Feldmanet al.

Vol. 10, No. 7/July 1993/J. Opt. Soc. Am. B

1191

Polarization instability in a twisted birefringent optical fiber

Sandra F. Feldman,* Doreen A. Weinberger,' and Herbert G. Winful
Department of Electrical Engineering and ComputerScience, Universityof Michigan,Ann Arbor, Michigan 48109
Received December 4, 1991; revised manuscript

received January 19, 1993

The fast-axis polarization instability arises in a weakly birefringent fiber as a result of competition between
the natural fiber birefringence and the nonlinear ellipse rotation. Direct observation of the fast-axis polarization instability is reported. A full theoretical development of the polarization instability in a twisted, birefringent optical fiber is presented. The theory includes the derivation of and full solutions for the evolution of light
in a twisted fiber as well as stability analysis and phase-plane representation of the solutions. The experiment
is described in detail; good agreement is obtained between theory and experiment. As a result of the instability, very small variations in either the input power or the input polarization to the fiber result in large
changes in the output polarization. A crossed polarizer at the fiber end converts the polarization variation into
intensity information. Thus the modulation depth of an input pulse has been increased from 15% to 100%.
Modulation gains of as much as 10 times are theoretically possible.

INTRODUCTION
The interaction between natural, linear fiber birefringence and birefringence induced through the nonlinear
index of refraction has aroused interest for both pulseshaping and switching applications. Pulse shapers may
be used to clip the wings or winglike pedestals often
present in ultrashort pulses.'13 Intensity- or polarizationdependent switches may be used in fiber-optic logic gates,4
shutters, 5 or modulators.6 Each of these effects is based
on the same basic phenomenon. As a result of the nonlinear refractive index, more intense light experiences a different net birefringence than less intense light. Light
waves of different intensities therefore evolve through different polarization states as they propagate in a fiber. If
one appropriately orients a polarizer at the fiber output,
the high- and low-power components may be effectively
separated.
In a weakly birefringent fiber, the interplay between
natural birefringence and nonlinear effects, including the
nonlinear refractive index and the nonlinear ellipse rotation, leads to a fast-axis polarization instability.6 9 Unavoidable twisting of the fiber in the laboratory induces
additional circular birefringence,' 0 which complicates the
analysis but does not change the basic manifestation of the
polarization instability."" 2 In the study reported here we
have used long pulses from a Q-switched ND:YAGlaser to
make a direct observation of asymmetry between the fast
and the slow axes of a birefringent fiber arising owing to
polarization instability.'3" 4 Good qualitative agreement is
obtained between experiment and theory. Additionally, a
sixfold increase in the depth of a periodic modulation on
the input beam is observed in the neighborhood of the
polarization instability. Previous researchers used circularly polarized input to demonstrate dramatic pulse shaping that is consistent with a polarization instability.' 5
However, the researchers did not directly probe the fiber
principal axes, nor did they observe modulational gain.
The polarization instability could be used to enhance the
operation of nonlinear switches and pulse shapers as well
as to construct a novel amplifier to increase the modula0740-3224/93/071191-11$06.00

tion depth of a signal. On the other hand, in a fiber device that relies on the preservation of linear polarization
to function correctly, such a polarization instability could
be detrimental. It is necessary to understand the polarization instability both to determine how the instability
may be used in nonlinear devices and to determine the
critical parameters for the instability so that it may be
avoided if necessary.

NONLINEAR POLARIZATION EVOLUTION IN

TWISTED FIBER
Derivation of Propagation Equations

The equations governing polarization evolution in the fiber are derived in the cw approximation, so that timedependent effects are ignored. This approximation is
valid as long as the fiber length is short and the light
pulses are long enough that dispersion may be neglected.
The equations including dispersion have been considered
both for strongly birefringent fibers, in which case it is
valid to neglect nonlinear ellipse rotation, 6 and for weakly
birefringent fibers (n < 10-6), in which case one obtains
a soliton polarization instability that is similar to the cw
polarization instability.' 7 The equations are not integrable when time-dependent effects are included and
must be investigated numerically. When the time dependence may be neglected, the equations may be solved
exactly.

Since the equations governing the evolution of the polarization state of a light wave in an untwisted fiber are contained as a special case of the equations for the twisted
fiber, and since small amounts of twist are nearly impossible to avoid in the laboratory, the theory is developed
here for a twisted fiber." When one twists the optical
fiber there is a twofold effect: the first is a strictly geometric effect that is due to the precession of the fiber principal axes'" and the second is shear-strain-induced circular
birefringence. 0 A fiber twist rate of q rad per unit length
results in circular birefringence a = hq, where h
0.130.16 for silica fiber. Wedefine a normalized birefringence
1993 Optical Society of America

1192

Feldman et al.

J. Opt. Soc. Am. B/Vol. 10, No. 7/July 1993

by -q= Aa/r. We take the z direction as the direction of

propagation. In a coordinate frame that twists along
with the fiber the dielectric permittivity tensor becomes

1.
2

n1.

(1)

polarized basis fields of Eq. (6). We consider the slowly

varying envelope approximation, which applies when the
field changes gradually as it propagates, and so the terms
that involve second derivatives with respect to distance
may be neglected. After much algebra, a pair of coupled
differential equations is obtained for the two circularly
polarized modes, c+ and c_:

62

The real diagonal terms account for the natural linear fiber birefringence, while the imaginary off-diagonal terms
represent the twist-induced circular birefringence. In an
isotropic fiber the diagonal terms el and 62 would be equal.
Since Maxwell's equations are simpler in an untwisted
frame, we transform the dielectric tensor into the laboratory coordinate system, using the rotation matrix

cos(qz)

-sin(qz)

sin(qz)

+ Eacos(2qz)

where eo = 1/2(el
and ea = 1/2(el birefringence.

+
62)

-iq

62)

ca sin(2qz)

+ iq

c - ea cos(2qz)

I, (3)

n 2 is the average permittivity

is proportional to the linear fiber

We follow the derivation of Winfull9 and write the

electric-field vector as a superposition of the two orthogonally linearly polarized modes, such that E = [LEj(z) +
SEy(z)]f(x,y)exp(-icot).
E and Ey are the complex

linearly polarized mode amplitudes, and f(x,y)

(7a)
d

-c

dz

= iK

is

y)dxdy = 1. The mode pattern does not change

significantly as the light propagates down the fiber. The
field components propagating in the fiber must satisfy
Maxwell's wave equation,

=
d2E,+() E=o2p.NL

(xeo/2)[(E E*)E + 1/2(E E)E*],

(5)

1.5.

