showing how to

model it, measure it and use it in circuit analysis and simulation
 SLIDE # 4

What Is Magnetic Coupling?

Two windings are coupled when some of the magnetic flux produced by
currents flowing in either of the windings passes through both windings

Only part of the flux produced by a current in one winding reaches other
windings
 SLIDE # 5

Magnetic Coupling Modeling Options

Electric circuit: inductances and couplings
Linear model

Model parameters can be determined from circuit measurements

Parameters can be measured with fairly high accuracy if appropriate

measurement procedures are followed

No information on flux paths or flux levels

 SLIDE # 6

Magnetic Coupling Modeling Options

Magnetic circuit: reluctance circuit and electrical-to-magnetic
interfaces for each winding
Explicitly shows flux paths as magnetic circuit elements

Flux paths and reluctances are only approximations

Works with linear or nonlinear reluctance models

Electrical circuit parameters can be calculated from magnetic circuit

parameters, but not vice-versa
 SLIDE # 7

V-I Equations For an Isolated Inductor

(Current flows into

positive terminal)

di
Time domain v=L
dt
Frequency domain v = jLi
 SLIDE # 8

Time-domain Equations For Two windings

di1 di2
L11 = Self inductance of winding 1 v1 = L11 + L12
dt dt
L22 = Self inductance of winding 2
di1 di2
L12 = L21 = Mutual inductance v2 = L 21 + L 22
dt dt
 SLIDE # 9

Time-domain Equations For N Windings

di1 di di di
v1 = L11 +L12 2 + L13 3 ++ L1N N
dt dt dt dt
di1 di di di
v2 = L12 + L22 2 + L23 3 ++ L2 N N
dt dt dt dt

di1 di2 di3 diN

v N = LN 1 + LN 2 + LN 3 ++ LN N
dt dt dt dt

v1 L11 L12 L1N i1

v L L 22 L 2 N d i2 d
2 = 21 [v] = [L] [ i ]
dt dt

v N LN 1 LN 2 LN N iN
 SLIDE # 10

Frequency-domain Equations For Two Windings

L11 = Self inductance of winding 1

L22 = Self inductance of winding 2 v1 = jL11 i1 + jL12 i2

L12 = L21 = Mutual inductance v2 = j L21i1 + j L22 i2
 SLIDE # 11

Frequency-domain Equations For N Windings

v1 = j L11i1 + j L12 i2 + j L13i3 ++ j L1N iN

v2 = j L12 i1 + j L 22 i2 + j L23 i3 ++ j L 2N iN

v N = j LN 1i1 + j LN 2 i2 + j LN 3i3 ++ j LN N iN

v1 L11 L12 L1N i1

v L L 22 L 2N i2
2 = j 21 [v] = j[L][ i ]


v N LN 1 LN 2 LN N iN
 SLIDE # 12

Inductance Matrix Symmetry

The inductance matrix is symmetric due to the reciprocity theorem (This is

required for conservation of energy)
Lqr = Lrq
L. O. Chua, Linear and Nonlinear Circuits. New York: McGraw-Hill,
1987, pp. 771-780

This principle can be derived from Maxwells equations:

C. G. Montgomery, R. H. Dicke, and E. M. Purcell, Ed., Principles of

Microwave Circuits. New York: Dover Publications, 1965

It can also be derived from a stored energy argument:

R. R. Lawrence, Principles of Alternating Currents. New York: McGraw-

Hill, 1935, pp. 187-188
 SLIDE # 13

Coupling Coefficient For Two Windings

L12 L21 M
k= = = 1 k 1
L11L22 L11L22 L1L2
 SLIDE # 14

Coupling Coefficients For N Windings

Lqr
k qr = k qr = k rq
Lqq Lrr

L12 L13
1
1 k12 k13 L11L 22 L11L33
L 21 L 23
K = k 21 1
k 23 = 1
L 22 L11 L22 L33
k31 k32 1
L31 L32
L L 1
33 11 L33 L 22
 SLIDE # 15

Coupling Measurement Techniques:

Series-aiding Series-opposing Method
Coupling measurements are made for each pair of windings

Measure the inductance of each pair of windings connected in the series-

aiding manner

Measure the inductance of each pair of windings connected in the series-

opposing manner

Laid Lopp L12 Laid Lopp

L12 = k= =
4 L1L 2 4 L1L 2

L1 L2 L1 L2

series-aiding, Laid series-opposing, Lopp

 SLIDE # 16

Series-aiding Series-opposing Coupling Formula

Derivation

Start with the fundamental VI equations

v1 = j L11i1 + j L12 i2

v2 = j L12 i1 + j L 22 i2
1 k12 1
Write down what is known for the series-aiding configuration
i1 i2
i2 = i1
L1 L2
vaid = v1 + v2
vaid = j Laid i1 series-aiding, Laid
 SLIDE # 17

