Sei sulla pagina 1di 5

Note that these weight coefficients tend to those of the regular trapezoidal rule as (x6t) -0.

The fact that only 5(t) has to vary linearly over 5t means that we can take the meash size to be relatively coarse in many cases of practical interest. APPENDIX B The moments about the origin are m =(x')=e'cel202/2, 1=1, 2,..,
(B1])

Neither yj or yz depends on C. Furthermore, we have 71 > 0, which indicates a long tail to the right of the maximum of the lognormal PDF.
1
2

A. Papoulis, Probability, Random Variables and Stochastic

Processes (McGraw-Hill, New York, 1965).


H. Bateman, Tables of Integral Transforms (McGraw-Hill,

New York, 1954), Vol. 1, pp. 15 and 74.


H. Cramer, MathematicalMethods of Statistics (Princeton U.

and increase very rapidly with 1. The corresponding 10 central moments are ,ug=((ok(A2)

P., Princeton, N. J., 1946), Sec. 17.4.


4Handbook of Mathematical Functions, edited by H. Abramowitz

MX)l
/ ) (I)(-J) (I-J-1/2 (

2)

where c-expcT 2 and 2=w(w-(1)e2B The coefficient of skewness yi and the coefficient of excess 72 are
'Y -A, /4-2,
t/( )

3)

(o=(e + 2),

(E 4)

and I. A. Stegun, Nat. Bur. Std., U.S. Appl. Math. Ser. (U. S. Government Printing Office, Washington, D. C., 1964; Dover, New York, 1965). 'R. L. Mitchell, J. Opt. Soc. Am. 58, 1267 (1968). 6 H. S. Wall, Analytic Theory of Continued Fractions (Chelsea, New York, 1967), Sec. 87. 7 J. A. Shohat and J. D. Tamarkin, The Problem of Moments (American Math. Soc., Providence, R. I., 1943). 8C. C. Heyde, J. Roy. Statist. Soc. 26, 392 (1963). 9 J. Strohbehn, in Progress in Optics, edited by E. Wolf (NorthHolland, Amsterdam, 1971), Vol. 9. This critical review contains references to many of the pertinent papers. 'OR. Wartmann, Mitteil. Math. Stat. 8, 83 (1956).

Curvature loss formula for optical fibers


Dietrich Marcuse
Bell Laboratories, Crawford Hill Laboratory, Holmdel, New Jersey 07733

(Received 23 July 1975) The loss formula for optical fibers with constant radius of curvature of their axes is derived by expressing the field outside of the fiber in terms of a superposition of cylindrical outgoing waves. The expansion coefficients are determined by matching the superposition field to the field of the fiber along a cylindrical surface that is tangential to the outer perimeter of the curved fiber. This method is a direct extension of my derivation of the curvature-loss formula for slab guides. INTRODUCTION It is the purpose of this paper to test a novel method for computing curvature losses by applying it to the bent optical fiber. The same method can be used to compute radiation losses of more general curved fiber geometries. It is well known that dielectric waveguides lose power by radiation if their axes are curved. For dielectric slab guides it is possible to derive this curvature loss by solving the eigenvalue equation of the problem. 1.2 However, this direct approach is not necessarily the simplest from a mathematical point of view and it is practically inapplicable for dielectric waveguides of finite cross section such as an optical fiber. Other approximate solutions of the curvature-loss problem have thus been devised. Lewin 3 solved the problem by constructing an appropriate solution for the electromagnetic field. Arnaud 4 couples the field of the fiber to the whispering gallery mode of a curved metal surface proximation in the Kirchhoff-Huygens diffraction integral in its vector form. White 6 has successfully derived the curvature-loss formula of the slab as well as the fiber by using this approach. Shevchenko 7 derived the loss formula by adapting the known radiation mechanism for the slab to the case of a fiber. Finally, Chang and Kuester8 utilize an approach that resembles quantum mechanical perturbation theory. I have derived the curvature-loss formula for slab waveguides by using the following approach. 9 The curved slab guide is best described in a cylindrical polar coordinate system, r, 0, z. The field solution outside of the slab in the infinite cladding must be a simple Hankel function H(2) (nkr) (n is the refractive index of the medium outside of the core and k is the free-space propagation constant). Only the order number gu and the field amplitude remain undetermined. If we now utilize the fact that the field of the curved guide near the core must resemble closely the field of the straight waveguide, we may immediately identify the order number with 1 = fl,R,
Copyright 1976 by the Optical Society of America

which, in the end, is allowed to move to infinity.


