Foundations for low-loss fiber gradient-index lens

pair coupling with the self-imaging mechanism

Martin van Buren and Nabeel A. Riza

A fiber-optic collimator that emits a Gaussian beam with its beam waist at a certain distance after the
exit face of the lens is labeled a self-imaging collimator. For such a collimator, the waist of the emitted
Gaussian beam and its location are partly dependent on the properties of the gradient-index 共GRIN兲 lens.
Parameters for the self-imaging collimator are formulated in terms of the parameters of a GRIN lens 共e.g.,
pitch, core refractive index, gradient index, length兲 and the optical wavelength. Next, by use of the
Gaussian beam approximation, a general expression for the coupling power loss between two self-
imaging-type single-mode fiber 共SMF兲 collimators is, for the first time to our knowledge, derived as a
function of three types of misalignment, namely, separation, lateral offset, and angular tilt misalignment.
A coupling experiment between two self-imaging collimators with changing separation distance is suc-
cessfully performed and matches the proposed self-imaging mechanism coupling loss theory. In addi-
tion, using a prism, lateral offset, as well as angular tilt, misalignments are experimentally simulated for
a two self-imaging collimator coupling condition by a single collimator reflective test geometry. Exper-
imental results agree well with the proposed loss formulas for self-imaging GRIN lenses. Hence, for the
first time to our knowledge, the mathematical foundations are laid for employing self-imaging-type fiber
collimators in SMF-based free-space systems allowing optimal design for ultra-low-loss coupling. © 2003
Optical Society of America
OCIS codes: 110.2760, 060.2310.

1. Introduction coupling power loss is dependent on the distance be-

Coupling loss is a key issue for the design of both
integrated-optic and fiber-optic components and sub-
systems involving free-space light interaction. Ear-
lier research that used mode mismatch analysis was
conducted for insertion loss determination for fiber-
to-integrated-optic waveguide chip coupling by indi-
vidual lenses such as a gradient-index 共GRIN兲 lens.1
Free-space-based fiber-optic components such as iso-
lators, circulators, attenuators, switches, and wave-
length division multiplexers and demultiplexers2–5
frequently make use of single-mode fiber 共SMF兲 col-
limators to couple light between fibers. The main
advantage of these collimators is the moderate cou-
pling power loss for large separation distance. The
The authors are with the Photonic Information Processing Sys-
tems Laboratory, School of Optics, Center for Research and Edu-
cation in Optics and Lasers, University of Central Florida, 4000
Central Florida Boulevard, Orlando, Florida 32816-2700. The
e-mail address for N. A. Riza is riza@creol.ucf.edu.
Received 2 April 2002; revised manuscript received 3 September
2002.
0003-6935兾03兾030550-16$15.00兾0
© 2003 Optical Society of America
tween the input GRIN lens and the output GRIN lens
and the misalignment of these lenses. There are
three types of misalignment: separation, lateral off-
set, and angular tilt misalignment. Recently, a gen-
eral formula to determine the coupling loss of two
SMF collimators with these three misalignments was
given.6,7
The fundamental mode that is the only mode al-
lowed to travel through the SMF propagates by ap-
proximation as a Gaussian beam in free space. A
quarter-pitch GRIN lens that is directly connected to
a SMF emits by approximation a Gaussian beam
with its beam waist on the exit face of the GRIN lens.8
These lenses were used in Ref. 6. However, GRIN
lenses with pitches other than 0.25 are also used in
free-space fiber-optic interconnections. Further-
more, an air gap between the SMF and the GRIN lens
is also typical as this prevents backreflection.9 An-
other advantage of an air gap between the SMF and
the GRIN lens is that the magnitude of the gap can be
adjusted to set the location of the beam waist at a
certain distance after the exit plane of the GRIN lens.
We call such a lens a self-imaging lens. There can be
a minimum coupling power loss when these GRIN
lenses are separated by a certain distance that is

550 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003

labeled the self-imaging distance.10 Ideally there
will be zero coupling power loss when two identical
GRIN lenses are used and the separation distance
between the collimator end faces is twice the GRIN
lens beam-waist distance. We call this phenomena
the self-imaging mechanism because the symmetry of
the optical components placement 共i.e., fibers, GRIN
lenses, and free-space air gaps兲 forms an imaging-
type situation leading to minimal coupling loss.
Recently, this self-imaging effect has been experi-
mentally confirmed.11 Note that the air gap can be
replaced by a tapered GRIN lens solid-optic SMF
interconnect design leading to the same beam char-
acteristics that lead to the self-imaging mechanism.
Also note that the optical beam characteristics of
what we call the self-imaging mechanism has been
known and utilized in optical designs such as laser
cavities12 and perhaps in other commercial free-
space optics.
We begin this paper by providing the basics of a
given GRIN lens and related Gaussian beam charac-
teristics that are required to lay out the proposed
foundations for GRIN lens free-space coupling loss.
The separation distance for which the power loss be-
tween the two coupled GRIN lenses is the least de-
pends on the location of the beam waist. In this
paper, using ray matrix theory, we find an analytical
expression for this beam waist location as a function
of the parameters of the GRIN lens and the SMF. In
addition, the power loss that is due to misalignment
of two free-space coupled self-imaging collimators is
analyzed. The coupling power loss is determined
with the aid of the coupling loss coefficient defined as
the overlap area of the electric field amplitude of the
two collimator output beams.13 Considering the
common case of small angular tilt misalignment 共e.g.,
⬍0.3 deg兲, approximations are made leading to an
analytical expression for the power coupling loss that
is written as a function of the separation, lateral
offset, and angular tilt misalignment. Experiments
are performed to match the misalignment-based cou-
pling loss formulas for self-imaging collimators.
2. Basics of a Gradient-Index Lens
A. Design and Coupling-Loss Analysis of
Self-Imaging Collimators
The classic GRIN lens design, such as that by Nippon
Sheet Glass Japan, used a radial refractive-index
variation to realize the rod lens effect. Specifically,
this GRIN lens has a parabolic-shaped refractive-
index profile where this refractive index is expressed
as
冉
nr ⫽ n0 1 ⫺
Ar
2
2
冊 , (1)
where n0 is the refractive index at the center of the
GRIN lens, 公A is the gradient constant, and r is the
distance to the central axis. More recently, another
GRIN lens design called GRADIUM by LightPath
Technologies has emerged in which the refractive in-
Fig. 1. Propagation of the Gaussian beam through a GRIN lens.
The solid curves represent the beam width of the Gaussian beam.
The complex radii of curvature at the end of the SMF, at the left-
and right-hand side of the GRIN lens, and at the location of the
beam waist are denoted as q0, q1, q2, and q3, respectively.
dex varies in the axial direction. The specific axial
index change profile is, to our knowledge, not public
information. Because our earlier research6,7 used
radial-type GRIN lens theory, we continue in this
paper with radial GRIN lens analysis. Neverthe-
less, the concepts and approach taken in this paper
can be applied in general to any GRIN lens, including
the GRADIUM lens.
In general, a SMF is not directly connected to the
GRIN lens. Specifically, in a typical fiber collimator
package there is a small air gap between the SMF
and the GRIN lens 共of the order of 1 ␮m兲, where this
gap also prevents backreflection.9 When the light
has crossed the gap, it is captured and collimated by
the GRIN lens, finally traveling through the air
again. Figure 1 shows schematically how the
Gaussian beam propagates through the GRIN lens.
The objective is to determine the location of the beam
waist of the Gaussian beam after the GRIN lens as a
function of the magnitude of the gap between the end
of the fiber and the entrance face of the GRIN lens
and some parameters of the GRIN lens.
To determine the state of a Gaussian beam after it
has passed through an optical system, e.g., a lens or
a mirror, one can make use of ray matrices and the
ABCD law.8 If the gap between the end of the SMF
and the GRIN lens has a length L, the ray matrix of
this gap is14
Mgap ⫽ 冋 册
1 L
0 1
. (2)
Subsequently, the Gaussian beam will enter a GRIN
lens. The ray matrix of the GRIN lens is given by15
冤 sin共 冑AZ兲
冥
1
cos共 冑AZ兲
MGRIN ⫽ n 0 冑A .
⫺n 0 冑A sin共 冑AZ兲 cos共 冑AZ兲
(3)
Here Z is the length of the lens. The pitch of a
GRIN lens is defined as p ⫽ 公AZ兾共2␲兲. A quarter-
pitch GRIN lens 共 p ⫽ 0.25兲 means that 公AZ ⫽ ␲兾2.
A quarter-pitch GRIN lens will transform the light
emitted by a point source at an edge of the lens into
a parallel light beam at the other edge of the lens.
However, in this paper we do not use a quarter-pitch
lens to enable the proposed self-imaging collimator.
20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 551
Finally, after the light has left the GRIN lens, the
Gaussian beam travels through the air again. The
ray matrix is similar to Eq. 共2兲 and equals

