reconstruction of the radial refractive index profile of ... · reconstruction of the radial...
TRANSCRIPT
Reconstruction of the radial refractive index pro�le ofoptical waveguides from lateral di�raction patterns
Seminar der AG Computerorientierte Theoretische Physik, 28. Juni 2017
Gunnar Clauÿen1
1Fachbereich IngenieurwissenschaftenJade Hochschule am Standort Wilhelmshaven
OUTLINE
1 Introduction and motivation
2 TheoryMie theory for cylindersInverse problems: IRGN algorithm
3 The research project
4 Results: Diameter of a homogeneous cylinder
5 Work in Progress: Strati�ed cylinders
6 Conclusion and outlook
28.06.2017 Reconstruction of the radial refractive index pro�le... 1
INTRODUCTION AND MOTIVATION (I)
Motivation
Fibres in communication technology (waveguides)
Distinct types: Gradient and step-index �bres
Desired: Measurement and control of parameters (refractive index pro�le)during production, i.e. the drawing process
Current knowledge
Methods exist for the determination of the outer diameter
These methods work well for intransparent cylinders
No algorithm is known for the refractive index pro�le n(d) or {dj , nj}
28.06.2017 Reconstruction of the radial refractive index pro�le... 2
INTRODUCTION AND MOTIVATION (II)
General approach
Illumination of a cylinder with plane-wave incidence, ζ = 90◦
Refraction on layer boundaries and di�raction on edges
Di�raction pattern is measured through an array of CCD cells
Evaluation of the di�raction pattern for parameter determination
Experimental setup
θ
k
r=d/2
yCCD
y
x
nm
λ
ns
Please note
Setup is the same for strati�edcylinders
Distance yCCD has to be knownaccurately
All e�ects add up to scattering
28.06.2017 Reconstruction of the radial refractive index pro�le... 3
INTRODUCTION AND MOTIVATION (III)
Theoretical approach: Inverse problem
Direct problem: Parameters known, result unknown
Inverse problem: Result known, parameters unknown
Illustration of an inverse problem in [Bohren & Hu�man 1983]
28.06.2017 Reconstruction of the radial refractive index pro�le... 4
MIE THEORY FOR CYLINDERS (I)
Light scattering: Basics of Mie theory
Propagation of EM waves is expressed through Maxwell equations
Especially: Vector wave equation ∇2 ~E + k2 ~E = 0 and continuity atboundaries, [ ~E2 − ~E1]× n̂ = 0
Mie theory: Transformation to speci�c coordinates corresponding to geometryallows expansion of scattered �elds into Bessel functions
Scattering on cylinders
Incident wave: ~Ei =∑∞n=−∞En ~N
(1)n
Scattered wave: ~Es =∑∞n=−∞En
[bnI ~M
(3)n + ianI ~N
(3)n
]For normal incidence (ζ = 90◦), only z-components of ‖-polarized
components are relevant: Es‖(z) = −∞∑
n=−∞Enbn‖Nn(z)
28.06.2017 Reconstruction of the radial refractive index pro�le... 5
MIE THEORY FOR CYLINDERS (II)
Scattering coe�cients
General calculation for homogeneous (r = 1) or strati�ed (r > 1) cylinders:
bn=mrJn(xr)[J′n(mrxr)+Tr−1
n N′n(mrxr)]−J′n(xr)[Jn(mrxr)+Tr−1n Nn(mrxr)]
mrHn(xr)[J′n(mrxr)+Tr−1n N′n(mrxr)]−H′n(xr)[Jn(mrxr)+T
r−1n Nn(mrxr)]
(1)
Tsn=
msJn(xs)[J′n(msxs)+Ts−1n N′n(msxs)]−J′n(xs)[Jn(msxs)+Ts−1
n Jn(msxs)]
msJn(xs)[J′n(msxs)+Ts−1N′n(msxs)]−N′n(xs)[Jn(msxs)+Ts−1n Jn(msxs)]
(2)
Recursive calculation: Layers s = 1...r, start with T0 = {0}Depends only on the relative refractive index ms =
ns
ns+1and the �size
parameter� xs = ks+1 · rs of the s-th layer
Outline of the calculation of the di�raction pattern
1 Calculate scattering coe�cients bn up to nmax = dx+ 4 · 3√x+ 2e
2 Transform CCD array to polar coordinates (x, y) 7→ (ρ, ϕ)
3 Calculate vector-harmonic generating functions Nn(z) for these coordinates
4 Calculate sum Es‖(z) = −∑nmax
−nmaxEnbn‖Nn(z).
