NONLINEAR PHENOMENA

IN POWER ELECTRONICS

Soumitro Banerjee

Department of Electrical Engineering

Indian Institute of Technology
Kharagpur 721302, India

NONLINEAR PHENOMENA. . . 1
 All power electronic circuits share the following properties:

Switches make the circuit toggle between two or more different

topologies (different sets of differential equations) at different
times.

Storage elements (inductors and capacitors) absorb energy

from a circuit, store it and return it.

The switching times are nonlinear functions of the variables to

be controlled (mostly the output voltage).

The basic source of nonlinearity:

feedback controlled switching

NONLINEAR PHENOMENA. . . 2
 In addition there are parasitic nonlinearities

1. the nonlinear v i characteristics of switches,

2. nonlinear inductances and capacitances,

3. electromagnetic couplings between components.

However, the main source of nonlinearity is the ubiquitous

switching element which makes all power electronic systems
strongly nonlinear even if all components are assumed to be ideal.

Therefore

Power electronics engineers/researchers are invariably dealing

with nonlinear problems.

NONLINEAR PHENOMENA. . . 3
 Example 1: The voltage-mode controlled buck converter

Vin Vo

NONLINEAR PHENOMENA. . . 4
 Example 1: The voltage-mode controlled buck converter

Vin Vo

Vramp

Vcon Vref

NONLINEAR PHENOMENA. . . 5
 Example 1: The voltage-mode controlled buck converter

Vin Vo

Vramp

Vcon Vref

NONLINEAR PHENOMENA. . . 6
 Nominal periodic behavior

0 0.001 0.002 0.003 0.004

(a)
But such a behavior occurs only within certain limits of the external
parameters.

NONLINEAR PHENOMENA. . . 7
 Change of behaviour due to fluctuation of input voltage:

0 0.001 0.002 0.003 0.004

(b)
The change in behavior occurred when the input voltage changed
from Vin = 24 V to Vin = 25 V.

NONLINEAR PHENOMENA. . . 8
 When Vin is increased to 35V, the behavior becomes aperiodic, or
chaotic.

NONLINEAR PHENOMENA. . . 9
 (a) (b)

(a) The experimental period-2 attractor

and (b) the chaotic attractor.

NONLINEAR PHENOMENA. . . 10
 CHAOS

Aperiodic waveform

Seemingly random, noise-like behavior

Completely deterministic

The orbit is sensitively dependent on the initial condition

Statistical behaviour (average values of state variables, power

spectrum etc.) completely predictable.

Unstable at every equilibrium point, but globally stable.

Waveform bounded.

NONLINEAR PHENOMENA. . . 11
 Example 2: The current mode controlled boost converter

L D
V in S C R

i Switch off when

+
R i=I ref
I ref Q
S
Switch on at the
Clock next clock

Instability and transition to period-2 subharmonic at the critical duty

ratio of 0.5.

NONLINEAR PHENOMENA. . . 12
 How to probe such phenomena?

Averaged model
dx
= f (x, , t)
dt

x(t)

(t)

Simple, but details destroyed. Eliminates nonlinear effects.

NONLINEAR PHENOMENA. . . 13
 Sampled-data (discrete) model

xn+1 = f (xn , )

Continuous Discrete
time orbit Observations
y
ac te
sp sta
e

t
0 T 2T 3T 4T

x
Relatively complex, but accurate. Captures nonlinearity.

NONLINEAR PHENOMENA. . . 14
 vcon vramp
(vn,i n) (vn+1,i n+1) (vn+2,i n+2)

off on off on

Sampled data model: (vn , in ) 7 (vn+1 , in+1 )

NONLINEAR PHENOMENA. . . 15
 The procedure: stacking of solutions

Start from an initial condition (xn , yn ) at a clock instant.

Using the on-time equation and the value of Iref , obtain the
length of the on-period.

Obtain the state vector at the end of the on-period.

Use this state vector as the initial condition in the off-time

equations, and evolve for (T Ton ). This gives (xn+1 , yn+1 ) at
the next clock instant.