Md ⫽ 冋 册
1 d
0 1
. (4)

Here, d is the distance the Gaussian beam has trav-

eled after the GRIN lens. It is assumed that the
light has its beam waist at the end of the fiber.
Hence the complex curvature parameter at the end of
the fiber can be written as12,16
␲nw 02
q0 ⫽ i ⫽ iz 0. (5)
␭
where w0 is given by17
w0
⬵ 0.65 ⫹ 1.619V ⫺3兾2 ⫹ 2.879V ⫺6,
a Fig. 2. Distance d between the edge of the GRIN lens and the
beam waist as a function of the pitch of the GRIN lens according to
1.2 ⬍ V ⬍ 2.405. (6) Eq. 共9兲. For the gap L between the SMF and the GRIN lens a
By applying the ABCD law with Eq. 共2兲 as a matrix, value of 1 ␮m was taken. For a pitch of approximately 0.26, the
beam-waist distance has its maximum value of 40 mm.
we can determine the complex beam radius when the
Gaussian beam reaches the entrance face of the
GRIN lens. The complex beam radius at this point
is denoted as q1. To obtain the complex beam radius
at a distance d after the GRIN lens, denoted as q3, we of ␭ ⫽ 1550 nm is used in the calculations. It can
apply the ABCD law two more times using Eqs. 共3兲 be observed that, for a pitch of approximately 0.25,
and 共4兲 leading to the distance from the GRIN lens edge to the beam
waist becomes zero. The value of the pitch at
␣␥ ⫹ ␤␦ ␤␥ ⫺ ␣␦ which d becomes zero depends on the gap between
q3 ⫽ ⫹d⫹i 2 , (7) the fiber and the GRIN lens. A sharp maximum is
␥2 ⫹ ␦2 ␥ ⫹ ␦2
recognizable for a value of the pitch of approxi-
where mately 0.26.
Next the value of the beam waist is found. The
␣ ⫽ L cos共 冑AZ兲 ⫹ sin共 冑AZ兲,
1 beam waist is denoted as wT, where the T refers to
(7a)
n 0 冑A transmitter, and is given by

␤ ⫽ z 0 cos共 冑AZ兲, (7b)

册 冤 冥
1兾2
␤␥ ⫺ ␣␦
␥ ⫽ ⫺n 0 冑AL sin共 冑AZ兲 ⫹ cos共 冑AZ兲,
冋 1兾2 ␭
(7c) ␭ Im共q 3兲 ␥2 ⫹ ␦2
w ⫽ wT ⫽ ⫽ . (10)
␦ ⫽ ⫺n 0 冑Az 0 sin共 冑AZ兲. (7d) ␲n ␲n

The next objective is to determine the location after

the GRIN lens where the Gaussian beam has its
beam waist. The beam waist can be found at a cer-
tain value for d such that the condition
Re共q 3兲 ⫽ 0
is satisfied. This implies that the beam waist can be
found at a distance
␣␥ ⫹ ␤␦
d⫽⫺
␥2 ⫹ ␦2
after the GRIN lens. Figure 2 shows a graph of
this distance as function of the GRIN lens pitch.
Here the gap between the fiber and the GRIN lens
has a value of 1 ␮m. The GRIN lens has a core
refractive index of n0 ⫽ 1.6 and gradient constant of
公A ⫽ 0.3 ⫻ 103 m⫺1. Furthermore, a wavelength
B. Gradient-Index-Lens–Gradient-Index-Lens Coupling
Loss in the Symmetric Self-Imaging Case
In this subsection, a formula is derived to predict the
power coupling loss between two misaligned GRIN
(8) lenses. In Subsection 2.A the Gaussian beam ap-
proximation and the propagation of a Gaussian beam
through a GRIN lens were discussed. The parame-
ters that were investigated in Subsection 2.A, e.g.,
the beam waist and its location, are used in the the-
(9) ory of the coupling loss formula. To describe the
electric fields belonging to a Gaussian beam, a coor-
dinate frame is introduced. The origin is located at
the center of the exit face of the GRIN lens, and the
z axis coincides with the propagation axis of the
Gaussian beam 共see Fig. 3兲. If a fiber is directly
coupled to a quarter-pitch GRIN lens, the beam waist
would be found at the exit face of the GRIN lens.

552 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003

Fig. 3. SMF directly coupled to a quarter-pitch lens and the beam Fig. 5. GRIN lens 2 in coordinate frame 2. The location of the
width of the Gaussian beam. The beam waist is located at the beam waist is z⬘ ⫽ d1.
edge 共called the exit face兲 of the GRIN lens.

should be replaced by 共 z ⫺ d1兲 in Eq. 共11兲, and the

electric field in coordinate frame 1 can be described by
E x共 x, y, z兲 ⫽ Ẽ x共 x, y, z ⫺ d 1兲

⫽ E1
wT
w共 z ⫺ d 1兲 再
exp ⫺i关k共 z ⫺ d 1兲

Fig. 4. General case in which the beam waist does not coincide
⫺ ␩共 z ⫺ d 1兲兴 ⫺ r 2 冋 1
w 共 z ⫺ d 1兲
2

with the edge of the GRIN lens. The beam waist is located at z ⫽
d 1. ⫹i
k
2 R共 z ⫺ d 1兲 册冎 . (16)

The x component of the electric field can be expressed Because a typical transmissive interconnection ge-
as8 ometry uses two GRIN lenses, a second GRIN lens is

再
introduced for the coupling loss analysis. In general,
wT this lens depicted as GRIN lens 2 in Fig. 5 will be able
Ẽ x共 x, y, z兲 ⫽ E 1 exp ⫺i关kz ⫺ ␩共 z兲兴
w共 z兲 to capture only a part of the light emitted by GRIN lens

冋 册冎
1 whereas the other part is lost. The magnitude of
1 k this coupling loss is determined by the overlap of the
⫺ r2 ⫹i , (11)
w 共 z兲
2
2 R共 z兲 electric field of the beam produced by GRIN lens 1 with
the beam equivalent of GRIN lens 2. This beam
where E1 is the maximum electric field that is located equivalent can be treated in the same way as the beam
at the origin 共 x ⫽ y ⫽ z ⫽ 0兲 and r is the distance to produced by GRIN lens 1. In analogy with GRIN
the z axis, lens 1, the entrance face of GRIN lens 2 is placed at
2␲n the origin of another coordinate frame, as shown in
k⫽ , (12) Fig. 5. The new coordinates are distinguished with
␭ primes 共x⬘, y⬘, z⬘兲 and are called coordinate frame 2.

