Rep. Prog. Piiys. 58 (1995) 169-267.

Prinled in the UK

The generation of ultrashort laser pulses

P M W French
Femtosecond Oplin Group, Blackett Laboratory, Imperial College, Prince Consort Road. London, SW7 ZBZ, UK

Abstract This article is intended to provide an introduction to ultrafast laser development for scientists new to the field. It also aims to provide a snapshot of the state-of-the-art of ultrafast lasers and to indicate how this state evolved over the last thirty years. I n the first section, the main issues concerning ultrashort pulse generation are discussed and the most important ultrafast laser media are briefly reviewed. An extensive historical survey of mode-locking and pulse compression from 1964 until 1994 is then presented which covers the most important developments and aims to put recent advances and state-of-the-art femtosecond lasers in context. This review also anticipates future developments in practical ultrafast lasers for real-world applications. The basic techniques of mode-locking are then reviewed at a tutorial level. These include active mode-locking, passive mode-locking with real, resonant saturable absorbers and passive mode-locking with the optical Kerr effect. Emphasis is placed on ultrafast solid-state lasers, dye lasers and fibre lasers. Group velocity dispersion and self-phase modulation are introduced and their interaction discussed in some detail. Fibre-optic pulse compression is described and the significance of soliton shaping and solitary lasers is highlighted. Some of the phenomena limiting the minimum achievable duration of laser pulses are identified. The final section describes techniques for measuring ultrashort pulses including electrooptic streak cameras and second harmonic generation autocorrelation.

This review w a s received in August 1994.

0034-4885/95/O20169 t 96%59.50 Q 1995 IOP Publishing Ltd

169

I70

PM C V French

Contents

I. Introduction I , I , General remarks I .2. Issues in ultrashort pulse generation 1.3. Ultrafast laser media 2. Historical overview 2.1. Introduction 2.2. First generation: picosecond lasers 2.3. Second generation : femtosecond dye lasers 2.4. Third generation : femtosecond solid-state lasers 2.5. Fourth generalion: useful ullrafast lasers 3. Principles of mode-locking 3.1. Introduction 3.2. Active mode-locking 3.3. Passive mode-locking 3.4. Hybrid mode-locking 4. Role of group velocity dispersion and self-phase modulation in ultrafast lasers 4.I. Group velocity dispersion 4.2. Self phase modulation 4.3. Pulse compression using SPM and GVD 4.4. Soliton propagation and soliton shaping 5. Measurement of ultrashort light pulses 5.1. Introduction 5.2. Linear techniques 5.3. Second harmonic autocorrelation References

Page 171 171 172 174

180 180 181 183

188 194
198 198

200 206 223 224 224 234 235 237 245 245 245 246 252

The generation of uhashort laser pulses

171

1. Introduction

1.1, General remarks

This article is intended to review the physics and technology concerned with the generation of ultrashort laser pulses. It will hopefully also convey some of the excitement that the dramatic advances of the last few years have provoked within the laser community. The recent advent of tunable femtosecond lasers which are reliable and user-friendly has already had a significant impact on spectroscopic studies within diverse areas of physics, chemistry and biology. In particular, the development of room-temperature vibronic solid-state lasers and the exploitation of the optical Kerr effect for pulse shaping in ultrafast lasers have decoupled femtosecond pulses from dye lasers. One consequence of the increased convenience of the technology is that commercial ultrafast laser systems are now state-of-the-art and routinely outperform home-made research lasers for the first time. Progress contiuues to be extremely rapid and the keenly anticipated compact, diode-pumped, all-solid-state femtosecond lasers, tuning throughout the visible and near infra-red spectral regions, will be of enormous importance for metrology, medical diagnostics, communications, data storage and many other applications. A vital aspect of this modest miniaturization (from a few metres to tens of cm) will be that of a dramatic cost reduction-making state-of-the-art picosecond and femtosecond lasers viable for OEM applications. It is hoped that this will spur an explosion of technological innovations to complement the revolution in scientific research that is being driven by the ready availability of a -IO6 improvement in temporal resolution and the concomitant increase in peak optical power. The physics of compactjlow-cost lasers, however, does not differ significantly from that of their larger ancestors and most of this article will deal with the so called large-frame lasers. In the space available, it is not possible to provide a complete record of the progress made in ultrashort pulse generation, or even to provide a comprehensive list of references. This review is inevitably drawn from a personal perspective and will concentrate on developments and ideas which have been important to the authors work and understanding of the field. The central aim is to explain the state-of-the-art of ultrafast lasers, to show how this has evolved and to project current trends to the near future, This review is not intended to answer specialized questions but to provide a basic and hopefully intuitive introduction to an important, and relatively mature, research field which is increasingly being adopted by non-specialists as a means to understand ultrafast processes in many diverse branches of science and technology. Written by an experimentalist, this article includes only such mathematics as is necessary for the understanding of the basic tools of the trade. There are many excellent works covering theoretical aspects of every topic included in this review and some of these have been referenced for the benefit of those who wish to study a particular topic in detail. There are also many excellent text books which give a more thorough introduction to the basic topics covered here. References [1-4] are a personal selection. The discussion here will be limited to ultrashort pulse generation from continuous wave (cw) mode-locked lasers, with particular reference to dye lasers and vibronic solid-state lasers, since these are the most important for ultrashort pulse generation.

172

P M W French

Semiconductor lasers and fibre lasers will be mentioned in passing. Most ultrafast lasers are predominantly homogeneously broadened and no account here will be taken of inhomogeneous broadening or of the physics particular to specific laser gain media. Space constraints also preclude the possibility of discussing the wealth of ultrafast diagnostic and time-resolved spectroscopic techniques which have been developed. Some attention will be paid, however, to the standard ultrafast diagnostics of sffG autocorrelation and streak cameras since they are essential to any ultrafast laboratoty. The first section of this article introduces the essential issues of ultrashort pulse generation and some of the laser media which have been employed to this end. The second section is a historical overview of the evolution of ultrafast lasers. This assumes some knowledge of the field and a newcomer may be well advised to first read sections 3 and 4 which discuss the basic principles in more depth. Section 3 discusses the various techniques for mode-locking lasers and is intended to be comprehensible to a reader with an elementary knowledge of laser physics. Section 4 deals with the roles of group velocity dispersion and self-phase modulation which are crucial to the generation of femtosecond pulses. Finally, section S discusses the most common techniques for measuring ultrashort pulses.

1.2. Issues it1 ultrashort pulse generation

For the users of ultrashort pulses, the first issue is to find a suitable laser which provides radiation at the wavelength of interest. In the past this has mainly confined the applications to those spectral regions covered by laser dyes which may be excited by argon ion lasers, krypton ion lasers and the fundamental output of the Nd: Y A G laser and its harmonics. While this provided broad and relatively convenient coverage for the visible spectrum, the infra-red was less accessible due to the scarcity of suitable dyes and the difficulties associated with the toxicity and poor photochemical stability of those dyes that did exist. In the near infra-red, tunable ultrashort pulses could be obtained from cryogenic colour-centre lasers but this was not considered a convenient option by many researchers who relied on the Nd-doped solid-state lasers at 1.06 pm and 1.3 pm, The recent development of broadly tunable, vibronic solid-state media which operate at room temperature has transformed ultrafast laser physics, providing unprecedented spectral coverage and the last decade has seen tremendous advances in the technology. These media will be briefly discussed in section 1.3. Having established that a suitable laser exists, the next step is to generate the ultrashort pulses by mode-locking. The many techniques for mode-locking lasers are reviewed in section 3. Essentially they require some form of amplilude modulation applied to the laser radiation which has a period equal to the cavity round trip time. This modulation may be derived externally, as in the case of active mode-locking, or i t may be derived passively from the radiation itself via an intensity-dependent loss mechanism. There are alternative approaches to ultrashort pulse generation using gainswitching in which femtosecond pulses may be generated by pumping a cascade of short cavity lasers with ever shorter pump pulses e.g. .[SI.This works well with dye lasers which can exploit uv excimer pump lasers and achieve broad spectral coverage. In general, however, most users prefer cw lasers for their experiments which generally produce shorter and cleaner pulses because the pulse shaping takes place over many 100s to 1000s of round trips. Indeed, passively mode-locked cw lasers offer one of the highest peak-to-background ratios of any ultrafast signal that can be synthesized. The precise mode-locking technique employed will depend on the characteristics of the laser

