THESIS

Laser Frequency Stabilization

with a Cryogenic Optical Cavity

Takafumi Ushiba

Department of Physics, Faculty of Science,

University of Tokyo.

December 2015
 Abstract

In the field of precise measurements such as gravitational-wave detectors and

optical atomic clocks, the frequency instability of a laser determines the per-
formance. Especially in the case of state-of-the-art optical atomic clocks, the
frequency noise of a probe laser limits their stability. Therefore, improvements
of the frequency stability of a laser are necessary to make the performance of
optical atomic clocks better.
In this research, we have aimed to improve the laser frequency stability in
order to make better optical lattice clocks. Since the cycle time of a measure-
ment of the atomic transition frequency at an optical lattice clock is about 1
second, the laser frequency instability at around 1 second determines the clock
instability through the Dick effect.
Currently, an optical cavity with a rigid spacer gives the best stability at
around 1 second. However, the stability of several cavity-stabilized lasers has
already reached the thermal-noise limit of the cavity components, which is
one of the fundamental noises. Thus, reducing the cavity thermal noise is
indispensable for obtaining better frequency stability. For this purpose, new
developments are under way around the world through several approaches;
for example, a longer cavity length, crystalline mirror coating with a high
mechanical Q factor, and operation at cryogenic temperature.
A state-of-the-art cavity-stabilized laser has been reached an Allan devi-
ation of 1 × 10−16 at 1 second, which is close to the thermal-noise limit of
this cavity. For further reductions of the thermal noise to reach a stability of
1 × 10−17 , utilizing a cryogenic cavity below 20 K is inevitable. Thus, several
groups are currently trying to make an optical cavity to work at a very low
temperature, but there is no reported result that shows an instability less than
1 × 10−16 , because a cryogenic cavity at below 70 K needs a cryogenic refriger-
ator that can easily induce additional noises. A group in Germany reported an
Allan deviation of 2 × 10−13 at around 1 second and 2 × 10−14 at around 1000

iii
Abstract

seconds using a silicon cavity operated at a very low temperature; they also
discussed the thermal property necessary to reach 2×10−17 stability. However,
the noises that limit the current instability have not been shown.
In this thesis, we construct a laser frequency-stabilizing system with a
monocrystalline silicon cavity operated at a very low temperature. In order
to realize a temperature-stable and quiet environment at less than 20 K, we
used a cryogenic refrigerator with helium recondensation as well as a multiple
stack structure with glass balls for thermal isolation. We used a passive and
active vibration isolation system for reducing the effect of seismic motion and
vibration from the refrigerator. In addition, several noise-reduction system
such as Doppler noise cancellation, laser power stabilization, etc, were installed.
We evaluated the performance of our system quantitatively. The instability
−15
−0.34 × 10
result has been evaluated to have an Allan deviation of σy = 7.03+0.38 .
This instability is lowest around 1 second using an optical cavity operated
at a very low temperature. Moreover, we evaluated any noise contributions
for the frequency instability due to various effects, such as seismic motion,
temperature fluctuation, etc, and found that the acceleration noise limits the
stability of our laser frequency.
In this work, we showed that laser stabilization with a very low-temperature
optical cavity can reach a 10−15 level at around 1 second. Moreover, we showed
prospects to reach a stability below 10−16 by using a cryogenic optical cavity.

Thesis supervisor: Masaki Ando

Thesis title: Laser Frequency Stabilization with a Cryogenic Optical Cavity

iv
 要旨

重力波検出器や光格子時計などの精密測定分野では、レーザーの周波数安定度
がそれらの性能を決める。特に世界最高水準の光格子時計ではプローブレー
ザーの周波数雑音が時計の安定度を制限しており、より高性能な時計を作るに
はレーザーの周波数安定度を向上させる必要がある。
本研究ではより高性能な光格子時計開発のために、レーザー周波数の安定
化を行っている。光格子時計では原子の遷移周波数測定にかかる時間が 1 秒程
度であるため、1 秒付近のレーザー周波数の安定度がディック効果を介して時
計の性能を制限する。
1 秒付近の周波数基準として現在最も安定なものは光共振器である。しか
しながら、世界最高水準の光共振器に安定化されたレーザーの安定度は原理
的な雑音の一つである光共振器の熱雑音レベルに到達している。したがって、
レーザーの周波数安定度をより向上させるには光共振器の熱雑音低減が必要不
可欠であり、長い光共振器や機械的 Q 値の高いコーティングを用いた共振器、
低温化した共振器を利用した周波数安定化が試みられている。
世界最高の光共振器に安定化されたレーザーの安定度は 1 秒のアラン標準
偏差で 1 × 10−16 を実現しており、その共振器の熱雑音限界近くまで到達して
いる。そのため、1 × 10−17 の周波数安定度を目指した更なる熱雑音低減のた
めには、20 K 以下の極低温光共振器を用いることが必要不可欠である。現在
いくつかの研究グループが安定な極低温光共振器の製作を試みているが、70 K
以下の温度領域に到達するためには新たな雑音源となりうる冷凍機を用いる必
要があるため、未だ 10−16 以下の安定度は報告されていない。ドイツのグルー
プのみが極低温の光共振器に安定化したレーザーの周波数安定度を報告してお
り、そのアラン標準偏差は 1 秒付近で 2 × 10−13 、1000 秒付近で 2 × 10−14 であ
る。この共振器の熱特性によって決まる安定度の限界は 2 × 10−17 とされてい
るが、現在の安定度を制限する要因についてはわかっていない。
そこで、我々は極低温動作の光共振器による周波数安定化レーザーを構築
した。本実験では高い温度安定性と静寂な環境を実現するため、ヘリウムガス
の凝縮を利用したパルスチューブ冷凍機を用いている。また、高い温度安定性

v
要旨

の確保のためにガラス球を用いた多段のスタック構造を用いている。さらに、
地面振動や冷凍機の振動が光共振器に伝わらないように受動防振と能動防振を
組み合わせた防振装置を用いている。加えて、ドップラー雑音除去やレーザー
の強度安定化などの雑音低減措置もとられている。
−15
−0.34 ×10
この周波数安定化システムを定量的に評価することによって 7.03+0.38
のアラン標準偏差を得た。この結果は極低温の光共振器を用いた周波数安定化
レーザーとして最も安定なものである。また、地面振動や温度揺らぎなどの雑
音の寄与を定量的に調べることにより、私たちの周波数安定度が共振器に加わ
る加速度の影響によって制限されていることを示した。
これにより、極低温の光共振器を用いても 10−15 台の安定度のレーザーが
構築できることを示した。また、極低温の光共振器を用いた周波数安定化に
よって 10−17 台の安定度を達成するための具体的な方法を示した。

指導教員：安東正樹（准教授）
論文題目：極低温光共振器を用いたレーザー周波数安定化

vi
 Acknowledgements

This experiment has been supported by many people. Here, I would like to
express my gratitude to them.
First, I would like to thank my supervisors. My former supervisor, Kimio
Tsubono, gave me a chance to start this experiment. If I could not have become
acquainted with him, I could not pursue to this work. My current supervisor,
Masaki Ando, held meetings for my doctoral thesis and discuss the schedule
of the experiment and writing. I could not have finished writing this thesis
without him. My second supervisor, Hidetoshi Katori, in the ALPS course
often gave me a useful advice for my experiment. He also gave me a chance to
do this work in his laboratory at RIKEN.
I am glad to have worked with my co-researchers. Yoichi Aso is one of
proposers of this experiment. He contrived the conceptual idea using thermal
insulation stacks. Noriaki Ohmae is a very reliable person for me. Almost
all the progress and failures of this work were reported to him at first and he
always showed me the next way to go. He also help me directly to construct
the experimental setup when I did that work that was difficult to do alone.
Moreover, the narrow linewidth laser that is used in the beat frequency mea-
surement was developed by him. Ayaka Shoda taught me the concept of a
feedback system involving AVIS. The circuit made by her was also used in the
experiment.
I am obliged to our group members. Yuta Michimura, who is a current
assistant professor of our group, gave me some experimental advice. Kentaro
Komori and Yuya Kuwahara, who are the 2nd-year master course students,
were good for discussing the experimental and theoretical matters. Naoki
Aritomi and Tomofumi Shimoda, who are 1st-year master course students,
helped me to cope with incidental tasks at Hongo campus while in RIKEN.
I would also like to thank past members of our group. Wataru Kokuyama
taught me how to progress the research and pursue life as a PhD student.

vii
Acknowledgements

Nobuyuki Matsumoto often gave me fundamental questions concerning the

experiment, and thus I could reconsider my experiment deeply each time.
I am grateful to the engineers of the Faculty of Science. My work could
not have been completed without Shigemi Otsuka and Yoshikatsu Nanjo.
This research is mainly supported by Prof. Katori’s ERATO project, Inno-
vative Space-Time project, Japan Science and Technology Agency, and JSPS
Grant-in-Aid for JSPS Fellows Number 15J11096.   Also, this research is
also performed in Prof. Katori’s laboratory, Quantum Metrology Laboratory,
RIKEN.
Finally, I greatly thankful my parents for their continuous support of my
life.

viii
 Contents

Abstract iii

要旨 v

Acknowledgements vii

Contents xii

Glossary xiii

1 Introduction 1

2 Frequency Stabilization 7
2.1 Fabry-Perot cavity . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Transmittance and Reflectance . . . . . . . . . . . . . . 7
2.1.2 Free spectral range and finesse . . . . . . . . . . . . . . . 9
2.1.3 Reflectance and its differential function . . . . . . . . . . 10
2.1.4 Time constant of a cavity . . . . . . . . . . . . . . . . . 11
2.1.5 Response for frequency fluctuation . . . . . . . . . . . . 12
2.2 Schawlow-Townes limit . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Pound-Drever-Hall method . . . . . . . . . . . . . . . . . . . . . 16
2.3.1 Error signal . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.2 Response for frequency fluctuation . . . . . . . . . . . . 18
2.4 Resonant frequency and cavity length . . . . . . . . . . . . . . . 19
2.5 Noise source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5.1 Thermal noise . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5.2 Shot noise . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.3 Acceleration noise . . . . . . . . . . . . . . . . . . . . . . 24
2.5.4 Temperature-fluctuation noise . . . . . . . . . . . . . . . 24
2.5.5 Residual amplitude modulation noise . . . . . . . . . . . 25

ix
Contents

2.5.6 Residual gas noise . . . . . . . . . . . . . . . . . . . . . . 26

2.5.7 Doppler noise . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5.8 Other noises . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.6 Summary of this chapter . . . . . . . . . . . . . . . . . . . . . . 30

3 Experimental Design 31
3.1 Design concept . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 Overview of the experimental design . . . . . . . . . . . 31
3.1.2 Silicon optical cavity . . . . . . . . . . . . . . . . . . . . 32
3.1.3 Vibration isolation system . . . . . . . . . . . . . . . . . 34
3.1.4 Cryogenic refrigerator . . . . . . . . . . . . . . . . . . . 35
3.1.5 Thermal isolation system . . . . . . . . . . . . . . . . . . 36
3.1.6 Input and output optics . . . . . . . . . . . . . . . . . . 39
3.2 Noise budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.1 Thermal noise . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.2 Shot noise . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.3 Acceleration noise . . . . . . . . . . . . . . . . . . . . . . 43
3.2.4 Temperature-fluctuation noise . . . . . . . . . . . . . . . 46
3.2.5 Residual amplitude modulation noise . . . . . . . . . . . 49
3.2.6 Residual gas noise . . . . . . . . . . . . . . . . . . . . . . 49
3.2.7 Doppler noise . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2.8 Design performance . . . . . . . . . . . . . . . . . . . . . 52
3.3 Summary of this chapter . . . . . . . . . . . . . . . . . . . . . . 54

4 Performance Evaluation 55
4.1 Silicon optical cavity . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Vibration and its isolation systems . . . . . . . . . . . . . . . . 58
4.2.1 Seismic vibration level . . . . . . . . . . . . . . . . . . . 58
4.2.2 Passive vibration isolation system . . . . . . . . . . . . . 60
4.2.3 Active vibration isolation . . . . . . . . . . . . . . . . . . 63
4.2.4 Vibration sensitivity . . . . . . . . . . . . . . . . . . . . 66
4.3 Temperature fluctuation . . . . . . . . . . . . . . . . . . . . . . 69
4.3.1 2nd stage of the refrigerator . . . . . . . . . . . . . . . . 69
4.3.2 Outer shield . . . . . . . . . . . . . . . . . . . . . . . . . 70

x
Contents

4.3.3 Laser power effect . . . . . . . . . . . . . . . . . . . . . . 71

4.4 Feedback system . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5 Input and output optics . . . . . . . . . . . . . . . . . . . . . . 76
4.5.1 Laser power stabilization . . . . . . . . . . . . . . . . . . 76
4.5.2 Doppler noise cancellation . . . . . . . . . . . . . . . . . 78
4.6 Summary of this chapter . . . . . . . . . . . . . . . . . . . . . . 81

5 Stability Estimation 83
5.1 Noise estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.1.1 Thermal noise . . . . . . . . . . . . . . . . . . . . . . . . 83
5.1.2 Residual gas noise . . . . . . . . . . . . . . . . . . . . . . 83
5.1.3 Acceleration noise . . . . . . . . . . . . . . . . . . . . . . 84
5.1.4 Temperature-fluctuation noise . . . . . . . . . . . . . . . 85
5.1.5 Laser power fluctuation noise . . . . . . . . . . . . . . . 87
5.1.6 Circuit noise . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.1.7 Shot noise . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.1.8 RAM noise . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.1.9 Doppler noise . . . . . . . . . . . . . . . . . . . . . . . . 90
5.1.10 Total noise . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.2 Frequency instability measurement . . . . . . . . . . . . . . . . 92
5.2.1 Experimental setup . . . . . . . . . . . . . . . . . . . . . 92
5.2.2 Instability measurement . . . . . . . . . . . . . . . . . . 94
5.3 Estimation without the acceleration noise . . . . . . . . . . . . . 102
5.4 Summary of this chapter . . . . . . . . . . . . . . . . . . . . . . 104

6 Conclusion 105
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.2 Future prospects . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

A Gaussian Beam and Cavity Mode 109

A.1 Gaussian beam . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A.2 Cavity eigen mode of a FPC . . . . . . . . . . . . . . . . . . . . 111

xi
Contents

B Allan Deviation 113

B.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
B.2 Allan deviation and power spectrum . . . . . . . . . . . . . . . 114

C Electrical Circuit in our Experiment 117

D Photos of Experimental Setup 121

Bibliography 125

xii
 Glossary

Symbols √
i imaginary unit i = −1
c speed of light c = 299792458 m/s
h Planck’s constant h = 6.626 × 10−34 Js
ℏ Dirac constant ℏ = h/2π = 1.055 × 10−34 Js
kB Boltzmann constant kB = 1.381 × 10−23 J/s
σB Stefan-Boltzmann constant σB = 5.670 × 10−8 J/s
e elementary charge e = 1.602 × 10−19 C
λ laser wavelength
ν laser frequency [Hz] (ν = c/λ)
Ω laser angular frequency [rad/sec] (Ω = 2πν)
f Fourier frequency [Hz]
n refractive index
L length of a FPC
F finesse of a FPC
p0 pressure of standard condition gas
V0 volume of standard condition gas
T0 temperature of standard condition gas
NA Avogadro’s constant NA = 6.022 × 1023

xiii
Glossary

Acronyms
ADC analog-to-digital converter
AR anti-reflective, anti-reflection
AOM acousto-optic modulator
AVIS active vibration isolation system
AVIT active vibration isolation table
BS beam splitter
CTE coefficient of thermal expansion
DAMI differential arm length Michelson interferometer
EOM electro-optic modulators
ECDL external cavity diode laser
FDT fluctuation dissipation theorem
FEM finite element method
FPC Fabry-Perot cavity
FSR free spectral range
FWHM full width at half maximum
HR high-reflection
HWP half-wave plate
LD laser diode
OFC oxide-free copper
PBS polarizing beam splitter
PD photodetector, photodiode
PDH Pound-Drever-Hall (method)
PSD power spectral density
PVIS passive vibration isolation system
PZT lead zirconate titanate (piezoelectric transducer)
QWP quarter-wave plate
RMS root mean square
RAM residual amplitude modulation (noise)
TIS thermal isolation stack
VCXO voltage-controlled crystal oscillator

xiv
 Chapter 1
Introduction

Since the development of the laser in 1960 [1], lasers have been used in many
fields of academic research, such as material science, quantum physics, and
astronomy. As these researches became more complicated and precise, the im-
portance of a frequency stabilized laser has been increasing. In resent years,
very stable lasers are required in several fields, such as gravitational-wave de-
tectors [2–4], gravity measurements [5], and optical clocks [6]. Especially in the
field of frequency standard, the stability of several clocks, for example Sr and
Hg optical lattice clocks, is limited by the frequency fluctuations of a probe
laser, which detects the atomic transition frequency [7, 8].
The frequencies of several optical clocks are used for secondary represen-
tations of the second [9], and these are candidates for the next definition of
second. Therefore, improving the stability of these clocks is very important for
many fields of science. Moreover, optical clocks are also useful for fundamen-
tal physics. Stable clocks can be used for testing the local position invariance
by measuring the constancy of the fine-structure constant [10]. Therefor, im-
provements of optical clocks are essential for state-of-the-art science.
To focus on optical lattice clocks, the instability of a probe laser at around
1 second determines their instability [7] through the Dick effect [11]. Thus,
improving the laser frequency instability at around 1 second is important for
making better optical lattice clocks [12]. Optical lattice clocks are achieved
by stabilizing the probe laser frequency to the atomic-transition frequency [6].
√
Since their stability typically improves in proportion to 1/ τ [13], where τ
is the averaging time of the frequency measurement, the single measurement
accuracy of atomic transition frequency determines the clock instability. Since
the minimal feedback cycle time in optical lattice clocks is about 1 second

1
1 Introduction

Dusseldolf

MPQ
MPQ NPL VCQ

JILA NIST
LNE-SYRTE
PTB
PTB

2000 2005 2010 2015

Year

Figure 1.1: The history of the frequency stability of the major cavity-stabilized
lasers. Red dots are the frequency-stabilized lasers with a cavity at room
temperature [16–21]. The yellow dot is when using a long cavity of about 48
cm [24]. The green dot is when using crystalline coating mirrors that have a
high mechanical Q factor [22]. The light-blue dot is when using a cryogenic
optical cavity at around 124 K [12]. The blue dot is when using a cryogenic
cavity below 24 K [23]. The stability is the Allan deviation or modified Allan
deviation at around 1 second. Detailed characteristics are given in Table 1.1

[13–15], the stability of optical lattice clocks has been improved when the laser
frequency stability at around 1 second becomes better.
Frequency stabilization at around 1 second is often performed using an
optical cavity with a rigid spacer, because it provides the best current stability
at around 1 second. Figure 1.1 shows the history of the major cavity-stabilized
lasers [12, 16–24]; Table 1.1 gives their stability and several characteristics.
In 2004, it was suggested that the frequency instability of several narrow-
linewidth lasers was limited by the Brownian thermal noise of the cavity com-
ponents [25]. The thermal-noise level of lasers discussed in this paper was
estimated for an Allan deviation of σy ≃ 3 × 10−16 . Therefore, thermal-noise
reduction is necessary to achieve frequency stability below 3 × 10−16 . One
approach for reducing the thermal noise is to use a long cavity [24]. The

2
 Institute Instability Length Temperature Coatings
MPQ [16] 4 × 10−15 27 cm about 300 K dielectric multilayer
NPL [17] 1 × 10−15 10 cm about 300 K dielectric multilayer
JILA [18] 1 × 10−15 7 cm about 300 K dielectric multilayer
MPQ [19] 2 × 10−15 7.8 cm about 300 K dielectric multilayer
LNE-SYRTE [20] 6 × 10−16 10 cm about 300 K dielectric multilayer
NIST [21] 3 × 10−16 29 cm about 300 K dielectric multilayer
PTB [12] 2 × 10−16 21 cm about 124 K dielectric multilayer
VCQ [22] 1 × 10−15 3.5 cm about 300 K crystalline multilayer
Dusseldolf [23] 2 × 10−13 25 cm below 24 K dielectric multilayer
PTB [24] 1 × 10−16 48 cm about 300 K dielectric multilayer

Table 1.1: The stability and several characteristics of each frequency stabilized
laser shown in Fig. 1.1. The instability of PTB [12] is the modified Allan
deviation at around 1 second. The instability of the other results is the Allan
deviation at around 1 second. The instability is read from the graph of the
Allan deviation or the modified Allan deviation of lasers in Ref. [12, 16–24].

estimated thermal-noise level of the cavity, which is about 48 cm long, is

σy ≃ 5.4 × 10−17 . The instability of the laser is σy ≃ 9 × 10−17 at around 100
seconds and σy ≃ 1 × 10−16 at around 1 second. This result is the highest fre-
quency stability at around 1 second, which is close to the thermal-noise limit.
Another approach is to construct a reference cavity with high mechanical Q-
factor coating mirrors [22]. The estimated thermal noise level of the cavity is
σ ≃ 1 × 10−15 and the instability of the laser is σy ≃ 1 × 10−15 at around 1
second. The thermal-noise level of this cavity is not so good because the cavity
length is short, 3.5 cm long. However, its stability reaches a stability which
cannot be achieved with conventional dielectric multilayer coatings. Thus, this
result is very important in terms of making high mechanical Q-factor coatings.
Another approach is to cool an optical cavity at low temperature [12,23]. One
way is to cool the optical cavity at around 124 K, which is the zero-cross
temperature of the coefficient of thermal expansion (CTE) of monocrystalline
silicon [12]. The thermal-noise level of the cavity is estimated at the modified
Allan deviation of mod σy ≃ 5 × 10−17 , and the frequency stability of the
laser is mod σy ≃ 2 × 10−16 at around 1 second. Another way is to cool the
optical cavity below 24 K [23]. The estimated thermal noise level of the cavity

3
1 Introduction

is σy ≃ 6 × 10−18 . The frequency instability of the laser is σy ≃ 2 × 10−14 at

around 1000 second and σy ≃ 2 × 10−13 at around 1 second.
In view of the above results, the laser instability with these approaches,
except for utilizing a very low temperature cavity, is close to another thermal
noise limit, and thus using an optical cavity operated at a very low temperature
is inevitable for any further improvements of the laser stability. However, the
laser frequency instability with a very low-temperature cavity is currently far
from the state-of-the-art frequency stability: about 1 × 10−16 . Moreover, the
noise sources that limit the current instability are not determined, though the
noise level of the thermal effects has been well discussed [23]. Other than the
noises due to the thermal effects, there are many fundamental and technical
noises that disturb frequency stabilization, such as the shot noise, acceleration
noise, and residual amplitude modulation (RAM) noise. Therefore, we should
research all of the noise sources, at least all of the noise sources that are
well known as noise sources that disturb frequency stabilization, in order to
show that the very low-temperature optical cavity is a strong tool for further
improvements of the laser stability.
In this thesis, we constructed a frequency-stabilized laser with a cryogenic
optical cavity, and estimated many noise sources that may limit the laser
frequency stability, both systematically and quantitatively. By doing this sys-
tematic and quantitative evaluation, we showed the prospects to improve the
laser stability with a cryogenic optical cavity operated at very low tempera-
ture. Using a very low-temperature optical cavity is one of strong methods
for reducing the thermal noise. Thus, our work can be an important step for
making an ultrastable laser.
The organization of this thesis is as follows: Chapter 2 introduces the
method of frequency stabilization with a reference cavity. This chapter also
shows the noise sources that disturb the stabilization of the laser frequency
and their formulation. Chapter 3 describes the design and features of our ex-
perimental setup. The noise budget and the design stability of our experiment
are also described in this chapter. Chapter 4 considers the performance of our
experimental setup. Chapter 5 describes a noise estimation using the results
of Chapter 4 and that of a direct stability measurement of our laser. Chap-
ter 6 summarizes the results of this work. This chapter also explains future

4
prospects of the experiment.