␩共 z兲 ⫽ tan⫺1冉 冊 ␭z
␲nw T2
, (13)
For initial analysis, a special case is considered,
namely, a symmetric optical coupling setup. In
other words, the two GRIN lenses are identical, and

冋 冉 冊册
w 2共 z兲 ⫽ w T2 1 ⫹
␭z
␲nw T2
2

, (14)
the gap between GRIN lens 2 and SMF2 is the same
as the gap between GRIN lens 1 and SMF1. This
causes the distance from the entrance face of GRIN

冋 冉 冊册 2 lens 2 to the beam waist to be d1 again. Below, in

␲nw T2 Subsection 2.C, this setup is generalized. Analo-
R共 z兲 ⫽ z 1 ⫹ . (15)
␭z gously to the beam produced by GRIN lens 1, in co-
ordinate frame 2 the x⬘ component of the electric field
Consider now the general situation in which the
belonging to the beam equivalent of GRIN lens 2 in
GRIN lens is not a quarter-pitch lens and there might
coordinate frame 2 can be written as
be a gap between the fiber and the GRIN lens. In
Subsection 2.A a relationship for the location of the E x⬘共 x⬘, y⬘, z⬘兲 ⫽ Ẽ共 x⬘, y⬘, z⬘ ⫹ d 1兲

再
beam waist was derived in terms of the magnitude of
the gap between the fiber and GRIN lens and the wT
pitch of the GRIN lens. If it is assumed that the ⫽ E1 exp ⫺i关k共 z⬘ ⫹ d 1兲
w共 z⬘ ⫹ d 1兲

冋
distance between the exit face of a certain GRIN lens
1 and the location of the beam waist of the Gaussian 1
⫺ ␩共 z⬘ ⫹ d 1兲兴 ⫺ r⬘ 2
beam is equal to d1 共see Fig. 4兲, then the coordinate w 2共 z⬘ ⫹ d 1兲

册冎
frame is shifted a distance d1 to the left compared
with the previous setup 共see Fig. 3兲. This frame is k
⫹i , (17)
called coordinate frame 1. The shift implies that z 2 R共 z⬘ ⫹ d 1兲

20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 553

 bols Z0, X0, and ␪, respectively. Because the funda-
mental mode of the Gaussian beam is rotationally
symmetric around its axis of propagation, it is neither
necessary to distinguish a ␪x 共in the x, z plane兲 and a
␪y 共in the y, z plane兲 nor to distinguish a lateral offset
misalignment in the x and y directions. First, the
separation misalignment is taken into consideration,
and the lateral offset and angular tilt misalignment
are not considered. In this special case, the relation
between the two coordinate frames is
x ⫽ x⬘, y ⫽ y⬘, z ⫽ z⬘ ⫹ Z 0, r ⫽ r⬘. (19)

With the aid of this coordinate transformation, Eq.

共16兲 can be expressed in terms of x⬘, y⬘, and z⬘. Sub-
stituting z⬘ ⫽ 0 leads to
Fig. 6. Three types of misalignment: 共a兲 separation misalign-
ment 共Z0兲, 共b兲 lateral offset misalignment 共X0兲, 共c兲 angular tilt
misalignment 共␪兲. d1 ⫽ d2 ⫽ 100 mm, wT ⫽ wR ⫽ 0.5 mm, ␭ ⫽
E x共 x, y, z兲兩 z⬘⫽0 ⫽ E 1
wT
w共Z 0 ⫺ d 1兲 再
exp ⫺i关k共Z 0 ⫺ d 1兲
1550 nm.
⫺ ␩共Z 0 ⫺ d 1兲兴 ⫺ 共r⬘兲 2 冋 1
w 共Z 0 ⫺ d 1兲
2

where r⬘ represents the distance to the z⬘ axis.

As mentioned above, the coupling loss depends on
the overlap of the beam produced by GRIN lens 1
⫹i
k
2 R共Z 0 ⫺ d 1兲 册冎 . (20)

with the beam equivalent of GRIN lens 2, or the Substituting Eqs. 共17兲 and 共20兲 into Eq. 共18兲 and us-
overlap of Eq. 共16兲 with Eq. 共17兲. The coupling coef- ing the standard integral
ficient ␩c at z⬘ ⫽ 0 takes the form18,19

␩c ⫽
2
␲E 12w T2 兰兰 E x共 x, y, z兲兩 z⬘⫽0 兰兰 exp共⫺␣r 2兲dxdy ⫽
␲
␣
, (21)

⫻ E x⬘*共 x⬘, y⬘, z⬘兲兩 z⬘⫽0dx⬘dy⬘. (18) we obtain

2 expi关k共2d 1 ⫺ Z 0兲 ⫹ ␩共Z 0 ⫺ d 1兲 ⫺ ␩共d 1兲兴

再 冋 册冎
␩c ⫽
1 1 ik 1 1
w共Z 0 ⫺ d 1兲w共d 1兲 2 ⫹ 2 ⫹ ⫺ . (22)
w 共Z 0 ⫺ d 1兲 w 共d 1兲 2 R共Z 0 ⫺ d 1兲 R共d 1兲

If there is no loss, the coupling coefficient should The power transmission coefficient indicates what
equal one. The magnitude of the coupling coefficient part of the power emitted by GRIN lens 1 is captured
is dependent on the position of GRIN lens 1 relative by GRIN lens 2. This power transmission coefficient
to GRIN lens 2, or in other words, the position of the is written as T ⫽ 兩␩c兩2 and becomes

冋 册
T⫽ 2 . (23)
w 共d 1兲
2
w 共Z 0 ⫺ d 1兲
2
k2 1 1
⫹ 2 ⫹ ⫹ w 2
共Z 0 ⫺ d 1 兲w 2
共d 1 兲 ⫺
w 共Z 0 ⫺ d 1兲
2
w 共d 1兲
2
4 R共Z 0 ⫺ d 1兲 R共d 1兲

coordinate frame 1 relative to coordinate frame 2.
This relative position is determined by three types of
misalignment between the two lenses. These mis-
alignments are separation misalignment 关Fig. 6共a兲兴,
lateral offset misalignment 关Fig. 6共b兲兴, and angular
tilt misalignment 关Fig. 6共c兲兴 represented by the sym-
For d1 ⫽ 0, this result is exactly the same as the
formula found in earlier studies.6,20 From Eq. 共23兲,
it is verified that there is no loss 共T ⫽ 1兲 if the
separation distance equals 2 times d1. This is the
critical condition that implements the earlier pro-
posed self-imaging technique for coupling between

554 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003

 Fig. 8. GRIN–GRIN coupling case in which we combined the
three types of misalignment, i.e., separation misalignment 共Z0兲,
lateral offset misalignment 共X0兲, and angular tilt misalignment 共␪兲.

typical case for GRIN lens misalignment with SMF

coupling兲 and that
Fig. 7. Relationship between the coupling loss and the separation
misalignment, where there exists a mirror symmetry in which
Z0 ⫽ 2d1. For a separation distance of Z0 ⫽ 2d1, there is theo- w 12共 x⬘ sin ␪ ⫹ Z 0 ⫺ d 1兲 ⬇ w 12共Z 0 ⫺ d 1兲, (28)
retically no loss. Lateral offset and angular tilt misalignments
are assumed to be nonexistent.
R 1共 x⬘ sin ␪ ⫹ Z 0 ⫺ d 1兲 ⬇ R 1共Z 0 ⫺ d 1兲, (29)
two GRIN lenses.10 This calculated result is
shown in Fig. 7. For this example, the beam waist ␩ 1共 x⬘ sin ␪ ⫹ Z 0 ⫺ d 1兲 ⬇ ␩ 1共Z 0 ⫺ d 1兲. (30)
of both GRIN lenses was wT ⫽ wR ⫽ 0.5 mm and
their beam waist distance was d1 ⫽ d2 ⫽ 100 mm. Furthermore, the approximation cos ␪ ⬇ 1 has been
Infrared light with a wavelength of 1550 nm is made, because ␪ is small. These assumptions lead
assumed in the simulation. It is convenient to ex- to an approximation for Ex at z⬘ ⫽ 0 to be
press the loss in decibels, where L共dB兲 ⫽ ⫺10 log T.
In Fig. 7 the coupling loss in decibels is plotted as a
function of Z0. Note that there will be a loss when wT
Z0 is equal to zero, or in other words when the E x共 x⬘, y⬘, z⬘兲兩 z⬘⫽0 ⬵ E 1 exp兵⫺i关k共 x⬘ sin ␪
w 1共Z 0 ⫺ d 1兲
lenses are directly next to each other. Also note
that there exists a mirror symmetry for the interval ⫹ Z 0 ⫺ d 1兲 ⫺ ␩ 1共Z 0 ⫺ d 1兲兴其
0 ⬍ Z0 ⬍ 4d1 with Z0 ⫽ 2d1 as the symmetry axis.