The generution of ultrashort laser pulses

173

media. the desired pulse duration and practical considerations such as cost, the stability and pulse quality requirements and the need for synchronization to other signals. Requirements for pulse durations vary with application : many electronic detection systems are limited to temporal resolutions of tens of ps while pump/probe measurements can exploit pulses shorter than 1Ofs. In general, it is not significantly easier to generate picosecond pulses than femtosecond pulses if high temporal and amplitude stability is required. A picosecond laser is not simply a bad femtosecond laser and, in fact, a conventional actively mode-locked solid-state laser would probably have a higher component cost than a state-of-the-art Ti: sapphire laser generating 20 fs pulses. Once the desired pulse duration and wavelength have been determined, the first consideration is the bandwidth of the system. By the uncertainty principle, an ultrashort pulse in the time domain must have a correspondingly broad spectrum in the frequency domain. It is therefore necessary to establish that the gain medium has a sufficiently broad linewidth to amplify the ultrashort pulses and that the other components (mirrors, modulators etc) are sufficiently broadband in their reflection/transmission responses to transmit the pulses without filtering the pulse spectrum. In particular, if the laser is to be spectrally tuned using an element such as a prism or a birefringent (Lyot) filter, then this filter must have a sufficiently broad pass-band. Note that where picosecond pulses are required, it is often necessary to restrict the bandwidth of the laser in order to discriminate against noise. This is the case when a relatively weak mode-locking technique such as synchronous pumping is employed. To generate pulses much shorter than a picosecond, it is necessary to take account of group velocity dispersion (cvD)-the phenomenon which causes different frequency components to travel at different speeds, thereby broadening the pulses as they circulate in the laser cavity (see section 4). The spectral width of the pulses is inversely proportional to the temporal duration and so as pulses become shorter, they become more susceptible to CVD. For picosecond lasers, CVD is usually neglected although it is optimized in the latest picosecond Ti :sapphire lasers which pennit the pulse duration to be continuously adjusted up to -100 ps. Clearly excessive CVD will prevent the generation of short pulses and so it should be minimized. This is usually achieved by including components in the laser cavity which provide adjustable CVD of the opposite sign to the laser medium and other components. Thus the net cavity round-trip CVD experienced by the pulse may be arbitrarily set to zero or some optimum value. A further property of ultrashort pulses is that they tend to exhibit high peak powers. Typically the average power from a cw laser is not greatly reduced by the mode-locking process. Most lasers deliver -100 mW average output power, an order of magnitude, although there are obviously exceptions. With an output coupler of a few YO, one can estimate average intracavity power levels of -1-100 W. There is no fundamental reason why mode-locking should dramatically reduce the average power from the laser although the insertion loss of the components required to achieve mode-locking will usually entail some penalty. Suppose that the laser is mode-locked such that it generates pulses of 1 ps duration at a pulse repetition rate of 100 MHz-corresponding to a 10 ns cavity round-trip time. The intracavity pulse energy will therefore be -10 nJ-I pJ and the peak powers will be IO kW-l MW. For most lasers, cavity radiation is focused to a beam waist of tens of p m in the gain medium in order to achieve a sufficiently high population inversion. Thus peak intensities reach levels of 1O8-1Oi0 W cm-* in a beam of 100 p m diameter. At such high intensities, which are actually quite modest estimates, the refractive index of the intracavity media can become non-linear and this gives rise to se@+liu.w modulation (SPM) due to the optical Kerr effect. SPM causes the pulse

174

P M Mi French

spectrum to broaden which can increase the loss at any spectral filter in the laser. It can also, together with GVD, cause the pulses to either broaden or compress and it can cause the laser to become unstable. This interaction is discussed in section 4. In practice, to generate femtosecond pulses, first GVD must be reduced so that ultrashort pulses can propagate in the laser cavity. Secondly, the interaction of SPM and GVD must be exploited in such a way as to obtain shorter, stable pulses. The ability to control this interaction determines how short a pulse can be generated. The optical Kerr effect may also be exploited to achieve mode-locking. Either temporal SPM or self-focusing may be used to generate an intensity-dependent loss for the intracavity radiation which results in powerful pulse shaping. This technique is possible when the intracavity power levels are high enough to access the non-linear refractive index of the laser medium. The development of high average power ultrafast solid-state lasers has demonstrated the potential of this approach. Its major drawback is that it does require high intensities to work and this means that lasers mode-locked in this manner are often not scr-starting, i.e. they do not become modelocked as soon as they are switched on. The physics underlying the self-starting of lasers is not yet fully understood and the self-starting issue remains one of the most serious impediments to the widespread development of ultrafast lasers. In practice, many laser systems use a 'conventional' self-starting mode-locking technique to generate picosecond pulses and then use the optical Kerr effect to compress the pulses to the femtosecond regime. The final duration is then determined by the interaction of SPM and CVD. The discussion of applications of ultrashort pulses is beyond the scope of this article but I will make a few observations in this section. First, the most important scientific application of ultrashort pulses is time-resolved spectroscopy. Widely tunable femtosecond lasers permit time-resolved studies of many physical, chemical and biological processes. Pulses of a few femtoseconds duration enable molecular dynamics to be resolved. Femtochemistry in particular is growing into a huge field and there appears to be a real chance of directly controlling chemical reactions using appropriate ultrafast optical signals. Solid-state physics can be studied with sufficient resolution to resolve electron dynamics in semiconductors and there is much current investigation of ultrafast light-matter interactions. I t is now possible to generate pulses shorter than the homogeneous dephasing times of many systems and coherent phenomena are routinely studied. In particle accelerators, femtosecond pulses are being used in photo-injectors to generate extremely short bursts of electrons. Potential real-world applications of high lime resolution include ultrahigh bit-rate optical signal processing and communications, high speed electronics and time-gated imaging through turbid media which may find widespread medical application. The ability to coherently extract large amounts of stored energy from laser amplifiers in a very short time using femtosecond oscillators and amplifiers has produced relatively compact (table-top) laser systems delivering peak powers in the tens of TW range. These have been applied to multiphoton and atomic physics experiments and have been used to generate extremely bright bursts of x-rays, At more modest levels, powerful ultrashort pulses have been used to study a wide variety of non-linear optical effects. The list of applications is vast and the number of users of ultrafast laser technology is increasing rapidly. This article is intended to illustrate some of the physics underlying the technology and will perhaps assist the growth of the ultrafast community.
1.3. Ulmfusr laser media

1.3. I . Fi\.ed-wauclengfh lasers. The first laser to be mode-locked was the helium-neon laser [6] but the extremely narrow gain-linewidth only permitted the generation of ns

The generation of ultrashort laser pulses

175

pulses. Mode-locking achieved picosecond pulse durations in a number of broader linewidth fixed-wavelength laser media including the ruby laser at 694 nm, the argon ion laser at 514 nm, the krypton ion laser at 647 nm, Nd:doped YAG and glass lasers at 1.064 pm and Er:doped glass at 1.55pm. The electronic transitions in Nd:uac and ruby are homogeneously broadened due to collisions of the ions with the crystal lattice phonons. Gas lasers exhibit electronic transitions which are predominantly inhomogeneously broadened due to Doppler broadening. The Nd:glass and Er:glass lasers, which exhibit the broadest gain linewidth of the common fixed-wavelength lasers and have yielded the shortest pulses, are predominantly inhomogeneously broadened due to local inhomogeneities in the glass host. The fixed-wavelength solid-state lasers tend to have relatively long upper state lifetimes ( p s to ms) and low gain cross sections. This leads to a high energy storage capacity, making them useful as laser amplifiers and efficient laser oscillators, but it also means that they are difficult to saturate and this has an impact on the mode-locking techniques which may be applied L o them. It also means that they have a strong tendency to Q-switch when there is a saturable absorber in the cavity. Noble ion gas lasers have a linewidth sufficient to support pulses of -50 ps duration but typically deliver pulses >IO0 ps when actively mode-locked. They are useful as pump sources for synchronously mode-locked dye lasers and, like other fixed-wavelength lasers, more often provide the c w laser pump source for tunable ultrafast lasers. I n its own right, however, the arc-lamp-pumped N ~ : Y A is G one of the most widely used ultrafast lasers, typically delivering -100 ps pulses with of the order of tens of W average output power. In recent years it has been partly supplanted by N ~ : Y Lwhich F generates pulses of -30 ps duration with comparable power levels. Semiconductor diode-pumped Nd: doped lasers have now generated rather shorter pulses and have provided the first all-solid-state, compact ultrafast lasers. Diode-pumped Nd: YAG and Nd: VLF lasers have generated pulses as short as 8.5 ps [7] and 6 ps [SI respectively using the technique of Kerr lens mode-locking. An arc-lamp-pumped Nd: YLF laser has generated pulses as short as 2.3 ps [9]. Laser-pumped and diode-pumped Nd:glass lasers have generated subpicosecond pulses in many configurations and the shortest pulses of -32 fs were obtained from a krypton ion laser-pumped Nd:glass fibre laser [IO]. Er: glass fibre lasers have received considerable attention as candidates for ultrafast optical telecommunications sources and diode-pumped erbium fibre lasers have generated pulses as short as -100 fs [ 1 I].
1.3.2. Dye lasers. The first tunable ultrafast lasers were c w dye lasers and the passively mode-locked dye laser held the record for the shortest pulses for many years. The shortest pulses ever generated, of 6 fs duration, were obtained from a fibre-optically compressed dye laser [12]. Since its demonstration in 1966 by Sorokin and Lankard [ 131, the dye laser has been one of the most versatile and most widely employed laser sources. Organic dyes are laser media whose absorption bands in the ultraviolet, visible or near infrared spectrum result from the presence ofconjugated bonds in the molecular structure with associated delocalized x-electrons. Laser transitions occur between the manifold of vibrational (and rotational) energy levels associated with the electronic energy levels. This type of laser action is described as vibronic (or vibrational-electronic). The ground state singlet electronic level (So) is separated from the singlet and triplet excited levels (SI, T I ) by IO 000-30 000 cm-I. There are vibrational and rotational levels associated with each electronic level which are separated by -1000cm-' and -10 cm-' respectively. Interaction with the solvent of the dye produces collisional and electrostatic perturbations that homogeneously broaden these levels resulting in

I16

P M M French

electronic levels which are essentially continua of states. I t is the range of possible transitions hetween these continua that provide the broad absorption and emission hands of laser dyes. Much work has been done selecting and engineering dye molecules to shift the central transition frequency from the ultraviolet to the near infra-red. Today the broad emission bands of organic laser dyes provide spectral coverage from -320 [I41 to -1800 nm [15]. All laser dyes tend to have high gain cross sections and short upper state lifetimes (-11s). They are the most easily saturable laser media and this has dominated the development of the mode-locking techniques with which they have produced ultrafast lasers. At first glance, their high gain cross sections, large homogeneously broadened linewidths and cost effectiveness made dyes almost the ideal ultrafast laser medium. Unfortunately they have a limited lifetime due to photochemical degradation, are inconvenient to handle and are often toxic and carcinogenic. There are problems associated with triplet-quenching in which the long-lived triplet state traps excited molecules and so inhibits laser action. This is avoided in c w lasers by flowing the dye solution rapidly through the pumped active volume in a jet stream and sometimes by the addition of chemical triplet-quenching agents such as COT (cycloctatetraene) or oxygen to the dye solution. For many applications dye lasers are being replaced by broadly tunable vibronic solid-state lasers such as titanium-doped sapphire but there remain many ultrafast dye lasers in operation and they provide useful coverage of the visible spectrum where there are no broadly tunable solid-state lasers. Although the first mode-locked dye lasers were flashlamp pumped [16], it was the c w dye laser which became the standard laboratory ultrafast laser and on which much of femtosecond laser physics was learned. The tuning range o f c w dye lasers is limited by the available pump sources. Argon ion lasers are able to provide c w pump radiation from wavelengths as short as 275 nm and the various emission lines of krypton and argon ion lasers provide pump power at wavelengths up to -800 nm, as indicated in table I . Nd :YAG lasers provide
Tablc 1. Common pump sources for

cw lasers
Available pump power (W) 25 2
10

Pump laser Argon ion laser

Pump wavelength (nm) 476-514 528.7 514.5 501.7 496.5 488.0 476.5 752.2-799.3 647.1-676.4 520.8-568.2 -670 (1GaAlP) -980 (InGaAs) I064 532

