Beruflich Dokumente
Kultur Dokumente
GENERATION, AMPLIFICATION
AND NONLINEAR EFFECTS
Thesis submitted for the degree of Doctor of Philosophy of the University of London
UNIVERSITY OF LONDON
Department of Physics
November 1990
1
2
ABSTRACT
solitons in an all optical communications system, and soliton evolution in the presence
of relatively high gain in the adiabatic regime has been investigated. In particular, the
produce femtosecond soliton structures, which in some cases may be useful. In others,
gain provided by Raman or Erbium doped fibre amplification. The sources have
various applications for soliton investigations, operating from 1.0 to 1.6μm. Using
these sources, amplification in doped fibres has been investigated in the picosecond
of fixed or tunable repetition rate (THz) soliton-like pulses. These ultra-high repetition
rate trains may be of importance for use as control signals in multiplexing circuits or
routing switches. The process has been induced through cross phase modulation and
through a four wave mixing interaction. Cross phase modulation has been used to
power.
3
Since noise can evolve into solitons in the presence of gain, a method of
considered, based on a nonlinear optical loop mirror. The role of gain or loss in such a
considerably.
4
TABLE OF CONTENTS Page No.
ABSTRACT 3
1 GENERAL INTRODUCTION
1.1 Introduction. 9
1.6 References. 48
2.4 References. 66
3.5 References. 89
5
4 PULSE AMPLIFICATION IN DOPED FIBRES Page No.
AMPLIFICATION
6.1 Soliton Raman Generation. 147
6
7 NONLINEAR OPTICAL LOOP MIRROR Page No.
PULSE PROCESSING
7.1 Introduction. 173
8 CONCLUSIONS 193
References 195
References 204
ACKNOWLEDGEMENTS 208
7
8
1. GENERAL INTRODUCTION
1.1 Introduction
magnitude greater than that obtainable with present day transmission media. However,
severe limitations exist in the optical power and speed with which information can be
relayed down a fibre. These limitations have only come to light in recent years, since
absorption losses have been reduced. The advent of low loss high qualtiy silica optical
fibres now enables transmission over more than ~20km with only half power loss.
the necessity for electronic repeaters which limit the system bandwidth, can extend the
system length to over 1000km. Over long lengths, cumulative nonlinearity causes
always present.
distances multiply the effect, which scales with length. Nonlinearity places a power
limit on the pulses used in transmission and exaggerates the effect of dispersion in
many cases where spectral broadening occurs. The peak power of a pulse induces a
can exactly balance the effect of nonlinear spectral broadening, resulting in the
considered, with relevance to long distance transmission and ultra high bit rate
9
communications.
exact requirements, as certain changes in system parameters can lead to large changes
in response. For example, in a high repetition rate system, small changes in inter pulse
spacing can lead to detection errors through pulse-pulse interactions. Optical fibre
studied.
and soliton-like structures, which are used to illustrate nonlinear coupling. The
application of all fibre amplification to long distance and to ultra-high bit rate
transmission is considered. Two methods are considered: active (doped) fibre and
power pulses are shown to results in soliton generation, and the influence of seed
doped fibre amplifiers are considered, in the subpicosecond and femtosecond regimes.
at Terahertz repetition rates, for use as signal or clock pulses in a soliton based
transmission system. The potential for producing Terahertz repetition rate pulse trains
ultrashort pulse generation through soliton shaping of weak signals in the anomalous
dispersion regime.
10
1.2 Historical Perspective - Solitons in Fibres
The basis of soliton propagation in optical fibres has been the subject of
scientific interest for more than one hundred years. Numerical and analytical work on
the solitary wave solutions of the nonlinear Schroedinger equation (NLS) have
produced exact solutions using the inverse scattering method1. The NLS describes the
coherent pulses in optical fibres. The two physical properties which make soliton
propagation possible are dispersion and nonlinearity, and hence the NLS is directly
applicable. Dispersion in fibres arises from three sources: modal, waveguide and
material, and always results in pulse broadening, irrespective of its sign, in the absence
of nonlinearity. A pulse of finite duration requires a finite bandwidth to support it, and
hence shorter pulses suffer more relative dispersion than longer ones. However it was
demonstrated numerically that nonlinearity, which relates the wave velocity to the
intensity, can exactly compensate for dispersion, resulting in the possibility of the
communications:
i) shape, width and speed of soliton pulses are preserved in the abscence of loss,
iii) solitons survive collisions whilst maintaining shape and speed (elastic
scattering)
iv) solitons represent the common asymptotic state for a variety of completely
11
This latter property is of special significance in chapter 6 of this thesis, where
propagation distances. Since dispersion occurs over a length scale characteristic of the
initial pulse duration and fibre parameters, ultrashort pulses must propagate over a
sufficient length for dispersive effects to become noticeable. For high power pulse
experiments awaited manufacture of optical fibres with low loss, wich can now be as
low as ~0.2dBkm-1 in the and 1.55μm region4, and high power lasers producing
dispersion regime. ~10ps pulses at ~1.5μm from an F-centre NaCl colour centre
source were used. The results were explained5,6,7 as an exact or over balence between
Self Phase Modulation (SPM)8 and negative Group Velocity Dispersion9. Group
Velocity Dispersion (GVD) arising from the second order derivative of the
propagation constant in the fibre, vanishes to zero around 1.3μm, becomes negative
above and positive below for standard single-mode fibre (see section 1.3). Propagation
of low intensity (linear regime) pulses in a medium with GVD, regardless of sign,
dependent change in the local refractive index across the pulse, through a process
12
induces an approximately linear chirp across the centre of a pulse. In the positive GVD
regime, the two chirping mechanisms (positive GVD and SPM) work together and the
result is enhanced temporal broadening. In the negative GVD regime however, the
processes oppose one another and may balance. Propagation can then occur without
range of powers around a critical value, an exact balance can be struck between
dispersion (if negative) and nonlinearity. A pulse of this nature propagates without
change in shape and is known as the fundamental (or N=1) soliton. At higher powers,
an overbalance occurs in favour of the nonlinearity, and the pulse shape undergoes a
series of continuous temporal and spectral reshapings over the soliton period.
Traditionally pulses whose temporal duration was limited only by their spectral
bandwidth, known as transform-limited pulses, have been used to excite solitons, since
this results in the lowest required energy and most predictable behaviour. In this
pulses consisting of noise are used to illustrate the universality of the steady state
soliton solution.
medium. In the presence of loss, the energy of a soliton is dissipated. It has been
shown 10,10a,10b,10c,10d that in the presence of a small loss, treated as a perturbation, the
In the presence of gain as opposed to loss, the reverse applies, hence optical
pulse reshaping can be achieved through amplification. Two techniques for optical
amplification are discussed in this thesis: doped fibre and Raman amplification. Over
many thousands of kilometres, using Raman gain to compensate for loss, distortionless
transmission has been demonstrated over 6000km using 50ps pulses11 (more recently
13
100km transmission has been achieved11a). By exceeding optical loss with gain,
soliton compression has been achieved, in which the pulse duration shortens in order
All of the properties desribed above refer to the propagation of bright envelope
solitons derived from pulses which would normally disperse under linear conditions.
The durations are less than ~100ps, in order for dispersive broadening to become
noticeable over the typical fibre lengths used. An interesting mode of propagation
exists however if high power continuous signals or very long (>50ps) quasi-
continuous pulses are used. It has been shown that these waves are unstable, and
a train of solitary waves. The process results in the growth of an amplitude modulation
nonlinearity and dispersion, and is ultimately limited by higher order dispersive terms.
In this thesis, techniques for generating pulses at up to 2THz repetition rate are
change which is proportional to the local pulse intensity. The central portion therefore
undergoes a larger phase change than the wings. Hence by incorporating nonlinearity
reshaping and switching of optical pulses15. A device called a Nonlinear Optical Loop
Mirror (NOLM) has been proposed as a means of switching optical pulses, since it
used to 'clean up' the pedestals on noisy pulses through its intensity dependent
transmission.
Optical fibres confine intense optical radiation within a core which is typically
14
less than 100μm in area, over extremely long lengths, they are ideal media for
investigating and exploiting many nonlinear effects occurring at relatively low average
powers. Many of those observed so far include: Self Phase Modulation (SPM)8,
for example. Some of the consequences are exploited in this thesis, to provide tunable
sources of radiation and to study coupling between signals. All of the relevant physical
material such as silica of very high purity, surrounded by a cladding layer of similar
material but with a very slightly lower refractive index. Fig 1.3.1. shows a schematic
illustration of a standard, step index fibre, showing the step-like refractive index
fibre radius, b, is much larger (typically 60μm) than the core, and has little effect on
propagation within the limits considered here. The core diameter 2a (typically 4-
10μm) determines the number of transverse modes which can propagate. Larger core
diameter fibres can support more than one transverse mode (multimode) but are not
considered here.
15
Fig.1.3.1 Schematic of the cross-section and refractive index profile of a step-index
fibre.
where λ is the guided light wavelength, and Δn=n1-n2. It can be shown that for
V<2.405, only a single transverse mode is allowed to propagate. Higher order modes
all have a different propagation constant, resulting in substantial differences in flight
times if a pulse is transmitted in a multimode fibre. This results in a large intermodal
dispersion occurring in multimode fibres, however only single mode fibres are
considered here.
16
exists below which the fibre will become multimode, known as the cut-off
wavelength, λc. Above this wavelength, the beam waist becomes progressively larger,
Typically a fibre with Δn=0.017, a=2.5μm, and a λc=1.2μm single mode cutoff
wavelength will from Eq(1.3.1) have a V-number of V=1.9. From Fig.1.3.2., the ratio
of the beam waist to the core diameter a for V=1.9 is w/a=1.3. Hence the mode-field
area is 33μm2, which is substantially larger than the core area, which is 20μm2.
Throughout this thesis, the mode-field area is quoted where relevant, as many of the
calculations are intensity dependent.
17
The major cause of pulse spreading from dispersion in single mode fibres
arises from the finite bandwidth of the signal source. The group delay for each
component in the wave packet is dependent on the wavelength through both material
U
Waveguide dispersion can be visualised as the small change in the group velocity
which occurs as more of the mode field lies in the cladding rather than the core of the
fibre, as the signal wavelength deviates from the single-mode cutoff value.
βω =nω ω
c (1.3.2)
where n(ω) is the refractive index at frequency ω, and c is the vacuum speed of light.
The dependence of the propagation β on the wavelength λ is characterised by a series
of differentials:
18
m
β m= d β where m=0,1,2...
dωm
(1.3.3)
the relation:
β1= 1
Vg (1.3.4)
β2 is given by:
β 2 = d 1 = -1 dVg
dω Vg Vg dω (1.3.5)
19
and is therefore a measure of the dependence of the group velocity on frequency or
wavelength. From Fig.1.3.3. it can be seen that GVD is positive below and negative
above 1.27μm, at which wavelength it vanishes to zero. This is know as the zero GVD
wavelength λ0. However total dispersion does not vanish altogether. The total can be
expanded as a Taylor series around a central frequency ω0 as follows:
the exact shape of the refractive index profile function with radial distance given in
Fig.1.3.1.
single mode fibre with λo=1.31μm, on the same axes as the measured group delay
function.
20
Fig.1.3.4. Measured group delay and computed total dispersion parameter D(ps nm-1
km-1) vs wavelength for a standard telecommunications grade fibre.(Courtesy BTRL).
The group delay function is measurable, from which the total dispersion can be
approximately deduced using Eq(1.3.5). Note that although β2 is expressed in units of
dispersion in Fig.1.3.4. The dispersion parameters are important since they quantify
the amount of broadening experienced by a pulse with a known bandwidth or duration,
and also determine how long signals at different wavelengths will overlap and
possibly interact within a fibre length, the so-called 'walk-off length'.
21
2
LD = T
β2 (1.3.8)
then it can be shown26,27 that the electric field within the fibre can be reduced to a
radial part and a longitudinal part. The radial part is neglected25 by assuming it does
not change with propagation in Z, and the following equation for the evolution in Z is
obtained:
β 2 Ž2 U
- i ŽU + =0
ŽZ 2 ŽT2
(1.3.10)
This equation shows how the pulse shape U(Z,T) distorts with propagation distance Z.
Since ∂2U/∂T2 becomes arbitrarily large for shorter pulses, the envelope distortion is
exaggerated for shorter pulses.
For an initially Gaussian pulse shape with (1/e intensity point) duration τ0 the
it can be shown25 that the solution of Eq(1.3.10) yields an output pulse of the form:
22
-1 - T2
iβ 2 Z 2
U Z,T = 1+ exp iβ 2 Z
τ20 2τ20 1+
τ20 (1.3.12)
From this result it can be seen that the pulse envelope broadens according to
the following well know formula28 for the output pulse duration τ1:
2
iβ 2 Z 2
τ1 = τ0 1+ = τ0 1+ LZ
τ20 D
(1.3.13)
order dispersion terms such as ∂3β/∂ω3 will begin to dominate in Eq(1.3.6) and would
need to be considered in propagation as they too result in pulse distortion. Pulse
propagation around λo is very important in linear communications, and provides the
highest bit rate transmission provided no nonlinear effects are allowed to manifest
themselves. In fact low distortion picosecond pulse propagation has been shown in
lengths of fibre at λ0=1.32μm, with pulses of 5ps over 1km28a. At the zero dispersion
1
τmax (ps) = 1.4 L 3 (km) (1.3.13a)
and hence for a 20km fibre, τmax is 3.8ps, which corresponds to a maximum
23
Eq(1.3.13) can be rewritten:
which indicates that the phase of the output pulse is evolving with a quadratic
dependence on time, and with a sign of curvature given by that of β2. For λ<λ0, i.e. a
pulse in the normal dispersion regime, β2>0 and the long wavelength (red shifted)
components travel faster along the fibre than the short (blue) components. The reverse
occurs for β2<0, in the region λ>λ0 known as the anomalous dispersion regime. The
time derivative of the instantaneous pulse phase defines a local frequency within the
pulse, which acquires a linear characteristic in the central region of the pulse known as
a frequency chirp. A chirped pulse has a longer duration than that determined by the
Fourier-transform limit of its spectral bandwidth. The chirp induced through quadratic
dispersion can be either positive or negative, depending on the sign of the dispersion
parameter D. Dispersive broadening always occurs for a finite β2, unless the intensity
of the signal pulse becomes sufficiently high that nonlinear effects start to dominate.
This is discussed in the next section.
Fig 1.3.5. Dispersive broadening of a Gaussian input pulse (shown dashed) at Z=2LD
and Z=4LD (Arbitrary scales).
24
Fig.1.3.5. shows the temporal dispersive broadening induced on an input pulse
launched at Z=0 (shown dashed), after propagating over lengths Z=2LD and Z=4LD,
where LD is the dispersion length given in Eq.(1.3.8). Clearly, dispersion places a
definite upper limit to the transmission bandwidth of an optical fibre transmission line
enabling only a small fraction of the available bandwidth around the low disperion,
low loss wavelength region in the fibre to be used. The lowest loss wavelength region
is around 1.5μm , as shown in Fig.1.3.6.
Fig.1.3.6. Recorded loss profile for a standard single mode fibre. The dashed curve
represents intrinsic loss present through Rayleigh scattering and absorption in silica.
(Courtesy BTRL).
The zero dispersion wavelength can be tailored to coincide with the lowest loss
region around 1.5μm. This then offers an available low dispersion bandwidth of
around 200nm, as shown in Fig.1.3.6. The peak around 1.4μm is due to an overtone of
the fundamental water absorption occurring at 2.73μm, and can be reduced by careful
exclusion of water in the manufacturing process. The dotted curve represents the
fundamental intrinsic loss represesented through Rayleigh scattering by random
density fluctuations fused into the silica. This loss is inversely proportional to λ4, and
is the dominant loss mechanism around 1.5μm, where all other absorption losses are
lowest. The loss at 1.5μm can be reduced to less than 0.2dB km-1, which
25
represents a 1/e absorption length of:
1 = 4.34 (km)
α α dB (1.3.15)
where αdB is the loss in dB km-1. From 1500-1700nm, the loss in silica optical fibres
option, however, is that presented by the influence of nonlinearity when high power
pulses are used as signals. These effects are discussed in the next section.
In the previous section, only propagation in the linear intensity regime was
considered. Waveguides such as silica optical fibres however, do not generally
respond in a linear manner to a non-zero guided intensity. A change in power at the
input may cause a smaller change in the output accompanied by a qualitatitive change
in the output such as a wavelength shift. Silica has probably the lowest coefficient of
nonlinearity of any transparent material, however the high intensity within a fibre,
coupled with the long confinement length causes an accumulation of the nonlinear
effect, and can result in observable nonlinearity at only moderately high powers. A
beam focussed to a waist of radius ω0 in a bulk media achieves a maximum intensity
I=P/πω02, where P is the focussed power. The confocal parameter28 Z0=πω02/λ
approximately defines the focussed interaction length over which the beam area
doubles from that at the waist. Since nonlinear interactions scale with power and
interaction length, it is useful to quantify the power length product for focussing in a
bulk medium as:
26
πω20 P
IZ0 = P =
πω20 λ λ (1.4.1)
loss length is typically α-1=20km at 1.55μm, the enhancement in the power length
product becomes:
IZ0 (fibre)
≈ πωP
2
1 λ= λ
α P πω20 α
IZ0 (bulk) 0 (1.4.2)
~300x106. Despite the low nonlinear coefficient, this enhancement can result in
significant nonlinearity occurring over hundreds of metres of fibre with only a few
milliwatts of peak signal within the fibre. This nonlinearity can have drastic effects on
ultrashort pulse propagation in fibres, and is the major cause of pulse broadening in
linear (normally dispersive) transmission systems. The fundamental nonlinearity to be
included in the evolution equation describing the pulse envelope in a weakly nonlinear
waveguide can be expanded in terms of the electric field E within the pulse. The
induced nonlinear polarisation within the dielectric waveguide PNL can be described
by:
where ε0 is the vacuum permittivity and χ(j) (j=1,2,3...) is the jth order susceptibility.
The dominant term in Eq(1.4.3) is the linear term χ(1), from which the
absorption coefficient α and the refractive index n can be deduced. χ(2) is responsible
for nonlinear effects such as Second Harmonic Generation (SHG), and is zero in
centrosymmetric materials such as silica. Only if the symmetry of the host is broken
27
can χ(2) become finite and SHG be experienced. The lowest order nonlinear effects
observed in silica optical fibres generally arise from the χ(3) contribution, responsible
for third harmonic generation (which requires phase matching and is generally not
observed efficiently), four wave mixing and other nonlinear four wave interactions.
Only those processes which can be phase matched are observed efficiently, and most
of the latter nonlinear interactions are normally phase matched through a nonlinear
dependence of the waveguide refractive index on intensity, called the nonlinear Kerr
effect.
PNL= ε0 χ 1 + χ 3 E 2 E (1.4.4)
n = n0 +Δn (1.4.5)
which contains a linear part, and a nonlinear part which is dependent on the local field
intensity through:
χ 3 n0
Δn = E 2
2 (1.4.6)
Δn = n2 I (1.4.7)
28
where I is the field intensity expressed in W m-2.
The coefficient n2 has a value 3.2x10-20 m2W-1 for silica and is called the Kerr
coefficient. This is approximately two orders of magnitude smaller than that quoted
for CS2, but the enhancement obtained in long lengths of fibre overcomes this. If two
waves copropagate, the refractive index change experienced by one, n1, is perturbed
by the intensity of the other, I2, and vice-versa, with a resultant change dependent on
Δ n1 = n2 I1 + 2I2 (1.4.8)
Δ∅ = Δn k L
= n2 I k L (1.4.9)
for a single wave, where k=2π/λ is the wavenumber and L is the fibre length. This
intensity dependent phase shift results in a time dependent phase shift across a pulse
envelope, and is called Self Phase Modulation (SPM) if only one pulse is involved, or
Cross Phase Modulation (XPM) if two pulses with non-overlapping spectra are
involved. A phase shift across a wave results in a frequency shift determined by the
time derivative of the phase given by:
dω t = - d d∅ t
dt
dI t
= - n2 k L
dt 1.4.10
This frequency shift is sensitive to the wave envelope intensity, and results in a
29
local change in frequency across a pulse. SPM was first observed in fibres having a
hollow core filled with CS231. Only moderate power levels are required in silica based
fibres32,33, as the fibre can be kilometres long. Based on this length dependent
nonlinearity, it is useful to define a nonlinear length LNL(c.f. dispersion length LD in
Eq(1.3.8)) over which the induced phase change becomes comparable with unity:
LNL = 1 = 1
γP0 n2 kI (1.4.11)
where P0 is the peak power of the signal and γ is related to the fibre parameters as
follows:
2πn2
γ=
λAeff (1.4.12)
where Aeff is the mode field area (sometimes called the effective core area). If the
effect of SPM is included in the evolution equation for the pulse envelope propagation
(1.3.10), the result is called the Nonlinear Schroedinger Equation (NLS):
ŽU β 2 Ž2 U
i - + γ U 2U = 0
ŽZ 2 ŽT2
(1.4.13)
ŽΨ Ž2 Ψ
i + + VΨ = 0
Žt Žx2 (1.4.14)
30
balance can be struck, a stationary envelope function can be generated, and this
situation is considered later.
ŽU
i + γ U 2U = 0
ŽZ (1.4.15)
The envelope solution of this equation does not undergo any change in shape,
and for an initial pulse envelope function U(Z=0,T)=U0(T) the output pulse can be
shown to be25:
Fig.1.4.1. Temporal dependence of the local phase change dφ, and frequency shift dω
through SPM for a Gaussian input pulse envelope.
Since n2kZI=γPZ, the effect scales with length and peak power, and has a
characteristic lengthscale LNL. In Fig.1.4.1. the normalisation γP=1 is made and the
propagation length is LNL. The leading edge of the pulse undergoes a frequency
31
downshift (or red shift) and the trailing edge an upshift (or blue shift). Over the central
region of the pulse this leads to an approximately linear frequency 'chirp' on the pulse,
with the largest shift obtained at the inflexion points of the envelope profile. This
chirp is very sensitive to pulse shape, but is qualitatively similar for all "bell" shaped
pulses. The maximum frequency shift Δωmax can be obtained from the maximum phase
shift ΔΦmax from Eq(1.4.9) to obtain:
Δωmax = f Δ
∅max
T (1.4.17)
where f=0.86 for a Gaussian pulse shape34. For an unchirped (transform limited) input
pulse the total output bandwidth is the sum of the initial, plus the change in
bandwidth. It can be shown 34 that the total bandwidth due to SPM is:
A theoretical power spectrum35 calculated for a 10W peak power pulse of 30ps
duration at a 1μm wavelength launched into 500m of fibre with a 100μm2 mode field
area is shown in Fig.1.4.2.
