Development of a

Spatially Incoherent Laser Source

Entwicklung einer
räumlich inkohärenten Laserquelle

Wissenschaftliche Arbeit zur Erlangung des Grades

M. Sc. im Studiengang Physik der kondensierten Materie
an der Fakultät für Physik der Technischen Universität München.

Betreut von Prof. Dr. Michael Knap

Prof. Dr. Immanuel Bloch
Dr. Christian Groß

Eingereicht von David Wei

Eingereicht am 13. Februar 2019

Abstract

Projected optical dipole potentials for ultracold atoms suffer from coherent speckle noise,
significantly degrading the potential shaping quality. In this thesis, the generation of
incoherent light and its use to suppress coherent artefacts in the light field is demon-
strated. Passing a coherent laser beam through a rotating optical diffuser or modulating
the angular beam incidence onto the diffuser using an acousto-optic deflector are both
shown to produce quasi-monochromatic light fields with controllably reduced spatial co-
herence. An alternative approach involves conversion of temporal to spatial incoherence,
where modal dispersion in a square-core multimode optical step-index fibre efficiently
induces dephasing. This results in temporally fast decorrelating, spatially incoherent,
flat-top light complying with the requirements of off-resonant dipole traps with high trap
frequencies. To characterize the light source, a lateral shifting Michelson interferometer
has been constructed in order to measure the spatiotemporal coherence function. Fi-
nally, femtosecond lasers have been identified as a suitable spectrally broad light source
with low temporal intensity noise.
Contents

Abstract i

1 Introduction 1

2 Theory of Coherence and Dipole Traps 5

2.1 Coherence Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Scalar Diffraction Theory . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Correlation Functions . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.3 Propagation of Correlations . . . . . . . . . . . . . . . . . . . . . . 12
2.1.4 Reduction of Coherent Noise . . . . . . . . . . . . . . . . . . . . . 15
2.2 Optical Dipole Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Atomic Light Shifts . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.2 Heating of Trapped Atomic Systems . . . . . . . . . . . . . . . . . 18
2.2.3 Influence of Spatial Noise . . . . . . . . . . . . . . . . . . . . . . . 20

3 Diffuser-Controlled Decoherence 23
3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.1 Speckle Noise Statistics . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.2 Static Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.3 Dynamic Correlations . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.1 Rotating Diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.2 Beam Deflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.3 Image Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Static Speckles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Modulation by Diffuser Rotation . . . . . . . . . . . . . . . . . . . . . . . 38
3.5 Modulation by Angular Incidence . . . . . . . . . . . . . . . . . . . . . . . 41
iv Contents

4 Fibre-Controlled Decoherence 47
4.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.1 Step-Index Fibres . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.1.2 Broadband Light Sources . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.1 Implementation of Fibre Approach . . . . . . . . . . . . . . . . . . 60
4.2.2 Lateral Shift Interferometer . . . . . . . . . . . . . . . . . . . . . . 62
4.3 Temporal Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.4 Spatial Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4.1 Coherence Function . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4.2 Residual Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.3 Fibre Launch Conditions . . . . . . . . . . . . . . . . . . . . . . . 74
4.5 Intensity Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

5 Conclusion 83

Bibliography 85

Abbreviations 93

List of Figures 97

Acknowledgements 99
1 Introduction

Strongly interacting quantum many-body systems arise in a variety of fields, including

condensed-matter physics, quantum chemistry and high-energy physics [1]. Their under-
standing is hindered by the complexity that grows exponentially with system size, ren-
dering their full description classically intractable. Quantum simulators pose a solution
by mapping the original many-body problem onto an equivalent problem implemented
in another well-controlled quantum system [2].
Beginning with the first realization of a Bose-Einstein condensate [3, 4], ultracold
atoms have developed into an excellent simulation platform with optically tunable trap-
ping geometries [5, 6]. Artificial crystals have been realized using optical lattices, leading
to the observation of the superfluid-to-Mott insulating phase transition [7] as the first
simulated strongly correlated system. These systems have since allowed the study of i.a.
non-equilibrium dynamics [8] or topological matter [9]. Tunable contact interactions
have been added using Feshbach resonances [10], and long-range interactions are also
studied [11, 12].
The development of quantum-gas microscopes for both bosons [13, 14] and fermions
[15, 16, 17, 18] has enabled single-atom resolved fluorescence imaging, providing the
detection of local observables required for correlation studies. Subsequently, the imple-
mentation of spatial light modulators has permitted arbitrary initial state engineering
[19], allowing the observation of spin dynamics [20, 21] and thermalization experiments
[22, 23].
Optical tweezers constitute an alternative simulation tool [24], where atomic positions
can be arbitrarily arranged, providing flexibility in trap geometry [25], but the atoms
themselves are fixed in place. Both flexible trap arrangement and atomic tunneling
can be in principle retained by projecting arbitrary potentials using the spatial light
modulator in a quantum-gas microscope.

Coherent Disorder in Projected Potentials

However, due to the coherent nature of the commonly used light sources, wavefront
distortions, e.g. induced by scatterers in the optical path, can lead to undesired random
2 1 Introduction

interference artefacts (“speckle” [26]), significantly degrading the projected image and
altering the physics to be studied [27]. Particularly, note that such static speckle patterns
have been used to study localization of cold atoms [28]. While in interference pattern
[29] or Fourier mask [13] generated potentials coherence is an intrinsic requirement, the
use of spatially incoherent light sources for real-space imaging is expected to greatly
enhance performance [30].
Note that since speckle noise is a very universal wave phenomenon, emergent from
wavelength-sized scatterers, it is present in a variety of fields, not only in optical full-
field imaging (e.g. laser projection [31], holography [32], optical coherence tomography
[33]) but also, most notably, in ultrasound imaging [34] and synthetic-aperture radar
imaging [35].

Incoherent Light Generation

The use of coherent lasers in imaging is mainly justified by the high light intensities
achievable. With this requirement, multiple approaches to generating light with reduced
coherence properties have been developed.

Direct Spatial Coherence Reduction

One approach for speckle reduction, where the use of a chosen coherent light source can
be continued, involves actively reducing the spatial coherence [36]. For this, a spatially
dependent phase changing, transmitting optical element is used to control the wavefront.
This purposefully introduces coherent noise in a diffraction plane, but temporally mod-
ulating the wavefront allows averaging of different realizations of speckle fields, resulting
in spatial coherence reduction.
The modulation can be realized by mechanically moving (e.g. rotating [37], vibrating
[38] or shifting [39]) a bulk optical diffuser (with random or engineered [40] diffractive
surface structure), where the rotating random diffuser approach is tested in sec. 3.4 and
has found commercial use in laser projectors [41]. Another possibility is to modulate the
incidence angle on a diffuser, where the angle can be introduced by e.g. piezoelectrically
actuated mirrors [42] or acousto-optic deflectors (AODs) (sec. 3.5).
Speckle also arises when coherent light passes through multimode optical fibres [43],
where the particular realization of the pattern is determined by the fibre mode structure.
Mechanically vibrating the fibre [44] or modulating the fibre coupling [45] thus gives rise
to speckle averaging.
 3

Temporal Incoherence Conversion

Alternatively, one can use the temporal incoherence of spectrally broad light sources to
reduce spatial coherence, which is accomplished by inducing spatially varying propaga-
tion delays onto the initial light beam. When the differential delays are larger than the
coherence times, they mutually dephase and become incoherent.
The time delays can be realized using bulk optics, e.g. echelons [46] or tilted etalons
with a transmissivity gradient [30]. An alternative to such thick structures are optical
fibres, where distinct propagation modes travel with different speeds. For this, e.g. a
multi-core fibre where the cores have varying lengths [47] or multimode fibres with a
single large core [48] (ch. 4) can be used. Similarly, amplified spontaneous emission
(ASE) fibre amplifiers with large cores have been shown to produce light with reduced
spatial coherence [49].

Low Spatial Coherence Light Sources

Finally, lasers with complex cavities and feedback mechanisms can be built, which in-
trinsically have low spatial coherence [50].
One approach uses vertical-cavity surface-emitting laser (VCSEL) arrays combining
multiple independently lasing (i.e. mutually incoherent) coherent light sources in one
light field [51]. Spatial multimode output also results from highly transversally multi-
mode cavities, realized e.g. as degenerate lasers [52], where an imaging system is placed
within the resonator. Speckle reduction is also achieved for random lasers [53], where
lasing occurs in highly disordered media.

Outline
Chapter 2 gives an introduction into how the coherence properties of light fields can lead
to spatial and temporal noise, and how this affects optical dipole trapping of cold atoms.
In chapter 3 diffuser-based spatial coherence reduction techniques are discussed. In par-
ticular, speckle pattern averaging is implemented using a rotating diffuser and using an
acousto-optically modulated angle of incidence. Chapter 4 demonstrates spatial coher-
ence reduction in multimode optical fibres based on modal dispersion induced dephasing
of broadband light sources. In chapter 5 the decoherence results are summarized and
compared.
2 Theory of Coherence and Dipole Traps

In this chapter the generation of arbitrary potentials for ultracold atoms and the influence
of coherent noise is described. Sec. 2.1 provides an introduction to Fourier optics and
coherence theory, and sec. 2.2 shows the mechanisms of optical dipole traps.

2.1 Coherence Theory

In order to describe the effects of optical coherence on imaging, properties of partially

coherent light fields are derived.
First, scalar diffraction theory (sec. 2.1.1) is introduced explaining the propagation
of fully coherent light. Then correlation functions (sec. 2.1.2) used to quantify co-
herence are discussed, followed by the diffraction properties of partially coherent fields
(sec. 2.1.3). Finally, the coherence properties are related to coherent noise reduction in
imaging systems (sec. 2.1.4).

2.1.1 Scalar Diffraction Theory

An electromagnetic field E(r, t) at spatial position r = (x, y, z) and time t in a source-

free, linearly responding, isotropic, non-magnetic, stationary medium with refractive
index n(r, ω) and temporal angular frequency ω evolves according to the homogeneous
wave equation [54]

n2 (r, ω) 2
∇2r E(r) + 2∇r (E(r, t) · ∇r ln n(r, ω)) − ∂t E(r, t) = 0, (2.1)
c2

with the local speed of light in the medium c/n(r, ω). In homogeneous media, n(r, ω) =
n(ω) (or in a weak waveguide, sec. 4.1), the central term vanishes and all spatial dimen-
sions obey the same wave equation, thus allowing consideration of only one component
of E → Ei ≡ E for any spatial dimension i. For a polarized wave the intensity is given
by I(r, t) = c n(ω)ε0 |E(r, t)|2 . For convenience, the scalar value A ∝ E absorbs the
prefactors to obey |A|2 = I.
6 2 Theory of Coherence and Dipole Traps

A time-domain Fourier transform decouples the frequencies ω (implicitly assumed by

n(ω)) and yields the Helmholtz equation

∇2r A(r, ω) + k 2 (ω)A(r, ω) = 0, (2.2)

with wavenumber k(ω) = k0 (ω)n(ω) and vacuum wavenumber k0 (ω) = ω/c. As a conse-
quence of linearity, the total field is thus a superposition of independent monochromatic
fields.

Diffraction Integral

Considering a light field A(ρ̃, z̃) where the field distribution in a transversal plane (ρ̃, z̃) =
(x̃, ỹ, z̃) is known, the problem of diffraction is to deduce the field in a plane at arbitrary
positions z, where z is the optical axis. The Rayleigh-Sommerfeld diffraction formula in
the approximation z − z̃  λ [54], with wavelength λ, formalizing the Huygens-Fresnel
principle of spherical wave superposition, gives the solution as

k(ω)(z − z̃) eik(ω)∆r 2

Z
A(ρ, z, ω) = A(ρ̃, z̃, ω) d ρ̃, (2.3)
2πi ∆r2

with difference vector ∆r = (ρ − ρ̃, z − z̃). In particular, it can be written as a super-

position integral with an impulse response function h,
Z
A(ρ, z, ω) = hz−z̃ (ρ, ρ̃, ω)A(ρ̃, z0 , ω). (2.4)

Propagation through a system can then be modelled by evaluating the diffraction integral
between optical elements and changing the wavefront according to the effect of the
element.

Paraxial Propagation

If the propagating field has low divergence, |ρ− ρ̃|  |z −z0 |, the paraxial approximation
can be applied, which involves expanding the positional difference (from the previous
section) as ∆r/∆z ≈ 1 + (|∆ρ|/∆z)2 /2 + O((|∆ρ|/∆z)4 ).
The quadratic wavefront approximation, known as Fresnel approximation, (dropping
the independent frequency ω dependence) thus leads to a propagation function of

keik∆z
 
ik 2
h∆z (ρ2 , ρ1 ) = exp (ρ2 − ρ1 ) , (2.5)
2πiz 2z
2.1 Coherence Theory 7

whereas for the constant wavefront (far-field, Fraunhofer) approximation one obtains the
Fourier transform kernel

keik∆z
   
ik 2 ik
h∆z (ρ2 , ρ1 ) = exp ρ exp − ρ1 · ρ2 . (2.6)
2πiz 2z 2 z

It can be shown that in Fresnel approximation a thin lens converts the light field in
its front focal plane into its Fraunhofer diffraction pattern in its rear focal plane [54].
Two lenses with focal lengths f1,2 in such a configuration Fourier transform the incident
field twice and thus constitute a 4f-imaging system with magnification M = f2 /f1 . In
an ideal, diffraction-limited imaging system the magnification-normalized propagation
function then becomes shift-invariant,

h(ρ2 , ρ1 ) = h(ρ2 /M − ρ1 ), (2.7)

and the diffraction integral becomes a convolution integral whose Fourier transform
defines the amplitude transfer function.

Gaussian Beam

The Fresnel-approximated Helmholtz equation does not only have the paraxial spherical
waves as solutions but also a set of Gaussian beams. Typical laser beams are well-
approximated by the lowest order mode of the set (“Gaussian beam”), which is given
by
w0 −ikρ2 /2R(z) −ρ2 /w2 (z)
A(ρ, z) = A0 e e , (2.8)
w(z)
p
where w(z) = w0 1 + (z/zR )2 is the beam width, with waist w0 and Rayleigh range
zR = kw02 /2, and R(z) = z + zR
2 /z is the wavefront curvature radius (neglecting the

purely longitudinally dependent Gouy phase).

2.1.2 Correlation Functions

The previous section fully describes the behaviour of deterministic light fields. However,
when random processes are involved, due to the light field itself or because of changes
of the optical system, only statistical predictions can be made.
Generally, a correlation function describes the expectation value of a product of ran-
dom variables (i.e. their mutual similarity), where it is called autocorrelation if the
factors describe the same variables, and cross-correlation if they denote different vari-
8 2 Theory of Coherence and Dipole Traps

ables. In optical coherence theory, one is interested in the correlation properties of a

light field A(r, t) and can generally define an n-th order correlation function as

n
" #
(n)
Y
ΓA ({ri }, {r0i }, {ti }, {t0i }) =E A ∗
(ri , ti )A(r0i , t0i ) , (2.9)
i=1

where E[X] denotes the expectation value of X over its statistical ensemble. First-order
correlation functions are of particular importance as they give lowest order insight into
the fluctuations of a quantity of interest and are suitable to describe noise characteristics.
The first-order correlations of the field A(r, t) and of its intensity I(r, t) = |A(r, t)|2 are
denoted as

(1)
ΓA (r1 , r2 , t1 , t2 ) ≡ ΓA (r1 , r2 , t1 , t2 ) and
(1) (2)
ΓI (r1 , r2 , t1 , t2 ) ≡ ΓI (r1 , r2 , t1 , t2 ) = ΓA (r1 , r2 , r1 , r2 , t1 , t2 , t1 , t2 ).

Since any observable field is a deterministic function, one typically evaluates the time-
average of the correlation function

ΓA (r1 , r2 , τ ) = E [hA∗ (r1 , t)A(r2 , t + τ )it ] , (2.10)

R T /2
where hX(t)it = limT →∞ −T /2 X(t)dt/T and ΓA is called the mutual coherence function.
In the case of a random signal the observable deterministic function can be thought of
as one (probable) realization from the ensemble, thus the statistical average is kept. The
combination of both averages is denoted as hX(t)i = E[hX(t)it ].
The first-order coherence function is defined as the normalized correlation function
and is given by
ΓA (r1 , r2 , τ )
γA (r1 , r2 , τ ) = p (2.11)
ΓA (r1 , r1 , 0)ΓA (r2 , r2 , 0)
such that the equal-space-time coherence is unity. The modulus of the coherence func-
tion, |γA (r1 , r2 , τ )|, is called the degree of coherence. Note that the equal-space-time
first-order correlation is the mean intensity, ΓA (r, r, 0) = hIA (r, t)i.
If one is only interested in the spatial or temporal length scale up to where correlations
appear, one can define the coherence length as [55]
Z ∞
sc = |γA (s)|2 ds, (2.12)
−∞

where s = τ for the coherence time and s = |r| for the spatial coherence length. Here,
2.1 Coherence Theory 9

the implied coherence function is γA (s) ≡ γA (s0 , s0 + seδs ) at a position s0 measuring

the coherence length along a direction eδs . For an exponential distribution e−|s|/σ , this
2 2
definition yields a coherence length of sc = σ, and for a Gaussian distribution e−s /2σ
√
it is sc = σ π. Similarly, one can define a coherence area Ac = |γA (s)|2 d2 s and a
R

coherence volume Vc = |γA (s)|2 d3 s.

R

First-Order Temporal Correlations

The magnitude of coherent noise (interference patterns) in longitudinal direction is de-

scribed by the temporal first-order correlation function given by the equal-space generic
first-order function and, dropping the spatial dependence, is written as

F
ΓA (τ ) = hA∗ (t)A(t + τ )i −−→
τ
|A(ω)|2 = SA (ω), (2.13)

where Ft denotes the asymmetric Fourier transform operator defined as Ft {f (t)}(ω) =

f (t)e−iωt dt. If the field A(t) is deterministic, its Fourier transform A(ω) exists and,
R

using the convolution theorem, the Fourier transform relation is proven. If A(t) is a wide-
sense stationary random process, different frequency components become uncorrelated
and the Wiener-Khinchin theorem [56] can be applied, stating the relation. SA (ω)
is called the power spectral density of the field (“optical spectrum”) and describes the
chromatic composition of a light field. Note that normalization by the intensity hIA (t)i =
R Fτ
SA (ω)dω/2π yields the normalized form in both time and spectral domain, γA (τ ) −−→
sA (ω). Also note that for real-valued signals A(t), using the convention of a two-sided
power spectral density, it is symmetric in frequency SA (−ω) = SA (ω).
If a deterministic and a random field have the same spectral composition, the differ-
ence between them lies in the phase relation of the Fourier components. While in the
deterministic case each frequency has one associated phase, the stochastic nature of the
random field requires a phase ensemble averaging, thus the Fourier transform A(ω) does
not necessarily exist. If a stochastic A(ω) is given, it is defined through SA (ω) and called
the amplitude spectral density.
Using the Wiener-Khinchin theorem (2.13), a definition of the spectral bandwidth can
be given as [55]
∆ω 1
∆ν = =R 2 , (2.14)
2π sA (ω)dω
which, using Parseval’s theorem and the definition for the coherence length (2.12), is
shown to obey the reciprocal relation ∆ντc = 1.
10 2 Theory of Coherence and Dipole Traps

Second-Order Temporal Correlations

Temporal fluctuations in intensity can be described with the second-order temporal

correlation function

F
ΓI (τ ) = hI(t)I(t + τ )i = hA∗ (t)A(t)A∗ (t + τ )A(t + τ )i −−→
τ
SI (ω), (2.15)

where the Wiener-Khinchin theorem is again applied and SI (ω) is the intensity power
spectral density (“noise spectrum”).
If A(t) is a deterministic field (e.g. single-mode laser), insertion of its Fourier transform
A(ω) yields the relation SI (ω) = | hA∗ (ω̃)A(ω̃ + ω)i |2 . Thus temporal intensity noise
can be interpreted as correlations in the frequency domain. Particularly, this describes
mixing of coherent signals such as mode beating.
If A(t) is a random field obeying zero-mean Gaussian statistics (“chaotic light”, e.g.
thermal sources), the moment theorem [55] yields the Siegert equation

ΓI (τ ) = Γ2A (0) + |ΓA (τ )|2 , (2.16)

relating noise in intensity to first-order (field) coherence properties. Particularly, the

carrier-normalized temporal relative intensity noise (RIN) sI (ω) can be related to the
first-order coherence function γA (τ ) by sI (ω) = |γA (τ )|2 e−iωτ dτ , which is inverse to
R

and decays in the order of the optical bandwidth.

Spatial Correlations

Instead of considering temporal correlations, one can analogously study the spatial cor-
relation properties of the field A(r, t) and define the spatial second-order correlation
function as
ΓA (r1 , r2 ) = hA∗ (r1 , t)A(r2 , t)i , (2.17)

which describes the visibility of a spatial interference pattern (or transversal coherent
noise). Note that temporal effects can be included using the cross-spectral density W
R
[55], with ΓA (r1 , r2 ) = W (r1 , r2 , ω)dω since different frequencies behave independently
(2.2).
Similarly to temporally random fields with vanishing temporal coherence, there also
exist spatially random fields where spatial coherence decays. In analogy to temporally
statistically stationary fields, one is often interested in Schell model sources, for which
2.1 Coherence Theory 11

the spatial coherence function depends only on its positional difference,

γA (r1 , r2 ) ≡ γA (r2 − r1 ). (2.18)

Note that fields with Gaussian coherence function and Gaussian intensity envelope are
called Gaussian Schell models [57].