C. Gradient-Index-Lens–Gradient-Index-Lens Coupling 再
⫻ exp ⫺关共 x⬘ ⫹ X 0兲 2兴

冋
Loss in a General Self-Imaging Case
In Subsection 2.B we discussed the special case of 1
⫻
the symmetric setup where only the separation mis- w 1 共Z 0 ⫺ d 1兲
2

alignment is examined. In this subsection, we con-
sider the general case of a nonsymmetric setup.
Furthermore, not only the separation misalignment
is taken into account, but both lateral offset as well
as angular tilt misalignments are investigated. In
Fig. 8 the three types of misalignment are com-
bined. From Fig. 8 we can derive
x ⫽ x⬘ cos ␪ ⫺ z⬘ sin ␪ ⫹ X 0, (24)
z ⫽ x⬘ sin ␪ ⫹ z⬘ cos ␪ ⫹ Z 0, (25)
册冎
k
⫹i . (31)
2 R 1共Z 0 ⫺ d 1兲
In approximations 共28兲–共30兲 a subscript 1 is added to
the functions w, ␩, and R to indicate that these func-
tions belong to GRIN lens 1. Because the general
case of a nonsymmetric setup is studied, we assumed
that the beam-waist distance of GRIN lens 2 equals
d2 instead of d1. Thus, instead of Eq. 共17兲 for the
symmetric situation, we use

y ⫽ y⬘, (26) E x⬘共 x⬘, y⬘, z⬘兲 ⫽ E 1

wT
w 2共 z⬘ ⫹ d 2兲 再
exp ⫺i关k共 z⬘ ⫹ d 2兲

r 2 ⫽ x 2 ⫹ y 2 ⫽ 共 x⬘ cos ␪ ⫺ z⬘ sin ␪ ⫹ X 0兲 2 ⫹ y 2. (27) ⫺ ␩ 2共 z⬘ ⫹ d 2兲兴 ⫺ r⬘ 2 冋 1

w 2 共 z⬘ ⫹ d 2兲
2

册冎
Using Eqs. 共24兲–共27兲 we can express Eq. 共16兲 in terms
of x⬘, y⬘, and z⬘. To maintain an analytical solution k
⫹i . (32)
for ␩c, we assume that sin ␪ is small 共e.g., ␪ ⬍ 0.30, a 2 R 2共 z⬘ ⫹ d 2兲

20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 555

The subscripts 2 in the functions ␩2, w2, and R2
means that wT in Eqs. 共13兲–共15兲 is replaced by wR.
The subscript R refers to receiver. Hence

␩ 2共 z⬘兲 ⫽ tan⫺1 冉 冊
␭z⬘
␲nw R2
, (33)

w 22共 z⬘兲 ⫽ w R2 1 ⫹冋 冉 冊册 ␭z⬘

␲nw R2
2

, (34)

R 2共 z⬘兲 ⫽ z⬘ 1 ⫹冋 冉 冊册
␲nw R2
␭z⬘
2

. (35)

Substituting relation 共31兲 and Eq. 共32兲 into Eq. 共18兲 Fig. 9. Coupling loss as a function of the lateral offset misalign-
and using the integral ment for different values of the separation misalignment with no
angular tilt misalignment. The plots belonging to Z0 ⫽ d1 and

冑 冉 冊
Z0 ⫽ 3d1 coincide. Also the plots belonging to Z0 ⫽ 0 and Z0 ⫽ 3d1

兰 ␲ ␤ 2 ⫺ 4␣␥
⫹⬁
are exactly the same. The loss equals zero when Z0 ⫽ 2d1 and
exp关⫺共␣x 2 ⫹ ␤x ⫹ ␥兲兴dx ⫽ exp , X0 ⫽ 0.
⫺⬁
␣ 4␣
(36)
be split up into real and imaginary parts, which are
denoted as J and ␺2, respectively. It follows that
we obtain

␩c ⫽
2
exp共i␺ 1兲exp 冉
G 2 ⫺ 4FH
. 冊
J ⫽ Re 冉 G 2 ⫺ 4FH
4F
冊
w 1共Z 0 ⫺ d 1兲w 2共d 2兲 F 4F F r共G r2 ⫺ G i2 ⫺ 4F r H r兲 ⫹ F i 共2G r G i ⫺ 4F i H r兲
(37) ⫽ ,
4共F r2 ⫹ F i2兲
(42)

冉 冊
Here
G 2 ⫺ 4FH
␺ 2 ⫽ Im
1 1 4F
F ⫽ F r ⫹ iF i ⫽ 2 ⫹ 2
w 1 共Z 0 ⫺ d 1兲 w 2 共d 2兲 ⫺F i 共G r2 ⫺ G i2 ⫹ 4F i H i兲 ⫹ F r共2G r G i ⫺ 4F r H i兲
⫽ .

冋 册
4共F r2 ⫹ F i2兲
k 1 1
⫹i ⫺ , (38) (43)
2 R 1共Z 0 ⫺ d 1兲 R 2共d 2兲
Using Eqs. 共42兲 and 共43兲, we can write the coupling
coefficient as
2X 0
G ⫽ G r ⫹ iG i ⫽
w 1 共Z 0 ⫺ d 1兲
2
2
␩c ⫽ exp关i共␺ 1 ⫹ ␺ 2兲兴exp共 J兲.

冋 册
w 1共Z 0 ⫺ d 1兲w 2共d 2兲 F
X0
⫹ ik ⫹ sin ␪ , (39) (44)
R 1共Z 0 ⫺ d 1兲
The power transmission coefficient can be calculated
with T ⫽ 兩␩c兩2 and the loss in decibels with L共dB兲 ⫽
X 02 kX 02 ⫺10 log T. This loss becomes
H ⫽ H r ⫹ iH i ⫽ ⫹

冋 册
i ,
w 12共Z 0 ⫺ d 1兲 2 R 1共Z 0 ⫺ d 1兲 4 exp共2J兲
(40) L ⫽ ⫺10 log . (45)
w 1 共Z 0 ⫺ d 1兲w 22共d 2兲兩 F兩 2
2

For d1 ⫽ d2 ⫽ 0, Eq. 共45兲 is in agreement with the

␺ 1 ⫽ k共d 1 ⫹ d 2 ⫺ Z 0兲 ⫹ ␩ 1共Z 0 ⫺ d 1兲 ⫺ ␩ 2共d 2兲.
For F, G, and H, the subscripts r and i represent the
real and imaginary value of that number, respec-
tively. Next, the argument of the last exponent can
(41)
results found by earlier research.6,20 To compare
this relation with Eq. 共23兲, belonging to the symmet-
ric setup, we assumed that d2 ⫽ d1 and that wR ⫽ wT.
In Fig. 9, Eq. 共45兲 is plotted in absence of angular
tilt misalignment for different values of the separa-

556 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003

tion distance between the two GRIN lenses in terms
of d1. Note that some plots in Fig. 9 coincide, e.g.,
the graph of Z0 ⫽ d1 is exactly the same as the graph
for Z0 ⫽ 3d1. This is also the case for the graphs
belonging to Z0 ⫽ 0 and Z0 ⫽ 4d1 and is in agreement
with the symmetry of Fig. 7. Furthermore, it can be
observed that there is no loss for Z0 ⫽ 2d1 and X0 ⫽
0, which corresponds with the self-imaging condition.
To make this graph, we assumed that the beam waist
of both lenses was wT ⫽ wR ⫽ 0.3 mm and their
distances to their beam waist d1 ⫽ d2 ⫽ 200 mm. A
wavelength of ␭ ⫽ 1550 nm was assumed for the
calculation.
If there is no angular tilt misalignment, an expres-
sion for the loss as a function of the lateral offset loss
can be found. This loss, denoted as Loffset, can be
written as

L offset ⫽ L共Z 0, X 0, ␪ ⫽ 0兲 ⫺ L共Z 0, X 0 ⫽ 0, ␪ ⫽ 0兲

Fig. 10. Loss as a function of the angular tilt misalignment for
⫽ A offsetX 02, (46) different values of the separation misalignment. The lateral off-
set misalignment is fixed at zero. When there is no angular tilt
in which the offset coefficient Aoffset is equal to misalignment, the loss for both the pairs Z0 ⫽ d1 and Z0 ⫽ 3d1 as

再冋
well Z0 ⫽ 0 and Z0 ⫽ 4d1 is the same. There is no loss when Z0 ⫽
⫺5 1 4 k2 2d1, and the angular tilt misalignment is equal to zero.
A offset ⫽ F r ⫺
ln 10 兩 F兩 2 w 14共Z 0 ⫺ d 1兲 R 12共Z 0 ⫺ d 1兲