5
 Chapter 2
Frequency Stabilization

Laser frequency stabilization has been performed since the invention of a laser.
Especially since 2000, the stability of the laser frequency has gradually im-
proved. In this chapter, we give an overview of frequency stabilization.
We introduce the characteristic of a Fabry-Perot cavity (FPC) in Sec-
tion 2.1. In the next Section, we discuss the fundamental limit of the frequency
stability of a laser, called the Shawlow-Townes limit. We describe the Pound-
Drever-Hall method for frequency stabilization in Section 2.3. We explain the
relation between the frequency stability and the cavity length fluctuation in
Section 2.4. Lastly, we discuss the noise sources of laser frequency stabilization
in Section 2.5.

2.1 Fabry-Perot cavity

2.1.1 Transmittance and Reflectance
A Fabry-Perot cavity (FPC) consists of two mirrors facing each other, like
in Fig. 2.1. A part of the input light to a FPC is transmitted through a
front mirror, and plies between two mirrors. After that, the light is partly
transmitted through a FPC, and the rest is reflected.
The amplitude of the reflected light can be calculated as

∑
∞ ( )
( )
−2iδ n rE t2F e−2iδ
Er = ErF +Et2F (−rE )e−2iδ rF rE e = E rF − , (2.1)
n=0
1 − rF rE e−2iδ

where rF and rE are the amplitude reflectivity of the front and end mirrors; tF

7
2 Frequency Stabilization

Front mirror End mirror

Figure 2.1: FP cavity

is the amplitude transmittance of the front mirror and δ = 2πL/λ is a phase

change of the light traveling along the cavity length.
The amplitude of the transmitted light can be described as

∑
∞
( )n tF tE e−iδ
−iδ
Et = EtF tE e rF rE e−2iδ =E , (2.2)
n=0
1 − rF rE e−2iδ

where tE is the amplitude transmittance of the end mirror.

We then obtain the amplitude reflectance and transmittance of a FPC,
respectively, as follows:

rE t2F e−2iδ
rFPC = rF − , (2.3)
1 − rF rE e−2iδ

tF tE e−iδ
tFPC = . (2.4)
1 − rF rE e−2iδ
The power of the reflected and transmitted light can be calculated as
2
{(rF2 + t2F )rE − rF } + 4rF rE (rF2 + t2F ) sin2 δ 2
Pr = |Er |2 = |A| , (2.5)
(1 − rF rE )2 (1 + F sin2 δ)

t2F t2E
Pt = |Et |2 = |A|2 , (2.6)
(1 − rF rE )2 (1 + F sin2 δ)
where F = 4rF rE /(1 − rF rE )2 . The power of the transmitted light becomes

8
 2.1 Fabry-Perot cavity

Figure 2.2: Example of the transmitted laser power. The Y-axis is normalized
to make the maximum value 1.

maximum at δ = nπ. When the power of the transmitted light is maximum,

the cavity state is called ’on resonance’.

2.1.2 Free spectral range and finesse

The power of the transmitted light has a sharp peak when the amplitude
reflectance of the front and end mirrors satisfies rF , rE ∼ 1. Figure 2.2 shows
an example of the transmitted laser power.
The laser frequency resonant with a FPC can be calculated as

2πL 2πLν nc
δ= = = nπ ⇔ ν = . (2.7)
λ c 2L
The difference between the resonant frequencies next to each other is called
the free spectral range (FSR); which can be described as
c
νFSR = . (2.8)
2L

Next, we calculate the full width at half maximum (FWHM) frequency,

9
2 Frequency Stabilization

νFWHM . Assuming that rF , rE ∼ 1 and using Eq. (2.6), νFWHM satisfies the
following equation:

1 1
2 = . (2.9)
1 + F sin (2πνFWHM /νFSR ) 2

Then, if we assume νFWHM ≪ νFSR , νFWHM can be written as

c
νFWHM = √ . (2.10)
πL F

The ratio of νFSR and νFWHM is called the ’finesse’, which can be expressed
as √
νFSR π rF rE
F= = . (2.11)
νFWHM 1 − rF rE
This is an index of the sharpness of the cavity resonance, and only depends on
the amplitude reflectance of the front and end mirrors.

2.1.3 Reflectance and its differential function

Frequency stabilization is achieved using the reflected light of a FPC. We
therefore describe the behavior of reflectance of the cavity in this subsection.
Using Eq. (2.3), the reflectance of a FPC and its differential as a function
of δ can be written as

rE t2F e−2iδ
rFPC = rF − , (2.12)
1 − rF rE e−2iδ

′ −2it2F rE e−2iδ
rFPC = . (2.13)
(1 − rF rE e−2iδ )2
Figure 2.3 shows the behavior of the reflectance as a function of δ.
When a FPC is on resonance, Eq. (2.12) and Eq. (2.13) can be changed to

t2F rE
rFPC = rF − , (2.14)
1 − rF rE

′ 2it2F rE
rFPC = . (2.15)
(1 − rF rE )2
′
Then, |rFPC | becomes minimum and |rFPC | has the maximum value on reso-

10
 2.1 Fabry-Perot cavity

Figure 2.3: Absolute value and phase of the reflectance of a FPC. The param-
eters are rF = rE = 0.99, tF = 0.13.

nance.

2.1.4 Time constant of a cavity

When a FPC is in the steady state, the laser light stays in a FPC for a while.
Now, considering the situation that the input laser is suddenly stopped, the
laser beam staying in a FPC gradually reduces. In this part, we formulate the
time scale of this decay.
The round-trip loss of the amplitude of the light inside a FPC can be
written as
∆Eintra = −(1 − rF rE )Eintra . (2.16)

Then, the attenuation rate satisfies the following equation:

dEintra c
= − (1 − rF rE )Eintra . (2.17)
dt 2L

11
2 Frequency Stabilization

The solution to Eq. (2.17) can be described as

Eintra = E0 e− 2L (1−rF rE )t ,
c
(2.18)

where E0 is the amplitude of the intra-cavity light when the laser is stopped.
Therefore, the intra-cavity laser power can be calculated as

Pintra = |Eintra |2 = |E0 |2 e− L (1−rF rE )t .

c
(2.19)

Now, we define the time constant of a FPC as the time that it takes to
obtain the intra-cavity power, 1/e. Thus, the time constant, τFPC , can be
written as
L
τFPC = . (2.20)
c(1 − rF rE )
Using Eq. (2.11) and assuming rF , rE ∼ 1, Eq. (2.20) can be rewritten as

1
τFPC = . (2.21)
2πνFWHM

Thus, we can know the necessary νFWHM to measure the time constant of
a FPC.

2.1.5 Response for frequency fluctuation

When the input laser has some frequency fluctuation, we can describe the
amplitude of the input laser as

Ein = Eei(Ωt+δϕ(t)) , (2.22)

where δϕ(t) is the phase fluctuation, which satisfies

∂δϕ(t)
= δΩ(t) = 2πδν(t), (2.23)
∂t
where δν(t) is the laser frequency fluctuation. Assuming δϕ(t) ≪ 1, Eq. (2.22)
can be approximated as

Ein ≃ EeiΩt (1 + iδϕ(t)). (2.24)

12
 2.1 Fabry-Perot cavity

Then, using the inverse Fourier transformation (IFT),

∫ ∞
δϕ(t) = ˜
δϕ(ω)eiωt
dω, (2.25)
−∞

Eq. (2.24) can be rewritten as

∫ ∞
Ein = Ee iΩt
+ iE ˜
δϕ(ω)ei(Ω+ω)t
dω. (2.26)
−∞

Therefore, since the amplitude reflectance of the FPC in Eq. (2.12) is a

function of the laser frequency, the amplitude of the reflected light can be
expressed as
∫ ∞
Er = r(Ω)Ee iΩt
+ iE ˜
r(Ω + ω)δϕ(ω)e i(Ω+ω)t
dω
−∞
( ∫ ∞ )
iΩt r(Ω + ω) − r(Ω) ˜ i(Ω+ω)t
= r(Ω)Ee 1 + iδϕ(t) + i δϕ(ω)e dω .
−∞ r(Ω)
(2.27)

Here, r(x) stands for rFPC (x). Now, considering that the laser angular fre-
quency, Ω, satisfies the resonant condition, Eq. (2.27) can be approximated
as
( )
t2F rE
Er ≃ rF − Eei(Ωt+δϕ(t))
1 − rF rE
( ∫ ∞ )
t2F rE sin γ −iγ ˜
× 1−2 iωt
e δϕ(ω)e dω ,
−∞ rF − rE (rF + tF ) 1 − rF rE e
2 2 −2iγ

(2.28)

˜
where γ = Lω/c. Since we can describe δϕ(ω) as

˜
2π δν(ω)
˜
δϕ(ω) = , (2.29)
iω
Eq. (2.28) can be rewritten as
( ∫ ∞ )
Er rF − rE (rF2 + t2F ) ˜
= 1−i HFPC (ω)δν(ω)e iωt
dω , (2.30)
Ein 1 − rF rE −∞

13
2 Frequency Stabilization

where HFPC (ω) is the response function of a FPC for frequency fluctuation,
and it can be expressed as

4π t2F rE sin γ
HFPC (ω) = e−iγ . (2.31)
ω rF − rE (rF + tF ) 1 − rF rE e
2 2 −2iγ

Here, after calculating the absolute value of HFPC (ω), it can be written as

4π t2F rE | sin γ|
|HFPC (ω)| = √ . (2.32)
ω(1 − rF rE ) rF − rE (rF + tF ) 1 + F sin2 γ
2 2

Assuming γ ≪ 1, Eq. (2.32) can be rewritten as

4πL t2F rE 1
|HFPC (ω)| = √ (√ )2 . (2.33)
c(1 − rF rE ) rF − rE (rF2 + t2F ) FL
1+ c
ω

Then, the behavior of the frequency response of a FPC for frequency fluctua-
tion is the same as that of a 1st-order low pass filter, and its cut-off frequency
can be described as
c ν
fc = √ = FWHM . (2.34)
2πL F 2
Figure 2.4 shows the absolute value and phase of HFPC as a function of the
Fourier frequency.

2.2 Schawlow-Townes limit

The laser is a tool for making coherent light by using the stimulated emission
radiation from atoms. Figure 2.5 shows a schematic diagram of the structure
of a laser.
The fundamental frequency fluctuation the noise source of a laser is the
spontaneous emission of atoms, which causes a random phase fluctuation of
the stimulated emission [28]. The limit to the frequency stability of this effect
is called the ’Schawlow-Townes limit’. The frequency fluctuation spectrum of

14
 2.2 Schawlow-Townes limit

Figure 2.4: Absolute value and phase of the response function, HFPC (ω). The
parameters used for this calculation are those in our optical cavity: rF = rE =
0.999989, tF = 3.24 × 10−3 , and L = 0.2.

this effect can be described as [29]

√
δν 2h √
= νFWHM [/ Hz], (2.35)
ν Pν

where νFWHM , h, P , and ν are the FWHM of an optical cavity, the Planck
constant, the output laser power, and the frequency of a laser. Because of this
limit, it is very difficult to construct a laser whose stability is below the 10−16
level.

However, it is well-known that we can reduce the frequency fluctuation

below the Schawlow-Townes limit by using an active feedback system for fre-
quency stabilization [29]. Thus, the stability of a laser can reach the 10−17
level of stability by using an external reference optical cavity.

15
2 Frequency Stabilization

mirror pump source mirror

output
laser medium

Figure 2.5: Schematic diagram of the structure of a laser. A laser consists of

three parts: a laser medium, a pump source, and an optical cavity. The laser
medium is the source of stimulated emission of radiation. The pump source is
an energy source for making the inverted population of the laser medium. The
optical cavity plays the role to store the light of the spontaneous emission and
the stimulated emission.

2.3 Pound-Drever-Hall method

The Pound-Drever-Hall (PDH) method is a strong method for frequency sta-
bilization using phase modulated light [30].

2.3.1 Error signal

Considering light with phase modulation, the amplitude of this light can be
described to be
∑
∞
i(Ωt+m sin ωm t)
Ein (t) = Ee =E Jn (m)ei(Ω+nωm )t , (2.36)
−∞

where m is the modulation depth and ωm is the angular frequency of the phase
modulation. Jn (m) is a Bessel function of the n-th order. Here, assuming that
m is small, the amplitude of the light can be approximated as
( )
Ein (t) ≃ E J0 (m)eiΩt + J1 (m)ei(Ω+ωm )t − J1 (m)ei(Ω−ωm )t . (2.37)

Equation (2.37) shows that there are three different components, called the
carrier and sidebands, which have angular frequencies of Ω and Ω ± ωm .

16
 2.3 Pound-Drever-Hall method

This light enters the optical cavity. Noting that the amplitude reflectance
of a FPC in Eq. (2.12) is a function of the angular frequency of light, the
reflected power can be written as
 ( )2
Pr (t) = r(δω)J0 (m) + J1 (m) r+ (δω)eiωm t − r− (δω)e−iωm t  Pin , (2.38)

where r(δω) stands for rFPC (δω) and r± (δω) ≡ r(δω ±ωm ). δω is the difference
between the angular frequency of the input light and the nearest resonant angu-
lar frequency of a FPC. Pin is the input laser power. By expanding Eq. (2.38),
we obtain
Pr (t)
= |r(δω)|2 J0 (m)2 + |r+ (δω)|2 J1 (m)2 + |r− (δω)|2 J1 (m)2
Pin
[{ } ]
+ 2J0 (m)J1 (m)Re r∗ (δω)r+ (δω) − r(δω)r− ∗
(δω) eiωm t
[ ∗
]
− 2J1 (m)2 Re r+ (δω)r− (δω)e2iωm t . (2.39)

Here, considering demodulation of the reflected power by sin ωm t, we obtain

an error signal, ϵ, given by
ϵ [ ]
= −2J0 (m)J1 (m)Im r∗ (δω)r+ (δω) − r(δω)r−
∗
(δω) . (2.40)
Pin

Figure 2.6 shows an example of the error signal as a function of the laser
frequency. If the demodulation frequency, ωm , is far from FWHM and δω ≪
νFWHM , we can use the following approximations:

r± (δω) ∼ 1, (2.41)

r(δω) ≃ r(0) + r′ (0)δω. (2.42)

Then, Eq. (2.40) can be rewritten as

ϵ ′
= 4J0 (m)J1 (m)Im[rFPC (0)]δω. (2.43)
Pin

Here, using Eq. (2.15), Eq. (2.43) can be rewritten as

ϵ 8t2F rE L
= J0 (m)J1 (m)δω. (2.44)
Pin (1 − rF rE )2 c

17
2 Frequency Stabilization

Figure 2.6: Example of an error signal according to the PDH method.

Thus, by assuming rF , rE ∼ 1 and t2F ≃ 1 − rF2 , which means that the loss at
the front mirror is sufficiently small, Eq. (2.44) can be further approximated
as
ϵ 4J0 (m)J1 (m)F
= δω. (2.45)
Pin πνFSR
Equation (2.45) shows that we can obtain a signal proportional to the difference
between the laser frequency and the resonant frequency of the FPC. Thus, by
feeding back the error signal to the laser frequency, we can make the laser
frequency resonant with the frequency of the FPC.

2.3.2 Response for frequency fluctuation

The input laser with both some frequency fluctuation and phase modulation
can be described as
Ein = Eei(Ωt+δϕ(t)+m sin ωm t) . (2.46)

18
 2.4 Resonant frequency and cavity length

Using Eq. (2.37), Eq. (2.46) can be approximated as

( )
Ein ≃ E J0 (m)ei(Ωt+δϕ(t)) + J1 (m)ei((Ω+ωm )t+δϕ(t)) − J1 (m)ei((Ω−ωm )t)+δϕ(t) .
(2.47)

Then, by considering the situation that the angular frequency, Ω, satisfies

the resonant condition of a FPC, the amplitude of the reflected light can be
expressed as
( ∫ ∞ )
Er ≃ r(0)J0 (m) 1 − i ˜
HFPC (ω)δν(ω)e iωt
Eei(Ωt+δϕ(t))
−∞
( )
+ J1 (m) r+ (δΩ)e iωm t
− r− (δΩ)e−iωm t Eei(Ωt+δϕ(t)) , (2.48)

where δΩ is the angular frequency fluctuation, defined in Eq. (2.23). Now,

assuming that the frequency fluctuation is much smaller than the modulation
frequency, we can use the approximation r± (δΩ) ∼ 1. Therefore, the reflected
power can be calculated as
 ( ∫ ∞ ) 2
Pr  
= r(0)J0 (m) 1 − i HFPC (ω)δν(ω)e dω + 2iJ1 (m) sin ωm t .
˜ iωt
P −∞
(2.49)
Demodulating the reflected power by sin ωm t, we can obtain the error signal
as [∫ ∞ ]
ϵ ′ ˜ ′ iω ′ t ′
= 4r(0)J0 (m)J1 (m)Re HFPC (ω )δν(ω )e dω . (2.50)
P −∞

This equation means that an error signal for the ω component of the frequency
fluctuation reduces to the transfer function of a FPC. So, the error signal gets
small when the frequency, f, > νFWHM /2.

2.4 Resonant frequency and cavity length

In the previous section, we explained that we can make laser frequency res-
onant with an optical cavity. Thus, the performance of the laser frequency
stabilization is limited by the stability of the resonant frequency of the FPC.
The laser frequency that is resonant with an optical cavity can be written

19
2 Frequency Stabilization

as
nc
ν= , (2.51)
2L
where n is the number of anti-nodes of the standing wave in an optical cavity.
Here, considering a small fluctuation of the cavity length, δL, the variation of
a resonant frequency of a FPC can be expressed as

nc −δL ν
δν = 2
= − δL. (2.52)
2 L L
Equation (2.52) means that the resonant frequency of a FPC changes when the
cavity length fluctuates. Therefore, the stability of the laser frequency that is
tuned to the resonant frequency of a FPC is determined by the cavity-length
fluctuation.

2.5 Noise source

In this section, we describe several noise sources of frequency stabilization.