Since the formal definitions of coherence always apply infinite time averaging, no state-
ment is made about the rate of convergence to this residual coherence value. Consider
a spatially incoherent light source. For an infinitesimal amount of time the wavefront
of the field does not change and is thus spatially coherent. In order to become spatially
incoherent, the relative phases across the observation plane must change, which happens
on the time scale of the coherence time. Therefore, the light field is effectively only
spatially incoherent when the observation dynamics happen on time scales which are
much longer than the coherence time. If the evolution of the field can be described as
a Gaussian random process, the Siegert relation (2.16) can be used to show that the
residual coherence of two mutually incoherent fields decays with integration time scale
T  τc as r
τc
|γA | = , (2.19)
T
with coherence time τc [58]. The coherence value |γA,res | ≡ |γA (∆s → ∞, T → ∞)| will
be referred to as “residual coherence”, where ∆s denotes spatial distance.

The “level of incoherence” of a field is determined by the number of its intrinsic

degrees of freedom (“effective number of mutually incoherent modes”) M , which can
be defined as the ratio of the beam area to the coherence area, M = AI /Ac . Let
i(ρ) = I(ρ)/ I(ρ)d2 ρ denote the normalized intensity envelope, then the coherence
R

i(ρ1 )i(ρ2 )|γA (ρ1 , ρ2 )|2 d2 ρ1 d2 ρ2 . Defining the

RR
area (2.12) can be generalized as Ac =
beam area analogously, AI = i(ρ)d2 ρ, the number of incoherent modes thus yields [59]
R

|ΓA (ρ1 , ρ2 )|2 d2 ρ1 d2 ρ2

RR
Ac
Z
1 2
=
2 ≈ 2 IA (ρ)d2 ρ, (2.20)
M P
R
IA (ρ)d ρ2
A

IA (ρ)d2 ρ is the power of the light field, and the approximation is valid
R
where PA =
for Schell model sources in the incoherent limit. Note that for non-uniform illumination,
the integration area is reduced to a region of interest, with a cutoff typically set at the
2 /2W 2
FWHM of the intensity (fig. 2.1). For a Gaussian intensity profile e−ρ , this results
in M = πW 2 /3Ac .
12 2 Theory of Coherence and Dipole Traps

single coherent mode

1.25

Normalized intensity I
incoherent superposition
1.00

0.75

0.50

0.25

0.00
−0.75 −0.50 −0.25 0.00 0.25 0.50 0.75
Position ½=w

Figure 2.1: Number of mutually uncorrelated areas.

The light grey graph shows the Gaussian intensity envelope of an incoherent
field with waist w while the blue graphs show the coherence areas generating
the beam. In the illustrated (1D) case, the number of mutually incoherent
modes is thus M = 9 for a full width at half maximum (FWHM) cutoff (dark
grey).

2.1.3 Propagation of Correlations

When studying the coherence properties of a light field, rather than at the position of
light generation, the properties at a distant plane after propagation through an optical
system is typically of interest.
Consider an initial, quasi-monochromatic transversal field a(ρ̃) propagating with func-
tion hz (ρ, ρ̃) (sec. 2.1.1) to a target plane A(ρ). Suppose that the propagation function
hz is deterministic (i.e. a known optical system), then one can write ΓA as the propa-
gated result of the initial correlation function Γa ,
ZZ
ΓA (ρ1 , ρ2 ) = h∗z (ρ1 , ρ̃1 )hz (ρ2 , ρ̃2 )Γa (ρ̃1 , ρ̃2 )d2 ρ̃1 d2 ρ̃2 . (2.21)

Propagation of Schell Model Sources

If a spatially incoherent light field is produced by multiple randomly distributed, inde-

pendent emitters, their emission statistics are stationary and the light source can be
assumed to be a Schell model source. Then it is convenient to express the correlation
function in relative, ∆ρ, and mean coordinates, ρ, i.e. ρ1,2 = ρ ± ∆ρ/2.
Consider the field propagating to a Fraunhofer diffraction plane (2.6). In the incoher-
ent limit, the coherence function falls off much faster than the intensity envelope, hence
p
Ia (ρ1 )Ia (ρ2 )γa (∆ρ) ≈ Ia (ρ)γa (∆ρ). Then the correlation integrals factorize such that
2.1 Coherence Theory 13

100

50
Beam width w=¸

−50
sc = 5¸
sc = 10¸
sc = 20¸
−100
sc = 100¸

−1000 −750 −500 −250 0 250 500 750 1000

Position z=¸

Figure 2.2: Beam divergence dependence on spatial coherence.

For the same beam waist w0 = 10λ (with wavelength λ), spatially incoherent
beams (coherence length sc → 0) exhibit a much larger divergence w/z than
coherent beams (sc → ∞). Finite numerical aperture (NA) apertures set a
bound to divergence, which limits the possible degree of incoherence within
an optical system.

one obtains the generalized van Cittert-Zernike theorem [58]

 2 Z Z
k − ik ik
γA (∆ρ)IA (ρ) = Ia (ρ̃)e z
ρ̃·∆ρ 2
d ρ̃ γa (∆ρ̃)e− z ∆ρ̃·ρ d2 ∆ρ̃. (2.22)
2πz

Looking at the dependences of the integrals, one observes that the coherence function in
the diffraction plane is the Fourier transform of the intensity envelope in the initial plane
and vice versa. This can be understood by picturing each coherence area as an individual
Gaussian beam. If the coherence area (i.e. waist) of γa is small, the divergence becomes
large, explaining the broader envelope of intensity IA . The more coherence areas (i.e.
individual beams) exist (which is proportional to the width of the intensity envelope Ia )
for a fixed divergence, the finer the diffraction (i.e. interference) pattern γA becomes.
Since the coherence areas (i.e. beams) are mutually incoherent, γA decays quickly, as
expected for the coherence function of an incoherent field.
This Fourier transform relation implies that a reciprocity relation (“M-squared value”)
analogous to (2.14) between source size and divergence is given that remains invariant
throughout propagation. For real, finite-sized optical systems, its maximal number is
14 2 Theory of Coherence and Dipole Traps

thus limited by the element with the smallest value for the product of field size in a
focus and NA. If the field is too incoherent, it will be clipped, reducing the light power
efficiency. Fig. 2.2 shows the divergence of a Gaussian Schell model beam, whose
evolving beam width is given by [57]
 2  !
2z 1 π
w2 (z) = w02 1+ 2 + 2 , (2.23)
kw0 w0 sc

where w0 denotes the intensity width of a Gaussian beam, sc the spatial coherence length
and k the wavenumber.

Imaging Resolution

For an imaging system, the propagation integral, which maps an input field a(ρ̃) to
an output field A(ρ), is given by the convolution A(ρ) = h(ρ − ρ̃)a(ρ̃)d2 ρ (2.7),
R

where w.l.o.g. the magnification is assumed to be unity, M = 1. Thus, expressing this

statement in Fourier space reduces the relation to a product,

A(ν) = H(ν)α(ν), (2.24)

where ν is the spatial frequency, H(ν) = Fρ {h(ρ)}(ν) the amplitude transfer function
and A, α the Fourier transforms of the fields A, a. Fourier transforming the correlation
propagation function (2.21), GA (ν1 , ν2 ) = Fρ1 ,ρ2 {ΓA (ρ1 , ρ2 )}(ν1 , ν2 ), thus yields the
intensity spectrum
Z
IA (ν) = Fρ {IA (ρ)}(ν) = H∗ (ν̃)H(ν̃ − ν)Ga (ν̃, ν̃ − ν)d2 ν̃. (2.25)

As a transfer function, H(ν) determines the imaging resolution of a certain spatial

frequency ν, where H(ν) = 1 means a perfect reproduction in the image plane.
In the coherent limit Ga (ν1 , ν2 ) = α∗ (ν1 )α(ν2 ), one thus obtains a spatial intensity
spectrum of IA (ν) = H∗ (ν̃)α∗ (ν̃)H(ν̃ − ν)α(ν̃ − ν)d2 ν̃, whereas the incoherent limit
R

Ga (ν1 , ν2 ) = Ia (ν1 − ν2 ) yields IA (ν) = Ia (ν) H∗ (ν̃)H(ν̃ − ν)d2 ν̃.

R

These relations can be most easily understood when considering a quadratic pupil,
Q
which is equivalent to a rectangular transfer function H(ν) = i={x,y} Θ(ν0 − |νi |),
with spatial cutoff frequency ν0 and heaviside function Θ. In the coherent case, all
frequencies up to ν0 are perfectly reproduced, resulting in a high contrast, accurate image
for large scale structures. In the incoherent case, one has to consider the convolution
2.1 Coherence Theory 15

10-1

10-2
Resulting relative noise ±I
10-3

10-4

10-5 j°AB j = 0:00

j°AB j = 0:01
10-6 j°AB j = 0:05
j°AB j = 0:10
10-7 j°AB j = 0:30
j°AB j = 1:00
10-8
10-4 10-3 10-2 10-1
Relative strength of noise field ±Icoh

Figure 2.3: Relative noise level dependence on degree of coherence.

The log-log plot compares the relative noise for fully coherent superpositions
δIcoh with the relative noise for partial coherence δI at different degrees
of coherence |γAB | (2.27). For sufficiently small fluctuations, noise levels
decrease proportional to the degree of coherence.

of the amplitude transfer function, i.e. a triangular function H∗ (ν̃)H(ν̃ − ν)d2 ν̃ =

R
Q
i={x,y} (1 − |νx |/2ν0 )Θ(2ν0 − |νi ). The sub-unity values for finite spatial frequencies
indicate a reduction in image contrast, but the larger achievable spatial frequencies (up
to 2ν0 ) enable higher resolution. In summary, incoherent imaging leads to higher resolved
but more blurred images and reduces ringing artefacts.

2.1.4 Reduction of Coherent Noise

The goal of an imaging system is to propagate a given intensity distribution from an

object plane to an image plane, where a scaled, diffraction limited copy of this intensity
distribution is produced. In real systems, imperfections always lead to deviations in the
light field.
Consider a target intensity distribution IA (ρ) and a noise field B(ρ, t). The resultant
intensity is then given by the noise-averaged, coherent field superposition
p
IA+B (ρ) = IA (ρ) + hIB (ρ, t)it + 2 IA (ρ) hIB (ρ, t)i Re[γAB (ρ, t)], (2.26)
16 2 Theory of Coherence and Dipole Traps

noise in object plane R=1 noise in objective R=5

Figure 2.4: Wave redundancy for noise in different planes.

The illustration shows five mutually incoherent (blue) rays propagating
through an imaging system. If a noise source (generating red rays) is lo-
cated in the object/an intermediate image plane (left), the same coherence
area is imaged onto the image plane, so the wave redundancy is R = 1 and
spatial incoherence does not suppress noise. In the typical case of noise in the
objective plane (right), a point in the image plane is illuminated by multiple
rays, here R = 5, such that coherent noise is suppressed.

where the interference induced noise magnitude is proportional to the modulus of the
cross-correlation |ΓAB | (the last term). Assuming that the noise source cannot be re-
moved or avoided, the incoherent intensity addition gives the lower noise bound. Writing
the noise terms relative to the target intensity, δI = IA+B /IA −1, gives the relation plot-
ted in fig. 2.3,
p
δI(|γAB |) = δIcoh − 2(1 − |γAB |)( 1 + δIcoh − 1) ≈ |γAB |δIcoh , (2.27)

where δIcoh = δI(|γAB | = 1) is the relative noise at full coherence.

Estimation of the actual noise reduction is difficult as it does not only depend on
the noise properties and position, but also on the illuminating light field (which for an
arbitrary image can vary strongly). To provide a qualitative picture, imagine the image
formation process as the superposition of different modes (fig. 2.4). One can formally
define the wave redundancy [60] given by

X |an |2
R(ρ) = , (2.28)
maxk |ak |2
n∈N (ρ)

where an denotes the weight of mutually incoherent modes and the summation goes over
N (ρ), which are the modes constituting the field at an observation point ρ. R(ρ) → 1
means that all the light at ρ in the image plane comes from a single mode of the noise
plane, whereas R(ρ) → ∞ means that the noise source only contributes an infinitesimal
amount. Thus the effective coherence is expected to be limited by |γAB | → |γAB | + 1/R.
2.2 Optical Dipole Traps 17

First, consider planar reflections (e.g. at plane-parallel windows) such that the noise
P
field is given by the time-delayed superposition of the target field B(t) = n ηn A(t−τn ),
where ηn is the relative strength of the n-th reflection. Then the wave redundancy yields
P P
R = 1/ n ηn and the interference term is thus given by δI ≈ IA n ηn γA (τn ) such that
coherence times smaller than the reflection delays can significantly reduce noise.
Second, consider scatterers in an arbitrary plane of the image path, as illustrated in
fig. 2.4. If the noise source is located in a Fourier plane, every image point is illuminated
by all modes, such that the wave redundancy is maximal, R → ∞. Since typically the
scatterer is located at the objective, this case is often fulfilled and eq. (2.27) can be
used to estimate the noise reduction. However, if the scatterer is located in the object
or an intermediate image plane, the wave redundancy is minimal for a perfect imaging
system, R = 1. For a finite point spread function, the noise is correspondingly slightly
reduced.

2.2 Optical Dipole Traps

In this section optical dipole traps for neutral atoms are introduced and the influence of
partial coherence on spatial inhomogeneities and system heating is discussed.

2.2.1 Atomic Light Shifts

When considering the interaction between a light field E(r, t) and an atom in state |ii,
the realization that electronic dynamics happen on much faster time scales than nuclear
dynamics allows for an adiabatically approximated approach, separating the dynamics
of the “internal” electronic part from the nuclear centre-of-mass part. As typical extents
of atoms are in the subnanometre regime, i.e. much smaller than optical wavelengths
at hundreds of nanometres, the electric dipole approximation can be applied and the
field can be treated as locally homogeneous. If only one valence electron is involved, the
electronic Hamiltonian reduces to a hydrogen-like problem in a spatially homogeneous
field E(t). The electronic field-free eigenenergy difference between states |ji and |ii will
be denoted as ~ωji = Ej − Ei .
If only single-photon effects are taken into account, different photonic frequencies ω
and polarizations p can be considered independently, resulting in the Hamiltonian
X
Ĥ 0 (ω, p) = er̂ · E p (ω) = ~dji (p)Ep (ω) |ji hi| , (2.29)
i,j
18 2 Theory of Coherence and Dipole Traps

where Ep (ω) is the respective Fourier component of the field E(t), dji (p) = hj| er̂ · p |ii
is the dipole matrix element and the Rabi frequency Ωji (ω, p) = dji (p)Ep (ω) can be
defined. For small coupling with respect to detuning, ∆ji (ω) = ω − ωji  Ωji (ω, p),
second-order perturbation theory yields an eigenenergy shift for the state |ii of [61]

sp (ω) ~|dji (p)|2

∆Eji (ω, p) = 2ε0 cI, (2.30)
2π ∆ji (ω)

where Sp (ω) = Isp (ω) ∝ |Ep (ω)|2 denotes the (two-sided) optical power spectral density
(2.13) and I is the total light intensity. Note that using the Wigner-Eckart theorem, the
dipole matrix element dji (p) can be factorized into the spherically symmetric reduced
matrix element kdji k and a branching factor cji (p) [62].
The total light shift is obtained by summing the contribution from each frequency,
R P P
polarization and coupling state, ∆Ei = dω p j6=i ∆Eji (ω, p). For the typical case,
where |ii is a ground state, the detuning ∆(ω) is large compared to the energy splitting
ω0 between relevant excited states |ji and the light polarization is spectrally independent
sp (ω) = sp s(ω), the light shift can be reduced to

3πc2
Z
dω αi Γ
∆Ei = s(ω) I, (2.31)
2ω03 2π ∆(ω)

where Γ = ω03 kd0 k2 /3πε0 ~c3 denotes the spontaneous decay rate and the factor αi =
2
P P
p j6=i sp |cji (p)| accounts for selection rules. Note that apart from αi , this result
recovers the classical dipole potential, where the light field interacts with an induced
atomic dipole.
Since the dynamics of the whole atom depends only parametrically on the electronic
state, its total energy varies according to the light intensity. Spatially varying inten-
sity distributions I(r) thus allow the creation of arbitrary, internal state |ii dependent
potential landscapes Ui (r) = ∆Ei (r) ∝ I(r).

2.2.2 Heating of Trapped Atomic Systems

Since one is typically interested in atoms in well-defined quantum states, heating pro-
cesses are undesirable, as they induce uncontrolled state changes. Electronic (internal)
state changes pose a problem since trapping potentials, detection mechanisms and ob-
servable physics are often state-selective. When studying itinerant physics, it is also
important to avoid motional state changes induced by fluctuating potentials.
2.2 Optical Dipole Traps 19

Photon Scattering

One such process is spontaneous photon scattering, which can be seen as coherent pho-
ton absorption k1 and reemission k2 , with which a momentum transfer is associated,
depositing an energy of Esc (k1 , k2 ) = ~2 (k2 −k1 )2 /2m, where m is the mass of the atom.
Assuming the energy splittings within ground states to be small compared to the light
frequency and demanding energy conservation up to Esc implies quasi-elastic scattering
ω1 ≈ ω2 . Neglecting polarization and anisotropy effects, the heating rate for state |ii is
thus given by [61]
~2 c2
Z
dω
Psc,i = s(ω)ω 2 Γsc,i (ω), (2.32)
m 2π
where Γsc,i (ω) denotes the photon scattering rate.
Again only considering single-photon effects, scattering can be modelled within a
homogeneously broadened two-level system. In the large detuning limit the steady-state
solution of the optical Bloch equation then yields
2
3πc2

Γ
Γsc,i (ω) = I, (2.33)
2~ω03 ∆(ω)

with the spontaneous emission rate Γ. Since the scattering rate decreases faster with
detuning than the light shift, ∆Ei /Γsc,i ∝ 1/∆, far off-resonant light is typically used
for dipole traps.

Parametric Heating

Apart from photon scattering, the motional state of an atom can also be directly changed
by classical fluctuations of the optical dipole potential (i.e. the light intensity). For
analysis, the potential V (r, t) = V0 (r) + V (r, t) is decomposed into a static target
potential V0 (r) and a spatiotemporal noise term r (r) ≡ h(r, t)i.
First, the case of a spatially coherent field is considered, such that perfect correlations
across the field exist and the dependence on the position factorizes, (r, t) = r (r)t (t).
The noise effects can be described in terms of transition rates between motional states
of the atoms within first-order time-dependent perturbation theory and are given by

1
Γmn = | hm| V r (r) |ni |2 St (ωmn ), (2.34)
~2

where ~ωmn is the energy difference between the corresponding states and St (ω) is the
relative temporal intensity power spectral density (2.15). It is thus evident that resonant
20 2 Theory of Coherence and Dipole Traps

noise has to be minimized to avoid driving these transitions. If one is interested in

dynamics in the temporal regime τ where the |mi ↔ |ni transitions are temporally
resolved, ωmn . 1/τ , fluctuations in this frequency regime can lead to random adiabatic
state changes. Therefore temporal noise is only acceptable for fluctuation frequencies
much larger than any relevant transition frequency.

Second, if the field is incoherent, one can distinguish between fluctuations which are
mutually independent between different coherence areas and global fluctuations influ-
encing all coherence areas simultaneously. Since the light field is required to appear
spatially incoherent for the atomic system (2.19), it can be assumed that fluctuations
at the time scales of the prior are much faster and therefore insignificant. Taking into
account only the global fluctuations, the temporal fluctuation statistics are position in-
dependent, allowing (r, t) = r (r)t (t) to again be factorized, yielding the above result.

Note that the energy deposited in the system is directly given by the transition energy
Eh = ~ωmn .

As an example consider a harmonic trap potential V0 (x) = mω02 x2 /2 with trap fre-
quency ω0 . Assuming that the trap potential itself is globally fluctuating r (x) ∝ V (x),
one obtains a transition rate of Γmn ∝ | hm| x̂2 |ni |2 Sεt (ωmn ) ∝ δn+2,n Sεt (2ω0 ) and, since
the energy levels are equidistantly spaced, exponential heating [63].