Pump wavelength
(nm)

Available Pump power

W)
1.2 0.75
I.4

1.5 2.5 8
2.8

472.7 465.8 457.9 454.5 351,1-385.8 333.6-363.8 275.4-305.5 468.0-530.9 406.7422.6 337.5-365.4
-800-900

I .9 3 7 0.6

Krypton ion laser

I.6 4.6 3.6

1 . 5 1.3 2
-10
. I

Laser diodes

- 10
- 3 22 3

(GaAIAs)
1480
( InGdAsP)

Nd:vhc laser and harmonics

355 266

2 I

The generation of ultraslzorr laser pulses

177

useful pump power at 1.064 pin and I .3 pin although there are not many infra-red dyes which are sufficiently srnble for cw operation. Reference [ I71 provides a comprehensive listing of laser dyes, their properties and references on previous work. The physics of dyes and dye lasers are discussed further in [IS] and [ 191.
1.3.3. Tunable solid-stnte lasers. The Ti: sapphire laser has become almost synonymous with femtosecond laser technology and it, together with other room-temperature vibronic laser media, looks set to replace ultrafast dye lasers in almost all applications. Table 1 gives a list of potential pump sources for c w vibronic solid-state lasers. Such lasers offer all the convenience of solid-state laser media, they have long (-,us) lifetimes and gain linewidths which exceed those of dyes by almost an order of magnitude. The broad absorption/emission linewidths arise as a consequence of the strong coupling between the vibrational energy states of the host crystals and the electronic energy states of the active ions-hence the label vibronic solid-state lasers. This strong coupling between vibrational and electronic energy states depends critically on the precise interaction between the active ion and host crystal fields. Vibronic lasers are usually transition earth meld1 ions, such as Ti3, Cr and Cr4+doped into suitable hosts. The transition metal ions are appropriate since their electronic structure is that of argon with the valence electrons filling the 3d shell outside the inert 3s2, 3p6 configuration. Thus these electrons are exposed and can interact strongly with the host crystal field. In contrast, the fixed-wavelength lasers tend to be rare earth ions (e.g. Nd?, Er3+etc) with the valence electrons filling the 4f shell inside the 5s2, 5p6 shells. In this case, the electrons responsible for the laser transition are shielded from the host crystal field by the inert gas structure. Consequently there is no significant interaction and the transitions tend to be narrow. Note that the transition metal ions do not always form vibronic lasers in crystals. Cr+:sapphire is ruby, a narrow linewidth, fixed wavelength laser medium, while c r : L i s ~ ~provides a vibronic tunable laser. The difference arises because the local crystal field in LisnF is stronger than in sapphire, Even in ruby, however, the absorption bands are broad as a consequence of the coupling between vibrational and electronic energy states. The physics of solid-state lasers is discussed at length in [20-221. Ti:sapphire was first lased in 1982 by Moulton and subsequently characterized [23] as a promising tunable laser medium with an upper state lifetime of 3 ps and one of the highest gain cross sections of any solid-state laser. It tunes from -680-1 130 iimthe widest percentage tunability of any laser. With improvements in the crystal quality, argon-ion laser-pumped cw. Ti: sapphire lasers deliver several W of average output power and have generated pulses as short as -10 fs. The design of such lasers is rather similar to cw dye lasers with the dye jet stream being replaced by the Ti:sapphire laser rod and the folding angle at the focusing mirrors being adjusted to compensate for astigmatism e.g. [24]. Typically the pump and laser radiation are focused to <50pm beam waist over the 2cm rod length and the non-linearity induced by the high intracavity power over this relatively long interaction length (compared to that found in dye lasers) is largely responsible for the mode-locking techniques which were developed on Ti :sapphire lasers and which have now been applied to most other solid-state lasers. The principle disadvantage of Ti:sapphire is that its absorption bands are not suitable for pumping directly with flashlamps, arc lamps or laser diodes. The success achieved with Ti:sapphire has prompted researchers to look for other vibronic laser media. Many Cr-doped crystals have been developed (or rediscovered) to provide tunable c w lasers in the near infra-red. Cr+: LiSAF, Cr+ :LiCAF and similar

178

PM M Fwiich

crystals are promising materials which tune over the region -750-1000 nm and which have absorption bands suitable for pumping with diode lasers at 670 nm and with cotiventional flashlamps and perhaps arc lamps. They have an upper state lifetime -7Ops and so are well suited as laser amplifiers and low threshold diode-pumped oscillators. Pulses as short as 33 fs have been generated from an argon-ion pumped Cr3+:LisAF laser [25] and already the first diode-pumped femtosecond Cr: LisAF lasers have been demonstrated [25]. In thenear infra-red, Cr: Forsterite and Cr4t :YAG provide coverage from 1.13-1.37pm and 1.35-1.58pm respectively. Kerr lens modelocking has yielded pulses as short as 25 fs from Cr4t : Forsterite [27] and pulses as short as 70 fs have been generated from YAG lasers (281. These Cr4t: doped lasers are readily pumped by cw Nd :YAG lasers at 1.06 pm and may be directly diode-pumped into the same absorption band e.g. using pump laser diodes at 980 nm. Their upper state lifetimes and gain cross sections are comparable l o Ti:sapphire and so one can expect similar performance from the cw lasers. Up to now, the crystal quality is inferior to Ti :sapphire and there are problems with thermal lensing and excited stale absorption. Nevertheless, these lasers can deliver -1 W average output power for -10 W pump

SH
8H
1W

SH SH

C+*:ForstterIte

err
1200
1400
1600

C = F
n:sapphire

400

GOO

800

TOO0

1800

Wavelength (nm)

Figure 1.

Tuning range of roomtemperature c w vibronic solid-state lasers.

power. Figure I shows the tuning range of just four room-temperature cw vibronic solid-state lasers. It will be seen that with second harmonic generation, most of the visible and near infra-red spectrum is covered up to -1.6 pm. There is still a small gap around 600 nm which is, ironically, where (ultrafast) c w dye lasers perform best. It is likely that this region will be covered in the near future, perhaps by Pr+-doped lasers. There are a number of other tunable vibronic lasers which operate at cryogenic temperatures. These include divalent transition metal lasers such as NiMg0 and CO :MgF2 and alkali-halide colour centre lasers. The practical difficulties of working at 77 K have precluded the widespread application of these lasers but they have been demonstrated as c w mode-locked systems. The former transition metal lasers can be pumped at 1.3 pm by a Nd:unc laser and tune from -1.6-1.73 pm and 1.62-2.1 p m respectively. They have been actively mode-locked to yield pulses of tens of ps 1291. Colour-centre lasers provide spectral coverage form -800 nm to -3 pm and may operate either c w or pulsed. The energy levels ofthese media are associated with defects in alkali-halide crystals. These are vacancies in the crystals which may be filled by either electrons or small alkali ion impurities. Some common colour-centre active media are NaCI, NaF, NaCl:OH-, KCI:Li, KCI:Na and KCI:TI. Reference [30] provides a useful review of colour-centre lasers. They have received a considerable amount of attention as ultrafast lasers and the NaC1:OH- and KCI:TI lasers in particular have

The generation o f ultrashort laser pulses

1I9

been used to generate femtosecond lasers tunable through the telecommunicationswindow around 1.5Spm e.g. [31]. The upper-state lifetimes and gain cross sections of colour-centre lasers vary considerably from values which are similar to dyes (tens of ns) to parameters which are more like transition metal lasers (e.g. 1.6 p s for KC1:TI). As well as requiring cryogenic operating temperatures, c w colour-centre lasers also suffer from photo-degradation and these practical inconveniences mean that they will be replaced by either vibronic room-temperature solid-state lasers or optical parametric oscillators. 1.3.4. Semiconductor lasers. Semiconductor lasers exhibit linewidths of the order of -1Onm and generated pulses as short as a few 100 fs duration. They are clearly a strong candidate for compact ultrafast laser sources and will almost certainly find realworld applications as such. Ultrafast semiconductor laser physics is a huge field in itself and semiconductor lasers are outside the scope of this article although they will be mentioned in passing. Their upper state lifetime and gain cross sections are comparable to those of dyes and so gain (and absorption) saturation plays a strong role in their mode-locking. In general, although it is possible to generate -600 fs pulses from monolithic semiconductor structures e.g. [32], femtosecond pulse generation requires the semiconductor amplifier to be used in an external cavity. One reason for this is that they exhibit high non-linearity when interacting with ultrashort pulses and attempts at generating ultrashort pulses by directly mode-locking semiconductor lasers are usually complicated by unwanted phase modulation which arises from the carrier dynamics of gain switching and depletion. External cavities permit the incorporation of appropriate chirp compensation devices. They also permit the cavity length to be extended such that an external K F drive signal may be synchronously applied for active mode-locking. The shortest pulses to be directly generated from an actively mode-locked semiconductor laser are 580 fs [33]. External compensation of the frequency modulation has generated pulses as short as 200 fs [34] from a hybridly mode-locked semiconductor laser which was externally amplified and compressed with an appropriate dispersive delay line. Reference [34] gives a review of ultrashort pulse generation from semiconductor lasers. In practice the degree of complexity required to generate clean ultrashort pulses from semiconductor lasers makes them no more simple than a diode-pumped solidstate laser. It is very difficult to simultaneously optimize all the parameters of semiconductor diode lasers. Although diode lasers and diode arrays can deliver tens or even hundreds of watts of average output power, it is not generally possible to achieve a high quality spatial and temporal profile at the same time. One reasonably successful technique to achieve good beam quality is to use the tapered amplifier geometry in a master oscillator-power amplifier (MOPA) scheme which can provide high output powers in a diffraction-limited beam [35]. An all-semiconductor device can be constructed by injection-locking the power amplifier with a low power, diffraction-limited ultrafast diode laser. An alternative scheme, which can provide comparable output powers, better beam qudity and rather shorter pulses for a similar level of complexity, is to use diode lasers or diode arrays to pump a mode-locked (vibronic) solid-state laser. This approach is being applied to Cr: LisAF, as discussed in the previous section. I t has already been applied to Nd: YAG and Nd: YLF lasers for picosecond pulse generation and to Nd: glass and Er:glass fibre lasers for both picosecond and femtosecond operation. Table I indicates the wavelengths at which high power diode laser pump sources are available. The rapid development of new vibronic solid-state media and ingenious pumping schemes make this approach extremely promising.