We resolve the field into complex circularly polarized

components by explicitly writing the field amplitude and
phase as
=

1/V2[E(z)

2 + 21c+1
2 )c_,
i3(Ic_1

cfn2/ 3 cAeff and the fiber

iEy(z)].

=--

a.

d Ie+ = cc-Isintr,
d
=
d-Ie-I =. -iIc.IsinP,

(8a)

dz

(8b)

q+KCsI)(cII+
2q + K(cos/)

-1

- II
+
Ic-I(

(6)

The dependence on distance that is due to the linear propagation term is explicitly written as exp(ikoz), where ko =
wn/c. We now expand Eqs. (4) and (5), using the circularly

(Ic-I2

- Ic+12).
(8c)

(4)

where X = 8nn 2/3. In silica fiber the nonlinear index coefficient n2 3.2 X 10- cm2/W and the fiber core index

c(z)exp(ikoz)

dzIC+I

where i,j = x, y. The nonlinear polarization pNL leads

to the familiar phenomenon of the intensity-dependent
refractive index and nonlinear ellipse rotation. 7 2 0 The
first term in the expansion of the nonlinear polarization
in an optical fiber is the third-order term2 0

2n

gral f If(x, y)14dxdy. The normalized birefringence K =

ln/A = koan/2n.
The fiber beat length is defined as
Lb = 7/K and corresponds to the fiber length required in
the linear regime for the polarization state to undergo one
full evolution and to return to its initial polarization.
Finally, we further simplify the equations by explicitly
rewriting the modes in terms of amplitudes and phases so
that c(z) = Ic(z)Iexp[i45(z)]. It is also convenient to
define T =+ - 0 - 2qz.; Equating the real parts of
the resulting equations yields differential equations for
the amplitudes of c+ and c_. Equating the imaginary
parts of the equations yields differential equations for the
phases + and 0-, which in turn yields a single compact
equation for T:

f f2 (x,

exp(-i2qz)c+ - i-c_

(7b)

the transverse-mode pattern normalized so that

pNL

2C

effective-modal area Aeff is the inverse of the inte-

cos(qz)

e = R(-qz)etwj1 tR(qz)
ea sin(2qz)

i(C1

where ,B = XcW/4ncAeff=

In the laboratory coordinate system the dielectric tensor

transforms to

Eo

d c+ = i c+ + i exp(i2qz)c_ + iP(Ic+2 + 21cj 2)c+,

dz 2n

The polarization of the light is completely determined

once solutions for c+I,Ic-I, and T are obtained.
It is relevant to note that the theory presented above
corresponds to the limiting case of a vanishing modulation
frequency in the theory presented by Wabnitz for modulational polarization instability, which includes the effect of
chromatic dispersion.2'
Derivation of Conservation Equations

Equations for conservation of power and momentum are

derived from Eqs. (8). It will prove to be convenient to
normalize the powers in the circularly polarized components by /3/ 2 K so that u = IC+1 2 /3/2K and v = ICI|2 3/2K.
The power conservation equation is easily obtained from
Eqs. (8a) and (8b). The equation d/dz(lc+12 + IcI 2) = is
readily apparent since Ic+IdIcI/dz (1/2)dlc+12/dz, and a
similar equation holds for Ic-1. With the normalized powers u and v it is clear that

p = u + v,

(9)

Vol. 0, No. 7/July 1993/J. Opt. Soc. Am. B

Feldman et al.

where p is a constant that expresses the normalized, total

power in the system. The equation for the conservation
of momentum is derived from the identity
d (1c+lIc-lcosP) = cosPd

(lc+lIc-I) - Ic+IIc-IsinT dP
(10)

the derivative d(Ic I Ic-I1)/dz,using Eqs. (8a)

and (8b), and we recast the expression for Ic+IIcJdP/dz,
using Eq. (8c). After these substitutions and additional
algebra, a second equation for the conservation of momentum is obtained:

We rewrite

r = Vuvcost - u( + , - p),(
where jL= q/K(1 h/2n). The two conservation equations are used to solve the set of coupled differential equations describing the evolution of the light polarization in a
twisted fiber.
Solutions for Evolution of Light in Twisted Fiber
We use the normalized power variables u and v as well as
Azto recast Eqs. (8) as follows:

du
-=

2KN/UsinP,

(12a)

dv

sin,

(12b)

dz

-= -2K
dz

d = K , - TACOS T -

+ 2K(V - U).

2Kt

1193

Once u(z) is found, v(z) is immediately obtained from

the conservation of power. The relative phase difference
I is determined from the conservation of momentum.
However, is not unambiguously defined by the inverse
cosine operation implied in Eq. (11). To determine the
correct value for T, refer to Eqs. (12). If du/dz 2 0, then
0 and P is assigned in the range 0 ' ' -,. If
sin
du/dz < 0, then sinP < 0 and T is in the range - ir 5
T < 0. Once u, v, and T are known, the output light polarization is calculated. The azimuth of the polarization
ellipse in laboratory coordinates is 6 lab= T/2 + qz. However, it is more useful to consider the orientation of the
polarization ellipse relative to the principal fiber axes, in
which case 0 = /2P The angle 0 = 00 corresponds to the
slow axis, while 0 = 90 corresponds to light polarized
along the fast axis. The ellipticity is given by e = (r - 1)/

(r + 1),wherer = I.
Determination of Eigenmodes and Stability Analysis
of Solutions

The eigenmodes of this system are those polarization

states that propagate along the fiber without changing.
Once the eigenmodes are known, a stability analysis is
performed and the results are diplayed in bifurcation diagrams. To find the modes that propagate without changing, one rewrites Eqs. (8) and sets the derivatives with

(12c)

Equation (11),which expresses the conservation of momentum, may be used to reduce Eq. (12a) to the quadrature,
f|U0

= f+

2Kdz' =

2Kz,

wherethe quartic Q(u) = u(p - u) - [F + u(u

(13)

p+

and uo is the input power in the right-circularly polarized

component. One chooses the sign on the right-hand side
of Eq. (13) by evaluating sin P at the fiber input. If
sin T > 0, then the positive sign is chosen; otherwise the
negative sign is used.
Equation (13) may be inverted to solve for u in terms of
elliptic functions.22 The exact solutions depend on the

nature of the quartic Q(u)= -(u

(a)

0.5

- U1)(U - U 2 )(U - U3)

Q(u) may have four real roots or two real roots

and two complex-conjugate roots. In case 1, Q(u) has four
real roots, ul 12 2 U3 2 U4, as seen in Fig. 1(a), where
-Q(u) is plotted versus u. If one considers -Q(u) as a
potential, it becomes clear that the allowable values of u
are confined to either (a) ul U 2 U2 or (b) U3 U11 U4.
12, and
U
In case 2 [Fig. 1(b)], Q(u) has two real roots, ul
two complex roots, U3 = U4*. In this case u is constrained
by ul - U 2 U2 . Separate solutions exist for each of these
cases [Figs. 1(a), 1(b), and 2] and are summarized in
Table 1. The solutions of Table 1 were checked in selected cases by direct numerical integration of the coupled
differential equations.
(U -

Right-Circularly Polarized Power Component

IL)]

U4 ).