2.2.3 Influence of Spatial Noise

In order to determine the effect of static spatial noise in optical potentials, consider
the a setting in which atoms are trapped in a two-dimensional (2D) square lattice
V (ρ) = V0 (sin2 (klat x) + sin2 (klat y)) with lattice wavenumber klat . For bosons, a contact
interaction and a tight binding approximation for the Wannier functions hρ|ii = wi (ρ) =
w(x − xi ) w(y − yi ) yields the Bose-Hubbard model [64]

UX
â†i âj +
X X
Ĥ = −J n̂i (n̂i − 1) + (i − µ)n̂i , (2.35)
2
hi,ji i i

where â†i (âi ) denotes the particle creation (annihilation) operator at lattice site i, n̂i =
â†i âi is its number operator and µ is the chemical potential. The kinetic energy term J
describes hopping between adjacent lattice sites, U the contact interaction and i the
2.2 Optical Dipole Traps 21

local on-site energy. These parameters are given by

√ √
J = hj| Ĥ0 |ii ≈ 2Er v0 e−2 v0 ,
√
Z
4π~as
U= |wi (ρ)|4 d2 ρ ≈ 4as Er v0 ,
m
√
i = hi| V̂ |ii ≈ Er v0 ,

where Ĥ0 = p̂2 /2m + V̂ is the atomic Hamiltonian, v0 = V0 /Er is the normalized
potential magnitude, m is the atomic mass, as is the normalized scattering length and
Er = ~2 klat
2 /2m is the recoil energy. The approximations are obtained for harmonic

expansion of the potential and are valid in the Lamb-Dicke regime, i.e. V0  Er . Since
scattering lengths are short and the exponential term small, the on-site energy is the
largest term i  U, J.
First, consider the case in which the lattice potential is projected and the trap fre-
quencies vary at certain sites, i.e. a proportional scaling factor is introduced such that
locally V (ρ) → (1 + δ)V (ρ). Since the absolute value i is by far the largest compared to
the other parameters, fluctuations δ can easily lead to on-site energy changes larger than
√
U, J, significantly impacting dynamics. The on-site energy thus scales as i → i δ.
However, projected potentials are typically diffraction limited and cannot realize
tightly spaced lattice sites. A more probable type of speckle noise is a potential variation
which can be assumed to be constant over one lattice site. In this case a local offset is
introduced, V (ρ) → V (ρ) + V0 δ, and the on-site energy scales as i → i + V (ρi ) δ.
Following eq. (2.27) the use of light with reduced spatial coherence |γA | proportionally
p
reduces fluctuations δ → |γA | δ. Therefore noise is reduced by a factor of |γA | or |γA |,
respectively.
3 Diffuser-Controlled Decoherence

Since coherent light sources provide the advantage of high control and intensity, it is
desirable to continue to use these sources and change their coherence properties. As
spatial coherence means that a light beam is mutually phase stable in a given propagation
plane, coherence can be reduced by dynamically shaping the wavefront where the phase-
modulation pattern is chosen in such a way that the mean phase change vanishes.
In sec. 3.1 the properties of diffusers and modulation-controlled decoherence effects are
derived. Sec. 3.2 describes the experimental setup used to obtain the results for static
speckle characterization (sec. 3.3) and coherence reduction due to a rotating diffuser
(sec. 3.4) and acousto-optic beam deflection (sec. 3.5).

Diffuser-Based Wavefront Modulation

When spatially and temporally coherent, polarized light illuminates an object structured
on the scale of the wavelength (“diffuser”), the light is scattered at the microscopic
diffractive elements that locally distort the wavefront. Considering a large number of
such illuminated grains, the observed light field is a coherent superposition of the scat-
tered light, thus producing an interference pattern. As the diffracting structures (“scat-
terers”, “grains”) of diffusers are random surface height fluctuations, the interference
pattern is also random and is called a speckle pattern (individual patches of similar
intensity are called “speckles”).

Figure 3.1: Temporal averaging of uncorrelated speckle patterns.

Integrating a series (middle) of uncorrelated speckle patterns (left) over time
averages out the mean intensity (right). As the averaging process between
two distant points are uncorrelated, the resulting field is spatially incoherent.
24 3 Diffuser-Controlled Decoherence

When the diffuser structure changes in time, the speckle pattern also changes and
temporal integration leads to a blurring of the observed field (fig. 3.1). As the statistics
of the diffracted field reflects the statistics of the diffuser structure, globally uncorrelated
modulation patterns produce the desired spatially incoherent light source. Assuming a
fixed spot illuminating the diffuser, the required phase modulation can be obtained by
moving the diffuser relative to the spot (e.g. moving the diffuser) or by varying the
illumination conditions (e.g. the angle of incidence).

3.1 Theory

In this section the statistical properties of speckles are derived for both the static and
the dynamically modulated case.
A diffuser is a transparent plane plate whose surface is roughened on the light exiting
side, thus its bulk approximately does not change the incident wavefront and its only
effect is a position-dependent phase delay due to varying surface heights of the optically
dense diffuser material. Assuming these phase delays to be stochastic, this results in
a random phase screen, where propagation obeys Fresnel diffraction (2.5). In order
to analyse speckles to deduce information on the second-order coherence properties of
the resulting light field, knowledge about one-point (first-order statistics in sec. 3.1.1)
and two-point (second-order statistics in secs. 3.1.2 and 3.1.3) correlation functions is
necessary.

3.1.1 Speckle Noise Statistics

Consider a spatiotemporally coherent light beam incident on a random phase screen and
uniformly illuminating a large number of scatterers. In the far field near the optical
axis, the spatial dependence can be dropped for first-order statistics as all scatterers
contribute to the field. The resulting complex scattering field à is thus given by a
coherent superposition of the diffraction fields ãn created by each emitter (i.e. scatterer)
and yields à = n ãn = Aeiθ , where (A, θ) denotes its polar decomposition.
P

Since the scatterers are uniformly illuminated, and considering the emitters as inde-
pendent identically distributed complex random variables an eiθn , the problem becomes a
random walk in amplitude space and the central limit theorem (CLT) can be applied for
the real and imaginary parts, yielding Gaussian distributions with standard deviation
σA . Transforming this probability distribution into polar representation gives rise to the
3.1 Theory 25

Rayleigh distribution
A −A2 /2σ2
pA,θ (A, θ) = 2 e
A, (3.1)
2πσA
where pA,θ denotes the joint probability density function of A and θ. Replacing the
amplitude by intensity I = A2 and performing the trivial phase integral over θ gives an
exponential probability density function for the intensity,

1 −I/2σ2 1 −I/E[I]
pI (I) = 2 e A = e , (3.2)
2σA E[I]

where the statistical moments E[I q ] = q! E[I]q are substituted and E[X] indicates the
ensemble average of X. In particular, note that the relative root mean square (RMS)
intensity deviation, called speckle contrast in this context, is given by

∆I
C= = 1, (3.3)
E[I]

with variance ∆I 2 = E[I 2 ] − E[I]2 , thus indicating a large range of possible intensity
values typical for interference patterns.
Since all points around the centre of the observation plane are assumed to be sta-
tistically identically illuminated but uncorrelated, the spatial distribution of intensities
reflects the first-order statistics, resulting in a light field with large intensity fluctuations,
i.e. the speckle pattern. Patterns with unity contrast, i.e. fulfilling eq. (3.3), are called
fully developed speckles.
An analysis for finite numbers N of coherent superpositions yields a speckle contrast
p
of C = 1 − 1/N [58]. Considering this behaviour as an estimate of the convergence to
the above probability distribution and noting that diffractive grain sizes are commonly
well below ten times the beam size, the exponential behaviour can be seen as practically
always valid.

Intensity Superposition

When temporally modulating the phase screen, different uncorrelated speckle patterns
are sequentially realized. If the modulation is faster than the observation response, the
intensity is temporally integrated such that an incoherent (intensity) superposition of
fields is measured.
Independent speckle patterns arise when the scattering emitters, being random vari-
ables, individually change their particular realization. The degrees of freedom to achieve
this include illumination of uncorrelated scattering areas, polarization and wavelength
26 3 Diffuser-Controlled Decoherence

diversity or the shape of the illuminating wavefront. Incoherent superpositions can thus
be induced by sequentially changing above parameters.
Considering the intensity superposition of N independent, fully developed speckle
patterns I = N
P
n=1 In with the same mean intensity hIn i, it can be shown [58] that the
probability distribution of the intensity is given by a gamma density function

N N I N −1 −IN/E[I] N →∞ 1 2 2
pI (I) = N
e −−−−→ √ e−(I−E[I]) /2∆I (3.4)
Γ(N )E[I] 2π∆I

which, due to the CLT, quickly converges to a Gaussian. The distribution has mean
intensity E[I] = N hIn i and variance ∆I 2 = N hIn i2 , which gives the contrast

1
C=√ (3.5)
N

and thus has a scaling with N expected for statistically independent sums. It turns out
that for monochromatic fields the contrast is a measure for the residual spatial coherence
|γA,res | ≈ C (sec. 3.1.3).
The above probability distribution is reasoned for superpositions of discrete speckle
patterns. For continuous superpositions, where correlations decrease smoothly, a speckle
decorrelation time can be defined, such that after discretisation the gamma density
function remains valid [58].

3.1.2 Static Correlations

In order to provide a spatially dependent description of the field, the two-point correla-
tion function of the diffracted spatial field profile A(ρ, z) is evaluated, where ρ = (x, y)
is the transversal and z is the longitudinal coordinate. Given the correlation properties
of the field immediately after the random phase screen a(ρ), the correlation propagation
formalism (sec. 2.1.3) is used to obtain the correlations in the observation plane. Note
that this section treats static (ensemble-averaged) correlations Γ̃A and not temporally
averaged correlation functions ΓA , i.e. Γ̃A (ρ1 , ρ2 ) = E[A∗ (ρ1 , t)A(ρ2 , t)] is implied.

Field Correlations

For a coherent incident light field ain (ρ), the correlation function can be written as
a product of the deterministic field and a random phase delay φ due to the diffuser,
Γ̃a (ρ1 , ρ2 ) = a∗in (ρ1 )ain (ρ2 )µφ,φ (ρ1 , ρ2 ), where µφ,φ (ρ1 , ρ2 ) = ei(φ(ρ2 )−φ(ρ1 )) is the
(probability theoretical) joint characteristic function of φ. Note that since in this model
3.1 Theory 27

the phase delay results from surface height fluctuations, it has a chromatic dependence
φ(ρ) = k0 nh(ρ), with vacuum wavenumber k0 , refractive index of the diffuser material
n and its local surface height h(ρ) along the optical axis.
In order to evaluate the characteristic function, assume that the phase delays are
Gaussian distributed, with width σφ , and spatially wide-sense stationary, i.e. the cor-
relation of the phase delays can be written as Γ̃φ (∆ρ) = σφ2 γ̃φ (∆ρ), with correlation
coefficient γ̃ and difference coordinate ∆ρ = ρ2 − ρ1 . Then the characteristic function
becomes [58]
2
µφ (∆ρ) = e−σφ (1−γφ (∆ρ)) . (3.6)

If the incident field is a plane wave, resulting phase fluctuations are purely introduced
by the diffuser such that its (normalized) correlation function follows the characteristic
function, γ̃a → µφ . If the wavefront is curved, propagation does not change on average,
but scattering introduces an angular spread which increases divergence.
As the spatial dependence of µφ is only given by its phase delay correlations, the den-
sity of individual scattering grains on the diffuser determine the shape around the coher-
ence maximum. At large separations, when the phase delay correlations have decayed,
2
the field correlations exhibit an asymptotic normalized correlation value of µφ,res = e−σφ ,
so residual correlations are determined by how evenly the phase delays are distributed.
As the phase delays are proportional to the local diffuser surface height, the depth of the
grains determine the fraction of scattered light. Note that since the phase is periodic,
increasing σφ corresponds to an increasingly uniform distribution.

Intensity Correlations

Since recorded speckle patterns are intensity images, it is of interest to relate the in-
tensity correlation function to the field coherence. As the fields ai = a(ρi ) are circular
complex Gaussian random variables, Reed’s theorem [65] ha∗1 a∗2 a1 a2 i = ha∗1 a1 i ha∗2 a2 i +
ha∗1 a2 i ha1 a∗2 i is applicable such that the intensity correlation becomes

Γ̃I (ρ1 , ρ2 ) = I(ρ1 )I(ρ2 )(1 + |γ̃A (ρ1 , ρ2 )|2 ), (3.7)

in analogy to the Siegert relation (2.16) for chaotic light in the temporal domain.

Gaussian Schell Model for Speckle Field

Typically, the diffuser phase delay is assumed to have a Gaussian

 spatial correlation
 with
2 /r 2
width rc [58], giving the characteristic function µφ (∆ρ) = exp −σφ2 (1 − e−∆ρ c ) . For
28 3 Diffuser-Controlled Decoherence

1.0 ¾Á = ¼=3
¾Á = ¼=2
Field correlation after diffuser ¹Á

¾Á = ¼
0.8
¾Á = 2¼
¾Á = 3¼
0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Displacement r=rc

Figure 3.2: Field correlations immediately after a Gaussian diffuser.

The characteristic function µφ of a diffuser with Gaussian first-order (sur-
face roughness distribution width σφ ) and second-order statistics (correlation
length rc ) describes the field correlations γa immediately after the diffuser
for planar wavefront incidence. The dashed lines show the Gaussian approx-
imation to the exact solid lines. The short-range approximation fits well, but
the asymptotic residual coherence is not captured.
3.1 Theory 29

sufficient phase randomization, σφ & π (i.e. most incident light is scattered, fig. 3.2), it
√
can be approximated as Gaussian with width ρ̃c = rc / 2σφ [66]. For illumination with
a centred, monochromatic, Gaussian beam with waist w, the resulting light field thus
constitutes a Gaussian Schell model (2.18). Using the generalized van Cittert-Zernike
theorem (2.22), the correlation function in a Fraunhofer diffraction plane then becomes
 2
1 rc w 2 /2W 2 (z) 2 /2ρ2 (z)
Γ̃A (ρ, ∆ρ) = I0 e−ρ e−∆ρ c , (3.8)
2 σφ

where the first exponential term corresponds to the scattered intensity envelope and the
last term is the normalized correlation function. The respective Gaussian widths are
given by
2z z
ρc (z) = and W (z) = . (3.9)
kw k ρ̃c
Here ρc denotes the Gaussian width of the field, thus (3.7) the Gaussian width of the
√
intensity (“speckle size”) is xc = ρc / 2 and the coherence length, as defined in eq.
√
(2.12), yields sc = xc 2π.

The number of mutually uncorrelated areas (2.20) for a cutoff at the FWHM of the
√
intensity, W 2 ln 2, is given by
√
π W 2 (z) 1 W 2 (z) π (w/ 2)2
M (z) = = = = M0 , (3.10)
3 Ac (z) 6π x2c (z) 3 (2π ρ̃c )2

where M0 is the number of uncorrelated areas immediately after the diffuser. Thus,
although the speckle size xc (z) scales upon propagation, the speckle number remains
constant.

If the requirement of a deep random phase screen is not fulfilled, σφ . π, a significant

part of the light power is coherently transmitted and gives rise to an intensity spot much
smaller than the diffracted width W .

3.1.3 Dynamic Correlations

As the diffuser is used to reduce the coherence of a given light source, the spatial and
temporal coherence functions are calculated for the experimentally used modulation
schemes. Monochromatic Gaussian beam illumination is assumed as incident light field.
30 3 Diffuser-Controlled Decoherence

Moving Diffuser

Consider a static beam illuminating a diffuser which moves transversally along the path
d(t). As the effect of the diffuser is a random phase delay, the only modification required
of the static correlation function is to introduce a parametric time dependence of the
position ρ → ρ + d(t). Thus using the static diffraction result (3.8) and averaging over
time gives the spatiotemporal coherence function [66]

γA (∆ρ, τ ) = hγ̃A (∆ρ + d(t + τ ) − d(t))it . (3.11)

Assuming a linear motion d(t) = vt with periodicity d(t) = d(t + T ), one obtains
2 +(vτ )2 )/2ρ2
γA (∆ρ, τ ) = e−(∆ρ c . Thus the coherence time is given by the time required
to move the diffuser out of its correlation area and the spatial coherence length is given
by the speckle size. The residual coherence is limited on the diffuser side by its residual
2
phase delay correlations e−σφ and on the modulation side by the number of sampled
uncorrelated areas M = vT /sc (2.19) determined by the periodicity T , thus |γA,res | =
√ 2
C & 1/ M + e−σφ .
Note that a Fresnel correlation propagation analysis allows the estimation of the near-
field decorrelation time, which is obtained by replacing the speckle size xc with the
illumination beam size w [66].

Angled Incidence

Consider a Gaussian illuminating light field ain (ρ) which temporally varies its angle of in-
cidence, i.e. it introduces a linear phase gradient a(ρ, t) = ei(k0 sin(θ(t))·ρ+φ(ρ) cos θ(t) ain (ρ),
where θ = (θx , θy ) is the angle from the optical axis (the sine function is interpreted
vectorally) and the cosine term is added since the phase delay depends on the wavevector
in propagation direction (sec. 3.1.2). Similarly as before, a static correlation function
for the parameter θ is constructed which, in the uncorrelated limit rc → 0, yields [58]
2 2 (∆ρ/z+sin θ(t )−sin θ(t ))2 /2
γ̃A (∆ρ, t1 , t2 ) = e−σφ (cos θ(t1 )−cos θ(t2 )) e−(kw/2) 2 1
, (3.12)

where the characteristic function describes a chromatic decorrelation due to to a short-

ened wavevector along the optical axis (the same effect is achieved by normal incidence
with shorter wavelength). The spatially dependent exponential describes a translation
of the whole speckle pattern in eθ direction. Note that due to its k dependence, a
wavelength change additionally produces a similar but scaled speckle pattern.
For small angles θ  1 the characteristic function term is negligible and the sine
3.2 Setup 31

L75 L50 L75 RD

Figure 3.3: Rotating diffuser setup.

A variable magnification telescope prepares a collimated beam incident on a
rotating diffuser (RD).

functions can be linearised sin θ ≈ θ. Then the duality between ∆θ and ∆ρ/z becomes
apparent. For uniform linear angular modulation θ(t) = Ωt with modulation angular
frequency Ω, one thus obtains the same result as for the moving diffuser, γA (∆ρ, τ ) =
2 +(Ωτ z)2 )/2ρ2
e−(∆ρ c , where the residual coherence is limited by the phase delay distribution
and sampled correlation areas M = ΩT z/sc , with modulation period T .

3.2 Setup
In this section the diffuser setups used to generate spatially incoherent light and speckle
analysis methods are described.
The random diffusers studied here are of surface scattering type [58] and include
ground-glass diffusers (DGs) and holographic diffusers (DHs). The random surface
height fluctuations of DGs result from sand-blasting the non-polished side of a glass
plate and are specified by the particle size (grit). The surface reliefs on DHs are pro-
duced by holographically recording a speckle pattern which is used to form the surface
structure. They consist of polycarbonate and are directly specified by their FWHM
divergence.

3.2.1 Rotating Diffuser

The spatially decorrelating moving diffuser is implemented as a rotating diffuser with
the setup shown in fig. 3.3. For this, a hole is water-jet cut into the centre of a 50 mm
diameter Thorlabs N-BK7 220 grit (DG220) diffuser and the diffuser is mounted onto a
high rotational speed brushless motor.
The diffuser is illuminated by the 780 nm Toptica DL pro (NBL780) external cavity
diode laser (ECDL), where the beam hits the diffuser near its edge, thus the effective
diffuser motion is nearly linear with a speed of v = 2πf R, with rotation frequency f
and radial beam position R. In order to vary the correlation lengths, the beam is first
32 3 Diffuser-Controlled Decoherence

FMh AMh FMv AMv

DDS DDS DDS DDS

W2
AO
Dh
L100 W2 L100 AODv L100 L25 D Lc

Figure 3.4: Acousto-optic deflector setup.

An incoming linearly polarized monochromatic beam is rotated into horizon-
tal polarization by a half-waveplate (W2) to match the input polarization of
a horizontally deflecting AOD (AODh). The Bragg diffracted beam is im-
aged into a vertically oriented AOD (AODv) with a 4f-imaging system where
another half-waveplate (W2) again allows matching of the polarization. The
vertically diffracted beam (not drawn) is imaged onto the diffuser (D) with
another 4f-imaging system. The scattered light is collected by the final lens
(Lc). To drive each AOD, a DDS generated, frequency modulated (FM)
signal is mixed with an amplitude modulated (AM) envelope signal and is
power amplified. FM leads to a varied Bragg diffraction angle (light red
beam).

expanded by a telescope. A collimated incidence is used to keep the scattering angles

low and fulfil the planar wavefront approximation.

3.2.2 Beam Deflection

Decorrelation by diversity of angular incidence is achieved with the setup shown in

fig. 3.4. An AOD is used as angular modulator to the incoming 420 nm Toptica TA-
SHG (NBL420) laser beam, and collimated illumination of the same spatial spot is
realized using a 4f-imaging system. The diffusers used in this part are the holographic
diffusers Edmund Optics 80◦ FWHM divergence (DH80) and Edmund Optics 15◦ FWHM
divergence (DH15). Note that the figure shows two AODs and imaging systems to allow
modulation in both spatial dimensions.
AODs rely on Bragg reflection at a travelling acoustic wave to deflect an incident laser
beam by θ = 2θB where the first-order Bragg angle is given by [67]

λ λ
sin θB ≈ θB = = f, (3.13)
2Λa 2va
3.2 Setup 33

with optical wavelength λ, acoustic wavelength Λa and acoustic frequency f . Modulation

of the frequency thus allows scanning of the deflection angle. In contrast to acousto-
optic modulators (AOMs) using longitudinal waves, AODs use more slowly propagating
transversal shear modes in birefringent crystals [68] where the anisotropic acousto-optic
interaction requires a specific linear polarization of the input light. Since acoustic waves
have a finite velocity, modulation frequencies are limited by the acoustic propagation
time necessary to pass the optical beam, i.e. τa (w) = w/va for a Gaussian beam with
waist w.
In order to achieve fast and arbitrary frequency modulation (FM) signals, a direct
digital synthesis (DDS) board [69] is used which can digitally generate the FM signal.
The acoustic power dependence of the AOD diffraction efficiency is used to compensate
for its deflection angle dependence. For this, a feed-forward scheme is implemented, in
which another (synchronized) DDS produces an amplitude modulation (AM) envelope
signal which is mixed with the FM signal.
The AOD model used is a Crystal Technology AODF 4120-3 using the shear mode of
a TeO2 crystal with acoustic velocity va = 650 m/s at centre frequency fc = 120 MHz
with nominal bandwidth fBW = 50 MHz and aperture D = 3 mm. Using eq. (3.13) and
taking the demagnification M = 1/4 of the imaging system into account, the deflection
rate is θ/∆f = λ/va = 1.2 mrad/MHz. Note that the decorrelation performance is
independent of the demagnification since the increased angular incidence is cancelled by
the reduced beam size. Still demagnification is chosen as the light collection efficiency
depends on the product of beam width and divergence (sec. 2.1.3), and the scattering
based divergence is larger than the modulation angles.
The 400 MHz bandwidth Analog Devices AD9910 DDS in a home-built reference im-
plementation is used to generate the AM and FM signals, which are mixed and amplified
by a home-built driver. Note that the reference implementation includes a high-pass fil-
ter at the output which is removed for the AM output as the modulation frequency is
much smaller than the radio frequency (RF) carrier frequency.