I80

P rM W French

2. I-listorical overview
2. I . In/rod~ictio~i

The first laser to be mode-locked was a helium-neon laser in 1964 [36]. Nanosecond pulses were generated using an infracavity loss modulator and this was explained as active mode-locking. Subsequently many new mode-locking techniques have been demonstrated and pulse durations have been reduced to the the -10 fs regime. There are many excellent reviews of mode-locking which cover the historical development in more depth than I shall attempt here. In particular, Smith has provided a useful review of the early research on mode-locking from a 1970 perspective [37], New has described the development until 1983 [38] and Krausz e/ a1 have summarized the important advances of the last decade [39], It is not possible to reference here all the significant papers in this field or to acknowledge all the contributions to its development. This is a personal overview and I have cited those papers which have come to my attention, trying particularly to include those related to significant developments in the field. Over three decades, the development of ultrafast lasers has been described in terms of three generations [39]which have been closely related to the breakthroughs in the laser media and appropriate pumping schemes. The first generation of mode-locked lasers were mostly lamp-pumped solid-state lasers, dye lasers or gas lasers. They were mode-locked using generic active or passive techniques to generate pulses of tens of picoseconds duration. Active modulation techniques alone were shown to be unable to mode-lock the whole of the available gain linewidths and the pulse durations achieved by passive mode-locking were limited to the recovery time of the resonni?/ saturable absorbers. Furthermore, the fact that the flashlamp pump pulses were limited to a few 100 {is duration nieant that there were only a limited number of cavity round-trips on which the mode-locking mechanism could act on the pulses. This issue was addressed for the second generation of ultrafast lasers which was engendered by the demonstration of the c w dye laser [40]. This quickly led to the c w mode-locked dye laser [4l J which could also take advantage of the strong gain/absorption saturation of dyes to achieve more powerful pulse compression than was hitherto possible. Essentially, by using two separate mechanisms to shape the leading and trailing edges of the pulses, they achieved durations shorter than the recovery times of the resonant non-linearities they exploited. The resulting lasers generated subpicosecond pulses and it was then necessary to take group velocity dispersion (GVD)into account. This led to a greatly improved understanding of the interactioii of GVD and non-linear frequency chirp and resulted in the generation of sub-30 fs pulses whose duration was limited by the gain linewidths of the dyes. Extracavity pulse compression yielded sub-10 fs pulse durations. The third generation was ushered in by the demonstration of the soliton laser [42] which showed that ~zo~z-reso~un~ mode-locking techniques based on the optical Kerr effect could be used to produce femtosecond pulses. A parallel and equally important development was the emergence of 'Ti:sapphire as a broadly tunable vibronic solid-state laser medium which did not exhibit strong gain saturation under cw pumping. This laser required the non-resonant mode-locking techniques and prompted the intensive research which has produced widely tunable ultrafast lasers generating sub-100 fs pulses over a broad spectral range. One is templed to say that the development of mode-locking techniques is almost at an end since we have the means to mode-lock almost any laser medium, given sufficient intracavity power. Certainly a major part of ultrafast laser development in the near future will be to develop new broadband gain media and apply existing

The generation of ultrashort laser pulse's

181

mode-locking techniques. J envisage the fourth generation of ultrafast lasers to be compact practical devices appropriate to rea-world applications. These lasers will almost certainly be diode-pumped and they will be versatile in terms of pulse repetition rate and energy. With the help of non-linear frequency shifting techniques such as harmonic generation and optical parametric amplification, they will provide a broad spectral coverage. Finally, they will hopefully be designed to be robust and low-cost such that they will become standard instrumentation in laboratories and applications sites.

2.2. First generation: picosecond lmem

After the experimental observation of the mode-locking of the He-Ne laser, theoretical descriptions of mode-locking were published e.g. [43,44]. FM modulation was used to mode-lock a He-Ne I451 soon after and active mode-locking techniques were applied to argon ion [46] and ruby [47] lasers. In 1965, the use of a fast saturable absorber dye to passively mode-lock a laser was demonstrated with ruby [48] and was quickly applied to Nd :glass lasers [49] whose inhomogeneously broadened linewidth could potentially support subpicosecond pulses. Unfortunately, the poor thermal properties of the glass host precluded c w operation (cf the arc-lamp-pumped N ~ : Y A laser) G for most ultrafast experiments and investigations were limited to @switched, mode-locked pulse trains of uncertain reproducibility at 1.06 pm. A great improvement in tunability came with the invention of the dye laser which had a huge impact of ultrafast laser development. While still operating in pulsed mode, the spectral coverage of such lasers meant that they became significantly more useful research tools which in tum increased the motivation to improve ultrafast laser technology. Flashlamp pumped dye lasers were passively mode-locked using fast saturable absorber dyes [50] in 1969 and yielded pulses as short as 6 ps [51]. Subsequent investigations of new laser amplifier/saturable absorber combinations extended the tunability of ultrafast Rashlamp pumped dye lasers from the blue to the near infra-red [38,521. The demonstration of mode-locking by synchronous pumping of a dye laser [53-55) using a frequency-doubled Nd :glass laser more o r less completed the range of mode-locking techniques available to the first generation of ultrafast lasers. The technique of passive mode-locking was especially successful and this possibility to generate such short pulses prompted the invention of new techniques for ultrashort pulse measurement such as second harmonic autocorrelation [56] (which is discussed in section 5 ) and two-photon fluorescence (TPF) [57]. The minimum pulse duration achievable with a fast saturable absorber is limited to its absorption recovery time [58] for which the lower limit is -10 ps (e.g. for the Kodak dyes A9740 or ,49860). Initially there were many over-optimistic interpretations of pulse width measurements from solid-state lasers due to the fact that noise bursts give similar TPF traces to pulses with comparable spectral widths. Understanding the correct contrast ratios for TPF profiles o f mode-locked pulses [59] led to more consistent experimental reports and usually the pulse durations were rather longer than the minimum possible values determined by the gain linewidth and the saturable absorption recovery times. Streak cameras were also employed to measure pulsewidths and to follow the dynamics of ultrafast lasers e.g. [60]. One problem with passive mode-locking was that the saturable absorber dye also Q-switched the lasers and terminated laser action before a steady-state could be established, even for cw pumping. A second problem was that the high pulse energies

182

P M M French

(and correspondingly high intensities) produced strong non-linear self-phase modulation of the pulses which, together with GVD, led to spectral and temporal broadening. I t was shown that for passively mode-locked flashlamp pumped lasers, the mode-locking was non-deterministic with the pulses growing from stochastic fluctuations in the cavity radiation e.g. [61,62]. To perfom reasonably reproducible experiments with pulses of constant energy, it was necessary io switch a single pulse out of the evolving Q-switched mode-locked pulse train, either by cavity dumping the laser during the evolution of the pulse train e.g. [63], or by selecting and switching out a pulse from the laser output [64]. A relatively recent alternative approach to the operation of pulsed ultrafast lasers has been to include a negative feedback device in the cavity which functions like an inverse saturable absorber (i.e. the loss increases with intensity). This frustrates the Qswitch and prolongs the mode-locked pulse train for the duration of the pump excitation e.g. [65], thus providing many more cavity round-trips for the mode-locking process to act on the pulses and achieving durations close to the minimum possible. Active mode-locking via amplitude or frequency modulation was applied to pulsepumped and cw lasers. In particular, lhe actively mode-locked Nd:uno laser [66] received much attention since the possibility of cw pumping (using lungsten filament lamps and arc lamps) furnished a relatively efficient laser with high average power, delivering pulses of -30 ps estimated duration [67]. A theoretical descriplion of active mode-locking using A M and FM, together wilh an experimental investigation of the Nd:uno laser was published in the seminal papers by Kuizenga and Siegman [68] which predicted the achievable pulse duration in an actively mode-locked laser as a function of the modulation and laser parameters. Their model worked reasonably well for the narrow linewidth gain media such as N ~ : Y A G bu! was less successful with broadband gain media. Experimentally, it proved difficult to generate pulses much shorter than -100 ps with arc-lamp-pumped Nd:unc lasers and this became a standard F of a very commercial ultrafast laser. Recently actively mode-locked N ~ : Y Llasers similar design have generated pulses of -30 ps due to the broader gain linewidth. During the early research on mode-locking, the phenomenon of spontaneous modelocking or self-mode-lockingwas observed in He-Ne lasers [69] and discussed in terms of the non-linear behaviour of the laser medium. It was also observed in ruby [70], Nd:glass [71], argon ion [72] and other lasers. Self-locking was analysed in terms of frequency pulling of combination tones (beat frequencies) of oscillating modes to other mode frequencies e.g. 1731 or in the time domain in terms of efficient interaction of socalled n-pulseswith the population inversion [74]. Spontaneous mode-locking did not prove a reliable approach for the generation of ultrashort pulses and was more or less abandoned in favour of actively or passively mode-locked c w lasers but the ideas are once again important in the effort lo understand the start-up of the third generation femtosecond solid-state lasers which are often described as self-mode-locked. The role of non-linear frequency chirp was also studied in first generation ultrafast lasers as a means to passive FM mode-locking [75, 761 and as a source of instability e.g. [77]. Researchers used the time-bandwidth product (see section 3) as a means of assessing the quality of the mode-locking and, particularly for solid-state lasers, often measured values far in excess ofthe theoretical minimum. To explain this, some authors considered linear frequency chirp arising from dispersion in the laser medium e.g. [78,79], but for the case of passively mode-locked Nd:glass lasers, it became clear it was due to non-linear self-phase modulation (SPM) arising from the optical Kerr effect [so, 811. Techniques for compressing chirped optical pulses date back to the interferometer proposed by Gires and Tournois in 1964 [82] and compression of pulses after