Fig.1.4.2. Theoretical power spectrum through SPM of a 30ps, 10W pulse in a 500m
fibre35.
32
It is clear from Fig.1.4.1. that each particular frequency shift occurs for two
temporal regions within the pulse envelope. Interference in the power spectrum can
occur resulting in the characteristic spectral modulation shown in Fig.1.4.2. If
dispersion is now included in Eq(1.4.13), then over lengths where LD and LNL become
comparable, dispersion operates on the SPM broadened spectrum. In the β2>0 (normal
Fig.1.4.3. Experimentally obtained spectrum of SPM using 100ps, 90W signal pulses
in a ~2.2km length of fibre. Note spectral sidelobes are due to optical wave breaking
(see text).
33
Temporal broadening through dispersion of a SPM spectrally broadened pulse
can be several orders of magnitude greater than that caused by dispersion alone37.
However the approximately linear chirp which initially only covers about 1/3 of the
pulse through SPM alone can be extended to cover almost all of the pulse through
dispersion. This is accompanied by a "squaring off" of the pulse shape, making the
broad pulse more suitable for high efficiency compression using a negatively
dispersive delay line such as a grating pair38,39. In fibre grating compression, negative
dispersion from a grating pair is used to reverse the effect of normal (β2>0) dispersion
in the fibre. Operating on a pulse whose spectrum has been broadened substantially
through SPM in a fibre, the pulse can be made to compress as the new frequency
components undergo different flight times in the delay line and eventually overlap
each other temporally. The phase delay function of a pair of diffraction gratings
arranged with parallel faces is given by:
∅ω = ∅0 - aω2
(1.4.19)
where a is a constant of the grating compressor related to the grating separation. Since
the frequency chirp is approximately linear across the whole pulse when significant
dispersion accompanies nonlinear SPM in the fibre, the phase must be approximately
quadratic. A quadratic compressor with a phase delay similar to that of Eq(1.4.19) is
capable of almost completely compensating for this chirp, and highly efficient
compression is possible. The output pulse duration is of the order of that given by the
Fourier transform limit of the SPM spectral bandwidth, however, higher order terms in
the chirp also require dispersion compensation , and hence perfect transform limited
pulses are difficult to generate.
34
up to 1.5kW of output peak power. This source is very useful for soliton studies, and a
full characterisation of the technique can be found in the references42,42a.
Fig.1.4.4. Scematic of the fibre grating compressor used as a signal source in this
thesis42.
Since β2 is negative above the zero dispersion wavelength λ0, the fibre itself
ŽU β 2 Ž2 U
i - + γ U 2U= 0
ŽZ 2 ŽT2
(1.4.20)
35
In order to reduce the units of the equation to those of the Schroedinger
equation, it is useful to impose the following normalisation:
A= U , ξ= Z , τ= T
P0 LD T0 (1.4.21)
where P0 is the peak power in the fibre, LD is the previously defined dispersion length
and T0 is the duration of the input pulse. This yields:
ŽA sgn β2 Ž2 A
i = - N2 A 2 A
Žξ 2 Žτ 2
(1.4.22)
where sgn(β2) is the sign of β2, i.e -1 for anomalous dispersion. The parameter N is
given by:
N2 = LD
LNL (1.4.23)
By defining :
1
LD γτ20 2
u = NA = A= A
LNL β2 (1.4.24)
equation (1.4.22) can now be simplified to yield the well known standard NLS:
2
1 Ž u + i Žu + u 2 u = 0
2 Žτ2 Žξ
(1.4.25)
For anomalous dispersion, this equation can be solved using the inverse
scattering formalism, to yield solitary solutions for input pulses of hyperbolic secant
shape. The simplest solution is the fundamental soliton, whilst further solutions exist
which are integer multiples in amplitude of the fundamental. The fundamental soliton
36
has a solitary or stationary envelope function which is retained during propagation,
whilst higher order solitons exist which are periodic with propagation distance. The
fundamental soliton solution is given by:
iξ
u(ξ, τ) = sech(τ) exp
2 (1.4.26)
The importance of this function for pulse propagation lies in the independence
of the pulse envelope sech(t) on propagation distance parameter ξ. Hence if a
hyperbolic secant shaped pulse of peak power sufficient to make N=1 in Eq(1.4.23) is
propagated in a lossless fibre, the pulse will propagate without change in shape over
arbitrarily large distances. It is also important to note that the soliton phase is
independent of local pulse intensity, and thefore no frequency chirp exists (the
fundamental soliton is perfectly transform limited). The phase rotates at a linear rate
equally across the entire pulse. This makes solitons perfect for switching based on
interferometers, as described in chapter 7.
For optical communications, the soliton is ideal, since provided the power is
maintained, dispersion does not dominate transmission. From Eq(1.4.23) the peak
power required to excite the fundamental soliton is given by:
LD
=1
LNL (1.4.27)
β2 3.11 β 2
P1 = =
γτ20 γτ2FWHM (1.4.28)
At this peak power level, the L.H.S of Eq(1.4.22) is zero, indicating that the
pulse envelope is stationary. In measurable units,
37
3
3.11 Dλ Aeff
P1 =
4π 2 c n2 τ2FWHM (1.4.29)
where D is the GVD in units of ps nm-1km-1, and the full width half maximum pulse
duration τFWHM =1.763T0, where T0 is the half width (at 1/e fold intensity) of the
pulse. For power levels below P1, the dispersion in the fibre dominates over the
nonlinearity, and broadening always occurs, albeit at a lesser rate if the power is
comparable with P1.
To excite a higher order soliton, the power has to exceed P1 by a factor N2,
PN = N 2 P1 (1.4.30)
When a higher order soliton is excited, the nonlinearity exceeds that required
to balance the dispersion and maintain the pulse shape, and the pulse shape undergoes
compressions and rarefactions at a periodic interval related to the rate of phase
rotation in Eq (1.4.26). The period in units of fibre length is given by5
Z0 = 0.322π cτ = 0.322π τ
2 2 2
2 2
λ D 2 β2
(1.4.32)
If Aeff=80μm2, a fibre with ⎥D⎥= 10ps nm-1km-1 at 1.4μm would require 18W
of peak power to excite a fundamental soliton of 10ps duration. The soliton period for
this 10ps pulse would be 48.6m of fibre, increasing to 4.86km for an input pulse of
100ps duration with adequate peak power. The soliton period is related to the
dispersion length LD by:
Z0 = π LD
2 (1.4.32)
The numerical evolution of the envelope of an n=1,2 and 3 soliton are shown
in Fig.1.4.5., for a propagation length normalised to the soliton period Z0.
38
Fig 1.4.5. Theoretical evolution of the fundamental (N=1) and high order (N=2,3)
solitons with normalised propagation distance Z/Z0 .Note the change in intensity scale
between solutions.
39
CW signals under modulated gain, amplified modulational instabilities (self-pulsing
instabilities), and amplified sub-fundamental soliton power pulses all evolving into
soliton pulses. This is of major importance for regenerative amplification operating
periodically in an optical communication system, as the regenerated pulses may not
be exact soliton solutions. The fact that noise can develop into pulses is of great
importance, and requires strict control.
So far loss has been neglected. If the fibre length is comparable with the loss
length α-1, which is typically 20km at the lowest loss wavelength around 1.5μm, the
pulse energy will experience significant attenuation through absorption, and loss must
be considered. Loss can be treated as a small perturbation provide it is small over the
evolution length of the soliton defined by the soliton period. This limit is said to be
adiabatic provided the perturbation length is longer than the soliton period, and hence
loss can be incorporated into the NLS in terms of the normalised amplitude in
Eq(1.4.25) to yield:
Žu Ž2 u
i +1 + u 2 u = -iΓu
Žξ 2 Žτ 2
(1.4.33)
where Γ=αLD/2. The term on the R.H.S. is effectively an inverse driving term in the
NLS, and can be treated as a weak perturbation44 to yield a solution for an input pulse
of the form u(0,τ)=sech(τ). The solution undergoes an increase in pulsewidth in the
presence of loss, governed by the ratio of output to input pulsewidth:
40
τ1 = e 2Γξ = e αZ
τ0 (1.4.34)
This indicates that as the propagation length increases, the amplitude of the
soliton will decrease with a subsequent increase in duration in the presence of loss.
Both N=1 and N=2 soliton propagation have been modelled in the presence of loss45,46,
with the conclusion that dispersive broadening occuring on a soliton is less than that
experienced by a non-soliton pulse. Many schemes have been proposed for amplifying
solitons to overcome absorption loss and maintain the signal duration47,48,49.
Stimulated Raman Scattering (SRS) has been demonstrated as a distributed amplifier
in a long distance transmission experiment over many thousands of kilometres49a,50,51.
Erbium doped fibre amplifiers have also been employed in transmission experiments
over 10,000km52.
41
pulses in either single-pass or oscillator arrangements54, although the pulses do have
a significant temporal jitter. Details of the SRS process and Raman amplification are
given in the next section.
All of the nonlinear processes discussed in this thesis are governed by the third
order nonlinear susceptibility χ(3), and involve four wave interactions. Those in the last
section left the dielectric medium unexcited, and are elastic. Those in this section
involve an energy exchange which is assumed to be very small compared with the
incident energy, and are inelastic. Both SRS and SBS were discovered early in
fibres55,56,57, and are analogous to the phenomena seen in bulk materials, except for the
exclusion of certain scatttering phenomena which cannot be phase matched in an
approximately two dimensional dielectric waveguide. SRS is caused by the interaction
of the incident field with the optical phonon modes of the medium, and SBS involves
42
the acoustic phonon modes which are much lower frequency, resulting in much
smaller frequency shifts. For both processes, phase matching is automatic for the
stimulated process which occurs through χ(3) at high intensities, since the momentum
deficit is provided by the scattering phonon involved. Although similar in origin, the
dispersion relations applying to SRS and SBS are different, resulting in SBS only
being observed in the reverse direction and only for the narrowest linewidth pump
sources58. Hence SBS is generally not observed in the forward direction in a fibre or
from pulsed sources where the pulse bandwidth is larger than the Brillouin linewidth.
For this reason SBS is not covered here, as the bandwidth of the sources used for
soliton studies (much less than 1ns pulsewidths) always resulted in suppression of the
effect.
dIs
= g Ip Is
dZ (1.5.1)
where Ip and Is are the pump and Stokes field intensities respectively, and g is the
Raman gain coefficient. This gain is present in both directions relative to the pump.
The Raman gain coefficient, g, is shown as a function of frequency detuning from the
43
pump in Fig.1.5.1. for silica based optical fibres55. The gain coefficient is maximum
for a shift of approximately 440cm-1, and scales inversely with excitation wavelength
(1.32μm for this graph).
Fig.1.5.1. Raman gain coefficient, g, vs frequency shift in silica fibres. Vertical scale
shows magnitude for excitation at 1.32μm.
The lineshape is very broad due to the inhomogenous crystal field of the glassy
fibre, resulting in a spread of the molecular vibrational frequencies into a continuum.
The inclusion of codopants in the fibre core can modify the Raman gain curve, but
dopants used in manufacture have no significant qualitative effect.
As seen in Eq(1.5.1), weak signals within the Stokes band which copropagate
with the pump experience stimulated amplification. This means the process can be
used as a broad-band amplifier , and to produce tunable sources in either single pass or
Raman oscillator arrangements. The amplification is given by:
gI0Z
Is Z = Is 0 e
(1.5.2)
where Is(0) is the Stokes intensity at Z=0, and Z is the interaction length. In practice, if
44
pump (if the pump is pulsed as is often the case), and that it exceeds the Raman noise
intensity, which will otherwise dominate the process. The FWHM bandwidth of the
Raman gain curve shown in Fig.1.5.1 is in excess of 8THz, and can hence amplify
pulses as short as 100fs in the low loss wavelength region. Once stimulated parametric
Raman scattering occurs in a fibre, a large optical phonon density makes anti-Stokes
scattering possible, from molecules vibrationally excited as a consequence of Stokes
SRS. It evolves according to the following equation:
-gI0Z
Ias Z = Ias 0 e
(1.5.3)
and by analogy with Eq(1.5.2) the anti-Stokes signal intensity Ias(0) is absorbed
exponentially into the pump intensity. Effectively the anti-Stokes becomes a new
pump and the old pump becomes the new Stokes band in this interaction. It then
follows that with sufficient pump intensity the process can cascade to produce
multiple orders of Stokes bands.
Although the SRS process does not have a fixed threshold, a pump power
threshold has been defined57 as that power Pth at which the Stokes band becomes equal
kAeff
P th ≈
gL (1.5.4)
where Aeff is the mode field area of the fibre, and k is a fibre constant in the region 16-
20. Pump depletion is ignored in this estimate, and so is fibre loss. The value of k is
~16 in the forward and ~20 in the reverse direction to the pump field, and hence
forward Stokes grows first unless the backward Stokes is pre-seeded at the Stokes
wavelength.
approximately 0.5W Continuous Wave (CW) for a fibre of ~20km, which is the loss
45
length. For pulsed excitation, this peak power is easily obtained and the process is
highly efficent, but for ultrashort pulses dispersion quickly separates the Stokes signal
from the pump through 'walk-off', and the Stokes receives no more gain. The
interaction length Z in Eq(1.5.2) needs to be replaced by the walk-off length ΔL given
by:
ΔL = Δ t
DΔλ (1.5.4)
As a result of the high efficiency of the SRS process, a variety of schemes have
been devised to employ it as the gain medium in an oscillator arrangement61. By using
the relatively large bandwidth available to amplify ultrashort pulses, in conjunction
with a negatively dispersive delay line to overcome the effect of normal dispersion on
the Stokes signal from the fibre below λ0, short pulses have been generated directly
If the Stokes field wavelength falls in the anomalous dispersion regime, then
negative disperision can be provided by the fibre itself above λ0. In a variety of
46
The primary importance of soliton Raman generation is that the Raman gain provides
adiabatic amplification of the Stokes pulse, which may consist of noise spikes or MI
peaks. This process is the reverse of that described in Eq(1.4.34), where loss induced
soliton pulse broadening. If the energy of a soliton is increased by amplification, the
duration decreases adiabatically in order to maintain the fundamental soliton
condition. Higher order solitons cannot normally be formed through Raman
amplification, as the energy cannot be injected into the soliton quickly enough for
non-adiabatic compression to occur except for extremely intense pump fields. During
Raman amplification, the soliton pulse normally sees only a small increase in energy
over a gain length relative to the soliton period. This adiabatic form of amplification is
dicussed further in chapter 6, where it is utilised to generate solitons from a variety of
non-soliton signals.
The adiabatic scenario discussed so far operates in the relatively small gain
regime, with αgZ0<0.05, where αg is the gain length for the amplification process and
Z0 is the soliton period. In this thesis results are presented above this limit, where the
pulse energy undergoes dramatic changes in energy over a relatively short distance. In
chapter 6, sub-fundamental soliton power pulses are amplified through large
synchronous Raman gain to reconstruct solitons from dispersive pulses.
47
continuum. Experimental work to quantify the effect73 has confirmed theory74 which
predicts a shift in mean wavelength of a soliton of duration τ proportional to τ−4 or the
soliton peak power squared. As the pulse shifts to longer wavelength, the fibre
dispersion parameter increases, and the pulse broadens in order to preserve its soliton
nature. The effect is most noticeable when the input pulse spectrum exceeds ~1THz,
either through multisoliton compression which occurs as a high order soliton evolves
or through simple ultrashort soliton propagation. For this reason, solitons with τ>10ps
are normally used in soliton transmission systems, to avoid any SSFS. The SSFS
process can be used in conjunction with spectral selection to produce a source of
femtosecond solitons75, tunable by contol of pulse duration or peak power beyond
1.55μm.
1.6 References.
48
(10) A.Hasegawa & Y.Kodama. Proc.IEEE. 69,1145(1981) U U
(25) K.J.Blow & N.J.Doran. Lecture notes. NATO Advanced Summer Institute
on 'Nonlinear Waves in Solid State Physics", Erice,Sicily July (1989)
(26) A.W.Snyder & J.D.Love. "Optical Waveguide Theory".
Chapman & Hall (1983)
(27) A.Hasegawa. "Optical Solitons in Fibres". 2nd Ed. Springer Verlag 1990
(28) A.Yariv. "Optical Electronics". Holt-Saunders (1985)
(28a) Bloom et al. Opt.Lett. 4,279(1979)U U
49
(36) W.J. Tomlinson, R.H.Stolen & A.M.Johnson. Opt.Comm.54,377(1985) U U
50
Opt.Lett. 12,625(1987)
U U
51
52
2. FIBRE RAMAN LASER SOURCES
For a pulsed pump, the forward Stokes wave generally never depletes the
pump completely, due to walk-off. However a backward Stokes signal can see
continuous gain as it walks through the pump , either pulsed or Continuous Wave
(CW). By careful choice of fibre parameters, the group velocities of the pump and
Stokes signals can be matched in the forward direction, either through group velocity
matching around the dispersion minimum wavelength, or by propagation in different
transverse modes.In this way complete pump depletion can result in theory by
allowing the Stokes signal to walk through the pump completely. The actual limit
depends on higher order Stokes generation, and is a complicated problem. Highly
efficient conversion of pump light into the Stokes band is possible, with almost
complete pump conversion in an oscillator arrangement. It is hard to obtain narrow
53
linewidth operation of a Raman oscillator due to the inhomogoneous nature of the gain
for small signals, making them ideal for pulsed amplification. However in the large
signal limit, if pump depletion occurs, the gain becomes saturated and operation at one
laser wavelength can influence the gain at another, resulting in more of a
homogenously broadened system. Thus if the gain is saturated, the whole homogenous
linewidth is available for gain.
If mirrors are coupled to the ends of a fibre the Raman amplifier can form an
oscillator, tunable by the insertion of appropriate tuning elements within the cavity.
Only enough feedback is required to maintain operation above the single pass Raman
threshold, which in a low loss region of the fibre would enable typically 90% output
coupling of the Raman signal. Multiple Stokes orders have been made to oscillate
simultaneously through the use of multiple mirrors4.
54
2.2 A CW fibre Raman Laser at 1.41μm
kAeff
P th ≈
gL (2.2.1)
For non-polarisation preserving fibre used here, k=30 in the forward direction.
The Raman gain coefficient g=0.74x10-13mW-1 for a pump wavelength of 1.32μm. Aeff
is the fibre mode field area (typically 100μm2), L is the gain length neglecting loss and
walk-off, which did not occur as the pump in this case was CW. However at the
Stokes wavelength, which was around 1.4μm, the loss in the fibre was approximately
1.5dBkm-1, being enhanced through OH- absorption as shown in Fig.1.3.6. This loss
represented a loss length of αL~3km. The fibre used here was 9km in length,and hence
L in eq(2.2.1) should be replaced by an effective length of 3km. The fibre had a mode
field area of ~54μm2, and hence the calculated Pth for single pass Raman gain was
7.3W.
Only 1W of CW average power was available, and hence the high threshold
pump power was not attainable in a single pass configuration, although CW single
pass generation and oscillation has been reported elsewhere at other wavelengths5,6. In
order to reduce the threshold value, an oscillator was arranged so that the single pass
spontaneous Raman provided a seed for the double pass gain, reducing the threshold
accordingly. Fig.2.2.1. shows the experimental arrangement of the oscillator.
55
Fig.2.2.1. Experimental arrangement of the CW Raman oscillator.
preserving silica fibre (F) had an 8.3μm core diamter and a Δn of 0.0045. The output
was collected and collimated via (L2), an identical lens to (L1), and directed onto
grating (G) via mirror (M) which was highly reflective around 1.4μm. The grating (G)
was of a holographic type, with 1200 lines/mm, and was blazed for ~70% diffraction
efficiency at 1.4μm for 75o angle of incidence into the -1 diffractive order. The
diffracted signal was then directed of beansplitter (BS1) back into the input via (L1) to
form a ring cavity. Polarisation strainer (P) ensured that the Raman signal emerging
from the fibre was predominantly P-polarised, enabling the maximum gain to be
extracted from the P-polarised pump, whilst minimising the loss from the diffraction
grating. The output from the oscillator was monitored by inserting a ~2% reflecting
pellicle at the fibre output end.
The single pass output spectrum from the fibre for the maximum launched
pump power of 1.2W is shown in Fig.2.2.2(a)., with no detectable Raman signal
around 1.4μm because of the high loss in the system (~13dB total). The spectrum does
however show an interesting signal continuum around the 1.33-1.38μm wavelength
region, the origin and nature of which is unknown. It does however represent an extra
loss to the Raman process at this stage.
56
The total loss in the ring (including that from coupling, fibre absorption which
is high at 1.4μm, and diffraction) was estimated at 18dB, which is a X63 power loss.
The Raman threshold can be calculated from Eq(1.5.2) to give a Raman gain of X63 in
a single pass of the fibre
ln 63 Aeff
Pth = = 505mW
2gL (2.2.2)
57
one of 8% reflectivity, resulting in a 15% increase of pump threshold from 500mW to
~600mW.
58
When a Raman oscillator is pumped by intense laser pulses, the Raman shifted
pulses experiences a number of effects which make them suitable for compression
using a negatively dispersive delay line. The Raman pulses are generated at the peak
of the pump pulse, where the intensity and hence Raman gain is highest, within a short
propagation length for intense excitation. The Raman shifted signal propagates faster
than the pump pulse through normal dispersion, and hence walks through the front of
the pump pulse, severely depleting it. The Raman pulses generated are of roughly the
same duration as the pump, and can suffer Self Phase Modulation (SPM) as it
propagates and Cross Phase Modulation (XPM) from the pump. The combined effects
of SPM, XPM from the Raman depleted pump, and considerable normal dispersive
broadening result in a complicated frequency chirp across the Raman pulses, a portion
of which is approximately linear. By arranging the fibre in a synchronously pumped
ring geometry, the reinjected Raman signal can receive further amplification from a
pump pulse injected in synchronism, and oscillation can occur. Since the Raman pulse
is chirped, adjustment of the cavity length synchronises different frequency
components of the Raman pulse with the pump pulse, and the oscillator can be
tuned7,7a.