3.2.3 Image Analysis

As the degree of coherence is directly related to the normalized intensity covariance (3.7),
spatial coherence properties can be measured by recording different speckle intensity
images on camera and performing image processing steps to deduce the quantities of
interest. The speckle patterns are recorded with the AlliedVision Manta G-145B NIR
charge-coupled device (CCD) camera.
34 3 Diffuser-Controlled Decoherence

raw filtered processed

Figure 3.5: Speckle pattern image processing.

The directly recorded (raw) speckle pattern often has a non-uniform intensity
envelope which can be estimated by applying a Gaussian filter. Normalizing
the raw image with the filtered image yields the processed speckle pattern
suitable for statistical speckle analysis.

In order to avoid speckle blurring due to finite-sized pixel areas, measurement distances
are appropriately chosen to resolve a typical speckle by at least a 3x3 pixel array. Then
the speckle contrast (sec. 3.1.1) deviation can be estimated to be already below 1 % [58].

Residual Coherence Estimation

Residual coherence properties can be obtained by their relation to speckle contrast (3.5)
and involve measuring the first-order speckle statistics, namely the probability density
distributions of intensity. Despite first-order statistics being technically defined for dif-
ferent realizations of speckle fields at a fixed observation position, the self-averaging
property of speckles allows the measurement of the spatial intensity distribution of a
single speckle pattern instead.
Imaging the whole speckle pattern creates the problem that the mean intensity is not
uniform, such that the intensity distribution is superposed by the intensity envelope. To
compensate this effect (fig. 3.5), each image is centred around its centroid and sym-
metrically resized. Imaging a large number of speckles ensures that speckle fluctuations
appear in a much higher spatial frequency regime than the intensity envelope, thus a
Gaussian filter with a smoothing radius much larger than the typical speckle size can be
applied to suppress low frequency variations. The resulting image (”processed speckle
pattern”) is obtained by normalizing the raw image by the filtered image, from which
the statistics can be taken.
For speckle contrast analysis, the mean and variance is directly determined from the
raw data. A comparison value arises from least-squares fitting the gamma distribution
(3.4) onto the probability density function. Then the effective number of incoherent
3.3 Static Speckles 35

superpositions of fully developed speckle patterns can be extracted, which yields the
contrast value.

Spatial Coherence Estimation

Since the (squared) modulus of the spatial coherence function is given by the static
speckle intensity autocovariance function (3.7), it can be constructed from a single static
speckle image. As the illuminating light spot has a Gaussian shape, the spatial coherence
function is also expected to be Gaussian. Thus a 2D least-squares Gaussian fit with offset
is performed on the processed speckle patterns, which yields the Gaussian intensity
correlation length xc (3.9).

Temporal Coherence Estimation

Temporal coherence can be estimated both statically and dynamically. As diffuser-based

decorrelation mechanisms temporally modulate a speckle-defining parameter (e.g. dif-
fuser displacement, sec. 3.1.3), a cross-correlation analysis with respect to this parameter
can be performed, i.e. for varying parameters (or equivalently at different time steps)
images of speckle patterns are recorded, and the correlation function between these pro-
cessed speckle patterns is calculated. In the case of the moving diffuser, the diffuser is
placed on a micrometre stage to capture speckle patterns at different lateral displace-
ments. In the case of angled incidence, the patterns are measured for different fixed
AOD RF frequencies.
To avoid the computationally expensive three-dimensional (3D) (two image and one
parameter dimensions) correlation calculations, the images are sliced along the modu-
lation axis and multiple slices with a spacing larger than the speckle size are selected.
Each of these 2D slices is autocorrelated, then the results are averaged and fitted with
a 2D tilted Gaussian.
Dynamically, the coherence time can be estimated by the decay of the speckle contrast
(3.5), as it is determined by the number of uncorrelated parameter (or time) steps. For
the rotating diffuser, this is achieved by varying the camera exposure time, and for the
angled incidence, the angular modulation range is adjusted.

3.3 Static Speckles

In this section the static correlations of speckle patterns are measured, which determine
coherence lengths and the intrinsic residual correlations of the diffuser.
36 3 Diffuser-Controlled Decoherence

DG220 DG600
10 10
6 6

Normalized intensity I=hIi

Normalized intensity I=hIi

8 8
5 5
Position y [mm]

Position y [mm]
4 6 4 6

3 3
4 4
2 2
2 2
1 1

0 0 0 0
0 2 4 6 0 2 4 6
Position x [mm] Position x [mm]

Figure 3.6: Speckle patterns generated by ground-glass diffusers.

Both images are recorded for the same illumination condition in the Fourier
plane of a collection lens. While the diffuser ground by 220 grit particles
(DG220) shows proper diffusion behaviour, a large power fraction is coher-
ently transmitted through the 600 grit diffuser (DG600), rendering it unsuit-
able for coherence reduction.

Diffuser Types

Fig. 3.6 shows the speckle patterns of two ground-glass diffusers with different grain
statistics at identical illumination conditions. The speckle sizes are similar, but whereas
the coarser diffuser produces a homogeneous speckle pattern, the finer one retains a
coherent, undiffracted beam. As argued in sec. 3.1.2, this results from a too smooth
surface height (first-order) distribution, insufficiently randomizing the phase, making it
unsuitable for coherence reduction. This can be attributed to the fabrication process
where smaller sand grains fail to carve deep structures into the glass, despite surface
height (second-order) correlations likely to be shorter than those of the coarser diffuser.
It is thus beneficial to use a holographic diffuser instead, as first-order fluctuations
can be tuned independently from the second-order fluctuations, which allows for smaller
scattering angles while still ensuring little residual correlations. This issue is in principle
also solved by Lambertian opaque glass diffusers. However, due to volume scattering
(multi-scattering within diffuser), its transmittance is low, resulting in power efficiencies
of typically ηopaque ∼ 30 %.
In contrast, the efficiencies of both the holographic DH15 diffuser ηDH15 ∼ 95 % and
the ground-glass DG220 diffuser ηDG220 ∼ 75 % are measured to be significantly higher.
The FWHM divergence of the DG220 is given by θDG220 ∼ 16°.
3.3 Static Speckles 37

35
DH80
DG220
30

25
Speckle size xc [µm]

20

15

10

200 400 600 800 1000 1200 1400 1600 1800

Gaussian beam waist w [µm]

Figure 3.7: Speckle size dependence on beam size.

The speckle size xc (3.9) depends inversely on the illuminating beam width
w. In accordance with the van Cittert-Zernike theorem (3.8) plotted in grey,
sufficiently random diffusers show the same correlation behaviour in Fourier
plane (measured at a distance of z = 30 mm for the wavelength λ = 780 nm),
as verified by a 220 grit ground-glass diffuser (DG220) and a 80° FWHM
divergence holographic diffuser (DH80).

Speckle Size Dependence On Illumination Area

In order to evaluate the speckle size dependence on the illumination area, the speckle
size is measured for different incident beam sizes on various diffuser plates and plotted
in fig. 3.7.

The speckle patterns are recorded in the Fourier plane of a f = 30 mm lens with a
numerical aperture chosen not to cut off the Gaussian intensity envelope. This ensures
high collection efficiency and speckle sizes should only depend on illumination conditions
and not on any imaging optics. Applying eq. (3.6) agrees well with the data. Since
the correlation length determines the coherence length when modulating the diffuser to
obtain incoherent light, this indicates that the illumination area provides control over
the required spatial coherence length.
38 3 Diffuser-Controlled Decoherence

DH80
35
Displacement correlation length dc [µm]

30

25

20

15

10

200 300 400 500 600 700 800 900

Gaussian beam waist w [µm]

Figure 3.8: Displacement correlation length dependence on beam width.

The Gaussian displacement correlation length dc (proportional to coherence
time) falls off inversely with illuminating beam width w, similar as the spatial
speckle correlation function, indicating a translational shift of the speckle
pattern as main decorrelation mechanism. The data is measured for the 80°
FWHM divergence holographic diffuser (DH80) at a distance of z = 30 mm
from the diffuser without lens.

3.4 Modulation by Diffuser Rotation

The first implementation of coherence reduction uses a rotating diffuser (sec. 3.2.1) to
integrate a series of speckle patterns. The coherence properties of the resulting light
field is described in this section.

Temporal Coherence

In order to obtain the short-term decorrelation characteristics, i.e. the coherence time,
a static speckle pattern cross-correlation analysis (sec. 3.2.3) is performed, with the
horizontal (transversal) displacement of the DH80 diffuser as parameter. By fitting a
Gaussian along this displacement correlation function, one obtains a correlation length
dc which can be associated with a correlation time tc = dc /v for linear motion.
Fig. 3.8 thus effectively shows the decrease of the temporal coherence length and
agrees with the expected curve. Deviations are likely introduced by finite displacement
precision resulting in limited resolution and thus an uncertain Gaussian fit. Since the
3.4 Modulation by Diffuser Rotation 39

DG220
100

Speckle contrast C

10-1

0 1 2 3 4 5 6
Exposure time t [ms]

Figure 3.9: Residual coherence convergence of rotating diffuser.

The logarithmic plot shows the speckle contrast (i.e. the residual coherence)
decrease when averaging over increasing numbers of uncorrelated speckle
patterns. The data is obtained for a 220 grit ground-glass diffuser (DG220)
rotating with a period of T = 4.2 ms where the contrast decrease ceases. The
speckle size is measured as xc = 18 µm, recorded at a distance of z = 55 mm.

displacement correlation length behaves identically as the spatial correlation length,

particularly having a z-position dependent scaling (3.9), decorrelation performance varies
upon the position of the diffraction plane.

Residual Coherence

The capabilities of the rotating diffuser approach to reduce residual coherence are tested
with a motor rotation period of T = 4.2 ms and an outer radial position of the illu-
minating spot at R = 20 mm maximizing the diffuser velocity v = 2πR/T ∼ 30 m/s.
To estimate the expected decorrelation time tc , the static speckle size is measured as
xc = 18 µm ∼ vtc . The observation distance of z = 55 mm from the diffuser is deduced
from static measurements using the linear relation between distance and speckle size.
The number of mutually uncorrelated modes (2.20) is thus M ∼ 842 .
Fig. 3.9 shows an initially fully developed speckle contrast which decreases for longer
exposure times since more uncorrelated areas are averaged, limited due to the rotational
periodicity to Cmin ∼ 0.033 giving a coherence time of τc ∼ 4.3 µs. The data agrees with
40 3 Diffuser-Controlled Decoherence

Figure 3.10: USAF resolution target imaged with different light sources.
The images with a bar line width of 15 µm are recorded under illumination
with a narrowband laser, a static speckle pattern, rotating diffuser light
and an light-emitting diode (LED) (left to right). The coherent noise gen-
erated by the dirty imaging optics is clearly reduced when using the rotating
diffuser light in comparison to the coherent laser.

the model, where the decrease rate is obtained from above estimation and a constant
offset C0 = 0.02 is fitted to account for coherent transmission through the diffuser.

Intensity Noise

Since the temporal coherence function is determined by the spatial correlation func-
tion of the random speckle field (sec. 3.1.3), the Siegert relation (2.16) can be used
to derive its intrinsic intensity noise spectrum yielding a theoretical RIN of sI (ω) =
√ 2 2 2
2πxc e−xc ω /2v /v. In the low frequency limit using the parameters from the previ-
ous section this results in sI ∼ −60 dBc/Hz, thus performing much worse compared to
free-running lasers.
Global (total beam width) intensity noise measurements show that the fast spinning
motor furthermore introduces strong vibrations due to an unbalanced diffuser plate (off-
axis mounting hole with excessive tolerances) leading to diffuser plate tilting. This
modulates the light intensity in the rotation frequency regime at a magnitude of up to
−15 dBc/Hz, thus requiring multi-period averaging.

Imaging

In order to qualitatively demonstrate coherent noise reduction of spatially incoherent

light, images of a resolution test target are recorded with different focused light sources
through a dirty microscope objective (fig. 3.10). The target is back-illuminated by a
coherent laser, a static speckle pattern, a rotating diffuser and a LED, serving as optimal,
incoherent reference. Whereas the coherent laser produces fine interference rings that
3.5 Modulation by Angular Incidence 41

lead into the dark area, the incoherent LED does not produce any coherent artefacts.
The same is valid for the light produced by the rotating diffuser where the large scale
interferences are blurred and thus do not contribute to noise. The remaining stripes
originate from the finite illumination contrast and are a consequence of the diffuser
motion.

Conclusion

Rotating diffusers present a method to generate partially coherent, pseudo-thermal light

with tunable coherence properties. Precision-manufactured diffuser-motor combinations
could in principle also overcome currently present vibration issues.
However, decorrelation times are limited by the rotation speed of the (already fast-
spinning) motor. Considering that, when increasing the rotation frequency by an order
of magnitude, the diffuser already moves near sonic speed renders the moving diffuser
approach impractical for high-speed incoherent light generation.

3.5 Modulation by Angular Incidence

In order to overcome temporal averaging limitations due to the physical motion of the
diffuser, an angled incidence approach (sec. 3.1.3) is suitable as in this case only the
beam itself has to be modulated.
The deflection scheme is thus required to provide fast modulation of a large angular
deflection bandwidth. For this an AOD is tested with which the beam deflection can be
realized much faster than with mechanical means like micro-electro-mechanical systems
(MEMS).

One-Dimensional Angular Scan

To verify the picture that the angular decorrelation of speckle patterns in a Fraunhofer
plane mainly originates from pattern translation (sec. 3.1.3), a cross-correlation analy-
sis with one-dimensional (1D) varying angles of incidence is performed, confirming the
model (similar to fig. 3.8).
For a residual coherence measurement, the AOD is linearly frequency modulated at
fFM = 10 kHz. Since for fast modulation the camera exposure time is too long to observe
coherence reduction on different time scales, the fact that the modulation is periodic is
used to instead vary the scanning frequency range while keeping the exposure time at
its multiple texp = 10/fFM .
42 3 Diffuser-Controlled Decoherence

100 DH15
DH80
Speckle contrast C

10-1

0 25 50 75 100 125 150 175

Angular bandwidth ¢µ [mrad]

Figure 3.11: Residual coherence convergence due to angular diversity.

The logarithmic plot shows the speckle contrast due to AOD-based angular
incidence modulation at fFM = 10 kHz for holographic diffusers with 15°
(DH15) and 80° (DH80) FWHM divergence. Since a larger angular mod-
ulation bandwidth comes with averaging over more uncorrelated speckle
patterns, the contrast decreases, in accordance with the translational pat-
tern shifting model obtained for a speckle size of xc = 40 µm in the Fourier
plane of a f = 75 mm lens (solid lines).
3.5 Modulation by Angular Incidence 43

Ratio 2/3 Ratio 1/10

Vertical angle µy [mrad]

100

50

0 100 0 100
Horizontal angle µx [mrad] Horizontal angle µx [mrad]

Figure 3.12: Image of the two-dimensional angular scanning pattern.

The plots show beam images recorded without the diffuser (thus the inci-
dence angle is mapped to the spatial position of the camera sensor) with
sufficiently long exposure times to trace the scan pattern with a ratio rx /ry
between horizontal and vertical FM frequency. Variations in intensity result
from inhomogeneous AOD diffraction efficiencies at different input angles
and variations in the angular bandwidth are a consequence of astigmatism
due to relative positioning errors of AODs and diffuser.

Fig. 3.11 is obtained for a static speckle size of xc = 40 µm in the Fourier plane of
a f = 75 mm lens for two holographic diffusers. The number of mutually uncorrelated
modes (2.20) for the DH15 (DH80) diffuser is thus M ∼ 482 (M ∼ 3082 ). Again the
decorrelation with modulation bandwidth follows the predicted reduction rate, where
the discrepancy between the diffusers results from the coherent transmission part and is
fitted to C0 = 0.01 (C0 = 0.03). The achievable minimal speckle contrasts at C ∼ 0.1
show that the small deflection ranges of AODs (compared to mechanical deflectors)
severely limit the number of averaged uncorrelated speckle patterns.

Two-Dimensional Angular Scan

An effectively larger angular bandwidth ∆θ is obtained using a second, perpendicularly

deflecting AOD allowing incidence from a solid angle, i.e. the number of decorrelated
patterns determining the residual coherence becomes 1/C 2 ∝ θc /∆θ → (θc /∆θ)2 with
the linear decorrelation angle θc (sec. 3.1.3). However, since the solid angle has to be
rasterised the required linear modulation speed would also have to increase by the same
factor.
Therefore instead a Lissajous curve is modulated where the AOD signal of each di-
mension i is linearly swept up and down with frequency fi,FM = ri fFM (fig. 3.12)
44 3 Diffuser-Controlled Decoherence

0.10 Ratio 2/3

Ratio 1/10

0.09
Speckle contrast C

0.08

0.07

0.06

0.05
0 20 40 60 80 100
Fundamental FM frequency fFM [kHz]

Figure 3.13: Residual coherence dependence on modulation speed.

The speckle contrast (i.e. the residual coherence) increases when the FM
frequency approaches the AOD response time of 1/τ ∼ 450 kHz in either
deflection direction (modulation speed ratio rx /ry ). The solid line shows
the expected contrast for the DH15 diffuser without taking into account
modulation speeds, measured for a speckle size of xc = 40 µm in the Fourier
plane of a f = 75 mm lens.

where fFM is the fundamental (2D) modulation frequency and ri is the corresponding
multiple.
q In the linearised version of eq. (3.12) the 2D incidence angle is given by
θ = θx + θy2 , thus a geometric consideration yields an effective AOD FM bandwidth
2
q
of fFM, eff = 2fFM rx2 + ry2 . The factor of two is introduced to include the up- and
down-ramp when there is a finite phase between the horizontal and vertical modulation.
Fig. 3.13 shows how the speckle contrast evolves when the FM speed is increased,
measured with the DH15 and the same illumination conditions as in fig. 3.11. The
camera exposure time is chosen to match the fundamental frequency. In the lower
kHz regime the contrast behaves as expected, where the model takes into account the
previously estimated offset correlation C0 = 0.03 and the reduced vertical scan range
(fig. 3.12). For the scan frequency ratio 2/3 the contrast strongly increases from fFM &
50 kHz, whereas in the 1/10 case a slower increase is visible starting already around
fFM ∼ 10 kHz.
The increase can be attributed to the finite response time of the AOD determined
by the finite travelling time of the acoustic wave crossing the beam τ = w/va ∼ 2.2 µs.
3.5 Modulation by Angular Incidence 45

At these time scales the beam is not continuously deflected but rather simultaneously
deflected resulting in simultaneous illumination of the diffuser from a range of angles
reducing the speckle suppression capabilities. For the same fundamental frequency the
vertical modulation of the ratio 1/10 case reaches this regime earlier than in the 2/3
case, thus its increase begins first. As the horizontal scan is slower in the 1/10 case, the
contrast increases more slowly.

Intensity Noise and Efficiency

An intensity noise estimation in analogy to the rotating diffuser approach (sec. 3.4)
gives a theoretical intensity noise value of S ∼ −90 dBc/Hz in the low-frequency limit
for an effective modulation speed of fFM, eff = 100 kHz.
While using the AOD in one dimension resulted in diffraction efficiencies of ηAOD ∼
65 % across the deflection bandwidth (specification: ηAOD & 75 %), tilted incidence into
the second, orthogonal AOD strongly reduced diffraction efficiency, giving a combined
power efficiency below η ∼ 20 %. Accounting for feed-forward stabilization losses and the
finite NA of the collection lens after the diffuser resulted in an overall light utilization of
only η ∼ 10 %. Note that a light collection cutoff at the intensity FWHM of a Gaussian
results in a collection efficiency of ηFWHM = 50 %, such that the theoretical maximum
2
is η ∼ ηAOD ηFWHM ∼ 28 %.

Conclusion

As a diffuser-based coherence reduction approach, the light generated by AOD-scanned

angular diffuser incidence has similar properties as the one from the rotating diffuser
approach.
In comparison, significantly higher modulation speeds in the 105 Hz regime can be
achieved. This comes at the expense that less independent speckle patterns can be
averaged (i.e. higher residual coherence) since the deflection ranges of AODs are limited,
and that light utilization is reduced.
If power efficiency and low temporal intensity noise is not the main concern and
quasi-monochromatic light is required, this approach can be used for spatial coherence
reduction. However, for tight optical confinement of atoms, trapping frequencies easily
reach the achievable modulation frequency, requiring yet another coherence reduction
approach.
4 Fibre-Controlled Decoherence

Active modulation techniques reducing spatial coherence in a light field rely on deter-
ministically shaping the wavefront of a possibly coherent beam in such a way that spatial
correlations are minimized. Despite having in principle full control over the decoherence
properties, this approach is limited by the speed of the available modulation techniques.
Since the characteristic times for spatial decoherence are determined by the coherence
time (2.19), an opposite approach is to start from an intrinsically fast fluctuating light
field. Applying dispersive spatiotemporal coupling allows the conversion of incoherence
in the time domain into incoherence in the spatial domain. If an application does not
require quasi-monochromaticity, broadband light sources enable essentially arbitrarily
fast decoherence.
In sec. 4.1 the coherence properties of optical fibres are derived in scalar wave theory,
particularly taking into account the effects of different core geometries. The experimental
setup is shown in sec. 4.2, which includes a description of the studied light sources and
the interferometric coherence measurement approach. In the subsequent sections the
results for temporal coherence (sec. 4.3), spatial coherence (sec. 4.4) and intensity noise
(sec. 4.5) can be found.