2.5.1 Thermal noise

The spacer and mirrors of an optical cavity thermally fluctuate because these
components have a finite temperature. The thermal fluctuation causes a
cavity-length fluctuation, called ’thermal noise’. Thermal noise is one of the
fundamental noises of laser frequency stabilization.
Thermal noise has been theoretically well studied and a theory, called
fluctuation-dissipation theorem (FDT), tells that the thermal fluctuation is
related to the loss of a system. According to the FDT, the power spectral
density (PSD) of the thermal noise of a harmonic oscillator can be described
as [ ]
4kB T 1
Gx (ω) = − Im 2 [m2 /Hz], (2.53)
Mω ω − ω02 {1 + iϕ(ω)}
where M, T, ω0 , ϕ(ω) are the mass, temperature, resonant angular frequency,
and loss angle of the harmonic oscillator. If we assume a structure damping

20
 2.5 Noise source

model, a loss angle can be written as

1
ϕ(ω) = , (2.54)
Q

where Q is the quality factor. We can therefore rewrite Eq. (2.53) as

4kB T ω02
Gx (ω) = [m2 /Hz]. (2.55)
M ωQ ω 4 − 2ω 2 ω02 + ω04 (1 + 1/Q2 )

Figure 2.7 is an example of the thermal noise power spectral density. Then,
by considering that the Fourier frequency is far from the resonant frequency
and Q ≫ 1, Eq. (2.55) can be approximated as

4kB T
Gx (ω) = 2
(ω ≪ ω0 ) [m2 /Hz], (2.56)
M ω0 ω Q

4kB ω02 T
Gx (ω) = 5
(ω ≫ ω0 ) [m2 /Hz]. (2.57)
Mω Q

In the case of a real cavity spacer and mirrors, they have many elastic
modes. Thus, the thermal noise of these components is the summation of
the thermal noises of all elastic modes. At frequencies below the resonant
frequencies of the elastic modes, the thermal noise of a cylindrical cavity spacer,
mirror substrate, and mirror coatings can be described as [25]

4kB T L
Gspacer (f ) = 2
[m2 /Hz], (2.58)
ωQspacer 3πR Yspacer
4kB T 1 − σ2
Gsubstrate (f ) = √ [m2 /Hz], (2.59)
ωQsubstrate πYsubstrate w0
4kB T 2(1 + σ)(1 − 2σ)d
Gcoating (f ) = [m2 /Hz], (2.60)
ωQcoating πYsubstrate w02

where QX , YX , R, σ, and w0 are the quality factor of X, Young’s modulus of X,

the radius of the cavity spacer, Poisson’s ratio of a mirror substrate, and the
beam spot size on the mirror. Then, the spectrum of the normalized frequency

21
2 Frequency Stabilization

Figure 2.7: An example of a PSD of the thermal noise. The mass, M , is 1

kg, the temperature, T , is 300 K, and the resonant angular frequency, ω0 , is
2π × 100 rad/s. The blue and red lines have difference quality factors: Q = 104
for the red and Q = 102 for blue.

fluctuation of the thermal noise can be expressed as

√ √
δν Gspacer (f ) 4kB T 1 √
= = [/ Hz], (2.61)
ν L ωQspacer 3LπR2 Yspacer

√ √
δν Gsubstrate (f ) 4kB T 1 − σ2 √
= = √ 2 [/ Hz], (2.62)
ν L ωQsubstrate πL Ysubstrate w0
√ √
δν Gcoating (f ) 4kB T 2(1 + σ)(1 − 2σ)d √
= = [/ Hz]. (2.63)
ν L ωQcoating πL2 Ysubstrate w02

2.5.2 Shot noise

The number of photons on a photodetector has some fluctuation since photons
are quanta and this fluctuation is called the shot noise. This noise is also a
fundamental noise for laser frequency stabilization.

22
 2.5 Noise source

The PSD of the shot noise can be expressed as

GI (f ) = 2eIDC [A2 /Hz], (2.64)

where e is an elementary charge. IDC is the DC current of a photodetector

(PD). Using the DC laser power to a photodiode, PDC , IDC can be written as

ePDC η
IDC = , (2.65)
hν
where h is the Planck constant and η is the quantum efficiency (QE) of a PD.
Thus, we can rewrite the PSD of the shot noise as
( )2
hν 2hν
GP (f ) = GI (f ) = PDC [W2 /Hz]. (2.66)
eη η

In the PDH method, an error signal can be written as in Eq. (2.43). Then, the
spectrum of the frequency fluctuation of the shot noise can be described as
√ √
√ 1 GP (f ) 2hνPDC /η √
Gν (f ) = = ′
[Hz/ Hz]. (2.67)
2π ϵ/δω 8πPin J0 (m)J1 (m)Im[r (0)]

The laser power of the reflected light of the PDH method can be written
as in Eq. (2.39). Then, the average power of the reflected light in the situation
under a laser resonating with a FPC can be written as
[ ]
Pr = PDC = r2 (0)J02 (m) + 2J12 (m) Pin . (2.68)

Here, we use the approximation r± (0) ∼ 1. Therefore, Eq. (2.67) can be

rewritten as
√
√ 2hν [r2 (0)J02 (m) + 2J12 (m)] /(ηPin ) √
Gν (f ) = [Hz/ Hz]. (2.69)
8πJ0 (m)J1 (m)Im[r′ (0)]

Then, the spectrum of the normalized frequency fluctuation of the shot noise
can be obtained as
√ √
δν Gν (f ) 2h [r2 (0)J02 (m) + 2J12 (m)] /(ηPin ν) √
= = [/ Hz]. (2.70)
ν ν 8πJ0 (m)J1 (m)Im[r′ (0)]

23
2 Frequency Stabilization

Here, if we assume rF , rE ∼ 1 and the amplitude loss of an input mirror is

small enough, Eq. (2.45) is satisfied. Thus, Eq. (2.70) can be approximated as
√
δν 1 hcλ √
= [/ Hz]. (2.71)
ν 8J0 (m)LF ηPin

2.5.3 Acceleration noise

The acceleration of a FPC causes an elastic deformation, and changes the
cavity length. The main source of the cavity acceleration is a seismic vibration.
A spectrum of the cavity length fluctuation caused by the acceleration of
a FPC can be written as
√ √ √
Gx (f ) = AL Ga (f ) [m/ Hz], (2.72)
√
where Ga (f ) is the spectrum of the cavity acceleration and A is the cou-
pling constant, called the vibration sensitivity, from acceleration to the laser
frequency fluctuation. The vibration sensitivity depends on the shape of the
cavity spacer, the way of cavity support, the material of a FPC and the cavity
support, etc.
We can thus describe the spectrum of the normalized frequency fluctuation
of the acceleration noise as
√
δν Gx (f ) √ √
= = A Ga (f ) [/ Hz]. (2.73)
ν L

2.5.4 Temperature-fluctuation noise

The temperature fluctuation of an optical cavity leads to thermal expansion,
which changes the cavity length.
The spectrum of the temperature-fluctuation noise can be expressed as
√ √ √
Gx (f ) = αL GT (f ) [m/ Hz], (2.74)
√
where α and GT (f ) are the coefficient of thermal expansion (CTE) and
the temperature-fluctuation spectrum of the FPC. Then, the spectrum of the
normalized frequency fluctuation of the temperature-fluctuation noise can be

24
 2.5 Noise source

written as √
δν Gx (f ) √ √
= = α GT (f ) [/ Hz]. (2.75)
ν L

2.5.5 Residual amplitude modulation noise

In practice, phase modulated light by using an EOM has a superfluous ampli-

tude modulation at the same frequency of the phase modulation. There are
many sources of this noise, such as scattered light, the temperature fluctuation
of the EOM, etc. So, in this part, we show its effect.
Considering light with a phase and amplitude modulation, we can express
its amplitude as [33]

Ein = E (1 + ma sin(ωm t + ϕ)) ei(Ωt+m sin ωm t)

( )
≃ E (1 + ma sin(ωm t + ϕ)) J0 + J1 eiωm t − J1 e−iωm t eiΩt (2.76)

where ma is an amplitude-modulation index, and ϕ is a phase difference be-

tween a phase modulation and an amplitude modulation. J0 and J1 represent
J0 (m) and J1 (m), respectively. Then, considering a laser frequency resonant
with the FPC, the amplitude of the reflected light can be written as

Er = E (r(0)J0 + 2iJ1 sin ωm t) eiΩt

− ima E (iJ0 sin(ωm t + ϕ) − r(0)J1 cos ϕ + J1 cos(2ωm t + ϕ)) eiΩt .
(2.77)

Here, we assume that r+ (0), r− (0), r+ (ωm ), r− (−ωm ) ∼ 1. Therefore, a

reflected power can be approximated as

Pr
≃ (r2 (0)J02 + 2J12 − 2J12 cos 2ωm t)
Pin
( )
+ 2 r(0)J02 + 2r(0)J12 + J12 ma cos ϕ sin ωm t
( )
+ 2 r(0)J02 + J12 ma sin ϕ cos ωm t − 2J12 ma sin(3ωm t + ϕ). (2.78)

Here, we use an approximation: m2a ≃ 0.

Then, by demodulating the reflected power by sin ωm t, we can obtain the

25
2 Frequency Stabilization

error signal as
[ ]
ϵRAM = 2 r(0)J02 + 2r(0)J12 + J12 ma Pin cos ϕ, (2.79)

Therefore, the spectrum of an error signal of the RAM noise can be described
as
√ [ ] √
GϵRAM (f ) = 2 r(0)J02 + 2r(0)J12 + J12 Pin cos ϕ Gma (f ), (2.80)
√
where Gma (f ) is a spectrum of the amplitude-modulation index.
Therefore, the frequency fluctuation spectrum of the RAM noise can be
expressed as

√ r(0)J02 + 2r(0)J12 + J12 √ √

Gν (f ) = cos ϕ Gm (f ) [Hz/ Hz]. (2.81)
4πJ0 J1 Im[r′ (0)] a

Thus, the spectrum of a normalized frequency fluctuation of the RAM noise

can be written as
√
δν Gν (f ) r(0)J02 + 2r(0)J12 + J12 √ √
= = cos ϕ Gm (f ) [/ Hz]. (2.82)
ν ν 4πνJ0 J1 Im[r′ (0)] a

Now, assuming that rF , rE ∼ 1, and that the amplitude loss of the input mirror
is sufficiently small, we can use the approximation r(0) ∼ 0 and Eq. (2.45).
Thus, Eq. (2.82) can be rewritten as

δν J1 (m)νFSR √ √
= cos ϕ Gma (f ) [/ Hz]. (2.83)
ν 4νJ0 (m)F

2.5.6 Residual gas noise

The inside of an optical cavity is held at a high vacuum for reducing any
fluctuation of the refractive index in a FPC. However, due to residual gas,
there is some fluctuation of the refractive index, which changes the resonant
frequency of a reference cavity.
A PSD of an effective cavity length fluctuation of the residual gas noise

26
 2.5 Noise source

can be described as [34]

( )( )
2(n0 − 1)2 p T0 τR √
Gx (f ) = [m2 / Hz], (2.84)
(NA /V0 )πw2 p0 T 2

where n0 , A0 , V0 , p0 , T0 are the refractive index of the gas, the Avogadro

constant, and the volume, pressure, and temperature of the gas under the
standard condition; τR is defined as
√
w T0
τR ≡ , (2.85)
u0 T

where u0 is the mean speed of the gas molecules under the standard condition.
Then, the spectrum of the normalized frequency fluctuation of the residual gas
noise is
√ √ ( )( )
δν Gx (f ) n0 − 1 V0 p T0 √
= = τR [/ Hz]. (2.86)
ν L Lw πNA p0 T

2.5.7 Doppler noise

In general, we transfer the frequency-stabilized light from a FPC to an appli-
cation system. However, while a frequency stabilized laser is transferred from
a FPC to an application system, the beam undergoes some phase modulation
because of the optical path-length fluctuation. This phase noise is called the
Doppler noise.
Now, considering that the beam, EeiΩt , travels some distance, the ampli-
tude of the light after traveling is described as
nl
EAT = Eei(Ωt+ λ ) , (2.87)

where n and l are the refractive index and the path length along which the
light travels. Then, the angular frequency of the light after traveling can be
written as ( )
′ dn dl 1
Ω =Ω+ l+ n . (2.88)
dt dt λ
The path-length fluctuation is mainly caused by seismic vibration and tem-

27
2 Frequency Stabilization

perature fluctuation of the optical path, and the fluctuation of the refractive
index is basically caused by winds and sounds. From Eq. (2.88), the angular
frequency fluctuation of the Doppler noise can be described in the frequency
domain as ( ) iω
′ ˜
Ω̃ (ω) = lñ(ω) + nl(ω) . (2.89)
λ
Then, the frequency-fluctuation spectrum of the Doppler noise can be ex-
pressed as
√ if √ 2 √
Gν (f ) = l Gn (f ) + n2 Gl (f ) [Hz/ Hz], (2.90)
λ
where Gn (f ) and Gl (f ) are the spectrum of the reflective index and that of the
path length, respectively. Therefore, the spectrum of the normalized frequency
fluctuation of the Doppler noise is written as

δν √ if √ 2 √
= Gν (f )/ν = l Gn (f ) + n2 Gl (f ) [/ Hz]. (2.91)
ν c

2.5.8 Other noises

In this subsection, we consider those noise sources that are hard to be formu-
lated.

Laser-power fluctuation noise

If the mirrors of a FPC have some absorption loss, a part of the input laser
power changes to heat at the mirror surface. This effect makes the coatings
locally heat up, which causes thermal expansion. In this way, the input laser-
power fluctuation is changed to a cavity-length fluctuation.
However, formulating this effect has some difficulty, since any local heating
of the coatings highly depends on the configuration of an optical cavity. We
should thus check this noise effect in a real configuration of our experiment,
and stabilize the intra-cavity laser power.
On the other hand, due to the very large thermal conductivity of monocrys-
talline silicon, the laser-power fluctuation noise should be small in our exper-
iment. Moreover, since the laser-power fluctuation noise can be reduced by
changing the operation parameters, like laser power, we can reduce the effect

28
 2.5 Noise source

of this noise after constructing our experimental setup. The laser-power fluc-
tuation noise can therefore be well reduced for reaching a 10−17 level stability.

Circuit noise

For frequency stabilization, the electrical feedback system, consisting of a PD,

amplifiers, and analog circuit filters, is necessary. However, such a system has
some electrical noises, which causes frequency fluctuation of the laser.
There are many electrical noise sources such as the dark current of the
PDs, noises of the op-amps, thermal noise of the resistance, etc. Then, the
formulation of the circuit noise is very difficult. So, measuring the circuit noise
in the frequency stabilization and estimating the noise level of an electrical
circuit are necessary.
However, the circuit noise can be reduced by changing the operation pa-
rameters like the laser power and the modulation depth. We can therefore
reduce this noise to below the 10−17 level.

29
2 Frequency Stabilization

2.6 Summary of this chapter

• We explained the properties of a FPC for characterizing the absolute
frequency value and frequency fluctuation.

• The PDH method is a strong method for obtaining the signal of the
difference between laser frequency and resonant frequency of an optical
cavity using phase-modulated light.

• The stability of the laser frequency resonating with a FPC is determined

by the cavity length stability of a reference cavity.

• We considered various noise sources of the cavity length fluctuation and

the form of the spectrum of the normalized frequency fluctuation using
the parameters of a FPC, laser power, etc.

30
 Chapter 3
Experimental Design

In this chapter, we show the experimental setup for achieving 10−17 level fre-
quency stabilization. We explain the design concept of our experiment in
Section 3.1. We describe the noise budget of our experimental setup in Sec-
tion 3.2. Several designs mentioned in this chapter were developed by my
collaborators: Dr. Yoichi Aso and Dr. Noriaki Ohmae. Their contributions
are shown in each subsection.

3.1 Design concept

3.1.1 Overview of the experimental design

A schematic of our experimental setup for the frequency stabilization is shown
in Fig. 3.1. Outside the cryostat, there is an input laser system on an optical
table. In this part, we produce phase-modulated light and transport it to
the cryostat. Moreover, there are also Doppler noise-cancellation systems [35]
located on this optical table.
In a vacuum chamber, there are several components for reducing any cavity
length fluctuation. At the bottom of a cryostat, there are rubber legs for high-
frequency vibration isolation. On top of them, there is an active vibration
isolation table (AVIT) for low frequency vibration reduction. There are some
optics on the AVIT for matching the laser mode to the cavity eigen mode.
There are also thermal-isolation stacks on the AVIT for making a temperature
gradient from 300 K to 18 K, which is close to the zero-cross point temperature
of the CTE of monocrystalline silicon. The optical cavity is put on thermal-

31
3 Experimental Design

refrigerator

cryostat

optical table

radiation
feed back shield
PD
PD PD
cavity

ECDL EOM BS

thermal
insulation
stack

AVIT
fiber

rubber leg

Figure 3.1: Schematic of our experimental setup. See the text for details.

isolation stacks, and there are three radiation shields around it. At the top of
the cryostat, there is the refrigerator for cooling the reference cavity.

3.1.2 Silicon optical cavity

The optical-cavity design is very important for reducing several noises, such
as the thermal noise, the acceleration noise, and the temperature fluctuation
noise. The cavity of our experiment was mainly designed by Dr. Aso. He
decided on the material of the cavity, the shape of the cavity spacer, the
direction of the crystal axis, etc.
One of the fundamental noises for the frequency stabilization is the thermal
noise of the cavity components. Thus, reducing the thermal noise is neces-
sary for making an ultrastable laser. Considering Eq. (2.61), Eq. (2.62), and
Eq. (2.63), the main approaches for reducing the thermal noise are as follows:

1. using a high mechanical quality factor material (large Q),

2. cooling the optical cavity (small T ),

3. using a long optical cavity (large L).

32
 3.1 Design concept

Therefore, we use monocrystalline silicon, which has a very high mechanical

quality factor [36] for the material of the cavity spacer and mirrors. Moreover,
we cool the optical cavity to around 18 K. Since the mechanical quality factor of
monocrystalline silicon at a very low temperature is larger than that at a room
temperature [36], we can reduce the thermal noise effectively. Furthermore,
since monocrystalline silicon has a zero-cross point temperature of the CTE
at around 18 K, reducing the temperature-fluctuation noise is also expected
[23, 37]. The cavity length is 20 cm, which is long enough for our goal.
Considering Eq. (2.62) and Eq. (2.63), it is also useful to make the beam-
spot size large for reducing the thermal noise of the mirrors and coatings.
Therefore, we use a curved mirror, whose radius of curvature is 1 m for the
input mirror and a flat mirror for the end mirror of the reference cavity (see
Appendix A). Moreover, the larger Poisson’s ratio is also effective for reducing
the thermal noise from the mirror substrates and coatings. Since the Poisson
ratio of monocrystalline silicon along the [100] axis is larger than that along
the [111] axis [39], the spacer and mirrors are made of monocrystalline silicon
with a [100] orientation along the optical axis of the cavity.
Taking into account Eq. (2.71) and Eq. (2.83), a larger finesse is better for
reducing the shot noise and RAM noise. Therefore, the coatings of the mirrors
are dielectric multilayers of SiO2 and Ta2 O5 , which have been currently used
for a very high-finesse cavity.
The optical cavity of our experiment is shown in Fig. 3.2. The mirrors are
optically contacted to the cavity spacer. The radius of the cavity spacer is 20
cm long.

33
3 Experimental Design

spacer

mirror

cavity stage

Figure 3.2: Image of a silicon optical cavity. The cavity spacer and mirrors
are all made of silicon. They are supported by four rubber legs (blue part of
this figure). These are on a silicon stage for reducing any cooling stress due to
the difference in the CTE between the optical cavity and the cavity stage.

3.1.3 Vibration isolation system

The acceleration noise is one of the most problematic technical noise sources
for the frequency stabilization. There are therefore two vibration isolation
systems in our setup.
One is the passive vibration isolation system (PVIS) using rubber legs. At
the bottom of the cryostat, there are three rubber legs for high-frequency vi-
bration isolation. At the frequency which is higher than the resonant frequency
of the rubber legs, seismic vibration is hardly transferable to the top of them.
The other is the active vibration isolation system (AVIS) using a hexapod-
type active vibration isolation table (AVIT). Figure 3.3 is an image of the
AVIT. It has six legs and there are PZTs in each leg. Thus, we can control
six degrees of freedom to change the voltage applied to the PZTs. In addition,
there are six vibration sensors on the top plate of the AVIT for measuring the
vibration. Therefore, by feeding back these vibration sensors’ signals to the
voltage applied to PZTs, we can reduce any vibration at the top of the AVIT.

34
 3.1 Design concept

Figure 3.3: Design of the AVIT. Red parts are PZT legs and blue structures
are aluminum pots for maintaining the pressure around the vibration sensors
at atmospheric pressure.