Snyder 5 has proposed solving the curvature-loss problem by using the field of the straight fiber as a first ap216 J. Opt. Soc. Am., Vol. 66, No. 3, March 1976

(1)
216

I
I

CYLINDRICAL vTANGENT SURFACE

(This approximation is used only for convenience, it is in no way essential for the general method.) The E, and E 0 components of the electric field of the guided mode for r' >a are given by the equations' 2

E =A [Jv(Ka)/IHI(l(iya)]HV' )(iyr') cosvo efilRl


and E 0 =AY Is*
R 0p I

(3)

J Hn 2 iv

[H' )(iyr') sin(v +1)0 +1 )e(iya)


+ H2, (iyr')sin(v
-

1) 01 e -OgR

(4)

with k = 27r/X = wVE 0


an= (d 22 2)1/2

0,

(5)

FIG. 1. The fiber is bent into a torus with radius of curvature R. The dotted vertical lines indicate a cylinder that is tangential to the torus at radius r=R+a.

and

(6) The amplitude is related to the power P carried by the mode,

K = M lk2 - 2)1/2

where 0, is the propagation constant of the guided mode in the straight guide and R is the radius of curvature of the bent slab. The unknown amplitude coefficient of the field is obtained by comparing the general field solution with the field of the straight slab near the waveguide core. The radiation loss caused by the curved slab is then obtained by calculating the amount of power outflow AP per unit length of the waveguide and by substitution into the loss formula 2coa = AP/P The factor 2 is necessary since ae is defined as the amplitude-loss coefficient of the guided wave; P is the power carried in the straight guide. We show in this paper that the approximate method for computing the curvature loss of the slab may be extended to waveguides of arbitrary shape using the round optical fiber as an example. We use again cylindrical polar coordinates for the field outside of the bent dielectric waveguide and express the most general solution of Maxwell's equations as a superposition of outgoing cylindrical waves. The superposition of waves, instead of just one simple cylindrical wave used in case of the slab, is necessary because the field must in general vary in the direction of the z axis-the direction perpendicular to the plane defined by the bent guide axis. The order number of the Hankel function used in the expansion is again given by Eq. (1) and the unknown expansion coefficients are determined by matching two of the four transverse field components of the field expansion to the field of the guided mode along a cylindrical surface that is tangential to the curved guide. The curvature loss is once more calculated by using Eq. (2), DERIVATION OF THE CURVATURE-LOSS FORMULA The geometry of our problem is sketched in Fig. 1. We consider a bent fiber of circular cross section. However, the method described here in principle also works for fibers of arbitrary cross sections. The formulas are simplified by using the weak guidance approximation9 " 1 1 for the field of the guided mode. ,1 217 J. Opt. Soc. Am., Vol. 66, No. 3, March 1976 (2)

A
with

4(A01E)l /2 2p 4 \1/2 \e"vn V' IJv..(Ka) Jv1 (Ka) 1

(7a)

e 2, 1, and

v=o p0 (7b)

(8) Hi1 ) is the Hankel function of the first kind and order v, J, is the Bessel function, n1 is the refractive index of the fiber core, and n2 the index of the (infinitely extended) cladding. The only modification from the field expressions for the modes of the straight fiber consists in replacing Agz' in the exponent of the propagation factor with ,gR0. Outside the fiber the field must also be expressible as a superposition of fields of the following form:
2 ij

= k a (nl2

-n2)

= BjH(2) (pr) e-iu

e-. aS

(9) (10)

ac,,j

=BjFjH(-) (pr)e-"" e-','Z

The other field components can be obtained from the z components of the electric and magnetic fields by differentiation. 13 The time factor exp(iwt) is omitted from all field expressions, but this time dependence ensures that the Hankel function of the second kind, H(2 ), describes outgoing waves at infinite distance r. The parameter p is defined as
p= (n k
2 2

- P2)1/2

(11)

The amplitude coefficient Fj can assume two different values belonging to two sets of mutually orthogonal modes. Of the infinitely many ways of choosing F, and F2 , we use
F, n2 (IL/M
0 )"

{H )[p(R + a)]/H(5)'[p(R + a)]}

(12)

and
F2 =n 2 (E /MUO)1/2 {H(,)'[p(R +a) ]/H(2) [p(R+a)]}. Dietrich Marcuse -

(13)
217

The prime indicates differentiation of the Hankel function with respect to its entire argument. The magnetic permeability of vacuum g,, should not be confused with the order Aiof the Hankel functions. The only requirement on the choice of values for F, and F2 is the need for making the two sets of modes with j = 1 and j = 2 mutually orthogonal. Our choice was based on the desire to let the expressions for Fj appear simple and symmetrical. The amplitude coefficient Bj is related to the following normalization of the modes (the asterisk indicates complex conjugation): dp
00

By following an argument analogous to the approximation of the Hankel function explained in Ref. 9, pp. 402403, we obtain H (2)[ p(R + a)] i( 2r ) 1 e
'R) exp(2'
2

Xexp(Q

1)

(23)

and
H (2)1 = _ (7/n2k)H (2) .