Thus we obtain the map

(xn+1 , yn+1 ) = f (xn , yn )

NONLINEAR PHENOMENA. . . 16
 Dynamics in discrete time

Iterate the map starting from any initial condition.

Obtain a sequence of points in the discrete state space.
Plot the discrete-time evolution, called the phase-portrait.

state
space (x n+3,yn+3 )
(x n+2,yn+2 )
(x n+1,yn+1 )
f (x n ,yn)

(x n ,yn)

NONLINEAR PHENOMENA. . . 17
 For periodic systems, after some initial transient, all the iterates fall
on the same point in the discrete state space. The fixed point of
the map is stable period-1 attractor.

y
control voltage

time x

Continuous-time Discrete-time

NONLINEAR PHENOMENA. . . 18
 If the system is period-2 (the same state repeats after 2 clocks),
there will be 2 points in the discrete-time state space
period-2 attractor.

y
control voltage

time x

Continuous-time Discrete-time

NONLINEAR PHENOMENA. . . 19
 If the system is period-n (the same state repeats after n clocks),
there will be n points period-n attractor.

If a system is chaotic, there will be an infinite number of points in

the phase portrait. chaotic attractor.

NONLINEAR PHENOMENA. . . 20
 The phase portrait for the buck converter in the chaotic mode.

NONLINEAR PHENOMENA. . . 21
 Question: Why, and in what ways, does the system behaviour
change with the change in a parameter?

Studied through Bifurcation diagrams

(panoramic view of stability status).

Sampled variable at steady state versus parameter, e.g.,

iL (nT ) vs. R.

Bifurcation diagrams can be plotted against the variation of any

of the system parameters, e.g., Vin , R, Iref etc.

NONLINEAR PHENOMENA. . . 22
 An experimental bifurcation diagram of the buck converter.
x-coordinate: input voltage Vin ,
y-coordinate: sampled value of the control voltage.
Parameter values are: R = 86, C = 5F, L = 2.96mH, VU = 8.5V,
VL = 3.6V, clock speed 11.14 kHz.

NONLINEAR PHENOMENA. . . 23
 What is a bifurcation?

Nothing but loss of stability.

In a linear system, a loss of stability means the system

collapses.

In a nonlinear system, when a periodic orbit loses stability,

some other orbit may become stable.

Thus, at a specific parameter value, one may observe a

qualitative change in the character of the orbit. This is a
bifurcation.

NONLINEAR PHENOMENA. . . 24
 Bifurcation: qualitative change in system behaviour with the
change of a parameter

Study of bifurcations:

Derive a discrete time map xn+1 = f (xn , ).

Obtain the fixed point xn+1 = xn .

Examine the Jacobian at the fixed point and find the loci of the
eigenvalues when a bifurcation parameter is varied.

Identify the condition for the eigenvalue(s) moving out the unit circle in
the complex plane.

NONLINEAR PHENOMENA. . . 25
 Two types of bifurcation seen in power electronics

Smooth bifurcations (found in other systems as well)

Border collision (characteristic of power electronics)

Abrupt change of behavior due to a structural change

Structural change in switching converters = Alteration in topological

sequence e.g., change of operating mode, reaching a saturation
boundary.

NONLINEAR PHENOMENA. . . 26
 For voltage

control voltage
mode control:

time
control voltage

Loss of stability,
No structural change
(smooth bifurcation)

control voltage time

Structural change
(border collision bifurcation)

time

NONLINEAR PHENOMENA. . . 27
 on off on off on off on off on off on off

Loss of stability, No structural change

(Smooth bifurcation)

on off on off on off on off on off on off

NONLINEAR PHENOMENA. . . 28
 on off on off on off on off on off on off

Change of topological sequence

(involves Border collision bifurcation)

on on off on on off on on off

NONLINEAR PHENOMENA. . . 29
 The two classes of bifurcations

Smooth bifurcations Border collision

Loss of stability without Loss of stability due to struc-
structural change tural change

Standard appearance of bi- Non-standard appearance in

furcation diagrams, e.g., pe- bifurcation diagrams, e.g.,
riod doubling cascade, peri- bendings, sudden transition
odic windows etc. to chaos.

