348

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

Coupled Lines

(10.1.1)

I2
I2
I1
V2
V1
V2
= L2
Lm
,
= C2
+ Cm
z
t
t
z
t
t
When Lm = Cm = 0, they reduce to the uncoupled equations describing the isolated
individual lines. Eqs. (10.1.1) may be written in the 22 matrix forms:


L1 Lm I
V
=
Lm L2 t
z
(10.1.2)


C1 Cm V
I
=
C2
Cm
z
t
where V, I are the column vectors:

10.1 Coupled Transmission Lines

Coupling between two transmission lines is introduced by their proximity to each other.
Coupling effects may be undesirable, such as crosstalk in printed circuits, or they may
be desirable, as in directional couplers where the objective is to transfer power from one
line to the other.
In Sections 10.110.3, we discuss the equations, and their solutions, describing coupled lines and crosstalk [486503]. In Sec. 10.4, we discuss directional couplers, as well
as ber Bragg gratings, based on coupled-mode theory [504525]. Fig. 10.1.1 shows an
example of two coupled microstrip lines over a common ground plane, and also shows
a generic circuit model for coupled lines.

I1
V1
V2
= C1
+ Cm
z
t
t

V1
I1
I2
= L1
Lm
,
z
t
t

10

V=

V1
V2


I=

I1
I2


(10.1.3)

For sinusoidal time dependence ejt , the system (10.1.2) becomes:

dV
= j
dz
dI
= j
dz

L1
Lm

Lm
L2

C1
Cm

Cm
C2

(10.1.4)
V

It proves convenient to recast these equations in terms of the forward and backward
waves that are normalized with respect to the uncoupled impedances Z1 , Z2 :

a1 =
a2 =

V1 + Z1 I1


,
2Z1

b1 =

V2 + Z2 I2


,
2Z2

b2 =

V1 Z1 I1


2Z1

V2 Z2 I2


2Z2

a=

a1
a2


,

b=

b1
b2


(10.1.5)

The a, b waves are similar to the power waves dened in Sec. 12.7. The total average
power on the line can be expressed conveniently in terms of these:
Fig. 10.1.1 Coupled Transmission Lines.

For simplicity, we assume that the lines are lossless. Let Li , Ci , i = 1, 2 be the
distributed inductances and capacitances per unit length when the lines are isolated from
each other. The corresponding propagation velocities and characteristic impedances




are: vi = 1/ Li Ci , Zi = Li /Ci , i = 1, 2. The coupling between the lines is modeled
by introducing a mutual inductance and capacitance per unit length, Lm , Cm . Then, the
coupled versions of telegraphers equations (9.15.1) become:
C is related to the capacitance to ground C
1
1g via C1 = C1g + Cm , so that the total charge per unit
length on line-1 is Q1 = C1 V1 Cm V2 = C1g (V1 Vg )+Cm (V1 V2 ), where Vg = 0.

P=

1
1
1
Re[V I]= Re[V1 I1 ]+ Re[V2 I2 ]= P1 + P2
2
2
2


 
 
 

= |a1 |2 |b1 |2 + |a2 |2 |b2 |2 = |a1 |2 + |a2 |2 |b1 |2 + |b2 |2
= a a b b

where the dagger operator denotes the conjugate-transpose, for example, a = [a

1 , a2 ].
Thus, the a-waves carry power forward, and the b-waves, backward. After some algebra,
it can be shown that Eqs. (10.1.4) are equivalent to the system:

da
= jF a + jG b
dz
db
= jG a + jF b
dz

347

(10.1.6)

d
dz

a
b


= j

F
G

G
F



a
b


(10.1.7)

www.ece.rutgers.edu/orfanidi/ewa

349

with the matrices F, G given by:


F=


G=


(10.1.8)

where 1 , 2 are the uncoupled wavenumbers i = /vi = Li Ci , i = 1, 2 and the
coupling parameters , are:







1
1
Lm
Lm
Cm
=

Cm Z1 Z2 =
1 2



2
2
Z1 Z2
L1 L2
C1 C2
(10.1.9)







Lm
Cm
1
1
Lm






+ Cm Z1 Z2 =
1 2
+
=
2
2
Z1 Z2
L1 L2
C1 C2
A consequence of the structure of the matrices F, G is that the total power P dened
in (10.1.6) is conserved along z. This follows by writing the power in the following form,
where I is the 22 identity matrix:


P = a a b b = [a , b ]



a
b

350

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

where a0 , a1 are real such that |a0 | = |a1 |. Such matrices form a commutative subgroup
of the group of nonsingular 22 matrices. Their eigenvalues are = a0 a1 and they
can all be diagonalized by a common unitary matrix:
1
2



F
G

G
F





V Z0 I
b=

2Z0


F=

a=

V + Z0 I
,
2


G=

G
F

 

a
b

Lm
Cm

L0
C0

V Z0 I
2

=0

a0
a1

a1
a0


= a0 I + a1 J ,

I=

1
0

0
1


,

J=

=
A

A+

0
1


=

d2 a
= B2 a ,
dz2

1
e =
2

1
1

(10.1.12)

1
0

(10.1.14)

a0 + a1

a0 a1

(10.1.15)

(10.1.16)