Coherence Conversion in Multimode Fibres

The idea behind temporal coherence conversion is to induce a spatially dependent prop-
agation time delay. If the relative delay between two such propagation modes exceeds
the coherence time of the light source, they mutually dephase which leads to coherence
reduction. Since the modes correspond to distinct positions in transverse space, the
modal decoherence also translates into spatial decoherence.
Multimode fibres are optical fibres with large core sizes with respect to the trans-
mitted wavelength, such that they can carry many distinct spatial modes with varying
propagation speeds. This phenomenon of modal dispersion can be intuitively understood
by applying a ray optics picture drawn in fig. 4.1. If the light beam incident on the
fibre is divergent, it consists of rays propagating at various angles. As rays with higher
incidence angle travel a longer optical path, their propagation time is longer which leads
48 4 Fibre-Controlled Decoherence

multimode optical fibre modal dispersion

Figure 4.1: Coherence conversion in multimode fibres.

Light rays travelling with larger angles of incidence cover a longer optical
path, therefore their longitudinal propagation speed is reduced. If their dif-
ferential time delay after transmission through the fibre exceeds the temporal
coherence time of the incoupled light, the field becomes decorrelated and the
output face of the fibre behaves as a spatially incoherent light source.

to the desired differential time delay. When using broadband light sources, additionally
chromatic dispersion effects arise. Its main contribution is material dispersion due to the
fibre core material and has a waveguide dispersion correction due to evanescent waves
travelling in the optically thinner fibre cladding. As typical multimode fibres use silica
as material (which has low dispersion) and large cores confine the light within the core,
modal dispersion is by far the dominant effect in step-index fibres.

4.1 Theory

In this section the influence of multimode optical fibres on the temporal and spatial
coherence properties of an incoupled broadband light beam is derived, showing coherence
reduction.

Fibre Propagation

Neglecting bendings and imperfections, optical fibres can be regarded as longitudinally

invariant waveguides with refraction index n(ρ, ω) where ρ = (x, y) is the transversal, z
is the longitudinal position, and ω is the angular frequency of the light field. Applying
a separation of variables ansatz of the light field Ã(r, ω) = A(ρ)Az (z)A(ω) on the
Helmholtz equation (2.2) decouples the longitudinal term into a harmonic oscillator
with longitudinal propagation constant β and a transversal Helmholtz equation [67]

∇2ρ A(ρ) + (k 2 (ρ, ω) − β 2 (ω))A(ρ) = 0, (4.1)

4.1 Theory 49

with local wavenumber k(ρ, ω) = n(ρ, ω)k0 (ω) and vacuum wavenumber k0 (ω) = ω/c.
This eigenvalue equation describes the propagation of transversal fields, which has a
finite number of solutions (“propagation modes” |mi with eigenvalue km and field profiles
ψm (ρ) = hρ|mi) since only guided (i.e. totally internally reflected) modes are considered.
Because the longitudinal propagation speeds are of main interest, the propagation
wavenumber
2
βm (ω) = k 2 (ω) − km
2
(ω) (4.2)

is considered as a dependent variable and yields the dispersion relation. The explicit
dependence on m describes modal dispersion, and chromatic dispersion is identified as
material dispersion n(ω) and waveguide dispersion km (ω).
The propagation function hL (2.4), relating an incident field a(ρ) at z = 0 with the
propagated field A(ρ) at the fibre output z = L, can thus be expanded in terms of the
propagation modes as
X
A(ρ, ω) = hm|ai ψm (ρ)e−iβm (ω)L , (4.3)
m

∗ (ρ)a(ρ)d2 ρ denotes the input field overlap integral.

R
where hm|ai = ψm
Assuming the field to be initially spatially coherent, one can factorize its spatial and
temporal part as a(ρ, t) = a(ρ)at (t) such that the space-time correlation function (2.10)
becomes Γa (ρ1 , ρ2 , τ ) = ha∗ (ρ1 )a(ρ2 )i ha∗t (t)at (t + τ )i. Inserting the mode expansion
yields the output correlation function
Z
dω iωτ X
∗
ΓA (ρ1 , ρ2 , τ ) = e sa (ω) ha|mi hn|ai ψm (ρ1 , ω)ψn (ρ2 , ω)ei(βm (ω)−βn (ω))L ,
2π m,n
(4.4)
where sa (ω) is the normalized (statistical) optical power spectral density (2.13).
If the relative spectral bandwidth (2.14) is small, ∆ν  ν, and the fibre is highly
multimode (as studied here), one can approximate the mode profiles as spectrally inde-
pendent ψm (ρ, ω) ≈ ψm (ρ). Note that a similar derivation can be found in [70].

Spatial Coherence Function

In order to determine spatial coherence, consider the equal-time correlation function at

the fibre output, ΓA (ρ1 , ρ2 ) = ΓA (ρ1 , ρ2 , τ = 0), with which the intuitive picture of
decorrelating propagating light rays can be formalized.
In the limit of vanishing modal dispersion, km = kn , the phase evolution in eq. (4.4)
50 4 Fibre-Controlled Decoherence

cancels and the frequency integral evaluates to unity, indicating no coherence reduction,
γA (ρ1 , ρ2 ) = 1. This corresponds to common cases like free propagation in homogeneous
media or single-mode fibres.
In the opposite case of vanishing chromatic dispersion, modal dispersion remains the
only effect. For a spectrum centred at frequency ω0 the dispersion relation (4.2) can be
expanded as βm (ω) ≈ βm (ω0 ) + (ω − ω0 )/vm (ω0 ), where the group velocity
s
c βm (ω0 ) c 2
km
vm (ω0 ) = = 1− 2
(4.5)
n k(ω0 ) n k (ω 0)

is identified. The frequency dependent part of the exponent in the correlation function
(4.4) thus reduces to ω∆tmn , with relative modal time delays ∆tmn = L/vm − L/vn .
Then the frequency integral Fourier transforms the optical spectral density into the
initial temporal coherence function γa (2.13), such that the spatial correlation function
becomes [70]
X
∗
ΓA (ρ1 , ρ2 ) = ψm (ρ1 )ψn (ρ2 ) ha|mi hn|ai γa (∆tmn )ei(∆βmn (ω0 )L−ω0 ∆tmn ) , (4.6)
m,n

with propagation constant difference ∆βmn (ω) = βm (ω) − βn (ω).

This equation can be understood by first considering one position on the fibre output
face, ρ1 , illuminated by only a single mode. Let a second point ρ2 be equally illuminated
√
by a multitude of modes, ψn (ρ2 ) = 1/ M , where the weight of each mode is given by
its occupation, i.e. the overlap hn|ai. The total mutual coherence between the two
points is then given by the mutual coherence of each mode, weighted by its occupation.
Since each mode is a time-delayed copy of each other, the coherence in question is the
original temporal coherence. In order to allow the field at ρ2 to have a non-equal mode
composition, the overlap ψn (ρ2 ) is introduced, which weights the effect of each mode.
By also allowing arbitrary mode compositions for ρ1 , one finally obtains eq. (4.6).

Residual Spatial Coherence

In order to estimate the degree of coherence without knowledge of the mode structure,
the average over all pairs of points in the output plane is considered. Since no specific
correlations at the output plane are assumed, the overlaps ψm (ρ) can be seen as indepen-
dent identically distributed random variables (i.e. they have common statistics and are
uncorrelated with respect to mode m and position ρ), such that their cross-correlation
∗ (ρ )ψ
factorizes hψm ∗
1 n6=m (ρ2 )i = hψm (ρ1 )i hψn6=m (ρ2 )i.
4.1 Theory 51

Furthermore note that the mode decompositions ψm are normalized by definition,

thus the average mode occupation is |ψm |2 = 1/M , where M is the number of fibre
modes. As expectation values of phase-uniformly distributed complex numbers vanish,
hψi = 0, the complex spatial correlation function also vanishes, hΓA (ρ1 , ρ2 )i = 0. Its
modulus however is finite but not linear in the cross-correlation terms. Instead its
p
upper bound given by Jensen’s inequality [71] h|ΓA |i ≤ h|ΓA |2 i is evaluated, which
involves calculating a four-point expectation value. Assuming Gaussian statistics, Reed’s
∗ ψ ψ ∗ ψ i = |ψ |2
theorem [65] can be applied to obtain hψm |ψn |2 δmk δnl , such that
n k l m
the residual correlation (2.19) becomes

M
1 X
|ΓA,res |2 ≡ |ΓA (ρ1 , ρ2 )|2 = | hm|ai |2 | hn|ai |2 |γa (∆tmn )|2 , (4.7)
M2
m,n=1

where | hm|ai |2 describes the occupation of mode m. With the expected value of the
PM 2
intensity hIA (ρ)i = m=1 | hm|ai | /M , one can define the residual spatial degree of
coherence |γA,res | = |ΓA,res |/ hIA i.
Eq. (4.7) reproduces the idea (sec. 4) that mutually time-delayed modes ∆tmn dephase
according to the decaying temporal coherence function |γa |. As the whole occupation
correlation matrix is summed in the formula, best coherence reduction is achieved by
exciting modes which maximize the mean differential mode delays. Under the require-
ment of a large number of excited modes, homogeneous fibre launch conditions (exciting
modes of different group velocities equally) are therefore favoured.
Instead of summing the discrete modes, it is often convenient to express them in terms
of their time delays ∆tm . Transforming into a time-delay distribution p(∆t) thus yields
ZZ
|γA,res |2 = p(∆t1 )p(∆t2 )|γa (∆t1 − ∆t2 )|2 d∆t1 d∆t2 . (4.8)

Note that p has the properties of a probability density function to ensure normalization.

Spatial Coherence Function in Incoherent Limit

As in the previous section the output positions are averaged, no statement can be made
about the shape of the spatial coherence function. If again no assumptions on the mode
structures are made, one can provide a simple estimate due to [72] in the fully incoherent
limit.
Consider the central point in the plane of interest and let it be incoherently and
uniformly illuminated with light incident up to an angle α. In free space propagation,
52 4 Fibre-Controlled Decoherence

free-space
fibre NA
equivalent
source

Figure 4.2: Fibre equivalent of the van Cittert-Zernike theorem.

The guiding condition of the optical fibre, i.e. its numerical aperture (NA),
limits the angular spread of light illuminating a point at the fibre end face
(on the right hand side). By extrapolating this angle, one can identify an
equivalent free-space source, to which the van Cittert-Zernike theorem (2.22)
can be applied to obtain the spatial coherence function.

one can identify the incoming light with an incoherent source filling the corresponding
solid angle (fig. 4.2), such that the van Cittert-Zernike theorem (2.22) gives the spatial
coherence function. Since in the case of a fibre the angle is limited by the guiding
condition, one can replace tan α ≈ NA to obtain

J1 (k0 ∆ρNA)
γA (∆ρ) ≈ 2 (4.9)
k0 ∆ρNA

for a circular light source with vacuum wavenumber k0 , the NA of the fibre and a radial
displacement ∆ρ. Integration of eq. (2.12) yields a coherence length of sc = 32/3πk0 NA.

Temporal Coherence

For determining temporal coherence, consider the equal-space correlation function at the
fibre output, ΓA (τ ) = ΓA (ρ, ρ, τ ).
Consider again the two cases discussed before. For vanishing modal dispersion the
spatial terms lose frequency dependence, thus the temporal coherence function remains
invariant γA (τ ) = γa (τ ). For vanishing chromatic dispersion and linear modal dispersion,
the correlation function yields
X
∗
ΓA (τ ) = ha|mi hn|ai ψm (ρ)ψn (ρ)γa (τ + ∆tmn )ei(∆βmn (ω0 )L−ω0 ∆tmn ) , (4.10)
m,n

with symbols as defined for eq. (4.6).

Using the same statistical assumptions as the ones used to derive the residual co-
∗ ψ i = δ
herence, especially hψm n mn /M , the averaged temporal coherence function also
reduces to the input function hγA (τ )i = γa (τ ), so on average the temporal characteris-
tics of the light source do not change.
4.1 Theory 53

4.1.1 Step-Index Fibres

Step-index optical fibres consist of a light-guiding core with homogeneous refractive

index n1 and size a, surrounded by a cladding with homogeneous but optically less
dense refractive index n2 , which allows for total internal refraction. The cladding is
chosen thick enough to fully confine the evanescent waves of the guided modes. From
the condition for total internal reflection, the fibre numerical aperture is defined as
p
NA = n21 − n22 .
For the typically small values NA  1, the fibre is in the weak guiding regime,
i.e. it obeys the paraxial approximation and the treatment in sec. 4.1 is applicable.
The polarization-degenerate solutions of the transversal Helmholtz equation (4.1) are
accordingly called linear polarized (LP) modes.

Round Core

For round-core step-index fibres with core radius a, cylindrical symmetry suggests a
separation ansatz decoupling the azimuthal dimension ϕ from the radial one ρ. The
propagation solutions are thus given by A(r) = Aρ (ρ)e−ilϕ e−iβz with azimuthal mode
number (and eigenvalue) l. The radial mode profile is given by [67]

J (k ρ) ρ ≤ a
l lm
Aρ (ρ) ∝ (4.11)
K (κ ρ) ρ > a
l lm

with the core-cladding relation NA2 k02 = klm

2 + κ2 , the lth Bessel function of the first
lm
kind Jl , the lth modified Bessel function of the second kind Kl and the radial mode
number m with transversal eigenvalue klm . The boundary conditions lead to the char-
acteristic equation
Jl±1 (klm a) Kl±1 (κlm a)
klm = ±κlm (4.12)
Jl (klm a) Kl (κlm a)
whose discrete solutions give rise to the propagation modes. Approximating the solutions
by the roots of the Bessel functions yields klm = (|l| + 2m)π/2a. As the field has to fulfil
the guiding condition klm < NAk0 , i.e. free in the core but decaying in the cladding,
the equation has a limited number of solutions, determining the total (polarization-
degenerate) number of modes M = 2(k0 aNA/π)2 [67].
Fig. 4.3 shows some LPlm propagation eigenmodes, where the real part of the complex
mode profiles are shown. Higher order modes exhibit oscillations, which indicate a
tilted wavefront. Thus one can identify growing mode numbers m as increasingly angled
54 4 Fibre-Controlled Decoherence

LP 01 LP 61 LP 06 LP 66

Figure 4.3: LP modes of round-core step-index fibres.

The plots show the real part of the polarization-degenerate linear polarized
LPlm modes with azimuthal mode number l and radial mode number m.
Angled light rays correspond to oscillating fields whose frequencies grow with
mode number. Its absence in LP01 identifies it as the straightly propagating
fundamental mode, whereas e.g. the ring-shaped LP61 mode corresponds to
skewed beams.

propagating light rays, l = 0 as meridional rays and growing |l| as increasingly skewed
rays. Particularly note that high |l| modes tend to depopulate the fibre centre. As
the group velocity reduces with increasing mode number, this results in a spatially
inhomogeneous velocity distribution, affecting the fibre induced decorrelation (as it relies
on variety of differential modal delays, i.e. differences of the inverse group velocities).

Square Core

For square-core step-index fibres with core side length 2a, no separation of variables
is possible since the refractive index profile does not factorize. Instead the separable
problem with n1  n2 (neglecting the evanescent part of the mode in the cladding,
2 +k 2 <
analogously to the infinite quantum well) is solved under the guiding condition km n
NA2 k02 . Here the transversal profile is written as A(ρ) = Ax (x)Ay (y), which leads to
decoupled harmonic oscillators with modes |mi and |ni. The mode profile in the x
dimension is thus given by

sin(k x)
m m ∈ 2Z
Ax (x) ∝ (4.13)
cos(k x)
m m ∈ 2Z + 1

π
with discrete solutions km = 2a m. As the y dimension behaves analogously the com-
2
bined, transversal eigenvalue is kmn = km2 + k 2 and the total (polarization-degenerate)
n
number of modes M = (k0 aNA)2 /π.
In contrast to the modes of the round-core fibres, the square-core LPmn modes, some
4.1 Theory 55

LP 11 LP 61 LP 16 LP 66

Figure 4.4: LP modes of square-core step-index fibres.

The plots show the real part of the LPmn modes with mode numbers m and
n corresponding to the x and y dimensions. LP11 represents the fundamen-
tal mode and higher order modes correspond to light ray propagation with
steeper angles. In contrast to the round-core modes, the square-core modes
are evenly distributed across the fibre core.

shown in fig. 4.3, do not have a pronounced area with propagation speed separation and
light power is more evenly distributed.

Differential Mode Delay and Residual Coherence

In telecommunication applications it is common to describe modal dispersion in terms of

the differential mode delay τf = ∆tmax − ∆tmin , i.e. the delay difference between fastest
and slowest mode. In order to assess the fibre decoherence properties, a histogram of
the time delay per unit length ∆tm /L = 1/vm with the modes as ensemble is evaluated
(fig. 4.5). The parameters are chosen to reflect the experimentally studied round-core
(square-core) fibres with core radius a = 62.5 µm (a = 75 µm) and numerical aperture
NA = 0.22 (NA = 0.39) at the central wavelength λ = 780 nm. The differential mode
delays are τf /L = 38 ps/m (τf /L = 118 ps/m).
Fig. 4.5 also shows that the modal delay distribution is well-approximated by a
uniform distribution p(0 ≤ ∆t ≤ τf ) = 1/τf . Evaluating eq. (4.8) thus gives a simple
relation for the residual coherence

1 τf
 
|∆t|
Z
2 τc
|γA,res | = 1− |γa (∆t)|2 d∆t → (4.14)
τf −τf τf τf

|γa (τ )|2 dτ and where the limit

R
by identifying the temporal coherence length τc =
(which is also an upper bound) is valid if |γa (τ )|2 has substantially decayed on the scale
of τf .
Additionally, this continuum approximation does not take into account the finite mode
56 4 Fibre-Controlled Decoherence

200
MFR
Mode density Mp [(ps/m)-1]

MFS
150

100

50

0
0 20 40 60 80 100 120
Modal delay per unit length ¢t=L [ps/m]

Figure 4.5: Distribution of modal delays.

The graphs show the numerically calculated (4.12, 4.13) modal histogram
of the differential modal delay per unit length for a round-core step-index
fibre (MFR) with core radius a = 80λ, numerical aperture NA = 0.22, and a
square-core step-index fibre (MFS) with core half-side length a = 96λ, NA =
0.39. The total number of modes are given by M ∼ 2150 and M ∼ 17500,
respectively, and p denotes the normalized distribution. The larger NA of
the MFS allows for larger propagation angles (and thus larger differential
mode delays) than the MFR, and additionally its larger core size contributes
to a larger number of propagation modes.
4.1 Theory 57

Spectral density SA single-mode multimode mode-locked spontaneous

Wavelength ¸ Wavelength ¸ Wavelength ¸ Wavelength ¸

Figure 4.6: Spectral properties of light sources.

The graphs depict typical optical spectra of single-mode, multimode, fem-
tosecond (mode-locked) lasers and amplified spontaneous emission sources.
Single-mode and mode-locked lasers can produce very narrow linewidths,
while multimode laser linewidths are broader since individual modes can fluc-
tuate independently. Spontaneous emission sources show continuous random
spectra without discrete mode structures.

number. Since coherence reduction results from statistical averaging of uncorrelated

fields, it is limited by the inverse number of degrees of freedom (the non-degenerate
mode number M̃ ), such that |γA,res |2 > 1/M̃ . From the ideal fibre calculations above
√
one would expect M̃ ∼ M , but real fibres are always subject to mode coupling lifting
the degeneracy, such that one can expect a scaling M̃ ∝ M [73].

4.1.2 Broadband Light Sources

Since the fibre-based spatial coherence reduction relies on temporal incoherence, the
properties of the illuminating light source (fig. 4.6) are of major importance not only to
temporal intensity noise but also to coherence reduction performance.

Incoherent Sources

Temporally incoherent light sources have random emission characteristics and are nat-
urally widely available (e.g. thermal sources, spontaneous emission sources). The chal-
lenge is to achieve high intensities and is accomplished using ASE, where spontaneously
emitted radiation reaches a gain medium and is coherently amplified.
A compact and affordable class of light sources are semiconductor-based sources, which
consist of electrically driven, optically active, edge-emitting pn-junctions, and produce
a spontaneous emission spectrum when driven at low currents. If the emitted light is
fed back, a lasing threshold current exists over which selective modes are exponentially
58 4 Fibre-Controlled Decoherence

amplified, leading to a narrowband spectrum. However, if due to low light feedback

spontaneously emitted light is predominant in the active region, ASE arises and the
whole spontaneous, incoherent spectrum experiences significant gain. Superluminescent
diodes (SLDs) enforce this behaviour [74] by tilting the end facets, resulting in minimized
backreflections. If a single-mode waveguide structure is used as active region, only a
single transversal mode is excited while an unrestricted number of longitudinal modes
are allowed, resulting in broadband but spatially coherent output.
Producing random emission, the Siegert relation (2.16) for chaotic light sources can
be used to estimate the intensity noise, which in the low-frequency limit yields a RIN of
sI (ω → 0) = 2/∆ν where ∆ν is the optical bandwidth (2.14).