Substitute assumptions for series aiding into V-I equations

v1 = j L11i1 + jL12 i1

v2 = j L 21i1 + j L22 i1

Recall vaid = v1 + v2

Substitute vaid = ( j L11i1 + jL12 i1 ) + ( j L12 i1 + j L 22 i1)

Simplify vaid = j (L11+ 2L12 + L 22 )i1

Recall vaid = j Laid i1

Therefore Laid = L11+ 2 L12 + L 22

 SLIDE # 18

Write down what is known for the series-opposing configuration

i2 = i1 i1 i2

vopp = v1 v2 L1 L2

vopp = j Lopp i1
series-opposing, Lopp

Substitute assumptions for series-opposing into V-I equations

v1 = j L11i1 jL12 i1
v2 = j L21 i1 j L22 i1

Recall vopp = v1 v2

Substitute vopp = ( j L11i1 jL12 i1 ) ( jL12 i1 j L22 i1)

 SLIDE # 19

Simplify vopp = j (L11 2L12 + L22 )i1

Recall vopp = j Lopp i1

Therefore Lopp = L11 2 L12 + L 22

Recall Laid = L11+ 2 L12 + L 22

Subtract Laid Lopp = 4L12

Laid Lopp
Solve L12 =
4
L12
Recall k=
L11L22
Laid Lopp
Therefore k=
4 L11L22
 SLIDE # 20

Coupling Measurement Techniques:

Voltage Ratio Method
Coupling measurements are made for each pair of windings

Measure the voltage at each winding when a voltage is applied to the

other winding

vocqr = voltage measured at winding q when winding r is driven

vdrq = voltage measured at winding r when winding r is driven and the

voltage at winding q is being measured

voc 21 voc12
k=
vd 12 vd 21
Measurements can be corrupted by capacitance and loading

Particularly suited for measurements between windings on different legs

of three-phase ungapped transformers
 SLIDE # 21

Voltage Ratio Method

Coupling Formula Derivation
Start with the fundamental VI equations

v1 = j L11 i1 + jL12 i2
v2 = jL12 i1 + j L22 i2

Find i1 and v2 when winding 1 is driven and winding 2 is unloaded

vd 12
v1 = vd 12 = j L11 i1 i1=
j L11
v2 = voc 21 = j L12 i1

Substitute i1, vd 12 voc 21 L12

simplify and voc 21 = j L12 =
j L11 vd 12 L11
rearrange
 SLIDE # 22

Find i2 and v1 when winding 2 is driven and winding 1 is unloaded

vd 21
v2 = vd 21 = j L22 i2 i2 =
j L22
v1 = voc12 = j L12 i2

Substitute i2, vd 12 voc12 L12

simplify and voc12 = j L12 =
j L22 vd 21 L22
rearrange
voc 21 L12 voc 21 voc12
Recall = k=
vd 12 L11 vd 12 vd 21
voc 21 voc12 L12 L12
Multiply =
vd 12 vd 21 L11 L22
 SLIDE # 23

voc 21 voc12 L12

Take the square root = =k
vd 12 vd 21 L11L22

voc 21 voc12
Therefore k=
vd 12 vd 21
 SLIDE # 24

Coupling Measurement Techniques:

Self and Leakage Inductance Method
Coupling measurements are made for each pair of windings

Measure the self inductance of each winding (L11 and L22)

The inductance measured at a winding when another winding is shorted

is called the leakage inductance

Lleakqr = inductance measured at winding q when winding r is shorted

Lleak12 Lleak 21
k12 = 1 k 21 = 1
L11 L22

If perfect measurements are made, k12 = k21

 SLIDE # 25

Self and Leakage Inductance Method

Coupling Formula Derivation
Start with the fundamental VI equations

v1 = j L11i1 + jL12 i2

v2 = jL12 i1 + j L 22 i2

Find the leakage inductance at winding 1 when winding 2 is shorted, L12s

Write down what is known for that condition

v1 = j Lleak 12 i1

v2 = 0 = j L12 i1 + j L22 i2
Simplify 0 = L12 i1 + L22 i2
 SLIDE # 26

Solve for i2 L12

i2 = i1
L 22

Combine
equations for v1 v1 = j Lleak 12 i1= j L11 i1 + jL12 i2

Simplify Lleak 12 i1= L11 i1 + L12 i2

Substitute L212
value of i2
Lleak 12 = L11
L 22
 SLIDE # 27

Divide by L11 Lleak 12 L212

= 1
L11 L11L 22

Rearrange L12s L212

1 =
L11 L11L 22

Take square root Lleak 12 L12

1 =
L11 L11L 22

Therefore Lleak 12
k = 1
L11
Similar formula Lleak 21
when winding 1 is k = 1
shorted L22
 SLIDE # 28