(10.1.13)

d2 b
= B2 b
dz2

However, it is better to work with (10.1.7) directly. This system can be decoupled by
forming the following linear combinations of the a, b waves:
A = a b
B = b a

(10.1.11)

Using the property FG = GF, and differentiating (10.1.7) one more time, we obtain
the decoupled second-order equations, with B as dened in (10.1.16):

(10.1.10)

The matrices F, G commute with each other. In fact, they are both examples of
matrices of the form:

A=

1
1

Such matrices, as well as any matrix-valued function thereof, may be diagonalized

simultaneously. Three examples of such functions appear in the solution of Eqs. (10.1.7):

Lm
Cm
+
L0
C0

Z0 I)(Z
+ Z0 I)1 Q
= (Z Z0 I)(Z + Z0 I)1 = Q(Z

where, for simplicity, we removed the common scale factor 2Z0 from the denominator
of a, b. The parameters , are obtained by setting Z1 = Z2 = Z0 in (10.1.9):

1
e+ =
2

= [e+ , e ] ,




+ G)(
G)Q
B = (F + G)(F G) = Q (F


F
1 Q
+ G)(
G)
Z = Z0 (F + G)(F G)1 = Z0 Q (F

F
G

b=

1
1

,
A = Q AQ

the latter following from the conditions F = F and G = G. Eqs. (10.1.6) and (10.1.7)
form the basis of coupled-mode theory.
Next, we specialize to the case of two identical lines that have L1 = L2 L0 and




C1 = C2
C0 , so that 1 = 2 = L0 C0 and Z1 = Z2 = L0 /C0 Z0 , and speed
v0 = 1/ L0 C0 . Then, the a, b waves and the matrices F, G take the simpler forms:
V + Z0 I
a=

,
2Z0

1
1

so that we have QQ = Q Q = I and Ae = e .

The eigenvectors e are referred to as the even and odd modes. To simplify subsequent expressions, we will denote the eigenvalues of A by A = a0 a1 and the
. Thus,
diagonalized matrix by A

Using (10.1.7), we nd:

dP
= j[a , b ]
dz

Q=

A
B


=



a
b


(10.1.17)

The A, B can be written in terms of V, I and the impedance matrix Z as follows:

A = (2D)1 (V + ZI)
1

B = (2D)

(V ZI)

V = D(A + B)

ZI = D(A B)

D=

Z + Z0 I
2Z0

(10.1.18)

Using (10.1.17), we nd that A, B satisfy the decoupled rst-order system:

d
dz

A
B


= j



A
B

dA
= jBA ,
dz

dB
= jBB
dz

(10.1.19)

with solutions expressed in terms of the matrix exponentials ejBz :

A(z)= ejBz A(0) ,

B(z)= ejBz B(0)

(10.1.20)

www.ece.rutgers.edu/orfanidi/ewa

351



V(z) = D ejBz A(0)+ejBz B(0)


ZI(z) = D ejBz A(0)ejBz B(0)

VG =

G = (ZG I Z)(ZG I + Z)1

L = (ZL I Z)(ZL I + Z)1

(10.1.22)

V(l)= ZL I(l)

(10.1.23)

B(l)= L A(l)

(10.1.24)

But from (10.1.19), we have:

ejBl B(0)= B(l)= L A(l)= L ejBl A(0)

B(0)= L e2jBl A(0)

(10.1.25)

Inserting this into (10.1.24), we may solve for A(0) in terms of the generator voltage:
1

A(0)= D


2jBl 1

I G L e

Z(Z + ZG I)

VG

(10.1.26)

Using (10.1.26) into (10.1.21), we nally obtain the voltage and current at an arbitrary
position z along the lines:
V(z) = [ejBz + L e2jBl ejBz
I(z) = [ejBz L e2jBl ejBz

I G L e2jBl
I G L e2jBl

1
1

Z(Z + ZG I)1 VG
(Z + ZG I)1 VG

(10.1.27)

These are the coupled-line generalizations of Eqs. (9.9.7). Resolving VG and V(z)
into their even and odd modes, that is, expressing them as linear combinations of the
eigenvectors e , we have:
VG = VG+ e+ + VG e ,

where

V(z)= V+ (z)e+ + V (z)e ,

VG =

VG1 VG2

V (z)=

V1 (z)V2 (z)

(10.1.28)

In this basis, the matrices in (10.1.27) are diagonal resulting in the equivalent solution:
V(z)= V+ (z)e+ + V (z)e =

ej+ z + L+ e2j+ l ej+ z

Z+
VG+ e+
1 G+ L+ e2j+ l
Z+ + ZG

Z
ej z + L e2j l ej z
VG e
1 G L e2j l
Z + ZG

The

matrices D, Z, G , L , , B all commute with each other.