Coherent Sources

Broadband light can also be coherently generated, which has the advantage that second-
order temporal correlations (intensity) are independent of first-order correlations (spec-
trum), and is achieved by mode-locking [75].
(sp)
Consider a laser with a broadband gain spectrum SA (ω) and a resonator with linearly
spaced longitudinal modes eiωn t+ϕn centred at frequencies ωn . If running freely above
threshold, lasing can occur on multiple modes simultaneously and independently, i.e. the
phase ϕn of each mode fluctuates independently (equivalently stated, the instantaneous
resonance frequencies can drift relative to each other). Temporal mode-locking refers
to periodically modulating the gain with frequency ωrep such that only one phase value
ϕn = ϕ is amplified and others are suppressed. Then one obtains a fully deterministic
field with a comb-like spectrum

(fc)
X
SA (ω) = SA (ω) δ(ω − (ωceo + nωrep )), (4.15)
n

where ωrep is the pulse repetition rate, ωceo is the carrier-envelope offset (CEO) which
accounts for the difference of carrier and envelope frequency, and the spectral envelope
(sp)
SA (ω) is determined by SA . If all phases are aligned, (in analogy to the Fraunhofer
diffraction pattern of a grating in space) interference leads to sharp pulses in the time
domain. As the modes are mutually coherent, noise affecting one mode is distributed
to all modes, leading to narrow linewidths of the individual modes. The RIN thus
behaves like a narrowband laser, but is peaked at the repetition rate frequencies sI (ω) ∼
δ(ω − nωrep ).
If a spectrum with a nanometre-wide bandwidth needs to be mode-locked, the gain
4.2 Setup 59

modulation is passively achieved, for fibre lasers often nonlinear optical loop mirrors are
used [76]. Phase-aligned fields thus emit femtosecond pulses which can be dispersively
elongated.

4.2 Setup

In this section the experimental setup for fibre-based generation of incoherent light is
described, as well as the measurement principles to evaluate the coherence properties.

Light Sources

In order to compare the effects of different optical spectra, multiple light sources with
fundamentally varying light generation processes are studied in this chapter. All light
sources are transmitted through a single-mode fibre before the light is modified to reduce
coherence.
One of the tested temporally incoherent light sources is the single-mode fibre pigtailed
Superlum SLD-261 (SLD670) (Superlum DBUT), which is driven by its associated driver
at a current of ISLD670 = 127.5 mA. The other incoherent broadband light is generated
by the laser diode Eagleyard EYP-RWE-0780 (LD780) (in a home-built linear laser
casing whose feedback mirror is removed), and is powered by a Toptica DC110 driver at
a current of ILD780 = 120 mA. In order to stay below the lasing threshold and retain the
broad diode spectrum, reflective feedback is avoided by placing a Faraday isolator after
the output.
As a coherent light source, the 780 nm Menlo M-Comb/M-VIS (FC780) mode-locked
femtosecond laser [75] (as part of the Menlo FC1500 frequency comb) is tested. The
FC780 oscillator is an 1560 nm Erbium-doped fibre ring-laser (M-Comb module) operat-
ing at a repetition rate of frep = 250 MHz and is passively mode-locked using a nonlinear
optical loop mirror [76]. Its output is amplified in another Erbium-doped fibre and fre-
quency doubled to 780 nm (M-VIS module) at 200 mW average power. For practical
reasons the light is coupled into an 80 m single-mode fibre, which in turn is coupled via
a fibre-coupler into another 20 m single-mode fibre leading to the optical table.
In order to have a comparison value for the generated incoherent light, the NBL780,
a grating-based ECDL, serves as a coherent reference light source.
60 4 Fibre-Controlled Decoherence

Lc MF L11

SMF L8 L60 L200

Figure 4.7: Setup for multimode fibre-controlled spatial decoherence.

The single-mode fibre (SMF) coupled light is collimated with an f = 8 mm
lens (L8), is widened by a M = 3.5 telescope to a Gaussian beam waist of
w = 4 mm at the fibre plane (long propagation distances lead to diverging
beams) and is coupled into different multimode fibres (MF) with varying
lenses (Lc). The incoherent output light field is collimated with a f = 11 mm
lens (L11).

Optical Fibres

The multimode optical fibres studied in this chapter are multimode silica fibres. Tab.
4.1 gives an overview over the tested fibre model, length L, core diameter 2a, numer-
ical aperture NA and the theoretically expected differential mode delay τf (4.14) at a
wavelength of λ = 780 nm.

Fibre model (abbreviation) Length L Core size 2a NA Delay τf

Thorlabs FG105LCA (MFR5) 5m 105 µm 0.22 190 ps
Thorlabs FG105LCA (MFR20) 20 m 105 µm 0.22 760 ps
Thorlabs FP150QMT (MFS5) 5m 150 µm 0.39 590 ps
Thorlabs FP150QMT (MFS25) 25 m 150 µm 0.39 3000 ps
Thorlabs GIF625 (MFG5) 5m 62.5 µm 0.275 1 ps [77]

Table 4.1: Overview over optical fibre properties.

The graded-index fibre Thorlabs GIF625 (MFG5) provides a reference as a multimode

fibre with low modal dispersion.

4.2.1 Implementation of Fibre Approach

The setup used to generate spatially incoherent light with a multimode fibre is shown
in fig. 4.7. Since decorrelation is fully passively achieved in the fibre and the incoupled
light is spatially single mode, the collimated beam only has to be expanded with a
telescope to a beam size which produces the desired convergence when coupling into the
fibre. Here a Gaussian beam waist of w = 4 mm is chosen, which for the coupling lens
focal lengths f = 11, 4 mm amounts to converging angles corresponding to numerical
apertures of NA = 0.18, 0.5. The actual experimental implementation uses flip mirrors
4.2 Setup 61

USP

UVCO PD
AOM L1000
L300 L30

Lc MF L11 BS

Figure 4.8: Intensity stabilization setup.

A given spatially coherent input beam passes the first diffraction order of
an AOM placed within the focus of a large focal length telescope for beam
expansion, and is coupled into the multimode fibre (MF). A beam sampler
(BS) splits a small fraction of the light to a photodiode (PD) whose sig-
nal feedbacks a proportional-integral (PI) control loop with setpoint SP. A
voltage-controlled oscillator (VCO) generated RF signal is mixed with the
PI control signal to modulate the AOM driving power.

to direct incoming light onto a common path, and further flip mirrors to switch between
coupling lenses.

Intensity Noise

Light intensity power spectral densities are measured by illuminating a sufficiently fast
photodiode which outputs a linear voltage signal, such that a spectrum analyser can
measure light power fluctuations. Low frequency (up to 100 kHz) spectra are obtained
using the Stanford Research Systems SR760 which records voltage traces in the time
domain and uses fast Fourier transform (FFT) to calculate the spectrum. For higher
frequencies the Agilent N9320 RF spectrum analyser is used. Calculation of normalized
spectra also requires the mean power which is measured using a time-averaging voltmeter.
Note that clipping the light beam might influence the result: Whereas under- or over-
filling the photodiode active area is insignificant for single spatial mode illumination,
simultaneous incidence of different numbers of coherence areas of a spatially multimode
beam can in principle lead to varying noise spectra since the locally independent fluctua-
tions are averaged. Assuming spatially invariant temporal noise statistics, homogeneous
√
illumination with M mutually incoherent areas leads to a noise suppression factor 1/ M .
Thus selectively blocking a part of the beam using a pinhole before the photodiode allows
the recording of spatially dependent intensity noise properties.
In order to correct fluctuations in light power, the intensity stabilization setup shown
62 4 Fibre-Controlled Decoherence

in fig. 4.8 is used. Hereby, intensity control is realized using the first-order diffracted
output of an AOM driven in a regime where the diffraction efficiency is linear in RF
power. The error signal is generated by comparing a setpoint voltage from a low noise
voltage source with the output of a photodiode measuring the light intensity. Using
this signal a PI controller outputs a control signal setting the RF signal power driving
the AOM. As a stable intensity is required after propagation through the multimode
fibre but acousto-optic Bragg diffraction efficiency is wavevector selective (focusing an
incoherent beam to typical AOM aperture sizes leads to significant divergence, i.e. a
large wavevector distribution), the control AOM is placed before the multimode fibre
whereas the feedback photodiode is placed thereafter.

4.2.2 Lateral Shift Interferometer

In order to characterize the coherence after propagation through the multimode fibre,
a laterally shifting Michelson interferometer is used, which is able to measure the full
spatiotemporal coherence function (fig. 4.9, [78]).

Principle

The interferometer is based on a Michelson interferometer whose planar mirrors are

replaced by retroreflectors (“cat’s eyes”). When a light beam is incident on the non-
polarizing beam splitter, each of the beams is reflected from a horizontally oriented
retroreflecting mirror. This introduces a double-reversal of the wavefront of the beam
and a horizontal displacement when off-centrally incident. After recombination at the
beam splitter, the measurable intensity profile at one output is given by

Ilt (ρ) = |Al (ρ) + At (ρ)|2 = Il (ρ) + It (ρ) + 2 Re ΓA,lt (ρ), (4.16)

where l, t denote the interferometer paths, Ilt the intensity of the coherent superposition
of fields Al , At and ΓA,lt the cross-correlation thereof. As there is free space access to the
individual paths, the coherence function can be obtained by recording single-path images
√
(blocking either beam) and performing the normalization γA,lt = (Ilt − Il − It )/2 Il It .
If the beams on both paths have the same path lengths and are identically displaced,
the recombination produces identical, perfectly superposed beams Al = At .
Longitudinally translating a retroreflector by z/2 does not alter the transversal dis-
placement, so for perfect spatial overlap one has Al (ρ, t) = At (ρ, t + z/c). Therefore
the perfect overlap configuration reduces to a Michelson interferometer with which the
4.2 Setup 63

ND CCD
MF L11 L500

MS BS

Rt PC

Rl
PZS
PZC

Figure 4.9: Implementation of the lateral shift Michelson interferometer.

The interferogram generating part starts with a 50/50 non-polarizing beam
splitter (BS) splitting the incoming beam (red). The transmitted part (pur-
ple) is backreflected by the retroreflector Rl whose position can be longitu-
dinally scanned by a piezo actuated translation stage (PZS). The reflected
part (yellow) also returns from a retroreflector (Rt) which can be laterally
shifted by a manual translation stage (MS). At the BS the beams are re-
combined (red) but have an offset due to the shifted Rt (purple and yellow).
The interferogram is recorded by a CCD camera after passing a neutral den-
sity filter (ND). The process is automatized using a computer (PC) which
receives the camera images and sets the piezo actuator voltage via a piezo
controller (PZC). The lenses (L11, L500) before the BS are used to image
the light field at the output of the optical fibre (MF) onto the camera.
64 4 Fibre-Controlled Decoherence

temporal correlation function can be measured. In the opposite case, transversally trans-
lating a retroreflector by s/2 does not alter the longitudinal displacement and introduces
an equal-time lateral shear of the output beam At (ρ, t) = Al (ρ + ses , t).
As one is interested in the degree of coherence |γA | and the camera records the inter-
ference fringes Re ΓA , one can sweep the relative longitudinal displacement across the
extremal equal-time position. The extrema of the spatially resolved envelope then give
the spatial correlation function.

Implementation

Since longitudinal interference fringes have a periodicity of an optical wavelength, the

displacement of the longitudinal retroreflector is realized with a translation stage com-
prising a linear piezo actuator, theoretically allowing positioning with nanometre pre-
cision. Being an open-loop transducer, the piezo hysteresis [79] is accounted for by
performing a calibration measurement with the NBL780 laser. The transversal retrore-
flector is mounted on a manual micrometre stage since the precision requirements are
less stringent.
This is ensured by a 4f-imaging system installed before the interferometer, with which
the fibre output face is magnified by M = 45. As the interferometer only linearly shears
the beam, image formation is not altered and aberrations (minimized by long focal
lengths) are equally introduced in both interferometer paths. The images are recorded
with an AlliedVision Manta G-145B NIR CCD camera, where saturating intensities are
attenuated with absorptive neutral density filters.

Measurement

The coherence function is measured by sequentially, manually adjusting the transversal

displacement and measuring the temporal coherence function for each displacement po-
sition. For this the piezo voltage is set to the minimal value, where the normalization
images are recorded. Then the piezo actuator voltages are discretely stepped, thus ob-
taining the spatially resolved temporal correlation functions. For most measurements
the relative longitudinal displacement of 2 · 20 µm is scanned in 500 steps ensuring each
fringe to be sampled about 10 times.
The equal-time degree of coherence for a given transversal displacement is approxi-
mately given by the extrema of the temporal degree of coherence function of each pixel.
Regarding this quantity as a function of transversal displacement yields the spatial de-
gree of coherence |γA (s)|. The coherence length is deducted by integrating the measured
4.2 Setup 65

250 maximum 1.2 extrapolated

minimum processed
trace raw
1.0
200 mean

Degree of coherence j°A j

Intensity I [ADC]

0.8
150
0.6
100
0.4

50
0.2

0 0.0
−50 −25 0 25 50 −200 0 200
Temporal delay ¿ [fs] Displacement s [µm]

Figure 4.10: Interferometer measurement data processing.

For a given transversal displacement s of the retroreflector, a longitudinal
(i.e. temporal) interferogram is taken pixelwise. The left plot shows an
exemplary measurement for one pixel (here s = −40 µm). First, images of
the individual interferometer paths are recorded whose incoherent sum is
plotted as “mean”. Then the longitudinal trace is recorded and its extrema
are registered. Normalization according to eq. (4.16) yields a spatial coher-
ence data point.
Repeating this procedure for all transversal displacement steps gives a spa-
tial coherence function |γA (s)| for each pixel (right plot, raw). These func-
tions are linearly transformed to ensure that their maximum is unity and
located at zero displacement (processed). In order to obtain the spatial
coherence length sc (2.12), an exponential decay (extrapolated) is assumed
for the tail contribution.
For measuring the residual coherence |γA,res |, a single large transversal dis-
placement is chosen. Then a pixelwise histogram of coherence values is
taken, from which the mean and the standard deviation is obtained.
66 4 Fibre-Controlled Decoherence

(finite-extent) spatial coherence function (2.12). The missing tail mass is estimated by
assuming an exponential decay from the boundary values |γA (smin, max )| (fig. 4.10).

The residual degree of coherence |γA,res | is measured by transversally displacing the

beam far away from the main coherence function peak, typically s = 1 mm is chosen.
Then the mean and the standard deviation of the coherence value of each pixel is taken,
where the standard deviation includes both the actual spread of the distribution and
measurement uncertainties (this quantity will be plotted as error bars). Note that pix-
elwise averaging (of coherence length or residual coherence) implicitly assumes a Schell
model source (2.18).

The main measurement error originates from the imperfectly mounted retroreflector
(with an estimated relative angular misalignment of ∼ 0.9° between the retroreflectors),
resulting in a mutually rotated beam profile. The consequence is that for a fixed transver-
sal stage displacement the actual beam displacement becomes image-position-dependent.
To correct (reduce) this effect, the point of zero lateral displacement is shifted pixelwise
to the maximum of the measured (raw) spatial coherence function. The function is also
normalized such that the maximum is unity.

Additionally, the finite dynamic range of the camera (8 bit) sets a limit to coher-
ence measurement uncertainty. Let the images from the two interferometer arms have
analogue-digital converter (ADC) values of Nl,t , then eq. (4.16) implies that the de-
√
duced coherence value cannot fall below δ = 1/2 Nl Nt , leading to an uncertainty of
∆|γA | = δ/2. In order to avoid an excessive uncertainty, values below a cutoff (here
∆|γA | = 0.01 is chosen) are masked pixelwise and not taken into account in further
analyses.

4.3 Temporal Coherence

As short coherence times are not only desirable to be able to illuminate quickly re-
sponding atomic systems, but are a central requirement for spatial decorrelation after
propagation through the multimode fibre, the temporal coherence properties of the stud-
ied primary and decorrelated light sources are discussed in this section. Additionally,
the temporal coherence functions deduced from the optical spectra are verified by direct
interferometric measurements.
4.3 Temporal Coherence 67

0.16 NBL780
SLD670
0.14
Spectral density SA [1/nm] LD780
0.12 FC780

0.10

0.08

0.06

0.04

0.02

0.00
660 680 700 720 740 760 780 800
Wavelength ¸ [nm]

Figure 4.11: Optical spectra of the studied light sources.

The light sources include a narrowband laser (NBL), a superluminescent
diode (SLD), a laser diode (LD) and the frequency comb (FC), where the
numbers denote the specified centre wavelength. Note that the NBL780 has
been scaled down to fit the canvas.

Optical Spectra of Light Sources

The optical spectra of the light sources are measured by coupling the light into a short
optical fibre connected to the Yokogawa AQ6315A grating-based optical spectrum anal-
yser. The results are depicted in fig. 4.11, normalized to their total power.
The LD780 emits a broad, noisy spectrum centred at 770 nm indicating a spontaneous
emission spectrum and operation below lasing threshold. The spectral fluctuations are
introduced by substantial uncontrolled feedback, leading to random multimode noise.
The lack of any large peaks suggests that external reflective feedback is negligible, such
that the diode can be treated as a spatially single-mode broadband light source. The
SLD670 behaves similarly and emits a symmetric spectrum around 670 nm. Instead
of the strong spectral noise it features a more homogeneous, continuous spectrum, as
expected from specifically designed ASE sources. The remaining spectral ripple is also
indicative of residual feedback towards the gain area and typical for SLDs.
The femtosecond laser from the FC780 operates at around 790 nm and consists of
a two-peak profile roughly following the (frequency-doubled) Erbium gain profile [80].
The narrowband reference laser NBL780 has a linewidth much smaller than the optical
68 4 Fibre-Controlled Decoherence

0.14 10 mW
25 mW
0.12 50 mW
Spectral density SA [1/nm]

85 mW
0.10 125 mW
145 mW
0.08
180 mW

0.06

0.04

0.02

0.00
775 780 785 790 795 800 805 810
Wavelength ¸ [nm]

Figure 4.12: Spectral broadening of femtosecond laser.

The spectra of the FC780 are measured after 100 m of single-mode fibre
propagation, where the nonlinear effect of self-phase modulation (SPM)
broadens the spectrum depending on the transmitted light power.

spectrum analyser resolution and is centred at 780 nm.

Variation of Spectra

For high (peak) power light sources like the femtosecond laser, spectral broadening can
be achieved by self-phase modulation (SPM) in the long (L = 100 m) single-mode fibre
leading to the optical table (sec. 4.2). This nonlinear effect becomes significant when
the intensity reaches I ∼ λ/n(2) L [81]. With wavelength λ ∼ 790 nm, the silica Kerr
nonlinearity n(2) ∼ 2.7 × 10−9 µm2 /kW and mean output power P ∼ 100 mW, this
regime is reached, as seen in fig. 4.12, granting a possibility to tune the desired spectrum.
To estimate unexpected spectral changes due to the subsequent multimode fibre prop-
agation, the spectrum after the single-mode fibre is compared to the spectrum after
both fibres, for which no significant difference is found. This can be attributed to the
larger core size and dispersion in the single-mode fibre, stretching the pulse duration and
decreasing the peak power [82]. If higher peak intensity femtosecond lasers are used, un-
wanted spectral broadening can thus be always avoided by dispersing the ultrashort
pulses before incoupling, e.g. using positively chirped mirrors [83].
4.3 Temporal Coherence 69

1.00 FC780 10 mW
FC780 180 mW
0.75
SLD670
Degree of coherence j°A j

0.50 LD780

0.25

0.00

−0.25

−0.50

−0.75

−1.00
−300 −200 −100 0 100 200 300
Temporal delay ¿ [fs]

Figure 4.13: Temporal coherence functions after multimode fibre.

The opaque lines show the spectrally expected degree of temporal coherence
|γA |, and the semi-transparent lines (background colour) show the oscillat-
ing coherence functions Re γA directly measured with the interferometer.
While the coherence functions of the superluminescent diode (SLD) and
laser diode (LD) monotonically decay, the double peaked spectrum of the
femtosecond laser (FC) leads to an oscillating behaviour. Increasing the
FC780 power decreases its coherence time below the SLD670.

Coherence Function

As the temporal coherence function of a spatially coherent light field is fully determined
by its optical spectrum (2.13), the previously measured spectra can be used to calculate
the input temporal coherence functions. Using the interferometer (sec. 4.2.2) in the
Michelson configuration allows for the spatially resolved measurement of the temporal
coherence function, detecting potential propagation induced spectral changes.
Fig. 4.13 shows the directly measured temporal coherence functions in comparison
to the spectrally expected ones. Plotting the spatially resolved coherence function,
one cannot distinguish any significant differences, such that one can assume that the
spectral features in the regime relevant for induced coherence reduction (O(nm)) remain
unchanged. This agreement with the theoretical estimation (4.10) indicates that there
is no significant spatiotemporally coupled losses.
Tab. 4.2 gives an overview over the coherence times τc (2.12) associated with the light
sources.
70 4 Fibre-Controlled Decoherence

Light source (abbreviation) Remark Coherence time τc

Superlum SLD-261 (SLD670) 118 fs
Eagleyard EYP-RWE-0780 (LD780) 55 fs
780 nm Menlo M-Comb/M-VIS (FC780) 10 mW 170 fs
780 nm Menlo M-Comb/M-VIS (FC780) 180 mW 95 fs

Table 4.2: Overview over light sources.

Since the differential mode delays τf of the step-index fibres (4.14) all by far exceed
the coherence times τc  τf (4.14), the spatial decoherence is expected to be mode-
excitation limited.
Note that since the FC780 is a coherent light source, its temporal coherence function
has periodic revivals set by the repetition rate frep . Thus if the modal dispersion of
the fibre used for coherence reduction exceeds the pulse spacing, no more decorrelation
is possible. The critical length can be estimated from the differential mode delay (sec.
4.1.1) as Lrep ∼ 2cn/frep NA2 ∼ 25 m, which is not exceeded by the studied fibres.