(24)

We need the amount of power outflow from the fiber at r - o in order to calculate the curvature loss from Eq. (2): 2a( = p r
X

= (2 7)2 6 .L, 6 J, 6(j3 -

1')NA3

(14)

C|2(,0Z adz}

3Zf

2*>)

dg. (25)

We obtain N

= in 2WE

r(i

)*
2

H(2)\* I 2(

(15) (16)

Using the approximation of the Hankel functions for large arguments and substitution of the relevant terms and integration over 1 leads to
2a = Y /2 Io 12 e2ya exp[23

13/)R ]J

(Ka)

72V2 I J,1(Ka)J,,.(Kca)I I H (1)(iya)I1-r ear


N0a2=

(26)

[(Vf;

- 1] H()H )I*IB2I2

The arguments of the Hankel functions are p(R +a) and the primes on these functions indicate again differentiation with respect to the whole argument. We now require that a superposition of the cylindrical waves be equal to the field of the guided mode of the fiber at r=R+ a,
.C1(1) Ez= i

The integral over (26) has the form

that is implicitly contained in Eq.

2 exp[- (yR/n2k ) 12] dO= (ff/yR)/2 n 2 k.

(27)

For large values of the radius of curvature R only the region very near 13=0 contributes substantially to Eq. (25), justifying our assumption that 1may be replaced by 13=0 in If. We also used the approximation ogln 2k that is valid for weakly guiding fiber. We replace the Hankel functions of imaginary argument with the modified Hankel function using
2 H"l)(iya) = (2/iT) e-iv(7/ )Kv(ya). 5

sz,1+ C2 (1)&J

2 ]do,

(17) (18)

(28)

E. =

jf

[c,(13)'S1,+ c2 (0)8 0 , 2 ] d:.

A general field expansion would require summation over the index ji, but since the two fields must have the same (P dependence we use the relation (1) to determine M and refrain from summing over this index. Atr=R+a, we have r't=(a2 + 421/2 and (19)

(29)

The eigenvalue equation' allows us to replace the Bessel functions in Eq. (26) with Hankel functions via the relation
J~v

K2

I JIJ.,I

IH (1) I 2 =

H ( I H ~'11 |7 y2 K K2

22

e = arctan(z/a).

(20)

4
-4

The expansion coefficients c] and c2 can now be determined with the help of the orthogonality relations (14): 1
-iABj [E3C* j -EJCj*0]r=R+adz J w(nca)0 H < fE
)*[ p(R + a)]

y1Kv,. (ya)Kv,1('Yaj'

(30)

A discussion of the integral (22) for 1=0 is given in the Appendix, its value is I = e-^(T/2)(2/iy) e-ra (31) The curvature loss formula may thus be written in the form
2(71/02)R] 2[_ 2=e,7 3/2 V1NfRK,,1(-ya)Kv,,1('ya)

(21)

2N H,(')(iya)' 0 with

(32)

*V= f

2 2 H ')[iy(a 2 + z )1 ]cos[varctan(z/a)] e"'dz.

(22)

It shall become apparent that only the region around

/3=0 contributes appreciably to power radiation at re-m.

The parameters e, and V are defined by Eqs. (7) and (8), respectively. As explained in the next section, e, =2 holds for HE and EH modes. Two special cases are of interest. For well guided modes we have ya>>l. In this case Eq. (32) may be apDietrich Marcuse
218

For this reason we have neglected in Eq. (21) terms proportional to 13.
218 J. Opt. Soc. Am., Vol. 66, No. 3, March 1976

proximated as 2 2aK2 e2ra exp[23(y 3/2)R]

In this case we may also approximate the parameter 6 as the root of the Bessel function1
J,(Ka) = 0 .