-0.51

-1

2.5
2
1
1.5
0.5
-0.5
0
Right - Circularly Polarized Power Component

(b)

Fig. 1. Roots and allowed values of the quartic Q(u). The quartics are calculated for circularly polarized light in the same fiber
(a) with no twist and (b) with a twist rate of one-half twist per
beat length. The normalized power P = 2.2 in both cases. The
roots are indicated by filled circles. The value of u is constrained
to lie within one of the regions bounded by arrows.

1194

J. Opt. Soc. Am. B/Vol. 10, No. 7/July 1993

Feldman et al.

Table 1. Solutions for the Nonlinear Polarization Evolution in a Twisted Optical Fiber
Four Real Roots
Parameters

U1

U > U2 > U3 > U4

U1

(Ul - U2)(U3 - U4 )/(U, - U3 )(U2 - U4)

2/\/(ul

'ko

sin'l

uWz)

U3(U1 - U2) -

- U3)(U2 - U4)
[(iU
U3)(Uo- U2) 1/2
L(U - U2)(UO- U3)

(1

Two Real Roots U1

Two
Complex Roots U3
2
A = ( - a)2 + b2;B 2

Four Real Roots

2U3 2 U > U4

(u - U2)(U3- U4 )/(Ul - U3)(U2 - U4)

[(U1 - U2 )2 - (A -

2/\/(ul - U3 )(U2 - U4)

1/VA7_

sin

sn2 (x Im)U 2 (u -U3).,

U2) - sn2 (x Im)(u, - U3)

U2

cos-1

[(U

- U3)(UO4)11/2
(3 - U4)(U1 - UO)

U1(U3- U4) + sn 2(xIm))U4 (Ul - U3)

(U3 - U4) + sn2 (x Im)(ul - U3)

(,

U >

U2

a + ib
(122- a)2 + b2

U4*

B)2]/4AB

- uo)B - (UO- 2)A1

uo)B + (WO- U2 )AJ

L(u,-

(u2A - uB)cn(x m) + (u2 A + uB)

(A - B)cn(x m) + (A + B)

ax = 2z/g + F0o1m); F is an elliptic integral of the first kind.

bsn

and cn are elliptic functions.

respect to distance (d/dz) equal to 0 as follows:

du= f,(u',v',P)=

KV'sinP n

dz'
dz=f2(U',V',>P)=

-/CU'sinP=>

dz = f 3(U',V',P)= K(-

(14a)

0,

(14b)

0,

)(cos ) - 2L

+ 2K(V'2 - U'2) => 0,

where u' =

V|7Ic+|

state of the input is close to that of an unstable eigenmode

(but does not exactly reproduce the SOP of the eigenmode),
then the light may evolve in a complicated way in the fiber,
and the output SOP need not remain close to the input
polarization state. One determines the stability of the
eigenmodes by forming
the Jacobian matrix J and examining its eigenvalues.2 3 The Jacobian matrix is defined as

(14c)

and v' =

\"17/Kc-j.

azimuth of the polarization ellipse in the fiber frame,

/2% is equal to

or 90. The fiber eigenmodes will

always be aligned with the principal axes. The ellipticity

of the eigenmodes may be determined from Eq. (14c).
Since T is known to be either 0 or 1800, cos T = +1 for
eigenmodes aligned along the slow and fast axes, respectively. We use the power conservation equation, p =
ui + v2, square both sides, and collect terms to obtain the
following quartic for u'2:
16(U,2)4 + 16 (,u - 2p)(u'2 ) 3 + 4(1 - 6 p + 5p2 +

p2 + 2p2A

Of,

aP

af2

af2

af2

(16)

2)(U2)2

p3)(U'2) + p2 = 0. (15)

To determine whether the solutions to Eq. (15) correspond

to the fast- or slow-axis eigenmodes, the solutions must
be substituted back into Eq. (14c). As already stated, if
the positive or negative sign holds, the solution corresponds to an eigenmode aligned with the fast or slow axis,
respectively.
Once the fiber eigenmodes are known, one completes
the analysis by determining which of the eigenmodes are
stable and which are unstable. If the polarization state of
the input light is close to that of a stable eigenmode, then
it will not evolve far from that state of polarization (SOP)
as it propagates. On the other hand, if the polarization

af3

af3

au'

av'

af3

oP

where fl, f2, and f are defined above in Eqs. (14). The
derivatives are evaluated, and one obtains
0

L- Kv'(cos'I) -

+ 4(-p -

of,
Ov'

au' av'

To satisfy Eqs. (14a) and (14b) simultaneously, we require

sin P = 0, corresponding to P = 0 or T = 1800. Thus the
=

of,
au'

K'

KU'5((cos ) + 4 KV'

Kv' cos

0] ~,

-KU'cosP1

(17)

where 6 = (1/1IU'
2 + 1/Iv'I2 ). One evaluates the matrix J,
using the values for u' v' and P for each polarization eigenmode. For each eigenmode, three eigenvalues, j
(j = 1, 2,3), of J are determined. If for all the eigenvalues Re(uji) < 0, then the eigenmode being evaluated is
stable. If for any one of the eigenvalues Re(,.j) > 0, then
the eigenmode is unstable.
The bifurcation diagrams in Fig. 2 show the fiber eigenmodes as a function of input power. Figure 2(a) shows the
fast- and slow-axis eigenmodes for the untwisted fiber.
The slow axis remains a stable guiding center for linearly
polarized light at any power. Below the critical power
(P < 1) the fast axis maintains stable guiding for linearly
polarized light. At the critical power (P = 1) the fast
axis undergoes a pitchfork bifurcation. Although linearly
polarized light remains an eigenmode, the eigenmode is
no longer stable. Two new, stable, elliptically polarized
eigenmodes emerge, which correspond to those polarizations at which the nonlinear ellipse rotation exactly balances the effect of the natural, linear fiber birefringence.
Figure 2(b) shows a similar diagram for a fiber with a
twist rate of only 0.3 twist per natural beat length, corresponding to a small induced circular birefringence.

Vol. 10, No. 7/July 1993/J. Opt. Soc. Am. B

Feldman et al.
SLOW- AXIS EIGENMODE
I 1.