Such an AVIS enables us to reduce any vibration at low frequencies easily,

while low-frequency vibration reduction using a PVIS is difficult.
Moreover, the AVIS can easily reduce vibration from a refrigerator, while
it’s difficult for the PVIS to decrease, because such vibration is directly trans-
mitted to a cooling target. Thus, we can reduce vibration not only from seismic
motion, but also from a refrigerator.

3.1.4 Cryogenic refrigerator

The refrigerator can be a problematic vibration and temperature-fluctuation
source. We thus use a helium recondensation-type pulse tube refrigerator. It is
expected to be a low vibration and temperature fluctuation refrigerator. This
cryogenic refrigerator was selected by Dr. Aso to reduce these noises.
Figure 3.4 is a schematic diagram of our refrigerator. The vibration level
of a cold head of the pulse tube-type refrigerator is smaller than that of other
refrigerators like a Stirling refrigerator, because the pulse tube-type cryocooler
has no fundamental vibration section at the cryogenic part. Moreover, since
there is no mechanical connection between the cold heads and the cooling
targets in our refrigerator, the vibration of cold heads does not easily transfer
to an optical cavity [38].
A helium recondensation-type pulse tube refrigerator also has a benefit
concerning the temperature-fluctuation noise. A pulse tube refrigerator makes

35
3 Experimental Design

connected to
compresser
supply port
for He gas

1st stage
1st base
( 100 K)
gas He

2nd stage

liq. He

2nd base ( 4 K)

Figure 3.4: Schematic diagram of the refrigerator used in our experiment. Cold
heads of the refrigerator are inside of the chamber. This chamber is filled with
He gas. The 1st stage and the 1st base perform heat exchange via He gas. If
the temperature of the 2nd stage is low enough, a part of He gas is changed to
liq. He, cooling the 2nd base. The temperature of the 1st base is about 100 K
and that of the 2nd base is about 4 K.

the temperature fluctuation synchronize with its compression. However, by

using liquid helium to transfer heat, temperature fluctuation of the 2nd stage
hardly transmits to the 2nd base because the liquid helium works as a heat
bath.

3.1.5 Thermal isolation system

The temperature-fluctuation noise is one of the most problematic noises for
frequency stabilization.
In our experiment, we use an AVIT for active vibration reduction. How-
ever, since the vibration sensor on AVIT cannot be used at the cryogenic
temperature, only the part around an optical cavity can be cooled. We thus
use thermal-isolation stacks, which are original structures in our experiment
for making a high-temperature gradient between a reference cavity and the
AVIT.
Figure 3.5 is a schematic diagram of the thermal-isolation stacks. The

36
 3.1 Design concept

Connected to Connected to
1st base 1st base
Connected to Inner radiation shield Connected to
2nd base & silicon optical cavity 2nd base

Top plate of AVIT

Figure 3.5: Structure of thermal-isolation stacks. The gray parts are aluminum
plates coated by nickel. The dark black parts are SUS inserts that have a 3
mm diameter hall. The blue parts are 5 mm BK7 spheres. There are five
stacks; the third floor is connected to the 1st base of the refrigerator and the
forth floor is linked to the 2nd base of the refrigerator. The temperatures of
the 4th and 5th floors are about 18 K and that of the 3rd floor is about 100 K.
The temperature of the top plate of an AVIT is almost the same as the room
temperature.

thermal-isolation stacks are made of three parts: aluminum plates, SUS inserts,
and glass balls. Aluminum plates are coated by nickel for making the emissivity
low. A SUS insert has a small hall, and a glass ball is put on this hall. In this
structure, since the supporting surface between the hall and the sphere is a
circle, the lower plate and upper plate has an ideal one-dimensional connection.
Then, the thermal conduction between two plates can be very small, and it is
possible to make a high-temperature gradient between them. The conceptual
idea of this thermal-isolation stacks was invented by Dr. Aso.
Another enemy for cooling is heat transfer from radiation. The heat flux
from radiation can be described as

Q = εσB AT 4 , (3.1)

where ε, σB , A, and T are the emissivity, Stefan-Boltzmann constant, surface

area, and temperature. Thus, even if only a small area sees 300 K parts, the
heat flux becomes very large and disturbs the cooling. Then, there are three
radiation shields in our experiment to reduce the heat flux from radiation.

37
3 Experimental Design

refrigerator

super
insulation outer ( 100 K)

middle ( 18 K) spacers
inner ( 18 K)

BK7
window
transmitted light
input light

thermal isolation stacks

Figure 3.6: Schematic diagram of the radiation shields. The middle and outer
radiation shields are connected to stacks with thin OFC wires and aluminum
foils for being able to move an AVIT independently from the radiation shields.
The temperature of the outer shield is 100 K, and those of inner and middle
shields are 18 K.

Figure 3.6 shows a schematic diagram of the radiation shields. An inner

radiation shield neighbors an optical cavity, and its temperature is almost the
same as that of a reference cavity. It is made of aluminum for reducing the mass
of the inner radiation shield because we should decrease the mass to below the
allowable level for the AVIT. A middle radiation shield is connected to the 2nd
base of the refrigerator through the SUS spacer which plays a role for making a
temperature gradient between the 2nd base and the middle radiation shield. It
is made of oxide-free copper (OFC), which has very high thermal conductivity.
Moreover, the surface of the middle radiation shield is coated by gold because
OFC is easily oxidized. Furthermore, since the emissivity of gold is very small,
it can reduce the heat flux from the radiation. An outer radiation shield is
connected directly to the 1st base of the refrigerator. It is made of aluminum,
and there are super insulation sheets around it for reducing the heat flux from
the radiation.
Radiation shields have halls for the input and transmitted light. However,
even though the halls are very small, the heat flux from the room-temperature

38
 3.1 Design concept

part is very severe. We therefore use BK7 glass windows on the outer and
middle radiation shields. BK7 has a high transmittance of around 1.4 µm
and a low transmittance of over 2 µm. Therefore, only a laser beam can be
transmitted through BK7 windows, and we can reduce the heat flux from
radiation.

3.1.6 Input and output optics

The input and output optics is optics outside of the cryostat. This part has
many roles, for example sending laser light to the cavity, obtaining PDH sig-
nals for frequency stabilization, and reducing several noises. One of the most
important roles in this part is to measure any instability of our laser. Since
the laser instability is evaluated by measuring the beat frequency between two
lasers, we should prepare another laser that is no less stable than our laser.
Such a stable laser was invented Dr. Ohmae. We can therefore measure the
frequency stability of our laser directly.
Our laser source is an external cavity diode laser (ECDL); its wavelength
is about 1397 nm for applying to a Sr optical lattice clock. Figure 3.7 is a
schematic diagram of our ECDL. Mainly, there are three components: the
laser diode (LD), the diffraction grating, and the PZT actuator. When the
light of the LD enters the diffraction grating, the 1st-order diffracted light
returns to the LD. The returned light resonates with the cavity configured
by the diffraction grating, and the LD and provides the optical feedback to
the LD. When we change the angle of the diffraction grating using the PZT
actuator, the wavelength of the resonated beam is changed. Moreover, we
can change the frequency of the ECDL to change the current applied to the
LD. Therefore, the wavelength of the ECDL is tunable over a wide range, for
example several tens of nanometers. In our ECDL, we use a grating with the
high diffraction efficiency: about 50%. The output power is about 70 mW at
1397 nm when the supply current to the LD is about 500 mA.
Figure 3.8 is the design of the optical system used in our experiment. We
make phase modulated light with the EOM, and obtain the PDH signal using
the PD1. The PDH signal is fed back to the PZT actuator and the LD current,
and our laser frequency is stabilized to the resonant frequency of the optical

39
3 Experimental Design

diffraction grating temperature

GR25-1208 (THORLABS)
0th order controller
TED200C
PZT (THORLABS)
1st order

to input and
LD output optics
SAF1093H
(THORLABS)

base plate
laser box

DC current source
Signals from the circuit
for frequency stabilization LDC205C
(THORLABS)

Figure 3.7: Schematic diagram of our ECDL. A pink part is a laser diode, and
a blue part is a diffraction grating. A green part is the PZT actuator. These
components are put in the laser box for blocking winds and sounds. The
LD and the PZT are connected with the feedback circuit for the frequency
stabilization (details in Appendix C). The temperature of the LD is stabilized
using the temperature controller. Several model numbers of the components
are shown in this figure.

cavity. The frequency fluctuations at low frequencies and high frequencies are
fed back to the PZT and the LD, respectively. Doppler-noise cancellation using
the differential arm length Michelson interferometer (DAMI) [35] is achieved
using the signals of the PD2 and the PD3. These signals are fed back to the
RF frequencies applied to the AOM1 and AOM2. Details of the Doppler noise
cancellation are given in Section 3.2.7. Laser-frequency stabilization is achieved
by feeding back the signal of the PD5 to the amplitude of the RF signal applied
to the AOM1. Direct frequency instability evaluation is achieved by measuring
the beat frequency between our laser and the stable laser using the PD6. The
instability of the stable laser is at the 10−16 level on a time scale from 0.1
second to 10 second when the linear drift of the laser is subtracted. The beat
frequency is measured using the frequency counter, and the Allan deviation of
the beat signal is calculated according to its definition (see Appendix B). Our
cavity mirrors have some wedge for separating the reflected light at the HR
surface from that at the AR surface. The reflected light at the AR surface is
detected by the PD4, and used for the RAM noise estimation.

40
 3.2 Noise budget

ECDL (1397 nm) isolator QWP HWP PBS HWP coupler

wind shield

collimeter
cryostat PD2

PD5

isolator
collimeter lens

coupler

isolator

HWP
AOM1

EOM
HWP
BS PD1

PBS
AOM2

wind shield coupler BS

frequency PD4
counter BS
PD6 collimeter PD3

stable BS
laser

Figure 3.8: Design of the input and output laser system. A stable laser has
a 10−16 level of frequency stability. All optical fibers in this experiment are
polarization-maintaining fibers. The PD6 is a detector for measuring the beat
signals between our laser and the stable laser. The beat frequency is measured
by a frequency counter.

3.2 Noise budget

In this section, we show the noise budget of our experimental design.

3.2.1 Thermal noise

The thermal noise of a FPC depends on the materials composing the optical
cavity and its configuration, such as the cavity length, and curvatures of mir-
rors. Table 3.1 gives our experimental design parameters that determine the
thermal noise levels. Using these parameters and Eq. (2.61) to Eq. (2.63), the
thermal noise levels can be calculated as

δν 7.2 × 10−20 √
≃ √ / Hz (spacer), (3.2)
ν f

δν 2.3 × 10−18 √
≃ √ / Hz (substrate), (3.3)
ν f
δν 2.0 × 10−17 √
≃ √ / Hz (coatings). (3.4)
ν f

41
3 Experimental Design

cavity length L = 0.2 m

cavity radius R = 0.1 m
Q factor Qspacer = Qsubstrate = 1.0 × 107 [36]
Qcoating = 2.5 × 103 [25]
Young’s modulus Yspacer = Ysubstrate = 1.69 × 1011 Pa [39]
Poisson’s ratio σ = 0.28 [39]
beam spot size on a front mirror wF = 4.2 × 10−4 m
beam spot size on an end mirror wE = 4.7 × 10−4 m
thickness of coatings d = 1.2 × 10−5 m

Table 3.1: Parameters for calculating the thermal noise.

3.2.2 Shot noise

The shot-noise level depends on the cavity length, input laser power, wave-
length, modulation depth, and the QE of a PD. Table 3.2 gives our experi-
mental parameters that determine the shot-noise level. Then, using Eq. (2.71),
we can calculate the shot-noise level as
δν √
≃ 3.6 × 10−19 / Hz. (3.5)
ν

wave length λ = 1.4 × 10−6 m

cavity length L = 0.2 m
cavity Finesse F = 1.0 × 105
modulation depth m = 0.27 rad
input laser power P = 1.0 × 10−4 W
QE of a PD η = 0.88

Table 3.2: Parameters for calculating the shot noise.

42
 3.2 Noise budget

3.2.3 Acceleration noise

The noise level of the acceleration noise is determined by the vibration sensi-
tivity and the acceleration of a reference cavity.
There are too many factors to determine the vibration sensitivity theoret-
ically, for example, the configuration of a cavity spacer, support points, and
support materials of an optical cavity. However, a vibration sensitivity below
1 × 10−11 /(m/s2 ) in all translational axes was achieved in 2012 [12]. We there-
fore assume the vibration sensitivity of our cavity design to be 1×10−11 /(m/s2 )
in all translational axes.
The seismic vibration level depends on the place, time, weather, and so on.
However, the typical RMS seismic motion spectrum density can be described
as [40]
√ 1.0 × 10−7 √
Gx (f ) ≃ m/ Hz. (3.6)
f2
Then, the acceleration from the seismic motion can be expressed as
√ √ √
Ga (f ) = (2πf )2 Gx (f ) = 3.9 × 10−6 (m/s2 )/ Hz. (3.7)

Since the vibration at the place of a FPC is isolated passively and actively, the
acceleration of a reference cavity is smaller than Eq. (3.7).
The passive vibration isolation can reduce the vibration at high frequency.
The equation of motion of the mass on the rubber legs can be written as

ẍ + 2ξω0 ẋ + ω02 (x − X) = 0, (3.8)

where x and X are the displacements of the upper and lower parts of the
rubber legs, respectively. ω0 is the resonant angular frequency and ξ is a
damping ratio. In the frequency domain, Eq. (3.8) can be written as
( )
−ω 2 + 2iξω0 ω + ω02 x̃ = ω02 X̃. (3.9)

Then, the transfer function of the vibration can be described as

x̃ ω02
HP (ω) = = , (3.10)
X̃ −ω 2 + iωω0 /Q + ω02

43
3 Experimental Design

Figure 3.9: Example of the transfer function of passive vibration isolation.

ω0 = 2π × 9 rad/s and Q = 3.0. These are the same values as the design
parameters.

where Q = 1/2ξ is the Q factor. Figure 3.9 is an example of the vibration

isolation ratio of the passive vibration isolation using the rubber legs. Thus,
the vibration level on the rubber legs is smaller than that on ground at high
frequency.

On the other hand, active vibration isolation reduces the seismic vibration
at low frequency. The designed open-loop transfer function of an AVIS is as
follows:
if (1 + if /f0 ) (1 + if /f1 )
HA (f ) = G0 . (3.11)
(1 + if /f2 )4
Figure 3.10 shows the absolute value of the designed HA (f ).

Then, the acceleration noise level of our experimental design can be written
as  
δν 
−17  HP

= 3.9 × 10  . (3.12)
ν 1 + HA 

Table 3.3 shows the parameters for calculating the acceleration noise.

44
 3.2 Noise budget

Figure 3.10: Absolute value of the designed open-loop transfer function of an

AVIS. f0 = 0.5 Hz, f1 = 20 Hz, f2 = 3.3 Hz, and G0 = 3

vibration sensitivity A = 10−11 /(m/s2 )

√ √
seismic acceleration spectrum Ga (f ) = 3.9 × 10−6 (m/s2 )/ Hz
Passive vibration isolation ω0 = 2π × 9 rad/s
Q = 3.0
Active vibration isolation G0 = 3
f0 = 0.5 Hz
f1 = 20 Hz
f2 = 3.3 Hz

Table 3.3: Parameters for calculating the acceleration noise.

45
3 Experimental Design

3.2.4 Temperature-fluctuation noise

The temperature-fluctuation noise depends on the CTE and the temperature

fluctuation of an optical cavity.
Figure 3.11 shows the thermal structure of our experiment. The sources
of temperature fluctuation of a reference cavity are the temperature of the
laboratory, the 1st stage, and the 2nd stage of the refrigerator. However, the
temperature fluctuations of the 1st stage and the laboratory are transferred
through an outer shield. We thus don’t have to consider the temperature
fluctuations of the 1st stage and must do so for the laboratory, but the 2nd
stage and the outer shield. The temperature fluctuation of the 2nd stage is
transferred to the cavity through a middle shield. The temperature fluctuation
of an outer shield is also transferred to the cavity via a middle shield.
Considering the heat transfer through solid materials, any temperature
fluctuation at high frequency diminishes during the heat transfer according
to the thermal resistance and the heat capacity. The transfer function of the
temperature fluctuation in such a situation can be described as

HT 1
HX→Y (f ) = , (3.13)
1 + if RX,Y CY

where the subscript X → Y means the situation where temperature fluctuation

of X travels to Y. RX,Y and CY are the thermal resistance between X and Y,
and the heat capacity of Y, respectively.
On the other hand, considering the heat transfer from radiation, the heat
flux between two materials, X and Y, can be described as
( )
X,Y σB AX TX − AY TY ,
QX→Y = εeff 4 4
(3.14)

where
εX εY
εeff
X,Y = (3.15)
εX + εY − εX εY
is the effective emissivity between X and Y. AX and TX are the surface area
and the temperature of X. Then, the fluctuation of the heat flux can be written
as
( )
δQX→Y = 4εeff X,Y σB AX TX δTX − AY TY δTY .
3 3
(3.16)

46
 3.2 Noise budget

Room
Radia!on 1st stage
Room → Outer
Outer shield
Radia!on
Outer → Middle 2nd stage

Radia!on FPC
Middle → FPC

Middle shield

Figure 3.11: The designed thermal structure of our experiment.

We can therefore describe the temperature fluctuation of Y as

∫ ( ∫ ∫ )
δQX→Y 4εeff
X,Y σB
δTY = dt = AX TX δTX dt − AY TY δTY dt .
3 3
CY CY
(3.17)
Now, considering Eq. (3.17) in the frequency domain, it can be rewritten as

4εeff ( )
X,Y σB
δT˜Y = AX TX3 δT˜X − AY TY3 δT˜Y . (3.18)
iωCY

The transfer function of the temperature fluctuation from radiation is thus

expressed as

RT δT˜Y AX TX3 1
HX→Y (f ) = = . (3.19)
δT˜X AY TY 1 + if CY 2εeff σπB AY T 3
3
X,Y Y

Now, assuming that the temperature fluctuation spectra of the 2nd stage
of the refrigerator and the outer radiation shield to be

√ 1.0 × 10−4 √
GT (f ) = K/ Hz, (3.20)
f

47
3 Experimental Design

the noise level of the temperature-fluctuation noise can be written as

δν √ ( HT )
= α GT (f ) HMiddle→FPC RT
+ HMiddle→FPC ×
ν √
2 2
√
HT HT RT
(H2nd→Middle ) + (HOuter→Middle + HOuter→Middle ) [/ Hz].
(3.21)

Table 3.4 gives the parameters of our experiment for calculating the temperature-
fluctuation noise.

temperature TOuter = 100 K

TMiddle = TFPC = 18 K
√ √
temperature fluctuation GT (f ) = 1.0 × 10−4 /f K/ Hz
thermal resistance R2nd,Middle = 70 K/W
RMiddle,FPC = ROuter,Middle = 103 K/W
heat capacity CMiddle = 130 J/K
CFPC = 120 J/K
surface area AOuter = 0.75 m2
AMiddle = 0.63 m2
AFPC = 0.51 m2
emissivity εOuter = εFPC = 0.1 [41]
εMiddle = 0.04 [41]
cavity CTE α = 1.0 × 10−9 /K

Table 3.4: Parameters for calculating the temperature-fluctuation noise

48
 3.2 Noise budget

3.2.5 Residual amplitude modulation noise

The noise level of the RAM noise can be described as Eq. (2.83). Therefore,
by using the parameters given in Table 3.5, it can be calculated as

δν √
= 1.3 × 10−17 / Hz. (3.22)
ν

modulation depth m = 0.27 rad

√ √
amplitude-modulation index fluctuation Gma (f ) = 1.1 × 10−5 / Hz
phase difference ϕ = 0 rad
cavity finesse F = 1.0 × 105
cavity FSR νFSR = 7.5 × 108 Hz
laser frequency ν = 2.14 × 1014 Hz

Table 3.5: Parameters for calculating the RAM noise.

3.2.6 Residual gas noise

The effect of the residual gas depends on the temperature, pressure, beam size,
kind of residual gas, etc. Using Eq. (2.86) and the parameters in Table 3.6,
the noise spectrum of the residual gas noise can be calculated as

δν √
= 1.0 × 10−19 / Hz. (3.23)
ν

pressure p = 1 × 10−4 Pa
temperature T = 18 K
beam size w = 4.5 × 10−4 m
cavity length L = 0.2 m
refractive index n0 = 1 + 2.52 × 10−4

Table 3.6: Parameters for calculating the residual gas noise. We assume the
kind of residual gas to be water.

49
3 Experimental Design

3.2.7 Doppler noise

The laser light acquires frequency noise, like Eq. (2.91), while traveling. How-
ever, when Doppler noise cancellation is applied, the effect from the fluctuation
of the traveling path will be negligible, and the effect from the length fluctua-
tion of a short arm of a DAMI will be significant [35].

Figure 3.12 is a conceptual image of the Doppler noise cancellation. The

amplitude of a laser beam after M1 can be written as
( )
Ω(L1 +L2 )+ΩA L2
i (Ω+ΩA )t−
EAM1 = tBS tM1 Ee c
, (3.24)

where tBS , tM1 are the transmittances of a BS and an M1 mirror, respectively.