28.06.2017 Reconstruction of the radial refractive index pro�le... 6
MIE THEORY FOR CYLINDERS (III)
...and it works!
Agreement for an intransparent d = 370 µm cylinder, λ = 632.8 nm
x-position [mm]-10 -8 -6 -4 -2 0 2 4 6 8 10
Inte
nsi
ty [
a.u.]
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
MeasurementCalculationDifference
28.06.2017 Reconstruction of the radial refractive index pro�le... 7
IRGN ALGORITHM (I)
Inverse problem (now formally...)
Generally, a operator F on a parameter set r creates (maps to) a far-�eld
pattern u∞, i.e. a simple operator equation:
F : r 7→ u∞ or F(r) = u∞ (3)
Direct problem: F and r are known, u∞ is to be determined
Inverse problem: F and u∞ are known, r is sought
Solution of the inverse scattering problem
Solution through the iteratively regularized Gauss-Newton (IRGN) algorithm:
||F ′[rn]hn + F(rn)− uδ∞||2 + αn||hn + rδn − r0||2!= min (4)
In each iteration n: Calculate alteration hn of the parameter set rn
Iteration terminates after N steps, if ||F(rN )− uδ∞||2 ≤ τδ
28.06.2017 Reconstruction of the radial refractive index pro�le... 8
IRGN ALGORITHM (II)
Regularisation
Special treatment for noisy data, weighting of far-�eld pattern
The step update hn of parameters rn is calculated according to
hn = −F′[rδn]
∗(F (rδn)− uδ∞) + αn(rδn − r0)
F ′[rδn]∗F ′[rδn] + αnI
(5)
with regularisation parameter αn = αn0 , α0 = 1/2
Notes
Derivative F ′[rn] to the parameters rn has to be known
For homogeneous cylinders: Analytcally...
For strati�ed cylinders: Numerically...
Analytical calculation does not pay o� via a considerable time advantage!
28.06.2017 Reconstruction of the radial refractive index pro�le... 9
THE RESEARCH PROJECT (I)
General properties
Research project at Jade Hochschule
Applied and carried out by Prof. Dr. Werner Blohm since 2014
Aim: Improvement of the precision of di�raction-optics methods
Doctoral research is embedded in the project through Jade2Pro
Several related �milestones� have been proposed for both the applicant (Prof.Blohm) and the doctoral researcher (me)
�Applicant� side ended in 2016, but research will be supported further
28.06.2017 Reconstruction of the radial refractive index pro�le... 10
THE RESEARCH PROJECT (II)
Structogram of the Project
28.06.2017 Reconstruction of the radial refractive index pro�le... 11
THE RESEARCH PROJECT (III)
Formal development of the doctorate
Cooperation with Zentrum für Technomathematik (ZeTeM), Fachbereich 3,Bremen University
Supervision by Prof. Dr. Armin Lechleiter, head of AG Inverse Probleme
Guest in Bremen in WS 2015/16, seminar talk
Beginning of WS 2016/17: Ph.D. student at Bremen
In WS 2016/17: Recognition of the research topic by FB3
Complete Title: �Bestimmung des radialen Brechungsindexpro�ls optischer
Lichtwellenleiter aus lateralen Streulichtverteilungen�
In WS 2016/17: Project successfully evaluated by Jade2Pro
Project-related �publicity� so far
Talks at 2nd, 3rd und 5th Jade2Pro-Kolloquium, Oldenburg
Poster presentation at the DPG Spring Meeting 2017 of the Atomic,Molecular, Plasma Physics and Quantum Optics Section (SAMOP), Mainz
28.06.2017 Reconstruction of the radial refractive index pro�le... 12
RESULTS: CYLINDER (I)
The actual inverse scattering problem
Operator F : Mie theory for cylinder scattering
Parameters r: Only diameter d, everything else is �xed (ns, nm, yCCD, λ)
Far-�eld u∞: Pattern on CCD array at distance yCCD
Fréchet derivative F ′ for d is analytically known
(d-dwahr
)/λ-10 -8 -6 -4 -2 0 2 4 6 8 10
||F(
d)-
F(d
wah
r)||
2
0
0.5
1
1.5
2
2.5
λ = 523 nmλ = 633 nmλ = 780 nmλ = 1046 nm
Encountered �problem�
Over d, the patterns resemble each other
An IRGN algoritum iteratively minimizes theresidual ||F(d)− u∞||2The residual, when compared to a referencedreal, exhibits local minima...