Self Inductance Measurement Tips

The ratios of the self inductances of windings on the same core will be
nearly equal to the square of the turns ratios

The ratios of the inductances are more important than the exact values

Measure at a frequency where the Q is high

(It doesnt have to be at the operating frequency)

Avoid measuring inductance close to self-resonant frequencies

Take all measurements at the same frequency

 SLIDE # 29

Self Inductance Measurement Tips

For ungapped cores:

The ratios of the inductances are more important than the exact values

Inductance measurements may vary with amplitude of the test signal

Best results are obtained if the core excitation is equal for all
measurements

If possible, adjust the measurement drive voltage to be proportional to the

turns in order to keep the flux density constant

Inductance measurements can be made at flux levels near normal

operating levels using a power amplifier and a network analyzer
 SLIDE # 30

Leakage Inductance Measurement Tips

Avoid measuring inductance close to self-resonant frequencies

Winding resistances decrease the measured coupling coefficients when

using the self and leakage inductance method

Measure at a frequency where the Q is high

(It doesnt have to be at the operating frequency)

For each pair of windings, shorting the winding with the highest Q gives
the best results.
 SLIDE # 31

Coupling Coefficient Value Discrepancies

Use the same value for k12 and k21 when performing circuit calculations

If k12 k21, the one with the largest value is usually the most accurate
 SLIDE # 32

Coupling Coefficient Significant Digits

Inductances can typically be measured to at least two significant figures

Use enough significant digits for coupling coefficients in calculations and

simulations to be able to accurately reproduce the leakage inductance

k = 1
Lleak 12
L11
( 2
Lleak12 =L11 1 k12 )
Generally use at least four significant digits when the coupling coefficient is
close to 1 (k is known to more significant digits than the measured
inductances it was derived from)
 SLIDE # 33

Coupling Polarity Conventions

Polarity dots indicate that all windings which have dots of the same style
will have matching voltage polarities between the dotted and un-dotted
terminals when one winding is driven and all of the other windings are
open-circuited

By convention, positive current flows into the terminals that are labeled as
being positive regardless of the polarity dot positions

Voltage reference polarities can assigned in whatever manner is

convenient by changing the polarity of the coupling coefficient

1 k 1

Pin-for-pin equivalent electrical behavior

 SLIDE # 34

Three-legged Transformer Coupling Coefficient

Dot Convention
Three types of polarity dots are required for a three-legged transformer
There are three coupling coefficients
The triple product of the three coupling coefficients is negative
It is simplest to make all three of the coupling coefficients negative

k12 k 23 k31 0
 SLIDE # 35

Equivalent Circuits For Two Windings

Many possible circuits, but only three parameters are required
My favorite is the Cantilever model: two inductors and one ideal
transformer
k La 1: Ne

L11 L22 1 Lb 2

Ideal

La = Lleak12 L 22
Lb = L11 La Ne =
Lb
Measure the inductance Measure the self Measure the self inductance of
at winding 1 with winding inductance of winding 1 winding 2 and calculate above
2 shorted and subtract La equation
Cantilever Model Derivation
k La 1: Ne

L11 L22 1 Lb 2

Ideal

If winding 2 is shorted, then the inductance measured at winding 1 is

simply La because the ideal transformer shorts out Lb. Thus La is equal to
the leakage inductance measured at winding 1.
La = Lleak12
To convert the coupled inductor model
to the cantilever model, recall that (
Lleak12 = L11 1 k 2 )
Therefore (
La = L11 1 k 2 )
 k La 1: Ne

L11 L22 1 Lb 2

Ideal

By observation, Lb must equal the open-circuit inductance of winding 1

minus La
Lb = L11 La

To convert the coupled inductor model to

the cantilever model, recall that (
La = L11 1 k 2 )
Substitute (
Lb = L11 L11 1 k 2 )
Therefore Lb = k 2 L11
 k La 1: Ne

L11 L22 1 Lb 2

Ideal

The inductance of winding 2, L22 , is equal to Lb reflected

through the ideal transformer L 22 = Ne2 Lb

L 22
Solve for the turns ratio Ne =
Lb
Recall Lb = k 2 L11
1 L 22
Thus, in terms of the coupled Ne =
inductor model k L11
 SLIDE # 39

Two Leakage Inductance Model

Two Leakage Inductance model allows the actual turns ratio to be used,
but it adds unnecessary complexity