The generation o f ultrashort laser pulses

183

active phase modulation were proposed [83] and demonstrated [84]. In 1968 Treacy demonstrated that pulses from a passively mode-locked Nd:glass laser could be compressed by an appropriate dispersive delay line [U]. He proposed the use of diffraction gratings [86] and this, combined with the use of a non-linear medium (CS2) to chirp pulses for exlracavify compression [87], essentially laid the foundations for the srM/ G V D pulse compressor which was demonstrated to compress 20ps pulses from a Nd:glass laser to -2 ps [SS]. One further idea to exploit the optical Kerr effect was to use self-focusing for pulse compression [89], thereby anticipating the invention of Kerr lens mode-locking and todays state-of-the-art femtosecond lasers. 2.3. Second generation: femtosecond dye lasers The development of the cw dye laser was an important breakthrough for ultrafast lasers since it provided a tunable broadband laser in which the various pulse-shaping techniques of both active and passive mode-locking could act on the circulating pulses with cumulative effect until a steady-state was reached. As well as greatly increasing the effectiveness of the mode-locking techniques, the c w mode-locked laser was also a reproducible source of high quality pulses which were reasonably straightforward to characterize and to apply to ultrafast measurements. c w colour centre laser also provided much improved performance through passive mode-locking with fast semiconductor saturable absorbers whose recovery times were subpicosecond e.g. [90]. Semiconductor lasers were also passively mode-locked using fast semiconductor saturable absorbers to generate pulses of 1.6 ps duration [9l]. The pulses from such lasers were observed to be chirped and subsequent compression yielded durations of 0.83 ps [92]. Actively ]node-locked semiconductor lasers generated pulses as short as 580 fs 1931. Most of the research effort directed towards femtosecond pulse generation, however, concentrated on mode-locking c w dye lasers which provided the first widespread ultrafast laser technology. The two most powerful approaches were synchronous pumping and passive mode-locking with a slow saturable absorber (see section 3). The same techniques were also applied to colour-centre lasers and semiconductor lasers with less success but it was from these systems that the techniques for the third generation of femtosecond solid-state lasers evolved. As well as the general references already cited, [94] provides an excellent review of femtosecond dye lasers. The first c w mode-locked dye laser was domonstrated in 1972 by Ippen el a1 [95]. This was a Rhodamine 6G dye laser mode-locked with a cell of the saturable absorber dye, DODCI, in the cavity. Pulses as short as 1.5 ps were generated which was surprising since the recovery time of the absorber was several hundred picoseconds. A theory to explain this result, and anomalously short pulses obtained from passively mode-locked flashlamp-pumped dye layers, was presented by New [96,97] who explained how the combined action of saturable absorption and saturable gain could generate much shorter pulses than the recovery times of either the gain or absorption. Rapid progress was made experimentally with the introduction of free-flowing dye jet streams, rather than flow cells [98], and the first subpicosecond pulses were generated by Shank and Ippen using a mixture of amplifier and absorber dye flowing in a single jet stream [99]. This laser also included an acousto-optic cavity-dumper and thereby produced pulses of 60.5 ps duration with -4 kW peak power. I n a later experiment with separate amplifier and absorber dye jets, the same authors used a diffraction grating pair to compress chirped output pulses to 300 fs [ 1001. At this time, Haus published a theory of mode-locking with a slow saturable absorber which built on the work of New and

I84

P M W French

derived a closed form solution for the pulse shape which was shown to be a hyperbolic secant [loll. The experimental dye laser was further refined to incorporate colliding pulse mode-locking (CPM). In the first implementation, the saturable absorber was flowed through a cell contacted onto the end mirror of the cavity such that the retroreflecting pulses set up a standing wave in the absorber and the coherent addition of the interfering light fields saturated the absorption more efficiently. Pulses of 300 fs were obtained directly from the laser [IO21 which were a factor of two shorter than obtained without this effect. The next advance was to construct a laser with no intracavity bandwidth limiting elements such as prisms and pulses of 200 fs were generated using only the dielectric coatings of the mirrors to control the laser wavelength [103]. This approach was combined with the colliding pulse geometry in a linear and a ring configuration and generated sub-100 fs pulses for the first time directly from the laser oscillator [104]. This landmark experiment prompted much research on CPM ring dye lasers and significantly shorter pulses (e.g. 55 fs [IOS]) were obtained from a ring laser with carefully (but empirically) selected mirrors. The CPM mechanism was studied in detail e.g. [IO61 but eventually proved to be essentially unnecessary once the crucial role of frequency chirp and G V D was elucidated. One interesting variation of the CPM technique was to use an anti-resonant ring configuration [IO71 which had been previously demonstrated with passively mode-locked solid-state lasers [lox]. It was the systematic optimization of the intracavity frequency chirping mechanism that led to the state-of-the-art CPM dye lasers and ultimately to the femtosecond solidstate lasers of today. Dietel et a/ [log, 1101 showed that the pulses from a CFM dye laser exhibited a negative chirp and demonstrated that they could minimize the pulse duration by changing the intracavity glass-path in the laser. The glass-path was introduced by a prism which was located such that the cavity focusing optics reduced its tendency to limit the available bandwidth [ l l I]. Time-dependent saturation of the saturable absorption was proposed as a source of non-linear chirp [70] leading to pulse compression with the GVD of the prism glass [ 1121. Formation of frequency chirp in saturable absorbers received considerable attention, e.g. [ I 131 and the optimization of c p h i lasers was shown to be sensitive to the precise detuning from the absorber resonance i.e. to the laser wavelength. The next important step was the proposal and demonstration of using prism pairs to provide adjustable intracavity GVD [ I 141. Initially this did not improve the pulse duration because the optimum GVD had already been achieved by mirror selection. The GVD contribution from dielectric mirrors was explicitly modelled [ 115, 1161 and this completed the basis on which to optimize GVD in femtosecond lasers. Martinez et a/ [ 1171 pointed out that soliton shaping in such lasers, could lead to the generation of shorter pulses in a manner analogous to compression in optical fibres. Subsequent more sophisticated theoretical models considered the interaction of GVD with frequency chirp arising from both SPM due to the Kerr effect and the time-dependent saturation oftheabsorption [ I 18, 1191. In 1985 Valdmaniset a/[120]demonstrated a c r M ring dye laser which generated pulses as short as 27 fs. This laser was designed to maximize the positive frequency chirp due to SPM in the dye jets and minimize the negative contribution due to absorption saturation [ 1211. This work prcmpted much more investigation of the pulse shaping mechanisms in CPM dye lasers (see sections 3 and 4) although significantly shorter pulses were not reliably generated since the durations were already limited by the linewidths of the amplifier and absorber dyes. Salin et a/ 11221 observed high order soliton propagation in a CPM dye laser and proposed this as a compression mechanism and a perturbed non-linear SchrBdinger equation (NLSE) as a model of the laser. Haus and Silberberg [I231 predicted that soliton shaping