4.4 Spatial Coherence

In this section the spatial coherence properties of broadband light fields after transmis-
sion through a multimode fibre are studied and the influence of fibre type, incoupling
conditions and spectral properties is discussed.

4.4.1 Coherence Function

The short range behaviour of the spatial coherence function is of interest since it de-
termines how fine imaged structures can be to still remain mutually incoherent. If a
statistically large number of mutually incoherent modes are excited and each point in
the output plane is composed of many modes, one expects that the spatial coherence
function (by definition peaked at value 1) quickly decays on a characteristic coherence
length scale sc (4.9).
Fig. 4.14 shows the beam-averaged spatial coherence function |γA (s)| at the output of
the fibres Thorlabs FG105LCA (MFR20) and Thorlabs FP150QMT (MFS25), measured
for the LD780 with a f = 4 mm incoupling lens. These step-index fibres are compared
to the MFG5 graded-index fibre where decorrelation is not yet achieved (4.14) and a
single-mode fibre (SMF) which is spatially coherent. Both are illuminated with the
SLD670.
4.4 Spatial Coherence 71

1.0

0.8
Degree of coherence j°A j

0.6 SMF
MFG5
MFR20
0.4 MFS25

0.2

0.0

−200 −150 −100 −50 0 50 100 150 200

Displacement s [µm]

Figure 4.14: Profile of spatial degree of coherence.

The beam-averaged spatial coherence function is measured for central incou-
pling by a f = 4 mm lens and has a similar main peak for both the round-
core (MFR) and square-core (MFS) step-index fibre. Insufficient differential
mode delay leads to large coherence lengths in the case of graded-index fi-
bres (MFG), bounded by the fully coherent single-mode fibre (SMF).
The light coloured lines show the coherence functions expected for fully
incoherent light (4.9). Due to peak position variations across the beam,
averaging flattens the side peaks. The discrepancy at the sides between
the fibres indicates slower decay of the MFR, leading to a larger coher-
ence length. Accounting for the 45x imaging magnification, the main peak
FWHM of the step-index fibre outputs is on the order of 2 µm.
72 4 Fibre-Controlled Decoherence

For the step-index fibres the spatial correlations exhibit a main peak which quickly
decays on a similar length scale for both fibres. The coherence function has a dip after
the main peak, in accordance with the oscillating Bessel function from the estimation in
eq. (4.9) (where averaging has flattened the dip in the figure). The mutual discrepancy
visible at larger displacements is an indication of the larger NA of the square-core step-
index multimode fibre (MFS) as on the same cross sectional area more angular spread
is possible. The respective integrated coherence lengths (2.12) are given by sc,MFR20 =
139 µm (sc,MFS25 = 88 µm). Accounting for the 45x imaging magnification (sec. 4.2.2),
the number of mutually incoherent modes (2.20) is given by MMFR20 ∼ 202 (MMFS25 ∼
432 ) at 780 nm, which is also an estimate for the number of fully incoherent sites usable
for optical potential generation. Note that the coherence length definition also takes
into account the function tail, such that few slowly decaying functions lead to a large
discrepancy in sc despite the good overlap of the averaged main peaks.
These properties are in stark contrast to the other fibres. While the single-mode fibre
is unsurprisingly fully coherent, the graded-index fibre also exhibits high correlations due
to the reduced modal dispersion (three orders of magnitude lower than the step-index
fibres).

4.4.2 Residual Coherence

Apart from the numbers of coherence areas (the short range behaviour of the coherence
function) discussed in the previous section, the residual coherence (4.14) is of importance
since it determines the overall minimal achievable degree of coherence. Fig. 4.15 sum-
marizes the residual coherence measurements obtained for different light sources (tab.
4.2) and step-index fibres (tab. 4.1).
To isolate the effects of the coherence time, a measurement series with the combination
FC780/MFS25 is taken where the spectral changes are introduced only by the nonlinear
spectral broadening (sec. 4.3). Particularly, the optical path after the single-mode fibre
output remains untouched, so the fibre launch conditions are identical. The results
show barely any change in residual coherence. This agrees with the estimation (sec.
4.3) that the decorrelation is limited by mode degeneracy and initial mode excitation.
Note that the slight increase in residual coherence with coherence time is below the
measurement uncertainty, but indicates that larger spectral widths can lead to smaller
residual coherence also in the mode-limited regime.
The second graph in the figure shows the fibre geometry and fibre length dependence
of the residual coherence, and is measured with the SLD670 and the LD780. Since the
4.4 Spatial Coherence 73

0.08
FC780 MFS25 SLD670 MFR
LD780 MFR
0.07
SLD670 MFS
Residual coherence j°A; res j

LD780 MFS
0.06

0.05

0.04

0.03

0.02
100 120 140 160 103
Coherence time ¿c [fs] Differential mode delay ¿f [ps]

Figure 4.15: Spectral and fibre dependence of residual coherence.

The left plot shows the spectral dependence (fig. 4.12) of the residual coher-
ence |γA,res | measured for the femtosecond laser (FC780) and the square-core
fibre (MFS25). The minor increase of residual coherence with coherence
time indicates operation in the mode excitation-limited, fully decorrelated
regime. The dashed line indicates the minimal achieved value. Note that
the vertical bars indicate the standard deviation of the residual coherence
ensemble measured across the beam and the estimation of the mean is more
precise (sec. 4.2.2).
The fibre dependence of the residual coherence is shown on the right, where
measurements are taken with the superluminescent diode (SLD670) and
the laser diode (LD780) for the round-core (MFR, 5 m and 20 m) and the
square-core fibres (MFS, 5 m and 25 m), which are plotted in terms of their
differential mode delay τf (tab. 4.1). The results indicate overall smaller
residual coherences for the MFS, and a reduction with increasing fibre length
and spectral bandwidth. However, note that these values can vary signifi-
cantly due to the high mode excitation sensitivity (sec. 4.4.3).
74 4 Fibre-Controlled Decoherence

fibre launch conditions are not perfectly fixed, leading to varying coherence reduction
performance, a conclusive statement cannot be stated. However, a general tendency that
longer and higher NA fibres reduce residual coherence can be observed. Furthermore, the
overall lower coherence values for the LD780 again suggest better decorrelation properties
for broader spectra.
Similar as in sec. 4.4.1, the measurement results can be compared to graded-index mul-
timode and single-mode fibres. The combination SLD670/MFG5 results in (τc , |γA,res |) =
(118 fs, 0.39), revealing the reduced modal dispersion in graded-index fibres, and the
combination LD780/SMF verifies full spatial coherence (τc , |γA,res |) = (55 fs, 1.01).

4.4.3 Fibre Launch Conditions

As the incoupling conditions of the light beam define the distribution of excited fibre
modes which in turn determine the intermodal propagation delays, the resulting output
beam profile and spatial coherence function can vary strongly as seen in fig. 4.15. There-
fore in this section the coherence reduction dependence on the incoupling conditions is
studied.
To maximize the spread of propagation constants of the excited modes, a tightly
focused launch of the LD780 is chosen, but is limited to ensure high light power efficiency
inherent to this method. Different launch divergences are realized using different coupling
lenses with focal length f , and different input angles and positions by misaligning the far
coupling mirror (fig. 4.7) at a distance of z = 200 mm. By linearly stepping the mirror
angle from the (misaligned) transmission minimum θ = −θc via the maximum θ = 0
to the minimum on the other side θ = θc allows tracing of the correlation properties
as a function of input. Since from geometric optics the coupling mirror angle θ is
linearly related to both the incidence angle θin = θ (1 − z/f ) and the incidence position
ρin = f θ/(z/f − 1), the critical angle θc is determined by either the maximal angle
obeying the guiding condition of the fibre, θin ≤ θfibre ∼ NA/n, or (in this case) the
position given by the fibre radius ρin ≤ a.

Beam Profiles

Fig. 4.16 shows the beam profile at the end face of the multimode fibres and is the
most striking difference between the fibres. At central, vertical, focused incidence (with
respect to the fibre end face) only low-order modes are excited in the round-core fibre (fig.
4.3). These meridional rays (passing the optical axis) make up only a small fraction of
available modes and have a high mode density in the fibre centre. When misaligning the
4.4 Spatial Coherence 75

1.0

Normalized intensity I/Imax

MFR20

0.8

0.6

0.4
MFS25

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Normalized coupling mirror angle /c

Figure 4.16: Fibre output beam profiles for different incoupling conditions.
The images are resized to the respective fibre core diameters and the colours
are normalized to the maximum intensity of each image. The top row shows
the 20 m round-core fibre, the bottom row shows the 25 m square-core fibre,
both illuminated with the LD780. The coupling mirror angle θ is stepped
from vertical incidence on the left to the maximal angle with significant
power transmission on the right. The fibre mode overlap with the Gaussian
beam input is significantly better for the round-core fibre, indicating only
selective mode group excitation. The intensity distribution of the square-
core fibre is highly homogeneous despite the Gaussian input and very robust
against input conditions.
76 4 Fibre-Controlled Decoherence

input beam, the off-axis incidence induces a ring-shaped intensity distribution. Circular
symmetry reflects the mode symmetries and the finite radial extent shows that mainly
skew rays (not passing the optical axis) are excited. The radial broadening of the rings
suggests a more homogeneous mode excitation due to the fact that the diverging parts of
the input beam can couple into more non-degenerate mode groups. In the critical, barely
guided case, most energy is lost into radiative cladding modes, which also reduces mode
diversity. The intensity dip in the fibre centre again indicates a spatial separation of
propagation mode groups. The white spot at the bottom of the beam profile originates
from a damaged fibre end face. The same measurement with the square-core fibre
reveals its fundamentally different mode structure (fig. 4.4). As the propagation modes
have much more homogeneous profiles, any incoupling position leads to a spatially well-
distributed intensity.

Residual Coherence and Coherence Area

Fig. 4.17 shows the dependence of the beam-averaged residual coherence on incoupling
position, angle and divergence, measured for a lateral displacement of s = 1 mm (i.e.
s = 0.11 aMFR = 0.075 aMFS relative to the fibre core size). Again the square-core
fibre is insensitive to the fibre coupling angle and maintains a low residual coherence of
|γA,res | ∼ 0.03. The round-core fibre on the other hand has an overall higher residual
coherence of |γA,res | ∼ 0.05 with a peak at normal incidence. This can be explained
by the low mode diversity as skewed rays are only accessible for misaligned conditions.
The graph also shows the results for varying incoupling lens focal lengths, i.e. beam
divergence, and hints at better performance for larger divergence due to more excited
mode groups.
This picture can be verified when looking at the spatial structure of the coherence
lengths, which allows for observation of mode structure dependences. Fig. 4.18 shows
measurements performed for an off-axis incoupling condition matching the minimum
residual coherence depicted in fig. 4.17. Similar to the beam profile, the spatial coherence
length is also highly homogeneous for the MFS25, where only the edges show slightly
increased coherence lengths. The MFR20 reveals its modal structure as the ring-shaped
intensity profile separates an inner and an outer part. While the inner part is rather
homogeneous, the outer part varies in coherence, where the area of short coherence
length likely corresponds to the input position. As the skew modes are restricted from
the centre and the input beam is angled towards the centre, only few skew modes are
excited, resulting in the long coherence lengths in the outer part.
4.4 Spatial Coherence 77

0.10 MFR20 L11

MFR20 L4
0.09
MFS25 L11
MFS25 L4
Residual coherence j°A; res j

0.08

0.07

0.06

0.05

0.04

0.03

0.02
0.0 0.2 0.4 0.6 0.8 1.0
Normalized coupling mirror angle µ=µc

Figure 4.17: Residual coherence for different incoupling conditions.

The graph shows results for the round-core (MFR) and square-core (MFS)
fibres with incoupling lens focal lengths of f = 11 mm (4 mm) denoted as
L11 (L4). The MFS reduces coherence significantly better regardless of
fibre launch conditions, indicating spatially homogeneous modal delay dis-
tributions. Central launch into the MFR on the other hand only excites low
order modes resulting in an increased coherence. Note that the difference
in fibre length is comparatively insignificant (fig. 4.15).
Since decoherence in the MFS is insensitive to coupling position and angle,
the dependence on input beam divergence can be analysed. Comparing the
blue graphs, larger divergence is likely to further reduce residual coherence
at the expense of decreased efficiency. Note again that the vertical bars
denote the standard deviation of the residual coherence ensemble measured
across the beam and is larger than the measurement uncertainty.
78 4 Fibre-Controlled Decoherence

MFR20 MFS25
300 200

Coherence length sc [µm]

Coherence length sc [µm]

5 6
180
Position y [mm]

Position y [mm]
250
4
4 160
200
3 140
2
2 150
120

100 0 100
4 6 2.5 5.0 7.5
Position x [mm] Position x [mm]

Figure 4.18: Spatial structure of coherence length.

For each position in the beam the spatial coherence length is calculated
and plotted in a colour map for both a round-core (MFR) and a square-
core (MFS) fibre (note the different colour scales). In both cases the LD780
is off-axis coupled into the fibre through a f = 4 mm lens. The homoge-
neous mode excitation of the MFS leads to a homogeneous coherence length,
while the selective mode excitation of the MFR results in spatially strongly
varying properties. The white spots are a consequence of dirt blocking light
propagation.

4.5 Intensity Noise

In this section the Fourier-domain correlation properties, i.e. the intensity noise, after
propagation through the multimode fibre are studied for the different light sources. As
spatial incoherence inherently requires a finite spectral bandwidth (2.19), Fourier-domain
correlations are inevitable and in turn lead to a non-vanishing intensity power spectral
density. As such global temporal fluctuations directly drive heating transitions (2.34),
it is important to characterize the noise levels in the relevant frequency regime.

Noise Origin

Intensity noise can be generally separated into intrinsic noise determined by the physics
of the light generation process itself and technical noise that emerges from the actual
experimental implementation [84].
The intrinsic noise consists of the following statistically independent terms. First,
electromagnetic field quantization leads to frequency-independent shot noise, which in-
(sh)
creases the RIN (2.15) by sI (ω) = ~ω/P0 , with light power P0 = I0 (ρ)d2 ρ, and is
R

thus negligible for RIN spectra at typical off-resonant dipole trap powers. Second, spon-
taneous emission gives rise to an incoherent continuous spectrum leading to an excess
4.5 Intensity Noise 79

−70 background
NBL780
Relative intensity noise SI [dBc/Hz]
−80 SLD670
FC780
−90

−100

−110

−120

−130

−140

−150
102 103 104 105
Frequency f [Hz]

Figure 4.19: Relative intensity noise of light sources.

The RIN comparison between the free-running single-mode fibre coupled
light sources shows that the coherent light sources (NBL780 and FC780)
have significantly lower noise levels than the incoherent SLD670. The noise
in the 1 kHz regime originates from fibre coupling noise. As the illuminating
light power exceeds P > 1 mW for all cases, the photon shot noise limit
(sh)
sI < −155 dBc/Hz is negligible.

(sp)
noise term (sec. 4.1.2) sI = 2β 2 /M ∆ν, where ∆ν is the spectral bandwidth (2.14)
of the spontaneous emission, β = Psp /P0 is the empirically obtained fraction of sponta-
neous emission contributing to the total power [85] and M is the spatial and polarization
mode degeneracy [86].
Technical noise sources include thermal noise (scaling with temperature) which de-
scribes the fluctuations generated by the measurement electronics, electronic modula-
tion noise (up to MHz) if the light properties are actively controlled, and mechanical
noise (up to kHz) of the passed optical elements. Typically, the main technical noise of
semiconductor sources is induced by fluctuating diode driving currents as it is directly
carried over to the light intensity.

Relative Intensity Noise of Light Sources

Fig. 4.19 shows a comparison of the RIN of a narrowband laser (NBL), a SLD and a
femtosecond laser (frequency comb (FC)). The power spectral densities of the coherent
80 4 Fibre-Controlled Decoherence

(incoherent) sources are measured with a Thorlabs PDA10A-EC (Thorlabs PDA100A-

EC at 20dB gain) photodiode at mean voltage Udc = 1.2 V (Udc = 1.05 V) and the
FFT spectrum analyser with exponentially increasing resolution bandwidths, chosen to
fit the frequency range. The noise around 1 kHz present for the coherent sources can be
identified as mechanical single-mode fibre coupling noise, since the noise vanishes when
performing a reference free-space RIN measurement (before coupling into the fibre). The
NBL RIN shows a spectrum typical for an ECDL, having a similar level as the FC. This
is enabled by the fact that mode locking coherently couples the individual longitudinal
modes such that excess noise is distributed among all modes, resulting in a similar noise
generation process (sec. 4.1.2, [87]).
Longitudinal mode beating noise is not relevant as the NBL has a comparatively short
cavity length (with free spectral ranges ∆νFSR ∼ O(GHz))) and the FC repetition rate
is locked at frep = 250 MHz. Purely random light sources do not produce such noise
as quasi-continuous spectra imply fast decoherence of interfering frequencies. For the
SLD670 with its coherence time of τc ∼ 118 fs, the expected spontaneous emission excess
(sp)
noise floor is sI ∼ −126dBc/Hz. Note that it has been reported [84] that SLD light
amplification using gain-saturated semiconductor optical amplifiers can reduce this noise
floor. Since the SLD shows a significantly higher white noise at kHz frequencies, it is
most likely limited by driving current noise.
Due to low intensity, a spatially resolved RIN measurement involving only one spatial
coherence area could not be performed. A pinhole-filtered measurement (2 % of total
light power) for the FC780 passing the MFS25 does not show significant differences to
fig. 4.19. Furthermore, since the output properties of the MFS25 are insensitive to
fluctuations in the incoupling conditions, the noise within a single coherence area is not
expected to be considerably larger.

Intensity Stabilization

As long-term (compared to measurement times) intensity drifts lead to varying mean

intensities, low frequency noise has to be actively corrected and is achieved using a
negative feedback system (sec. 4.2.1). The main difference to typical stabilization setups
lies in the incoherent illumination on the feedback photodiode, which is not expected to
be an issue when choosing a sufficiently large diode area. The maximum stabilization
bandwidth is determined by both the differential mode delay τf ∼ 1/300 MHz and the
propagation time through the long fibre T ∼ 1/8 MHz. The bandwidth is thus only
limited by the electrical response of the PI box.
4.5 Intensity Noise 81

background
FC780
Relative intensity noise SI [dBc/Hz]

−80 FC780 stabilized

−100

−120

−140

102 103 104 105 106

Frequency f [Hz]

Figure 4.20: Relative intensity noise after multimode fibre with stabilization.
The intensity fluctuations of a free-running and an AOM-controlled, PI
stabilized femtosecond laser is compared. The low frequency noise floor is
significantly reduced, apart from power line noise introduced by the AOM
driver and the PI controller servo bump around 250 kHz. The background
noise bump above 100 kHz results from changing the measurement device
from a FFT to a RF spectrum analyser which has a high noise floor at low
frequencies.
82 4 Fibre-Controlled Decoherence

Fig. 4.20 shows that active stabilization significantly reduces the global intensity noise
from the femtosecond laser. In order to measure a large spectral range, the large band-
width (150 MHz) fixed gain silicon photodiode Thorlabs PDA10A-EC is used, which
is independent of the slower feedback photodiode used for intensity stabilization. The
increased noise of the stabilized light, peaked at 250 kHz, results from the finite stabi-
lization bandwidth as the PI controller fails to follow faster fluctuations. Tuning the
gain parameters allows for flattening of the bump, resulting in a reduced peak at the
cost of a higher low frequency noise floor.
5 Conclusion

In this thesis three methods for the generation of spatially incoherent light for suppression
of coherent noise in arbitrary optical dipole traps have been demonstrated.
In the first approach, an optical ground-glass diffuser was coherently illuminated and
rotated, which produced an intensity-averaged speckle pattern. Second, the incidence
angle of a coherent laser on a holographic diffuser was modulated by two orthogonal
AODs, which resulted in a similar speckle-averaging effect. Third, a broadband light
source was passed through a square-core step-index optical multimode fibre whose modal
dispersion lead to dephasing and thus incoherence.
In an imaging scheme, the number of mutually incoherent modes M (2.20) of the
generated light field determines how many independently illuminated image pixels can
be obtained. For the diffuser approaches, this number could be well-controlled by tuning
the incident beam size. In the particular data shown, the number of modes (defined
with a cutoff at the FWHM) was given by Mrot ∼ 842 for the rotating diffuser and
MAOD ∼ 482 for the AOD approach. In the multimode fibre case, the core size of
the multimode fibre determined the number of its mutually incoherent modes, yielding
MMFS ∼ 432 . If larger numbers are required, a larger fibre core or a shorter wavelength
can be used.
The residual coherences |γ| determine the fraction of the remaining relative speckle
noise and depend on the observing integration time T  1/f (3.5), where f is the fastest
relevant transition frequency in the atomic system. For the diffusers these were measured
to be |γrot | ∼ 3.5 % at Trot ∼ 1/240 Hz and |γAOD | ∼ 6 % at TAOD ∼ 1/50 kHz. Using
a femtosecond laser, the residual coherence of the multimode fibre could be suppressed
to |γMFS | ∼ 3 % where the integration time has to average out the pulses, thus T ∼
1/250 MHz. For tight traps, e.g. optical lattices, with atomic trapping frequencies in
the kHz regime, the diffuser approaches are clearly too slow, whereas the short coherence
times of the broadband sources make the fibre approach viable.
Since this thesis serves as a proof of concept, next steps include imaging and efficiency
tests with parameters exactly reflecting an ultracold atoms experiment, as well as an
actual implementation with higher powered light sources in the atomic system. Potential
84 5 Conclusion

improvements to the fibre approach can be achieved by specifically designing waveguides

which maximize modal dispersion or scrambling the incoupled wavefront to achieve larger
mode excitation diversity.
Bibliography

[1] Georgescu, I. M., Ashhab, S. & Nori, F. Quantum simulation. Reviews of Modern
Physics 86, 153–185 (2014).