BU

U
-1
0

1
0.5

2
1.5
1
FAST- AXIS EIGENMODES

2.5

.................................................

0
-l1

0.5

2.5

2
1.5
Normalized Power

(a)
SLOW- AXIS EIGENM[ODE

0
X1 -1

o0

0.5

v2.

2
I
1.5
FAS T- AXIS EIGENMIDDES

2.5

2
1.5
Normalized Poweer

2.5

m Ir
0
If
-0I

0.5

(b)

Fig. 2.

Bifurcation

diagrams

showing thE

estable(solidcurves)

and unstable (dotted curves) eigenmodes for both (a) an untwisted fiber and (b) a fiber with a twist rat.eof 0.3 twist per beat
length (after Ref. 14).

The slow axis again maintains stable guiding, albeit for

somewhat elliptically polarized light. Similarly, at low
power the fast-axis eigenmode is also *stable and elliptically polarized. The fast axis again1 bifurcates, and the
eigenmode that is most nearly linearly I polarized becomes
unstable. Although the nature of th e instability is unchanged, the onset of instability is ddelayed to a higher
power because of the need to use the nonlinearity to balance both the natural linear birefringe ace and the induced
circular birefringence.
The effect of twist on the polarizati [on evolution of the
light in the fiber depends strongly on tf ie ratio between the
twist-induced birefringence and the na tural fiber birefringence. In a slightly twisted but highly linearly birefringent fiber, light input along either axis should remain
oriented along that axis, twisting slowly in the laboratory
frame along with the fiber axes. In the other extreme,
a fiber with low linear birefringence (8n < 10-6) that is
strongly twisted (many twists per beat length) will develop
circular eigenmodes. To propagate v ithout a change in
the SOP, the light must be input with a circular polarization. This fiber is also stable against the polarization instability, since it requires extremely high power for the
nonlinear ellipse rotation to counter the strong induced
circular birefringence. In the interm ediate case the circular birefringence induced by the fibejr twist is of an order
of magnitude comparable with that cof the natural birefringence, resulting in an elliptical bi refringence. Light

1195

input with the correct elliptical polarization will propagate without changing, while all other states of polarization will evolve periodically as they propagate. The
period of this evolution is given by the modified beat length
2
Lb = VI/K(1 + p)1 . In general,however,it is alwayspossible to find a configuration in which a pair of orthogonal,
linearly polarized input states leads to linearly polarized
output states. These states are referred to as the nominal axes." In practice it is often convenient to use these
nominal axes as a laboratory reference.
Phase-Plane Presentation of Polarization Evolution

One uses phase portraits to visualize the solutions to sets

of coupled differential equations such as Eqs. (8). The
phase portrait plots ellipticity (e) versus azimuth () as a
function of propagation distance at a given light intensity.
In the phase plane a rectilinear coordinate system is used.
Ellipticity is indicated along the y axis, so that linear
polarization (e = 0) corresponds to a horizontal line across
the center of the plot. The upper half of the phase plane
represents right-elliptically polarized light with the rightcircular polarization along the top edge, and the lower half
represents left-elliptically polarized light with the leftcircular polarization along the bottom. The polarization
angle is plotted along the x axis. The angle 0 = 00 corresponds to the slow axis, while 0 = 900 corresponds to the
fast axis. Therefore the phase plane may be thought of as
having been wrapped around the surface of a cylinder.
Phase portraits are entirely equivalent in function to plots
on the Poincar6 sphere.24 In the Poincar6-sphere representation, the Stokes parameters are plotted on the surface of a sphere.

Linear polarization

is represented

along

the equator, while the circular polarizations occur at the

north and the south poles. The phase plane and the
Poincar6 sphere are used in the same way: an input SOP
is chosen and located on the plot. The trajectories on the
phase plane or sphere indicate how the light will evolve as
it propagates in the fiber.
Phase portraits are included in Fig. 3 for untwisted and
twisted fibers at powers less than and greater than the
critical power. The low-power phase portrait for an untwisted fiber [Fig. 3(a)] consists of two sets of concentric
ovals, centered on linearly polarized light oriented along
the fast and slow axes. The central point of these ovals
corresponds to a fiber eigenmode. The circular paths
surrounding the eigenmode indicate that the light polarization undergoes oscillatory behavior; the azimuth oscillates back and forth between the two angles 6 and passes
through both right- and left-elliptical polarizations in each
period. This oscillatory nature is clarified if one notes
the polarization evolution explicitly drawn beneath the
phase portrait. As the power is increased to three times
the critical power [Fig. 3(b)], the phase plane is divided
into two distinct regions by a separatrix orbit. The separatrix orbit passes through an unstable saddle point corresponding to linearly polarized light oriented along the fast
axis. Oscillatory behavior is again observed about the
slow axis for light input near that axis. However, near
the fast axis the evolution is drastically changed by the
presence of the saddle point. Linearly polarized light input near the fast axis but slightly off axis (inside the separatrix) will now oscillate about the slow axis. Light that
is precisely aligned with the axis but is slightly elliptically

1196

Feldmanet al.

J. Opt. Soc. Am. B/Vol. 10, No. 7/July 1993

RCP

121

LIN

14

-aI.LCP

FAST AXIS

80 90
FAST AXIS

SLOW AXIS
Azimuth

(C)

( )

C\()

10

11

Azimuth

12

13

14

15

16

I11

(a)
RCP

0
Azimuth
LIN

-90 -80 -60

FAST AXIS

-40

-&I LCP
80 90
FAST AXIS

-20
0
20
SLOW AXIS
Azimuth

(d)

(b)

Fig. 3. Phase portraits displaying the state of polarization of the

light as it propagates in both untwisted and twisted fibers at
powers less than and greater than the critical power (after
Ref. 14). (a) Low-power phase portrait for an untwisted fiber
(P = 0.5). Note the two regions of oscillatory motion centered
on linearly polarized light aligned with the principal axes.
(b) High-power phase plane for an untwisted fiber (P = 3.0).
Note the separatrix orbit (drawn as a heavy curve) separating the
regions of oscillatory and rotatory motion. (c) Low-powerphase
plane for a weakly twisted fiber (P = 0.01, twist rate = 0.3).
The phase plane now consists of two regions of oscillatory motion
as well as a region of rotatory motion. (d) High-power phase
plane for a weakly twisted fiber (P = 4.0, twist rate = 0.3). The
phase plane is once again divided by a separatrix orbit. Note
that the saddle point has moved from the linear polarization to a
slightly elliptical polarization.

polarized (outside the separatrix) executes rotatory motion: the light orientation traverses all the possible angles (-90 to +900) and always maintains the same
handedness. Again, the polarization evolution is explicitly displayed below the phase portrait to illustrate the dif-

ferent nature of oscillatory and rotatory motion. Note

also the two regions of oscillatory motion for elliptically
polarized light aligned along the fast axis; these correspond to the two additional stable eigenmodes seen in the
bifurcation diagram of Fig. 2(a).

e:)

10

11

':/L
7

12

Vol. 10, No. 7/July 1993/J. Opt. Soc. Am. B

et al.
Feldman

When the fiber is twisted, the eigenmodes become elliptical, as already observed from the bifurcation diagrams.
Thus at low power the phase plane consists of regions of
oscillatory motion centered on elliptical polarizations

exp(iO) _,N/P_11-=

orbit [Fig. 3(d)].