NONLINEAR PHENOMENA. . . 30
 How to analyse such loss of stability of power converters?

Bifurcation theory can help.

A bifurcation occurs when a fixed point of the discrete map loses

stability.

= At least one eigenvalue crosses the unit circle.

NONLINEAR PHENOMENA. . . 31
 If the map is smooth, there are three possibilities:

* * *
** * * **
*

(a) (b) (c)

(a) A period doubling bifurcation:

eigenvalue crosses the unit circle on the negative real line,

(b) A saddle-node or fold bifurcation:

an eigenvalue touches the unit circle on the positive real line,

(c) A Hopf or Naimark-Sacker bifurcation:

a complex conjugate pair of eigenvalues cross the unit circle.

NONLINEAR PHENOMENA. . . 32
 control voltage
on off on off on off on off on off on off
time

control voltage
on off on off on off on off on off on off time

Examples of period-doubling bifurcation, associated with one

eigenvalue becoming 1.

Common in all feedback-controlled dc-dc converters.

NONLINEAR PHENOMENA. . . 33
 A saddle-node bifurcation results in the creation of a new orbit (or
the destruction of an existing orbit).

A new periodic orbit coming into being may imply

Destruction of chaos for a range of parameters (periodic

window)

Creation of coexisting attractors (multistability).

NONLINEAR PHENOMENA. . . 34
 A Naimark-Sacker bifurcation causes a periodic orbit change into a
quasiperiodic orbit.

x1
x1
x3
t x2

(Combination of two incommensurate frequencies.)

In discrete time,

NONLINEAR PHENOMENA. . . 35
 Control Voltage (V)
Control Voltage (V) 6 6

4 4

2 2

0 0
0.2 0.6 1.0 1.4 1.8 2.2 0.2 0.6 1.0 1.4 1.8 2.2
Time (msec) Time (msec)
80 80

i n (mA)
i n (mA)

0 0

60 60
9 3 15 9 3 15
vcon (V) v con (V)

Quasiperiodic and mode-locked periodic behavior in a PWM buck

converter. Slow scale instability. Can also be predicted with the
averaged model.

NONLINEAR PHENOMENA. . . 36
 These are the three possible ways that a converter can lose
stability without structural change, i.e., without alteration of
topological sequence.

How can we analyse the loss of stability caused by structural

change?

NONLINEAR PHENOMENA. . . 37
 It has been found that sampled data modeling of all power
electronic circuits yield piecewise smooth maps.

The discrete state space is divided into two or more

compartments with different functional forms of the map.
The compartments are separated by borderlines. The
Jacobian changes discontinuously across the borderlines.

Borderline
xn+1 =f2 ( x n )
x2

xn+1 =f1 ( x n )

x1

NONLINEAR PHENOMENA. . . 38
 The current mode controlled boost converter

L D
V in S C R

i Switch off when

+
R i=I ref
I ref Q
S
Switch on at the
Clock next clock

I ref

NONLINEAR PHENOMENA. . . 39
 Structure of discrete state space in
in current mode controlled converters

Iref Iref

i
i i

T T T
(a) (b) (c)

(a) and (b): The two possible types of evolution between two
consecutive clock instants yielding two different functional forms in
the sampled-data model and (c): the borderline case.
7 Piecewise smooth map.

NONLINEAR PHENOMENA. . . 40
 Iref Iref

i
i i

T T T
(a) (b) (c)

 
m2 m2
Case (a): in+1 = f1 (in ) = 1 + m1 Iref m2 T m1 in .
Case (b): in+1 = f2 (in ) = in + m1 T
Boderline: Ib = Iref m1 T

in+1

Slo
1

pe
e
op

Sl

m2
/m
1
in

NONLINEAR PHENOMENA. . . 41
 Structure of discrete state space
in voltage mode controlled converters

(a) (b) (c)

(d) (e)

(a), (b) and (c): The three possible types of evolution between two
consecutive clock instants, and (d) and (e): the grazing situations
that create the borderlines.
7 Piecewise smooth map.