These leakage inductances are not equal to the previously-defined

conventional leakage inductances measured by shorting windings

L11 = Ll1 + LMag

2 Ll1 N1 :N 2 Ll 2
N2
L 22 = Ll 2 + LMag
N1 LMag
N2 L12
L12 = LMag k =
N1 L11L22 Ideal

N1 N1
LMag = L11L22 Lleak12 L 22 = L11L22 Lleak 21 L11
N2 N2
 SLIDE # 40

Two Leakage Inductance Model Derivation

Start with the fundamental VI equations

v1 = j L11i1 + jL12 i2

v2 = j L12 i1 + j L 22 i2

Write down the open-circuit primary voltage due to a current in the

secondary
N2
v1 = j L12 i2 = j LMag i2
N1
By inspection, the mutual inductance is

N2
L12 = LMag
N1
 SLIDE # 41

N1
Solve for LMag L Mag = L12
N2
2
Recall L11 = Ll1 + LMag N2
L 22 = Ll 2 + LMag
N1
Solve for the two leakage inductances
2
N2
Ll1 = L11 LMag Ll 2 = L 22 LMag
N1
Recall the equation for Lleak12 from the derivation of the Self and Leakage
Inductance model L2
Lleak 12 = L11 12
L 22
L212
By symmetry, Lleak 21 = L22
L11
 SLIDE # 42

Solve for L12 L12 = L11L 22 Lleak12 L 22 = L11L 22 Lleak 21L11

N1
Recall L Mag = L12
N2

N1
Therefore L Mag = L11L 22 Lleak12 L 22
N2

N1
Also, L Mag = L11L 22 Lleak 21 L11
N2
 SLIDE # 43

Equivalent Circuits For More Than Two Windings

Many possible circuits, but only N(N+1)/2 parameters are required

My favorite is the extended cantilever model
Exact correspondence to the coupling coefficient model
R. W. Erickson and D. Maksimovic, A multiple-winding magnetics model having
directly measurable parameters, in Proc. IEEE Power Electronics Specialists
Conf., May 1998, pp. 1472-1478

Extended cantilever model with parasitics:

K. D. T. Ngo, S. Srinivas, and P. Nakmahachalasint, Broadband extended
cantilever model for multi-winding magnetics, IEEE Trans. Power Electron., vol.
16, pp. 551-557, July 2001

K.D.T. Ngo and A. Gangupomu, "Improved method to extract the short-circuit

parameters of the BECM," Power Electronics Letters, IEEE , vol.1, no.1, pp. 17- 18,
March 2003
 SLIDE # 44

Magnetic Devices Can Be Approximately Modeled

With Magnetic Circuits
Reluctance paths are not as well defined as electric circuit paths
Accuracy is improved by using more reluctance elements in the model

R2 R3 R4 R5

R2 R3 R4 R5 R1 R6
R15

R1 R15 R6
MMF1 MMF2 MMF3

R13 R17 R18 R19 R14

R17 R19
R7 R12 R18
R16 R13
R14

R8 R9 R10 R11 R16

R7 R8 R9 R10 R11 R12

Winding 1 Winding 2 Winding 3

High-leakage Transformer
with two E-Cores
 SLIDE # 45

Winding Self Inductance

Replace the MMF source for each winding not being considered with a
short circuit
Determine the Thvenin equivalent reluctance, th ,at the MMF source
representing the winding for which the self inductance is being calculated
Inductance = turns squared divided by the Thvenin equivalent reluctance
R2 R3 R4 R5

R1 R6
R15
MMF1 Rth1

MMF1 N2 MMF2 MMF3

N1
N1 R17 R19 N3
R18 short
N12
R14
R13

L1 =
R7
R16

R12
short th1
R8 R9 R10 R11

Nx = turns of winding x
 SLIDE # 46

Leakage Inductances
Replace the MMF source for each winding not being considered with a
short circuit
Replace the MMF source for the shorted winding with an open circuit
Determine Thvenin equivalent reluctance, th , at the MMF source for the
winding where the leakage inductance is to be determined
Inductance = turns squared divided by the Thvenin equivalent reluctance
R2 R3 R4 R5

R1 R6
R15 MMF1 Rth1

MMF1 N2 MMF2 MMF3 N1

N1 R17 R19 N3 open
R18
R13
R14
N12
Lleak 13 =
R16
short th1
R7 R8 R9 R10 R11 R12

Nx = turns of winding x
 SLIDE # 47

Reluctance Modeling References

R. W. Erickson and D. Maksimovic, Fundamentals of Power Electronics,
2nd Ed., Norwell, MA: Kluwer Academic Publishers, 2001