L =

ZL Z
ZL + Z

(10.1.30)

V1 (z) =

ej z + L e2j l ej z
ej+ z + L+ e2j+ l ej+ z
V
V+ +

2
j
l
+
1 G+ L+ e
1 G L e2j l

V2 (z) =

ej z + L e2j l ej z
ej+ z + L+ e2j+ l ej+ z
V
V+

2
j
l
+
1 G+ L+ e
1 G L e2j l

V =

Z
Z + ZG

1
VG
= (1 G )(VG1 VG2 )
2

(10.1.31)

(10.1.32)

The parameters , Z are obtained using the rules of Eq. (10.1.15). From Eq. (10.1.12),
we nd the eigenvalues of the matrices F G:

They may be re-expressed in terms of A, B with the help of (10.1.18):

A(0)G B(0)= D1 Z(Z + ZG I)1 VG ,

ZG Z
,
ZG + Z

The voltages V1 (z), V2 (z) are obtained by extracting the top and bottom compo

nents of (10.1.29), that is, V1,2 (z)= V+ (z)V (z) / 2 :

where we dened:

The terminal conditions for the line are at z = 0 and z = l :

VG = V(0)+ZG I(0) ,

G =

(10.1.21)

To complete the solution, we assume that both lines are terminated at common
generator and load impedances, that is, ZG1 = ZG2 ZG and ZL1 = ZL2 ZL . The
generator voltages VG1 , VG2 are assumed to be different. We dene the generator voltage
vector and source and load matrix reection coefcients:

VG1
VG2

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

where are the eigenvalues of B, Z the eigenvalues of Z, and G , L are:

Using (10.1.18), we obtain the solutions for V, I :

352

Lm
1
(F + G) = ( + )= 1
= (L0 Lm )
L0
Z0

Cm
= Z0 (C0 Cm )
(F G) = ( )= 1
C0

Then, it follows that:



+ = (F + G)+ (F G)+ = (L0 + Lm )(C0 Cm )


= (F + G) (F G) = (L0 Lm )(C0 + Cm )


(F + G)+
L0 + Lm
Z+ = Z0
=
(F G)+
C0 Cm


(F + G)
L0 Lm
Z = Z0
=
(F G)
C0 + Cm

(10.1.34)

Thus, the coupled system acts as two uncoupled lines with

wavenumbers and characteristic impedances , Z , propagation speeds v = 1/ (L0 Lm )(C0 Cm ), and
propagation delays T = l/v . The even mode is energized when VG2 = VG1 , or,
VG+ = 0, VG = 0, and the odd mode, when VG2 = VG1 , or, VG+ = 0, VG = 0.
When the coupled lines are immersed in a homogeneous medium, such as two parallel
wires in air over a ground plane, then the propagation speeds must be equal to the speed

of light within this medium [496], that is, v+ = v = 1/ . This requires:

(L0 + Lm )(C0 Cm )= 
(L0 Lm )(C0 + Cm )= 

(10.1.29)

(10.1.33)

L0 =

C0
C20 C2m

Lm =

Cm
C20 C2m

(10.1.35)

Therefore, Lm /L0 = Cm /C0 , or, equivalently, = 0. On the other hand, in an

inhomogeneous medium, such as for the case of the microstrip lines shown in Fig. 10.1.1,
the propagation speeds may be different, v+ = v , and hence T+ = T .

www.ece.rutgers.edu/orfanidi/ewa

353

10.2 Crosstalk Between Lines

1
2
)
)
1 (1 G+ )(1 + L+ +
1 (1 G )(1 + L
V
V
V2 (0) =
2
2
2
1 G L 2
1 G+ L+ +

(10.2.1)

1
1
(1 G+ )(1 + L+ )
(1 G )(1 + L )
1 +
1
V
V
V2 (l) =
2
2
2
2
1 G+ L+ +
1 G L

where we dened V = VG1 /2 and introduced the z-transform delay variables =

ejT = ej l . Assuming purely resistive termination impedances ZG , ZL , we may use
Eq. (9.15.15) to obtain the corresponding time-domain responses:

1
1
(1 G )V(t)+ 1 +
G
2

m=1

(10.2.2)


1
(1 G+ )(1 + L+ )
(G+ L+ )m V(t 2mT+ T+ )
2
m=0


1
(1 G )(1 + L )
(G L )m V(t 2mT T )
2
m=0

where V(t)= VG1 (t)/2. Because Z = Z0 , there will be multiple reections even when
the lines are matched to Z0 at both ends. Setting ZG = ZL = Z0 , gives for the reection
coefcients (10.1.30):

In this case, we nd for the crosstalk signals:

V2 (0, t) =

(10.2.3)

V2 (l, t) =

(10.2.5)


1
2+m V(t 2mT+ T+ )
V1 (l, t) = (1 2+ )
2
m=0


1
(1 2 )
2m V(t 2mT T )
2
m=0

= 0 =

,
2

Z = Z0 Z = Z0 Z0
T = T T = T T

v = v0 v0

(10.2.6)

V2 (0, t) =


1
1
(+ )V(t) + V(t 2T+ ) V(t 2T )
2
2

V2 (l, t) =

1
V(t T+ )V(t T )]
2

Using a Taylor series expansion and (10.2.6), we have to rst-order:

=
V

dV
dt

T)
V(t T )= V(t T T) V(t T)(T)V(t


V(t 2T), where we
Therefore, V(t 2T )= V(t 2T)(T)V
. It follows that:
ignored the second-order terms (T)V


1
(1 + + )V(t)(1 + )
2+m1 V(t 2mT+ )
2
m=1

2T) ,
V(t 2T )= V(t 2T T) V(t 2T)(T)V(t


1
2m1 V(t 2mT )
(1 + )V(t)(1 )
2
m=1

where T = l/v0 . Because the s are already rst-order, the multiple reection terms
in the above summations are a second-order effect, and only the lowest terms will contribute, that is, the term m = 1 for the near-end, and m = 0 for the far end. Then,