NONLINEAR PHENOMENA. . . 42
 Dynamics of Piecewise Smooth Maps

If a fixed point loses stability while in either side, the resulting

bifurcations can be categorized under the generic classes for
smooth bifurcations.
But what if a fixed point crosses the borderline as some
parameter is varied?

The Jacobian elements discretely

change at this point

NONLINEAR PHENOMENA. . . 43
 The eigenvalues may jump from any value to any other value
across the unit circle.

The resulting bifurcations are called

Border Collision Bifurcations.

Continuous movement of Discontinuous jump of

eigenvalues in a smooth eigenvalues in a border
bifurcation collision bifurcation

NONLINEAR PHENOMENA. . . 44
 Theories have been developed that can predict the outcome of a border
collision bifurcation (which orbit will become stable) depending on the
eigenvalues of the Jacobian matrix at the two sides of the border. These
transitions are usually sudden and abrupt.

x x x

x x x

NONLINEAR PHENOMENA. . . 45
 As a system parameter is varied, the change of dynamical
behavior of power converters exhibit a succession of smooth and
border collision bifurcations.

Generally the first bifurcation from a normal period-1 operation

is of smooth type.

The development of a smooth bifurcation sequence (e.g., a

period-doubling cascade) is interrupted by border collision
bifurcations.

NONLINEAR PHENOMENA. . . 46
 Smooth
Sampled inductor current Bifurcation

Border
Collision

Load resistance

Bifurcation diagram of the boost converter with load resistance as

variable parameter.

NONLINEAR PHENOMENA. . . 47
 Which type of instability is likely in which converter?

1. Voltage mode controlled buck converter (PWM-2): The first

bifurcation is always period doubling. Subsequently period
doubling cascade interrupted by border collision, leading to
direct transition to chaos. Hamill and Deane (1990,1992),
Fossas and Olivar (1996), Chakrabarty and Banerjee (1996)

2. Voltage mode controlled boost converter (PWM-2): Hopf

bifurcation leading to quasiperiodicity. Tse (1997)

3. Voltage mode controlled converters (PWM-1): Hopf bifurcation

leading to quasiperiodic and mode-locked periodic orbits.
Period doubling also occurs. El Aroudi et al. (2000), Maity and
Banerjee (2007)

NONLINEAR PHENOMENA. . . 48
 4. Current mode controlled boost converter: The first bifurcation
is always period doubling. Subsequently period doubling
cascade interrupted by border collision, leading to direct
transition to chaos. Hamill and Deane (1990,1992,1995),
Chakrabarty and Banerjee (1997), Chan and Tse (1997),

5. Current mode controlled Cuk converter: Period doubling route

to chaos, truncated by border collision. Tse and Chan (1995)

6. Converters operated in DCM: Period doubling. Tse (1994)

7. Converters undergoing transition from DCM to CCM: border

collision bifurcation leading to quasiperiodicity. Maity and
Banerjee (2007)

NONLINEAR PHENOMENA. . . 49
 In many applications the operating point constantly changes with
time:
Power factor correction power supplies;
Dc-AC inverters;
Audio amplifiers using dc-dc converters.
Bifurcations are expected even under normal operating condition.
6

4
Current [A]

-2
0.08 0.085 0.09 0.095 0.1
Time [s]

NONLINEAR PHENOMENA. . . 50
 How to analyze the stability of power electronic circuits?

If the initial condition is perturbed and the solution converges back

to the orbit, then the orbit is stable. The stability margin can be
assessed from the rate of convergence.

Suppose the initial condition is given a perturbation x(t0 ). If the

original trajectory and the perturbed trajectory evolve for a time t,

x(t) = x(t0 ).