⫺
4F r
册
w 1 共Z 0 ⫺ d 1兲
2
Ltilt is introduced for analysis. This loss indicates
what the influence is of the angular tilt in the absence

冋 册冎
of lateral offset misalignment. This loss is equal to
4 k
⫹ Fi ⫺ Fi . L tilt ⫽ L共Z 0, X 0 ⫽ 0, ␪兲 ⫺ L共Z 0, X 0 ⫽ 0, ␪ ⫽ 0兲
w 1 共Z 0 ⫺ d 1兲 R 1共Z 0 ⫺ d 1兲
2

(47) ⫽ A tilt sin2 ␪ ⬇ A tilt␪ 2, (48)

In other words, the offset loss curve is a parabola and
the coefficient in front of the square term, the offset
coefficient, is dependent on the separation distance Z0.
Figure 10 shows the coupling loss as a function of
the angular tilt misalignment for different values of
the separation distance Z0 according to Eq. 共45兲.
Here X0 is fixed at zero. Note that the coupling loss
is sensitive for angular tilt misalignment. Again, a
loss of 0 dB is reached when there is no angular tilt
misalignment and Z0 is equal to 2 times d1. Finally,
it can be observed that, if there is no angular tilt
misalignment, the loss for a separation distance of
Z0 ⫽ d1 is equal to the loss for a separation distance
of Z0 ⫽ 3d1. The loss is also the same for Z0 ⫽ 0 and
Z0 ⫽ 4d1, if angular tilt misalignment is absent. To
make the measurements shown in Fig. 10, the same
values for d1, d2, wT, wR, and ␭ were used as in Fig. 9.
The proposed theory indicates that, dependent on
the accuracy of each type of misalignment, the best
where the tilt coefficient can be written as
5F r k 2
A tilt ⫽ . (49)
ln 10兩 F兩 2
In addition to the three misalignments, there is
another possible source for the coupling loss, namely,
spot-size mismatch. When the beam waists of the
beams produced by the GRIN lenses are not exactly
the same, it is impossible in theory to find a zero-loss
setup. Assuming there is neither lateral offset nor
angular tilt misalignment, the least loss will be found
at a separation distance equal to the beam-waist dis-
tance of the transmitting lens plus the beam-waist
distance of the receiving lens 共Z0 ⫽ d1 ⫹ d2兲. In this
setup the theoretical coupling power loss that is due
to spot-size mismatch is calculated to be
4

冉 冊
setup to reduce the loss is different. Although the L ⫽ ⫺10 log 2 . (50)
ideal separation distance is Z0 ⫽ 2d1 when there is no wR wT
⫹
lateral offset and angular tilt misalignment, from wT wR
Fig. 10 it can be concluded, for example, that one
should prefer a separation distance of Z0 ⫽ d1 to a This result is the same when uses two non-self-
separation distance of Z0 ⫽ 2d1, if the angular tilt imaging GRIN lenses 共d1 ⫽ d2 ⫽ 0兲 as described in an
misalignment is larger than approximately 0.045 earlier study.6 Usually the difference between the
deg. The point of intersection belonging to this case beam waists wT and wR is quite small, so the spot-size
is indicated with an arrow in Fig. 10. mismatch loss is also small compared with the loss
Like a lateral offset loss Loffset, an angular tilt loss that is due to misalignment. It follows that, for

20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 557

Fig. 11. Overview of the mirror experiment setup. A circulator is used to send the light from the source toward the GRIN lens and the
reflected light toward the detector. This detector is connected to a powermeter that measures the reflected power. i兾o, in兾out.
0.7 ⬍ wR兾wT ⬍ 1.4, the loss that is due to spot-size
mismatch is less than 0.5 dB.
3. Experimental Verification of the Self-Imaging Theory
In this paper we have established the theory behind
the GRIN–GRIN coupling loss for the self-imaging
condition. To verify this theory, experiments are
performed. At first, some parameters of the Gauss-
ian beam produced by the used GRIN lenses are de-
termined. The location and magnitude of the beam
waist can be determined by means of a mirror exper-
iment and a razor blade experiment, respectively.
Second, the coupling loss for separation misalign-
ment is verified experimentally. Finally, suitable
experiments are performed to verify the coupling loss
theory for both lateral offset as well as angular tilt
misalignment. For these experiments we used a
transmitting GRIN lens and a movable triangular
glass prism, which created another imaginary receiv-
ing GRIN lens with a certain misalignment relative
to the transmitting GRIN lens. By moving this
prism, we could change the artificial lateral offset or
angular tilt. For the experiments a LightPath Tech-
nologies GRIN lens T5100 series is used because its
design exhibits the self-imaging behavior, and other
commercial GRIN lens manufacturers such as NSG
currently do not produce self-imaging-type lenses.
We denote this lens as GRIN lens 1.
A. Identical Gradient-Index Lens Coupling Case
A GRIN lens emits a Gaussian beam with its beam
waist a certain distance d after the exit face of the
GRIN lens. In this experiment this distance is de-
termined. In Subsection 2.B a theoretical formula
for the loss as a function of the separation distance of
two identical GRIN lenses was derived 关see Eq. 共23兲兴.
Here both lateral offset as well as angular tilt mis-
alignment were assumed to be absent. A simulation
of a setup in which two identical GRIN lenses are
facing each other without the two above-mentioned
types of misalignment can be realized by means of a
mirror placed perpendicular to the Gaussian beam.
This way the imaginary separation distance between
the two identical GRIN lenses is twice as large as the
distance from the GRIN lens to the mirror. A sche-
matic overview of the experimental setup is shown in
Fig. 11 where we placed a horizontally movable mir-
ror in front of the GRIN lens. The Gaussian beam
558 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003
emitted by the GRIN lens is reflected by the mirror
and is captured by the same GRIN lens. The power
of the captured beam can be measured with a detec-
tor connected to a powermeter by a circulator. A
maximum optical power will be detected when the
distance from the mirror to the GRIN lens is equal to
the beam-waist distance. By moving the mirror in
front of the GRIN lens, we can find this beam-waist
distance. It is important that the mirror is perpen-
dicular to the Gaussian beam. If this is not the case,
one creates an undesired lateral offset as well as an
angular tilt misalignment.
Two different mirror experiments were performed.
At first, the mirror is aligned after each horizontal
shift. In theory, this realignment is not necessary,
but in practice the mirror can be misaligned by small
vibrations during the horizontal movement of the
mirror. The disadvantage of this method is that the
mirror at one distance could be better aligned than at
another distance. This makes the experiment less
reliable. Nevertheless, we can compensate this
measurement error by taking many measurement
points. For GRIN lens 1, the results of the experi-
ment in which the mirror is realigned each time can
be observed in Fig. 12. To obtain reliable results,
this experiment is repeated several times. The re-
sults of the different measurements showed consis-
tency. In this plot the loss of the connectors 共0.1 dB
for each connector兲 and the loss of the circulator 共1.1
dB兲 are taken into account. According to this plot,
Fig. 12. Coupling loss result of the mirror experiment when we
adjust the mirror for GRIN lens 1. From these results it follows
that the beam-waist distance is equal to 4 cm.
Fig. 13. Coupling loss result of the mirror experiment with no
adjustment to the mirror. According to this experiment, GRIN
lens 1 has a beam-waist distance of 4 cm.
the beam-waist distance of GRIN lens 1 is approxi-
mately 4 cm. Note that this is an approximate dis-
tance because the minimum of the plot is smooth.
Another way to perform the experiment is to align
the mirror only once. This alignment is done at
the largest possible distance between the GRIN
lens and the mirror for the following reason. The
larger the distance from the GRIN lens to the mir-
ror, the more sensitive the detected optical power is
for misalignment. Hence, if the mirror is well
aligned at a large distance, it certainly is aligned for
smaller distances. To reduce vibrations, we use a
translational stage that is able to move smoothly.
The results can be observed in Fig. 13. This ex-
periment also indicates that the beam-waist dis-
tance for GRIN lens 1 is ⬃4 cm. The minimum
measured coupling loss for both experiments is ap-
proximately 0.35 dB, where a 0.18-dB loss is due to
the mirror. Note that, for larger separation dis-
tances between the mirror and the GRIN lens, a
small misalignment of the mirror leads to a larger
undesired offset misalignment 共see Fig. 14兲. This
leads to a higher loss for large separation distances,
which is an inevitable consequence of the used ex-
Fig. 14. 共a兲 For a large separation distance a certain misalign-
ment ␪ of the mirror leads to a small overlap area or a large loss.
共b兲 For a smaller separation distance the same misalignment ␪
results in a larger overlap, thus a smaller loss.
Fig. 15. Results for the razor blade experiment for GRIN lens 1.
For GRIN lens 1 the beam waist is 0.479 mm.
perimental setup. In addition, the presence of ab-
errations that is due to the GRIN lens makes it
practically difficult for insertion loss to go to zero.
The beam waist for the light emitted by the GRIN
lenses is determined with the knife-edge beam pro-
file method.21 Figure 15 shows the result for GRIN
lens 1. The beam waist for GRIN lens 1 is equal to
0.479 mm. This result is in good agreement with
the specifications.22 Theoretically, this experi-
ment could be used to find the beam-waist distance.
By repeating the experiment for several distances
between the GRIN lens and the plane of the razor
blade, we can determine the minimum beam
width—that is, the beam waist. The distance
where this minimum beam width occurs is the
beam-waist distance. In practice, the divergence
is too small and the minimum is too smooth to
determine an accurate value for d.
B. Gradient-Index-Lens–Gradient-Index-Lens Coupling
Loss with a Separation Misalignment
Two GRIN lenses were used for a self-imaging cou-
pling loss experiment. GRIN lens 1 was the trans-
mitting lens. As a receiving lens we used another
T5100 series GRIN lens. This lens has a beam
waist of wR ⫽ 0.48 mm and a beam waist distance
of d2 ⫽ 9 cm. We use these parameters to compare
the experimental data with the theoretical coupling
loss. The results of this separation coupling loss
experiment are shown in Fig. 16. Here the self-
imaging mechanism is confirmed. To make a bet-
ter comparison, we elevated the theoretical loss
curve 0.35 dB. This value of 0.35 dB is the mini-
mum experimental loss. This loss can be caused
by the connectors in between the fibers and possible
small undesired angular tilt and lateral offset mis-
alignments. More importantly, it can also be
caused by aberrations in the GRIN lens. Theoret-
ically, the minimum loss should take place at a
separation distance of Z0 ⫽ d1 ⫹ d2 ⫽ 13 cm. Us-
ing the Fig. 16 plot for this pair of GRIN lenses, we
can obtain a minimum loss at a separation distance
of Z0 ⫽ 12.5 ⫾ 1 cm.
20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 559