Z0 Z
=
Z0 + Z


1
2m1 V(t 2mT )
+ (1 + )V(t)(1 )
2
m=1

= = ,

(G L ) V(t 2mT )
m

m=1


1
V1 (0, t) = (1 + + )V(t)(1 + )
2+m1 V(t 2mT+ )
2
m=1

These expressions simplify drastically if we assume weak coupling. It is straightforward to verify that to first-order in the parameters Lm /L0 , Cm /C0 , or equivalently, to
rst-order in , , we have the approximations:

(G+ L+ ) V(t 2mT+ )

m

V2 (l, t) =

G = L =

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

Similarly, the near-end and far-end signals on the driven line are found by adding,
instead of subtracting, the even- and odd-mode terms:

When only line-1 is energized, that is, VG1 = 0, VG2 = 0, the coupling between the lines
induces a propagating wave in line-2, referred to as crosstalk, which also has some minor
inuence back on line-1. The near-end and far-end crosstalk are the values of V2 (z) at
z = 0 and z = l, respectively. Setting VG2 = 0 in (10.1.32), we have from (10.1.31):

1
1
V2 (0, t) = (1 G+ )V(t)+ 1 +
2
G+

354

(10.2.4)


1
(1 2+ )
2+m V(t 2mT+ T+ )
2
m=0


1
2m V(t 2mT T )
(1 2 )
2
m=0
V(t) is the signal that would exist on a matched line-1 in the absence of line-2, V = Z V /(Z +Z )=
0 G1
0
G
VG1 /2, provided ZG = Z0 .

V2 (0, t) =





1
(+ ) V(t)V(t 2T) = () V(t)V(t 2T)
2

V2 (l, t) =


1
dV(t T)
V(t T)(T)V
= (T)
V(t T)(T)V
2
dt

These can be written in the commonly used form:



V2 (0, t)= Kb V(t)V(t 2T)
V2 (l, t)= Kf

dV(t T)
dt

(near- and far-end crosstalk)

(10.2.7)

www.ece.rutgers.edu/orfanidi/ewa

355

356

The interaction between the two lines is seen better in the MATLAB movie xtalkmovie.m,
which plots the waves V1 (z, t) and V2 (z, t) as they propagate to and get reected from
their respective loads, and compares them to the uncoupled case V0 (z, t)= V(t z/v0 ).
The waves V1,2 (z, t) are computed by the same method as for the movie pulsemovie.m


of Example 9.15.1, applied separately to the even and odd modes.

where Kb , Kf are known as the backward and forward crosstalk coefcients:

v0

=
2
4

Kb =

Lm
+ Cm Z0
Z0

Kf = T

v0 T

Lm
Cm Z0
Z0

(10.2.8)

where we may replace l = v0 T. The same approximations give for line-1, V1 (0, t)= V(t)
and V1 (l, t)= V(t T). Thus, to rst-order, line-2 does not act back to disturb line-1.
Example 10.2.1: Fig. 10.2.1 shows the signals V1 (0, t), V1 (l, t), V2 (0, t), V2 (l, t) for a pair of
coupled lines matched at both ends. The uncoupled line impedance was Z0 = 50 .
Lm/L0 = 0.8, Cm/C0 = 0.7

Lm/L0 = 0.4, Cm/C0 = 0.3

1.2

1.2

0.8

0.8
line 1 near end
line 1 far end
line 2 near end
line 2 far end

0.6
0.4

0.4

0.2

0.2

0.2
0

line 1 near end

line 1 far end
line 2 near end
line 2 far end

0.6

10.3 Weakly Coupled Lines with Arbitrary Terminations

The even-odd mode decomposition can be carried out only in the case of identical lines
both of which have the same load and generator impedances. The case of arbitrary
terminations has been solved in closed form only for homogeneous media [493,496]. It
has also been solved for arbitrary media under the weak coupling assumption [503].
Following [503], we solve the general equations (10.1.7)(10.1.9) for weakly coupled
lines assuming arbitrary terminating impedances ZLi , ZGi , with reection coefcients:

Li =

a1
a
2
,
c=
b1
b2

dc
= jMc ,
dz

Fig. 10.2.1 Near- and far-end crosstalk signals on lines 1 and 2.

For the left graph, we chose Lm /L0 = 0.4, Cm /C0 = 0.3, which results in the even and odd
mode parameters (using the exact formulas):

Z+ = 70.71 , Z = 33.97 , v+ = 1.01v0 , v = 1.13v0

+ = 0.17 , = 0.19 , T+ = 0.99T , T = 0.88T , Kb = 0.175 ,

i = 1, 2

(10.3.1)

Z+ = 122.47 , Z = 17.15 , v+ = 1.36v0 , v = 1.71v0

+ = 0.42 , = 0.49 , T+ = 0.73T , T = 0.58T , Kb = 0.375 ,

M=
0

Kf = 0.05

Kf = 0.05

The weak-coupling approximations are more closely satised for the left case. Eqs. (10.2.7)
predict for V2 (0, t) a trapezoidal pulse of duration 2T and height Kb , and for V2 (l, t), a
rectangular pulse of width tr and height Kf /tr = 0.2 starting at t = T:

Kf 
dV(t T)
u(t T)u(t T tr )]
=
tr
dt

These predictions are approximately correct as can be seen in the gure. The approximation predicts also that V1 (0, t)= V(t) and V1 (l, t)= V(t T), which are not quite true
the effect of line-2 on line-1 cannot be ignored completely.