Here is the state transition matrix.

NONLINEAR PHENOMENA. . . 51
 The properties of :

xA xB xC

If xB = AB xA , xC = BC xB

then xB = AB xA , xC = BC xB

and xC = AC xA , AC = BC AB

NONLINEAR PHENOMENA. . . 52
 For LTI systems, the state transition matrix can be obtained as the
matrix exponential
(t, t0 ) = eA(tt0 )
so that the perturbation at time t can be written as

x(t) = eA(tt0 ) x0 .

NONLINEAR PHENOMENA. . . 53
 Suppose we are able to find the state transition matrices during the
ON period and the OFF period:

x(dT ) = A1 x(0)
x(T ) = A2 x(dT )

The product A2 A1 does not give the state transition matrix over
the whole cycle. One has to take into account how the
perturbations change when they cross the border.

M. A. Aizerman and F. R. Gantmakher, On the stability of periodic motions, Journal

of Applied Mathematics and Mechanics (translated from Russian), 1958, pages
1065-1078.

NONLINEAR PHENOMENA. . . 54
 x(tB+ ) = S x(tB ),

where S, the jump matrix or saltation matrix, can be expressed

as
(f + f )nT
S=I+ T .
n f + h/t

f : RHS of the differential equations before switching

f+ : RHS of the differential equations after switching
h(x, t) = 0: the switching condition (a surface in the state space)
I: the identity matrix
n: the vector normal to the switching surface.

NONLINEAR PHENOMENA. . . 55
 The state transition matrix over the complete clock cycle, called the
monodromy matrix is expressed as

cycle (T, 0) = S 2 off (T, dT ) S 1 on (dT, 0),

where S 1 is the saltation matrix related to the first switching event,

and S 2 is that related to the second switching event.

NONLINEAR PHENOMENA. . . 56
 The monodromy matrix then relates the perturbation at the end of
the clock period to that at the beginning:

x(T ) = (T, 0)x(0).

The eigenvalues of the monodromy matrix are also called the

Floquet multipliers. If all the eigenvalues are inside the unit circle,
perturbations will die down and the system is stable.

NONLINEAR PHENOMENA. . . 57
 The voltage mode controlled buck converter:

vcon < vramp

s = closed

v v(t) v(t)
in dv(t) i(t) R

, S is conducting

di(t) L
= , =
dt v(t) dt C
, S is blocking.

L

NONLINEAR PHENOMENA. . . 58
 control voltage
time

0.62

0.6 1 2 3 4 X(0) 6 7 8 9

0.58 1,9

0.56 8 2
5
current, A

0.54
7 3
0.52

6 4
0.5

0.48 5

X(dT)
0.46

0.44
11.4 11.5 11.6 11.7 11.8 11.9 12
voltage, V

NONLINEAR PHENOMENA. . . 59
 The switching hypersurface (h) is given by

vramp (t)
h(x(t), t) = x1 (t) Vref = 0,
A
 
t
vramp (t) = VL + (VU VL ) mod 1
T

The normal to the hypersurface is:

h(x(t), t)
x1 (t) 1

n = h(x(t), t) = =

h(x(t), t) 0
x2 (t)

NONLINEAR PHENOMENA. . . 60
 By defining x1 (t) = v(t) and x2 (t) = i(t), the system equations are

A x + Bu, A(x (t) V ) < v
s 1 ref ramp (t),
x =
As x, A(x1 (t) Vref ) > vramp (t).

Where,

1/RC 1/C 0
As = , Bu = Vin
1/L 0 1/L

NONLINEAR PHENOMENA. . . 61
 When the state goes from the off state to the on state,

x2 (t )/C x1 (t )/RC
f p = lim f (x(t)) = ,
tt x1 (t )/L

x2 (t )/C x1 (t )/RC
f p+ = lim f + (x(t)) = .
tt (Vin x1 (t ))/L

where t is the switching instant.