59
the 1.41μm Raman shifted wavelength of a 1.32μm pump is discussed. The pulses
obtained were ~2 ps duration with peak powers of ~100W. These pulses were
approximately transform limited, and hence perfectly suitable for soliton propagation
studies (see chapter 6),.in contrast to those generated from single pass soliton Raman
generators or soliton Raman fibre ring resonators8,9,10,11,12. The experimental schematic
of the ring laser is shown in Fig.2.3.1.
that the pump at 1.32μm and Raman shifted signal at ~1.4μm both experienced normal
GVD (β2>0). A CW mode-locked Nd:YAG laser operating at 1.32μm generating
100ps pulses at a 100MHz repetition rate with an average power of 2W (~200W peak
power) provided the pump radiation. This radiation was directed toward the input end
of the fibre via beam splitter (BS1) (100%R at 1.32μm , 95%T at 1.4μm for 45o). The
pump radiation was focussed into the fibre using an achromatic long working distance
X20 broadband Anti Reflection Coated Objective (ARCO) lens (L1). The Dispersion-
Shifted Fibre (DSF) was 400m long, with a core diameter of 7μm and was single mode
at both the pump and Stokes wavelengths.
The fibre loss was less than 0.6dBkm-1 in the 1.2-1.6μm spectral region, with
60
an OH- absoprtion peak at ~1.39μm where the loss rose to ~2dB/km. The DSF was
tailored to have a dispersion zero wavelength λ0=1.46μm, to ensure both pump and
Stokes were in the positive GVD (β2>0) regime. Fig.2.3.2. shows both the measured
Fig.2.3.2. Measured group delay and GVD vs. wavelength of the DSF. (Courtesy
BTRL)
As shown, the average value of the GVD between the pump and Stokes
wavelengths was ~6.4ps/nm/km. The Stokes shift in this fibre was ~80nm around this
pump wavelength, and hence the relative difference in group delay for these two
wavelengths was approximately 80 X 6.4 = 512ps/km of fibre. This is in good
agreement with the measured 520ps/km group delay difference shown in the second
trace in Fig.2.3.2. The time difference between the pump and Stokes over the 400m
fibre length was therefore ~210ps. This means the Stokes walked forward by ~210ps
through the front of the 100ps pump pulse, in 400m of fibre. The fibre was
approximately 2 walk-off lengths, which was ideal since the first Stokes becomes
maximal after 1.5-2 walk-off lengths.
Approximately 40% pump power coupling efficiency was achieved at the fibre
61
input, and the output radiation was collected and collimated with ARCO (L2) identical
to (L1). The output radiation was directed via aluminium mirror (M1) through a
negatively dispersive delay line, comprising a pair of holographic gratings (G1) and
(G2) arranged single pass with their faces parallel. The gratings had 1200 lines/mm,
and exhibited 75% diffraction efficiency into the -1 diffractive order at 70o angle of
incidence. Mirror (M2) directed the diffracted radiation from the dispersive delay line
filter centred at 1.4μm was inserted in the cavity to restrict the oscillating bandwidth
of the laser and prevent satellite pulses building up from noise in the cavity. The
output coupler (BS2) had a reflectivity of 20% at 1.4μm.
The fibre coupling block containing lens (L2) was mounted on a translation
stage to allow fine adjustment of the oscillator cavity length, with micron precision.
This allowed synchronisation of the selected Raman signal with a pulse pump from the
pump laser, by maximising the output signal for any particular wavelength governed
by the position of the aperture and the angle of the dielectric filter, which allowed a
small degree of tunability. Since the gratings were most efficient for p-polarised light,
as was the Raman conversion for the p-polarised pump, a fibre polarisation strainer
was inserted at the output end of the fibre.
The output from the oscillator was continually monitored using a vibrating
mirror real time SHG autocorrelator (see appendix), and a 1m scanning spectrograph,
allowing optimisation and characterisation of the sytstem. Whilst gradually increasing
the pump power, the amount of SPM and XPM experienced by the Raman pulses
circulating in the ring varied. To compensate for this variation the distance between
the gratings was adjusted. Through reoptimisation of the cavity length, the shortest
pulses with highest peak powers could be obtained by monitoring the autocorrelation.
Thus the optimum grating separation was found to be 18cm.
The oscillator was essentially unidirectional, as the synchronous pumping only
62
occurred in one direction, and hence the pump threshold for single pass Raman
generation Pth was given by
30Aeff
P th ≈
gL
for non-polarisation preserving fibre, where the mode field area Aeff~70μm2, and the
Raman gain coefficient g=0.74X10-13 mW-1 at 1.32μm. For an interaction length equal
to the walk-off length, i.e. 190m, this yielded a pump power of 82W peak,
corresponding to an average pump power of 820mW. By maintaining the pump power
at 400mW, no single pass Raman generation was detectable around 1.4μm. To
estimate the threshold for oscillation in a ring configuration, an estimate of the loss
was required. However the loss is dependent on factors such as the oscillator
bandwidth, which is highly dependent on pump power through SPM and XPM, and
hence could not be estimated.
Once optimised for the shortest pulses and highest power, the system routinely
generated 2.7ps pulses with up to 20mW of average output power available,
corresponding to 74W peak in the pulses at 100MHz repetition rate determined by the
pump source. The output bandwidth was ~2nm, being limited directly by the
intracavity dielectric tuning filter. For grating separations +/- 10 to 15% of the
optimum 18cm, the pulses became substantially longer. The system could be coarsely
tuned from 1390 to 1415nm by adjusting the horizontal position of the aperture across
the dispersed spectrum emerging from the grating pair, and finely tuned by carefully
tilting the intracavity dielectric filter, which altered the 2nm bandpass centre
wavelength over this range. Each time tuning took place, the cavity length required
reoptimisation by adjustment of the (L2) translation stage. A typical output spectrum
63
Fig.2.3.3. Typical output spectrum of the fibre raman ring oscillator.
Fig.2.3.4. Background free autocorrelation traces of the output from the dispersion
compensated fibre Raman ring laser. Oscillator cavity length (a) optimised and (b)
mismatched by +/-10μm.
64
In Fig.2.3.4(a). the cavity length was optimised for the best possible output,
resulting in an output pulse duration of 2.7ps, assuming a Gaussian pulse shape. For a
cavity length detuning of +10μm, the laser generated pulses with temporal
characteristics attributed to the autocorrelation function of a noise burst (see
appendix), as shown in Fig.2.3.4(b). This suggests that the oscillator is producing
bursts of noise of ~20-50ps duration, with temporal features as short as a few
picoseconds. Similar behaviour was noted for detunings of -10μm. When operated
close to threshold pump power, it was possible by carefully adjusting the polarisation
strainer in the fibre ring to generate pulses as short as 1.8ps, yielding an optimum time
bandwidth product of 0.54 which is close to transform limited operation, however
more stable operation was obtainable when the pump power was higher, where the
shortest pulses measured were 2.7ps. These pulses, assuming a Gaussian pulse shape,
were approximately twice the transform limited duration for the measured 2nm
oscillating bandwidth.
In conclusion, a dispersion compensated fibre Raman ring laser has been mode
locked through synchronous pumping, producing stable 2.7ps pulses with an average
output power of 20mW (74W peak) at a 100MHz repetition rate for a pump power of
400mW. The source was tunable over the range 1390-1415nm, and the output is close
to transform limited. The source was very useful for soliton studies, as the output
wavelength falls in the anomalous dispersion regime of standard optical fibres. The
output wavelength makes the pulses suitable for studying the effects of synchronous
65
Raman amplification, using pump radiation from the same or a similar Nd:YAG laser
operating at 1.32μm. The synchronously pumped oscillator cavity length could also be
misadjusted, to produce noise bursts of similar duration to the pump source,
containing picosecond ultrashort features, which are useful for noise amplification
studies. The source was used extensively to obtain results presented in chapter 6, from
a variety of experiments involving synchronous amplification and soliton generation.
2.4 References
(3) from "Fibre Raman Lasers". C.Lin. In "Tunable Lasers". Ed. L.F.Mollenauer &
J.White. Springer-Verlag Topics in Appl. Physics 59 U
66
3. MODE LOCKED DOPED FIBRE LASER SOURCES
3.1 Introduction
Rare earth doped optical fibres have a broad fluorescence linwidth, due to the
homogeneous nature of the glass host crystal field. A high gain can be obtained over
relatively short fibre lengths, provided pumping is efficient, making doped fibre
amplifiers also ideal for incorporation in fibre lasers. The gain can be so high that laser
action is obtainable using only the cleaved fibre facets as reflectors, although this
could lead to problems through amplification of spontaneous emission in long range
communications. Fibre amplifiers of this kind are directly compatible with optical
communications in the linear or nonlinear regime, with the possibility of injecting
pump radiation periodically using fused couplers.
Maximum transmission speeds in fibre based transmission still fall far short of
those determined by the low loss bandwidth, even in the very best soliton transmission
systems, and to increase capacity multiplexing in both time and frequency may be
necessary. In a number of application such as multiplexing/demultiplexing, stable
sources of pulses at ultra-high (100GHz up) repetition rates are required with high
distributive fanout. These sources will effectively be used to derive signal pulses for
switching and clock pulses throughout the system and for synchronisation. A soliton
based system would benefit from a source generating soliton-like pulses, to prevent
the propagation of intrapulse radiation shed in the soliton formation process, which
may result in long range distortion through soliton - soliton interactions. Erbium
doped fibre has a fluorescence band around the region of lowest loss at 1.55μm, and
hence much interest has been dedicated to the development of an Erbium fibre laser
producing ultrasort pulses at ultra-high repetition rates.
Semiconductor diode lasers are ideal sources for use in high bit rate
transmission, since they demonstrate high efficiency and controllability, but they yield
67
highly chirped pulses at even modest gain switching frequencies and low output
powers. The powers available from doped fibre lasers are much higher, and active
mode locking using bulk or 'pigtailed' integrated modulators offers controllability over
output, but repetition rates are limited to the electro-optic speeds. The marriage of
high speed laser diodes as pulse sources with amplification in doped fibre offers an
alternative way of reaching the high repetition rate, high power and the direct
controllability goal. Fibre sources such as Neodymium and Erbium are capable of
producing more than 0.5W of average output power, and since soliton transmission is
curently limited to pulse durations of around 10ps to avoid Soliton Self-Frequency
Shift (SSFS), this would represent peak powers well into the nonlinear regime, making
fibre lasers more than capable of exciting solitons. Fibre amplifiers have optical
bandwidths of around 1THz (e.g. Nd, Er) and are therefore capable of supporting
pulses of this duration.
Since optical fibres can be tailored so that the dispersion minimum wavelength
occurs beyond 1.5μm, a compact, Erbium based laser producing soliton pulses in the
low GVD regime is quite feasible. The process of Modulational Instability (MI) (see
chapter 5) occurring in the anomalous dispersion regime offers the oportunity to
generate pulses in the THz regime. This effect could also be incorporated into an
Erbium fibre laser to produce a source which is perhaps completely passive. MI can
grow from amplitude and phase noise fluctuations, or can be induced by amplitude
modulation from XPM or frequency mixing, and is therefore controllable. The
ultimate modulation rate is limited by high order dispersion terms which dominate
around λo.
In this chapter, three mode locked fibre lasers are presented offering the
possibility of high output power and high repetition rates. The Nd based fibre laser
uses a novel mode locking technique based on interferometrically induced nonlinear
coupling between two cavities. An Er doped fibre laser is presented based on
traditional amplitude modulation mode locking, and has been used in an experiment
68
presented in chapter 5, where modulational instability was induced from the output
pulses. A further system using both an Er doped fibre and a laser diode as a modulator
allows the benefits of both systems to be simultaneously exploited for the first time.
Amplitude Modulation.
The region of lowest loss in optical fibres occurs in the 1.5-1.6μm wavelength
region, as shown in Fig.1.3.6. Losses of less than 0.2dB/km are obtainable, allowing
transmission over 10-20km with only half power loss. Much interest is shown in the
application of Erbium doped fibre amplifiers in extending the range of transmission
systems based on optical fibres. Very high gain is available1 (up to 46dB pumped with
a 1480μm high power laser diode), and a very broad gain bandwidth2 around the
fluorescence band at 1.55μm.
69
described in chapter 5, and hence Er doped fibre was an ideal gain medium. Erbium
doped fibres can produce high average powers6, whilst they provide a bandwidth2
capable of amplifying pulses as short as 100fs7. In this section the construction of a
mode locked Erbium fibre laser is described, using active mode locking by amplitude
modulation.
The first reported mode locking of an Er fibre laser8 used a bulk electro-optic
phase modulator, producing 70ps pulses, but with only ~100μW of output power. This
represented only around 10mW of peak power, which would result in a nonlinear
length given in eq(1.4.12) longer than 40km. These pulses are useless for nonlinear
propagation studies, since the fibre loss length is much shorter than this value. Shorter
pulses have been generated9 by incorporating soliton shaping in a fused coupled, fibre
ring oscillator, but the output power was still as low. To facilitate higher power output
in the system decribed here, a fibre which demonstrated high CW average output
power was incorporated into a low loss linear cavity arrangement, pumped with a
small frame 2W Argon Ion laser operating at 514nm. Despite the pump wavelength,
excited state absorption of the pump signal did not dominate10, and high output power
was available (~100mw) The experimental arrangement is shown in Fig.3.2.1.
The 514nm pump radiation was directed through beamsplitter (BS) (95%T at
70
514nm, 95%R around 1.55μm) and into a X20 anti reflection coated objective
(ARCO) lens for high transmission in the 1.5μm region. The pump beam was focussed
through a cavity mirror placed in contact with the normally cleaved Er fibre end face,
and into the Er fibre. This 'butt coupling ' technique provided laser output coupling at
the fibre end over the coating range of the mirror (90%T at 514nm, 20%T at 1.56μm).
Of the 2W of pump power available, up to 800mW could be launched into the Er
doped fibre, which was ~3.2m long with an outer diameter of 130μm, a core diameter
of ~15μm. The fibre was highly multimode at the pump wavelength, and the large core
diameter made it weakly multimode at the laser wavelength as well. At 514nm the
absorption was such that 90% of the estimated launched pump radiation was absorbed
over the fibre length. In order to eliminate self laser action off the other fibre end face,
and to suppress sub cavity effects which may destroy mode locking, a quartz wedge
polished at 7o was placed in optical contact to the fibre end with index matching gel.
Radiation exiting the fibre end and passing through the wedge was collected and
collimated with a X10 ARCO lens, coated for ultra low loss over the 1.55μm region.
The beam emerging from the lens was passed through the amplitude modulator, which
was connected to a stabilised source of RF power.
71
polarisation strainer was incorporated at the fibre end nearest the grating. The aperture
helped restrict the oscillation bandwidth8 by spatially restricting the bandwidth of the
beam retrodiffracted from the grating, and ensured that high order transverse modes
did not oscillate. Stable mode locked oscillation required critical adjustment of the
polarisation strainer, and optimisation of the cavity length.
Initially the laser output from the 20% transmission end was analysed with a
photodiode and sampling oscilloscope providing a combined temporal resolution of
~320ps, as shown in Fig.3.2.2. An isolator was placed in the laser output path to
ensure that reflections from detector apparatus did not destroy the mode locking,
which was extremely sensitive to perturbations. The cavity length was adjusted to give
a round trip of approximately 40ns, to enable 4th harmonic mode locking of the
fundamental modulation frequency, which was ~100MHz.
Fig.3.2.2. Sampling oscilloscope trace of the mode locked Er fibre laser output
72
hybrid transverse mode oscillation, but the mode locked output was very irregular and
exhibited multiple pulsing.
Fig.3.2.3(a). shows a streak camera trace of the mode locked output, split by an
optical delay of 387ps. The measured FWHM pulse duration was 140ps, which would
require a mode locked bandwidth of 0.02nm assuming a transform limited sech2 pulse.
The CW spectrum obtained with the modulator switched off is shown in Fig.3.2.3(b),
when the laser was tuned as in (a) to 1.565μm. The measured bandwidth corresponded
approximately to the resolution limit of the spectrograph, which was ~0.07nm. With
the modulator switched on again, the mode locked bandwidth deconvolved from the
resolution limit was 0.2nm, as shown in Fig.3.2.3(c). This indicated the laser was
operating approximately 18X the transform limit, probably generating chirped pulses.
Semi-interferometric autocorreleation measurements (see appendix) taken later in
chapter 5 confirmed that the pulses were chirped, with fringe visibility over less than
10% of the pulse profile.
Fig.3.2.3.(a) Streak camera trace, (b) CW and (c) mode locked spectrum of fibre laser
output.
73
Tuning over the range 1530 to 1580nm was possible, without noticable pulse
duration change over the 1555 to 1575nm range of high reflection (peaking at 80%)
for the narrow band dielectric cavity mirror. Outside this range, mode locking was
accompanied by severe relaxation oscillations, indicating that loss in the cavity was
becoming significant. Mode locking could not be obtained without this cavity end
mirror as losses were assumed to be too high, and laser performance may have been
improved by using a broad band cavity end mirror.
For an average output power of 60mW average, the 140ps pulses at 100MHz
repetition rate corresponded to 4.3W of peak power in each pulse. This peak power
was high enough to induce weak nonlinearity within the laser cavity, however the
nonlinear length NNL corresponding to this intacavity power was a few hundred
metres, and the soliton period Z0 for this situation would be in excess of 100km.
74
solitons directly from an Erbium fibre laser.
Currently, very high average output powers can be obtained from Erbium
doped amplifiers and lasers. The experiment presented in the last section described a
laser producing 60mW of average power, and up to 500mW has been obtained from an
Er doped fibre amplifier at 1550nm13. In fact the output from fibre laser systems is
now comparable with many small frame bulk solid state lasers.
75
combined with the high average power capability of the Er doped fibre medium. Diode
lasers provide a cheap and simple alternative to bulk or pigtailed modulators used
previously to mode lock Er fibre lasers9,16,17, and require much lower RF drive signals.
Semiconductor lasers can be modulated by direct modulation of the pumping current
at rates exceeding 10GHz, making them extremely important components for high bit-
rate optical communication. They can be monolithically integrated with other
optoelectronic components to form hybrid processing circuits. The output beam has
compatible dimensions to those of typical silica based fibres, and the output
wavelength can be tailored to that of the lowest loss and dispersion.
A very short electrical pulse (<100ps) drive signal can readily be applied to a
laser diode device, allowing the device to operate as a very fast loss modulator, with
open times of approximately the same duration as the typical laser output pulse
durations, which can be as short as 15ps for modulation rates approaching 30GHz. In
this section the application of a GaInAsP ridge waveguide laser diode chip as a
modulator in an Erbium doped fibre laser is described, resulting in mode locking of
the fibre laser.
76
power for the Er fibre laser was once again provided by a small frame Argon Ion laser
producing 2W of CW radiation at 514nm. The pump radiation was directed through
beamsplitter (BS) (95%T at 514nm, 95%R around 1.55μm) and through dielectric
cavity mirror (M) (90%T at 541nm, 50%R around 1.55μm) into a X20 anti reflection
coated objective (ARCO) lens for high transmission in the 1.5μm region.
In order to suppress sub cavity effects and eliminate laser action which readily
occurred from reflection at the fibre ends, quartz discs with a 7o wedge were contacted
to the fibre ends with index matching gel. In this way laser action was effectively
suppressed off the fibre facets. Mirror (M) was mounted on a translation stage to allow
coarse tuning of the cavity length. Radiation emerging from the Er fibre was collected,
collimated and refocussed with two identical ARCO lenses (MO2 &MO3) into the
laser diode chip (LD). The lens coatings were for ultra low loss in the 1.55μm region.
77
Alignment was initially facilitated by amplifying radiation from the laser diode
when driven CW above threshold, in the Er fibre single pass. When driven CW from a
D.C. current supply at approximateley 100mA, the laser diode alone generated
average output powers of up to 30mW. This device, which was used as a modulator
was a GaInAsP ridge waveguide device, 280μm long, with one uncoated facet parallel
to the other, which was ~100% reflection coated around the lasing wavelength. This
highly reflective facet formed one cavity end mirror of the Er fibre laser, in a linear
configuration. Pulses to drive the semiconductor modulator were derived from an
amplified RF frequency synthesizer (RF Osc). This signal was used to drive a step
recovery diode comb generator (Hewlett-Packard 33004A) at ~500MHz, with an
average RF power of 600mW. The derived pulse train consisting of electrical pulses
with ~90ps duration was coupled into the semiconductor modulator via a bias-tee
network in conjunction with the current from a variable DC supply.
The output from the Er fibre laser collected from beamsplitter (BS) was passed
through an optical isolator, to ensure that reflections from detection apparatus did not
disrupt the operation of the laser. The output signal was then analysed with a
photodiode/sampling oscilloscope with a combined temporal resolution of 320ps to
initially obtain stable mode locking. The overall fibre laser cavity defined by dielectric
mirror (M) and the highly reflective facet of the laser diode chip was initially arranged
to have a length seventeen times that of the fundamental, determined approximately by
the 500MHz RF drive frequency supplied to the diode. With no DC bias current to the
diode, laser action in the Er fibre was suppressed, through absorption in the unpumped
laser diode chip. By increasing the DC bias current up to 150mA, laser action was
allowed to occur, dominated by self laser action of the diode at 1550nm. A DC current
threshold for laser action in this configuration of 45mA was measured.
78
readily detected with the fast photodiode. For a fixed cavity length, the RF drive
frequency was adjusted in order to obtain the minimum pulsewidth from the Er fibre
laser. At a drive frequency of 500.475MHz, routine operation with 40ps pulse
generation was achieved. Fig.3.3.2. shows a typical output pulse of the 37ps duration
pulses recorded under these conditions with a Hamamatsu OOS-1/IR optical sampling
oscilloscope.
Fig.3.3.2. Optical sampling oscilloscope trace of 37ps pulses from the Er fibre laser
mode locked with an intra-cavity laser diode as a modulator.
around the same wavelength, i.e.~1.53μm. This is evidence that the Er fibre gain is
providing mode locked pulses, rather than the output simply consisting of amplified
laser diode pulses, which would give a spectrum centred around 1.55μm. Single pass
amplification of the laser diode pulses in a similar configuration provided ~50ps
pulses centred around 1.55μm. A transform limited sech2 pulse of ~40ps duration at
79
1.531μm requires a mode locked bandwidth of 0.06nm, indicating that operation was
above the transform limit. This was not suprising considering the negatively chirped
nature of pulses produced from a laser diode of this kind, when modulated at high
frequency. The optical sampling oscilloscope trace may have been broader than that of
the pulse duration, if any jitter occurred between the output pulses and the trigger
signal, which was derived directly from the RF oscillator.