M=
0

dc
c,
= jM
dz

jMz

Its solution is given by c(z)= e

expressed in closed form as follows:

The generator input to line-1 was a rising step with rise-time tr = T/4, that is,


1
t
VG1 (t)=
u(t)u(t tr ) + u(t tr )
2
tr

The weak coupling assumption consists of ignoring the coupling of a1 , b1 on a2 , b2 .

This amounts to approximating the above linear system by:

The right graph corresponds to Lm /L0 = 0.8, Cm /C0 = 0.7, with parameters:

V2 (l, t)= Kf

ZGi Zi
,
ZGi + Zi

Gi =

t/T

t/T

V(t)=

ZLi Zi
,
ZLi + Zi

Working with the forward and backward waves, we write Eq. (10.1.7) as the 44
matrix equation:

0.2
1

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

jMz

ej1 z

j1 z

ej2 z )
(e
=

z
j
1

(e
ej2 z )

jMl

0
0

0
0
0

(10.3.2)

c(0), where the transition matrix ejMz can be

ej2 z

j1 z ej2 z )
(e

ej1 z

j1 z

(e

j2 z

j2 z

1 2

1 + 2
=

may be written in terms of the z-domain delay variables

The transition matrix e
i = eji l = eiTi , i = 1, 2, where Ti are the one-way travel times along the lines, that is,
Ti = l/vi . Then, we nd:

11
0
0
0
a1 (0)
a1 (l)

a (l)
1
1
1
1
1 2 ) 0 a2 (0)
(

(1 2 ) 2
2

(10.3.3)
b1 (l)
0
0
1
0 b1 (0)
1

b2 (0)
b2 (l)
(1 2 )
(1 2 ) 2
0

www.ece.rutgers.edu/orfanidi/ewa

357

358

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

These must be appended by the appropriate terminating conditions. Assuming that

only line-1 is driven, we have:

V1 (0)+ZG1 I1 (0)= VG1 ,

V2 (0)+ZG2 I2 (0)= 0 ,

V2 (0) =

V1 (l)= ZL1 I1 (l)

V2 (l)= ZL2 I2 (l)

jKf (1 + L1 G2 2 ) 1 + Kb (1 2 )(L1 + G2 ) 1
V2l
V2 (l) =
(1 G1 L1 2 )(1 G2 L2 2 )

which can be written in terms of the a, b waves:

a1 (0)G1 b1 (0)= U1 ,
a2 (0)G2 b2 (0)= 0 ,

b1 (l)= L1 a1 (l)
,
b2 (l)= L2 a2 (l)

U1 =

Z1

(1 G1 )

VG1
2

(10.3.4)

Eqs. (10.3.3) and (10.3.4) provide a set of eight equations in eight unknowns. Once
these are solved, the near- and far-end voltages may be determined. For line-1, we nd:

V1 (0)=

V1 (l)=


a1 (0)+b1 (0) =

Z1 
2

1 + L1 12

1 G1 L1 12

1 (1 + L1 )
V
a1 (l)+b1 (l) = 1
2
1 G1 L1 12


Z1 

(10.3.5)

where V = (1 G1 )VG1 /2 = Z1 VG1 /(Z1 + ZG1 ). For line-2, we have:

(11 21 )(L1 11 + L2 21 )+(
1 11 21 )(1 + L1 L2 11 21 )

V20
V2 (0) =
(1 G1 L1 12 )(1 G2 L2 22 )

V2 (l) =

1 11 21 )(L1 11 + G2 21 )
(11 21 )(1 + L1 G2 11 21 )+(

V2l
(1 G1 L1 12 )(1 G2 L2 22 )

(10.3.6)
where V20 = (1 + G2 )V = (1 + G2 )(1 G1 )VG1 /2 and V2l = (1 + L2 )V, and we
,
by:
dened

Z2
=

Z1
Z2
=

Z1

Z2

=
Z1 1 2
1 2 2

Z2

1
=
Z1 1 + 2
1 + 2 2

Lm
Cm Z2
Z1
Lm
+ Cm Z2
Z1

(10.3.7)

In the case of identical lines with Z1 = Z2 = Z0 and 1 = 2 = = /v0 , we must

take the limit:
lim

2 1

jKf (L1 + L2 ) 2 + Kb (1 2 )(1 + L1 L2 2 )

V20
(1 G1 L1 2 )(1 G2 L2 2 )

ej1 l ej2 l
d j1 l
e
= jlej1 l
=
1 2
d1

(10.3.9)

The corresponding time-domain signals will involve the double multiple reections
arising from the denominators. However, if we assume the each line is matched in at
least one of its ends, so that G1 L1 = G2 L2 = 0, then the denominators can be
eliminated. Replacing j by the time-derivative d/dt and each factor 1 by a delay by
T, we obtain:
2T)
V2 (0, t)= Kf (L1 + L2 + L1 G2 )V(t




+ Kb (1 + G2 ) V(t)V(t 2T) + Kb L1 L2 V(t 2T)V(t 4T)
(10.3.10)


3T)
T)+L1 G2 V(t
V2 (l, t)= Kf (1 + L2 )V(t


+ Kb (L1 + G2 + L1 L2 ) V(t T)V(t 3T)
where V(t)= (1 G1 )VG1 (t)/2, and we used the property G2 L2 = 0 to simplify the
expressions. Eqs. (10.3.10) reduce to (10.2.7) when the lines are matched at both ends.