Ill-posed problem (i.e. non-bijective)
Danger of false convergence
28.06.2017 Reconstruction of the radial refractive index pro�le... 13
RESULTS: CYLINDER (II)
�Solution�: Modi�cation of the IRGN algorithm
But: Minima are spaced periodically by δmin,s ≈ 2.18λ
This property may be utilized...⇒ Detect local convergence and then �jump� by ±2.18λ
Example for such a way of the algorithm
-18 -16 -14 -12 -10 -8 -6 -4 -2 0 20
0.2
0.4
0.6
0.8
1
1.2
local search
jumps and local searches afterwards
global minimum
resi
dual
||f
(d)-
f(d
real)|
| 2
distance (d-dreal)/λ
28.06.2017 Reconstruction of the radial refractive index pro�le... 14
RESULTS: CYLINDER (III)
Where does the periodicity δmin,s ≈ 2.18 stem from?
In the present case, ns = 1.458 ≡ ms =ns
nm...
Research: Compare di�raction patterns for a range of d with a reference dreffor several values of ms, measure minima spacing...
Dependence of minima spacing
0.0
2.0
4.0
6.0
8.0
10.0
1.0 1.5 2.0 2.5 3.0
Abst
and
der M
inim
a de
s Res
iduu
ms
||F(
d)-F
(dw
ahr)||
2
Brechungsindex ns
Daten
Fitfunktion f(ns) = a+b/(ns-1)c
Funktion 1/(ns-1)
-20.0
0.0
20.0
40.0
60.0
80.0
100.0
1.00 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08 1.09 1.10
Abst
and
der M
inim
a de
s Res
iduu
ms
||F(
d)-F
(dw
ahr)||
2
Relativer Brechungsindex ms
Daten: 2 Schichten, Variation von Schicht 1
Fitfunktion f(ns) = 2/3*(ms-1)-1 Result
δmin,s =1
ms − 1
...thus:1
1.458−1 = 2.183...
28.06.2017 Reconstruction of the radial refractive index pro�le... 15
RESULTS: CYLINDER (IV)
Ideas for further improvement
Improvement of initial values: Sample for a few values of d around d0, choosevalue of least residual
Approximation through intransparent cylinders:
Subtraction of the 3rd sinusoid component from the di�raction pattern
approximates the pattern for a intransparent cylinder of the same dFor intransparency, only a single global minimum exists...
GN iterations allow an approximation of d0 to dreal
Exact determination of dreal is however not possible this way
�Plausibility check�: Are minima also present at dN for further observingangles θ? Otherwise, it's certainly no reliable result...
28.06.2017 Reconstruction of the radial refractive index pro�le... 16
RESULTS: CYLINDER (V)
Precision of the algorithm: Numerical evaluation
n = 1.4, N = 2910 random dreal = 110...150 µm and d0 ∈ dwahr ± 25λ
The actual diameter dreal is met precisely (i.e. |dN − dreal| < 0.1 µm,de�nition of a �success�) for 99.62% of the cases
The calculation time exhibits no visible dependence on the distance betweeninitial and actual values of d
28.06.2017 Reconstruction of the radial refractive index pro�le... 17
RESULTS: CYLINDER (VI)
Precision in presence of noise
Noise level [%]0 1 2 3 4 5 6 7 8 9 10
Succ
ess
rate
[%
]
20
30
40
50
60
70
80
90
100
Med
ian d
evia
tion
|d-d
real|
[nm
]
0
1.25
2.5
3.75
5
6.25
7.5
8.75
10
Noise: Multiplicative, normally distributed
�Noise level�: Scaling factor for the width(variance) of the distribution
Deviation of the result calculated only for�successful� cases
Note: �Failures� still �nish within localminima
Summary
Method for diameter determination is very precise
Precision drops linearly with noise level
Improvements lead to constant calculation times
28.06.2017 Reconstruction of the radial refractive index pro�le... 18
STRATIFIED CYLINDER (I)
Basics
Aim: Extension of the IRGN algorithm for strati�ed cylinders
Calculation of di�raction patterns: Only di�erent scattering coe�cients
Experience from homogeneous cylinders: Periodic local minima
Expected di�culties
Parameters to be found: Layer diameters dj
More complex operation, more parameters must be optimised
Especially: How can the �jumps� of the modi�ed IRGN algorithm beimplemented?