Comparison of Figs. 3(b) and 3(d) shows

that the location of the saddle point is shifted slightly to

an elliptical polarization because of the fiber twist. For
the instability in a twisted fiber to be reached with linearly polarized light, the azimuth of the light must be a
few degrees from the fast axis. The similar evolution of
the polarization state at high power for the twisted and

untwisted fiber indicates that even when the fiber is

slightly twisted the instability will manifest itself in a

form similar to that predicted for the untwisted case.
This occurs because the twist-induced birefringence becomes a fairly small perturbation on the system compared
with the magnitude of the nonlinear effect.
OBSERVATION OF POLARIZATION

INSTABILITY
Estimation of Fiber Beat Length

Because of the long beat length of the fiber used in this

experiment, standard measurement techniques, such as
25
observation of the Rayleigh scattering in the fiber, cannot be used. To estimate the beat length, the following
procedure was developed that is particularly suited for
fibers the beat lengths of which range from a few hundred
centimeters to a few meters. The fiber is supported in
vacuum chucks in a flat position to eliminate any effects
caused by external stress or bending. First, one locates
one of the principal axes of the fiber by rotating linearly
polarized input light (using a half-wave plate) and simultaneously adjusting an analyzing polarizer until the light
out of the fiber is also highly linearly polarized. Next, the
light input to the fiber is rotated in small steps. At each
step, one determines the orientation of the output light by
locating the major and minor axes of the elliptical polarization, and the ellipticity is calculated from the intensity
of the light along the ellipse axes. Since both the input
and the output polarizations are known, one may estimate
the fiber birefringence by treating the fiber as a wave plate
of unknown retardance in the Jones matrix formulation.
P, and P11 are the partial powers transmitted through
crossed and uncrossed polarizers corresponding to the
powers along the minor and major axes of the polarization
ellipse (Fig. 4). PTOT is the total power of the light beam
being analyzed. Rin is the rotation angle of the light into
the fiber relative to a fiber axis, and Routis the orientation
of the light leaving the fiber relative to the same axis.
Then in the fiber coordinate system the Jones matrices
representing the system2 6 are

sin Rout1
-Tsin Rout cos Rout

0 1 cos Rin -sin Rinl['1

exp(iF/2) [sin Rin cos Rin L 0]'

exp(-iF/2)
0

along each axis [Fig. 3(c)]. A comparison of Figs. 3(a) and

3(c) shows the profound effect of fiber twist on the evolution of a weak light beam. The phase plane for the
twisted fiber contains not only the two regions of oscillatory motion about the eigenmodes as before but also a
large region of rotatory motion. The circular birefringence induced by the twist has an important part in determining how the polarization evolves in the fiber. When
the light intensity is increased above the bifurcation point
shown in Fig. 2(b) the phase plane is again separated into
regions of oscillatory and rotatory motion by a separatrix

,,p-I rcos Rout

-i VP,

1197

(18)

where exp(iO) represents an arbitrary phase factor introduced in traveling through the fiber and F is the relative
phase delay between the parallel and perpendicular light
components in the fiber. The allows for right- or leftTo solve Eq. (18) it proves con-

elliptically polarized light.

venient to take the magnitude of each side and to work

with powers rather than fields, so the sign of the on the
left-hand side of the equation does not enter the final
equation for F. The above equation is straightforward to
solve for F, yielding
2

sin2 (Rin + Rout)

2
(Rin + Rout)
sin (Rin - Rout)- sin
P1/PTOT

From r, the beat length

Lb

fiber length X 2ir

F (rad)

(19)

may be calculated from

fiber length X 360
r (deg)

(20)

When one takes the inverse cosine to find F, the first several possible values should be retained, yielding several
values for the beat length. Physically, this is a direct
consequence of the ability to estimate the beat length only
to within modulo 27mrFor example, this method cannot
distinguish between fibers 0.25 beat length long and those
1.025 beat lengths long. Repeating the procedure with a
different length of fiber will establish which value for the
beat length is correct. We used this method repeatedly
to obtain a beat length estimate of 4 m for the fiber used
in this experiment. Some representative data that were
used to calculate this value for the beat length are presented in Table 2. Our calculation of the beat length is in
reasonable agreement with the estimation of a 10-m beat
length by Corning Glass Works, which supplied the fiber.
The critical power, P, = 2K//3, yields P, = 250 W for a 4-m
beat length.
Experiment Description

In this experiment 200-ns (FWHM) Q-switched pulses

from a Nd:YAG laser (A = 1.064 ,um) were used. These

Fig. 4. Determination of the beat length of a weakly birefringent fiber by means of a generalized Jones matrix technique.
The measurements required in Eq. (19) are indicated.

1198

Feldmanet al.

J. Opt. Soc. Am. B/Vol. 10, No. 7/July 1993

Table 2. Representative Data Used to Calculate

Fiber Beat Length'
Rin

(deg)
10
20
40
50

Rout(deg)

P.,

E/2 (deg)

Beat Length (in)

8
12
48
64

0.075
0.24
0.51
0.41

36
36
28
24

4
4
3.8
3.7

a In all cases the fiber length L = 3.2 m and the total power into the fiber
P = 2.2 (relative units). Approximately 20 mW of power was transmitted
through the fiber during the measurement.

Pi1

Fiber lies in aluminum

support
IZ~

ND

X/2 lox
MO

Folding
l. ... ... P.Mirror
lox /2 Polarizer
MO

wave plate. Light was coupled into and out of the fiber
with 1OXmicroscope objectives. Typically 50-60% of the
incident light was successfully coupled into the fiber.
The analyzer consisted of a half-wave plate followed by a
polarizing beam-splitter cube. We set the analyzer by
aligning the linearly polarized light with the fiber axes at
low power. The polarizer was then crossed to the fiber
output. It proved convenient to null the output, which we
did by rotating an output half-wave plate while keeping
the orientation of the polarizer fixed. This way, the polarizing cube permitted simultaneous monitoring of both
P-L,the power transmitted when the polarizer is in the
crossed position, and PI1, the power rejected by the crossed
polarizer. EG&G FND-100 photodiodes in a simple circuit with no preamplifier were used to detect the pulses.
The 200-ns Q-switched pulses were long in comparison to
the photodiode rise time (< 1 ns), so that we could directly

detect the output without power averaging. Neutralin

Nd:YAG 1.06pm

0
.
/odn
Attenuator Collimating 90-10
Mirror
AteutrLenses
Bean splitter

Fig. 5. Schematic of the experimental layout. The attenuator is

composed of a half-wave plate followed by a polarizer.