NONLINEAR PHENOMENA. . . 62
 Thus
0
f p+ f p = Vin ,
L

0 0
T


f p+ f p n = Vin ,
0
L

x2 (t ) x1 (t )
nT f p = .
C RC

NONLINEAR PHENOMENA. . . 63
  
T VL + (VU VL )t
x1 (t) Vref
h(x(t), t) AT
=
t t
VU VL
= .
AT
Hence the saltation matrix is calculated as

1 0
Vin /L

S=

1
x2 (t ) x1 (t )/R VU VL


C AT

NONLINEAR PHENOMENA. . . 64
 For a buck converter with the parameters

Vin = 24V , Vref = 11.3V , L = 20mH, R = 22, C = 47F,

A = 8.4, T = 1/2500s, VL = 3.8V and VU = 8.2V

the switching instant was calculated to be 0.4993 T .

The state at the switching instants are

12.0222 12.0139
x(0) = and x(d T ) = .
0.6065 0.4861

NONLINEAR PHENOMENA. . . 65
 The saltation matrix is calculated as

1 0
S=
0.4639 1

The state matrix is

1/RC 1/C 967.12 21276.6
As = = .
1/L 0 50 0

NONLINEAR PHENOMENA. . . 66
 The state transition matrices for the two pieces of the orbit are:

1. Off period:

As d T 0.8058 3.8366
(d T, 0) = e =
0.0090 0.9802

2. On period:

As dT 0.8052 3.8468
(T, d T ) = e = .
0.0090 0.9800

NONLINEAR PHENOMENA. . . 67
 Hence the monodromy matrix is

(T, 0, x(0)) = (T, d T ) S (d T, 0)

0.8238 0.0131
=
0.3825 0.8184

The eigenvalues are 0.8211 0.0708j implying that at the above

parameter values the system is stable.

NONLINEAR PHENOMENA. . . 68
 An idea for increasing the stability margin:

(T, 0, x(0)) = (T, d T ) S (d T, 0)

The state transition matrices for the ON and OFF periods are given
by the parameters and the duty ratio. Set by the users specs.

There is another handle: the saltation matrix.

(f + f )nT
S=I+ T .
n f + h/t

NONLINEAR PHENOMENA. . . 69
 h/t can be manipulated by varying the slope of the ramp.

nT can be manipulated by using current feedback along with

voltage feedback.

14 14
Control Voltage (V)

Control Voltage (V)

10 10

6 6

2 2

2 2
22 24 26 28 30 32 34 22 24 26 28 30 32 34
Input Voltage (V) (a) Input Voltage (V) (b)

The experimentally obtained bifurcation diagrams (a) under normal

PWM-2 control, and (b) when the secondary control loop is added,
with VU = 0.7 V.

NONLINEAR PHENOMENA. . . 70
 Take home message:

All power electronic circuits are strongly nonlinear systems.

Power converters may exhibit subharmonics and chaos for

specific parameter ranges. Linear analysis does not predict
these instabilities.

Smooth as well as nonsmooth (or border collision) bifurcations

occur in such converters.

To analyse the smooth bifurcations, one has to obtain the

eigenvalues of the monodromy matrix. To analyse the border
collision bifurcations, one has to obtain the eigenvalues of the
same periodic orbit before and after border collision.

Bifurcation theory helps in delimiting the parameter space and

in devising better controllers to avoid instability,.

NONLINEAR PHENOMENA. . . 71
 For further reading:

1. Soumitro Banerjee and George C. Verghese, (Editors)

Nonlinear Phenomena in Power Electronics: Attractors,
Bifurcations, Chaos, and Nonlinear Control, IEEE Press,
2001.

2. C. K. Tse, Complex Behavior of Switching Power Converters,

CRC Press, USA, 2003.

3. Z. T. Zhusubaliyev and E. Mosekilde, Bifurcations and Chaos

in Piecewise-Smooth Dynamical Systems, World Scientific,
Singapur, 2003.

NONLINEAR PHENOMENA. . . 72
 THANK YOU

NONLINEAR PHENOMENA. . . 73