The generation of ullrashorf laser pulses

185

would decrease the achievable pulse duration only by a factor of two before instabilities set in. They suggested that the shortest pulses would be obtained for zero intracavity CVD and external linear compression but Avramopoulos et a1 [124], showed theoretically and experimentally that the dye laser, with its strong spectral filter, due to the finite linewidths of the gain, absorption and cavity optics, was unstable for zero GVD and produced the shortest pulses for net negative GVD. These pulses were transforinlimited (i.e. chirp free). This work also indicated that the higher order soliton compression was not appropriate for such a system and that the NLSE could only be used to describe the laser when the spectral width of the pulses was much less than that of the effective filter. Most of the development of femtosecond dye lasers was undertaken using the CPM ring laser based on the design of 11211 using Rhodamine 6GjDODCl as the saturable amplifier/absorber dyes. The same techniques were eventually successfully applied to other amplifier/absorber dye combinations (e.g. [ 1251 and references therein) providing spectral coverage with -100 fs pulse generation from the blue [I261 to the near infrared [127]. It was observed, however, that not all combinations of gain and absorber dyes would result in passive mode-locking, even though they fulfilled the criteria set out by New [96, 971. Empirical selection of successful saturable absorbers based on systematic lifetime measurements (e.g. [128]) confirmed the hypothesis that a short lifetime was required for the absorber to initiate mode-locking e.g. [125]. This was supported by a theoretical model of New and Rea [129]. Alternative techniques for providing adjustable GVD were employed including Gires-Toumois interferometers [ I 301 which offered the possibility of compensating for higher order G V D [I311 although this did not result in the generation of shorter pulses [132], suggesting that this was not an issue for the relatively narrow spectral widths imposed by the dye media. An alternative technique for adjusting the cavity CVD was to employ dielectric mirror coatings off-resonance by using an appropriate non-resonant wavelength [I331 or by varying the angle of incidence [ 1341. Although the passively mode-locked femtosecond dye laser was demonstrated in , 1974 1991. it was not until 1985 [ 1351 that altemative amplifier/absorber dye combinations were used to generate fs pulses a t wavelengths outside the -590-640 nm range of Rhodamine ~G/DODCI. This lack of tunability was largely met by c w mode-locked dye lasers synchronously pumped by mode-locked ion lasers or frequency-doubled Nd :YAG lasers. The first demonstrations of synchronous pumping using Q-switched mode-locked solid-state lasers [53-551 generated pulses of >IO ps duration. Improved pulse compression was achieved by using a c w pump source which provided many more cavity round-trips for the pulses to be compressed and the first experiments yielded pulses of 2.8 ps duration [ 1361. The technique was quickly extended to other laser dyes providing broad spectral coverage from the blue (pumped with a mode-locked argon ion laser in the u v [ 1371) to the near infra-red (pumped with a mode-locked krypton ion laser 11381 or with a mode-locked N ~ : Y A laser G [ 1391) and this approach subsequently became the principal commercial ultrafast laser system. Any c w laser which can be excited by a mode-locked c w laser can obviously be synchronouslypu~~~pe~, If the pulse energy of the slave laser is sufficient to saturate the gain, then the laser will become synchronously mode-locked. In practice this means that all laser dyes which could be pumped by the c w mode-locked ion lasers or N ~ : Y A lasers G were available for c w mode-locking. The technique has also been applied to colour-centre lasers e.g. [ 140, 1411 where the higher saturation flux makes the pulse shaping less effective but still achieves durations of a few ps. Subpicosecond operation was first demonstrated by Heritage

I86

P M I+ French

and Jain [I421 who used 5 ps pulses from a synchronously pumped Rhodamine 6G laser to synchronously pump a Rhodamine B laser. The pulse duration obtained with synchronous mode-locking was shown theoretically 11431 and experimentally [ 144, 1451 to decrease with the square root of the duration of the pump pulses, to increase with the square root of the intracavity bandwidth 11431 and to decrease with synchronous pump power [ 146, 1471. This prompted several researchers to synchronously pump dye lasers with fibre-optically compressed pulses from mode-locked Nd :YAG lasers, thereby obtaining pulses of 0 0 0 fs duration e.g. [148, 1491. The principal disadvantages of synchronous pumping were associated with the need to maintain an accurate match (within a few microns) between the lengths of the pump and slave lasers (see section 3). This was observed experimentally [150, 1511 and was in agreement with various theoretical models e.g. [ 152-1541, Subsequently it was proposed that the slave laser cavity should be slightly longer than the pump laser to account for the fact that the pulse experienced a forward time-shift upon amplification [ 1551. Experimental [144, 1561 studies revealed that for longer slave cavity lengths, single pulse trains were obtained but for shorter cavities, satellite pulses were also observed. Theoretical studies indicated that the precise cavity mismatch at which the shortest pulses were obtained depended on the pulse duration and the bandwidth of the spectral filter in the cavity [117, 1571 but generally the slave laser cavity needed to be longer than the pump cavity to eliminate the formation of satellites, but this condition did not correspond to the shortest pulsewidth measurements. The role of stochastic spontaneous emission in the synchronously mode-locked laser was shown by Catherall and New [ 1571 to introduce random fluctuations in the pulse trains which could not be eliminated although they would not be directly observed in time-averaged autocorrelation measurements [ I581 (but usually would produce the exponential sloping wings often observed in early subpicosecond synchronously mode-locked lasers). Precisely matching the cavities would cause the slave pulse to lose synchronism with the pump pulses because of the forward time-shift in the amplifier and therefore would impair the mode-locking. On the other hand, lengthening the slave cavity to accommodate this time-shift would cause the broadband noise fluctuations, which arise from spontaneous emission, to circulate at a dilferent repetition rate to the slave pulses and cause random substructure to appear on the pulse trains. This conflict and the corresponding instabilities could be greatly ameliorated by reducing the width of the cavity spectral filter to limit the bandwidth of the noise fluctuations and tbe pulses to the picosecond regime i.e. closer to that of the pump pulses 11591. This approach was adopted in the standard commercial synchronously mode-locked dye lasers and in many ultrafast applications laboratories. An alternative approach was to pump a dye laser with compressed pulses of a few ps or even subpicosecond duration [ 148, 149, 160, 1611 and obtain near transform-limited femtosecond pulses. Recently a technique has been proposed [ 1621 and demonstrated [ 1631 to reduce the deleterious effects of spontaneous emission by feeding back a small part of the laser output to act as a coherent seed which swamps the noise background but which does not affect the pulse-shaping dynamics. This has resulted in stable pulse trains of transform-limited pulses. A compromise between passive mode-locking and synchronous pumping, called hybrid mode-locking was also developed for ultrafast dye lasers. Essentially the performance was a reflection of this compromise: the selection of appropriate saturable absorber combinations was less restricted than for pure passive mode-locking but the requirement for accurate matching of the cavity lengths was retained. The saturable absorber tended to suppress the noise background, reduce the wings of the pulses and

The generation of ul1r.ashor.t laser pulses

I87

relieve the problems caused by the forward time-shift in the amplifier by offsetting it against a backward time-shift in the absorber, The precise mode of operation depended on the amplifier/absorber dyes used and on the pumping conditions. The earliest hybridly mode-locked laser was demonstrated by Shank and Ippen in 1974 and generated pulses of 0.5-1 ps from a laser in which Rhodamine 6G and D O D C ~were flowed in a single jet stream [ 1641. Similar schemes utilizing a composite gain/absorber medium were domonstrated to provide spectral coverage i n the visible from 535 nm to 61 1 nm e.g. [165, 1661 and in the near infra-red from 750 to 850 nm e.g. [167, 1681. Typically these lasers employed much weaker saturable absorber concentrations than the passively mode-locked lasers and generated pulse of several 100 fs duration. Pulse shapes tended to resemble Lorentzian profiles with sloping wings, indicative of fluctuations in pulse profile or duration. A composite dye jet hybridly mode-locked laser yielded pulses as short as 70 fs [ 1691 but this result was not reproduced by other workers and it may perhaps have been due to a fortuitous optimization of intracavity GVD. Hybridly modelocked dye lasers with separate absorber and amplifier dye jets were also developed and colliding-pulse-ring [ 1701 and anti-resonant-ring [ 171) laser cavities were employed to enhance the pulse shortening, resulting in the generation of pulses as short as 85 fs [ 1721. The incorporation of intracavity prism pairs led to shorter pulses of durations comparable to the contemporary passively mode-locked lasers e.g. [173,1741 although the sloping wings of the autocorrelatiou traces and the excess time-bandwidth products indicated that the pulse trains were not of the same quality. Linear dispersion-compensated cavities were found to generate similar duration pulses to the CPM configuration. This approach was useful for several infra-red amplifier/absorber dye combinations which would not work in pure passively mode-locked lasers e.g. [175, 1761 and references therein. Chesnoy and Fini [I771 published a simple scheme to lock the cavity length of the dye laser to its optimum value which many researchers found of practical value. An alternative scheme based on the strength of a second hannonic signal [I781 was also proposed and most workers reported a sub-micron sensitivity of the pulse duration to the cavity length of the dye laser. The ultimate hybridly mode-locked dye lasers generated close to transform-limited pulses of less than 30fs duration e.g. [ 179, I 801 and the cavity designs were very similar to their purely passively mode-locked counterparts. I n general, hybridly mode-locked dye lasers were experimentally more complex than purely passively mode-locked dye lasers and often gave inferior performance but found useful application in the near infra-red spectral region where the lifetimes of the amplifier/absorber dyes precluded passive mode-locking. Today this spectral region is directly covered by femtosecond Ti :sapphire lasers. An important development to femtosecond dye lasers was the fibre-optic pulse compressor (see section 4). Reference [ 1811 provides a useful introduction to this topic and [I821 provides a tutorial overview offibre-optic compression. Following the proposal to compress pulses with chirp generated using SPM arising from the optical Kerr effect [87], experiments with a liquid CS2-filled fibre domonstrated that a linear chirp could be obtained at power levels below the threshold for self-focusing or self-trapping [183]. The development of very low loss, single-mode optical fibres [I841 provided a more convenient non-linear medium for SPM [ 1851 and compression experiments were carried out using negatively dispersive delay lines (near-resonant sodium vapour) to achieve 1.1 ps duration [186]. Shank et al [I871 combined sphi in the silica fibre with negative GVD from the diffraction grating pair of Treacy [86] to compress 90 fs to 30 fs and so invented the fibre-grating compressor. It was shown theoretically that the positive GVD of the fibre in the visible spectral region served to make the chirp from the SPM