S.-A. El-Hamamsey and Eric I. Chang, Magnetics Modeling for Computer-

Aided Design of Power Electronics Circuits, PESC 1989 Record, pp. 635-
645

G. W. Ludwig, and S.-A. El Hamamsy Coupled Inductance and

Reluctance Models of Magnetic Components, IEEE Trans. on Power
Electronics, Vol. 6, No. 2, April 1991, pp. 240-250

E. Colin Cherry, The Duality between Interlinked Electric and Magnetic

Circuits and the Formation of Transformer Equivalent Circuits,
Proceedings of the Physical Society, vol. 62 part 2, section B, no. 350 B,
Feb. 1949, pp. 101-111

MIT Staff, Magnetic Circuits and Transformers. Cambridge, MA: MIT

Press, 1943
 SLIDE # 48

Energy Storage
The magnetic energy stored in one inductor is:
1 2
WM 1 = Li
2

The energy stored in a set of N coupled windings is:

1 T
WM N = [i] [L ][i]
2
The energy stored in a set of 2 coupled windings is:

1
2
L11 L12 i1 1
WM 2 = [i1 i2 ]
L L
= L i
i 2 11 1 (
2
+ 2 L i i
12 1 2 + L i
22 2
2
)
12 22 2
 SLIDE # 49

Stability of a Set of Coupled Inductors

As set of coupled inductors is passive if the magnetic energy storage is
non-negative for any set of currents

For two coupled inductors, this is guaranteed if: 1 k 1

A set of three coupled inductors is passive if the
following conditions are met:
2 2 2
k12 + k 23 + k31 2k12 k 23k31 1
1 k12 1 1 k 23 1 1 k31 1
Yilmaz Tokad and Myril B. Reed, Criteria and
Tests for Realizability of the Inductance Matrix,
Trans. AIEE, Part I, Communications and
Electronics, Vol. 78, Jan. 1960, pp. 924-926
 SLIDE # 50

Stability of a Set of Coupled Inductors

When there are more than three windings, the coupling coefficient matrix
defined below can be used to determine stability

1 k12 k1N
k 1
k2 N
K= 21 k qr = k rq


k N 1 k N 2 1
A set of coupled inductors is passive if and only if all of the eigenvalues
of K are non-negative
There are N eigenvalues for a set of N windings
Eigenvalue calculations are built-in functions of programs like Mathcad
and Matlab
 SLIDE # 51

Consistency Checks
The eigenvalue test can let you know if there are measurement errors, but
it wont help identify the errors

The ratios of magnetizing inductances on the same core leg should be

approximately equal to the square of the turns ratios

Set up test simulations to verify that the models match the test conditions

Test leakage inductances by shorting one winding and applying a signal to

the other winding (Compute the inductance indicated by the voltage,
current and frequency)

Check for coupling polarity errors, especially if there are windings on

multiple core legs

Compare test data with multiple windings shorted to simulations or

calculations with multiple windings shorted
 SLIDE # 52

Inverse Inductance Matrix

The inverse inductance matrix, , is the reciprocal of the inductance matrix

= L1
Each diagonal element is equal to the reciprocal of the inductance of the
corresponding winding when all of the of the other windings are shorted

Example application: You have a four-winding transformer. What is the

inductance at winding 1 when windings 3 and 4 are shorted?

Set up an inductance matrix, Lx, by extracting all of the elements from the
total inductance matrix that apply only to windings 1, 3 and 4
 SLIDE # 53

Inverse Inductance Matrix Example

Remove the outlined elements from L to get Lx

L11 L12 L13 L14

L L11 L13 L14
22 L 22 L 23 L 24
L= Lx = L 31 L 33 L 34
L 31 L 32 L 33 L 34
L41 L43 L44
L41 L42 L43 L44

Use a computer program to compute the x = L x1

inverse of Lx

1
is the inductance at winding 1 when windings 3 and 4 are shorted
x11
 SLIDE # 54

Modified Node Analysis

Spice uses Modified Node Analysis (MNA) to set up the circuit equations

This type of analysis is well suited for handling circuits with coupled
windings

A good description of the method is found in:

L. O. Chua, Linear and Nonlinear Circuits. New York: McGraw-Hill, 1987,

pp. 469-472
 SLIDE # 55

Modified Node Analysis Example

Assign node numbers

The reference node is node 0
Assign currents for each branch
Compute mutual inductances from the self inductances and coupling
coefficients
L12 = k L11 L 22
 SLIDE # 56

Write a KCL equation using the node-to-datum voltages as variables for

each node other than the datum node, unless the node has a fixed voltage
with respect to the datum node. In that case, write an equation that
assigns the node voltage to the fixed value.
 SLIDE # 57