Fig. 16. Comparison of the experimental data with our theoreti-
cal result for separation coupling loss between two GRIN lenses.
The theoretical curve was shifted 0.35 dB upward to make a better
comparison. The experiment shows good agreement with theory.
C. Coupling Loss with a Lateral Offset Misalignment
To verify the coupling loss formula for lateral offset
misalignment, an offset experiment is performed.
Using a movable glass triangular prism, we created an
artificial offset misalignment. The principle of this
idea is based on two total internal reflections. To un-
derstand the setup, we consider a straight-light beam
entering the glass prism perpendicularly, exactly in
the middle of the hypotenuse plane, as is shown in Fig.
17共a兲. We are dealing with a Gaussian beam instead
of a straight beam. However, it can be assumed that
the divergence of the Gaussian beam is small enough
relative to the dimensions of the prism. For the re-
flection by the prism, the beam can be treated as if it
were straight.兲 Here only the light paths of the upper
and the lower beam borders are shown. After two
total internal reflections the upper beam border will
coincide exactly with the lower beam border and vice
versa. In other words, the prism will reflect an in-
coming beam upside down. For a symmetric beam,
like the Gaussian beam, the prism will act as a mirror.
Figure 17共b兲 shows that a horizontal shift over a dis-
Fig. 17. Prism is reflecting a light beam 共a兲 without offset and 共b兲 with offset. When there is no lateral offset, the reflected beam exactly
coincides with the incoming beam. The offset of the reflected beam is twice as large as the offset of the prism relative to the incoming
beam.
560 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003
tance X0 leads to a lateral offset misalignment of 2X0.
In the lateral offset setup, the laser source, circulator,
detector, and powermeter are in the same way con-
nected to the GRIN lens as in the mirror experiment.
The main advantage of this method is that only one
GRIN lens is needed, and thus the setup can be con-
sidered as symmetric, i.e., the imaginary GRIN lens is
identical to the real GRIN lens.
The prism experiment is performed with GRIN lens
1 for which we found d ⫽ 4 cm and wT ⫽ 0.479 mm.
The experiment was repeated for several separation
distances between the GRIN lens and the prism. The
separation distance Z0 equals twice the distance from
the prism to the GRIN lens. Figure 18 shows the
experimental data for the separation distances of 8,
110, and 150 cm. In the graphs the normalized offset
loss is plotted versus the lateral offset misalignment,
which equals twice the offset of the prism relative to
the propagation axis of the Gaussian beam.
The normalized offset loss is defined as the loss
that is due to lateral offset only. This is the reason
why the minimum offset loss is set to zero. In reality
there is a certain minimum loss caused by the inter-
nal reflections at the right-angle edges in the prism,
imperfections of the prism, and possible small angu-
lar misalignments of the prism. We could denote
the normalized loss as
冉 冊
Pm
L ⫽ ⫺10 log , (51)
P0
in which Pm is the measured power in a certain setup
with a lateral offset and P0 is the maximum measured
power during the experiment, i.e., when the prism has
no lateral offset relative to the Gaussian beam.
The experimental data were compared with the the-
oretical result as found above. The theoretical curve
Loffset ⫽ AoffsetX02 is also plotted in each chart. For
each separation distance the offset coefficient accord-
ing to Eq. 共47兲 was calculated. The value of Aoffset
belonging to the separation distance is displayed in
Fig. 18. Comparison of the experimental data with the theoretical results for the normalized coupling loss that is due to lateral offset for
a separation distance of 共a兲 8 cm, 共b兲 110 cm, and 共c兲 150 cm. The experimental data match the theory.

each graph in Fig. 18. Here the parameters for GRIN lens. This way an artificial angular tilt mis-
GRIN lens 1 were used, that is, wT ⫽ wR ⫽ 0.48 mm alignment of
and d1 ⫽ d2 ⫽ 4 cm. In general, it can be concluded
that the experimental data match the theory. ␪ ⫽ tan⫺1 冉 冊
2X 0
f
(52)

is created. By moving the prism, we can change the

D. Coupling Loss with Angular Tilt Misalignment angle ␪. Unlike the setups with the mirror and the
In addition to lateral offset loss, the loss that is due to prism, this lens-based design is not a symmetric
angular tilt misalignment needs to be verified experi-
mentally. The setup is partly the same as described
in Subsection 3.C. Now a convex lens is placed in
between the prism and the GRIN lens. The geometry
of the setup is shown in Fig. 19. The GRIN lens emits
a Gaussian beam with a beam waist wT at a distance
d1 away from the GRIN lens. A convex lens is placed
in front of the GRIN lens at exactly its focal distance f.
The distance from the beam waist to the convex lens is
denoted as X and it equals X ⫽ f ⫺ d1. We place the
movable prism at a variable distance L after the con-
vex lens. As described in Subsection 3.C, the prism
will reflect the Gaussian beam with a lateral offset
Fig. 19. Geometry of the angular tilt setup. By placing a convex
twice as large as the offset of the prism. On its way
lens exactly at its focal distance f after the GRIN lens, we can focus
back the beam propagates through the lens for a sec- the reflected Gaussian beam on the exit face of the GRIN lens
ond time. Because the convex lens is exactly its focal under a certain angle ␪. The distance L between the convex lens
distance separated from the GRIN lens, the Gaussian and the prism can be adjusted to vary the imaginary separation
beam will be focused exactly on the exit face of the distance Z0.