10.4 Coupled-Mode Theory

In its simplest form, coupled-mode or coupled-wave theory provides a paradigm for the
interaction between two waves and the exchange of energy from one to the other as
they propagate. Reviews and earlier literature may be found in Refs. [504525], see also
[353372] for the relationship to ber Bragg gratings and distributed feedback lasers.
There are several mechanical and electrical analogs of coupled-mode theory, such as
a pair of coupled pendula, or two masses at the ends of two springs with a third spring
connecting the two, or two LC circuits with a coupling capacitor between them. In these
examples, the exchange of energy is taking place over time instead of over space.
Coupled-wave theory is inherently directional. If two forward-moving waves are
strongly coupled, then their interactions with the corresponding backward waves may
be ignored. Similarly, if a forward- and a backward-moving wave are strongly coupled,
then their interactions with the corresponding oppositely moving waves may be ignored.
Fig. 10.4.1 depicts these two cases of co-directional and contra-directional coupling.

Then, we obtain:
(11 21 ) jKf ejl = j

Kb =

v0
4

Lm
+ Cm Z0
Z0

l
2

Lm
Cm Z0 ejl
Z0

(10.3.8)

where Kf , Kb were dened in (10.2.8). Setting 1 = 2 = = ejl = ejT , we obtain the

crosstalk signals:

Fig. 10.4.1 Directional Couplers.

Eqs. (10.1.7) form the basis of coupled-mode theory. In the co-directional case, if
we assume that there are only forward waves at z = 0, that is, a(0)= 0 and b(0)= 0,

www.ece.rutgers.edu/orfanidi/ewa

359

then it may shown that the effect of the backward waves on the forward ones becomes
a second-order effect in the coupling constants, and therefore, it may be ignored. To
see this, we solve the second of Eqs. (10.1.7) for b in terms of a, assuming zero initial
conditions, and substitute it in the rst:

z
b(z)= j

ejF(zz ) G a(z ) dz

da
= jF a +
dz

z

GejF(zz ) G a(z ) dz

d
dz

a1
a2


= j

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

Codirectional coupler, / = 0



a1
a2


(10.4.1)

Codirectional coupler, / = 0.5

0.8

0.8

0.6

0.6
P1 (z)
P2 (z)

The second term is second-order in G, or in the coupling constant . Ignoring this

term, we obtain the standard equations describing a co-directional coupler:

da
= jF a
dz

360

P1 (z)
P2 (z)

0.4

0.4

0.2

0.2

0
0

0.5

1.5

z /

0
0

0.5

1.5

z /

For the contra-directional case, a similar argument that assumes the initial conditions

a2 (0)= b1 (0)= 0 gives the following approximation that couples the a1 and b2 waves:





a1
1
a1
d
(10.4.2)
= j
b2
2
dz b2
The conserved powers are in the two cases:
2

P = |a1 | + |a2 | ,

P = |a1 | |b2 |

(10.4.3)

Fig. 10.4.2 Power exchange in co-directional couplers.

10.5 Fiber Bragg Gratings

As an example of contra-directional coupling, we consider the case of a fiber Bragg
grating (FBG), that is, a ber with a segment that has a periodically varying refractive
index, as shown in Fig. 10.5.1.

The solution of Eq. (10.4.1) is obtained with the help of the transition matrix ejFz :

ejFz

cos z j sin z
jz

=e

j sin z

where

1 + 2

2
Thus, the solution of (10.4.1) is:

a1 (z)
a2 (z)

sin z

cos z + j
sin z

1 2

cos z j sin z

= ejz

j sin z

= 2 + 2



j sin z
a1 (0)

a2 (0)

cos z j
sin z

(10.4.4)

(10.4.5)

(10.4.6)

Starting with initial conditions a1 (0)= 1 and a2 (0)= 0, the total initial power will
be P = |a1 (0)|2 + |a2 (0)|2 = 1. As the waves propagate along the z-direction, power is
exchanged between lines 1 and 2 according to:

2
P1 (z)= |a1 (z)|2 = cos2 z + 2 sin2 z

2
P2 (z)= |a2 (z)|2 = 2 sin z = 1 P1 (z)

(10.4.7)

Fig. 10.4.2 shows the two cases for which / = 0 and / = 0.5. In both cases,
maximum exchange of power occurs periodically at distances that are odd multiples of
z = /2 . Complete power exchange occurs only in the case = 0, or equivalently,
when 1 = 2 . In this case, we have = and P1 (z)= cos2 z, P2 (z)= sin2 z.

Fig. 10.5.1 Fiber Bragg grating.