Ka

(34)

The other special case of interest is the limit ya << 1 which permits the following approximation of (32) for V = 0: 2(x =_ 2 1 2 9P2~ ) 2v"' V VIR(lnya)2 and, for v Q 0,
/K 2

(35)

(ya) 2 3/ 2a3/2 exp[- 3(3/i2)R] 22(v-')('v- 1)!(v+1)! V'

(36

The parameters K, y, and 3g are obtained from Eqs. (5) and (6), and from the eigenvalue equation of the guided modes," DISCUSSION We have derived the curvature-loss formula [Eq. (32)] for the bent fiber by determining the expansion coefficients of a field expansion in terms of cylindrical waves. The expansion coefficients were found by matching the field expansion to the guided mode field of the straight fiber. The curvature of the fiber was taken into account be replacing the propagation factor exp(- iiogz') by exp(- igR0) and expressing the field of the straight guide in a coordinate system that locally follows the curved waveguide axis. The curvature loss formula of the fiber is, of course, only an approximation. The most serious limitation to its validity is caused by using the undistorted field of the straight guide for its derivation. Even if the radiation loss is disregarded, the field changes its shape in the curved guide. The field is forced toward the outer wall in a manner resembling a centrifugal force effect. Only for very large values of R is it permissible to neglect this effect. For sharply curved guides the field distortion caused by the bend has a considerable influence on the curvature loss. Our formula (32) is in essential agreement with the results of all the other authors. However, we must remember that its derivation was based on expressing the guided mode fields by the approximate modes of the weakly guiding fiber (Gloge's LP modes). In particular we have used the cosine function for the azimuthal dependence of these modes. Had the sine instead of the cosine function been used we would have found that the curvature loss vanishes in this approximation. This feature is peculiar to the weakly guiding mode approximation. The LP modes (3) and (4) are not exact modes of the structure but may be obtained by superimposing an HE,., and an EH,_1 mode of the fiber. 9'1 The superposition of these two modes may be phased so that either the mode expressed by (3) and (4) or the corresponding mode with the sine function dependence results. In the first case we find that the radiation loss 219 J. Opt. Soc. Am., Vol. 66, No. 3, March 1976

of the superposition of the two modes is twice as high as the curvature loss of each mode alone. For this reason Eq. (32) is twice as large as the corresponding formula for the HE or EH modes of the fiber. Only for v= 0 does the simplified mode of the weakly guiding fiber coincide with the true HE,, mode so that Eq. (32) gives the correct loss value. If the superposition of an HE,,. and an EH,_1 mode is phased so that the weakly guiding mode with a sinvzE dependence results, the radiation contributions of the two modes cancel so that no radiation loss occurs. Formally we obtain the correct loss formula for HE or EH modes as the arithmetic mean of the formulas for the approximate modes. Since the loss of the mode with the sinvO dependence vanishes, this arithmetic mean is just half the value given by (32). Chang and Kuester 8 point out that the curvature losses of HE and EH modes are independent of polarization and mode orientation, at least in the case of weakly guiding fibers. ACKNOWLEDGMENT I am very grateful to Mrs. W. L. Mammel, who invented the ingenious proof for the vanishing of the sum over binomial coefficients appearing in the expression for the integral (22). I am indebted to I. White and A. W. Snyder of the Australian National University and to J. A. Arnaud for illuminating discussions on this subject. APPENDIX We sketch here the evaluation of the integral (22) for 1:=0. It is Io= af HV') [iya( 1+ X 2 )1/2] cos(v arctanx) dx

2a e ">/f

K,,[ya(l+x')'"2 ] (el+x

dx.

The step from the first to the second line was accomplished by writing the cosine function as the sum of two exponentials and by combining the two exponentials into one term by means of a change of variables from x to - x. Next we expand the term (1 + ix)> with the help of the binomial theorem and use Eq. 6. 596-3, p. 705, of Ref. 17:
I-4a -~
4=-i e kO -

[-f]
2k 2k

ki~f2)E2-(1 + /2)r(k
(ay)>i'l/2) K(-k-(112)(aY)

For integers v and k the modified Hankel function can be represented as' 8
Kv-k- ( / 2) (ay) ='Tr

e-av E

(v - k+ M - 1)1

With the help of the identity' 9 r~k+ =2i~r(2k - 1)! ) (2) (k- 1) 22k and a change of summation indices we can write our expression in the form
Dietrich Marcuse 219

it 0=

W/) h

a e-

E
xn-1
fk)

VIi
)n

)(+z--k!

W)(v2772(n-

21 -2k)!

with rVI [2-j f (n) = n

if

n> ri

if n <rv

[-] indicates the integer that is equal to or smaller than


v/2. The above expression solves our problem because the sum over k vanishes. The following proof for the vanishing of the inner sum over 1 has been invented by Mammel. Consider the expression

"= k(.2

)knl

,wn-1-2k

k=lk1(n -k)!I 0 Next we take the n - 1 derivative of this expression:

dnv1
dxnX

x-2=

f E(n) (-1)

(v+n-l-2k)l

kl(n-k)I(v-2k)1

For x = 1 the right-hand side of this expressionbecomes identical to n1 times the inner sum of 4,o. The left-hand side of the n - 1 derivative vanishes for x =1, proving that the sum does not contribute to Iw. We thus have finally,
1,0 =

(2/iy) e-"'/2

eta

'M. A. Miller and V. I. Talanov, "Electromagnetic Surface Waves Guided by a Boundary with Small Curvature," Zh.