The measured output power of the laser was ~20mW, which corresponded to a
peak power of ~1W in the 40ps pulses at 500MHz repetition rate. The fundamental
soliton power for a 40ps pulse in a fibre similar to the Er fibre used here was ~1mW
peak, but with a corresponding soliton period Z0>20km. The nonlinear length scale
LNL>500m, thus no soliton or nonlinear pulse shaping mechanisms would have been
strong enough to affect the laser, unless the pulses evolved over hundreds of cavity
transits.
It is possible that the 280μm diode laser device which was effectively placed
intra-cavity in the Er fibre laser was operating as a bandwidth limiting element, since
80
the output facet reflectivity was as high as ~30% on the uncoated side. This possibly
stabilised the laser output as well, which tended to be spectrally unstable if driven CW
without the diode intracavity. For an etalon corresponding to the geometrical length of
the laser diode, the free spectral range Δν is given by
Δν = c
2nL
where n is the semiconductor refractive index which is ~3.5. From this equation,
Δν=153GHz for the device used here, which corresponded to a free spectral range of
approximately 1.2nm at 1.53μm.
The finesse F of a similar etalon, defined as the ratio of the free spectral range
to the FWHM transmission bandwidth is given by
F = π R1 R2
1- R1 R2
where R1 & R2 are the reflectivities of the etalon mirrors. For facet reflectivities of
30% and 100%, typical of those involved here, the finesse F was approximately 2.46,
which would have yielded a transmissive bandwidth of approximately 0.5nm. Hence it
is reasonable to suggest that this element had a bandwidth limiting effect on the
operation of the laser, and would limit the shortest pulse obtainable to 5ps duration, if
dispersion and chirp could have been compensated for in some way, and operation
was transform limited.
81
imposed by the laser diode carrier density fluctuation, by the use of a normally
dispersive fibre or a dispersive delay line.
In conclusion active mode locking of an Erbium doped fibre laser has been
demonstrated using a semiconductor laser diode as a modulator. The technique allows
routine generation of 40ps pulses at a 500MHz repetition rate, limited only by the
electronic driving circuitry. The Er fibre laser has generated average powers of up to
20mW, indicating that this relatively inexpensive method of modulation has the
potential for producing substantially higher peak powers if more efficient pump
schemes are employed, or shorter pulses are generated. This laser could be completely
made up of optoelectronic devices if a semiconductor laser pump was used, and may
find application to the study of high bit rate nonlinear optical communication systems.
The repetition rate of the system could theoretically be extended to the highest
frequency that the semiconductor device could cope with.
82
through this process18,19, though only over limited duration pulse trains. Up to 100
GHz modulations have been generated in a CW train20, but Brillouin scattering
hampered the process, limiting the modulation depth and significantly attenuating the
signal, which would have required further amplification to make it useful.
Much interest has recently been shown in passive mode locking techniques
employing external nonlinear cavities. The soliton laser21 was the first demonstration
of how an optical fibre coupled to a laser to form an external cavity could dramatically
influence the laser performance, resulting in a considerable reduction of the output
pulse duration. Subsequent research determined that soliton shaping in the external
fibre cavity was not a prerequisite for the process, only SPM induced spectral
broadening21 - 30. The process appears to occur through interferometric addition of
main cavity pulses with spectrally broadened pulses from the external nonlinear
cavity, and can result in pulse shortening through peak enhancement and pedestal
suppression if the cavities are interferometrically stable over a large number of
transits.The technique has been succesfully applied to a variety of systems from colour
centre lasers24 - 27 and Ti:Al2O3 lasers28,29, to Nd:YAG and Nd:Glass lasers30 - 32. One
In this section a technique for mode locking a Nd doped fibre laser is presented
which is novel in that it requires no direct modulation at the cavity round trip time, yet
can yield quasi continuous pulse trains of ~50ms duration with repetition rates
determined only by the relative cavity lengths. The technique was first reported in a
CW Ti:Al2O3 laser33 generating pulses as short as 6ps in trains of ~20ms duration,
with pulse repetition rates up to 1GHz. Mode locking is observed when one of the
mirrors in either the laser or the linear external cavity is moved with a velocity of the
order ~0.05m s-1. The term 'linear' refers to the lack of any nonlinear medium in the
external cavity, and as such the mechanism involved which results in pulse formation
has not yet fully been identified, although it is surmised that key elements include:
83
(a) sweeping the length mismatch between the two cavities (and therefore their
resonance) at a rate appropriate for Q-switching the laser,
(b) some nonlinearity in the laser amplifier (probably the optical Kerr effect)
providing pulse compression in a manner similar to Additive Pulse
Modelocking (APM)22,24.
A Nd3+ doped single mode fibre (0.5% Nd3+ by weight) of 37cm length
provided the gain medium, with the resonator formed by butt coupling two dielectric
coated mirrors to the end facets of the fibre. Mirror (M1) was 100μm thick, coated for
100% reflection around 1.06μm, and 95% transmission at the pump wavelength which
was 514nm. Mirror (M2) formed the laser output coupler and was ~96% reflective
the available pump radiation was coupled into the fibre, which was highly multimode
at the pump wavelength.
The output emerging from the fibre laser through mirror (M2) was collected
and collimated with a X20 broadband ARCO lens, and coupled into the external cavity
via beamsplitter (BS) (97%R around 1.06μm). The external cavity was formed
84
between mirror (M2), beamsplitter (BS) and highly reflective mirror (M3), which was
mounted on an optical shaker unit (Bruel & Kjaer type 4810), which was of a similar
type to that found in a scanning autocorrelator.
The fibre laser had a pump power threshold of ~200mW, and exhibited 99%
pump absorption. With a pump power of approximately 1.25W coupled into the fibre,
an average output power of ~500mW was obtained at 1.06μm. In order to mode lock
the laser, the length of the external cavity (M2)-(BS)-(M3) was approximately matched
to that of the main fibre cavity (M1)-(M2), accounting for the refractive index of the
fibre. Mirror (M3) was then oscillated along the (BS)-(M3) axis, at peak velocities
between 0.03 and 0.18ms-1 to obain pulses whilst the mirror was in motion. The output
pulse train for an oscillation of +/-1.5mm peak to peak displacement and an oscillation
frequency of 38Hz is shown in Fig.3.4.2.
The signal in (a) was obtained with a fast diode and oscilloscope, and shows a
pulse train of ~13ms period, which corresponds to the oscillation period of the shaker.
The pulses were also monitored using a fast diode and sampling oscilloscope with an
overall temporal resolution of ~150ps. The shortest pulses measured were of 300ps
duration (not deconvolved with the 150ps combined temporal resolution) at a
repetition rate of 267MHz, as shown in (b). This repetition rate corresponded to the
inverse cavity round trip time for the Nd fibre cavity, assuming a refractive index of
~1.5. The deconvolved duration of the pulses was 260ps, substantially shorter than
85
those obtained in early experiments using acousto-optic modulators to mode lock Nd
fibre lasers.
similar technique35 investigated how the pulse duration varied across the generated
pulse trains. The shortest pulses obtained were sampled using an electro-optic shutter
from the centre of the train, where the mirror moved fastest, and yielded pulse
durations approximately half of that integrated over the whole train. It seems
reasonable to asume that the same applies here, and hence pulses as short as 130ps
may be generated at the centre of the train. A study was made of the performance of
the laser as a function of the mismatch between the laser and external cavity lengths.
While mirror (M3) oscillated about an average position, the length of the external
cavity was varied by moving the mirror along its oscillatory axis on a translation
stage. Fig.3.4.3 shows how the measured pulse duration (deconvolved from the
measured 150ps photodiode-oscilloscope resolution) and pulse modulation depth
varied with cavity mismatch length. The modulation depth was defined as the peak to
background ratio of the mode locked pulse train.
Note that mode locking was observed over a much larger range of cavity
mismatch than that reported for lasers employing APM. As the external cavity length
86
deviated from optimum, the pulse length broadened and the modulation depth
decreased as the pedestal grew. This was a similar effect to that seen in Ti:Al2O3,
however no change in modulation depth was observed for that system. The
performance of the mode locked laser was relatively insensitive to small changes in
either amplitude or frequency of mirror oscillation, or to pump power.
The fibre laser operated on a number of spectral modes between 1.05μm and
1.07μm, both free running and when mode locked by the influence of the external
cavity. The relative intensity of these modes was extremely sensitive to cavity
alignment and fibre strain, whilst the temporal output remained consistently stable
over the resolvable timescales in comparison. Spectra obtained sequentially showed
large changes in the distribution of spectral energy between modes, and hence
conveyed little information as to the mode locking mechanism or laser performance.
To investigate the effect of stronger coupling between the cavities, mirror (M2)
was removed, increasing the output coupling from 4% (transmission) to 96% (residual
facet reflectivity ~4%) . Owing to the stronger coupling, the laser exhibited large
relaxation oscillations which could not be suppressed, even by reducing the
reflectivity of the beamsplitter (BS) or mirror (M3) to reduce the coupling to the
previous value. This suggested that a low loss (high Q) main laser cavity and perhaps
weak coupling between the cavities may have been important for effective mode
locking.
The velocity of mirror (M3) did not appear to be sufficient to directly mode
lock the laser by modulating the effective reflectivity of the external cavity, as the
interference effect would have induced a modulation of the cavity loss in the 100KHz
range. This timescale did, however seem appropriate for Q-switching the laser. The
velocity was also many orders of magnitude less than that required to lock the
longitudinal modes together through Doppler shifting, even if the effect built up over
hundreds of transits. By changing the output coupler of the main cavity (M2) to a 20%
87
transmissivity mirror, thereby increasing the coupling from what appeared stable, Q-
switched and mode locked operation could be observed, as shown in Fig.3.4.4(a) and
(b) on two different timescales.
The Q-switched envelope duration was ~600ns, with the occurrence of further
relaxation oscillations at a period of ~10ms, which corresponds approximately to the
modulation frequency caused by interference between the two cavities at ~100KHz.
APM may have been responsible for the effects shown in this experiment. The
nonlinearity could have been generated in the fibre laser cavity instead of in the
external cavity, with peak power enhancement caused by massive relaxation
oscillations facilitating SPM. The nonlinear length LNL was less than 5m of fibre for
the intra-cavity mode locked peak power in this experiment, and since the laser had a
relatively high Q factor, it is feasible that some nonlinear broadening occured over a
number of transits. Relaxation oscillations could have been induced through
modulation of the reflectivity of the external cavity at frequencies comparable with the
upper state lifetime of the gain medium. This would explain why mirror oscillation
would only induce mode locking if the mirror velocity was within a given range, with
the exact amplitude or frequency set arbitrarily, since a particular velocity would give
rise to a specific modulation frequency. The minimum and maximum velocities were
0.03 and 0.18 ms-1 respectively.
88
Another mechanism36 which may be responsible for the mode locking
observed here is the influence of the linear external cavity on the initial stages of pulse
evolution in a synchronously pumped system. Any external perturbation can strongly
influence the initial pulse shaping in such a laser, through an independent evolution of
the radiation in each cavity, coupled with a beating between the longitudinal modes of
both.
external cavity an integer fraction of the main cavity length, harmonic modelocking
was achieved, although the generated pulses were much longer and had significant
pedestals.
In conclusion, mode locking has been induced in a Nd doped fibre laser using a
novel technique where linear external feedback is fed back into the gain cavity with an
oscillating mirror. Similar work on Ti:Al2O3 suggests that the system may be scaled to
produce much higher repetition rates35. The technique may also work with other gain
media, and further work using Er doped fibre lasers may lead to the development of a
truly simple and compact source of high repetition rate pulses for use in transmission
systems operating in the nonlinear regime.
3.5 References
89
(2) C.G.Atkins, J.F.Masicott, J.R.Armitage, R.Wyatt, B.J.Ainslie &
S.P.Craig-Ryan. Elec.Lett. 25,910(1989) U U
90
(33) P.M.W.French, S.M.J.Kelly & J.R.Taylor. To be published in Opt.Lett.
(34) I.P.Alcock, A.I.Ferguson, D.C.Hanna & A.C.Tropper.
Elec.Lett. 22,269(1986)
U U
91
4. PULSE AMPLIFICATION IN DOPED FIBRES
Flourozirconate glass has been found to be preferable as a host for operation at this
wavelength9, but since the loss is lower in silica fibre around 1.55μm, Er doped fibre
which exhibits higher gain has dominated research. Existing fibre optical transmission
systems only use a small fraction of the available transparent bandwidth of the fibre
over long distances, since up until now electronic circuits involving a detector,
amplifier and source have been used to regenerate attenuated signals after 20-100km.
To overcome this and other problems associated with high speed electronic
multiplexing, all optical amplification, signal conditioning and information processing
is a necessary requirement. In this way the signal remains optical throughout the
transmission system. Therefore highly efficient monolithic optical amplifiers are
92
required. Raman amplification has been extensively studied as an option, but there are
shortfalls involved, such as the limited maximum obtainable gain per unit length and
the Soliton Self-Frequency Shift (SSFS) (see chapter 1).
Efficient pump confinement and good mode overlap with pump radiation in the
core results in very high available gain in optical fibres. The doping levels used here
resulted in almost complete pump absorption occurring over the few metre length of
doped fibre, although this is not necessary in Nd since its four level system has no
ground state absorption at the gain wavelength. The purpose of this experiment was to
establish the transient gain available in a typical fibre amplifier based on Nd doped
fibre. Neodymium in a glass host has a 3dB fluorescence bandwidth of more than
10nm around 1.06μm, and is hence capable of pulsed amplification of less than ~200fs
sech2 pulses. Here the pulses amplified were 4ps in duration, generated using the well
known fibre grating compression technique described in chapter 1. A schematic of the
experimental arrangement is show in Fig.4.2.1.
93
Fig.4.2.1. Schematic of the experimental arrangement for pulse amplification in Nd
doped fibre.
94
Fig 4.2.2(a) shows the spectal output of the compressor fibre, showing SPM
broadening to a bandwidth of 0.8nm. This was capable of supporting ~2ps pulses
assuming a Gaussian pulse shape. Fig.4.2.2(b) is the corresponding collinear
autocorrelation of the compressed pulses emerging from the double pass grating
arrangement, with a measured deconvolved pulse duration of ~4ps. This yielded for
200-250mW of average power a peak power of ~500W, and a corresponding energy of
~2nJ per pulse.
The output signal pulses from the compressor were coupled via a variable
neutral density wheel (ND filter) into a conventional GeO2-SiO2 fibre with an 8μm
core, doped with 300ppm Neodymium, with a medium phosphorous content (1-
2%mol% P2O5).This single mode fibre had a loss of ~2dBkm-1 at 1.06μm. At 514nm
95
average launched pump power of ~10mW at 514nm. The fluorecence spectrum around
1μm is shown in Fig.4.2.3. A gain cross section of Nd in glass at 1.06μm of σ=3X10-
20cm2 yields a saturation energy fluence (energy density) of:
∅sat = h ν = hc ≈ 3Jcm-1
2 σ 2σλ
where h is Plack's constant, and λ=1.06μm. Gain saturation results from transitions
between the upper and lower laser levels due to stimulated emmission. This lowers the
population inversion and hence the gain if enough stimulated emmission takes place.
Gain saturation is only significant if more energy than that stored in the gain medium
is extracted within a time interval comparable with the upper state lifetime, which can
be measured. The fluorescence lifetime of the 4F3/2-4I11/2 transition at 1.06μm in the Nd
doped fibre used here was measured by pumping a short length (~20cm) of identical
fibre with ~150ns pulses at a 2KHz repetition rate. These ~3kW pulses were derived
fom a Q-switched, frequency doubled Nd:YAG laser operating at 532nm. The
fluorescence at 1.06μm was separated from the pump band and the fluorescence signal
at 900nm using a monochromator. The signal was detected with a vacuum photodiode
with <5ns response time, and is shown in Fig.4.2.4.
96
the saturation fluence within the gain recovery time. The saturation fluence of 3Jcm-2
corresponded to an average saturation energy of 1.5μJ for the fibre used here, which
had a mode field area of ~50μm2.Since the fundamental transverse mode occupied by
the signal had a questionable overlap with that occupied by the pump, which occupied
a hybrid set of modes, the pump and signal alignment were continuously optimised
throughout the experiment, to maximise the gain experienced.
To attenuate the relatively large average signal power available from the
compressor a neutral density wheel (ND filter) was inserted into the signal path prior
to amplification. A conventional collinear autocorrelator with <100fs resolution and a
synchroscan streak camera was used to directly measure the gain of the amplifier, and
to monitor changes in the amplified pulse shape. Throughout the investigation no
change in the duration of the pulse transmitted through the amplifier was measured at
the power levels used, indicating that the gain bandwidth of the Nd doped fibre
exceeded that of the signal pulse, which was ~1nm. Fig.4.2.5. shows an
autocorrelation trace of the (a) unamplified signal pulse for an average launched signal
power of 4mW and (b) the resultant amplified signal for a launched pump power of
40mW at 514nm. Both traces were recorded on the same vertical scale.
Fig.4.2.5. Autocorrelation traces showing signal at 1.06μm (a) without pump, and (b)
with 40mW of counter-propagating pump radiation at 514nm.
A measure of the gain could be deduced from the ratio of the nonlinear
97
autocorrelation intensities, but was in fact measured more accurately directly using a
synchroscan streak camera. The gain was measured for a fixed maximum launched
signal power of 25mW (corresponding to ~250pJ per pulse) as a function of launched
pump power. The resultant curve is shown in Fig.4.2.6.
Fig.4.2.6.Variation of peak gain with pump power for a constant average signal
power of 25mW.
The maximum peak gain experienced for this signal power was ~1.2 for a
launched average pump power of between 15 and 40mW. Saturation of the gain
readily occurred, due to the high signal intensity injected. Higher gains could be
observed by reducing the signal level by an order of magnitude. To investigate the
gain as a function of signal intensity, the average pump power was maintained at a
maximum level of 40mW, and the signal level (measured by transmission in the
absence of any pump radiation) was adjusted using the ND filter. The resultant trend is
shown in Fig.4.2.7.
Fig.4.2.7. Variation of peak gain at 1.06μm with average signal power for 2.5m of Nd
doped fibre pumped with 40mW of average power at 514nm.
98
For signal pulses of ~4ps at 100MHz repetition rate, a saturated gain of X1.5
was measured for an average signal power of 2.5mW. For signal powers below
~0.2mW the gain increased rapidly from X2 to X10 for 15mW of launched signal
power (0.15pJ per pulse). The observed gain saturation effect is in reasonable
agreement with that predicted for Nd:Glass, since a 2.5mW average signal power
constitutes approximately 1.5μJ of energy in 600μs, which is of the same order as the
measured fluorescence lifetime. The measured gain was substantially lower than the
small signal gain for signals above the 100μW level, hence the saturation behaviour is
attributed to the high signal powers used.
In conclusion, up to X10 gain for 4ps pulses with energies of ~0.2pJ per pulse
(~20μW average) at 100MHz repetition rate has been observed in a Nd doped fibre
amplifier, for 40mW of pump power at 514nm. The importance of gain saturation and
the requirement of a high saturation fluence and long gain lifetime are paramount for
pulsed amplification. Operation at 1.06μm in Nd doped fibre is however of little
consequence to optical communications. Results more relevant to optical
communications are presented in the next section, which deals with Erbium
amplification operating around the 1.55μm region of lowest loss in silica fibres.
Fig.1.3.6. in chapter 1shows a plot of the measured absorption profile over the
region of highest transparency from 900-1700nm for a Dispersion-Shifted Fibre, with
λ =1.56μm. By careful control of the OH- impurity content during manufacture, the
0
absorption loss can realistically be reduced to less than 0.3dBkm-1 over lengths
greater than 20km. Absorption loss has a broadening effect on solitons, since as the
pulse energy and peak power decreases, the duration must increase to maintain the
fundamental soliton requirement (see chapter 1). High order solitons degenerate into
lower orders through loss a similar manner, unless break up through SSFS occurs first.
Soliton propagation was predicted early in the 1970's, but fibre technology had to
99
significantly advance for losses to be <1dBkm-1 before soliton effects could be
experimentally observed within an absorption length.
The full gain bandwidth (5dB) of the best Er doped fibre is in excess of
25nm18, and is therefore capable of amplifying pulses as short as 100fs (assuming
sech2 shape). In the experiment presented here, ~200 fs soliton pulses were amplified
in a single mode Er doped fibre for the first time.
100
Fig.4.3.2. Fluorescence spectrum of Er3+ doped fibre. The bandwidth of a 200fs
soliton is shown as a bar for reference.
The peak of the fluorescence appeared around 1.53μm, with a secondary peak
at 1.55μm. For reference, the bandwidth required to support or amplify a 200fs soliton
is shown as a bar for reference. Although the spectrum was measured with a low pump
power to ensure that laser action did not occur from the fibre ends, the exact
fluorescence lineshape is strongly dependent on dopant level, fibre length, overall
absorption an pump power19. Thus the gain bandwidth for a specific pump power
depends on the fibre parameters. The energy level splitting of the Er3+ ions in the
random glass host crystal field results in a very broad 1.55μm [4I15/2 - 4I13/2] transition.
Two fluorescence peaks normally exist around 1535 and 1552nm, with the
1552nm peak only dominant for high pump powers. The fluorescence spectrum shown
in Fig.4.3.2. is perhaps more representative of amplified spontaneous emission, and
probably shows less FWHM bandwidth than that available for amplification. For even
higher pump powers, the gain spectrum normally encompasses both peaks, resulting in
>20nm gain bandwidth18.
101
Fig.4.3.3. Schematic of the cascade soliton Raman ring laser.
The signal source use in this experiment was an optimised cascade Raman
soliton laser20, which was synchronously pumped by a mode locked Nd:YAG laser
operating at 1.32μm. The schematic of the soliton Raman laser is shown in Fig.4.3.3.