20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 561

Fig. 20. Comparison of the experimental data with the theoretical results for the normalized coupling loss that is due to angular tilt for
共a兲 L ⫽ 70 cm and 共b兲 L ⫽ 15 cm.

setup because the Gaussian beam is distorted by the
convex lens. In other words, the imaginary receiv-
ing GRIN lens is not identical to the transmitting
GRIN lens as was the case in the above experiments.
However, by means of ray matrix calculations, the
beam waist wR and the beam-waist location Ybw can
be calculated in terms of X, f, L, and wT. For a
derivation of these expressions, see Appendix A.
The location and magnitude of the beam waist are
B 0 D ⫹ A 0 Cz 02
Ybw ⫽ ⫺ , (53)
B 1 D ⫹ A 1 Cz 02
Now L can be adjusted to vary the beam waist wR and
its location Ybw. The imaginary receiving GRIN lens
is assumed to be a quarter-pitch lens, i.e., it has its
beam waist on its edge or d2 ⫽ 0. Furthermore, we
assumed that ␪ is small enough to make the approx-
imation cos ␪ ⬇ 1 and that the separation distance
between the two GRIN lenses is equal to Z0 instead of
Z0 cos ␪. This assumption is justified for the maxi-
mum angle of 0.20 for which the experiment was
executed. From Fig. 19 it follows that the separa-
tion distance is equal to Z0 ⫽ f ⫺ Ybw. The way the
laser source, the circulator, the detector, and the pow-

再 冎 1兾2 ermeter are connected to the GRIN lens is similar to

␭ D2 ⫹ C2z02 the above setup.
wR ⫽ ,
␲n 关共A1 Ybw ⫹ A 0兲 D ⫺ 共B 1 Y bw ⫹ B 0兲C兴 z 02 Several experiments were performed, each with a
(54) different value for L. A comparison of the theory
with the experimental data was made. The results
respectively, where
for L ⫽ 70 cm and L ⫽ 15 cm are shown in Figs.

A1 ⫽ 冉 冊
2 L
f f
⫺1 , (55)
20共a兲 and 20共b兲, respectively. A plano– convex lens
with a focal distance of f ⫽ 0.75 m was used 共see Fig.
19兲. Furthermore, we assumed that the beam
2L waist of GRIN lens 1 was wT ⫽ 0.48 mm and d1 ⫽
A0 ⫽ 1 ⫺ , (56) 4 cm as was concluded from the razor blade and the
f mirror experiments. Inserting these values in
2共X ⫹ L兲 2LX Eqs. 共53兲 and 共54兲 and using Z0 ⫽ f ⫺ d1, we obtain
B1 ⫽ 1 ⫺ ⫹ 2 , (57) Z0 ⫽ 0.95 mm and wR ⫽ 0.48 mm for L ⫽ 70 cm and
f f Z0 ⫽ 23 cm and wR ⫽ 0.35 mm for L ⫽ 15 cm. It is
2LX assumed that the imaginary receiving GRIN lens
B 0 ⫽ X ⫹ 2L ⫺ , (58) has the beam waist of its beam equivalent at its exit
f face 共quarter pitch兲 and thus d2 ⫽ 0. Subse-

C⫽ 冉 冊
2 L
f f
⫺1 , (59)
quently, it follows from Eq. 共49兲 that Atilt ⫽ 6.34
dB兾mrad2 for L ⫽ 70 cm and Atilt ⫽ 3.75 dB兾mrad2
for L ⫽ 15 cm. For these values the theoretical
curve is also plotted in Figs. 20共a兲 and 20共b兲. This
2共X ⫹ L兲 2LX
D⫽1⫺ ⫹ 2 , (60) theoretical curve matches the experimental data for
f f L ⫽ 70 cm. Although the measured loss for L ⫽ 15
cm is somewhat low compared with the theoretical
␲nw T2 curve, the experimental data generally obey the
z0 ⫽ . (61)
␭ theory.

562 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003

4. Conclusion
A GRIN lens with its beam waist at a certain distance
after its exit face is referred to as a self-imaging
collimator. This collimator has the power to imple-
ment the described low-loss free-space coupling be-
tween GRIN lenses by the so-called self-imaging
mechanism. The beam-waist magnitude and loca-
tion for such a self-imaging collimator are calculated
in terms of the parameters of the radial-type GRIN
lens 共i.e., pitch, core refractive index, gradient index,
length兲, the air gap between SMF and GRIN lens, and
the optical wavelength. For the first time to our
knowledge, theoretical misalignment coupling loss
formulas have been derived to employ self-imaging-
type fiber collimators in SMF-based free-space sys-
tems. Experimental results described in this paper
agreed well with the proposed mathematically derived
expressions. For the first time to our knowledge, the
provided coupling loss theory in this paper allows pho-
through the air again. Next the overall ray matrix
of the described optical system is calculated.
Consider a self-imaging GRIN lens producing a
Gaussian beam with a certain beam waist wT at a
certain distance d1 from the GRIN lens exit face.
We take the beam-waist location as the starting point
rather than the exit face of the GRIN lens. Here the
complex radius of curvature equals
q 0 ⫽ iz 0, (A1)
where
␲nw T2
z0 ⫽ . (A2)
␭
From this point, the Gaussian beam travels a
straight section over a distance X. The ray matrix
belonging to this straight section is

冋 册
tonic engineers to improve designs for optimal ultra-
low-loss coupling in fiber-based free-space systems 1 X
M1 ⫽ . (A3)
such as dynamic wavelength equalizers, variable 0 1
attenuators,23–25 and large three-dimensional optical
cross-connect switches. Second, the Gaussian beam propagates through a
convex lens with focal distance f. For a thin lens the
Appendix A: Propagation of a Gaussian Beam through ray matrix14
the Setup

In Subsection 3.D, a setup containing a triangular
glass prism and a convex lens was proposed to create
an artificial angular tilt misalignment. By varying
the distance L between the prism and the convex
lens, we can change the separation distance Z0 be-
tween the GRIN lenses of the imaginary setup. An
inevitable consequence of the distortion of the convex
lens is that the beam waist of the imaginary GRIN
lens will be changed. In this appendix a derivation
of the expressions for the beam waist and its location
are given.
To determine the state of a Gaussian beam, after it
has passed an optical system, e.g., a lens, one can
冤 冥
1 0
M2 ⫽ 1 (A4)
⫺ 1
f
holds. After the lens the light travels a distance L
before it reaches the prism, which reflects the Gauss-
ian beam with a lateral offset of 2X0. Then the light
travels back until it enters the focal lens for a second
time. The prism and the lateral offset 2X0 are both
small compared with the length L. We assume that
the prism acts as a mirror at the plane perpendicular
to the Gaussian beam located at the straight angle of
the mirror. So the ray matrix

冋 册
make use of ray matrices and the ABCD law.8 When
a certain optical system k has the ray matrix 1 2L
M3 ⫽ (A5)

冋 册
0 1
Ak Bk
Mk ⫽ is applicable to the following sections: the path from
Ck Dk
the convex lens to the prism, the reflection by the
and the complex radius of curvature before the prism, and the path back from the prism to the con-
Gaussian beam enters the optical system is equal to vex lens. Subsequently, the laser light is again fo-
qk, then the complex radius of curvature after the cused by the lens for which matrix M2 holds again.
optical system equals Finally, the Gaussian beam will travel again
through the air over a certain distance Y. Similar to
Ak qk ⫹ Bk Eqs. 共A3兲 and 共A5兲, for this section the ray matrix
q k⫹1 ⫽ .