188

P M W French

relatively linear and so large compression factors were achievable [ 1881. The technique was applied to the output of a synchronously mode-locked picosecond dye laser and yielded tunable pulses of -450 fs duration 11891. This system was later refined and the compressed pulses were again compressed in a second fibre-grating stage to produce tunable pulses of 90 fs duration [ 1901. A useful analysis of achievable fibre-optic pulse compression was published by Tomlinson et a/ [I911 which suggested shorter pulses could be obtained with higher peak powers and shorter fibre lengths. The technique was successfully applied to amplified passively mode-locked CPM dye lasers to yield pulses of 16 fs at IOHz repetition rate [I921 and 12 fs at 500 Hz 11931. At the same time, the c w mode-locked Nd : Y A G laser was compressed by several groups to provide pulses as short as 1.8 ps with peak powers of -3.4 kW 11941. A @switched modelocked Nd :YAG laser was compressed to -3 ps at 500 Hz repetition rate yielding peak powers as high as 1.5 M W 11951. The largest compression factor ever achieved was obtained with a 1.3 pm Nd:u,\c laser which was compressed first from 90 ps to I .5 ps using a fibre-grating compressor with positively dispersive fibre and then to 33 fs using soliton compression in a fibre with negative G V D [196]. As described above, compressed cw mode-locked Nd:Yac lasers were used to synchronously pump dye lasers to obtain tunable femtosecond pulses, e.g. [148, 1491. An alternative approach to generate widely tunable femtosecond pulses was to improve on the experiment of [I901 and to include an amplifier between the two fibre-grating compressors. This yielded tunable pulses as short as 16 fs at a repetition rate of 200 Hz [I 971. The principal disadvantages associated with this scheme were the fact that the fibre-optic compressor tended to produce pulses with some of the energy in a pedestal and the experimental ditliculties associated with optimizing the fibre-grating compressors for fluctuating input signals from the synchronously mode-locked dye laser. Yet another scheme for obtaining tunable femtosecond pulses was to use the relatively high quality amplified, compressed pulses from a passively mode-locked CPM laser to generate a femtosecond white-light continuum (e.g. [198]) and either employ this directly as a probe signal or select the desired spectral component and amplify it further to the required energy [199]. This was a complex experimental system but provided significant wavelength flexibility. The development of the copper vapour laser-pumped dye amplifier permitted the fibre-grating compression of 40 fs pulses from a CPM ring dye laser to 8 fs at a repetition rate of 5 kHz [200]. The transform-limited duration of the associated spectrum was 6 fs and in a subsequent experiment using a fibre-grating compressor which also incorporated a prism sequence which permitted it to compensate for both second and third order phase variations, the 6 fs pulse duration was achieved [201]. These transformlimited pulses, of only three optical cycles, remain the shortest generated to date. The experimental inconvenience of working with dye lasers, however, is leading to their replacenienl in most ultrafast laser applications by femtosecond solid-state laser systems.

2.4. Third generation: ferntosecorid solid-state lasers

Although they have not produced significantly shorter pulses than the second generation cw femtosecond dye lasers, femtosecond solid-state lasers have transformed ultrafast laser physics-particularly in terms of convenience and application. As discussed earlier, the two crucial elements to this revolution were the exploitation of the optical Kerr effect as an effectivelyinstantaneous non-resonant saturable absorber and the development of ultra-broadband vibronic laser media such as Ti :sapphire. With

The geiieration of ultrasliort laser pulses

189

these elements and with the experience of optimizing intracavity GVD gained from cw femtosecond dye lasers, sub-100 fs generation has become routine. The much greater convenience and tunability of femtosecond solid-state lasers have had a huge impact on ultrafast science and are far more significant developments than a further reduction it! the pulse duration. While the second generation of femtosecond lasers were generally not commercially available and required several man-years of experience to operate them routinely, the commercial femtosecond Ti: sapphire lasers are certainly state-ofthe-art and usually provide superior performance, in terms of power and tunability, compared to home-built systems. Commercial lasers typically provide pulses with durations limited to -100 fs but most researchers do not require the sub-20 fs pulses achievable from specially optimized systems. The ease of use and high power available from commercial systems is leading to their increasingly widespread applications in physics, engineering, chemistry, biology and other disciplines. Reference [39] provides an excellent review of the development of femtosecond solid-state lasers and [202] provides a snapshot of the current state-of-the-art. Of course progress has continued in the development of other ultrafast lasers including dye lasers and semiconductor lasers (see [203] for a recent review), femtosecond solid-state lasers have dominated the field and this section will concentrate on their development. The potential of the optical Kerr effect to simulate a saturable absorber without compromising the tunability o f a laser was demonstrated by Dahlstrom [204] who used the optical Kerr effect to induce an intensity-dependent polarization rotation of the intracavity light. The presence of intracavity polarizers translates this to an intensitydependent loss which can provide passive mode-locking and/or Q-switching in an manner similar to a real saturable absorber. Earlier, Comley er a/ [205] had reported passively mode-locking a ruby laser using an intracavity optical Kerr cell without explicitly considering intensity-dependent loss. As mentioned earlier, an alternative approach to simulate saturable absorption with the optical Kerr effect by using self-focusing was proposed by Lariontsev and Serkin [89]. Initially, strong non-linearity based on the orientation dynamics of anisotropic niolecular liquids was used to provide the optical Kerr effect. Unfortunately, this non-linear response, and therefore that of any simulated saturable absorption, was temporally limited to the molecular orientation relaxation time which was typically greater than a few picoseconds and this in turn limited the minimum achievable pulse durations. The difficulties in exploiting the relatively weak response of faster electronic non-linearities and the impressive successes achieved by femtosecond c w passively mode-locked dye lasers resulted in little attention subsequently being paid to these non-resonant approaches to passive mode-locking. The breakthrough in non-resonant mode-locking which triggered the renaissance in ultrafast solid-state lasers was the demonstration of the soliton laser in 1984 by Mollenauer and Stolen [206]. This was a synchronously-pumped colour centre laser which achieved -100 fold pulse compression to 210 fs by exploiting the optical Kerr effect induced in a single mode silica fibre which was located in a coupled resonant cavity. It washighly significant that the practically instantaneous electronic noii-linearity of the glass could be accessed by the power levels available in a c w laser due to the long interaction length of the fibre geometry. This experiment prompted much work by other researchers to understand this surprising result and L o apply the technique to other laser media. Figure 2 illustrates schematically how todays solid-state femtosecond lasers evolved from the soliton laser (i.e. an actively mode-locked laser with a resonant non-linear external cavity that exhibited anomalous CVD and therefore pulse compression, as indicated in figure 21~)).Initially, the fibre in the external cavity was

190

P M W Frencli

Figure 2 . Schematic ofevolution offemlosecond solid-state lasers.

required to compress the picosecond pulses generated by the actively mode-lockedeolour centre laser, due to the soliton shaping effect of SPM and anomalous GVD. The eompressed pulses were then re-injected back into the main laser cavity and this 'seeding' with shorter pulses resulted in enhanced mode-locking. It was observed that the length of the external cavity needed to be interferometrically stable and an elegant active stabilization feedback loop was developed for this purpose [207]. At first this approach to mode-locking seemed restricted in scope since the laser was required to operate at wavelengths for which the fibre was anomalously dispersive. Several groups constructed theoretical models of this system and Blow and Wood predicted from numerical simulations [208] that soliton shaping (i.e. pulse compression) was not necessary in the optical fibre for this successful enhancement of the modelocking using a non-linear external cavity. This was confirmed experimentally using fibre whieli did not exhibit anomalous dispersion [209-21 I]. Using a non-/incur intcrfero~~zelcr (e.g. a non-linear external resonant cavity) to achieve compression had been proposed in 1986 by Piche and Ouellefte [212] and passive mode-locking with a coupied non-linear Michaelson interferometer was demonstrated using a CO2 laser [213]. In 1989, Mark et a1 [214] applied a non-linear interferometer model to a synchronously pumped colour centre laser with a non-linear external resonator and produced a simple

The generation of ultrashort laser pulses

191

and convincing description of the pulse compression/enhanced mode-locking mechanism which was described as Additive Pulse Mode-Locking or APM. Since anomalous GVD was no longer an issue for APM, the generic picture of the mode-locked laser became that of figure 2(6). This immediately suggested that the technique could be applied to a wide range of laser media covering the visible and near infra-red spectral regions. It was applied to Ti:sapphire [215], using an intracavity acousto-optic modulator to initially actively mode-lock the Ti: sapphire laser, and directly generated femtosecond pulses from this laser medium for the first time. Almost simultaneously, Goodberlet et ai [216] generated chirped pulses of 1.4 ps duration from a similar laser and demonstrated that the active mode-locking of the Ti:sapphire laser was not necessary since the mode-locking was sustained once the modulator was switched off. This confirmed that APM was truly a passive mode-locking mechanism. Thus the typical APM laser with a non-linear coupled cavity evolved to the stage indicated in figure 2(c) and was quicly adapted to other laser systems including diode-pumped [217] and lamppumped [218] N ~ : Y Alasers, G a krypton ion laser-pumped Nd:glass laser [219], lamppumped [220] and diode-pumped [221] N ~ : Y Llasers F and a Cr:Forsterite laser [222]. The pulses generated were sub-100 fs or close to the limits imposed by the linewidths of the gain media. The principle disadvantage of APM is the need to maintain the cavity lengths to interferometric stability. This issue was elegantly side-stepped in an experiment to passively mode-lock a N ~ : Y A G laser [223] using a non-linear Sagnac interferometer (or anti-resonant ring), rather than a coupled cavity (i.e. a Fabry-Perot interferometer). Such a configuration has the advantage that, since the pulses to be interferometrically combined propagate around the same path in the Sagnac interferometer (albeit in opposite directions), then any fluctuations in cavity length will be experienced equally by both arms of the interferometer and will not affect the coherent addition. An allfibre embodiment of a non-linear Sagnac interferometer, the Non-linear Optical Loop Mirror, or NOLM, had previously been proposed 1.2241 as an element with an intensitydependent reflectivity. This approach was subsequently applied to fibre lasers in which a Sagnac interferometer containing a length of fibre and a Nd:glass fibre amplifier was used to achieve pulse compression [225] and mode-locking [226], yielding pulses of 125 fs duration. The non-linear Sagnac interferometer incorporating the amplifier was termed a Non-linear Amplifying Loop Mirror, or NALM. Following the proposal to incorporate a non-linear Sagnac interferometer i n a unidirectional ring laser [227], this realization of APM was applied to Er:glass fibre lasers, resulting in the Figure of 8 Laser or FSL [228]. Pulses of a few 100 fs duration were obtained by several research groups, e.g. [229-2311 and this type of laser was the subject of intensive research which was particularly directed towards understanding the source of instabilities associated with soliton-like pulse propagation that were observed as the pulse duration approached its minimum value e.g. [232]. By carefully optimizing the GVD of mode-locked fibre lasers, sub-100 fs pulses can be obtained, e.g. [233]. This topic is discussed in greater depth in sections 3.3.3.2 and 4.4.4. An alternative implementation of APM in fibre lasers was to use non-linear polarization rotation, induced by the optical Kerr effect. Essentially the scheme proposed by Dahlstrom [204] can be implemented using a length of optical fibre to replace the Kerr cell. This was proposed by Stolen e/ a1 [234] as an intensity discriminator to remove low level pedestals from intense fibre-optically compressed pulses. Mode-locking by non-linear polarization rotation may be considered as APM in which the two polarization states of the fibre correspond to the excitations of an equivalent Mach-Zehnder fibre