Node 1 v 1 = vs

Node 2 i1 = j C1(v1 v2 ) = j C1(vs v2 )

v3 1
Node 3 i2 = + j C 2 v3 = + j C 2 v3
R1 R1
Current variables must be maintained for each coupled inductor
Other current variables can be replaced by equivalent expressions
 SLIDE # 58

Write an equation for the voltage dropped across each coupled inductor
with the node variables on one side and the inductance and current
terms on the other side

v2 = j L11i1 jL12 i2

v3 = j L12 i1 j L22 i2

Note that the minus signs are due to the fact that i2 was assigned to be
flowing out of the dotted end of the winding instead of into it
 SLIDE # 59

Rearrange the node equations to facilitate writing matrix equations

1
i1 = j C1(vs v2 ) v2 + i1 = vs
j C1

1 1
i2 = + j C 2 v3 + j C 2 v3 i2 = 0
R1 R1

v2 = j L11 i1 jL12 i2 v2 j L11 i1 + j L12 i2 = 0

v3 = j L12 i1 j L22 i2 v3 j L12 i1 + j L22 i2 = 0

 SLIDE # 60

Convert the node equations into matrix form G X =U

1 1
1 v 2 + i1 = vs 2 + j C 2 v3 i2 = 0
j C1 R1

3 v2 j L11 i1 + jL12 i2 = 0 4 v3 j L12 i1 + j L22 i2 = 0

1
1 0 0
1
j C1 v2 v s

0 1 v 0
2

+ j C 2 0 1 =
3

R1 i1 0

3
1 0 j L11 jL12 i
2 0
4 0 1 jL12 j L22 i2
 SLIDE # 61

Modified Node Analysis Example Conclusions

The matrix equation can be solved numerically in programs like Mathcad or

Matlab

The approach is straightforward and can be used with many windings

If you want to have a symbolic solution, simple cases can be handled with
equivalent circuits (the cantilever model is easiest for two windings)

Equivalent circuit equations can get very messy with more than two
windings

Is there a better way to find symbolic solutions by using Dr. R. D.

Middlebrooks Design-Oriented Analysis techniques? (I hope to find out)

See Dr. Middlebrooks web site: www.ardem.com for information on his

analysis techniques
 SLIDE # 62

Coupling Stability Example Using Mathcad

Figure 1. Three coupled windings with resistive loads.

Suppose we have a three-winding inductor with a resistor terminating

each winding as is shown in Fig. 1. If we externally force an initial
condition of currents in the windings, and then let the inductor "coast"
on its own, the currents should all exponentially decay to zero as the
stored energy is dissipated.

If the coupling coefficient matrix is not positive definite, however, at

least one of the eigenvalues for the system will be not be negative.
 SLIDE # 63

As is shown below, this produces an unstable situation where more

energy can be extracted from the inductor than was initially stored there.

When the initial energy is dissipated, the stored energy becomes

negative, and the inductor continues to deliver power. This, of course, is
impossible. Therefore, a set of coupling coefficients in which the coupling
coefficient matrix is not positive definite describes a coupled inductor that
is not physically realizable.

Because Spice and other circuit simulators allow such inductors to be

specified, it is instructive to examine what happens when nonphysical
sets of coupling coefficients are specified.
 SLIDE # 64

We first examine a physically-realizable case. After that, we show what

happens when the coupling coefficients are improperly chosen.

Enter the values of the L1 := 10 H k12 := 0.96

inductances and
coupling coefficients
L2 := 11 H k23 := 0.99

L3 := 10 H k13 := 0.98
 SLIDE # 65

1 k12 k13
Construct the
coupling K := k12 1 k23
coefficient matrix. k k 1
13 23

2.953
Compute the eigenvalues of
eigenvals ( K) = 0.041
the coupling coefficient
matrix. 5.654 10 3

All of the eigenvalues are positive. As is shown below, this leads to

negative eigenvalues for the time response of the system.
 SLIDE # 66

L1 K L1 L2 K L1 L3
1, 2 1, 3
Construct the inductance L := K1 , 2 L1 L2 L2 K L2 L3
2, 3
matrix.
K L1 L3 K L2 L3 L3
1, 3 2, 3

10.000 10.069 9.800

Evaluate the inductance L = 10.069 11.000 10.383 H
matrix.
9.800 10.383 10.000

Enter the values of the R1 := 1 R2 := 1 R3 := 1

resistances
 SLIDE # 67

Enter the initial values of the I1i := 1 A I2i := 0 A I3i := 0 A

currents

I1

Define a current vector I I2
I
3

I1i 1

Define an initial condition Ii := I2i Ii = 0 A

vector I 0
3i

We can think of this as a situation in which the current in winding 1 is

externally forced to be 1 Ampere, and then the circuit is left to dissipate
the stored energy starting at time t = 0.
 SLIDE # 68