The fibre used in the ring laser was 600m long, with a dispersion zero wavelength
λ0=1.46μm, and hence was anomalously dispersive only at the second Stokes Raman
band wavelength around 1.5μm for the 1.32 μm pump. The principle behind Raman
ring oscillators was introduced in chapter 2, and in this case, the disperion
compensation arises through soliton shaping from anomalous dispersion in the fibre.
More details of the processes involved in Raman soliton generation19 are given in
chapter 6.
Fig 4.3.4. shows spectra of the output of the soliton Raman ring laser. The
spectrum in Fig 4.3.4(a) corresponded to single pass Raman generation, and indicates
only the first Stokes generation at around 1.41μm for the power levels being used.
(The narrow line around 1.34μm was attributed to Raman gain induced in a weak
second line from the Nd:YAG laser.). Fig 4.3.4(b) show the fibre output spectrum in a
resonator configuration, optimised for operation at 1.5μm, in the second Stokes
Raman band. Third Stokes generation was apparent, simultaneously occurring with the
102
second Stokes. By fixing the pump power at around 400mW average, the ring laser
provided in excess of 150mW average output power in a broad band around 1.5μm.
Within this broad band several single solitons existed, the number and temporal
separation of which were not particularly controllable, as a result of the soliton Raman
continuum evolution from noise.
Fig.4.3.4. Spectra from the cacade soliton Raman ring laser for an average pump
power of 500mW at 1.32μm, (a) without and (b) with feedback.
Intensity autocorrelation of the ring laser output was the integration of all of
the pulses, which were of an average duration of ~200fs. By spectrally selecting a
portion of the spectral continuum, near transform limited output was obtained. In this
case, spectral selection around 1.53μm derived a source of solitons with ~200fs
duration and ~80mW average power, at the fluorescence peak of the Erbium fibre (see
Fig.4.3.2.). This spectrally selected signal was used as a signal for Erbium
amplification.
103
background free (non-collinear) autocorrelation of the signal from the ring laser,
spectrally selected around 1.53μm, indicating an overall duration of 200fs. The
Erbium doped fibre was pumped in a counter-propagating pump geometry similar to
that used for Nd doped fibre in the last section, with 1W of CW Argon Ion pump
radiation at 514nm available. Up to 300mW of average pump power at 514nm was
coupled into the exit end of the fibre.
The pedestal component in the background free autocorrelation was less than
half a percent of the total intensity, indicating that in excess of half the average power
was contained in the soliton pulses. A corresponding transform limited 200fs soliton
pulse would have a spectral bandwidth of 12.3nm, as indicated in Fig.4.3.2. as a
reference bar. It is however important to note that the signal pulses used here have
bandwidths in excess of twice this value, as more than one soliton is present within an
envelope of spectrally separated signals. The complete spectrum, including all solitons
present, contributes to the autocorrelation envelope.
These signal solitons were simultaneously launched into the Er doped fibre,
and amplified on application of the counter-propagating pump at 514nm. Fig.4.3.6.
shows the autocorrelation trace of the pulses emerging from the Er doped fibre.
104
Fig.4.3.6. Autocorrelation trace of the (a) unamplified and (b) amplified soliton
pulses. Both vertical scales are the same, but zero levels differ for clarity.
For the trace in Fig.4.3.6(a), the pump power at 514nm was adjusted for
lossless propagation of the soliton signal, determined by measurement of the average
signal power at 1.53μm. This step was necessary because the three-level laser system
in Erbium would have resulted in complete signal absorption of the signal from the
ground state in the abscence of pump radiation over this fibre length. In Fig.4.3.6(b).
the launched pump power was increased to 250mW at 514nm. A clear increase in
signal intensity was evident with no increase whatsoever in the soliton pulsewidth
from 200fs.
To measure the degree of gain saturation occurring in the amplifier, the gain
was measured for various average signal powers, using a variable neutral density
wheel to attenuate the soliton signal power from the ring laser. The pump power was
fixed at 300mW in the 10m length of fibre during this process and the resultant trend
in the gain is shown in Fig. 4.3.7.
105
Fig.4.3.7. Gain vs average soliton signal power for a fixed pump power of 300mW.
For signal powers above ~5mW average, severe saturation of the gain was
evident, as shown in Fig.4.3.7., with the gain falling to almost unity for 50mW of
signal (corresponding to ~2.5kW launched peak power). For signals of about 1mW
average power, a gain of 2.6 was exhibited, corresponding to a net gain of ~4.2dB.
It was necessary throughout the experiment to operate with signal power levels
of the 1mW level, since the fundamental soliton power in the Er fibre was ~22.5W
peak (0.45mW average). Below this power level, significant dipersive broadening
dominated, resulting in broader measured pulse durations in the absence of gain. The
spectral bandwidth of the Raman soliton laser signal was as large as ~30nm (limited
only by the spectral filtering used beyond the ring laser). It is therefore possible that a
higher gain was present, since the measured bandwidth of the amplifier was less than a
half of that of the signal at the pump powers used here. If bandwidth limited gain was
operating, however, some temporal reshaping of the amplified pulses would be
expected, and no such effect was observed, within the ~50fs resolution acheivable. For
high signal levels, the low gain measured was a result of amplifier saturation, similar
to that recorded for picosecond amplification in Nd doped fibre in the last section.
A ~4.2dB gain has been observed in an Er doped fibre amplifier for 200fs
soliton pulses at a ~1mW average power level (~50Wpeak). Similar gains have also
been reported when operating the system with 120fs signal pulses. The low gains
achievable were a result of the relatively large bandwidth spread of the input signal,
106
and amplifier saturation at higher signal powers. No temporal dispersive broadening
was evident in the amplified soliton signal. More gain (up to ~20dB) has been
measured for smaller signal powers, however significant temporal broadening was
observed below an average power of ~0.5mw, which corresponds to the fundamental
soliton power in the Er fibre for a 200fs soliton pulse.
4.4 References
107
(16) M.Nakazawa, K.Suzuki & Y.Kimura. IEEE.Phot.Tech.Lett. 2,216(1990)
U U
108
5. MODULATIONAL INSTABILITY
ŽA β 2 Ž2 A
+ γ A A = - αA
2
i - 2
ŽZ 2 ŽT 2
(5.1.1)
where a loss term has been included from Eq(1.4.33) on the R.H.S. For the special
case where the loss coefficient α=0, this equation has been shown to support solitary
wave solutions of hyperbolic secant envelope pulses, whose shape does not change
with propagation provided the conditions set in Eq(1.4.27) are satisfied, i.e. that the
dispersion length equals the nonlinear length.
A(Z,T) = P0 ei γ P 0Z
(5.1.2)
109
state solution, but a more important effect becomes apparent if small perturbations are
considered. If a perurbation in amplitude a(Z,T) is imposed on A(Z,T) the propagation
can be evaluated by substitution into Eq(5.1.1) with α=0 to obtain:
Ža β 2 Ž2 a
i = +γ P 0 (a+a* )
ŽZ 2 ŽT2
(5.1.3)
2 1/2
k = ± 1 β 2 Ω Ω + sgn β 2 Ωc
2
2 (5.1.5)
2 4 γ P0 4
where Ωc = =
β2 β 2 LNL
(5.1.6.)
and sgn(β2)=+/-1, depending on the sign of the dispersion parameter β2. The nonlinear
LNL = 1
γP 0 (5.1.7)
110
unstable for anomalous GVD, and in fact results in an exponential growth of the
perturbation a(Z,T) with propagation length. The process is referred to as
Modulational Instability (MI), resulting for a sufficiently intense pump field at ω0 in:
(i) spontaneous modulation growth from the steady state for the autonomous CW
field in the presence of noise fluctuations, and
(ii) growth of any seed modulation imposed on a CW carrier at ωo, provided the
seed modulation frequency ωs<Ωc given in Eq(5.1.6).
Modulational instability has been found to occur in other systems such as fluid
dynamics and plasma physics, and is commonly referred to as the self-pulsing or
Benjamin-Feir instability2. The conditions necessary to observe MI are similar to those
required to observe soliton propagation in fibres. An interaction between anomalous
group velocity dispersion and SPM can provide exponential gain with length for any
amplitude or phase fluctuations induced on a CW or quasi-CW carrier.
2 2 1/2
g Ω = 2Im(k) = β 2 Ω Ωc - Ω
= 2 γ P0 = 2
LNL (5.1.9)
This implies that any amplitude and/or phase modulated signal becomes
unstable for modulation frequencies Ω<Ωc. Fig.5.1.1. is a graph of the spectral
dependence of the gain for three peak power levels (1W,2W & 4W) with parameters
β2=-20ps2km-1 and γ=2W-1km-1, which are appropriate for a standard silica fibre at
1.55μm.
111
Fig.5.1.1. Frequency dependence of MI gain for three peak power levels, with
β2=-20ps2km-1 and γ=2W-1km-1.
The gain reaches a peak for two equidistant sideband frequencies +/-Ωmax,
where
2 γ P0 1/2
Ωmax = ± Ωc = ±
2 β2 (5.1.10)
2
gmax = g Ωmax = 1 β 2 Ωc = 2γ P 0
2 (5.1.11)
112
It should be noted that the bandwidth of a typical 300fs sech2 pulse at 1.5μm is
approximately 8nm. To exhibit MI the modulation period would have to be of the
order of 30fs, which would require extremely high powers, and hence ultrashort pulses
do not normally exhibit MI at moderate peak powers. The process is best observed
with CW pump signals of high intensity. As the pump power is increased, the gain
peak moves further from the carrier frequency, and higher harmonic modulation peaks
can develop at integer multiples of the fundamental modulation frequency. The
process eventually saturates when the CW pump signal is sufficiently depleted to
reduce the peak power at the carrier frequency. Almost 100% pump depletion is
possible.
matching, which is only achieved within the MI gain sidebands. The phase matching is
provided through anomalous GVD operating on the SPM broadened signal. From the
FWM perspective, it is easier to explain how the process can also be induced by
injection of a probe signal within the gain sideband of a CW carrier. If a probe at
113
frequency ω1=ω0+Ω, copropagates with a CW carrier signal which is of insufficent
spontaneous MI3,4, however the frequency shift of the sideband is no longer power
dependent, and is fixed at the probe frequency. This process is referred to as FWM or
probe induced MI.
114
The gain from modulational instability can be used as an amplifier, since a
probe signal within the MI sideband of an intense carrier will exponentially grow at
the expense of the carrier. The sidebands produced through MI can also become
amplified through Raman scattering, and act as a precursor to the production of
Raman solitons, significantly reducing the threshold for the process. This is dicussed
in chapter 6. Since gain saturation through carrier depletion eventually slows down the
growth MI sidebands, the generated modulations can then undergo soliton shaping,
resulting in the generation of a train of solitary waves, which when amplified can grow
into many single solitons.
Since modulational instability can be induced through XPM, the pump field
which induces nonlinear broadening in the relatively weak signal need not propagate
in the anomalous GVD regime. Coupling through XPM is independent of the sign of
dispersion. In this way the peak power of an intense pump signal, regardless of its
dispersive nature, can provide the nonlinearity necessesary for a weak signal
copropagating in the anomalous GVD regime to undergo soliton shaping. Sub-
fundamental soliton signals can undergo nonlinear pulse shaping, provided they are in
the anomalous GVD regime, and that a strong pump is present, copropagating to
provide XPM. In this chapter a series of experiments are presented which investigate
MI, induced MI through XPM and FWM, and XPM induced pulse shaping in the
anomalous GVD regime.
115
despite significant Brillouin depletion of the relatively narrow band source. Such
pump sources are suitable due to a high output power, and a wavelength which is close
to the zero dispersion wavelength of standard fibres at ~1.3μm.
The extremely low absorption loss in optical fibres around 1.55μm has meant
that much interest has been placed in Erbium doped fibre amplification, as discussed
in chapter 4. The medium is directly compatible with soliton systems, with improved
performance over long distances and at higher bit rates. In order to generate ultra-high
repetition rate pulse trains around this wavelength, nonlinear effects such as MI offer a
means of generating pulses at up to 5THz6. By shifting the zero dispersion wavelength
λ0 of the fibre to around 1.55μm, it should be possible to reduce the anomalous GVD
Brillouin Scattering (SBS) presents a dominant loss mechanism for intense CW beams
whose bandwidths are much less than one Brillouin linewidth (~10GHz in silica). In
fact reflectivities of 50% are typical5. Pulsed systems with large bandwidths are
therefore preferred for generating MI.
116
The source of radiation used in this experiment was decribed in detail in
section 3.2. This Er fibre laser was mode locked with an acousto-optic modulator, and
produced pulses as short as 140ps at a repetition rate of 100MHz, with an average
power output of 60mW. The output was tunable from 1530 to 1580nm, with short
pulses generated over the range 1555 to 1575nm. Outside this range, the reflectivity of
the output coupler fell below ~ 15%, and the laser did not exhibit mode locking. The
experimental schematic is shown in Fig.5.2.1.
Essentially, the output pulses from the laser were focussed into 1.2km of
Dispersion-Shifted Fibre (DSF) via an isolator to avoid reflections from the fibre or
lenses destroying the laser mode locking. The laser was tuned by alignment of the
intra-cavity grating to the wavelength region of low anomalous dispersion in the fibre,
close to λ0~1.54μm. The DSF had a 5μm core diameter, a Δn of 0.017 and a loss of
less than 0.4dBKm-1 at 1.55μm. At 1.559μm, by optimisation of the cavity length, the
laser generated 150ps pulses at 100MHz repetition rate, as displayed on a synchroscan
streak camera with an S-1 photocathode. Up to 60mW of average output power was
available corresponding to around 4W peak power, of which up to 30mW average
could be coupled into the DSF with a X10 ARCO objective. The GVD at this
wavelength was estimated to be D=1.5ps nm-1km-1.
117
2 2 c n2 P
Ωc = 3
λ Aeff D (5.2.1)
where all symbols have their usual meaning. Given that λ=1.559μm, P=2W peak,
Aeff=33μm2, and D=1.5ps nm-1Km-1, Ωc was predicted to be 430GHz. Hence under
these conditions, MI gain was expected for frequencies below 430GHz, peaking at:
Ωc
Ωm = ≈ 300GHz
2 (5.2.2)
Temporal
evidence of this
modulation through
autocorrelation would
therefore exhibit a period
of 3.3ps. Since the pump
pulses used here were of
~150ps duration, they
appeared CW as far as
MI was concerned.
118
the dispersion length for the pump pulses was quite long.
The peak power of the signal emerging from the DSF was no more than ~2W ,
making it very difficult to observe any temporal modulation by autocorrelation.
Instability in the pump laser when generating pulses as short as 150ps caused severe
jitter in the spectral sideband separation, which tended to wash out the temporal
manifestation of the MI. To improve the signal to noise ratio of the system, slow scan
autocorrelation was used with a scan time of ~1ps per second of scan, with a total
capture time of ~30s. The resultant semi-interferometric trace obtained showed a
marked increase in contrast due to the enhancement of the fringes within the transform
limited portion of the input pulse. The pump pulses were significantly chirped, and
hence the fringes only occurred over the central portion of the autocorrelation trace.
Fig.5.2.3. shows the resultant modulated semi-interferometric autocorrelation trace of
the DSF output under the same conditions as those for the spectra of Fig.5.2.2.
A clear temporal modulation with a 3.6ps period was evident, consistent with
119
the inverse sideband separation in the spectrum of Fig.5.2.2. Note that the interference
fringes were not perfectly resolved, and had a spacing resulting from signal aliasing
with the sampling oscilloscope resolution at this scan rate. The measured 3.6ps
modulation period was within good agreement with that predicted at 3.3ps, with any
discrepancy probably due to either pump depletion, or under estimation of the mode
field area Aeff.
120
predicted, indicating modulation periods (frequencies) of approximately 11.5ps
(88GHz), 16.5ps (61GHz) and 27ps (37GHz) for launched powers of 15mW, 9mW
and 5mW respectively. These frequencies spanned the range 37-277GHz, and
therefore represented a source of modulations at a tunable repetition rate around
~1.56μm. If these modulations were amplified, they would have been expected to
develop into deep solitary waves and eventually form solitons. This proved that MI
could be induced from peak powers present within an Er doped fibre laser, providing a
useful ultra-high repetition rate source over this wavelength range, for use as clock,
signal or synchronisation pulses in an all optical communications sytem.
In conclusion, an actively mode locked Er fibre laser has been shown to induce
MI in pulses of a few hundred picoseconds, with characteristic modulation frequencies
in the range 37-277GHz determined by the launched pump peak power (or pulse
duration). The process could hopefully be induced intra-cavity in a fibre laser, to
produce deep modulation of the laser filed, or extra-cavity to produce modulations
which could be amplified further into solitons. Further research could produce a CW
pumped source of ultra-high repetition rate pulses if a high power Er fibre amplifier7
were used as the gain medium. Higher modulation frequencies should be obtainable
with higher pump powers or lower GVD, by tuning as close as possible to λ0, although
the latter procedure has been found to level off the modulation frequency to between 5
and 10THz as higher (4th) order dispersion dominates, instead of increasing to infinity
as Eq(5.2.1) predicts.
In section 1.1 Self Phase Modulation (SPM) was introduced, where the
intensity dependent refractive index present in a fibre results in a variation in the phase
across an optical pulse, and the development of a positive frequency chirp. If two
waves copropagate in an optical fibre, they can perturb the refractive index seen by
themselves and by the other wave in exactly the same way, resulting in Cross Phase
121
Modulation (XPM). Neglecting coupling through Stimulated Raman Scattering (SRS)
or Stimulated Brillouin Scattering (SBS) and coupling processes such as phase
matched Four Wave Mixing (FWM), the coupling through XPM can result in exactly
the same effect as SPM if the two pulsed waves copropagate perfectly. The resultant
phase modulation is a result of both SPM and XPM for each of the two waves. In
accordance with the treatment of SPM given in chapter 1, and neglecting absorption in
the fibre, the coupling can be shown to be related to an intensity dependent refractive
index change:
n = n0 + Δ n (5.3.1)
Δ n = n2 E1 2 + 2 E2 2
(5.3.2)
where the nonlinear refractive index coefficient n2=3.2X10- 20m2W-1. The first term in
this equation is similar to that given in Eq(1.4.7), and results in the familiar SPM of
the wave by itself. The second term is responsible for XPM, and is dependent on the
intensity of the copropagating field intensity. Note how the coupled term is twice as
effective8a. The effect is sometimes referred to as Induced Phase Modulation (IPM)
when the coupling occurrs between two signals derived from the same source wave,
e.g. XPM from a pump wave to a Stokes wave through SRS9,10.
122
complicated by pump - signal walk-off13,14, resulting in spectral asymmetry and even
cancellation of the local temporal phase shift. With walk-off, signal downchirp
induced by the leading edge of the pump pulse can be rapidly followed and cancelled
by the upchirp induced by the trailing edge. Similarly, no spectral broadening may
result from the interaction of a very short signal with a very long, intense pump if no
walk-off occurs. Since the intensity of the pump is slowly varying on the signal
timescale, little phase shift with time is experienced, and no chirp is induced. With
similar pump and probe pulse durations and no walk-off, the probe experiences a
phase modulation through XPM similar, but twice in magnitude to that experienced by
the pump pulse through SPM. The spectral effect is exactly similar to SPM, apart from
discrepancies which may arise from the experimental nature of the signals. For
example a CW signal would yield a large CW spectral component, which would
remain unmodulated between the copropagating pump pulses.
123
Fig.5.3.1. Phase shift (left ) and spectral broadening experienced by a signal pulse
for (a) perfect tracking,(b) initial coincidence followed by walk-off and (c) complete
pulse-signal walkthrough.
For the case where pump and signal pulses perfectly track with each other (a),
the resultant effect on the signal is indistinguishable from SPM. In (b) the two pulses
initially coincide in the fibre, and then the pump walks off through the signal either
forwards or backwards. The resultant phase shift is clearly influenced by the sign of
curvature of the pump pulse edge which walks off through the signal, resulting in a
strongly asymmetric spectral broadening. If the pump walks completely through a
signal pulse of finite and similar duration as shown in (c), the effect of the two
intensity edges produces no net phase change, on the timescale of the signal pulse. The
spectrum undergoes no net change, although some slight spectral shaping may occur
through incomplete compensation or interference during the interaction. In the work
that follows, the influence of an intense pump on a weaker signal through XPM is
investigated for the cases of perfect tracking and walk-off. XPM can provide a means
of spectrally broadening a low level signal incapable of inducing spectral broadening
in itself, in order to perform dispersive compression with gratings, or dispersion
compensation through soliton shaping in the fibre itself. This latter technique is
124
responsible for the results presented in chapter 5.5 where pulses are generated from a
CW laser diode signal.
In section 5.1, MI was introduced, noting that the effect can only
spontaneously arise in the anomalous dispersion regime. This is true for MI which is
phase matched through SPM, and grows through soliton shaping into solitary waves.
Since the intensity of a signal is of importance only to induce nonlinearity, there exists
the possibility of inducing MI in a signal through the influence of a strong
copropagating pump field, which spectrally broadens the signal through XPM. In this
section, sub-picosecond solitary waves are experimentally derived at 1.32μm (just
within the anomalous GVD regime for a standard fibre) through the process of XPM
induced MI3. The MI is induced through copropagating an intense pump pulse at
1.06μm (which is well within the normal GVD regime), of approximately the same
duration as the signal pulse.
The conditions necessary for MI to occur are similar to those required for
soliton propagation. The signal must have sufficient peak power in order to overcome
the tendency for the modulation to collapse through dispersive broadening. The
necessary peak power can be provided by a copropagating pump, injected
simultaneously with a weak signal, even if the signal is of insufficient intensity to
induce MI by itself. By varying the power and/or frequency of the external modulation
provided by the pump, solitary waves of controllable width and repetition rate could
be produced4. The coupling between pump and signal through XPM can be
detrimental and lead to cross- talk in multichannel communication systems11, however
here it allowed a signal in the normal dispersion regime to provide the high intensities
required in the interaction. MI has been shown to occur for signals propagating in the
normal dispersion regime, with the required phase matching provided solely through
125
SPM and XPM, or by propagation in different fibre modes15,16, however the effect was
closer to phase matched degenerate FWM than soliton shaping.