The wave

plate and the polarizer marked with an asterisk form a second attenuator, permitting more accurate control over the pulse power.
The fiber is held in vacuum chucks and supported in an aluminum bar to minimize stress-induced birefringence. Neutraldensity (ND) filters are used as needed to prevent damage to the
photodiodes. lox microscope objectives (MO's) are used to
couple light into and out of the fiber.

pulses were long enough and the fiber was short enough
that the effects of dispersion were negligible and the
cw approximations made throughout the preceding calculations were valid. The pulse repetition rate was kept to
500 Hz to avoid fiber damage. Since damage to the fiber
end face is affected both by the pulse peak power and the
average power, we used a low repetition rate to obtain
higher peak power. Even at only 500 Hz, catastrophic
fiber damage becomes a problem at peak powers exceeding
1 kW. The pulses were modulated at 76 MHz because of
longitudinal mode beating in the laser. We included a
sinusoidal approximation to the modulation in the model,
rather than using injection seeding or an intracavity etalon to attempt to eliminate the modulation. A schematic
of the experiment is shown in Fig. 5. We controlled the
laser power by using a pair of attenuators consisting of
half-wave plates followed by polarizers to adjust the intensity precisely while still maintaining the polarization of
the light. Two attenuators were needed because the laser
itself was polarized only slightly better than 10:1. A
single attenuator did not permit precise control over the
light intensity, particularly at low powers. Collimating
lenses were necessary to prevent excessive diffraction of
the beam before we launched the beam into the fiber. A
beam splitter placed in front of the final set of polarizing
optics split 10% of the light into a detector, which allowed us to monitor the input pulse shape during the
experiment. The final folding mirror was followed by a
polarizer so that the light entering the fiber was linearly
polarized at better than 500:1. We could orient the polarization at any angle to the principal axes by using a half-

density filters were used as needed to prevent photodiode

damage.
The silica fiber used was equivalent to a step-index fiber with a core radius of 2.6 Axmand A = 4.8 X 10' and
was 3.2 m in length. The fiber was held at either end by
means of low-vacuum chucks and was allowed to rest in a
long, V-shaped piece of aluminum to minimize birefringence effects caused by clamping, bending, and twisting.
Although a more highly birefringent fiber would be more
resistant to external effects and therefore easier to handle,
the critical power would also increase. The information
used for calculating the fiber beat length was obtained at
a cw power level of -100 mW to minimize nonlinear effects. At the same time, a low-power polarization extinction ratio of better than 500:1 was measured. The beat
length (Lb) was estimated to be -4 m, so that the fiber
length L 0.8 Lb and the corresponding critical power
P, 250 W. At peak-pulse powers approaching 1 kW,
stimulated Raman and Brillouin scattering as well as
significant pump depletion resulting from the Brillouin
scattering became problems. Therefore the peak power
was always kept below this experimentally determined
stimulated-scattering threshold.
Results and Comparison with Theory

The results of the coupled-mode theory for a weakly

twisted fiber were used to predict the transmission

through the exit analyzer. Exact quantitative agreement
between theory and experiment could not be expected because of the difficulty of either mounting a fiber with absolutely no twist or precisely measuring the twist rate.
However, we found that we could obtain good agreement
between theory and experiment by adjusting the various
parameters in the theory.
To compare the experiment with the theory, we approxi-

mated the input pulses as Gaussian pulses with an

exponentially decaying tail and with 15% (peak-to-peak)
sinusoidal modulation riding on the pulse that was due to
longitudinal mode beating in the laser [Fig. 6(a)]. We
used the theory for a fiber with a twist rate of 0.1 twist
per beat length to calculate the expected output polarization state for each point of the (model) input to predict the
output pulse shapes. Simulations showing the predicted
output through the crossed polarizer (Pj), together with
the corresponding experimental oscilloscope traces, are
shown in Figs. 6(b)-6(f). Note that the peak power given

Vol. 10, No. 7/July 1993/J. Opt. Soc. Am. B

Feldman et al.
2.5

1199

PEAKPOWER= 2.5

1.5

0.5
-2C0

30(0

TIME(ns)

-200

(e)
PEAKPOWER= 2.5

PEAKPOWER= 0.45

PEAKPOWER= 1.0

0.01

TIME(ns)

(c)

(a)
0.02

300

TIME(ns)

~~~~~~~~~~0.01

0.2

~~~~~~~~-U

0.01~

-200

300

-1 1

TIME(ns)

-200

TIME(ns)

(b)

300

TIME(ns)

300

(f)

(d)

Fig. 6. Comparison of the oscilloscope traces from the experiment with the theoretical calculations for a weakly twisted fiber. The
upper half of each part of the figure shows the theoretically predicted pulse shape transmitted through the crossed polarizer. The upper
trace in the oscilloscope photographs corresponds to P11, while the lower trace corresponds to P 1. The lower oscilloscope trace should be
compared with the theoretical prediction. Note that the vertical scale on the oscilloscope plots is in arbitrary units, and therefore the
transmittance near the slow and fast axes may not be directly compared. In addition, after data were taken for (e), the neutral-density
filters in front of the photodiode

were changed from 1.9 to 3.1 for (f).