[2] Feynman, R. P. Simulating physics with computers. International Journal of The-

oretical Physics 21, 467–488 (1982).

[3] Anderson, M. H., Ensher, J. R., Matthews, M. R., Wieman, C. E. & Cornell, E. A.
Observation of bose-einstein condensation in a dilute atomic vapor. Science (New
York, N.Y.) 269, 198–201 (1995).

[4] Davis et al. Bose-einstein condensation in a gas of sodium atoms. Physical review
letters 75, 3969–3973 (1995).

[5] Bloch, I., Dalibard, J. & Zwerger, W. Many-body physics with ultracold gases.
Reviews of Modern Physics 80, 885–964 (2008).

[6] Gross, C. & Bloch, I. Quantum simulations with ultracold atoms in optical lattices.
Science (New York, N.Y.) 357, 995–1001 (2017).

[7] Greiner, M., Mandel, O., Esslinger, T., Hänsch, T. W. & Bloch, I. Quantum phase
transition from a superfluid to a mott insulator in a gas of ultracold atoms. Nature
415, 39 (2002).

[8] Polkovnikov, A., Sengupta, K., Silva, A. & Vengalattore, M. Colloquium : Nonequi-
librium dynamics of closed interacting quantum systems. Reviews of Modern Physics
83, 863–883 (2011).

[9] Goldman, N., Budich, J. C. & Zoller, P. Topological quantum matter with ultracold
gases in optical lattices. Nature Physics 12, 639–645 (2016).

[10] Chin, C., Grimm, R., Julienne, P. & Tiesinga, E. Feshbach resonances in ultracold
gases. Reviews of Modern Physics 82, 1225–1286 (2010).

[11] Yan, B. et al. Observation of dipolar spin-exchange interactions with lattice-confined

polar molecules. Nature 501, 521 (2013).
86 Bibliography

[12] Schauß, P. et al. Observation of spatially ordered structures in a two-dimensional

rydberg gas. Nature 491, 87 (2012).

[13] Bakr, W. S., Gillen, J. I., Peng, A., Fölling, S. & Greiner, M. A quantum gas
microscope for detecting single atoms in a hubbard-regime optical lattice. Nature
462, 74 (2009).

[14] Sherson, J. F. et al. Single-atom-resolved fluorescence imaging of an atomic mott

insulator. Nature 467, 68 (2010).

[15] Parsons, M. F. et al. Site-resolved imaging of fermionic ˆ6li in an optical lattice.

Physical review letters 114, 213002 (2015).

[16] Haller, E. et al. Single-atom imaging of fermions in a quantum-gas microscope.

Nature Physics 11, 738 (2015).

[17] Cheuk, L. W. et al. Quantum-gas microscope for fermionic atoms. Physical review
letters 114, 193001 (2015).

[18] Omran, A. et al. Microscopic observation of pauli blocking in degenerate fermionic

lattice gases. Physical review letters 115, 263001 (2015).

[19] Weitenberg, C. et al. Single-spin addressing in an atomic mott insulator. Nature

471, 319–324 (2011).

[20] Fukuhara, T. et al. Quantum dynamics of a mobile spin impurity. Nature Physics
9, 235–241 (2013).

[21] Fukuhara, T. et al. Microscopic observation of magnon bound states and their
dynamics. Nature 502, 76 (2013).

[22] Choi, J.-y. et al. Exploring the many-body localization transition in two dimensions.
Science (New York, N.Y.) 352, 1547–1552 (2016).

[23] Kaufman, A. M. et al. Quantum thermalization through entanglement in an isolated

many-body system. Science (New York, N.Y.) 353, 794–800 (2016).

[24] Labuhn, H. et al. Tunable two-dimensional arrays of single rydberg atoms for
realizing quantum ising models. Nature 534, 667 (2016).

[25] Barredo, D., Lienhard, V., Léséleuc, S. d., Lahaye, T. & Browaeys, A. Synthetic
three-dimensional atomic structures assembled atom by atom. Nature 561, 79
(2018).
Bibliography 87

[26] Rigden, J. D. & Gordon, E. I. Granularity of scattered optical maser light. Pro-
ceedings of the Institute of Radio Engineers 50, 2367 (1962).

[27] Bakr, W. S. et al. Probing the superfluid-to-mott insulator transition at the single-
atom level. Science (New York, N.Y.) 329, 547–550 (2010).

[28] Billy, J. et al. Direct observation of anderson localization of matter waves in a

controlled disorder. Nature 453, 891 (2008).

[29] Bloch, I. Ultracold quantum gases in optical lattices. Nature Physics 1, 23 (2005).

[30] Ma, R. Engineered potentials and dynamics of ultracold quantum gases under the
microscope: Dissertation (2014).

[31] Riechert, F. Speckle Reduction in Projection Systems: Dissertation (Univer-

sitätsverlag Karlsruhe, Karlsruhe, 2009).

[32] Bianco, V. et al. Strategies for reducing speckle noise in digital holography. Light:
Science & Applications 7, 48 (2018).

[33] Schmitt, J. M., Xiang, S. H. & Yung, K. M. Speckle in optical coherence tomogra-
phy. Journal of biomedical optics 4, 95–105 (1999).

[34] Burckhardt, C. B. Speckle in ultrasound b-mode scans. IEEE Transactions on

Sonics and Ultrasonics 25, 1–6 (1978).

[35] Lee, J.-S. Speckle analysis and smoothing of synthetic aperture radar images. Com-
puter Graphics and Image Processing 17, 24–32 (1981).

[36] Iwai, T. & Asakura, T. Speckle reduction in coherent information processing. Pro-
ceedings of the IEEE 84, 765–781 (1996).

[37] Stangner, T., Zhang, H., Dahlberg, T., Wiklund, K. & Andersson, M. Step-by-step
guide to reduce spatial coherence of laser light using a rotating ground glass diffuser.
Applied optics 56, 5427–5435 (2017).

[38] Kubota, S. & Goodman, J. W. Very efficient speckle contrast reduction realized by
moving diffuser device. Applied optics 49, 4385–4391 (2010).

[39] Farrokhi, H. et al. High-brightness laser imaging with tunable speckle reduction
enabled by electroactive micro-optic diffusers. Scientific reports 7, 15318 (2017).
88 Bibliography

[40] Trisnadi, J. I. Hadamard speckle contrast reduction. Optics Letters 29, 11 (2004).

[41] Trisnadi, J. I. Speckle contrast reduction in laser projection displays. Projection

Displays VIII 131–138 (2002).

[42] Akram, M. N., Tong, Z., Ouyang, G., Chen, X. & Kartashov, V. Laser speckle reduc-
tion due to spatial and angular diversity introduced by fast scanning micromirror.
Applied optics 49, 3297–3304 (2010).

[43] Epworth, R. E. (ed.). The phenomenon of modal noise in analogue and digital
optical fibre systems (1978).

[44] Mehta, D. S., Naik, D. N., Singh, R. K. & Takeda, M. Laser speckle reduction
by multimode optical fiber bundle with combined temporal, spatial, and angular
diversity. Applied optics 51, 1894–1904 (2012).

[45] Donini, L. A Spatially Incoherent Laser Source (2018).

[46] Lehmberg, R. H., Schmitt, A. J. & Bodner, S. E. Theory of induced spatial inco-
herence. Journal of Applied Physics 62, 2680–2701 (1987).

[47] Karpol, A., Reinhorn, S., Elysaf, E., Yalov, S. & Kenan, B. Method and apparatus
for article inspection including speckle reduction (2005).

[48] Nakano, H. et al. Partially coherent light generated by using single and multimode
optical fibers in a high–power nd:glass laser system. Applied Physics Letters 63,
580–582 (1993).

[49] Redding, B. et al. Low spatial coherence electrically pumped semiconductor laser
for speckle-free full-field imaging. Proceedings of the National Academy of Sciences
of the United States of America 112, 1304–1309 (2015).

[50] Cao, H., Chriki, R., Bittner, S., Friesem, A. A. & Davidson, N. Complex lasers
with controllable coherence. Nature Reviews Physics 1, 156–168 (2019).

[51] Redding, B., Bromberg, Y., Choma, M. A. & Cao, H. Full-field interferometric
confocal microscopy using a vcsel array. Optics letters 39, 4446–4449 (2014).

[52] Chriki, R. et al. Manipulating the spatial coherence of a laser source. Optics express
23, 12989–12997 (2015).
Bibliography 89

[53] Redding, B., Choma, M. A. & Cao, H. Speckle-free laser imaging using random
laser illumination. Nature photonics 6, 355–359 (2012).

[54] Goodman, J. W. Introduction to fourier optics (W.H. Freeman Macmillan learning,

New York, 2017), 4th edn.

[55] Mandel, L. & Wolf, E. Optical coherence and quantum optics (2008), reprint. with
corr edn.

[56] Cohen, L. Generalization of the wiener-khinchin theorem. IEEE Signal Processing

Letters 5, 292–294 (1998).

[57] Friberg, A. T. & Turunen, J. Imaging of gaussian schell-model sources. Journal of

the Optical Society of America A 5, 713 (1988).

[58] Goodman, J. W. Speckle phenomena in optics: theory and applications (Roberts

and Company Publishers, 2007).

[59] Zardecki, A., Delisle, C. & Bures, J. Generalized notion of the coherence area.
Optics Communications 5, 298–300 (1972).

[60] Upatnieks, J. & Lewis, R. W. Noise suppression in coherent imaging. Applied optics
12, 2161–2166 (1973).

[61] Grimm, R., Weidemüller, M. & Ovchinnikov, Y. B. Optical dipole traps for neutral
atoms. In Bederson, B. & Walther, H. (eds.) Advances in Atomic, Molecular, and
Optical Physics, vol. 42 of Advances In Atomic, Molecular, and Optical Physics,
95–170 (Elsevier textbooks, s.l., 1999).

[62] Steck, D. A. Quantum and Atomic Optics (2018), 0.12.3 edn.

[63] Savard, T. A., O’Hara, K. M. & Thomas, J. E. Laser-noise-induced heating in

far-off resonance optical traps. Physical Review A 56, R1095–R1098 (1997).

[64] Jaksch, D., Bruder, C., Cirac, J. I., Gardiner, C. W. & Zoller, P. Cold bosonic
atoms in optical lattices. Physical Review Letters 81, 3108–3111 (1998).

[65] Reed, I. On a moment theorem for complex gaussian processes. IEEE Transactions
on Information Theory 8, 194–195 (1962).

[66] Takai, N. Statistics of dynamic speckles produced by a moving diffuser under the
gaussian beam laser illumination. Japanese Journal of Applied Physics 13, 2025
(1974).
90 Bibliography

[67] Saleh, B. E. A. & Teich, M. C. Fundamentals of Photonics. Wiley Series in Pure

and Applied Optics (John Wiley & Sons, Chicester, 2013), 2nd edn.

[68] Chang, I. C. I. acoustooptic devices and applications. IEEE Transactions on Sonics

and Ultrasonics 23, 2–21 (1976).

[69] Analog Devices. A technical tutorial on digital signal synthesis. Analog Devices
Application Note 1–122 (1999).

[70] Hlubina, P. Spatial and temporal coherence of light in a fibre waveguide. Journal
of modern optics 40, 1893–1907 (1993).

[71] Jensen, J. L. W. V. Sur les fonctions convexes et les inégalités entre les valeurs
moyennes. Acta Mathematica 30, 175–193 (1906).

[72] Pask, C. & Snyder, A. W. The van cittert-zernike theorem for optical fibres. Optics
Communications 9, 95–97 (1973).

[73] Efimov, A. Spatial coherence at the output of multimode optical fibers. Optics
express 22, 15577–15588 (2014).

[74] Rossetti, M. et al. Superluminescent light emitting diodes: the best out of two
worlds. In Schenk, H., Piyawattanametha, W. & Noell, W. (eds.) MOEMS and
Miniaturized Systems XI, SPIE Proceedings, 825208 (SPIE, 2012).

[75] Haus, H. A. Mode-locking of lasers. IEEE Journal of Selected Topics in Quantum

Electronics 6, 1173–1185 (2000).

[76] Doran, N. J. & Wood, D. Nonlinear-optical loop mirror. Optics Letters 13, 56
(1988).

[77] Gloge, D. & Marcatili, E. A. J. Multimode theory of graded-core fibers. Bell System
Technical Journal 52, 1563–1578 (1973).

[78] Efimov, A. Lateral-shearing, delay-dithering mach-zehnder interferometer for spa-

tial coherence measurement. Optics letters 38, 4522–4525 (2013).

[79] Gross, R. & Marx, A. Festkörperphysik. De Gruyter Studium (De Gruyter, Berlin
and Boston, 2018), 3rd edn.

[80] Mears, R. J., Reekie, L., Im Jauncey & Payne, D. N. Low-noise erbium-doped fibre
amplifier operating at 1.54 μm. Electronics Letters 23, 1026–1028 (1987).
Bibliography 91

[81] Stolen, R. H. & Lin, C. Self-phase-modulation in silica optical fibers. Physical

Review A 17, 1448–1453 (1978).

[82] Walmsley, I., Waxer, L. & Dorrer, C. The role of dispersion in ultrafast optics.
Review of Scientific Instruments 72, 1–29 (2001).

[83] Szipöcs, R., Spielmann, C., Krausz, F. & Ferencz, K. Chirped multilayer coatings
for broadband dispersion control in femtosecond lasers. Optics Letters 19, 201
(1994).

[84] Shin, S., Sharma, U., Tu, H., Jung, W. & Boppart, S. A. Characterization and
analysis of relative intensity noise in broadband optical sources for optical coherence
tomography. IEEE photonics technology letters : a publication of the IEEE Laser
and Electro-optics Society 22, 1057–1059 (2010).

[85] Blazek, M., Hartmann, S., Molitor, A. & Elsaesser, W. Unifying intensity noise
and second-order coherence properties of amplified spontaneous emission sources.
Optics letters 36, 3455–3457 (2011).

[86] Hodara, H. Statistics of thermal and laser radiation. Proceedings of the IEEE 53,
696–704 (1965).

[87] Ho, P.-T. Phase and amplitude fluctuations in a mode-locked laser. IEEE Journal
of Quantum Electronics 21, 1806–1813 (1985).
Abbreviations
1D one-dimensional. 12, 41
2D two-dimensional. 20, 35, 44
3D three-dimensional. 35

ADC analogue-digital converter. 66

AM amplitude modulation. 32, 33
AOD acousto-optic deflector. 2, 32, 33, 35, 41, 43–45, 83
AOM acousto-optic modulator. 33, 61, 62, 81
ASE amplified spontaneous emission. 3, 57, 58, 67

CCD charge-coupled device. 33, 63, 64

CEO carrier-envelope offset. 58
CLT central limit theorem. 24, 26

DDS direct digital synthesis. 32, 33

DG ground-glass diffuser. 31
DG220 Thorlabs N-BK7 220 grit. 31, 36, 37, 39
DG600 Thorlabs N-BK7 600 grit. 36
DH holographic diffuser. 31
DH15 Edmund Optics 15◦ FWHM divergence. 32, 36, 42–44
DH80 Edmund Optics 80◦ FWHM divergence. 32, 37, 38, 42, 43

ECDL external cavity diode laser. 31, 59, 80

FC frequency comb. 67, 69, 79, 80

FC780 780 nm Menlo M-Comb/M-VIS. 59, 67–70, 72, 73, 79, 80
FFT fast Fourier transform. 61, 80, 81
FM frequency modulation. 32, 33, 43, 44
94 Abbreviations

FWHM full width at half maximum. 11, 12, 29, 31, 36–38, 42, 45, 71, 83

LD laser diode. 67, 69

LD780 Eagleyard EYP-RWE-0780. 59, 67, 70, 72–75, 78
LED light-emitting diode. 40, 41
LP linear polarized. 53–55, 97

MEMS micro-electro-mechanical systems. 41

MFG graded-index multimode fibre. 71
MFG5 Thorlabs GIF625. 60, 70, 74
MFR round-core step-index multimode fibre. 56, 71, 73, 77, 78
MFR20 Thorlabs FG105LCA. 60, 70, 76
MFR5 Thorlabs FG105LCA. 60
MFS square-core step-index multimode fibre. 56, 71–73, 77, 78
MFS25 Thorlabs FP150QMT. 60, 70, 72, 73, 76, 80
MFS5 Thorlabs FP150QMT. 60

NA numerical aperture. 13, 14, 45, 52, 56, 72, 74

NBL narrowband laser. 67, 79, 80
NBL420 420 nm Toptica TA-SHG. 32
NBL780 780 nm Toptica DL pro. 31, 59, 64, 67, 79

PI proportional-integral. 61, 62, 80–82

RF radio frequency. 33, 35, 61, 62, 81

RIN relative intensity noise. 10, 40, 58, 78–80
RMS root mean square. 25

SLD superluminescent diode. 58, 67, 69, 79, 80

SLD670 Superlum SLD-261. 59, 67, 69, 70, 72–74, 79, 80
SMF single-mode fibre. 60, 70, 71, 74
SPM self-phase modulation. 68

VCO voltage-controlled oscillator. 61

Abbreviations 95

VCSEL vertical-cavity surface-emitting laser. 3

w.l.o.g. without loss of generality. 14

List of Figures

2.1 Number of mutually uncorrelated areas. . . . . . . . . . . . . . . . . . . . 12

2.2 Beam divergence dependence on spatial coherence. . . . . . . . . . . . . . 13
2.3 Relative noise level dependence on degree of coherence. . . . . . . . . . . . 15
2.4 Wave redundancy for noise in different planes. . . . . . . . . . . . . . . . . 16

3.1 Temporal averaging of uncorrelated speckle patterns. . . . . . . . . . . . . 23

3.2 Field correlations immediately after a Gaussian diffuser. . . . . . . . . . . 28
3.3 Rotating diffuser setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Acousto-optic deflector setup. . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5 Speckle pattern image processing. . . . . . . . . . . . . . . . . . . . . . . . 34
3.6 Speckle patterns generated by ground-glass diffusers. . . . . . . . . . . . . 36
3.7 Speckle size dependence on beam size. . . . . . . . . . . . . . . . . . . . . 37
3.8 Displacement correlation length dependence on beam width. . . . . . . . . 38
3.9 Residual coherence convergence of rotating diffuser. . . . . . . . . . . . . . 39
3.10 USAF resolution target imaged with different light sources. . . . . . . . . 40
3.11 Residual coherence convergence due to angular diversity. . . . . . . . . . . 42
3.12 Image of the two-dimensional angular scanning pattern. . . . . . . . . . . 43
3.13 Residual coherence dependence on modulation speed. . . . . . . . . . . . . 44

4.1 Coherence conversion in multimode fibres. . . . . . . . . . . . . . . . . . . 48

4.2 Fibre equivalent of the van Cittert-Zernike theorem. . . . . . . . . . . . . 52
4.3 LP modes of round-core step-index fibres. . . . . . . . . . . . . . . . . . . 54
4.4 LP modes of square-core step-index fibres. . . . . . . . . . . . . . . . . . . 55
4.5 Distribution of modal delays. . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.6 Spectral properties of light sources. . . . . . . . . . . . . . . . . . . . . . . 57
4.7 Setup for multimode fibre-controlled spatial decoherence. . . . . . . . . . 60
4.8 Intensity stabilization setup. . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.9 Implementation of the lateral shift Michelson interferometer. . . . . . . . 63
4.10 Interferometer measurement data processing. . . . . . . . . . . . . . . . . 65
98 List of Figures

4.11 Optical spectra of the studied light sources. . . . . . . . . . . . . . . . . . 67

4.12 Spectral broadening of femtosecond laser. . . . . . . . . . . . . . . . . . . 68
4.13 Temporal coherence functions after multimode fibre. . . . . . . . . . . . . 69
4.14 Profile of spatial degree of coherence. . . . . . . . . . . . . . . . . . . . . . 71
4.15 Spectral and fibre dependence of residual coherence. . . . . . . . . . . . . 73
4.16 Fibre output beam profiles for different incoupling conditions. . . . . . . . 75
4.17 Residual coherence for different incoupling conditions. . . . . . . . . . . . 77
4.18 Spatial structure of coherence length. . . . . . . . . . . . . . . . . . . . . . 78
4.19 Relative intensity noise of light sources. . . . . . . . . . . . . . . . . . . . 79
4.20 Relative intensity noise after multimode fibre with stabilization. . . . . . . 81
Acknowledgements

First of all, I would like to thank my supervisor Christian Groß for the possibility to
work on this interesting project. His advice and support were of major importance for
the realization of the thesis.
I would also like to thank Immanuel Bloch for formally supervising the thesis and
providing me the opportunity to join his group.
Furthermore, I am grateful to Michael Knap for again supervising my thesis and giving
great lectures which left a lasting impression.
I am particularly thankful to Antonio Rubio Abadal for always helping me with any
issues, regardless of physical, organizational or presentational origin. I am also fortunate
to have had help by Simon Hollerith, especially when solving problems in the laboratory.
Additionally, I would like to express my appreciation to Johannes Zeiher for the dis-
cussions and feedback advancing the project. I am also thankful to Jun Rui for sharing
his knowledge on cold atoms with me.
For realizing the technical aspects of the project, I would also like to thank Anton
Mayer, Olivia Mödl and Karsten Förster.
Finally, I am deeply grateful to my parents for constantly supporting me, especially
during my studies.