Tekh. Fiz. 26, 2755 (1956). E. A. J. Marcatili, "Bends in Optical Dielectric Guides," Bell Syst. Tech. J. 48, 2103-2132 (1969). 3 L. Lewin, "Radiation from Curved Dielectric Slabs and Fibers," IEEE Trans. Microwave Theory Tech. MTT-22, 718-727 (1974). 4J. A. Arnaud, "Transverse Coupling in Fiber Optics Part III: Bending Losses," Bell Syst. Tech. J. 53, 1379-1394 (1974). 5 A. W. Snyder (private communication). 6 A. W. Snyder, 1. White, and D. J. Mitchell, "Radiation from Bent Optical Waveguides," Electron. Lett. 11, 332-333 (1975). 7 V. V. Shevehenko, "Radiation Losses in Bent Waveguides for Surface Waves," Radiophys. Quantum Electron. 14, 607614 (1973) (Russian original 1971). 8 D. C. Chang and E. F. Kuester, "General Theory of SurfaceWave Propagation on a Curved Optical Waveguide of Arbitrary Cross Section," Scientific Report No. 11, Electromagnetics Laboratory, Dept. Electr. Eng., Univ. of Colo., Boulder, Colo.; also, IEEE J. Quantum Electron. QE-11, 903-907 (1975). 9 D. Marcuse, Light Transmission Optics (Van Nostrand, Princeton, 1972), 398-406. 'OD. Gloge, "Weakly Guiding Fibers," Appl. Opt. 10, 22522258 (1971). " 1D. Marcuse, Theory of Dielectric Optical Waveguides, (Academic, New York, 1974). 2 1 Reference 11, Eq. (2.2-23), p. 65, and Eq. (2.2-25), p. 66. t3 Reference 9, Eqs. (8.2-7) through (8.2-10), p. 290. 14 M. Abramovitz and I. A. Stegun, Handbook of Mathematical Functions, Eq. 9. 2.4, p. 364. U. S. Department of Commerce, National Bureau of Standards, Appl. Math. Ser., 55. t 5Reference 11, Eq. (2.2-38), p. 68. tt Reference 11, Eq. (2.2-69), p. 73. t7I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and Products (Academic, New York, 1965). t 8Reference 17, Eq. 8.468, p. 967. 19Reference 14, Eq. 6.1. 18, p. 256.
2

Bragg diffraction of Gaussian beams by periodically modulated media*


Ruey-Shi Chu
Communication Systems Division, GTE Sylvania Inc., Needham Heights, Massachusetts 02194

Theodor Tamir
Department of Electrical Engineering and Electrophysics, Polytechnic Institute of New York, Brooklyn, New York 11201 (Received 3 October 1975) Analytical and numerical results are given on the diffraction of optical beams by a periodically modulated dielectric medium, which represents a thick hologram or an acoustic column. By using a rigorous representation for the field of a realistically bounded beam incident at a Bragg angle, we examine both the refracted beam and the beam due to Bragg scattering inside the periodic medium. The Bragg-scattered beam is

formed by continuous coupling of energy from the refracted beam into the Bragg-scattered wave. As the Bragg-scattered wave also couples part of its energy back to the refracted wave, the continuous coupling of energy back and forth between these two waves results in a diffusion of energy over a wide region. For a sufficiently thick modulated layer, both the refracted and the Bragg-scattered beams split into two beams. Because of this distortion of the beam profiles, the diffraction efficiency is found to be smaller than that accounted for by previous approaches using a single incident plane-wave model. The scattering of optical waves by a periodically modulated layer has been studied extensively in the context of light diffraction by sound'- 5 and in image reconstrucistic beams having bounded cross sections. By considering the scattering of a Gaussian beam incident on a periodic slab of arbitrary width, the present work shows that the diffraction process exhibits features that are not evident from the simpler (single plane-wave) analysis. In particular, the efficiency of diffraction into the
Copyright 1976 by the Optical Society of America 220

tion by holograms.

4-

llowovor, most theoretical in-

vestigations have dealt with the fields due to a single incident plane wave of infinite extent rather than with real220 J. Opt. Soc. Am., Vol. 66, No. 3, March 1976

Potrebbero piacerti anche