Compute the stored energy T 5

Ii L Ii = 1.000 10 J

The time response of the circuit of Fig. 1 can be described by the following
equation

I1 R1
d
I2 R2 L I (1)
I R dt
3 3

2.909 1.422 4.327

Compute the inverse
:= L
inductance matrix
= 1.422 5.263 6.858
1 1
H
4.327 6.858 11.462
 SLIDE # 69

17.251
eigenvals ( ) = 2.351
1
Compute the eigenvalues of .
H
0.033

I1 R1
d

Write (1) in terms of the inverse I I2 R2 (2)
inductance matrix dt I R
3 3

R1 0 0
d

Write (2) in a way that shows I I 0 R2 0 I (3)
in the right side dt 0 0 R
3
 SLIDE # 70

R1 0 0

Define G G := 0 R2 0 (4)
0 0 R
3

Substitute G into (3) d (5)

I G I
dt

1.725 107

= 2.351 106 s
Compute the eigenvalues of G -1
:= eigenvals ( G)

4
3.277 10
 SLIDE # 71

Compute the eigenvectors of G

0.295 0.77 0.565
:= eigenvecs ( G) = 0.501 0.628 0.595 s
-1

0.814 0.107 0.571

1 t
The solution to (5) will have the form c1 0 0 e

I 0 c2 0 e 2 t (6)
0 0 c
3 3 t
e
 SLIDE # 72

The values of the constants c1, c2 and c3 can be determined from the initial
conditions

c1 c1 c1 0.295
1
I0 c2 2 := Ii
c
c c c2 = 0.77 A s
3 3 c 0.565
3

c1 1 1

Check to see if the initial conditions c2 = 0 A Ii = 0 A

are satisfied c 0 0
3
 SLIDE # 73

Compute a coefficient matrix

c1 0 0 0.087 0.594 0.319

C := 0 c2 0 C = 0.148 0.484 0.336 A

0 0 c 0.24 0.083 0.323
3

(Note that the minimum index value for matrices in this document
is set at 1, not zero.)

1 t 2 t 3 t
Define current I1( t) := C e +C e +C e
1, 1 1, 2 1, 3
functions
1 t 2 t 3 t
I2( t) := C e +C e +C e
2, 1 2, 2 2, 3

1 t 2 t 3 t
I3( t) := C e +C e +C e
3, 1 3, 2 3, 3
 SLIDE # 74

Check the current values at a few points

I1( 0 s ) = 1 A I1( 0.1 s ) = 0.803A I1( 10 s ) = 0.23A

I2( 0) = 0 A I2( 0.1 s ) = 0.021A I2( 10 s ) = 0.242A

I3( 0) = 0 A I3( 0.1 s ) = 0.214A I3( 10 s ) = 0.233A

 SLIDE # 75

Plot the current values 9 7

t := 0 , 10 s .. 20 10 s
,

I1( t ) 0.5

I2( t )

I3( t )
0

0.5
7 6 6 6
0 5 .10 1 .10 1.5 .10 2 .10
t
Fig. 2. Winding Currents.

I1 drops quickly as the current builds up in the other two windings. All
of the currents then decay.
 SLIDE # 76

I1( t)
Define a current vector to facilitate
calculating the stored energy. I( t) := I2( t)
I ( t)
3

1 T
Define a function to compute the stored energy. W ( t) := ( I( t) ) L I( t)
2
 SLIDE # 77

7 5
t := 0 , 10 s .. 10 s

6
5 .10

6
4 .10
W( t )
6
3 .10

6
2 .10
6 6 6 6 5
0 2 .10 4 .10 6 .10 8 .10 1 .10
t
Fig. 3. Stored energy, J.

The stored energy is dissipated as expected.

 SLIDE # 78

We can now examine a case where the coupling coefficient are improperly
specified. Are relatively small change in one of the couple coefficients is all
that is required to create an unstable configuration.