28.06.2017 Reconstruction of the radial refractive index pro�le... 19
STRATIFIED CYLINDER (II)
Example: J = 2 layers, dref = [50, 120] µm, n = [1.55, 1.5]
Periodic behaviour and a global minimum at d ≡ dref exist
ResidualsDiagonal ADiagonal B
-20-15
-10-5
0 5
10 15
20
Distance (d1-d1,ref)/λ
-2-1.5
-1-0.5
0 0.5
1 1.5
2
Distance (d2-d2,ref)/λ
0 0.2 0.4 0.6 0.8
1
28.06.2017 Reconstruction of the radial refractive index pro�le... 20
STRATIFIED CYLINDER (III)
Further �ndings
Periodicity of minima for layer j < J : δmin,s =2
3
1
ms − 1
Remark: Refractive indices di�er only marginally between layers,|nj − nj+1| = 0.001...0.02⇒ nj+1 = 1.5 : mj ≈ 1.0006...1.013
Thus: For inner layers �large� periodicity can be expected,δmin,s ≈ 50....1000λ
For the outermost layer j = J the jump from nJ to nm is much larger ⇒ cf.homogeneous cylinder
28.06.2017 Reconstruction of the radial refractive index pro�le... 21
STRATIFIED CYLINDER (IV)
Even more further �ndings
The residuals along the diagonals in parameter space (d1, d2):
0
0.2
0.4
0.6
0.8
1
1.2
1.4
-4 -3 -2 -1 0 1 2 3 4 5
Res
idua
l ||f
(d)-
f(d ref
)|| 2
Factor (d-dref)/(lambda*δj,min)
Diagonal ADiagonal B
28.06.2017 Reconstruction of the radial refractive index pro�le... 22
STRATIFIED CYLINDER (V)
Ideas for layer diameter determination
IRGN steps lead into a �trench�
The �trench� is linear and described through the periodicities δj,min
Jumps should lead from one �trench� to another (diagonal A)
Motion within one �trench�: local Minima (diagonal B)?
Summary: A combination of IRGN, jumping and sampling methods seemspromising for determing both diameters d1 and d2
...and how can that be applied to J > 2?
28.06.2017 Reconstruction of the radial refractive index pro�le... 23
FURTHER OUTLOOK
Aims to be reached
Milestone: Algorithm for strati�ed cylinders is to be completed
Experimental veri�cation for homogeneous cylinder (experiment exists)
Optionally: Generalized Lorenz-Mie theory for spherical waves
Temporal constraints
Current position is running until April 2018
Probably extended by one year from remaining funds
Current di�culty: A. Lechleiter out of o�ce until further notice
28.06.2017 Reconstruction of the radial refractive index pro�le... 24
REFERENCES
References
Bohren, C.F.; Hu�man, D.R.: �Absorption and Scattering of Light by Small Particles�,
John Wiley & Sons, New York, 1983
Gerlinski, V.: �Bestimmung geometrischer Parameter von Glasfasern aus
Streulichtmustern�, Master thesis, Universität Bremen, 2016
Quinten, M.: �Optical Properties of Nanoparticle Systems. Mie and Beyond�, Viley-VCH,
Weinheim, 2011
Hohage, T.; Schormann, C.: �A Newton-type method for a transmission problem in
inverse scattering�, Inverse Problems 14, 1207�1227, 1998
Schäfer, J.-P.: �Implementierung und Anwendung analytischer und numerischer Verfahren
zur Losung der Maxwellgleichungen für die Untersuchung der Lichtausbreitung in
biologischem Gewebe�, Ph.D. thesis, Ulm, 2011
28.06.2017 Reconstruction of the radial refractive index pro�le... 25