冋 册
Ck qk ⫹ Dk
1 Y
M4 ⫽ (A6)
To apply the ABCD law, the matrix of the whole 0 1
optical system has to be found. This optical system
includes a straight section through the air to the is used.
convex lens, a distortion by the convex lens, a second To determine the overall ray matrix of the optical
straight section through the air to the prism, a re- system, we multiply the matrices as follows:

冋 册
flection by the prism, again a straight section through
the air back to the lens, a second propagation through A B
M5 ⫽ M4 䡠 M2 䡠 M3 䡠 M2 䡠 M1 ⫽ , (A7)
the lens, and finally the Gaussian beam will travel C D

20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 563

in which an angle ␪ with the normal instead of this distance

冉 冊
parallel to the normal. However, it is assumed that
2 L 2L ␪ is small enough so that cos ␪ ⬇ 1. This assumption
A⫽ ⫺1 Y⫹1⫺ , (A8)
f f f is justified for the maximum angle of 0.20 for which

冋 册
the experiment was executed.
2共X ⫹ L兲 2LX 2LX We can obtain the magnitude of the beam waist,
B⫽ 1⫺ ⫹ 2 Y ⫹ X ⫹ 2L ⫺ , denoted as wR, by inserting Ybw into Eq. 共A12兲.
f f f
From the definition of the complex radius of curva-
(A9) ture, Eq. 共A13兲, it follows that

C⫽
2 L
f f
冉 冊
⫺1 , (A10)
w ⫽ wR ⫽ ⫺ 冋 ␭
␲n Im共1兾q 1兲 册 1兾2

再 冎
2共X ⫹ L兲 2LX 1兾2
D⫽1⫺ ⫹ 2 . (A11) ␭ D 2 ⫹ C 2z 02
f f ⫽ .
␲n 关共 A 1 Y bw ⫹ A 0兲 D ⫺ 共B 1 Y bw ⫹ B 0兲C兴 z 02
Using the ABCD law, we can write the complex beam (A20)
parameter at distance Y to the left of the focal lens as
The authors acknowledge the experimental sup-
BD ⫹ ACz 02 共 AD ⫺ BC兲 z 02 port from Z. Yaqoob, M. A. Arain, and S. A. Khan.
q1 ⫽ 2 2 2 ⫹ i . (A12)
D ⫹ C z0 D 2 ⫹ C 2z 02 We also thank J. Wolter at the Technical University
in Eindhoven, The Netherlands, for suggesting that
At the beam waist the Gaussian beam has a planar M. Van Buren conduct a research project with N. A.
wave front, or in other words 1兾R ⫽ 0. Recalling the Riza’s laboratory at the Center for Research and Ed-
definition of the complex beam parameter, ucation in Optics and Lasers.
1 1 ␭ References
⫽ ⫺i , (A13)
q R ␲nw 2 1. R. März, Integrated Optics: Design and Modeling 共Artech
House, Norwood, Mass., 1995兲.
it can be seen that for the beam waist the complex 2. M. S. Borella, J. P. Jue, B. Ramamurthy, and B. Mukherjee,
beam parameter is purely imaginary. To determine “Components for WDM lightwave networks,” Proc. IEEE 85,
the location of the beam waist of the reflected Gauss- 1274 –1307 共1997兲.
ian beam, the value of Y for which the real part of q1 3. W. J. Tomlinson, “Application of GRIN-rod lenses in optical
is equal to zero has to be found. In terms of Y this fiber communication systems,” Appl. Opt. 19, 1127–1138
implies 共1980兲.
4. N. A. Riza and S. Yuan, “Low optical interchannel crosstalk,
共B 1 Y ⫹ B 0兲 D ⫹ 共 A 1 Y ⫹ A 0兲Cz 02 fast switching time, polarization independent 2 ⫻ 2 fiber optic
Re共q 1兲 ⫽ ⫽ 0, switch using ferroelectric liquid crystals,” Electron. Lett. 34,
D 2 ⫹ C 2z 02 1341–1342 共1998兲.
(A14) 5. N. Madamopoulus and N. A. Riza, “Directly modulated
semiconductor-laser fed photonic delay line with ferroelectric
in which liquid crystals,” Appl. Opt. 37, 1407–1416 共1998兲.

冉 冊
6. S. Yuan and N. A. Riza, “General formula for coupling-loss
2 L characterization of single-mode fiber collimators by use of
A1 ⫽ ⫺1 , (A15) gradient-index rod lenses,” Appl. Opt. 38, 3214 –3222 共1999兲.
f f
7. S. Yuan and N. A. Riza, “General formula for coupling-loss
2L characterization of single-mode fiber collimators by use of
A0 ⫽ 1 ⫺ , (A16) gradient-index rod lenses: errata,” Appl. Opt. 38, 6292
f 共1999兲.
8. A. Yariv, Optical Electronics in Modern Communications, 5th
2共X ⫹ L兲 2LX ed. 共Oxford U. Press, New York, 1997兲.
B1 ⫽ 1 ⫺ ⫹ 2 , (A17) 9. N. A. Riza and N. Madamopoulos, “High signal-to-noise ratio
f f
birefringence-compensated optical delay line based on a noise-
2LX reduction scheme,” Opt. Lett. 20, 2351–2353 共1995兲.
B 0 ⫽ X ⫹ 2L ⫺ . (A18) 10. N. Madamopoulos and N. A. Riza, “Reversible fiber-optic
f switched delay module using GRIN lens fiber-optic collimators
and ferroelectric liquid crystals,” in Proceedings of the IEEE-
The solution of this linear equation is denoted as Ybw
LEOS Annual Meeting 共Institute of Electrical and Electronic
and equals Engineers, New York, 1998兲, pp. 275–276.
11. N. Madamopoulos and N. A. Riza, “All-fiber connectorized com-
B 0 D ⫹ A 0 Cz 02
Y ⫽ Y bw ⫽ ⫺ . (A19) pact fiber optic delay-line modulus using three-dimensional
B 1 D ⫹ A 1 Cz 02 polarization optics,” Opt. Eng. 39, 2338 –2344 共2000兲.
12. W. T. Silfvast, Laser Fundamentals 共Cambridge U. Press,
So, theoretically at this distance at the left of the lens, Cambridge, UK, 1996兲.
the beam waist will be found. Actually one has to 13. J. C. Palais, “Fiber coupling using graded-index rod lenses,”
take this distance along the light path, which makes Appl. Opt. 19, 2011–2018 共1980兲.

564 APPLIED OPTICS 兾 Vol. 42, No. 3 兾 20 January 2003

14. F. L. Pedrotti and L. S. Pedrotti, Introduction to Optics, 2nd ed.
共Prentice-Hall, New York, 1996兲, p. 70.
15. H. A. Haus, Waves and Fields in Optoelectronics 共Prentice-
Hall, Englewood Cliffs, N.J., 1984兲, p. 140.
16. A. Yariv and P. Yeh, Optical Waves in Crystals: Propagation
and Control of Laser Radiation 共Wiley, New York, 1984兲.
17. D. Marcuse, “Loss analysis of single-mode fiber splices,” Bell
Syst. Tech. J. 56, 703–718 共1977兲.
18. H. Kogelnik, “Coupling and conversion coefficients for optical
modes,” in Proceedings of the Symposium on Quasi-Optics, J.
Fox, ed., Vol. 14 of Polytechnic Institute Microwave Research
Institute Symposia Series 共Polytechnic Brooklyn, Brooklyn,
N.Y., 1964兲, pp. 335–347.
19. S. Nemoto and T. Makimoto, “Analysis of splice loss in single-
mode fibers using a Gaussian field approximation,” Opt. Quan-
tum Electron. 11, 447– 457 共1979兲.
20. R. W. Gilsdorf and J. C. Palais, “Single-mode fiber coupling
efficiency with graded-index rod lenses,” Appl. Opt. 33, 3440 –
3445 共1994兲.
21. W. J. Smith, Modern Optical Engineering: The Design of Op-
tical Systems 共McGraw-Hill, New York, 1990兲.
22. LightPath Technologies, Product Information CD, Collimators
Catalogue, 2603 Challenger Tech Ct., Orlando, Fla. 32826, 2001.
23. M. J. Mughal and N. A. Riza, “Compact acousto-optic high
speed variable attenuator for high power applications,” IEEE
Photon. Technol. Lett. 14, 510 –512 共2002兲.
24. N. A. Riza, R. Akbar, S. Sumriddetchkajorn, F. Perez, and
M. J. Mughal, “47 dB dynamic range sub-microsecond switch-
ing speed variable fiber-optic attenuator for fast transient
fiber-optics,” in Photonics in Switching, Vol. 59 of 2001 OSA
Trends in Optics and Photonics Series Postconference Digest
共Optical Society of America, Washington, D.C., 2001兲, paper
PDP2.
25. M. J. Mughal and N. A. Riza, “65 dB dynamic range 2.8 mi-
croseconds switching speed variable fiber-optic attenuator,” in
the Twenty-Seventh European Conference on Optical Commu-
nication ECOC’01 共Institute of Electrical and Electronic Engi-
neers, New York, 2001兲, Vol. 6, pp. 56, 57.
20 January 2003 兾 Vol. 42, No. 3 兾 APPLIED OPTICS 565