The backward wave is generated by the reection of a forward-moving wave incident

on the interface from the left. The grating behaves very similarly to a periodic multilayer
structure, such as a dielectric mirror at normal incidence, exhibiting high-reectance
bands. A simple model for an FBG is as follows [353372]:

d
dz

a(z)
b(z)


= j

jKz

ejKz



a(z)
b(z)


(10.5.1)

where K = 2/ is the Bloch wavenumber, is the period, and a(z), b(z) represent the
forward and backward waves. The following transformation removes the phase factor
ejKz from the coupling constant:

A(z)
B(z)


=

ejKz/2
0

d
dz



jKz/2

e
A(z)
B(z)


= j

a(z)
b(z)


=

ejKz/2 a(z)
ejKz/2 b(z)



A(z)
B(z)


(10.5.2)


(10.5.3)

www.ece.rutgers.edu/orfanidi/ewa

361

362

= ejFl

A(l)
B(l)


,

with

F=

(10.5.4)

The transfer matrix ejFl is given by:



cos l + j
sin l
j sin l

U11 U12
ejFl =
(10.5.5)

U12 U11
sin l
cos l j
sin l
j

where = 2 ||2 . If || < ||, then becomes imaginary. In this
case, it is more
convenient to express the transfer matrix in terms of the quantity = ||2 2 :

jFl

cosh l + j sinh l

sinh l
j

sinh l

cosh l j sinh l

(10.5.6)

The transfer matrix has unit determinant, which implies that |U11 | |U12 | = 1.
Using this property, we may rearrange (10.5.4) into its scattering matrix form that relates
the outgoing elds to the incoming ones:

B(0)
A(l)


=



A(0)
B(l)


,

sin l

,
=

sin l
cos l + j

U12
,
U11

U12
,
U11

T=

U11

(10.5.7)

T=

1
cos l + j

sin l

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0
4

0
4

Fig. 10.5.2 Reectance of ber Bragg gratings.

at the center of the band, = 0, is given by ||2max = tanh2 |l|. The reectance at the
asymptotic band edges is given by:

||2 =

||2 sinh2 l
||2 sin2 l
= 2
2
2
2
cos l + sin l
cosh2 l + 2 sinh2 l

|l|2
,
1 + |l|2

at = ||

The

zeros of the reectance correspond to sin l = 0, or, = m/l, which gives
= ||2 + (m/l)2 , where m is a non-zero integer.
The Bragg wavelength B is the wavelength at the center of the reecting band, that
is, corresponding to = 0, or, = K/2, or B = 2/ = 4/K = 2.
By concatenating two identical FBGs separated by a spacer of length d = B /4 =
/2, we obtain a quarter-wave phase-shifted FBG, which has a narrow transmission
window centered at = 0. Fig. 10.5.3 depicts such a compound grating. Within the
spacer, the A, B waves propagate with wavenumber as though they are uncoupled.

(10.5.8)

If there is only an incident wave from the left, that is, A(0)= 0 and B(l)= 0, then
(10.5.7) implies that B(0)= A(0) and A(l)= TA(0).
A consequence of power conservation, |A(0)|2 |B(0)|2 = |A(l)|2 |B(l)|2 , is
the unitarity of the scattering matrix, which implies the property ||2 + |T|2 = 1. The
reectance ||2 may be expressed in the following two forms, the rst being appropriate
when || ||, and the second when || ||:

||2 = 1 |T|2 =

0.8

where , are the reflection coefcients from the left and right, respectively, and T is
the transmission coefcient. We have explicitly,

||2

||2

A(0)
B(0)

Fiber Bragg Grating, l = 6

Fiber Bragg Grating, l = 3

where = K/2 is referred to as a detuning parameter. The conserved power is given

by P(z)= |a(z)|2 |b(z)|2 . The elds at z = 0 are related to those at z = l by:

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

(10.5.9)

Fig. 10.5.2 shows ||2 as a function of . The high-reectance band corresponds to

the range || ||. The left graph has l = 3 and the right one l = 6.
As l increases, the reection band becomes sharper. The asymptotic width of the
band is || ||. For any nite value of l, the maximum reectance achieved

Fig. 10.5.3 Quarter-wave phase-shifted ber Bragg grating.

The compound transfer matrix is obtained by multiplying the transfer matrices of

the two FBGs and the spacer: V = UFBG Uspacer UFBG , or, explicitly:

V11

V12

V12

V11

U11

U12

U12

U11



ejd

ejd



U11

U12

U12

U11

(10.5.10)

where the Uij are given in Eq. (10.5.5). It follows that the matrix elements of V are:
2 jd
+ |U12 |2 ejd ,
e
V11 = U11



jd
V12 = U12 U11 ejd + U11
e

(10.5.11)

The reection coefcient of the compound grating will be:

comp =






jd
V12
U12 U11 ejd + U11
e
T ejd + Tejd
=
=
2 jd
V11
T ejd + ||2 Tejd
U11
e
+ |U12 |2 ejd

(10.5.12)

www.ece.rutgers.edu/orfanidi/ewa

363

364

Electromagnetic Waves & Antennas S. J. Orfanidis June 21, 2004

= /T and U11 = 1/T. Assuming a quarter-wavelength spacing

where we replaced U12
d = B /4 = /2, we have d = ( + /)d = d + /2. Replacing ejd = ejd+j/2 =
j ejd , we obtain:


T ejd Tejd
comp = jd
(10.5.13)
||2 Tejd
T e

At = 0, we have T = T = 1/ cosh ||l, and therefore, comp = 0. Fig. 10.5.4 depicts

the reectance, |comp |2 , and transmittance, 1 |comp |2 , for the case l = 2.
Compound Grating, l = 2

Compound Grating, l = 2
1

0.8

0.8

Transmittance

Reflectance

Fig. 10.6.1 Forward and backward intensities in stratied medium.