(a) Theoretical

approximation

and oscilloscope trace of the

Q-switched pulse input into the fiber. The modulation is due to longitudinal mode beating in the laser. (b) Light at the critical power
input at =3 to the slow axis. Little light leaks through the crossed polarizer. (c) Intense light (P = 2.5) input to the fiber at =30 to the
slow axis. Again, little pulse shaping is observed. (d) Low power light (P = 0.45) is input to the fiber at -10' to the fast axis. Nonlinear pulse-shaping effects begin to appear at a light intensity of only one half of the critical power. (e) Light at the critical power is input
to the fiber at =10 from the fast axis. Significant pulse shaping is observed at the critical power. The fiber-crossed polarizer system
is far more transmitting for the central region of the pulse than in the wings. (f) Intense light (P = 2.5) input at -10 from the fast
axis. Significant pulse shaping is observed. The pulse appears narrower since the power in the wings is so strongly suppressed relative
to the peak.

for each plot as well as the values along the power axis is

in units normalized to P,. On the oscilloscope traces, the

lower trace corresponds to P 1 and the upper trace corresponds to PI1. The theoretical predictions should be compared with the lower oscilloscope traces. Typically only a
10%)of the light is passed
small percentage (maximum
through the crossed polarizer; hence pulse-shaping effects
are most easily observed in P,.
For linearly polarized light input near the slow axis, a
fairly constant, low-transmission rate is expected through
the analyzing polarizer; the output pulse undergoes little
shaping and should be similar to the input pulse [Figs. 6(b)
and 6(c)]. Linearly polarized light input near the fast
axis undergoes pulse shaping that is highly power dependent. At power well below the critical power, the fast axis

is a stable eigenmode, the behavior is similar to that near

the slow axis, and a small fraction of the pulse intensity is
transmitted essentially unshaped. At a slightly higher
power, the fiber-and-analyzer combination results in a
flat-topped output pulse [Fig. 6(d)] as a result of powerdependent pulse shaping. As the peak power is increased
to the critical power, switching occurs, and the central region of the pulse is transmitted more strongly than the
wings [Fig. 6(e)]. Finally, as the power continues to
increase above the critical power, this nonlinear transmission becomes more pronounced, and the modulation depth
of the pulse also increases [Fig. 6(f)]. The pulse appears
narrower, since the wings are strongly suppressed relative
1
to the central peak. In addition, the peak of P 1tends to
become slightly flatter as more power is transferred to P,.

1200

Feldmanet al.

J. Opt. Soc. Am. B/Vol. 10, No. 7/July 1993

Amplitude Modulation Gain

An important feature of nonlinear dynamical systems is

their ability to amplify fluctuations when they are driven
near an instability. For an input power in the vicinity of
the critical power for the polarization instability, a small
relative modulation of the input intensity can result in a
much larger modulation in the power transmitted in the
orthogonal polarization.+ If we define the modulation
depth as AP/P, then the amplitude-modulation (AM) gain
is defined as

.ar_
0

IP_
1AP

gIAPi.J/Pin
I

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

Normalized Input Power

Fig. 7. AM gain as a function of normalized input power. No

AM gain is expected below the critical power. AM gain as high
as 10 may be realized.

1 DI

(21)

where gD = APa/APin reduces to the differential gain and

the transmissivity T = Pl/P1n. Figure 7 plots gAMversus
the normalized input power and shows that AM gains as
high as 10 can be expected from this system. The experimental results shown in an expanded view in Fig. 8 confirm the predicted AM gains. A 15% modulation in the
input intensity is transformed into a nearly 100%modulation in the output P,. The modulation in P11 remains the
same as that in the input. Essentially, the input to this
amplifier is a large amplitude pulse with a small modulation, while the output is a smaller amplitude pulse carrying a large modulation.

CONCLUSION

Fig. 8. Oscilloscope trace showing a nearly fully modulated output pulse for an AM gain of -6. This expanded view was taken
at the same time as the pulses in Figs. 6(d)-6(f) (after Ref. 13).

The inclusion of a small amount of twist in the theory

greatly improves the fit of the experiment to the theoretical curves, as shown by a comparison of Fig. 1 of Ref. 13
and Fig. 6 herein. Not only is the qualitative agreement
better, as evidenced by the better match of the curve
shapes, but the quantitative agreement is also improved.
In Fig. 6(e) the ratio of the peak power to the power on the
shoulder is approximately three for both theory and ex-

periment. The ratio between the peak powers for

Figs. 6(e) and 6(f) is 50 in the experiment and
-30 in the theory if twist is included but only 8 if the
theory neglects fiber twist.
In the alignment of the analyzing optics, the assumption was made that if linearly polarized light into the fiber
yields linearly polarized light out of the fiber at low power,
then the light is traveling along a principal axis of the
fiber. However, in a twisted fiber this condition is met
along the fiber nominal axes. Only for negligible or small
twist do the nominal and principal axes coincide. The
experimental assumption that the nominal axes and the
principal axes are coincident is justified in light of the improved agreement between the theory and the experiment
when a small twist is incorporated into the theory.

When the intensity of light incident upon an untwisted,

birefringent optical fiber is low enough that nonlinear effects are negligible, then linearly polarized light that is
oriented along either of the fiber principal axes maintains
its state of polarization as it propagates. With increasing
light intensity a nonlinear birefringence is induced by
means of the intensity-dependent refractive index. When
the induced birefringence becomes comparable with the
natural fiber birefringence, asymmetry results between
the fast- and slow-fiber modes. Along the slow axis the
induced birefringence adds to the natural birefringence;
the slow axis therefore maintains stable guiding. Along
the fast axis the induced birefringence opposes the natural birefringence. The two birefringences exactly cancel
each other at the critical power. For linearly polarized
light above the critical power the fast axis becomes an unstable saddle point, and even a slight deviation from perfect linear polarization or from perfect alignment along
the fast axis leads to completely different output polarization states. Additionally, even at a given input polarization small changes in the input intensity may lead to
different output polarization states. Thus a signature of
the polarization instability is that small changes in input
intensity may lead to large changes in the intensity transmitted through a crossed polarizer located at the fiber
exit. If a weakly modulated beam is input into the fiber
with sufficient intensity to excite the polarization instability, the output beam transmitted through the crossed
polarizer may have a significantly greater modulation
depth. This increase is referred to as AM gain.
The perturbation caused when the fiber is slightly

twisted does not significantly alter the dynamics described above. When the fiber is twisted, circular birefringence is induced that, in turn, leads to elliptically polarized eigenmodes. At any input power light must be
launched with an appropriate elliptical polarization to

maintain the state of polarization as the light propagates.

These elliptically polarized eigenmodes are aligned along
the principal axes of the fiber. In a twisted fiber it is no
longer appropriate to speak of the presence of fast or slow
axes because linearly polarized light launched along these
axes does not maintain its state of polarization. However, the principal axes for the linear, natural fiber birefringence remain a convenient reference for measuring
the orientation of the light polarization as it evolves in the
fiber. In this case, the birefringence induced by intense
elliptically polarized light may cancel the combined natural and twist-induced birefringence.
When long lengths of low-birefringence fiber are used
in nonlinear switching applications, the effect of twist can
be quite important. A number of recent papers have debirescribed polarization-switching experiments involving
27 28
In
fringent fibers that are multiple beat lengths long.
the absence of twist, theory predicts that the switching
power P, will depend on the polarization angle of the input
9
In particular,
field with respect to the birefringence
near the slow axis (0 0), P, scales with the fiber length
as 1L, while at 45 (or for circularly polarized input) it
scales as 1\/E. Near the fast axis, the switching power is
independent of length and depends only on fiber birefringence and nonlinearity. Unavoidable twists in a long fiber
will lead to some averaging over these length dependences
of the switching power. This would explain some discrepancies noted in Ref. 27 between experiments and the
twist-free theory. 9
In conclusion, we have presented direct observations of
AM gain and of asymmetry between the fast and slow
modes of a birefringent fiber resulting from polarization
instabilities. Good qualitative agreement is obtained between the experiment and the theory. We observed a sixfold increase in the modulation depth of the input beam by
probing the instability. Such AM gain has potential use
in a novel amplifier with moderate power and fiber length
requirements.