ΩA is the shift frequency of an AOM.

The amplitude of the light reflected at an M1 mirror and coming back to

the BS can be written as
( )
2(Ω+ΩA )(L1 +L2 )
i (Ω+2ΩA )t−
ERM1 = tBS rM1 Ee c
, (3.25)

where rM1 is the amplitude reflectance of the M1 mirror. In the same way, we
can describe the amplitude of the light reflected at the M2 mirror as being
2ΩL3
ERM2 = rBS rM2 Eei(Ωt− c ), (3.26)

where rM2 is the amplitude reflectance of the M2 mirror. Then, the laser power
at a PD can be described as
 ( ) 2
Pr  2(Ω+ΩA )(L1 +L2 ) 2ΩL3 
= rM1 e ( ) 
i (Ω+2Ω )t−
A c
− rM2 e i Ωt− c

t2BS rBS
2
Pin
{ }
Ω(L1 + L2 − L3 ) + ΩA (L1 + L2 )
= rM1 + rM2 + 2 cos 2ΩA t −
2 2
.
c
(3.27)

Now, by demodulating this signal by cos(2ωA t), where ωA is an invariable

50
 3.2 Noise budget

M2

AOM
BS

PD M1
feed back

Figure 3.12: Conceptual image of the Doppler noise cancellation. Using this
scheme, we can transfer laser light with a very small phase modulation.

angular frequency as large as ΩA , we can obtain the error signal ϵDNC as

[ ]
ϵDNC 1 Ω(L1 + L2 − L3 ) + ΩA (L1 + L2 )
= cos 2 (ΩA − ωA ) t − .
t2BS rBS
2
Pin 2 c
(3.28)
Then, by feeding back the signal to the shift frequency of the AOM to keep
ϵDNC = 0, the equation

Ω(L1 + L2 − L3 ) + ΩA (L1 + L2 ) π
(ΩA − ωA ) t − = (2n + 1) (3.29)
c 4
is obtained. Therefore, Eq. (3.24) can be rewritten as
( )
−ΩL3 +ΩA L1
i (Ω+ωA )t− + π4 (2n+1)
EAM1 = tBS tM1 Ee c
. (3.30)

Then, the frequency fluctuation in the frequency domain can be written as

( )
˜ = iω ΩL̃3 − ΩA L̃1 − L1 Ω˜A .
δΩ (3.31)
c
Here, considering the situation L1 ≪ L3 and Ω ≫ ΩA , Eq. (3.31) can be
approximated as
˜ = iωΩ L̃3 .
δΩ (3.32)
c
Then, the frequency fluctuation of the Doppler noise depends only on the
length fluctuation of the short arm of the DAMI.
There are many kinds of noise sources of the Doppler noise, such as seismic

51
3 Experimental Design

vibration, temperature fluctuation, and wind. Therefore, the noise level of

the Doppler noise cannot be estimated easily in general. However, using the
DAMI, which has a reliable wind shield and sound isolation, 10−17 level, the
frequency transfer is reported using a 30 km long optical fiber [42]. The fibers
used in our experiment are much shorter than 30 km: about 5 m. The Doppler
noise therefore is not problematic for reaching the 10−17 level of stability if we
use the Doppler noise canceling system in our setup.

3.2.8 Design performance

The total noise budget of this experiment is shown in Fig. 3.13. The designed
noise budget of our experiment is limited by the coating thermal noise, accel-
eration noise, and RAM noise of around 1 Hz.
The laser frequency stability is often evaluated by the Allan deviation (see
Appendix B). This deviation and the power spectrum are connected by the
following equation:
√ ∫
∞
sin4 (πτ f )
σy (τ ) = 2 Gy (f ) df , (3.33)
0 (πτ f )2

where τ is an averaging time. Gy (f ) is a PSD of the physical quantity y. This

equation means that the Allan deviation at τ is dominantly determined by the
power spectrum near f = 1/2τ . Using this equation, the Allan deviation of our
experiment at around 1 second which is an important time scale for applying
a frequency-stabilized laser to an optical lattice clock can be estimated as in
Fig. 3.14. So, we can achieve a 10−17 level of stability by using this design.

52
 3.2 Noise budget

Figure 3.13: Total noise budget of our experiment. The noise sources limiting
the frequency stability in our experimental design are the coating thermal
noise, acceleration noise, and RAM noise.

Figure 3.14: Designed stability of our experiment (blue curve). The stability
of our design can reach the 10−17 level of stability at 1 second (below the black
rectangle).

53
3 Experimental Design

3.3 Summary of this chapter

• We use a monocrystalline silicon optical cavity, and cool it down at 18
K for reducing the thermal noise.

• In our experiment, both passive and active vibration isolation systems

are used. PVIS reduces the acceleration noise at high frequency and
AVIS decreases at low frequency.

• The thermal insulation stacks and radiation shields are very important
structures for cooling a reference cavity at 18 K.

• The Doppler noise cancellation is a strong method for reducing the phase
fluctuation, which the laser beam obtains during traveling the optical
path.

• The designed noise budget of our experiment is limited by the coating

thermal noise, acceleration noise, and RAM noise at around 1 Hz.

• We have designed a laser frequency stabilization system that can reach

the 10−17 level.

54
 Chapter 4
Performance Evaluation

In this chapter, we show the performance of our experimental setup.

4.1 Silicon optical cavity

In this section, we consider the optical property of the FPC in our experiment.
According to the inspection sheet of our mirrors, the power transmittance
is
T = t2F = t2E = 1.05 × 10−5 . (4.1)

Then, the amplitude transmittance of the front and end mirrors can be calcu-
lated as
√
tF = tE = T = 3.24 × 10−3 . (4.2)

The reflectance of a mirror is related to the finesse of a FPC. So, it can be

estimated by measuring the finesse of an optical cavity. Using Eq. (2.11) and
Eq. (2.21), the finesse of a cavity can be estimated as

F = 2πτFPC νFSR . (4.3)

Here, νFSR in our experiment is

c
νFSR = = (7.49 ± 0.01) × 108 Hz. (4.4)
2L
The error is the systematic error of the length of our cavity. Thus, the finesse
can be calculated by measuring the time constant, τFPC , of a FPC.
A schematic of the experimental setup for measuring the time constant
of our optical cavity is shown in Fig. 4.1. The time constant of the FPC is

55
4 Performance Evaluation

feedback
cryostat

AOM PD1
3165-1
(Gooch & Housego)

ECDL EOM BS PD2

Model: 4063
(New forcus) swiching
signal oscilloscope
AFG3252
∼ cavity
(Tektronix) LO electrical function TDS2024C
swich generator (Tektronix)
ZX80-DR230-S+ AFG3252
(Mini-Circuit)

Figure 4.1: Schematic of the setup for measuring the time constant of our
optical cavity. The PD2 used in this measurement is not the InGaAs photo
diode G12181-020A but the InGaAs photo diode G12180-003A because the
response of the PD should be fast enough to measure the decay time of the
transmitted light. We use the trans-impedance circuit for changing the photo-
current to the voltage. Details of the circuit is shown in Appendix C. Several
model numbers of the equipments used for this measurement are shown in the
figure.

determined by measuring the time dependence of the transmitted light power

when the input light is stopped. At first, we stabilize the laser frequency to the
resonant frequency of the FPC. Then, a certain amount of transmitted light
enters the PD2. When the RF power is switched off, the laser light entering
the FPC stops, and the power of the transmitted light gradually decreases.
The switching signal from the function generator and the output signal of the
PD2 are saved using the data-storing function of the digital oscilloscope.

The wavelength of the laser is about 1397 nm, and the temperature of the
cavity is about 10 K during the measurement. The transmitted power at the
time t = 0 is about 15 µW. The sampling frequency and the resolution of the
oscilloscope are 10 MHz and 2 mV, respectively. The switching speed of the
electrical switch in this experiment is below 1 µs. Therefore, the decay time of
the transmitted power is determined by the time constant of the cavity when
the time constant of our cavity is much longer than 1 µs.

Figure 4.2 shows an example of the time dependence of the transmitted

56
 4.1 Silicon optical cavity

Figure 4.2: Example of the time dependence of the transmitted light reduction.
The blue line is the measured transmitted power and the black line is a fitting
curve with Eq. (4.5) using the transmitted power at over t = 25 µs.

power. By fitting this reduction curve with the following equation

P = Ae−t/τFPC + C, (4.5)

the time constant of our FPC can be estimated. In our experiment, the time
constant is measured as

τFPC = (3.07 ± 0.01) × 10−5 s. (4.6)

The error is dominantly determined by the fitting error of the measurement.

Then, the finesse of our optical cavity is determined to be

F = (1.445 ± 0.006) × 105 . (4.7)

Moreover, by using Eq. (2.21), FWHM can be calculated as

νFWHM = (5.19 ± 0.02) × 103 Hz. (4.8)

57
4 Performance Evaluation

Assuming that the amplitude reflectance of the front mirror and the end mirror
are the same, the amplitude reflectance of the mirrors of our FPC can be
estimated as
rF = rE = 0.99998913 ± 0.00000004. (4.9)

The power transmittance, T , power reflectance, R, and power loss, L, of a

mirror are related by an equation, as follows:

T + R + L = 1, (4.10)

where R = rF2 = rE2 is the power reflectance of a mirror. Then, the amplitude
loss of a mirror can be estimated to be
√
lF = lE = L = (3.35 ± 0.02) × 10−3 . (4.11)

4.2 Vibration and its isolation systems

In this section, we discuss the seismic vibration in our laboratory and the
performances of the vibration isolation systems.

4.2.1 Seismic vibration level

The ground vibration is a main source of the acceleration noise, which de-
pends on the environment, such as the stiffness of the ground, structure of the
building, etc. We should thus check the spectrum of the seismic motion in the
experimental environment.
The seismic acceleration can be measured with a vibration sensor. The
setup for the acceleration spectrum measurement is shown in Fig. 4.3. We
define the X axis as the direction along the optical axis of our cavity. We
also define the Y and Z axes as the direction orthogonal to the X axis in the
horizontal plane and the vertical direction, respectively. The signal of the
vibration sensor is converted into the voltage spectrum by the FFT analyzer.
Since the typical transfer function of our vibration sensor (L-22E, 510 Ω, IRIS

58
 4.2 Vibration and its isolation systems

Z
OR34 (OROS) cryostat
FFT cavity X
analyzer Y
vibration
sensor

floor of the lab.

Figure 4.3: Setup for measuring the seismic acceleration spectra.The vibration
sensors signal is amplified by an electric amplifier. The details of the circuit
are shown in Appendix C. The model number of our FFT analyzer is shown
in the figure.

PASSCAL) is given by [43]

√
2πf 1
H(f ) ≃ 1+ 2
[(m/s2 )/V], (4.12)
47.1 (f /2.0 Hz)

we can calculate the acceleration spectrum from the signal of the vibration
sensor.

Figure 4.4 shows the seismic acceleration spectra of the translational axes in
our laboratory. The distance between the vibration sensor and the cryostat is
about 1 m, and the cryogenic refrigerator is moving during this measurement.
We use a hamming window function for measuring the voltage spectrum.

In this result, the seismic vibration level of our experimental environment

is much larger than that we expected. We should therefore further reduce
the vibration levels using passive and active vibration isolation systems, and
employ a low-vibration sensitivity support for a reference cavity.

59
4 Performance Evaluation

Figure 4.4: Vibration level at our experimental room. The blue, green, red
lines correspond to the acceleration along the X, Y, and Z axes. The black
line is the acceleration spectrum that we expect from Eq. (3.7). The spectra
below ∼0.1 Hz are circuit noise of the amplifier of the vibration sensor. The
peak at around 0.2 Hz is the micro seismic noise and that at around 1.4 Hz is
from a vibration-reduction structure of the building for earthquakes.

4.2.2 Passive vibration isolation system

The transfer function of the passive vibration isolation can be written as
Eq. (3.10). The performance of the system is therefore determined only by
a resonant frequency and Q factor.
The setup for checking the performance of the PVIS is shown in Fig. 4.5.
If there is no resonance except for the resonance of rubber legs, the signal ratio
between the sensors on the AVIT and on the ground is the same as the transfer
function of the PVIS. We can use the same method described in Section 4.2.1
for measuring the acceleration spectra at the top plate of the AVIT and the
floor. This measurement is also done under the condition that the cryogenic
refrigerator is moving.
Figure 4.6 is the result of this measurement. By reading the peak values
and the half widths of the peaks, we can estimate the horizontal resonant

60
 4.2 Vibration and its isolation systems

Z
cryostat
optical axis of the cavity
Y X
FFT
analyzer vibration
sensor cavity and TIS

signal
AVIT

vibration
sensor
rubber leg floor

Figure 4.5: Setup for measuring the resonant frequencies of the rubber legs.
The vibration sensors are put on the top plate of the AVIT and on the ground.
The resonant frequencies depend on the mass on the rubber legs. So, the
mass on the AVIT is the real cavity and the thermal insulation stacks (TISs)
during this measurement. The vibration spectra are measured using the FFT
analyzer.

frequencies and Q values of the PVIS as:

fPX = 8.5 ± 0.2 Hz, (4.13)

fPY = 8.8 ± 0.2 Hz, (4.14)

QX = 2.9 ± 0.1, (4.15)

QY = 3.4 ± 0.1. (4.16)

The error is dominantly determined by the frequency resolution of the mea-

surement.
However, the resonant frequency and Q factor along the Z axis cannot
be determined by this measurement because there are many resonances of
structures inside the cryostat at over 10 Hz. Thus, the performance along the
Z axis is estimated by measuring the transfer function from the voltage applied
to the AVIT to the vibration of the top plate. Figure 4.7 shows a gain curve of
the transfer function. By reading the peak value and half width of the peak,
we can get the resonant frequency and Q factor as

fPZ = 21 ± 2 Hz, (4.17)

61
4 Performance Evaluation

Figure 4.6: Ratio between the seismic motion and the vibration on the AVIT.
These ratios represent the vibration isolation ratios of the PVIS along each
axis.

QZ = 1.2 ± 0.1. (4.18)

The error of this result is also determined by the frequency resolution.

In this result, the passive isolation system of the translational axes are
almost the same as the design, but that of a vertical axis has a higher resonant
frequency, as expected, which means that the vibration isolation performance
along the Z axis is worse. The main reason of this result is that the rubber legs
are compressed by the mass on them, and thus the hardness of the vertical axis
becomes higher than the other translational axes. It will therefore be solved
by using a structure whose hardness does not depend on the mass on it, for
example a mechanical spring.

62
 4.2 Vibration and its isolation systems

Figure 4.7: Gain curve of the transfer function of the AVIT along the Z axis.
The gain curve below at 2 Hz has a slope because of the frequency response
of the vibration sensor on the AVIT. The dip and the peak of the gain curve
at around 60 Hz and 70 Hz are the anti-resonance and resonance of the AVIT,
respectively.

4.2.3 Active vibration isolation

The active vibration isolation using the AVIT is achieved by feeding back the
vibration signal at the top plate of the AVIT to the voltage applied to the PZT
actuators. Figure 4.8 shows a block diagram of the AVIS.

The vibration signals at the top plate of the AVIT are diagonalized to six
degrees of freedom: three translational axes and three rotational axes. In the
real feedback system of our experiment, we do not use the signals of rotational
axes because the signal of the rotational vibration is too small to be measured
by the vibration sensors.

The open-loop transfer functions of the AVIS in our experiment are shown
in Fig. 4.9 to Fig. 4.11. In these results, the behaviors of the AVIS are better
than the design performance.

63
4 Performance Evaluation

[V/m] [V/V]
three vertical signals and
Vibraon three holizontal signals Input
Sensors Matrix

diagonalized signals

[V/V] Filter

PZT actuators six signals to each PZT Output

Matrix
[m/V] [V/V]

Figure 4.8: Feedback diagram of the AVIS. The signals from vibration sensors
are diagonalized by the input matrix of a digital feedback system. These diag-
onalized signals are independently filtered and changed to the voltage applied
to each PZT by the output matrix.

Figure 4.9: Open-loop transfer function of the X axis. The red dots are the
measured values and the green line is a theoretical curve of the feedback system.
The black line is the designed open-loop transfer function of the AVIS.

64
 4.2 Vibration and its isolation systems

Figure 4.10: Open-loop transfer function of the Y axis. The red dots are
the measured values and the green line is a theoretical curve of the feedback
system. The black line is the designed open-loop transfer function of the AVIS.

Figure 4.11: Open-loop transfer function of the Z axis. The red dots are
the measured values and the green line is a theoretical curve of the feedback
system. The black line is the designed open-loop transfer function of the AVIS.

65
4 Performance Evaluation

Figure 4.12: Acceleration spectra on the AVIT after controlling it. The blue,
green, red lines show vibrations along X, Y, and Z axes, respectively. The solid
lines represent the inloop spectra after controlling the AVIT. The dashed lines
express free seismic acceleration spectra. The black line shows our designed
acceleration spectrum noise level.

Figure 4.12 shows the vibration levels after controlling the AVIT. Then,
the acceleration noise at a cavity point does not reach the designed noise level,
especially at high frequencies, because the seismic acceleration level is much
larger than the acceleration that we expect from Eq. (3.7). We should therefore
use a lower resonant-frequency structure for the passive vibration isolation.

4.2.4 Vibration sensitivity

The vibration sensitivity depends on many factors, such as the cavity and
support materials, the shape of a reference cavity, support points, etc. So by
choosing these elements adequately, the vibration sensitivities along the X, Y,
and Z axes can be small at the same time. In this subsection, we discuss the
vibration sensitivity of our reference cavity.
Our cavity is supported by four rectangular parallelepiped rubber legs
which are placed symmetrically around the center of the cavity. The size

66
 4.2 Vibration and its isolation systems

Figure 4.13: Calculation results of the FEM. The blue, green, and red lines
show the vibration sensitivity along the X, Y and Z axes, respectively.

of the rubber legs is 5 mm in height and 8 mm on each side.

The vibration sensitivity can be roughly calculated by the finite element
method (FEM) [20]. Since the shape of the cavity spacer has already been
designed, we calculate the vibration sensitivity for deciding on the best support
points of the cavity. Figure 4.13 shows the calculation result of the FEM as
a function of the distance between the support point and the plane of mirror
surface. In this result, the dependence of the vibration sensitivity on the
support points is very small, especially for the horizontal axes. We thus choose
support points near to the zero-cross point of the vibration sensitivity along
the Z axis: about 4 mm.
We then check that the vibration sensitivities are as small as we calculated.
A schematic diagram of the setup for measuring the vibration sensitivity is
shown in Fig. 4.14. The vibration sensitivity is measured by the frequency
fluctuation during swinging the cavity. The acceleration applied to the opti-
cal cavity is measured by the vibration sensor at the top plate of the AVIT.
The frequency fluctuation is measured by the beat signal fluctuation between
our laser and another stable laser. The frequency fluctuation signal and the

67
4 Performance Evaluation

feedback cryostat Z
swing using AVIT
PD1 Y
X
EOM AOM cavity

ECDL BS
PD2
AVIT
stable laser
signals from supply voltage
to PZT
53230A frequency vibration sensors
(Agilent) counter function
generator
data logger DL900
(GRAPHTEC) AFG3252
FFT

Figure 4.14: Schematic of the setup for measuring the vibration sensitivity. X
axis is a direction along the optical axis, and Y axis is an perpendicular to X
axis in the horizontal plane. Z axis is the vertical direction. The stable laser
is the same one in Fig. 3.8.

acceleration signal are stored by the frequency counter and the data logger,
respectively. We calculate the spectra of each signal and take their ratio of
them at the swinging frequency. The swinging frequency is 5 Hz, and the
√
acceleration applied to the cavity is about 10−3 (m/s2 )/ Hz at 5 Hz..
In the case that the support point is about 4 mm, the vibration sensitivities
of the X, Y, and Z axes are as follows:

AX = (2.5 ± 0.3) × 10−10 /(m/s2 ), (4.19)

AY = (4.6 ± 0.5) × 10−11 /(m/s2 ), (4.20)

AZ = (7 ± 1) × 10−12 /(m/s2 ). (4.21)

The errors are dominantly determined by the systematic errors of the vibration
sensors.
The vibration sensitivity along the Z axis achieves our requirement, below
−11
10 /(m/s2 ), but those of the X and Y axes do not satisfy this requirement.
The reason why the horizontal vibration sensitivities are not small may be
that there are some asymmetries of the configuration of a reference cavity,
such as manufacturing errors of the cavity spacer, the mounting accuracy of
the mirrors, etc. However, the vibration sensitivities of our reference cavity is

68
 4.3 Temperature fluctuation

insensitive to the support point of the optical cavity in the FEM calculation
result. Therefore, we should change the shape of our cavity poor to reduce the
vibration sensitivity.