In this experiment, to ensure that the pump and signal propagate with similar
propagation constants17, the fibre was chosen so that the minimum dispersion
wavelength λ0 was between the pump and signal wavelengths. The group velocities
were made more similar in this way, although they still differed considerably and
hence walk-off still occurred. The pump pulses experienced SPM , and transferred a
chirp to the signal pulses of twice the magnitude via XPM, provided the two signals
did not walk-off. In the anomalous GVD regime, the chirped signal underwent soliton
compression, and developed MI from amplitude or phase fluctuations in the pump
field. The experimental arrangement is shown in Fig.5.4.1.
The signal pulses were derived from a CW mode locked Nd:YAG laser
operating at 1.32μm, generating 100ps pulses at 100MHz repetition rate with average
(peak) powers of up to 1.8W (180W). Although previously described theoretical MI
considerations have been based on CW signals, a pulsed source is nearly always used
to avoid loss through SBS, which did not occur here as the pump and signal
bandwidths were too broad. The high peak powers obtained with pulsed excitation
also facilitated autocorrelation, which was required to measure temporal modulations
on the timescale of the MI. However on this timescale, both the pump and signal
126
pulses were in fact approximately CW. The pump pulses were provided by a similar
Nd:YAG laser operating at 1.06μm, producing 100ps pulses with average (peak)
powers of up to 5W (500W). Power fluctuations in both lasers were less than 3%peak,
and the laser pulsewidths were continuously monitored by sampling both input and
output beams from the experiment on a synchroscan streak camera. This also
facilitated synchronisation of the pulses, which were temporally overlapped to within
20ps at the fibre input with an optical delay line.
The fibre used for the nonlinear interaction (F2) had a minimum dispersion
wavelength λ0=1.27μm, and hence was weakly anomalously dispersive for the probe
signal at 1.32μm, and normally dispersive for the pump at 1.06μm. The fibre had
GVD parameters |D|=5ps nm-1km-1 at 1.32μm, and |D|=26ps nm-1km-1 at 1.06μm.
Since the GVD parameters and hence group velocities still differed considerably, the
XPM interaction length was limited to the walk-off length which was computed as
follows: The propagation time delay between two signals with wavelength separation
Δλ is given by18:
Δ T = L D λ Δλ
c λ (5.4.1)
where L is the fibre length, λ is the average wavelength and D(λ) is the relative
normalised dispersion parameter in dimensionless units given by:
D λ = D cλ (5.4.2)
ΔT = D Δ λ L (5.4.3)
For the 100ps pump and signal pulses, ΔT=200ps, |D|=(26-5)=21ps nm-1Km-1,
and Δλ=255nm. The interaction length L was therefore 37m for the pulses to walk-off
127
by 200ps. The length over which MI develops is characterised by the nonlinear length
LNL=(γP0)-1 from Eq(5.1.7). For fibre (F2) used here, Aeff=96μm2, and hence for a
typical launched pump power of 1W peak, LNL=212m, which was over 5 times the
walk-off length. To increase the interaction length for the peak powers used in this
experiment, the 100ps pulses were initially broadened in fibre (F1) which was 2km
long, with a minimum dispersion wavelength λ0=1.48μm. The combined action of
SPM (which caused ~10nm spectral broadening) and the large positive GVD
parameter of 20ps nm-1km-1 broadened the pump pulses to a duration of approximately
500ps. The maximum launched power into fibre (F1) was maintained at 400mW using
a X10 microscope objective (MO1) to ensure that Raman scattering did not occur18,10,
which would have severely distorted the pump pulses and resulted in a highly
nonlinear induced phase shift of unknown quality in the signal, via XPM. The
resultant 500ps, 8W peak power pump pulses emerging from fibre (F1) had a SPM
broadened spectrum, with a linear chirp19 of approximately 0.02nm s-1. ΔT in
Eq(5.4.3) was now 500+100=600ps for the pump - signal interaction, yielding a walk-
off length of L=112m. This walk-off length was much closer to the nonlinear length
(212m), increasing the chance of generating MI. The use of 1.2km of fibre for (F2)
also meant that if total phase synchronism was lost between pump and signal pulses,
the pulses would still interact somewhere in the length of fibre (F2).
Both the 1.06μm pump pulses from (F1) and the 1.32μm signal pulses were
simultaneously launched into fibre (F2) using a dichroic beamsplitter (BS2) and X20
achromatic microscope objective (MO3). The radiation emerging from the fibre was
collected and collimated with an identical objective (MO4) into a 1m scanning
no Raman conversion took place. The pulses continued to broaden dispersively in this
fibre from the 500ps, 10nm bandwidth to 800ps in the presence of a GVD parameter
128
of 26ps nm-1Km-1 at 1.06μm. The average power of the 1.32μm radiation launched
into fibre (F2) was maintained at 20-25mW (2-2.5W peak), which was below the
The spectral output from fibre (F2) around the 1.32μm signal wavelength both
with and without the pump radiation at 1.06μm is shown in Fig.5.4.2. for various
levels of pump power.
129
Fig.5.4.2 .Variation of fibre (F2) output spectrum with pump power (a)0, (b)20mW
(0.4Wpeak), (c)30mW (0.6W), and (d)40mW (0.8W) at 1.06μm. The signal power was
25mW at 1.32μm.
With no pump power present (a), for 25mW average signal power at 1.32μm
(~2.5W peak) only slight SPM induced spectral broadening was evident. This is
consistent with that predicted from Eq(1.4.10), i.e.~1.1nm, provided the uncertainty in
launched pulse duration was accounted for. When 20mW (0.4W peak) of 1.06μm
chirped pump radiation was introduced simultaneously, a rapid growth of sidebands
was evident (b), separated from the central peak by ~8.7cm-1 (260GHz, or 1.5nm). A
further increase of the pump power at 1.06μm to 30mW (c) and 40mW (d) caused the
sideband intensity to increase accordingly. The sidebands in turn generated secondary
sidebands at a frequency separation of ~17.8cm-1 from the central peak. A five fold
increase in the intensity of the secondary peaks occurred for a 30% increase in pump
power, illustrating the rapid exponential nature of the MI gain process predicted
theoretically1. The fundamental modulation increased steadily with pump power, as
predicted in Eq(5.1.6).
130
bandwidth of the central feature was 1nm without pump radiation, increasing to
1.35nm for 40mW of launched average pump power (0.8W peak) at 1.06μm. This was
evident in the spectra shown in Fig.5.4.2(a) & (d). The 0.35nm bandwidth increase
was in good agreement with that predicted (0.37nm) for an interaction length of 112m
in fibre (F2).
131
launched signal power alone. In the presence of the 1.06μm pump pulses, a distinct
deeply modulated, sub-picosecond pulse structure appeared on the autocorrelation
trace as shown in Fig.5.4.4. The duration of the individual structured pulses was
measured (assuming sech2 profiles) to be 520fs, with a temporal period between the
pulses of 3.5ps, which was in good agreement with the inverse of the 290GHz
measured frequency separation of the MI sidebands shown in Fig.5.4.3(b).
Decreasing either launched pump or probe power, over a fixed fibre length,
resulted in a reduction of the temporal structure and an increase in the modulation
period. This was consistent with theoretical prediction11, and was indicative of optimal
compression of the modulation occurring earlier in the fibre for higher pump powers
with, as a result, the generation of more satellite pulses over the rest of the gain length
in the fibre. The femtosecond structure shown in Fig.5.4.4. was superimposed on a
long ~100ps background pulse represented by the input signal pulse. A modulation
depth at the centre of 60% maximum was indicated on the trace, indicating that the
pulse structure was 100% modulated6a,6b. This was confirmed by the large proportion
of energy present in the modulated sidebands of the spectrum in Fig.5.4.3(b).
132
channels may suffer from cross-talk induced MI. Coherent detection would be
rendered unusable if XPM altered the incoming phase of transmitted information.
For an M-channel system, the maximum power in each channel is restricted to:
P < 0.05 α
γ M-1 (5.4.4)
where α-1 is the fibre loss length and γ is the nonlinear coefficient defined in
Eq(1.4.12). This is consistent with experimental data obtained with laser diode signals
propagated over 15km of fibre22,23, which resulted in an imposed power limitation of
1mW peak to restrict the induced phase shift below ΔΦ=0.1. Since the soliton power
scales with γ-1, this suggests that WDM in a soliton transmission system may present a
problem, since low values of γ would result in a higher fundamental soliton power P0.
In section 5.3, Cross Phase Modulation (XPM) was introduced. It was shown
theoretically that XPM between copropagating waves with non-overlapping spectra
could lead to spectral broadening of one wave by the other (and vice versa). For the
case where both waves are pulses of approximately similar durations, the effect is
indistinguishable from SPM, apart from phase and spectral asymmetry resulting from
walk-off between the pulses. If dispersion begins to dominate over the fibre lengths
involved, i.e.
T ps < L β 2
where L is the fibre length, T is the pulse duration and |β2| is the the GVD in ps2km-1,
the asymmetric spectral reshaping can also lead to temporal reshaping. In this chapter,
XPM is applied to a low intensity CW laser diode signal in the anomalous GVD
regime, induced through copropagation with an intense 100ps pulse in the normal
dispersion regime. In this way, picosecond pulses are generated from the weak CW
133
diode laser signal through the temporal reshaping which results from spectral
broadening through XPM by the intense pump4. The effect relies primarily on the
soliton-like shaping which occurs when a spectrally broadened pulse propagates in the
anomalous GVD regime, but the spectral broadening arises through the high intensity
of the pump radiation in the normal GVD regime rather than SPM arising from the
signal.
same.
Fig 5.5.1. illustrates the effect of pump - signal walk-off on the spectrum of the
signal for the situations where the group velocities of the pump and signal are equal
(Vp=Vs), and for the situations where (Vp<Vs) and (Vp>Vs). These theoretical
predictions are based on XPM between a strong pump pulse and a weak signal pulse
of the same duration, which initially coincide in the fibre.
134
Fig.5.5.1. XPM spectra for signal walk-off through the front of the pump (Vp<Vs),
perfect tracking (Vp=Vs), and through the rear of the pump pulse (Vp>Vs).
In the first case (Vp<Vs) the signal pulse walks through the front of the pump
pulse, and hence only receives a net red shift resulting from the leading edge of the
pump pulse. The opposite occurs for the case where (Vp>Vs). For the perfect case
tracking where (Vp=Vs), the spectrum remains symmetric and indistinguishable from
SPM. For complete walkthrough of the pump and signal, which can only occur if the
pulses initially do not overlap, the net phase change is zero, as the signal first
experiences a net red shift followed swiftly by a net blue shift.
Symmetric spectral broadening with a linear, positive chirp should lead to soliton
135
pulse shaping of the CW signal area which overlaps with the pump pulse.
and the fibre was single mode above 1μm. Fig.5.5.2. is a plot of the measured group
delay versus wavelength with the computed GVD parameter for the fibre used in this
experiment. From this plot, it was estimated that the pump pulses at 1.06μm would
perfectly track at the same group velocity with a signal propagating at 1.53μm. This
permitted efficient overlap of the pump and signal over the entire fibre length,
producing symmetric spectral broadening. The GVD parameter at 1.06μm was 26 ps
nm-1km-1 (normal GVD) and at 1.53μm was 16 ps nm-1km-1 (anomalous GVD). For a
maximum launched pump power of 500mW average (50Wpeak), which represented a
power density of ~50MWcm- 2, the magnitude of the expected spectral broadening in
the 1.53μm signal was over 30nm, assuming perfect pump signal tracking and
neglecting Raman conversion in the pump. The maximum launched signal power was
60μW average, hence effects from SPM in the signal were negligible over the fibre
length involved.
136
Fig.5.5.2. Measured group delay (upper) and calculated GVD (lower), indicating
perfect tracking between the signal at 1.53μm and the pump pulses at 1.06μm.
137
Fig.5.5.3. Spectrally broadened laser diode signal through XPM from 480mWof
average power in 100ps pulses at 1.06μm.Diode laser tuned to (a)1459.5nm and
(b)1541.5nm.
In (a) the laser diode was tuned to 1459.5nm, resulting in enhanced broadening
occurring on the short wavelength side of the spectrum. Long wavelength
enhancement occurred for tuning to 1541.5nm, as shown in (b).The asymmetric
broadening was related to the influence of relative walk-off between the pump and
signal in the fibre, since in (a) (Vp<Vs), and in (b) (Vp>Vs).
138
Fig.5.5.4. Spectral broadening occurring in signal pulses through XPM, with and
without the presence of significant dispersion. The asymmetry results from walk-off
where (Vp>Vs).
Fig.5.5.5. Variation of spectral broadening width with pump power through XPM, for
a fixed signal power of 60μW at 1530.5nm.
139
The bars in Fig.5.5.5. relate to the bandwidth or spectral extent of the diode
signal broadened by the corresponding average pump powers shown. A trend in
increased spectral width was apparent, with little shift in the centre wavelength,
indicating that near perfect tracking was achieved for a signal near 1.53μm as
predicted by Fig.5.5.2.
For the highest launched average pump powers (500mW) significant Raman
conversion occurred from 1.06μm to 1.12μm, interfering with the XPM process by
depleting the pump pulse envelope. This was particularly noticeable for lower signal
wavelengths (e.g.1452nm), at which pump powers sufficient to generate Raman
conversion produced significant spectral broadening up to three times that normally
measured when Raman conversion was absent. The resultant broadening was also
asymmetric, being enhanced to the long wavelength side. This can be explained
qualitatively by the pump travelling slower than the signal, as shown in Fig.5.5.2. for a
signal wavelengh of 1452nm. During Raman conversion the pump pulse would have
become depleted and the Raman component would walk-off through the front of the
pump pulse, inducing a distinctly asymmetric chirp onto the pump through XPM13,10,
resulting in an enhancement on the long wavelength side. This chirp, if transferred
through XPM onto the laser diode signal, would have been responsible for the
observed spectral asymmetry at high pump powers.
Fig.5.5.6. Variation of spectral broadening through XPM with signal wavelength, for
60μW of signal and a fixed 480mW of average pump power.
140
Fig.5.5.6. illustrates the effect of group
velocity mismatch and hence walk-off between the
pump and signal on the XPM process by variation
of the signal wavelength. The average signal power
was maintained at 60μW, and the pump power at
480mW.
141
obtain autocorrelation of the weak output signal to measure the generated pulsewidths.
The average signal power was around 60μW, and of this only a small fraction (~1%
by duty cycle) was attributed to the XPM signal. The pulses generated through
anomalous dispersive compression of the spectrally broadened (positively chirped)
signal could therefore only constitute a maximum of ~0.6μW of average power. It was
however, possible to display the pulses on a conventional storage oscilloscope. An AC
coupled Germanium photodiode with ~100ps rise time was used to measure the
temporal distribution of the spectrally resolved signal. Two, 15dB gain, broadband
(500MHz) RF amplifiers were placed after the photodiode to improve the signal to
noise ratio. The spectral resolution was reduced ~1nm to obtain higher sensitivity. The
rejection ratio of the spectrograph prevented pickup of the pump signal when tuned to
the signal wavelength, and no modulation of the diode signal was observed in the
absence of the pump pulses in the fibre. Modulation required the presence of both
signals, and hence resulted from the XPM process. Fig.5.5.8. shows the spectrally
filtered signals monitored with the photodiode and oscilloscope with a combined
resolution of 2ns. The top trace shows the pump pulses at 1.06μm, for phase reference.
The trace marked CW (carrier wave) was obtained by spectrally resolving the central
region of the spectrally broadened signal in Fig.5.5.7., and indicates that depletion of
the carrier signal was occurring simultaneously with modulation by the carrier pulse.
This is consistent with XPM forcing energy away from the central CW region of the
spectrum periodically. The lower traces marked LSB and USB correspond to the lower
and upper sideband components of the spectrum respectively. Clearly the sidebands
are created in synchronism with the pump pulses, within the experimental error
invloved.
Assuming that the operational parameters involved here are reasonably close to
those used in the theoretical model which this experiment was designed to emulate11,
then pulses of only a few picosecond in duration may have been formed through the
XPM process. The fibre used in this experiment was substantially longer than that
which is optimum for short pulse generation, which has been shown theoretically11 to
142
lead to short pulse fragmentation at the expense of the central pulse. Both higher
sensitivity and temporal resolution would be required to confirm this predicted
behaviour.
Modulational Instability.
Induced Modulational Instability (MI) can occur when XPM induces spectral
broadening in a relatively weak signal propagating in the anomalous GVD regime, as
described in section 5.4. The effect can also be induced if a probe wave is launched
simultaneously with a pump wave, frequency shifted from the pump so as to fall
within the MI gain region. The process can be envisaged as occurring through
frequency mixing27 between the probe wave and the relatively intense pump, with the
probe acting as a source of seed photons instead of the process evolving from
spontaneous noise. If |Ω| is less than Ωc given by Eq(5.1.6.), then MI will amplify a
probe wave at frequency ω1, where ω1=ω0+Ω, and ω0 is the frequency of the intense
pump field. Physically, two photons from the pump at ω0 are parametrically converted
into one at ω1 and one at 2ω0-ω1, analogous to degenerate four wave mixing3,4.
The process can only be induced if the pump falls in the anomalous dispersion
143
regime, unlike XPM induced MI which has been shown to produce similar phenomena
under conditions where group velocity mismatch is facilitated by use of
birefringence5,28,16. Through application of an external probe signal copropagating
with a pump of insufficient intensity to produce MI by itself, MI can give rise to the
generation of ultrashort soliton-like pulses at repetition rates governed solely by the
MI sideband detuning from the carrier4. This should allow the generation of pulse
trains with repetition rates up to ~10THz, before frequency level-off through the
influence of higher order dispersion occurs5. These modulation rates are up to an order
of magnitude higher than those obtainable electronically or electro-optically.
144
spontaneous MI.
The fibre sample (F) was 500m in length, single mode around 1.3μm, with a minimum
dispersion wavelength λo=1.3μm. The fibre was therefore anomalously dispersive at
the pump wavelength around 1.32μm, enabling MI to be observed with relatively low
power single beam pump pulses. The signal emerging from the fibre was interrogated
using a 1m spectrograph with 0.1nm resolution, and a standard background free (non-
collinear - see appendix) autocorrelator with less than 50fs resolution.
145
Fig.5.6.2. Output spectra from 500m of fibre for input of (a) 70mW average power at
1.32μm alone and (b) with simulataneous injection of 300μw CW radiation at
1.307μm from laser diode.
146
Stimulated Raman amplification of the Stokes sideband did not dominate for
the relatively low pump powers used here by seeding on the anti-Stokes side of the
pump wavelength. Interference of the frequency components generated spectrally gave
rise to the evolution of pulse trains temporally. An autocorrelation trace of the output
corresponding to the spectra of Fig.5.6.2(b) is shown in Fig.5.6.3.
In Fig.5.6.3(a), the trace was taken over the approximate FWHM duration of
the pump pulse, which was ~100ps. On the broad pump pulse profile, trains of
femtosecond pulses separated by ~4.6ps had evolved. The period corresponded well
within experimental error to the 4.7ps modulation period inferred from the 1.2nm
spectral separation of the longitudinal modes of the laser diode. Around the peak of
the autocorrelation trace a 50% modulation depth was apparent. This suggests that the
generated pulse trains were almost 100% modulated6a,6b. Fig.5.6.3(b) shows the
autocorrelation trace detail of one of the pulse trains on an expanded time and
intensity scale. The 490fs measured pulse separation agreed remarkably well with the
475fs separation inferred from the 12nm wavelength difference between the 1.32μm
pump and 1.307μm probe wavelengths.
Assuming sech2 pulse profiles4, the individual pulses in the train were
measured to have 130fs duration, as shown in Fig.5.6.3(b). This duration would be
dependent on pump power, wavelength, dispersion magnitude and fibre interaction
147
length. The asymmetric nature of the sidebands in Fig.6.5.2(b). was due to slight
Raman conversion providing preferential gain on the long wavelength side of the
pump. This is a typical characteristic precursor to soliton Raman continuum
generation (see section 6.2), however no Raman shifted components were observed
over the entire spectral range.
5.7 References
148
(2) K.Tai, A.Hasegawa & A.Tomita. Phys.Rev.Lett. 56,135(1986) U U
149
6. SOLITON RAMAN GENERATION
AND AMPLIFICATION
between pump and Stokes pulses is minimised. The relatively broad bandwidth of the
Raman Stokes band (~100nm) extends from near 0 to over 1000cm-1 (peaking at
~440cm-1 in silica fibres), enabling amplification of pulses as short as ~100fs duration.
The pulse shaping mechanism responsible for generating a soliton Raman continuum
is however far more subtle than that involved when solitons are excited directly, by
launching transform limited pulses of appropriate peak power.
A consequence of there being finite Raman gain at near zero detuning from the
pump wavelength is the possibility of amplifying MI pulses, (see chapter 5) which
may develop on a long pump envelope if it falls in the anomalous GVD region. The
existence of MI, which is by no means a prerequisite to soliton Raman generation, can
substantially reduce the threshold powers required. Instead of amplifying MI, a small
signal injected within the Stokes Raman gain band could pre-seed the conversion
process6, experiencing considerable gain. Noise bursts or sub-fundamental soliton
power (linearly dispersive) pulses could also be amplified in a similar manner,
eventually achieving power levels sufficient to form fundamental solitons. In this
chapter a series of experiments are described demonstrating how soliton pulses form
150
the stable product of any amplified signal provided enough energy is present, and how
they can be regenerated at a point after which dispersion has reduced the intensity well
below the fundamental soliton power.
since the pump and Stokes will not significantly walk-off, resulting in considerable
conversion efficiency, but this is surprisingly less efficient than if MI acts as a
precursor. The pump does not undergo MI in this case, since it falls in the normal
GVD region, however the influence of XPM on the Stokes pulse leads to considerable
soliton shaping.
151
SRS wavelength conversion of the pump radiation begins either at the Stokes
MI sideband, the peak of the Raman gain, or that wavelength which walks off least
from the pump, depending on the exact characteristics of the fibre and pump signals.
Any number of pulse-like structures developing from amplified MI undergo soliton
compression as their energy increases, with an associated decrease in pulse duration in
the presence of gain10.
By virtue of the random nature of the process which essentially evolves from
noise, more than one soliton can develop per pump pulse, at no fixed wavelength. For
relatively high powers, pump pulse fragmentation occurs rapidly through Raman
enhancement of the MI signals12. These modulations are themselves solitary waves
and can be relaunched to produce solitons in their own right using Raman gain for
amplification13. The bandwidth of the amplified and compressed soliton fragments
becomes sufficient to produce intra pulse Raman scattering, and each soliton self-
frequency shifts (SSFS) to a wavlength, dependent on the soliton duration.