Enter the values of L1 := 10 H k12 := 0.96

the inductances and
coupling coefficients
L2 := 11 H k23 := 0.99

L3 := 10 H k13 := 0.99
(Previously 0.98)
 SLIDE # 79

1 k12 k13
Construct the
coupling K := k12 1 k23
coefficient matrix. k k 1
13 23

2.96
Compute the eigenvalues of
eigenvals ( K) = 0.04
the coupling coefficient
matrix. 5
6.757 10

One of the eigenvalues is negative. As is shown below on, this makes

one of the eigenvalues for the time response of the system positive.
 SLIDE # 80

L1 K L1 L2 K L1 L3
1, 2 1, 3
Construct the inductance L := K1 , 2 L1 L2 L2 K L2 L3
2, 3
matrix.
K L1 L3 K L2 L3 L3
1 , 3 2 , 3

10.000 10.069 9.900

Evaluate the inductance L = 10.069 11.000 10.383 H
matrix.
9.900 10.383 10.000

Enter the values of the R1 := 1 R2 := 1 R3 := 1

resistances
 SLIDE # 81

Enter the initial values of the I1i := 1 A I2i := 0 A I3i := 0 A

currents

I1i 1

Define an initial condition Ii := I2i Ii = 0 A

vector I 0
3i

Compute the stored energy T 5

Ii L Ii = 1.000 10 J
 SLIDE # 82

Compute the inverse 248.75 239.557 495

= 239.557 226.136 471.964
1 1
inductance matrix := L
H
495 471.964 980

1.457 103
1
Compute the eigenvalues of . eigenvals ( ) = 2.385
H
0.033
 SLIDE # 83

R1 0 0

Define G G := 0 R2 0 (4)
0 0 R
3

1.457 109

= 2.385 106 s
Compute the eigenvalues of G := eigenvals ( G) -1

3.27 104

One of the eigenvalues is positive, so the system is unstable.

 SLIDE # 84

Compute the eigenvectors of G

0.414 0.713 0.566
:= eigenvecs ( G) = 0.395 0.701 0.594 s
-1

0.82 0.023 0.572

1 t
The solution to (5) will have the form c1 0 0 e

I 0 c2 0 e 2 t (6)
0 0 c
3 3 t
e
 SLIDE # 85

The values of the constants c1, c2 and c3 can be determined from the initial
conditions

c1 c1 c1 0.414
1
c2 2 c2 = 0.713 A s
I0 c := Ii
c c c 0.566
3 3 3

c1 1 1

Check to see if the initial conditions c2 = 0 A Ii = 0 A
are satisfied c 0
0
3
 SLIDE # 86

Compute a coefficient matrix

c1 0 0 0.172 0.508 0.32

C := 0 c2 0 C = 0.164 0.5 0.336 A
0 0 c
3 0.34 0.016 0.323

1 t 2 t 3 t
Define current I1( t) := C e +C e +C e
1, 1 1, 2 1, 3
functions
1 t 2 t 3 t
I2( t) := C e +C e +C e
2, 1 2, 2 2, 3

1 t 2 t 3 t
I3( t) := C e +C e +C e
3, 1 3, 2 3, 3
 SLIDE # 87

Check the current values at a few points

I1( 0.01 s ) = 4 10 A I1( 0.1 s ) = 3.344 10 A

5 62
I1( 0 s ) = 1 A

I2( 0.01 s ) = 3 10 A I2( 0.1 s ) = 3.189 10 A

5 62
I2( 0 s ) = 0 A

I3( 0.01 s ) = 7 10 A I3( 0.1 s ) = 6.621 10 A

5 62
I3( 0 s ) = 0 A
 SLIDE # 88

Plot the current values 11 9

t := 0 , 10 s .. 4.2 10 s

2
I1( t )

I2( t )
0
I3( t )

10 9 9
0 5 .10 1 .10 1.5 .10
t
Fig. 4. Winding Currents.

I1 rises instead of decaying. The other currents start at zero

and then build up.
 SLIDE # 89

I1( t)
Define a current vector to facilitate calculating
the stored energy. I( t) := I2( t)
I ( t)
3

1 T
Define a function to compute the stored energy. W ( t) := ( I( t) ) L I( t)
2
 SLIDE # 90

6
5 .10

W( t ) 0

6
5 .10
9 9 9 9 9
0 1 .10 2 .10 3 .10 4 .10 5 .10
t

Fig. 5. Stored energy.

The stored energy decays to zero, but it then goes negative as our imaginary inductor
pumps out energy at a rapidly-increasing rate.

One of the things that we can learn from this example is that estimating values for
coupling coefficients can easily produce a nonphysical circuit. Trying to simulate
such a circuit may produce frustration as one tries to figure out why the circuit won't
converge.
Acknowledgement
Thanks to my manager, Dr. David J. Christie, for reviewing the
presentation and providing helpful comments and suggestions
Also, thanks to Dr. James H. Spreen for his continuing guidance
 SLIDE # 92

NASDAQ: AEIS
www.advanced-energy.com

Advanced Energy Industries, Inc. All Rights Reserved.