4.3 Temperature fluctuation

The temperature fluctuation is small at high frequencies in general because
the time constant of heat transfer at room temperature is very long. However,
since the pulse tube refrigerator cools the 1st and 2nd stages by compressing
and expanding heat exchange gas, the temperature fluctuation spectrum of
the refrigerator has some peaks at the cycle frequency. Moreover, the time
constant of the heat transfer becomes short at low temperature because the
heat capacity of the materials becomes small. Furthermore, since an optical
cavity gets heat from an input laser, the laser power fluctuation causes a
temperature fluctuation.

4.3.1 2nd stage of the refrigerator

One of the temperature-fluctuation sources of this experiment is the 2nd stage
of the refrigerator. The temperature of the 2nd stage of our refrigerator be-
comes more stable at low temperature than at high temperature because the
liquid helium plays the role of a heat bath. We therefore check the temperature
fluctuation of the 2nd stage when it operates at below 4 K, which is a temper-
ature well below the value to change helium gas to liquid helium. Figure 4.15
shows the temperature-fluctuation spectrum of the 2nd cold head.
The floor level of this result is dominantly determined by the circuit noise
for measuring the temperature. The actual temperature fluctuation of the 2nd
stage is therefore below this floor line. The peak structure does not exist in the
circuit noise spectrum. So, the sharp peak values of this measurement reflect
the actual temperature fluctuation of the 2nd stage caused by the action of
the compressor; it is much higher than that which we assumed in Eq. (3.20).
However, the thermal-isolation design of our experiment has a large margin,
for example 105 at 1 Hz. So, these peaks are not problematic for reaching the
10−17 level stability.

69
4 Performance Evaluation

Figure 4.15: Temperature-fluctuation spectrum of the 2nd stage of the re-

frigerator. The blue lines are the measured values and the black line is a
temperature-fluctuation spectrum that we assumed in our design. The tem-
perature during the measurement is about 3.6 K.

4.3.2 Outer shield

Other temperature fluctuation sources of this experiment are the temperature

fluctuations of the 1st stage of the refrigerator and the room temperature.
However, these temperature fluctuations do not transfer to a reference cavity
directly, because there is an outer shield next to the middle shield. Then,
it is sufficient to check the temperature fluctuation of the outer shield. The
temperature-fluctuation spectrum of the outer shield is shown in Fig. 4.16.

The floor level of this measurement is limited by the circuit noise for mea-
suring the temperature. This result is also larger than the temperature spec-
trum that we assumed in the experimental design. However, since the margin
of the temperature-fluctuation noise is much larger than the difference between
the measured temperature fluctuation and that which we expected, it is not a
problem.

70
 4.3 Temperature fluctuation

Figure 4.16: Temperature-fluctuation spectrum of the outer radiation shield.

The blue lines are the measured values, and the black line is the temperature-
fluctuation spectrum that we assumed in our design. The temperature during
the measurement is about 100 K.

4.3.3 Laser power effect

The input laser power fluctuation causes a temperature fluctuation locally,

which leads to a cavity length fluctuation through thermal expansion. In this
subsection, we discuss this effect.
The normalized frequency fluctuation of the laser power change can be
described as
δν √ √
= β GP (f ) [/ Hz], (4.22)
ν
where β is a coupling index and GP (f ) is the PSD of the transmitted laser
power fluctuation.
The coupling index, β, is determined by measuring the frequency fluctua-
tion when the transmitted power is modulated. Figure 4.17 shows the setup
for measuring the coupling index, β. We stabilize the laser to the cavity and
supply the amplitude modulated RF signal to the AOM. Then, the transmitted
light power and the laser frequency are modulated. The power of the trans-

71
4 Performance Evaluation

amplitude feedback
modulation
cryostat

PD1
PD2

ECDL EOM AOM BS

PD3
cavity
stable laser
frequency FFT data
counter logger

Figure 4.17: Setup for measuring the laser power-fluctuation effect. The PD1
is the InGaAs photo diode G12180-003A, and the PD2 is the InGaAs photo
diode G12180-250A. The stable laser is the same one as in Fig. 3.8

mitted light is measured using PD2 and the data logger. The laser frequency
is determined by measuring the beat frequency signal between our laser and
the stable laser in Fig. 4.17 using the frequency counter. The data of the
transmitted power and the laser frequency is converted to the spectra, and we
take the ratio between them to determine the coupling index, β.
The wavelength of the ECDL is about 1397 nm, and the temperature of our
cavity is about 10 K during the measurement. The amplitude of the RF signal
is about 36 dBm, and the modulation depth is 10% of it. The modulation
frequency is varied between 0.1 Hz to 10 Hz.
The result of this measurement is shown in Fig. 4.18. In this result, we can
regard β as being a constant. The value is estimated as

β = (1.6 ± 0.2) × 10−7 /W. (4.23)

The error is the standard deviation of this measurement.

72
 4.4 Feedback system

Figure 4.18: Coupling index, β, of our reference cavity. The blue dots are the
measured values, which are connected by the blue line.

4.4 Feedback system

We describe the frequency-stabilization feedback system in this section.

The laser source of our experiment is an ECDL. The frequency of an ECDL
can be stabilized by controlling the angle of a diffraction grating, and the
amperage of a laser diode (LD) current. Thus, the frequency fluctuation at
low frequencies is fed back to an angle of the grating, and high-frequency
fluctuation is used to control the laser diode current. Figure 4.19 shows a block
diagram of our feedback system. A detailed circuit is shown in Appendix C.
The open-loop transfer function of this feedback system can be described
as
G = −GPDH GPA G1 (FPZT APZT + FC AC ) . (4.24)

Since the transfer functions (G1 , G2 , GPA , FPZT , and FC ) can be calculated by
the components of the feedback circuit, it is sufficient to measure only GPDH ,
APZT , and AC .

73
4 Performance Evaluation

PDH method

[V/Hz] preamp. [V/V]

frequency
fluctuation
port 2 port 1 monitor1

amp. 2 [V/V]
amp. 1 [V/V]

[Hz/V] [V/V]
low freq.
PZT
filter monitor2

[V/V]
[Hz/V]
high freq.
LD
filter

Figure 4.19: Block diagram of our feedback loop for the frequency stabilization.
GPDH is the gain of the PDH method; GPA is the transfer function of the
preamplifier. FPZT and FC are the transfer functions of the low frequency and
high frequency filters, respectively. APZT and AC are actuator gains of PZT
and LD, respectively. G1 and G2 are the gains of amplifier1 and amplifier2,
respectively.

The actuator gain of the PZT has some resonance, so it is difficult to mea-
sure APZT over the resonant frequency. However, below the resonant frequency,
APZT is a flat response function, which is easy to measure. By applying the
voltage from port2 and measuring the frequency shift of the ECDL, we can
obtain the actuator gain APZT as

APZT = (3.0 ± 0.1) × 109 Hz/V (below 100 Hz). (4.25)

The PDH gain, GPDH , and the actuator gain, AC , can be evaluated by
measuring the transfer function between some signal ports during frequency
stabilization. The transfer function is measured by the frequency response
analyzer FRA5087 (NF’s product).
When we supply a modulated voltage to port2, the transfer function from
the input voltage at port2 to the output voltage from monitor1 can be written

74
 4.4 Feedback system

G1
Figure 4.20: Gain and phase of Hp2 . The red dots are the measured values
and the black curve is a theoretical line, assuming AC = 109 Hz/V.

as ( )
FPZT FC A C
G1
Hp2 ≃ 1+ . (4.26)
G2 FPZT APZT
Here, we assume the open-loop gain to be G ≫ 1. Figure 4.20 shows the
measured transfer function and a theoretical curve calculated by Eq. (4.26).
In this result, we can estimate the transfer function from the voltage to the
frequency through the LD current as

AC = (1.0 ± 0.1) × 109 Hz/V. (4.27)

When we supply a modulated voltage from port1, the transfer function

from the output voltage from the monitor1 to that from the monitor2 can be
written as
G
HGG1PA = = GPDH GPA (FPZT APZT + FC AC ) . (4.28)
G1
Figure 4.21 shows the measured transfer function and theoretical curve cal-
culated by Eq. (4.28). In this result, we can estimate the gain of the PDH

75
4 Performance Evaluation

Figure 4.21: Gain and phase of HGG1PA . The red dots are the measured values
and the black curve is a theoretical line assuming AC = 109 Hz/V and GPDH =
4.4 × 10−6 V/Hz.

method as
GPDH = (4.4 ± 0.4) × 10−6 V/Hz. (4.29)

4.5 Input and output optics

The laser power stabilization and Doppler noise cancellation are performed in
the input laser system. In this section, we check their performance.

4.5.1 Laser power stabilization

The laser-power fluctuation noise is caused by the fluctuation of an intra-cavity
power, and the transmitted light power is proportional to the intra-cavity laser
power. Therefore, we stabilize the laser power in order to make transmitted
light stable for reducing the laser-power fluctuation noise.
The setup for the laser power stabilization is shown in Fig. 4.22. Laser

76
 4.5 Input and output optics

feedback
cryostat

PD1
EOM AOM PD2

ECDL BS

cavity
feedback

FFT
analyzer

Figure 4.22: Setup for the laser power stabilization. The laser power is sta-
bilizaed so as to make the transmitted laser power 90 µW. The laser power
stability is estimated by using the signal of the PD2.

stabilization is achieved by feeding back the signal of the PD2 to AOM using
the feedback circuit (see detail in Appendix C). The output from the monitor
port of the feedback circuit is converted to the spectrum by the FFT analyzer,
and used for evaluating the laser power fluctuation.
The transmitted laser power is about 90 µW. The amplitude and the
frequency supplied to the AOM is about 160 MHz and 36 dBm, respectively.
The PD2 is the InGaAs PD G12181-020K and the trans-impedance gain of
the circuit is 104 . The window function for calculating the spectrum is the
hamming window function.
Figure 4.23 shows the power fluctuation spectra of the free-running and
stabilized laser.

77
4 Performance Evaluation

Figure 4.23: Spectra of the transmitted laser power fluctuation. The blue line
is the inloop spectrum with the laser power fluctuation being stabilized. The
green line is the free-run spectrum of the laser power.

4.5.2 Doppler noise cancellation

When Doppler noise cancellation is applied, the effect from the fluctuation of
the traveling path is negligible, and the effect from the length fluctuation of
a short arm of a DAMI will be significant. Moreover, there are some optical
paths to which the Doppler noise cancellation technique cannot be applied.
We therefore evaluate the phase modulation of these optical paths in this
subsection.
All the optical paths without Doppler noise cancellation, which can affect
the stability of our laser, is inside the windshield. Then, we make a Mach-
Zehnder interferometer and measure the differential phase fluctuation of the
two arms.
Figure 4.24 shows the setup for measuring the phase fluctuation inside the
windshield. The phase difference between two arms of the interferometer is
measured by the PD inside the figure, and this PD’s signal is converted to the
voltage spectrum by the FFT analyzer. The window function for calculating

78
 4.5 Input and output optics

windshield

laser BS

FFT
PD analyzer

BS

Figure 4.24: Setup for measuring the phase-fluctuation level in traveling the
optical paths inside the windshield. The PD in this figure is the InGaAs
photo detector G12180-003A with the trans-impedance circuit (details in Ap-
pendix C). The stable laser is the same one in Fig. 3.8.

the voltage spectrum is the hamming window function. Each arm length of
the interferometer is about 50 cm in this experiment, which is as long as the
optical path without the Doppler noise cancellation system in our input and
output optics. The windshield is a plastic box which is the same as that used
in our laser system.

In this setup, the PD voltage difference between the bright and the dark
state is measured as
∆V = 3.84 V. (4.30)

The phase-fluctuation spectrum can be calculated as

√
2π √ √ √
Gϕ (f ) = GV (f ) = 1.94 GV (f ) [rad/ Hz], (4.31)
∆V
where GV (f ) is a voltage spectrum of a PD. Then, we can estimate the phase
modulation inside the windshield as in Fig. 4.25.

79
4 Performance Evaluation

Figure 4.25: Phase-fluctuation spectrum of a Mach-Zehnder interferometer

inside the windshield.

80
 4.6 Summary of this chapter

4.6 Summary of this chapter

• We checked the performance of our experimental setup and measured
the behaviors of environmental noise sources.

• The results of the measurements and estimations are summarized in Ta-

ble 4.1

amplitude transmittance tF = tE = 3.24 × 10−3

FSR νFSR = 7.49(1) × 108 Hz
amplitude reflectance rF = rE = 0.99998913(4)
amplitude loss lF = lE = 3.35(2) × 10−3
cavity finesse F = 1.445(6) × 105
FWHM νFWHM = 5.19(2) × 103 Hz
FPC time constant τFPC = 3.07(1) × 10−5 s
PVIS cutoff frequency fPX = 8.5(2) Hz
fPY = 8.8(2) Hz
fPZ = 21(2) Hz
PVIS quality factor QPX = 2.9(1)
QPY = 3.4(1)
QPZ = 1.2(1)
vibration sensitivity AX = 2.5(3) × 10−10 /(m/s2 )
AY = 4.6(5) × 10−11 /(m/s2 )
AZ = 7(1) × 10−12 /(m/s2 )
PDH gain GPDH = 4.4(4) × 10−6 V/Hz
laser power coupling index β = 1.6(2) × 10−7
seismic acceleration spectrum shown in Fig. 4.4
AVIT open loop transfer function shown in Fig. 4.9 to Fig. 4.11
2nd head temperature fluctuation shown in Fig. 4.15
outer shield temperature fluctuation shown in Fig. 4.16
laser power fluctuation shown in Fig. 4.23
Doppler phase fluctuation shown in Fig. 4.25

Table 4.1: Table of various parameters of our setup

81
 Chapter 5
Stability Estimation

In this chapter we show the final stability of our frequency stabilization.

5.1 Noise estimation

In the previous chapter, we discussed the performance and environmental noise
sources of our experiment. In this section, we estimate the noise levels of several
noise sources using the results of the previous chapter.

5.1.1 Thermal noise

According to a theoretical calculation, we can estimate the noise level of the
thermal noise as
δν 2.0 × 10−17 √
= √ / Hz. (5.1)
ν f

5.1.2 Residual gas noise

The residual gas noise is determined by the vacuum level of the cryostat. The
vacuum level of our experiment is measured to be

p = 2.5 × 10−5 Pa. (5.2)

Using this pressure and the result of Eq. (3.23), we can estimate the residual
gas noise as
δν √
= 5.0 × 10−20 / Hz. (5.3)
ν

83
5 Stability Estimation

5.1.3 Acceleration noise

The acceleration applied to a reference cavity is shown in Fig. 4.12, and
the vibration sensitivity of our reference cavity is measured as Eq. (4.19) to
Eq. (4.21). Then, using Eq. (2.73), we can estimate the noise level of the
acceleration noise, as shown in Fig. 5.1.
In this result, the noise level is dominantly determined by the acceleration
along the X axis, which is parallel to the optical axis of the reference cavity.
This noise level is much larger than that we planned for in our designed.
The main reason for this result is that the vibration sensitivities are much
higher than we had assumed, especially the vibration sensitivity along the X
axis. If the vibration sensitivities satisfy the requirement, 10−11 /(m/s2 ) in
all translational axes, the acceleration noise could be about 20-times smaller
than that expressed in Fig. 5.1. Since the vibration sensitivity of our cavity is
insensitive to the support point of the cavity, we should modify the shape of
our FPC so as to reduce the vibration sensitivity.
Another reason is that the seismic acceleration level in our laboratory is
larger than that which we had anticipated from Eq. (3.7), especially at over
10 Hz. Since the AVIS cannot produce good vibration isolation at high fre-
quencies, we should use another passive vibration isolation structure whose
resonant frequency is much smaller than that of our PVIS.

84
 5.1 Noise estimation

Figure 5.1: Frequency-fluctuation spectrum of the acceleration noise. The

black line is the designed noise spectrum of our experiment.

5.1.4 Temperature-fluctuation noise

The temperature fluctuations of the 2nd stage and the outer shield are mea-
sured as shown in Fig. 4.15 and Fig. 4.16. The CTE of silicon is below 10−9 /K
when the temperature of an optical cavity is below 18 K [23]. Both the heat
capacity and the thermal conductivity have the same temperature dependence,
proportional to T 3 , at very low temperature. Then, the transfer function of the
heat transfer hardly changes the operation temperature of a reference cavity.
Thus, using the calculation result of the transfer function, we can estimate the
temperature-fluctuation noise, as shown in Fig. 5.2.
In this result, the frequency fluctuation of the temperature-fluctuation noise
is limited by the temperature fluctuation from an outer shield. Since this
noise level is much smaller than the noise spectrum that we designed, the
temperature-fluctuation noise does not disturb us from reaching the 10−17
level stability.

85
5 Stability Estimation

Figure 5.2: Frequency-fluctuation spectrum of the temperature-fluctuation

noise. The black line is the designed noise spectrum of our experiment.

86
 5.1 Noise estimation

5.1.5 Laser power fluctuation noise

The laser-power fluctuation noise can be estimated using Eq. (4.22). The cou-
pling index, β., is determined as in Eq. (4.23), and the laser-power fluctuation
is measured as shown in Fig. 4.23. The normalized frequency spectrum of the
laser-power fluctuation noise can thus be estimated as shown in Fig. 5.3.
In this result, the frequency fluctuation of this noise is small enough to
achieve the 10−17 level stability.

Figure 5.3: Frequency-fluctuation spectrum from the laser power-fluctuation

noise. The black line is the design noise spectrum of our experiment.

5.1.6 Circuit noise

The circuit noise can be estimated by using a feedback circuit. Using the
spectrum of the output voltage of amplifier1 in the feedback circuit (Fig. 4.19),
we can estimate the frequency-fluctuation spectrum as
√
δν GAG1 (f )
= , (5.4)
ν G1 GPA GPDH

where GAG1 (f ) is the PSD of the output voltage of amplifier1.

87
5 Stability Estimation

Figure 5.4: Frequency-fluctuation spectrum from the circuit noise. The black
line is the design noise spectrum of our experiment.

When we turn off the laser and measure GAG1 (f ), only the circuit noise
appears. Figure 5.4 shows the circuit noise level estimated by Eq. (5.4).
In this result, since the frequency fluctuation of the circuit noise is suffi-
ciently small, the circuit noise does not disturb us from reaching the 10−17
level stability.

5.1.7 Shot noise

When we use a laser without any phase modulation, the shot noise and the
circuit noise appear in the voltage spectrum, GAG1 (f ). Since the shot-noise
level depends on the laser power entering a PD, the laser power is adjusted
to be the same as the reflected light power at the time when the laser light is
resonate with a FPC.
The result of this measurement is the same as that of the circuit-noise
estimation. Then, the shot noise level of our experiment can be estimated as

δν √
< 1.7 × 10−18 / Hz. (5.5)
ν

This result is also small enough to reach the 10−17 level stability.

88
 5.1 Noise estimation

5.1.8 RAM noise

When we use the AR reflection signal of a PD4 instead of that of a PD1
(Fig. 3.8), the RAM noise, the shot noise, and the circuit noise appear in the
voltage spectrum, GAG1 (f ) [44]. The laser power is adjusted to be the same
value as the reflected light power resonating with a FPC, because the RAM
noise depends on the laser power entering a PD.
Using the spectrum GAG1 (f ) and Eq. (5.4), we can estimate the RAM noise,
as shown in Fig. 5.5. This result is dominantly determined by the RAM noise,
because the noise level of the shot noise and the circuit noise is smaller than
that in Fig. 5.5. The frequency fluctuation of the RAM noise is the same as
that which we designed. Thus, the RAM noise also does not disturb us from
achieving the 10−17 level stability.

Figure 5.5: Frequency-fluctuation spectrum of the RAM noise. The black line
is the design noise spectrum of our experiment.

89
5 Stability Estimation

5.1.9 Doppler noise

We use a Doppler noise cancellation system in our experiment. Thus, all of
the traveling paths that lead to a laser phase fluctuation is in the windshield.
We can therefore estimate the Doppler noise level by measuring the phase
fluctuation traveling inside the windshield. Figure 4.24 shows the experimental
setup for estimating the noise level of the Doppler noise.
The frequency-fluctuation spectrum caused by the phase fluctuation can
be expressed as √
√
Gν (f ) = f Gϕ (f ). (5.6)

Then, the spectrum of the normalized frequency fluctuation can be written as

√
δν f Gϕ (f )
= . (5.7)
ν ν
Using Eq. (5.7) and the phase-fluctuation spectrum, as shown in Fig. 4.25,
we can estimate the Doppler noise while laser light travels in the windshield,
as shown in Fig. 5.6. In this result, the noise level of the Doppler noise is much
lower than the designed noise budget. Therefore, the Doppler noise is not a
problem for reaching 10−17 level stability.

5.1.10 Total noise

Assuming that all of the noises we discussed in this section are uncorrelated, we
can estimate the total noise spectrum by calculating the square root of the sum
of squares of all noises. The total noise is shown in Fig. 5.8. Using this result
and Eq. (3.33), we can estimate the frequency stability of our experiment, as
shown in Fig. 5.8.
The frequency-fluctuation spectrum is dominantly determined by the accel-
eration noise. We can therefore improve the stability of our laser by reducing
the acceleration noise.
The estimated frequency stability of our laser is σy ≃ 2 × 10−15 at 1 second.
This stability is 20-times more unstable than our goal: the 10−17 level.