1201

Vol. 10, No. 7/July 1993/J. Opt. Soc. Am. B

Feldmanet al.

3. B. Nikolaus,

D. Grischkowsky,

and A. C. Balant,

"Optical

pulse reshaping based on the nonlinear birefringence of

single-mode optical fibers," Opt. Lett. 8, 189-191 (1983).
4. K.-I. Kitayama, Y. Kimura, and S. Seikai, "Fiber-optic logic
gate," Appl. Phys. Lett. 46, 317-319 (1985).
5. J. M. Dziedzic, R. H. Stolen, and A. Ashkin,

"Optical Kerr

effect in long fibers," Appl. Opt. 20, 1403-1406 (1981).

6. B. Daino, G. Gregori, and S. Wabnitz, "New all-optical devices

based on third-order noninearity of birefringent fibers," Opt.

Lett. 11, 42-44 (1986).
7. H. G. Winful, "Polarization instabilities in birefringent nonlinear media: application to fiber-optic devices," Opt. Lett.
11, 33-35 (1986).
8. Related theoretical

work was also done by K. L. Sala, "Non-

linear refractive-index phenomena in isotropic media subjected to a dc electric field: exact solutions," Phys. Rev. A
29, 1944-1956 (1984).
9. Recently researchers have theoretically shown that an analogous polarization instability may be expected for spatial
solitons in a waveguide: C. M. de Sterke and J. E. Sipe,
"Polarization instability in a waveguide geometry," Opt.
Lett. 16, 202-204 (1991).
10. R. Ulrich and A. Simon, "Polarization optics of twisted
single-mode fibers," Appl. Opt. 18, 2241-2251

(1979).

11. H. G. Winful and A. Hu, "Intensity discrimination with

twisted birefringent optical fibers," Opt. Lett. 11, 668-670
(1986).

12. F. Matera and S. Wabnitz, "Nonlinear polarization evolution

and instability in a twisted birefringent fiber," Opt. Lett. 11,
467-469 (1986).
13. S. F. Feldman, D. A. Weinberger, and H. G. Winful, "Obser-

vation of polarization instabilities and modulational gain

in a low-birefringence optical fiber," Opt. Lett. 15, 311-313

(1990).
14. S. F. Feldman, D. A. Weinberger, and H. G. Winful, "Polariza-

tion instability in a twisted optical fiber," in Nonlinear Dynamics in Optical Systems, N. B. Abraham, E. M. Garmire,

and P. Mandel, eds., Vol. 7 of OSA Proceedings Series (Optical

Society of America, Washington, D.C., 1990), pp. 471-474.
15. S. Trillo, S. Wabnitz, R. H. Stolen, G. Assanto, C. T. Seaton,

and G. I. Stegeman, "Experimental observation of polarization instability in a birefringent optical fiber, Appl. Phys.
Lett. 49, 1224-1226 (1986).
16. C. R. Menyuk, "Stability of solitons in birefringent optical
fibers. II. Arbitrary amplitudes," J. Opt. Soc. Am. B 5,
392-402 (1988).
17. K. J. Blow, N. J. Doran, and D. Wood, "Polarization

instabili-

ties for solitons in birefringent fibers," Opt. Lett. 12, 202204 (1987).

ACKNOWLEDGMENTS
The authors thank Corning Glass Works for the donation
of the low-birefringence fiber necessary to conduct the
experiment. This work is based on dissertation research
supported under a National Science Foundation Graduate
Fellowship awarded to S. R Feldman. The authors acknowledge National Science Foundation support under
grants EET-8552520 (to D. A. Weinberger) and EET8712877 (to H. G. Winful).

*Present address, General Electric Corporate Research

and Development, Schenectady, New York 12309.

'Present address, Department of Physics, Smith College,

Northampton, Massachusetts 01063.

18. P. McIntyre and A. W Snyder, "Light propagation in twisted

anisotropic media: application to photoreceptors," J. Opt.
Soc. Am. 68, 149-157 (1978).
19. H. G. Winful, "Self-induced polarization changes in birefringent fibers," Appl. Phys. Lett. 47, 213-215 (1985).
20. P. D. Maker, R. W Terhune,

and C. M. Savage, "Intensity-

dependent changes in the refractive index of liquids," Phys.

Rev. Lett. 12, 507-509 (1964).

21. S. Wabnitz, "Modulational polarization instability of light in

a nonlinear birefringent dispersive medium," Phys. Rev. A
38, 2018-2021 (1988).
22. P. F. Byrd and M. D. Friedman, Handbook of Elliptic In-

tegrals for Engineers and Scientists (Springer-Verlag,

New York, 1971).

23. R. Seydel, From Equilibrium to Chaos: Practical Bifurcation and Stability Analysis (Elsevier, New York, 1988).
24. M. Born and E. Wolf, Principles of Optics (Pergamon,
New York, 1983).

25. A. Papp and H. Harms, "Polarization optics of index-gradient

optical waveguide fibers," Appl. Opt. 14, 2406-2411

REFERENCES AND NOTES

1. R. H. Stolen, J. Botineau, and A. Ashkin, "Intensity discrimination of optical pulses with birefringent fibers," Opt. Lett. 7,
512-514 (1982).

2. N. J. Halas and D.Grischkowsky, "Simultaneous optical pump

compression and wing reduction," Appl. Phys. Lett. 48, 823825 (1986).

(1975).

26. A Yariv and P. Yeh, Optical Waves in Crystals (Wiley,

New York, 1984).

27. N. Finlayson, B. K. Nayar, and N. J. Doran, 'An ultrafast

multibeatlength all-optical fibre switch," Electron. Lett. 27,
1209-1210 (1991).
28. P. Ferro, M. Haelterman,

S. Trillo, S. Wabnitz, and B. Daino,

'All-optical polarisation switch with long low-birefringence

fibre," Electron. Lett. 27, 1407-1408 (1991).