1

0.6

0.4

0.2

forward. Similarly, the backward intensity, going from z + dz to z, will be decreased by

I (k + s)(dz) and increased by I+ s(dz). Thus, the incremental changes are:

0.6

dI+ = (k + s)I+ dz + sI dz

0.4

dI = (k + s)I dz + sI+ dz
0.2

or, written in matrix form:

0
4

0
4

d
dz

Fig. 10.5.4 Quarter-wave phase-shifted ber Bragg grating.

Quarter-wave phase-shifted FBGs are similar to the Fabry-Perot resonators discussed

in Sec. 5.5. Improved designs having narrow and at transmission bands can be obtained
by cascading several quarter-wave FBGs with different lengths [353373]. Some applications of FBGs in DWDM systems were pointed out in Sec. 5.7.

I+ (z)
I (z)


=

k+s
s

s
k s



I+ (z)
I (z)


(10.6.1)

This is similar in structure to Eq. (10.5.3), except the matrix coefcients are real. The
solution at distance z = l is obtained in terms of the initial values I (0) by:

I+ (l)
I (l)


= eFl

I+ (0)
I (0)


,

with

F=

k+s
s

s
k s


(10.6.2)

The transfer matrix eFl is:

10.6 Diffuse Reflection and Transmission

Another example of contra-directional coupling is the two-ux model of Schuster and
Kubelka-Munk describing the absorption and multiple scattering of light propagating in
a turbid medium [526542].
The model has a large number of applications, such as radiative transfer in stellar
atmospheres, reectance spectroscopy, reection and transmission properties of powders, papers, paints, skin tissue, dental materials, and the sea.
The model assumes a simplied parallel-plane geometry, as shown in Fig. 10.6.1.
Let I (z) be the forward and backward radiation intensities per unit frequency interval
at location z within the material. The model is described by the two coefcients k, s
of absorption and scattering per unit length. For simplicity, we assume that k, s are
independent of z.
Within a layer dz, the forward intensity I+ will be diminished by an amount of I+ k dz
due to absorption and an amount of I+ s dz due to scattering, and it will be increased by
an amount of I s dz arising from the backward-moving intensity that is getting scattered

U = eFl

cosh l sinh l
=
s

sinh l

s

sinh l

= U11

U21
cosh l +
sinh l

U12
U22


(10.6.3)

where = k + s and = 2 s2 = k(k + 2s). The transfer matrix is unimodular,

that is, det U = U11 U22 U12 U21 = 1.
Of interest are the input reectance (the albedo) R = I (0)/I+ (0) of the length-l
structure and its transmittance T = I+ (l)/I+ (0) , both expressed in terms of the output,
or background, reectance Rg = I (l)/I+ (l). Using Eq. (10.6.2), we nd:

R=
T=
These

U21 + U11 Rg
s sinh l + ( cosh l sinh l)Rg
=
U22 U12 Rg
cosh l + ( sRg )sinh l
1

U22 U12 Rg

cosh l + ( sRg )sinh l

are related to the normalized Kubelka [532] variables a = /s, b = /s.

(10.6.4)

www.ece.rutgers.edu/orfanidi/ewa

365

The reectance and transmittance corresponding to a black, non-reecting, background are obtained by setting Rg = 0 in Eq. (10.6.4):

R0 =
T0 =

U21
s sinh l
=
U22
cosh l + sinh l
1

U22

(10.6.5)

cosh l + sinh l

The reectance of an innitely-thick medium is obtained in the limit l :

R =

s
s


=
+
k + s + k(k + 2s)

k
(R 1)2
=
s
2R

(10.6.6)

For the special case of an absorbing but non-scattering medium (k = 0, s = 0), we

have = = k and the transfer matrix (10.6.3) and Eq. (10.6.4) simplify into:

U = eFl =

ekl

ekl

R = e2kl Rg ,

T = ekl

(10.6.7)

These are in accordance with our expectations for exponential attenuation with distance. The intensities are related by I+ (l)= ekl I+ (0) and I (l)= ekl I (0). Thus, the
reectance corresponds to traversing a forward and a reverse path of length l, and the
transmittance only a forward path.
Perhaps, the most surprising prediction of this model (rst pointed out by Schuster)
is that, in the case of a non-absorbing but scattering medium (k = 0, s = 0), the transmittance is not attenuating exponentially, but rather, inversely with distance. Indeed,
setting = s and taking the limit 1 sinh l l as 0, we nd:

U = eFl =

1 sl

sl

sl
1 + sl

R=

sl + (1 sl)Rg
,
1 + sl slRg

T=

1
1 + sl slRg

(10.6.8)

In particular, for the case of a non-reecting background, we have:

R0 =

sl
1 + sl

T0 =

1
1 + sl

(10.6.9)

10.7 Problems
10.1 Consider the practical case in which two lines are coupled only over a middle portion of
length l, with their beginning and ending segments being uncoupled, as shown below:

Assuming weakly coupled lines, how should Eqs. (10.3.6) and (10.3.9) be modied in this
case? [Hint: Replace the segments to the left of the reference plane A and to the right of
evenin equivalents.]
plane B by their Th