90
 5.1 Noise estimation

Figure 5.6: Frequency-fluctuation spectrum of the Doppler noise. The black

line is the design noise spectrum of our experiment.

Figure 5.7: Frequency fluctuation spectrum of the total noise of our experi-
ment. The black line is the designed noise spectrum of our experiment.

91
5 Stability Estimation

Figure 5.8: Result of our estimation for the frequency stability of our experi-
ment. Our goal is below the black rectangle.

5.2 Frequency instability measurement

We discussed the total noise of our experiment in Section 5.1. The frequency
stabilization of the laser is limited by the acceleration noise. In this section, we
consider how to measure the frequency instability directly, and check that the
acceleration noise actually limits the stability of our laser without any doubt.

5.2.1 Experimental setup

We show a schematic diagram of the experimental setup for the frequency-
instability measurement in Fig. 5.9. The frequency instability is determined
by measuring the beat frequency between two lasers. Since the beat-frequency
fluctuation is dominantly determined by a laser whose instability is the larger
of the two, we should prepare a more stable laser than ours in order to measure
the instability of our laser. The stable laser in Fig. 5.9 has a 10−16 level stability
from the averaging time, τ = 0.1 s, to τ = 10 s. Since the estimated instability
of our laser is at the 10−15 level, this laser is enough stable to measure the
instability of our laser. The beat frequency is measured using the PD5 and

92
 5.2 Frequency instability measurement

windshield
feedback for frequency stabilization

feedback for Doppler

noise cancellation cryostat
PD2 PD1 cavity
PBS PD4
EOM AOM1
ECDL HWP

AOM2
PD3
feedback for laser power stabilization
feedback for Doppler
noise cancellation
stable laser BS PD5
frequency
counter

windshield

Figure 5.9: Setup for measuring the instability of our laser. PD1 is used
for obtaining the PDH signal for frequency stabilization. PD1 , PD2, PD3,
and PD5 are InGaAs photo detectors G12180-003As, and PD4 is an InGaAs
photo detector, G12181-020K. Each PD has a trans-impedance circuit, and
each feedback system is achieved by an analog electric circuit. Details of these
circuits are given in Appendix C. The stable laser is the same one as in Fig. 3.8.
The frequency counter is an Agilent 53230A.

the frequency counter.

During this instability measurement, the Doppler noise cancellations and
the laser-power stabilization are operated. Doppler noise cancellations are
achieved by feeding back the signals of the PD2 and the PD3 to the frequency
of the RF signal to the AOM1 and the AOM2, respectively. Laser power
stabilization is achieved by feeding back the signal of the PD4 to the amplitude
of the RF signal to the AOM1.
Moreover, we use the simplified thermal isolation setup during the mea-
surement for reducing the initial cooling time. Figure 5.10 shows a schematic
of the simplified setup inside the cryostat. The main change is done in three
parts. The first is to remove the stainless thermal resistance between the 2nd
base of the refrigerator and the middle shield. The second is to take off the
inner radiation shield. The third is to reduce the number of floors of the
thermal-isolation stacks. Owing to these changes, the cavity is thermally well
connected to the 2nd base of the refrigerator. It takes therefore about two
days to cool a reference cavity, while it takes over two months for the designed
setup to cool.

93
5 Stability Estimation

refrigerator

super
insulation outer shield

temperature middle shield

sensor

heater BK7
window

AVIT top plate

Figure 5.10: Simplified setup for reducing the cooling time. The temperature
sensors are mounted at the forth floor of the thermal-isolation stacks, the
middle radiation shield, and the outer radiation shield (the green parts in the
figure). There are also heaters at the middle radiation shield and the outer
radiation shield for changing the temperatures (the pink part in the figure).

5.2.2 Instability measurement

We measure the beat frequency between the two laser frequencies. The gate
time of the frequency counter is 10 ms. The absolute value of the beat fre-
quency is about 220 MHz. The transmitted light power is about 90 µW. The
modulation frequencies of the EOM and the AOM are about 15 MHz and 160
MHz, respectively. The modulation voltages of the EOM and the AOM are
about 12 dBm and 36 dBm, respectively. The temperature around the cavity
is about 10 K during the measurement, and the wavelength of our laser is
about 1397 nm.
Figure 5.11 shows the beat frequency between our laser and the stable
laser without operating the AVIT. From this result, we can directly calculate
the Allan deviation of the frequency fluctuation, as shown in Fig. 5.12. The
stability seems to be dominantly limited by the stability of our silicon optical
cavity, because the stability of this result is much higher than that of the stable
laser used for the beat measurement. The stability at most stable time scales
is measured at an Allan deviation of

−15
−0.34 ) × 10
σy = (7.03+0.38 (τ = 0.8 s). (5.8)

94
 5.2 Frequency instability measurement

Figure 5.11: Beat-frequency fluctuation between our laser and another narrow-
linewidth laser. The gate time of the frequency counter is τ = 10 ms.

The error of this result is the 95% confidence interval of the χ2 distribution
of the measurement. In at previous work with the cryogenic optical cavity at
very low temperature, the stability of the laser was σy ≃ 2 × 10−14 at around
1000 seconds and σy ≃ 2 × 10−13 at around 1 second [23]. Therefore, our laser
is the most stable one with a very low temperature optical cavity at around
1 second, which is an important time scale for applying a laser to an optical
lattice clock. Note that the stability at around 1 second in the previous work
is read from a graph of the Allan deviation in Ref. [23].
Figure 5.13 shows the spectra of the beat frequency and estimated noise
spectrum using the vibration sensitivity and acceleration spectrum without
operating the AVIT. Since the spectrum estimated by the acceleration noise
coincides with that of the beat measurement at over 1 Hz, the stability of our
laser is definitely limited by the acceleration noise.
On the other hand, the spectrum below 1 Hz cannot be explained by the
acceleration noise. This noise level can be considered to be the temperature-
fluctuation noise from the 2nd stage of the refrigerator. Since we use the
simplified thermal isolation system during this stability measurement, the

95
5 Stability Estimation

Figure 5.12: Allan deviation of the beat frequency. The error bars represent
the 95% confidence intervals of the measurement. Our goal stability is below
the black rectangle.

Figure 5.13: Frequency-fluctuation spectrum. The blue line is the measured

value and the green line shows the estimated frequency-fluctuation spectrum
with the vibration sensitivity and the acceleration spectrum.

96
 5.2 Frequency instability measurement

temperature-fluctuation noise is much larger than that which we estimated

in Fig. 5.2. In simplified setup, the 2nd base of the refrigerator is directly
connected to the middle radiation shield, and the silicon optical cavity is put
directly on the fourth floor of the thermal-isolation stacks. Therefore, the tem-
perature fluctuation from the 2nd stage of the refrigerator travels well to the
optical cavity. Thus, the transfer function of the temperature fluctuation from
the 2nd stage to the cavity is dominantly determined by the connection part
of the middle shield and the stack: thin OFC wires and aluminum foils.

We can directly measure the transfer function of the temperature fluctu-

ation using the temperature sensors and heaters in the cryostat. Figure 5.14
shows a schematic of the setup for measuring the transfer function from the
middle shield to the thermal-isolation stack. Temperature sensors (DT-670,
Lake Shore Cryogenics) are mounted at the middle radiation shield and the
forth floor of the thermal-isolation stacks. A heater is also mounted at the
middle radiation shield. When we apply a periodic voltage to the heater, the
temperature sensor outputs the periodic temperature signals. We can there-
fore measure the transfer function by calculating the ratio of the amplitude
of the sensors. The heater is about 50 Ω, and the amplitude of the supplied
voltage to the heater is between 1 V and 10 V.

Figure 5.15 shows the measured gain of the transfer function from the
middle shield to the thermal-isolation stack. In theory, the transfer function of
the temperature fluctuation can be written as in Eq. (3.13). Thus, by fitting
these measured values with the function
1
|H| = √ , (5.9)
1 + (ωRC)2

we obtain
(2.2 ± 0.2) × 10−2
RC = Hz. (5.10)
2π
This error is the fitting error of the measurement. Note that we only use data
below 0.1 Hz for the fitting, because any measurement over 1 Hz is limited
by the noise of the temperature sensor. Using this transfer function and the
temperature spectrum in Fig. 4.15, we can roughly estimate the temperature
fluctuation at the optical cavity. Now, assuming the CTE of silicon to be

97
5 Stability Estimation

refrigerator

super
outer shield
insulation

function
generator
temperature heater
sensor

BK7
middle
shield window
data
logger
DL750
(YOKOGAWA)

temperature
sensor

AVIT top plate

Figure 5.14: Schematic of the setup for measuring the transfer function of the
temperature fluctuation.

α = 2.5 × 10−10 /K so that the hight of the characteristic peak around 1.5 Hz
is the same as the peak around 1.5 Hz in Fig. 5.13, the temperature-fluctuation
noise is calculated as shown in Fig. 5.16.
In this result, the frequency of the peak of the temperature-fluctuation noise
exactly coincides with the frequency of the peak in Fig. 5.13. Moreover, the
CTE of silicon at 10 K is at the 10−10 /K level [23]. Thus, the assumed value
of the CTE is not very different from the measured value. Note that the CTE
of the silicon at 10 K is read from the graph of the CTE in Ref. [23]. Since the
floor level of the temperature fluctuation spectrum in Fig. 4.15 is limited by
the circuit noise, the actual temperature-fluctuation noise below 1 Hz should
be smaller than that in Fig. 5.16. Therefore, this result is consistent with the
measurement. Thus, the noise source limiting the stability of our laser below
1 Hz is likely to be the temperature-fluctuation noise. If we use the designed
temperature isolation setup, the temperature-fluctuation noise can be reduced.
Therefore, the full setup operation is a subject of future work.

98
 5.2 Frequency instability measurement

Figure 5.15: Gain of the transfer function of the temperature fluctuation from
the middle radiation shield to the forth floor of the thermal-isolation stacks.
The red dots are the measured values and the green curve is a fitting curve.

-13
10

-14
10

-15
10
0.5 1 3

Figure 5.16: Frequency fluctuation of our laser and the estimated temperature-
fluctuation noise. The blue line is the frequency-fluctuation spectrum of the
beat measurement. The green line is the frequency-fluctuation spectrum of
the estimated temperature-fluctuation noise. The floor level of the green line
shows the upper limit of the frequency fluctuation due to the temperature
fluctuation. The inset is the enlarged view of the graph from 0.5 Hz to 3 Hz.

99
5 Stability Estimation

vertical sensor
horizontal sensor
g
spring
g
spring magnet
coil
x
magnet
to amp.
x

to amp.
coil
to amp. to amp.

Figure 5.17: Schematic diagram of our vibration sensor’s structure. Left is a

vertical sensor and right is a horizontal sensor. The components are a magnet,
a spring, and a coil. When the magnet moves, the coil gets induced voltage
proportional to the velocity difference between the magnet and the coil.

Next, we tried to operate the AVIT for reducing the acceleration noise.
However, when we operated the AVIT, the laser stability became worse than
the result in Eq. (5.8). The reason for this result may be considered to be the
tilt vibration of a reference cavity.

Figure 5.17 shows the structure of the vibration sensor in our experiment.
When the vibration sensor moves, it outputs a signal proportional to the vi-
bration sensor’s velocity. Then, the output of the vibration sensor for the
translational movement can be written as

VOut = HVS ẋ, (5.11)

where HVS is a response function of the vibration sensors (L-22E, 510 Ω),
which can be roughly approximated in the frequency domain as [43].
√
(f /2.0 Hz)2
HVS (f ) ≃ 47.1 [V/(m/s)]. (5.12)
1 + (f /2.0 Hz)2

On the other hand, if the vibration sensor is tilted, the velocity of the

100
 5.2 Frequency instability measurement

vibration sensor can be described as

∫ t
V
ẋ = g (1 − cos θ) dt ≃ 0 (vertical), (5.13)
0

∫ t ∫ t
H
ẋ = g sin θ dt ≃ g θ dt (horizontal). (5.14)
0 0

Therefore, the output signals are different between a vertical axis sensor and
a horizontal axis sensor, and these can be described as

V
Vout = 0, (5.15)
∫ t
H
Vout = HVS g θ dt. (5.16)
0

Therefore, only the horizontal vibration sensor responds to its tilt.

Now, considering the output signal ratio between the translational dis-
placement and the tilt movement in the frequency domain, it can be expresses
as
VOut x̃(ω)
RTR = H = −ω 2 . (5.17)
VOut g θ̃(ω)
Then, the signal of a horizontal vibration sensor is dominantly determined
by the tilt movement at low frequency. Since the rotational vibration of the
ground is very in general small, this effect does not appear at around 1 Hz.
However, in the AVIS, we use a feedback system for operating the AVIT.
Then, when the input matrix and output matrix, which play a role of diagonal-
izing six degrees of freedom, are not correct, the tilt-movement effect appears
in the output signal of the vibration sensor. Figure 5.18 shows the transfer
function of the AVIT. In this result, the X axis and the Y axis curves are
larger than the theoretical line, while Z-axis transfer function is comparable
to theory. This is evidence of coupling between the translational displacement
and the rotational movement below 1 Hz.
The rotational vibration of a reference cavity can be a large acceleration
through a gravitational acceleration of Earth. Therefore, the reason why the
AVIS does not work well is considered to be in the rotational vibration of a
reference cavity.

101
5 Stability Estimation

Figure 5.18: Transfer functions of the AVIT. The blue, green, and red dots
show the measured values of the X, Y, and Z axes. The black solid line is a
theoretical line without a tilt. The blue curve shows a theoretical curve with
a 10% tilt coupling. The green line shows a theoretical curve with 25 % tilt
coupling. All of the curves and dots are normalized so that the gain at 1 Hz
is equal to 0 dB.

5.3 Estimation without the acceleration noise

In the last section, we described that the noise sources limiting our laser-
frequency stability are the acceleration noise and the temperature-fluctuation
noise. We therefore discuss the effect of other noises in this section.
The estimated frequency-fluctuation spectrum of our experimental noises
without the acceleration noise is shown in Fig. 5.19. In this result, the noise
level is dominantly limited by the RAM noise and the thermal noise.
By calculating the Allan deviation with the spectrum in Fig. 5.19, we obtain
the Allan deviation, as shown in Fig. 5.20. This result suggests that if we make
the acceleration noise is small enough, and use the designed thermal isolation
setup, the instability of the laser of our experiment can reach the 10−17 level.

102
 5.3 Estimation without the acceleration noise

Figure 5.19: Normalized frequency spectrum of the noises in our experiment

without the acceleration noise. The black line is the designed spectrum of our
experiment.

Figure 5.20: Allan deviation calculated using the spectrum in Fig. 5.19.

103
5 Stability Estimation

5.4 Summary of this chapter

• We estimated the effects of many noise sources, and showed that the
acceleration noise is the most problematic one in our experiment.

• We directly evaluated the frequency fluctuation of our laser by measuring

a beat frequency between our laser and another stable laser, which had
a 10−16 level stability. The result of the direct measurement is consistent
with the acceleration noise level at over 1 Hz.
−15
• Our laser stability is σy = 7.0+0.38
−0.34 × 10 at 0.8 second.

• Our laser is the most stable one stabilized with a cryogenic optical cavity
at a very low temperature at around 1 second.

• The frequency fluctuation of our laser is probably limited by the temperature-

fluctuation noise at below 1 Hz.

• In the experiment, the AVIS did not work well, probably because of the
tilt coupling.

• The noise level of our laser can reach the 10−17 level if we make the
acceleration noise small enough.

104
 Chapter 6
Conclusion

In this chapter, we summarize our experiment and discuss the result. We also
show future plans for increasing the laser stability.

6.1 Summary
In this work, we constructed a frequency stabilizing system, and stabilized a
laser frequency with a very low-temperature optical cavity. Moreover, we esti-
mated the noise contributions due to various effects, such as the vibration and
temperature fluctuation, quantitatively. We then showed that the acceleration
noise is the most problematic noise source for our system. We also measured
the instability of our laser directly, and discussed the result. Then, we found
that the problematic noise sources of our stabilization system with a cryo-
genic optical cavity are the acceleration noise and the temperature fluctuation
noise. In addition, we also showed that other noise sources are small enough
for reaching the 10−17 level stability.
We stabilized a laser frequency using a cryogenic monocrystalline silicon
optical cavity. Since monocrystalline silicon has a very high mechanical Q fac-
tor, especially at cryogenic temperature, we can efficiently reduce the thermal
noise of the cavity spacer and mirrors. Moreover, the CTE of this material has
a zero-cross point temperature at around 18 K. We can therefore also reduce
the frequency fluctuation from the temperature fluctuation by controlling the
cavity temperature at around 18 K.
The vibration isolation system is a very important component of our exper-
iment, because the acceleration noise is one of the most problematic technical

105
6 Conclusion

noise sources for frequency stabilization with a reference cavity. A hexapod-

type AVIT can easily decrease the low-frequency vibration compared with the
PVIS. Moreover, AVIS can isolate the acceleration transferred through the
heat link from the refrigerator, while the PVIS cannot isolate it. The PVIS is
a strong structure for reducing the high-frequency vibration, while it is difficult
for the AVIS to reduce the high-frequency acceleration due to the resonances
of its structure.
Thermal-isolation stacks and radiation shields are other important com-
ponents of our experiment for cooling an optical cavity and reducing the
temperature-fluctuation noise. Thermal-isolation stacks are structures that
consist of aluminum plates, stainless-steel inserts and glass balls. It can make
a very steep gradient of the temperature from room temperature to very low
temperature. The radiation shield plays a role for reducing the heat transfer
from this radiation. There are glass windows at the input laser path that trans-
mits the laser beam, which do not transmit the radiation heat to an optical
cavity.
For concerning the input laser system, the laser-power stabilization and the
Doppler noise cancellation techniques are used. The laser-power stabilization
system with the transmitted light decreases the laser-power fluctuation noise.
The Doppler noise cancellation system reduces the phase fluctuation, while a
laser beam travels some optical paths. Each system is an essential component
for reaching the 10−17 level frequency stability.
Using these components, we have designed a reference cavity for reaching
−17
10 level stability. The designed performance is dominantly limited by three
noises: thermal noise, acceleration noise and RAM noise.
Using the designed setup, we have obtained a frequency stability of σy =
−15
−0.34 × 10
7.03+0.38 at 0.8 second. This laser is the most stable at around 1
second by using a cryogenic optical cavity at very low temperature. Note that
we turned off the AVIS, and used the simplified thermal isolation setup when
we must obtain the above frequency stability.
This stability is dominantly limited by the acceleration noise and the
temperature-fluctuation noise. Thus, we can improve the stability by reduc-
ing the acceleration noise and vibration sensitivity of a reference cavity and
constructing the designed setup for temperature-fluctuation noise reduction.

106
 6.2 Future prospects

6.2 Future prospects

The laser-frequency stability is limited by the acceleration noise in our exper-
iment as well as the temperature-fluctuation noise. Therefore, by decreasing
the acceleration noise and the temperature-fluctuation noise, the laser stabil-
ity should be improved. Since the temperature-fluctuation noise will be small
when we use the designed setup for thermal isolation, we only consider the re-
duction method of the acceleration noise in this subsection. There are several
methods to reduce this acceleration noise.
First, we should make the diagonalization of vibration signals on the AVIT
more exactly. In the recent configuration, the AVIS didn’t help us to improve
the stability of our laser. This is because the diagonalization matrix is not
exactly correct, and the top plate of the AVIT is tilted when we operate the
AVIT. Thus, by decreasing the coupling between the translational vibration
signals and the rotational ones, the AVIS should work better, and we can
reduce the low-frequency acceleration of a reference cavity.
The second method is to use a stronger PVIS, such as an air spring vibration
isolator, an inverted pendulum, a geometrical anti-spring filter, etc. These
passive vibration isolators have very low resonant frequencies, from around 0.1
Hz to 0.5 Hz, and small Q factor. Then, the vibration-isolation ratio improves
by more than 100-times larger at high frequency.
The third method is to decrease the vibration sensitivity of a reference
cavity. Our cavity has a much larger vibration sensitivity than that which we
designed; it is insensitive to the support point of an optical cavity because the
aspect ratio of our cavity is close to one. However, if we remake the cavity,
which has the relevant aspect ratio, for example at around four, which is the
aspect ratio used in a cavity whose vibration sensitivity is below 10−11 /(m/s2 )
in all translational axis [12], the vibration sensitivity should be more sensitive
to changing the support points, and we can adjust the support point for making
the vibration sensitivity smaller.
Fourthly, we could move our experimental setup to a place that has a
smaller seismic vibration, like Kamioka in Japan. The vibration levels from
1 Hz to 100 Hz at Kamioka are about 100-times smaller than those at our
laboratory at RIKEN in Saitama Prefecture [46]. The stability of the laser

107
6 Conclusion

frequency could be improved even if we do not change any configuration of our

experiment.