SSFS is a peak power and hence pulsewidth dependent effect, such that:
dω = - D
dz τ4
where D is the dispersion parameter of the fibre, and τ is the soliton pulse duration.
The power dependency arises from the decrease in soliton duration occuring through
Raman amplification. The SSFS is extremely sensitive to soliton duration τ, and
different pulses therefore shift to longer wavelengths by varying amounts. The spectral
development with pump power, of the output of a 500m length of fibre (λ0=1.3μm),
for a launched pump signal of 100ps pulses from a Nd:YAG laser at 1.32μm is shown
in Fig.6.2.1.11
152
Fig.6.2.1. Development of spectral Raman continuum generation with average pump
power11, showing evolution from MI. Intensity sensitivity is constant throughout.
Here the process was initiated by MI, as the pump propagated in the
anomalous GVD regime. For 60mW of launched average power, MI sidebands were
apparent at a frequency separation of 67cm-1 from the pump wavelength,
corresponding to a modulation frequency of ~2THz, or a period of 500fs. Background
free autocorrelations (see appendix) used to confirm this effect showed a broad pump
envelope with this modulation superimposed. Fig.6.2.(a)-(h) shows the temporal
development of the MI as it evolved into solitons, corresponding to the previous series
of spectra.
The modulation frequency, Ω, for which maximum gain occurs is related to the
pump power P0 through the following relationship (from Eq(5.1.10)):
153
Ωα 2 n2 P0
D
where n2 is the nonlinear refractive index coefficient and D is the dispersion
154
contributed to the autocorrelation pedestal through cross-correlation.
The centre of the pump pulse had the highest peak power, and hence
discontinuities or noise were most likely to give rise to MI in this region. The pulses
generated would receive Raman gain, at a longer wavelength than the pump, and
therefore travel slower than the pump as they were further from λ0. This resulted in the
signal pulses walking through to the back of the pump pulse. These then collided with
solitons in the early stages of formation, generated near the back of the pump pulse,
with any instantaneous enhancement of peak power and bandwidth during collision
resulting in further SSFS.
If a small external signal is injected within the Raman gain band of a pump
pulse in an optical fibre, this precursor signal would be expected to seed the Raman
conversion process in a similar manner to that produced when MI occurs. Problems
associated with the solitons being derived from spontaneous noise, or from the soliton
155
shaping that ensues from amplification of MI can in this way be overcome. To
facilitate this, a semiconductor laser diode, tunable over the Raman gain band, could
provide the seed signal which would substantially reduce the Raman conversion
threshold at one specific wavelength region. The exact wavelength at which the
conversion is initiated would be defined, and provided the pump power is controlled,
multiple pulse generation could be suppressed. The schematic for such an
experiment15 is shown in Fig.6.3.1.
156
from the high power pump pulses, which caused damage in the diode facet if
relaxation oscillations occurred in the pump laser.
A standard X20 microscope objective (MO1) focussed both pump and signal
beams into 1.98km of standard single mode fibre with a dispersion minimum at
λ0=1.34μm, and a loss of 0.5dBkm-1 around 1.3μm. An identical objective (MO2)
collected and collimated the fibre output beam into both a 1/2m scanning
monochromator and a background free (non-collinear - see appendix) autocorrelator
with resolutions of 0.2nm and ~20 fs respectively. To investigate the temporal
constituents of the broad spectral features generated, narrow bandpass dielectric filters
were inserted before the autocorrelator. Fig.6.3.2(a) shows the spectrum recorded for
an average launched pump power of 160mW(~16Wpeak) at 1.32μm in the fibre.
Fig.6.3.2. Spectra of fibre output for (a) 160mW average pump power at 1.32μm
alone, (b) as in (a) plus 60μW CW diode laser signal at 1.36μm, (c) 180mW pump
power alone, and (d) as in (c) plus 70μW CW diode laser signal at 1.36μm.
No MI was evident at any level of pump power, since the pump wavelength
here was in the normal GVD regime. This resulted, for equivalent fibre and input
pulse parameters, in a higher threshold for Raman soliton generation17. This higher
threshold was beneficial in the experiment, reducing the chances of producing
spontaneous Raman solitons from fluctuations of the pump pulse profile, at power
levels which would produce little gain in the laser diode signal. A low intensity
Raman component centred at 1.38μm was evident in (a) (the Raman gain peak occurs
around 1.4μm in fused silica fibres16), but at this power level (160mW) no temporal
157
soliton evolution was evident in the most sensitive autocorrelation measurements,
indicating that any radiation in the Raman component was of insufficient energy to
form solitons.
The laser diode signal was tuned ~20nm lower than the spontaneous Raman
peak to 1.36μm in order to define the wavelength of soliton Raman operation. When
60μW of laser diode signal was simultaneously coupled into the fibre with 160mW of
pump radiation, a large amplification of the diode radiation at 1.36μm took place, as
shown in Fig.6.3.2(b). It was possible to continuously tune the laser diode from 1.33 to
1.38μm, and obtain gain centred at the seed wavelength in a similar manner.
For the 160mW pump power case (b), the largest pulse/pedestal contrast ratio
was obtained. Below this level, the gain in the laser diode signal fell off rapidly. The
threshold for amplifying a 100μW laser diode signal was measured by spectrally
resolving the band centred at 1.36μm using a 10nm bandpass filter, and was estimated
158
at 140mW of average pump power. The autocorrelation trace corresponding to the
situation shown spectrally in Fig.6.3.2(d) is shown in Fig.6.3.3., with a 10nm bandpass
filter centred at 1.36μm placed before the autocorrelator.
The Fourier transform pulse duration limit imposed by the 10nm bandpass
filter would have the effect of limiting the measured pulse duration to about 200fs, if a
sech2 profile appropriate to soliton pulses is assumed. To ensure that the measured
pulses were not in fact shorter, the filter was replaced by a similar filter with 18nm
bandpass, yielding the same measured pulse duration. Provided that the pump power
was maintained at a180mW +/-5mW, the output pulses remained approximately n=1
solitons, tunable from 1.33 to 1.38μm. The spectra of Fig.6.3.2(d) indicates a FWHM
Raman bandwidth of ~20nm, which could theoretically support a 100fs soliton.
However no such short pulses were ever measured, using an unfiltered autocorrelator.
The reason for this is attributed to there being more than one soliton present in this
spectrum. The long wavelength shifted band around 1.38μm was in fact due to the
SSFS, and was emphasised for higher pump powers.
159
increasing pump power, with a fixed 100μW signal power.
Fig.6.3.4. Variation of the measured soliton pulsewidth at the 1.36μm seed wavelength
with pump power, for a fixed seed signal power of 100μW.
160
collisions occurring over long lengths of fibre could have resulted in the ascribed
broadening behaviour.
For example, in the fibre used here, the dispersion parameter |D| ~5ps nm-1km-1
at 1.36μm. For two 250fs solitons spectrally separated by as much as 20nm, the
interaction length would correspond to ~2.5m. The Raman conversion threshold given
by Eq(1.5.4) over 2.5m would yield a peak power value of ~60W, which is an
attainable peak power in the case of a seeded interaction. There is therefore sufficient
peak power for energy conversion and broadening in the results presented here.
Broadened pulses which fall below the fundamental soliton power disperse away,
contibuting to the measured autocorrelation pedestal. This effect is illustrated in
Fig.6.3.5.
161
the initial wavelength at which Raman gain occurred, reducing the conversion
threshold for one particular wavelength. Fig.6.3.6. shows how the seed signal power
level had very little influence on the generated soliton pulsewidth. The seed signal
only appeared to provide an alternative to growth from noise, or discontinuities arising
from imperfect pump pulses or MI, which normally self selects a wavelength for
maximum growth. The laser diode provided a more predictable and controllable seed
wavelength for the process, but the seed power made very little difference.
Fig.6.3.6. Variation of the measured soliton pulsewidth with laser diode seed signal
power at 1.36μm, for a constant pump power of 160mW.
For much higher launched pump powers, cascade into higher Raman Stokes
bands occurred, as indicated in Fig.6.3.7. The launched pump power level was
increased to 190mW average at 1.32μm, in optimised 80ps pulses.(24W peak). In the
absence of any laser diode seed signal, soliton generation occurred at 1.38μm, with no
significant component at 1.36μm, as shown in Fig.6.3.7(a).
162
Fig.6.3.7. Measured spectra with (a) 190mW average pump power (80ps pulses) at
1.32μm alone, and (b) with the addition of 160μW CW laser diode signal at 1.36μm.
When 160μW of 1.36μm CW laser diode radiation was added, severe pump
depletion occurred, and soliton generation was seeded at 1.36μm. This yielded 370fs
pulses within a 10nm, 1.36μm band, on a 30% intensity pedestal. The seeded pulses
experienced considerable SSFS, combined with generation of second Stokes radiation
at 1.48μm. By spectrally resolving this band, 220fs pulses on very low (~1%)
pedestals were measured by autocorrelation. This band was confirmed as a second
Stokes band since no variation of the centre wavelength occurred with pump power, as
would be expected from a signal undergoing SSFS.
163
6.4 Stimulated Raman Amplification of Noise Bursts into Solitons.
100ps pulses. For a cavity length mismatch from the optimum of +/-10μm, the ring
laser generated pulses with the temporal structure characteristic of a burst of noise, as
shown in the autocorrelation trace of Fig.6.4.1.
Fig.6.4.1. Background free autocorrelation of the ring laser output for a cavity length
detuning of +10μm. Similar structure occurred for equivalent negative mismatch.
164
invlove periodic, highly pertubative gain, since they show that noise can develop, with
sufficient gain, into non-dispersive soliton signals, which may lead to considerable
system error.
Fig.6.4.2. Schematic of the experiment for amplifying noise bursts into solitons.
165
Beam splitter (BS1) (25%R at 1.32μm) divided the pump beam form the mode
locked Nd:YAG master pump laser. Approximately 400mW of the 1.8W average
(180Wpeak) power available at 1.32μm was directed via beamsplitter (BS1) and an
optical delay line onto beamsplitter (BS2) to provide a sycnchronous pump for the
noise signal. The 20-40mW average power noise signal at 1.39μm was derived from
the detuned ring laser, which in turn was pumped by the remaining 1.2W of pump
radiation passing through beamsplitter (BS1). Both 1.32μm pump and 1.39μm noise
signal were combined with dichroic beamsplitter (BS2) and focussed with a standard
X10 microscope objective (MO1) into 500m of fibre with a minimum dispersion
wavelength λ0=1.38μm. The fibre had a small anomalous dispersion parameter
|D|=2ps nm-1km-1, and a mode field area of ~64μm2 at 1.4μm. The output from the
fibre was collected and collimated by an identical lens (MO2) into a scanning
166
occurring at these relatively low signal powers. The autocorrelation trace of the signal
output from the fibre is shown in the presence of synchronous gain in Fig.6.3.4., for
(a) 45mW and (b) 65mW of pump power at 1.32μm. In (a), the trace of the
unamplified noise signal in the absence of pump power is shown on a similar scale for
reference near the baseline.
Fig.6.3.4. Background free autocorrelation traces of the Raman amplified noise signal
(1mW average power) at synchronous pump pwers of (a )45mW and (b) 65mW
average at 1.32μm.
In (a) clear amplification of the noise signal was evident, with an estimated
gain from the autocorrelation of ~3.5, assuming no change in pulse duration. The short
coherence spike component of the trace began to dominate the measurement, and
some compression of this component occurred from 2.8ps to 1.16ps. This was
consistent with the ultrashort structures in the noise signal undergoing soliton
compression through amplification, as some of the structures approached a peak
power level close to the fundamental soliton power. For 65mW of average pump
power shown in (b), a further decrease in the FWHM of the spike occurred to 820fs,
and the trace began to exhibit features typical of a pulse with a pedestal, rather than a
bursts of noise. The pulse like structure became more significant, and the
autocorrelation peak/pedestal ratio increased from the 2:1 associated with a noise burst
to 5:1. For the highest launched synchronous pump power, which was 80mW average
at 1.32μm, the autocorrelation trace shown in Fig.6.3.5. was obtained.
167
Fig.6.3.5. Background free autocorrelation trace of the amplified noise signal for
80mW pump power.
The FWHM duration of the pulse structure was now as low as 620fs, above a
pedestal which extended over the pump pulse duration, accounting for ~12% of the
autocorrelation peak signal. By consideration of the respective areas of the
autocorrelation function, it was estimated that ~22% of the energy at 1.39μm was
contained in the short pulse component. By measuring the average spectral power in
this band, the average power in the pulse structure was estimated to be 1.8mW. This
band experienced an average power gain of approximately 8 times.
The fibre used for amplification in this experiment had a dispersion parameter
|D|=2ps nm-1km-1 at 1.4μm. For a pulse of 620fs duration at 1.39μm, the fundamental
soliton power is ~5.7W peak. This would correspond to an average power of 0.36mW
at 100MHz, with a predicted soliton period of 93m. From these calculations, it is
apparent that more than one soliton was present in the amplified noise signal shown in
Fig.6.3.5. In this case, ~5 solitons were generated, and hence due to the random nature
of the soliton evolution, the autocorrelation trace represented an integrated cross-
correlation between these pulses, contibuting to the pedestal intensity. Relaunching the
amplified noise signal into a second fibre of over 1km length, and similar
characteristics to the first yielded similar though slightly longer pulses at the outptut.
Since this was significantly longer than the soliton period (Z0~93m), which also
characterised the sub-fundamental soliton dispersion length LD from Eq(1.4.32), the
168
amplified noise signal had definitely partially evolved into soliton pulses. The pedestal
on the autocorrelation was retained, indicating that some sub-fundamental soliton
power structure was still present after propagation over 1km, as was expected.
169
6.5 Stimulated Raman Amplification of Dispersive Pulses
into Solitons.
6000km of transmission fibre21a,, with zero net signal loss and less than 8% energy
fluctuations. To reduce the Raman gain perturbation still further, quasi-CW gain can
be provided by periodic bidirectional pumping via dichroic fused couplers, resulting in
very small pulse energy fluctuations. In this way the total Raman gain becomes the
sum of two exponential functions, one decaying whilst the other grows.
170
to ascertain the fundamental soliton power. The fibre used here was 140m long, with a
minimum dispersion wavelength λ0=1.27μm, and an anomalous GVD parameter
|D|=15ps nm-1km-1 at 1.4μm. The mode field area of the fibre was estimated to be
~80μm2 at 1.39μm, resulting in a predicted fundamental soliton power of 8.1W peak
(~1.5mW average) for the ~1.8ps signal pulses. The predicted soliton period was
calculated to be 105m, and hence signal pulses would exhibit significant dispersion
below the fundamental soliton power. The output from the optimised Raman ring laser
is shown in the background free autocorrelation trace of Fig.6.5.2(a).
Fig.6.5.2. Autocorrelation traces of (a) output signal pulses from Raman ring laser
and from amplification fibre for (b) 12mW and (c) 2.5mW of average launched signal
power. In (b) & (c) the signal pulses are shown as an outer trace for reference.
This trace indicates a signal pulsewidth of ~1.8ps. Fig.6.5.2(b) shows the same
input signal pulse (outer trace) and the autocorrelation trace of the output from the
fibre for a launched average signal power of 12mW in the absence of any pump
power. The trace shows a clear high order soliton compression to 1.3ps, indicating that
the signal power exceeds the fundamental soliton power, and is in fact exciting a high
order soliton. This measured duration is considerably longer than the optimum pulse
duration occurring through high order soliton compression (~110fs), because this
compression occurs at a point ~27m from the input to the fibre.
171
By reducing the launched signal power to 2.5mW average, the autocorrelation
trace show in Fig.6.5.2(c) was obtained. This shows that the input and output signal
pulses are of approximately the same FWHM duration, i.e. ~1.8ps, with only a slight
envelope disparity in the wings of the trace due to a deviation of the launched signal
pulses from the ideal sech2 shape. This similarity in FWHM duration persisted for
average powers as low as 1.5mW, in good agreement with the predicted fundamental
soliton power.
For the amplification investigation, the test fibre was replaced with a 600m
length of fibre with a dispersion minimum wavelength λ0=1.38μm, and an anomalous
GVD parameter |D|=3ps nm-1km-1. The mode field area for this fibre was 95μm2,
yielding a predicted fundamental soliton power of 1.97W peak, or 350μW average for
the 1.8ps signal pulses at 100MHz, with a predicted soliton period of 525m. Once
again, below this power level, noticeable dispersion was expected to dominate over
the 600m fibre length. The calculated walk-off length between the 1.8ps signal pulses
at 1.39μm and the 100ps pump pulses at 1.32μm was ~400m, indicating that the gain
length and soliton period were of similar magnitude, with αZ0~1. The gain process
was thus still approximately adiabatic, and hence the soliton was expected to adjust its
duration to match its amplified energy accordingly. High order solitons would only be
excited for αZ0>>1, thus the gain length would have to have been comparable with the
soliton period, which for the pulses used here is practically impossible to achieve
using Raman gain without causing loss to higher order Stokes bands.
172
Fig.6.5.3. Autocorrelation traces of the output from the amplification fibre for an
input signal of 1.8ps duration at 250μW average power (a) without amplification and
(b) with 80mW of synchronous pump power at 1.32μm.
d τ-1 = α τ-1
dz
where α is the gain coefficient of the amplification process (Raman in this case, which
is proportional to pump power) and τ is the pulse duration. The gain saturates during
Raman amplification, limiting α, and therefore limiting the minimum pulsewidth
obtainable through Raman amplification. Thus the limitiation in the compression
173
achieved in this experiment is ascribed to gain saturation. Gain from e.g. Erbium
doped fibre amplifiers could perhaps provide an even higher compression factor, and
over shorter gain lengths enable investigation in the non-adiabatic regime with
αZ0>>1, where high order soliton excitation would be expected.
The absence of any SSFS effects in the spectral output of the amplifier was
once again ascribed to the prescence of considerable narrow band gain from the
copropagating pump field, which has been shown to suppress the effect through
frequency pulling20. The amplified pulses were verified to be solitons by relaunching
into ~3km of fibre, with only slight temporal broadening occurring due to coupling
losses and fibre parameter differences.
6.6 References
174
(8) A.S.Gouveia-Neto, A.S.L.Gomes & J.R.Taylor. Opt.Lett. 12,1035(1987) U U
175
7. NONLINEAR OPTICAL LOOP MIRROR PULSE PROCESSING
7.1 Introduction
The optical Kerr effect was introduced in chapter 1 which, to a first order
approximation, induces a nonlinear phase shift in the field across a pulse in the
presence of an intensity dependent refractive index change. The local phase of a pulse
undergoes a shift which is dependent on the local intensity, and as the intensity rises
and falls this results in a change in phase. A change in phase results in a frequency
chirp across the pulse, referred to as Self Phase Modulation (SPM). If the phase shift
is induced through the presence of another pulse of sufficient intensity, then Cross
Phase Modulation (XPM) occurs (see chapter 5).
Hence the phase shift (φ) across an intense pulse in an optical fibre becomes
intensity (I) dependent . Over a length of fibre L with coefficient of nonlinear
refractive index n2, Eq(1.4.9) gives:
Δ ∅ = n2 I k L (7.1.1)
where k=2π/λ is the wavenumber for a pulse centred at wavelength λ. If in the first
instance a CW wave is considered, Eq(7.1.1) shows that over moderate lengths of
fibre, a unity phase shift can be achieved for relatively low average powers. The
length at which Δφ becomes unity is in fact the nonlinear length LNL defined in
Eq(1.4.11). At 1.06μm, a fibre of 100m with a mode field area Aeff=38μm2 would
induce a π phase shift for 6.4W of input peak power. If two beams with different
relative intensity dependent phases overlap and interfere, the intensity of the resultant
field would clearly be sensitive to the parameters in Eq(7.1.1). By fixing the length,
for example, an intensity dependent switch could be constructed.
176
A number of both theoretical and experimental devices have been proposed for
switching signals in a fibre communications system, based on routing and logic
operations1 - 4. The devices rely on nonlinearity through the optical Kerr effect, which
provides a very fast response time in the order of 10fs. However the pulsed response
of such devices is not ideal, since the change in envelope intensity across a pulse gives
rise to a range of phase shifts, and hence signal pulse fragmentation. In section 2 of
this chapter the operation of such a device based on a Nonlinear Optical Loop Mirror
(NOLM) is described, both for non-soliton and soliton signal pulses4, illustrating how
a NOLM can provide complete pulse switching without fragmentation. In section 3, an
experiment is described which demonstrates how optical pulses can be 'cleaned up'
using the intensity discriminating nature of the NOLM.
The fused coupler is normally of the twisted evanescent type, with a power
splitting ratio α:(1-α), where {0<α<1). Due to the symmetric nature of the device, the
common optical path length for an input field E1, split by the coupler into fields E3 and
E4, is the same. This is similar in concept to an anti-resonant ring interferometer5, and
avoids the requirement for interferometric alignment of the path length difference.
177
This results in a linear response similar to that of a mirror for α=0.5 exactly, with an
incident field E1 reflected back as E01, with zero output in E02.
in the loop. This is achieved by adjusting the splitting ratio of the coupler, to be
different from α=0.5, so that fields E3 and E4 have different intensities. This intensity
Δ ∅ = 2π n2 L E3 2 - E4 2
λ (7.2.2)
For an evanescent coupler with power splitting ratio α:(1-α), the fields in the
loop are described as follows:
E3 = α E1 + i 1-α E2
E4 = i 1-α E1 + α E2
(7.2.3)
Now for a single input Ein, E2=0. The two fields generated after propagation around
2
E3 = α Ein exp i α Ein 2π n2 L/λ
2
E4 = i 1-α Ein exp i 1- α Ein 2π n2 L/λ (7.2.4)
These equations represent the through coupled and cross coupled fields derived
from Ein, with the addition of a phase factor induced through the nonlinear phase shift
(SPM). The phase difference between the two fields increases as α deviates from 0.5.
178
The total transmitted intensity through the loop mirror is given by :
2 2
E02 = Ein 1 - 2α 1-α 1+COS 1-2α Ein 22π n2 L/λ
(7.2.5)
The input signal is transmitted through the NOLM when |E02|2 = |Ein|2, i.e. when
(1-2α) |Ein|2 2π n2 L/λ=π . Hence, for any value of α, 100% transmission through the
NOLM is obtained if
2π n2 L/λ Ein = mπ
2
m=(1,3,5.....)