6.3 Conclusion
The laser-frequency stability has been gradually improving during the last
decade, and a state-of-the-art cavity-stabilized laser has been approaching the
thermal noise limit of the reference cavity. Therefore, it is inevitable for further
improvements of the laser stability to utilize a cryogenic cavity operated at a
very low temperature.
In this work, we stabilized the laser frequency with a cryogenic optical
cavity, and the stability of our laser is measured at an Allan deviation of
+0.38
σy = 7.03−0.34 × 10−15 . This stability is worse than that of the very stable
lasers in the world. However, our laser is the most stable laser stabilized with
a cryogenic optical cavity operated at very low temperature at around 1 second.
Moreover, we have studied various noise sources that may disturb the laser
stabilization, and found that our laser stability is limited by the accelera-
tion noise and the temperature-fluctuation noise. In addition, the other noise
sources, such as the laser-power fluctuation noise, the RAM noise, etc, are
small enough for achieving the 10−17 level of stability. In this way, our work
reveals the prospects for reaching the 10−17 level of stability.
Since cooling the cavity at a very low temperature is an indispensable
technique for further improving the laser stability, for example 1 × 10−17 , our
work is an important step for making an ultrastable laser in the future.

108
 Appendix A
Gaussian Beam and Cavity
Mode

In the main part of this thesis, we treated a laser beam using geometrical optics.
However, in reality, the electromagnetic field of a laser beam practically has
an intensity distribution, and we should consider this distribution in order to
calculate some parameters of a laser, like a beam size. In Appendix A, we
describe Gaussian beam optics.

A.1 Gaussian beam

The amplitude of a laser beam traveling in the direction along the Z axis can
be expressed as
E(t, x, y, z) = Ψ(x, y, z)eiΩt−ikz , (A.1)

where Ω is the angular frequency of a laser beam and k = Ω/c is the wave num-
ber. Ψ(x, y, z) is the distribution of an electromagnetic field, which satisfies
the following equation:
( )
∂2 ∂2 ∂
2
+ 2
− 2ik Ψ(x, y, z) = 0. (A.2)
∂x ∂y ∂z

One solution of this equation can be described as

√ (
x2 +y 2
)
k izR − q(z)
ik
2
Ψ(x, y, z) = e , (A.3)
πzR q(z)

109
A Gaussian Beam and Cavity Mode

where q(z) is the wavefront coefficient, which can be written as

q(z) = z + izR , (A.4)

where zR is the Rayleigh range, which is expressed as

kw02
zR = . (A.5)
2
Here, we consider the physical meaning of w0 . Using Eq. (A.3), the electro-
magnetic field distribution at z = 0 can be written as
√ (
2 2
)
k − x +y
2w0
Ψ(x, y, 0) = e . (A.6)
πzR

From this equation, it is found that the amplitude of the electromagnetic field
at a distance w0 is 1/e of the amplitude at the center.

We now consider the functions R(z) and S(z) defined as follows:

1 1 1
= + (A.7)
q(z) R(z) iS(z)

and define the function w(z) as

2S(z)
w2 (z) = . (A.8)
k
Using Eq. (A.7) and Eq. (A.8), Eq. (A.3) can be written as
√ [ ( ) ]
2 1 − 1
w2 (z)
ik
+ 2R(z) (x2 +y 2 )+i arctan(z/zR )
Ψ(x, y, z) = e . (A.9)
π w(z)

From the above calculation, R(z) and w(z) can be written as

[ ( z )2 ]
R
R(z) = z 1 + , (A.10)
z
√ ( )2
z
and w(z) = w0 1+ . (A.11)
zR

110
 A.2 Cavity eigen mode of a FPC

From Eq. (A.9), it is found that R(z) is the radius of curvature of the equiphase
plane of a laser beam, and w(z) is the beam size at a distance of z.

A.2 Cavity eigen mode of a FPC

In order to resonate the laser beam to a FPC, R(z) on the front and the end
mirrors should be matched with the radii of curvatures (ROC) of the mirrors.
So, R(z) satisfies the equations as follows:
[ ( )2 ]
zR
R(z0 ) = z0 1 + = −RF , (A.12)
z0
[ ( )2 ]
zR
and R(z0 + L) = (z0 + L) 1 + = RE . (A.13)
z0 + L
Here, z0 is the coordinate of the front mirror and L is the cavity length; RF and
RE are the radii of curvatures of the front and end mirrors, respectively. The
sign of ROC is positive when the mirror surface inside a FPC is concave. From
these equations, we can calculate z0 , zR , and w0 using the following equations:

L − RE
z0 = L, (A.14)
RF + RE − 2L

k 2 w04 L(RF − L)(RE − L)(RF + RE − L)

zR2 = = . (A.15)
4 (RF + RE − 2L)2
We can thus calculate the beam size on the front and end mirrors using
Eq. (A.11), Eq. (A.14), and Eq. (A.15).
We give some examples of the beam size on the mirrors in Table A.1

111
A Gaussian Beam and Cavity Mode

RF RE w(z0 ) w(z0 + L)
0.5 m 0.5 m 3.33 × 10−4 m 3.33 × 10−4 m
0.5 m 1.0 m 3.77 × 10−4 m 3.27 × 10−4 m
0.5 m ∞ 4.26 × 10−4 m 3.30 × 10−4 m
1.0 m 1.0 m 3.85 × 10−4 m 3.85 × 10−4 m
1.0 m ∞ 4.71 × 10−4 m 4.22 × 10−4 m

Table A.1: Several examples of the beam size on the mirrors of a FPC. The
parameters for the calculation are L = 0.2 m and k = 4.50 × 106 rad/m.

112
 Appendix B
Allan Deviation

When evaluating the frequency instability of a laser, the standard deviation

cannot be used for the criterion of the frequency stability because it can be
changed by the measurement time if the laser frequency has some drift. There-
fore, the Allan deviation is often used for estimating the frequency stability,
which is not varied by the measurement time, even if the frequency has a drift.

B.1 Definition
The Allan deviation is defined as
√
⟨(yi+1 − yi )2 ⟩
σy (τ ) = , (B.1)
2
where < Xi > means an ensemble average of Xi , and τ is a sampling interval;
yi is the i-th measured mean value of a physical quantity of y(t), which is
defined as ∫
1 ti +τ
yi = y(t) dt. (B.2)
τ ti
An ensemble average can be defined as

1 ∑
N
< Xi >= Xi . (B.3)
N n=1

113
B Allan Deviation

Then, Eq. (B.1) can be rewritten as

v
u
u 1 ∑
N −1
σy (τ ) = t (yi+1 − yi )2 . (B.4)
2(N − 1) n=1

In this thesis, we calculated the Allan deviation from the sampling data with
Eq. (B.4).

B.2 Allan deviation and power spectrum

The PSD of the physical quantity y and its Allan deviation are connected as
follows [47]: √∫
∞
σy (τ ) = Gy (f )H(f ) df , (B.5)
0

where H(f ) is a dimensionless function, defined as

sin4 (πτ f )
H(f ) = 2 . (B.6)
(πτ f )2

We show the behavior of H(f ) as a function of Fourier frequency in Fig. B.1.

Equation (B.6) means that the Allan deviation at an averaging time of τ is
dominantly determined by the power spectrum at around f = 1/2τ . Thus,
in this thesis we calculated the Allan deviation using the frequency spectrum
from about 0.01 Hz to 100 Hz with Eq. (B.5), because we care about the Allan
deviation at around 1 second.

114
 B.2 Allan deviation and power spectrum

Figure B.1: Value of the response function H(f ) (τ = 1 s).

115
 Appendix C
Electrical Circuit in our
Experiment

In Appendix C, we give details about the electrical circuits in our experiment.

10 kΩ

-+ output
ADA 4817-1
PD (Analog Devices)
GND

-3 V

Figure C.1: Electric circuit for the InGaAs PD G12180-003A and G12180-250A
(Hamamatsu Photonics’s products).

15 pF

100 kΩ

-+ output

PD OP27
(Analog Devices)

GND

GND

Figure C.2: Electric circuit for the InGaAs PD G12181-020K (Hamamatsu

Photonics’s product).

117
C Electrical Circuit in our Experiment

pre. amp. port1 & amp.1

offset switch
PLP-21.4+
(Mini-Circuit) CXO-043C monitor1
(IQD Frequency Products)
21.4 MHz 330 pΩ 12 kΩ 100 nF
to EOM LO 15 MHz 60 kΩ monitor2
LPF
port1
21.4 MHz PLP-21.4+ 47 kΩ
LPF
120 kΩ
200 kΩ 2 kΩ
PD signal × 5 MHz
LPF -+ -+
AD829
RAY-3 600 Ω AD829
(Mini-Circuit) PLP-5+ (Analog Devices) 47 kΩ 5
(Mini-Circuit)

51 Ω 47 pF
5.1 kΩ 68 pΩ
GND

1 kΩ switch high freq. filter

220 Ω 10 MΩ
-+ 3.9 kΩ
1 nF 60 Ω 2 kΩ
OP37
(Analog Devices)
1.6 kΩ
330 pF AD829
5.1 kΩ -+
510 Ω

1 kΩ
5 kΩ 1 μF to LD
switch

1 μF

200 kΩ port2 & amp.2

-+
1 kΩ 20 kΩ

OP297 10 kΩ 20 kΩ
(Analog Devices) offset

20 kΩ OP297
-+
low freq. filter 20 kΩ
port2
to PZT

Figure C.3: Electric circuit for frequency stabilization. Several model numbers
used in the circuit are given in this figure. Each block surrounded by red lines
represents the components of the block diagram in Fig. 4.19

10 nF
GND 1 kΩ

100 kΩ

82 kΩ

vibration
sensor
-+ -+ output
OP470 8.2 kΩ
L-22E, 510 Ω (Analog Devices)
OP470
(IRIS PASSCAL)

33 kΩ

Figure C.4: Electric amplifier for the vibration sensor. Several model numbers
used in the circuit are given in this figure. The gain of this circuit is about
1000.

118
 -15 V
10 kΩ 10 kΩ
10 μF 10 μF 10 kΩ
AOM
AD587
(Analog Devices) 100 kΩ
100 kΩ 100 kΩ OP27
-+
voltage
+15 V
voltage
reference -+ 100 kΩ
-+ OP27
10 kΩ
current
buffer variable
attenuator
10 μF OP27 -+ BUF634
(Analog Devices) ZX73-2500-S+
10 kΩ (Mini-Circuit)

OP470 switch VCXO

10 kΩ FXO-PC73
(Fox Electronics)

10 kΩ 100 kΩ
1 μF 1 μF
10 kΩ
10 kΩ
-+
10 kΩ
-+
PD signal 1 kΩ 1 MΩ 100 kΩ 1 kΩ
-+
1 kΩ
-+
1 kΩ
OP470
OP470 switch
OP27 -+
OP27 OP27
monitor port

Figure C.5: Electric circuit for laser power stabilization. Several model num-
bers used in the circuit are shown in this figure.

AFG3252 (Tektronix)
LO
1.5 nF 47 kΩ
frequency
doubler 1 MΩ
FD-2+
(Mini-Circuit)
PD signal × -+ OP37

RAY-3 510 Ω
1 kΩ
51 Ω
1 kΩ

GND
GND

switch 1 kΩ

1 kΩ
FXO-PC73
offset -+ VCXO AOM
5.1 kΩ
OP27 3.9 kΩ
+3.3 V

Figure C.6: Electric circuit for Doppler noise cancellation. The model numbers
used in the circuit are given in this figure.

119
 Appendix D
Photos of Experimental
Setup

In Appendix D, we show several photos from our experimental setup.

Figure D.1: Appearance of our setup. The cryostat is about 1.5 m in height
and 0.8 m in width. The rack is a digital feedback system for the AVIS using
a LabView real-time machine.

121
D Photos of Experimental Setup

Figure D.2: Monocrystalline silicon optical cavity.

Figure D.3: Thermal-insulation stacks and an inner radiation shield. A silicon

optical cavity is inside the inner shield.

122
Figure D.4: Appearance inside the cryostat. The optics in front of an outer
shield are ones for mode-matching. Behind an outer shield, there are some
optics for monitoring the transmitted light through a FPC. The outer shield
has two heaters for temperature control.

Figure D.5: Main part of an input laser system. This part is covered by a
wind shield for reducing the Doppler noise.

123
 Bibliography

1 Introduction

[1] T. H. Maiman, Nature 187, 493-494 (1960).

Stimulated Optical Radiation in Ruby

[2] K. Kuroda, Class. Quantum Grav. 27, 084004 (2010).

Status of LCGT

[3] G. M. Harry, Class. Quantum Grav. 27, 084006 (2010).

Advanced LIGO: the next generation of gravitational wave detectors

[4] T. Accadia, et al., J. Phys.: Conference Series. 203, 012074 (2010).

Status and perspectives of the Virgo gravitational wave detector

[5] B. D. Tapley, et al., Geophys. Res. Lett. 31, L09607 (2004).

The gravity recovery and climate experiment: Mission overview and early
results

[6] M. Takamoto, et al., Nature 435, 321-324 (2005).

An optical lattice clock

[7] M. Takamoto, et al., Nat. Photon. 5, 288-292 (2011).

Frequency comparison of optical lattice clocks beyond the Dick limit

[8] M. Yamanaka, et al., Phys. Rev. Lett. 114, 230801 (2015).

Frequency Ratio of 199 Hg and 87 Sr Optical Lattice Clocks beyond the SI
Limit

[9] http://www.npl.co.uk/science-technology/time-frequency/
optical-frequency-standards-and-metrology/units/
secondary-representations-of-the-second

125
Bibliography

[10] S. Peil, et al., Phys. Rev. A 87, 010102 (2013)

Tests of local position invariance using continuously running atomic clocks

[11] Santarelli, G. et al., IEEE Trans. Ultrason. Ferroelectr. Freq. Control 45,
887-894 (1998).
Frequency stability degradation of an oscillator slaved to a periodically
interrogated atomic resonator.

[12] T. Kessler, et al., Nat. Photon. 6, 687-692 (2012).

A sub-40-mHz-linewidth laser based on a silicon single-crystal optical cav-
ity

[13] M. D. Swallows, et al., IEEE Trans. Ultrason. Ferroelectr. Freq. Control

59, 416-425 (2012)
Operating a 87Sr Optical Lattice Clock With High Precision and at High
Density

[14] A. D. Ludlow, Doctoral Thesis, p170 (The university of Colorado, 2008)

The Strontium Optical Lattice Clock: Optical Spectroscopy with Sub-Hertz
Accuracy
https://jila.colorado.edu/yelabs/sites/default/files/uploads/
theses 2008 AndrewLudlow.pdf

[15] I. Ushijima. et al., Nat. Photon. 9, 185-189 (2015).

Cryogenic optical lattice clocks

[16] A.Yu. Nevsky, et al., Optic. Comm. 210, 91-100 (2002).

A Nd:YAG Laser with short-term frequency stability at the Hertz-level

[17] S. A. Webster, M. Oxborrow, and P. Gill, Optic. Lett. 29, 1497-1499

(2004).
Subhertz-linewidth Nd:YAG laser

[18] A.Yu. Nevsky, et al., Optic. Lett. 32, 641-643 (2007).

Compact, thermal-noise-limited optical cavity for diode laser stabilization
at 1 × 10−15

126
Bibliography

[19] J. Alnis, et al., Phys. Rev. A. 77, 053809 (2008).

Subhertz linewidth diode lasers by stabilization to vibrationally and ther-
mally compensated ultralow-expansion glass Fabry-Pérot cavities

[20] J. Millo, et al., Phys. Rev. A. 79, 053829 (2009).

Ultrastable lasers based on vibration insensitive cavities

[21] J. Millo, et al., Nat. Photon. 5, 158-161 (2011).

Making optical atomic clocks more stable with 10216-level laser stabiliza-
tion

[22] G. D. Cole, et al., Nat. Photon. 7, 644-650 (2013).

Tenfold reduction of Brownian noise in high-reflectivity optical coatings

[23] E. Wiens, et al., Optic. Lett. 39, 3242-3245 (2014).

Silicon single-crystal cryogenic optical resonator

[24] F. Häfner, et al., Optic. Lett. 40, 2112-2115 (2015).

8 × 10−17 fractional laser frequency instability with a long room-
temperature cavity

[25] Kenji Numata, Amy Kemery, and Jordan Camp, Phys. Rev. Lett. 93,
250602 (2004)
Thermal-Noise Limit in the Frequency Stabilization of Lasers with Rigid
Cavities

2 Frequency Stabilization

[26] T. Ushiba, Master Thesis, p5-p19 (The University of Tokyo, 2013).

「低温シリコン光共振器を用いた高安定化光源の開発」 (Development
of Highly Stabilized Laser using a Cryogenic Silicon Optical Cavity; in
Japanese)
http://granite.phys.s.u-tokyo.ac.jp/theses/ushiba m.pdf

[27] M. Ando, Master Thesis, p25-p35 (The University of Tokyo, 1995).

「Fabry-Perot 型重力波検出器の制御」 (Control of the Fabry-Perot type
gravitational wave detector; in Japanese)
http://t-munu.phys.s.u-tokyo.ac.jp/theses/ando m.pdf

127
Bibliography

[28] A. L. Schawlow and C. H. Townes, Phys. Rev. 112, 1940-1949 (1958).

Infrared and Optical Masers

[29] D. Shoemaker, et al., Optic. Lett. 14, 609-611 (1989).

Frequency-stabilized laser-diode-pumped Nd:YAG laser

[30] R. W. P. Drever, et al., Appl. Phys. B 31, 97-105 (1983).

Laser phase and frequency stabilization using an optical resonator

[31] E. D. Black, Gen. Am. J. Phys. 69, 79-87 (2001).

An introduction to Pound-Drever-Hall laser frequency stabilization

[32] K. Yamamoto, Doctoral Thesis, p25-p36 (The University of Tokyo, 2000).

Study of the thermal noise caused by inhomogeneously distributed loss
http://t-munu.phys.s.u-tokyo.ac.jp/theses/yamamoto d.pdf

[33] K. Kokeyama, et al., J. Opt. Soc. Am. A 31, 81-88 (2014).

Residual Amplitude Modulation in Interferometric Gravitational Wave
Detectors

[34] S. Moriwaki, 『重力波をとらえる』 edited by T. Nakamura, N. Mio, M.

Ohashi, p.224-228 (Kyoto University Press, 1998).

3 Experimental Design

[35] L. S. Ma, et al., Optic. Lett. 19, 1777-1779 (1994).

Delivering the same optical frequency at two places: accurate cancellation
of phase noise introduced by an optical fiber or other time-varying path

[36] D. F. McGuigan, et al., J. Low Temp. Phys. 30, 621-629 (1978).

Measurements of the Mechanical Q of Single-Crystal Silicon at Low Tem-
peratures

[37] K. G. Lyon, et al. J. Appl. Phys. 48, 865-868 (1977).

Linear thermal expansion measurements on silicon from 6 to 340 K

[38] Hartnett, J.G., et al., Ultrasonics, Ferroelectrics, and Frequency Control

57, 1034-1038 (2010).

128
Bibliography

Cryogenic sapphire oscillator using a low-vibration design pulse-tube cry-

ocooler: first results

[39] J. J. Wortman and R. A. Evans, J. Appl. Phys. 36, 153-156 (1965).

Young’s Modulus, Shear Modulus, and Poisson’s Ratio in Silicon and Ger-
manium

[40] P. R. Saulson Fundamentals of Interferometric Gravitational Wave De-

tectors, p.127 (World Scientific, Singapore, 1994).

[41] G. K. WHITE and P. J. MEESON,

”Experimental Thchniques in Low-Temperature Physics” (OXFORD SCI-
ENCE PUBLICATIONS)

[42] T. Akatsuka, et al., Jpn. J. Appl. Phys. 53, 032801 (2014).

30-km-long optical fiber link at 1397 nm for frequency comparison between
distant strontium optical lattice clocks

4 Performance Evaluation

[43] http://www.sercel.com/products/Lists/ProductSpecification/
Geophones specifications Sercel EN.pdf

5 Stability Estimation

[44] W. Zhang, et al., Optic. Lett. 39, 1980-1983 (2014).

Reduction of residual amplitude modulation to 1×10−6 for frequency mod-
ulation and laser stabilization

[45] A. Shoda, Doctoral Thesis, p50-p54 (The University of Tokyo, 2014).

Development of a High-Angular-Resolution Antenna for Low-Frequency
Gravitational-Wave Observation
http://granite.phys.s.u-tokyo.ac.jp/shoda/Thesis shoda.pdf

6 Conclusion

129
Bibliography

[46] T. Sekiguchi, JGW 1, 1402971 (2014).

Seismic Spectrum in Kamioka Mine
http://gwdoc.icrr.u-tokyo.ac.jp/DocDB/0029/T1402971/001/
141106 seismic noise.pdf

B Allan Deviation

[47] T. J. Witt, IEEE Instrum. Meas. 50, 445-448 (2001).

Using the Allan Variance and Power Spectral Density to Characterize DC
Nanovoltmeters

130