1-2α
(7.2.6)
and 100% reflection is obtained whenever m =(0,2,4....). Note that the switching
contrast, given by the modulation depth of the function |E02|2 / |Ein|2, increases as α
approaches 0.5, but the intensity required to reach switching (m=1) increases too, so a
trade-off exists between switching power and obtainable contrast. For example, for a
NOLM consisting of a coupler with α=0.4, and a fibre of 100m length with a mode
field area of 38μm2, the transmissivity will increase from 4% to 100%, for an increase
in coupled input power from 0 to 31.9W (accompanied by a complementary change in
the reflectivity fom 96% to 0%). The transmission function of a NOLM versus input
power for α=0.4 is shown in Fig.7.2.2. The axes show normalised units, with n2L/λ=1.
179
Fig.7.2.2. Transmitted power vs input power for NOLM with α=0.4 and n2L/λ=1.The
solid lines represent the minimum and maximum values obtainable.
Since the response of the electronic Kerr effect which induces SPM in silica
optical fibres is very fast (~10fs), this device offers the possibility of extremely high
frequency switching and signal processing. Therefore, the transient response of the
NOLM is important. The response to a pulsed input signal is, however, sensitive to the
instantaneous intensity of the pulse, resulting in different transmission function values
across the pulse profile. Only perfectly square pulses with durations short compared
with the loop transit time will give rise to a transmission function similar to the ideal
CW case considered up to now, and illustrated in Fig.7.2.2. For realistic bell-shaped
pulses output characteristics similar to that shown in Fig.7.2.3. are obtained.
180
Fig.7.2.3. Transmitted energy (or average power) vs input energy for a typical bell
shaped (sech2) pulse. The CW case of Fig.7.2.2. is shown dotted for reference.
iξ
u(ξ, τ) = sech(τ) exp
2 (7.2.7)
Higher order solitons are more complex, and have an envelope function which
periodically repeats with propagation in ξ with period π/2 (the soliton period). The
phase factor in the n=1 solution shown in Eq(7.2.7) φ=ξ/2 applies to the phase of the
whole pulse, and this indicates that solitons exhibit the special property of aquiring a
uniform phase as they propagate. The effect of soliton shaping in a NOLM can be
included if dispersion is considered, which imposes a requirement that the loop length
181
be of comparable length to the soliton period Z0 (or dispersion length LD) in order for
the phase shift imposed through SPM to be communicated throughout the pulse. This
length can be reduced by employing short pulse durations, or a high anomalous GVD
parameter. In general, the phase of a soliton evolves linearly with propagation
distance3 in accordance with Eq(7.2.7), and linearly with the amplitude of the soliton,
independently of the evolution of the soliton envelope function. This property makes
soliton pulses ideal signals for switching in a NOLM device, which would operate on
the whole pulse.
Neglecting pulse shaping effects through propagation through the loop and
XPM coupling between the counter-propagating pulses, the total energy output of the
NOLM for an input soliton u=Asech(τ) becomes4:
where Z=L/LD is the loop length in soliton units, and LD is the dispersion length which
is related to the soliton period Z0 by Eq(1.4.32). A=1 refers to the single or
fundamental soliton. For a fixed loop length, the soliton phase evolves with amplitude,
hence shorter pulses will achieve switching at lower average power levels, since the
phase evolves faster for a shorter soliton period. The energy of the soliton pulses
determines the reflectivity of the NOLM. Fig.7.2.4. shows the transmission
characteristics of a (α=0.42) NOLM as a function of input pulse energy for CW fields
or square pulses (ideal), bell-shaped non-square pulses (approximate), and solitons.
182
Fig.7.2.4. NOLM transfer function vs input energy for α=0.42, indicating ideal CW
response for square pulses, non-ideal response for bell-shaped pulses, and restored
ideal response for solitons.
183
Fig.7.3.1. Intensity profiles (top) and corresponding background free autocorrelation
functions (bottom) of (a) sech2 input pulses, and transmitted pulses through NOLM for
input peak powers of (b) 3, (c) 4, (d) 5, (e) 6, and (f) 8 normalised units (n2L/λ=1)
(Timescale arbitrary).
Fig.7.3.1.(a) shows the input sech2 pulse profile, and its autocorrelation (the
latter being broader by a factor 1.55; see appendix). In (b), the transmitted pulse
184
through a NOLM with α=0.4 is shown for an input peak power of 3 normalised units.
The peak of this pulse therefore coincides with the first transmission maximum of the
curve shown in Fig.7.2.2., with a resultant transmitted pulse profile which is
temporally shorter, and shows less energy in the pulse wings. The output pulse is also
squared off, because the transfer function changes only slowly near the transmission
peak, resulting in a triangulated autocorrelation trace, shown below. For higher input
peak powers (c)-(f), the pulse undergoes significant splitting and reshaping, whilst the
total energy transfer through the system represented by the area of the pulses shown,
fluctuates only slightly in accordance with the curve shown in Fig. 7.2.3. The
fractional energy transfer corresponding to the pulse shapes in Fig.7.3.1. was (from (b)
to (f)) 0.744, 0.601, 0.396, 0.380 and 0.653. respectively. The inadequacy of the
device in the normal dispersion regime for switching based on pulse energy is clear
from these simulations.
Fig.7.3.2(a) sech2 intensity profile with a 10% noise component (top) and
corresponding autocorrelation function (bottom). The filtered versions transmitted
through the NOLM are shown in (b) for 2.5 normalised units of peak power, with
185
n2L/λ=1.
186
The loop comprised a low loss directional coupler with power splitting ratio
α=0.52, and 200m of dispersion-shifted (λ0=1.62μm, |D|=1.6ps nm-1km-1 at 1.59μm)
non-polarisation preserving fibre, which was fusion spliced between the output ports
of the coupler. Splicing losses of ~1dB were incurred through splicing fibres with
different cross-sections. Since the fibre loop did not preserve polarisation, to ensure
efficient interference of the counter-propagating fields in the coupler, a strain-induced
birefringent polarisation rotator was inserted into the low power (α=0.48) arm of the
loop. This introduced a lump power loss of ~3dB, and was therefore the dominant
symmetry breaking element in the NOLM, effectively increasing the intensity
difference in the counter-propagating fields dramatically.
For a lump loss of power transmissivity TL in the fibre loop, the transfer
2
E02 Pout
2
= = TL 1 - 2α 1-α 1+COS 1-α TL- α φNL
Ein Pin
Pin 2π n2 L
where φNL =
λ Aeff
(7.2.8)
For no lump loss, i.e. TL=1, this reduces exactly to Eq(7.2.5). This results in an
overall attenuation in transmission by a factor TL, but a proportionate reduction in the
required switching power. This also allows the use of an α=0.5 coupler, with its
associated excellent switching contrast. The power transfer characteristic of a NOLM
with α=0.52 is shown in Fig.7.3.4.for a lump loss of both TL=0 and TL= 0.5 (3dB),
once again normalised by setting n2L/λ=1, and Aeff=1.
187
Fig,7.3.4. Power transfer characteristic of a NOLM for α=0.52, with and without a
lump loss TL=0.5 in the low power (α=0.48) arm. Units are normalised with
n2L/λAeff=1.
The normalised switching power, Ps, required for the transfer function to reach
the first transmission maximum in the absence of loss is given from Eq(7.2.6) by
(TL=0):
Ps 2π n2 L
= π
λ Aeff 1-2α
∴ Ps = 1
2 1-2α (7.2.9)
Hence for α=0.52, Ps=12.5 normalised units in the absence of lump loss, in
accordance with Fig(7.3.4). If the effect of the lump loss is considered, the normalised
switching power Ps is reduced to:
Ps = 1
2 1-α TL- α
(7.2.10)
188
For α=0.52 and TL=0.5, this yields a switching power Ps~1.8 units, as shown in
power output arm of the coupler in the NOLM, so as to enhance any intrinsic intensity
imbalance. The transfer function of a NOLM with α=0.52, and a lump gain of TG=2 in
Fig.7.3.5. Power transfer characteristic of a NOLM for α=0.52, with a lump gain
TG=2 in the high power (α=0.52) arm. Units are normalised with n2L/λAeff=1.
The input power Ps required to achieve the first transmission maximum in the
presence of a lump gain TG is given by:
Ps = 1
2 α TG- 1-α
(7.2.11)
189
coupler with α close to 0.5. The presence of gain enhances the signal in transmission
as well, making the device suitable for applications which require a high signal fanout.
Fig.7.3.6. is a plot of the switching power Ps required to obtain the first transmission
For moderate values of gain, e.g.10dB, the power splitting ratio, α, has very
little influence on the switching power Ps, whilst high contrast switching can be
achieved with α close to 0.5. In the example given for α=0.5 and a 10dB lump gain,
the switching power Ps is as low as 0.11. For a typical fibre with a 50μm2 mode field
area this represents a peak power - length product of ~260Wm, requiring only 2.6W
of peak power for switching in a 100m NOLM.
In the experiment described here, a lump loss of 3dB (TL=0.5) was presented
by the polarisation rotator. The fibre mode field area was ~50μm2, and hence the
predicted switching power - length product was 4185Wm. This corresponded over the
200m length of fibre to a 21W peak switching power. The NOLM transmission with
α=0.52 was expected to vary from a maximum of 50% to a minimum of ~0.2%,
190
yielding a very high contrast as illustrated in Fig.7.3.4. Throughout the experiment,
the polarisation rotator was adjusted so as to obtain maximum reflectivity of the
NOLM in the low intensity (linear) regime, by reducing the launched signal power to
an absolute minimum, and monitoring the unused (backward propagating) arm of the
power tap coupler. Fig.7.3.7. shows the experimentally observed autocorrelations of
the NOLM output as the launched signal power from the CCL laser was increased at
the NOLM input. The peak powers indicated were calculated from those measured
including the coupling losses incurred throughout the experiment.
Fig.7.3.7. Autocorrelation traces of (a) input signal pulses from CCL laser, and
transmitted pulses through the NOLM for launched peak powers indicated.
The input signal pulses from the CCL laser are shown in trace (a) for
comparison with the output pulses (b) - (f). For 21W launched peak power (b) a
191
triangular trace was observed, in accordance with that predicted in the simulation
shown in Fig.7.3.1(b). In fact all of the results were consistent with the predicted pulse
shaping effects, up to those shown in Fig.7.3.7(f), which represented the maximum
launched peak power obtainable. The results indicated a measured switching power
Ps~21W peak, which is identical to the predicted value when losses are considered.
through the NOLM gave a measured autocorrelation trace shown in Fig.7.3.8(b). The
device transmitted a narrower, relatively pedestal-free pulse, and clearly exhibited
properties suitable for suppression of low intensity noise and inter-pulse radiation.
In section 7.2, the response of the NOLM in the soliton (anomalous GVD)
regime was described, where it exhibited complete pulse switching, rather than the
partial pulse switching and reshaping illustrated above. In the soliton regime, the
pedestal suppression property of a NOLM is therefore removed, as any pulse which
develops into a soliton would undergo either complete transmission or reflection.
192
Nevertheless, the usefulness of such a device is not impaired, since the device would
be expected to operate as a soliton filter, allowing transmission of only soliton pulses
and reflecting inter-pulse radiation. This would require the incorporation of
anomalously dispersive fibre in the loop, with a length at least twice that of the soliton
period.
To investigate this possibility, optimised CCL laser pulses were injected into a
NOLM consisting of 200m of fibre with a dispersion minimum around 1.3μm. The
now anomalous GVD parameter |D| was ~18ps nm-1km-1 at 1.59μm, and the predicted
fundamental soliton power for the ~13ps pulses in the fibre was ~200μW peak power.
A typical launched peak power of ~20W represented excitation of a very high order
(n~300) soliton in the fibre, with a soliton period of Z0~3.5km. The loop length
(200m) was only a small fraction of the soliton period, and hence complete switching
of the entire ~13ps soliton was not possible. Since the switching power of this NOLM
was extimated to be Ps~150W peak, pulse shaping or indeed transmission of the ~13ps
was not expected. However since the launched pulse formed an n~300 soliton, high
order soliton formation was expected operate on the pulse envelope, compressing it to
less than 100fs, beyond which, pulse breakup was expected to occur. A femtosecond
soliton component in the loop would hence acquire sufficent peak power to cause
complete pulse switching of itself.
193
were launched in the soliton regime. The input pulse shape is shown for reference.
Note that the only component that switched through the NOLM was the short
component arising from the high order soliton breakup. Dispersive energy which
would normally be present in the process was effectively suppressed, and the device
operated as a soliton filter.
7.4 References
194
195
8. CONCLUSIONS
196
Amplification of spontaneous noise or seed signals in the presence of
anomalous dispersion can also give rise to modulational instability. Pulse trains
generated through modulational instability can have ultra-high repetition rates of up to
~5GHz. Hence the process is extremely useful for generating signals and clock pulses
for use in a time division multiplexed soliton transmission system. Both the
modulation and routing operations may be performed through nonlinear coupling such
as cross phase modulation, which has been shown to induce modulational instability
and induce pulse generation in the presence of anomalous dispersion. Anomalous
dispersion, gain and nonlinearity could be integrated into an Erbium laser, to produce
a CW train of control or signal pulses through intracavity modulational instability. The
feasibility of such a system has been demonstrated in the work presented here.
The use of high gain in a transmission system may lead to the evolution of
solitons from noise, which may be suppressed by the insertion of a natural 'noise eater
' or soliton filter based on the nonlinear loop mirror. Both gain and noise suppression
may be included within the device, which has been shown to allow transmission of
soliton pulses whilst reflecting non-soliton components, and clean up pedestals on
non-soliton pulses, suppressing inter-pulse radiation.
197
transmission over short distances over which collisions do not occur. This would find
application for local area network communication rather than transatlantic systems. At
ultra-high repetition rates, the most important limiting factor, apart from high order
nonlinear effects, would be gain saturation.
References
198
199
APPENDIX ULRASHORT PULSE MEASUREMENT:
AUTOCORRELATION.
The input signal was divided as close to 50/50 as possible with beamsplitter
(BS) to maximise the transmission through the interferometer. The pulses in one arm
were then delayed by adjusting the axial position of mirror (M2), and both signals
were then combined in the SHG crystal (LiIO3) via filter (F1) and lens (L). Filter (F1)
prevented stray ambient light around the second harmonic frequency (2ω) from
entering the detector.
200
Retro
BS
M SHG
L
ω
2ω
PMT
F2 F1
XTAL
M2
Background -free
(Non-collinear) Delay
IN
M1
BS
SHG
2ω+ω ω
PMT
F2 F1
XTAL M2
With Background
(Collinear)
Delay
IN
variation in delay. The PMT current was then displayed on the Y-channel of an
201
oscilloscope whilst a trigger signal in phase with the shaker oscillation was applied to
the timebase. The use of a digital storage oscilloscope allowed averaging to be made
over many integrations, greatly improving the signal to noise ratio in a number of
cases. In this way the real time autocorrelation was measured, facilitating optimisation
of the measured signal.
e1 t = Re E1 t e iωt
(A-1)
interacts with a nonlinear crystal capable of being phase matched for SHG, it
generates a pulse at the second harmonic frequency 2ω with envelope E2(t). The
optical field of the second harmonic pulse E2(t) can be shown to be of the form:
e2 t = Re E2 t e 2iωt (A-2)
α Re E 2 t e2 iωt
1 (A-3)
If the fundamental pulse is split into two and recombined coherently within the
SHG crystal with a relative delay τ imposed between the pulses, the total field for
equal intensity pulses incident on the crystal at frequency ω is:
202
etot t = Re E1 t eiωt + E 1 t-τ eiω t-τ (A-4)
= Re E1 t + E 1 t-τ e-iωτ eiωt (A-5)
Now, if
E t = E1 t + E1 t-τ e -i ωτ (A-6)
The second harmonic field emerging from the crystal is proportional to the
square of the complex amplitude of the fundamental, as shown in Eq(A-3). The second
harmonic envelope function E2(t) is therefore given by:
2 2
E2 t α Et = E1 t + E1 t-τ e -iωτ (A-8)
= E21 t + E21 t-τ e-2i ωτ + 2E1 t E1 t-τ e-i ωτ
(A-9)
If only the SHG signal at 2ω is allowed to pass into the PMT (using filter F2)
203
2 2
it α E2 t E*2 t = E1 t E*1 t + E1 t-τ E*1 t-τ
+ 4E1 t E*1 t E1 t-τ E*1 t-τ + S τ (A-10)
The term S(τ) contains harmonic functions in ωτ and 2ωτ, which tend to
average to zero by integration in the PMT as the delay is scanned, unless an extremely
slow trace is taken, resulting in an interferometric autocorrelation, which resolves the
interference fringes between the beams in a collinear geometry. Since the PMT cannot
resolve oscillations on a timescale of t, and averages these out to a DC level, terms in τ
can be collected in terms of the intensity I, given by EE*, where E is the field:
where I replaces the complex field amplitude multiplied by its complex conjugate.
iτ α 1+2G 2 τ (A-12)
where G(2)(τ) is given by:
I t I t-τ 2 2
G2 τ = = E1 t E1 t dt
I2 t -∞
(A-13)
G(2)(τ) represents the well known second order autocorrelation function, and is
non-zero when the pulse overlaps with itself, i.e. a photocurrent is enhanced when the
delay is adjusted such that the two pulses overlap temporally in the SHG crystal. The
pulse duration is related to the magnitude of the delay range Δτ over which an
enhancement in the SHG photocurrent occurs. Δτ was measured throughout the
experiments presented in this thesis by moving mirror (M2) with a motorised
204
translation stage, or with a shaker, and measuring the change in the photocurrent i(τ)
versus τ. For perfect relative overlap, i.e. τ=0, a coherent pulse of duration τ0 will
produce a PMT photocurrent i(0)=3 units, and for τ>>τ0, will produce a current
i(τ>>τ0)=1, since G(2)(τ=0)=1 and G(2)(τ>>τ0)=0 in Eq(A-12). This defines the
iτ α 2G 2 τ
(A-14)
205
Continuous Noise Noise Burst Coherent Pulse
Background free
(Noncollinear)
2
1 1
Δυ Δυ
3
1
(Collinear)
1
Δυ Δυ
2 Δτ
Δτ
1
0
Autocorrelation Delay (arbitrary units)
The first trace (left) is that of a continuous noise signal, with signal bandwidth
Δν, resulting in the generation of a coherence spike of duration Δν-1 on an infinitely
long background level. A semi-infinite noise burst gives a trace similar to that shown
in the middle. Note that the ratio of spike, pedestal and background intensities give a
good indication as to the nature of the signal when both collinear and non-collinear
autocorrelation are used. A fully coherent pulse yields a trace similar to that shown on
the right. The non-collinear geometry has no background level in this case, and is
hence a sensitive measure of proper mode locking in a laser, since inter-pulse radiation
is easily detected.
206
Table A-1. Deconvolution factors Δτ/ΔT and time bandwidth products ΔΤΔν for a
variety of common pulse shapes. (Top to bottom- Square, Gaussian, Sech2 and single-
sided exponential.)
The the factor Δτ/ΔΤ for determining the pulse FWHM duration ΔΤ from the
autocorrelation FWHM width Δτ is given in Table A-1 for a variety of commonly
encountered pulse shapes. The time bandwidth product ΔΤΔν of a Fourier transform
limited pulse of duration ΔΤ and bandwidth Δν is also shown for reference. The
-(ln2) t
exp 2 .110 throughout this
ΔT (t>0)
thesis was
directly shown to be ~<18fs5, limited by the bandpass of the SHG crystal, and any
spectrally narrowing or dispersive elements placed before it.
References.
207
JOURNAL PUBLICATIONS.
(11) Picosecond Pulse Generation From A Continuous Wave Diode Laser Through
208
Cross Phase Modulation In An Optical Fibre. E.J.Greer, D.M.Patrick,
P.G.J.Wigley and J.R.Taylor. Opt. Lett. 15(1990)
U U
(13) Mode Locking of a Continuous Wave Neodymium Doped Fibre Laser with a
Linear External Cavity. P.G.J.Wigley, P.M.W.French and J.R.Taylor.
Elec. Lett. 26(1990)
U U
(14) Active Mode Locking of an Erbium Doped Fibre Laser using an Intra-Cavity
Diode Laser Device. P.G.J.Wigley, A.V.Babushkin, J.I.Vukusic
and J.R.Taylor. Phot. Techn. Lett. 2(1990)
U U
CONFERENCE PUBLICATIONS.
209
(6) Optical Solitons. A.S.Gouveia-Neto, E.J.Greer, D.M.Patrick, P.G.J.Wigley
and J.R.Taylor. Proc. of the 2nd European Quantum Electronics Conference.
Dresden, E. Germany (1989)
(10) Picosecond Pulse Generation From A Continuous Wave Diode Laser Using
Cross Phase Modulation. E.J.Greer, D.M.Patrick, P.G.J.Wigley and
J.R.Taylor. Technical Digest of CLEO'90, Anaheim, California ,USA (1990)
(11) Mode Locking Of Solid State Lasers Using A Linear External Cavity.
P.M.W.French, S.M.J.Kelly, P.G.J.Wigley and J.R.Taylor.
Technical Digest of CLEO'90, Anaheim, California, USA (1990)
210
ACKNOWLEDGEMENTS
I would like to acknowledge the excellent supervision and guidance of Dr. Roy
Taylor throughout my Doctoral research, and for the critical reading of this
manuscript. For their encouragement and advice, I would like to thank Drs.Paul
French, and John Vukusic. I would also like to thank Dr.Stephen Kelly for many
useful discussions.
I would also like to thank Drs.Nick Doran, Keith Blow and Kevin Smith for
their assistance, and for allowing me to collaborate in their research at BTRL.
For their help in the lab, I would like to thank Dr. Arthur Gouveia-Neto, Elaine
Greer, David Patrick, and Nilay Pandit. I would also like to thank Elaine, Nilay and
Glenn Atkins for correcting my spelling.
The expert technical assistance of John Bean, Bob Wilkins, Dave Mountain,
Ted Bates, Charlie Cooper and Roy Morrison is also gratefully acknowledged.
The financial support for the work presented here was provided jointly by the
Science and Engineering Research Council and British Telecom Research
Laboratories.
211