192

P M W fieitch

interferometer [235,236]. The technique was applied to a dispersion-compensated Nd:glass fibre laser and achieved pulse durations as short as 70 Is [237]. Subsequently i t was applicd to Er:glass fibre lasers to yield picosecond [238] and femtosecond e.g. [239,240] pulses. These femtosecond Er:glass fibre lasers suffered from the same soliton instabilities as their Sagnac interferometer counterparts and, after careful optimization of the GVD, yielded pulses as short as 77 fs 12411 from a laser whose net GVD was slightly positive and so would not support soliton propagation. The shortest pulses yet obtained from a fibre laser were of 32 fs duration, obtained from an Nd fibre laser, mode-locked using mM/polarization rotation [242]. The essential experimental feature of APM is that it exploits optical fibre to provide the Kerr non-linearity in a geometry which provides a large interaction length. As discussed in section 3.3.3.2, this produces significant non-linear phase shifts for relatively low intracavity power, permitting self-starting, but the interferometric nature of the A m mechanism makes the laser operation sensitire to perturbations e.g. arising from environmental changes. While the Sagnac interferometer is insensitive to changes in cavity length since the interferometer arms share the same physical path, the polarization state of the light in the oplical fibre is sensitive to environmental perturbations. Any change in polarization state will effect the coherent addition of the signals in the interferometer and so degrade the APM laser performance. It is therefore desirable to design a femtosecond laser which does not require an optical fibre or active stabilization of the cavity length. One approach, initially demonstrated with Ti:sapphire and designated resonant passive mode-locking, or RPM, was to replace the fibre in the resonant saturable absorber [243]. This external cavity with a multiple-quantum well (MQW) laser was found to adjust its operating wavelength to accommodate any relative change in cavity length 12441 and so was stable to interferometrk fluctuations-at the cost of fluctuations in the laser wavelength. The use of a real saturable absorber compromised the tunability of this mode-locking scheme but this was somewhat alleviated by the demonstration of broadband bulk semiconductor absorbers in lieu of the MQW absorbers [245]. TypicaUy RPhf generated pulses of a few picoseconds duration. The pulse duration is limited by the absorption recovery dynamics of the saturable absorber and by any etalon formed by the difference in the lengths of the main and external cavities. The latter issue may be avoided by using the MQW semiconductor in the main laser cavity as a fast saturable absorber to generate pulses of less than -200 fs duration e.g. 12461. A n elegant refinement of the RPM technique was to construct a composite monolithic structure comprising a partially transmissive main laser cavity mirror, the saturable absorber and a high reflector which was spaced to provide the optimum interferometric condition for the addition of the radiation from the saturable absorber cavity and the main cavity. This device, which thus avoids all issues of cavity length stabilization, was described as an anti-resonant Fabry-Perot saturable absorber, or A-FPSA, and it has been successfully applied to mode-lock several laser media e.g. 12471 and to initiate mode-locking based on the optical Kerr effect e.g. [248]. RPM, including the A-FPSA, is reviewed in [249]. The intensive investigation of APM lasers, particularly Ti :sapphire, led to several surprising observations. In particular, it was shown that passive mode-locking of a Ti:sapphire laser could be obtained, even without a fibre or other non-linear element in the external cavity 12501. I t was observed that moving one of the mirrors in the laser or the linear external cavity was necessary to achieve mode-locking and that there was no need to match or to stabili7,e the cavity lengths. Figure 2(d) illustrates this moving mirror mode-locking. Typically pulses as short as a few ps were generated and the

The generation of ultrashort laser pulses

193

technique was shown to be applicable to Nd:glass fibre lasers [251]. Subsequently it was demonstrated that the external cavity was not necessary and simply moving the mirror of a cw (Ti:sapphire) laser at speeds of a few mm s- was sufficient to produce mode-locking [252]. This situation is represented in figure 2(e). The mode-locking mechanism was not explained until very recently but the techniqu? is reminiscent of an experiment by Smith who used a moving mirror to mode-lock a HeNe laser in 1967 [253]. Sargsjan er a/ [254] showed that the coupled linear cavity could increase the frequency shift imparted by the moving mirror due to the Doppler effect, although this frequency shift would still be much less than the longitudinal mode spacing. Cutler [255] showed empirically that the action,of the moving mirror would be to prevent the establishment of c w laser action by the Doppler shifting of the laser modes out of the gain linewidth. He also observed in numerical simulations that the presence of nonlinearity in the cavity with a moving mirror could lead to mode-locking. Wu et a1 [256] explained how the combined action of SPM with a finite gain linewidth would tend to pull the laser frequency towards the centre of the gain profile and this non-linear process could balance the Doppler shift due to the moving mirror. This mode-locking technique is discussed further in section 3.3.4.3. In general it is a weak mode-locking mechanism and generates picosecond pulses. It does, however, provide a simple means of providing high intensity radiation to initiate other non-linear mode-locking schemes. In 1990, Spence et a/ [257] reported an even more surprising experimental observation. They observed that they could obtain stable mode-locking and generate pulses as short as 60 fs duration [258] from a simple Ti: sapphire laser cavity with no modulator and no moving mirror. The mode-locking mechanism was not explained but it was observed that ]node-locking was obtained for slight misalignments of the cavity, relative to that which gave the highest c w power with a TEMW beam. This startling result was accompanied by anomalously short pulses generated from Ti :sapphire lasers modelocked using saturable absorber dyes e.g. [259,260] and active mode-locking e.g. [261]. Piche [262] provided an explanation of this mode-locking mechanism in terms of selffocusing in the laser rod due to the optical Kerr effect. The idea to exploit self-focusing to simulate a saturable absorber for Q-switching and pulse compression/mode-locking had previously been discussed by Lariontsev and Serkin [SS] and by Marconi et a/ [263]. Self-focusing was explicitly used L o generate femtosecond pulses from a Tizsapphire laser by Negus ef a/ [264] who provided the acronym Kerr lens modelocking, or K L M . Several other groups subsequently reported self-mode-locked or K L M Ti:sapphire lasers which generated femtosecond pulses e.g. [265]. In each case it was found that the mode-locking was not self-starting, although it was self-sustaining. Generally the mode-locking required to be initiated by the act of alignment (or tapping a mirror) e.g. [266], by an acousto-optic modulator e.g. [267], by a MQW semiconductor saturable absorber in an external cavity ( R P M ) e.g. [268], by synchronous pumping e.g. [269], by a MQW semiconductor saturable absorber in the main laser cavity e.g. [270] and by a moving mirror in an external cavity e.g. [271]. This was because, unlike APM with an optical fibre, or KPM with a real saturable absorber, the KLM Ti:sappliire laser exhibited a weak non-linearity due L o the relatively short non-linear interaction length in the laser rod-typically only 20 mm. The c w power levels were insufficient to access the optical Kerr effect and some other mechanism was required to initiate mode-locking and generate higher intensity levels. Thus the state-of-the-art may be represented by figure 2( f)in which the anipl$er medium provides both gain and non-/incar.refractive iFidex. It was pointed out that the self-focusing driving the K L M could be arranged to take place in an additional non-linear element in the cavity [272] relaxing the constraints

194

P M W French

on the focusing parameters in the gain medium and permitting media with higher nonlinear coefficienls to be used. This was exploited to Kerr lens mode-lock a diode-pumped N ~ : Y L laser F [273] and later to a lamp-pumped Nd:vLF laser [274]. K L M has also been successfully applied to other lasers including a diode-pumped Nd :YAG laser [275], to Nd:uAc laser-pumped Cr: Forsterile lasers e.g. [276-2781, to krypton/argon ion laser pumped Cr:LisAF lasers e.g. [279-2821, to a krypton ion laser-pumped Cr:LicnF laser [283] and to Cr:unc lasers e.g. [284,285]. Various authors studied the KLM laser numerically e.g. [286,287] and analytically e.g. [288,289]. The role of soliton shaping, i.e. the interaction between SPM and negative GVD in the laser cavity, was analysed and significantly shorter pulses o f -30 fs were obtained by reducing the amount of intracavity third-order GVD [290,291]. The optimization of these femtosecond solid-state lasers is discussed in detail in section 4.4.3. By using fused silica prisms and short Ti:sapphire laser rods, several authors reported the generation of sub-20 fs pulses e.g. [292-2941 and currently pulses of -10 fs duration are routinely generated from such lasers e.g. [295,296]. The optimization of femtosecond solid-state lasers has been the subject of considerable experimental and theoretical investigation and is discussed in section 4.4.3. One of the remaining objectives in the development of femtosecond solid-state lasers is to realize figure 2(g), i.e. to achieve self-starting without resorting to an additional mode-locking scheme which increases the complexity of an otherwise simple system and usually compromises performance in terms of tunability, pulse duration or power. This issue is briefly discussed in section 3.3.4. Self-starting has been observed in a KLM Ti:sapphire laser using a unidirectional ring configuration [297] which required a relatively high pump power of 11.4 W and which was not optimized for femtosecond operation. Recently a self-starting K L M Ti:sapphire laser generating sub 100 fs pulses has been reported 12981 which exploits self-focusing due to gain saturation, rather than the optical Kerr effect, as the initial mode-locking mechanism. Unfortunately, this technique, which relies on the resonant non-linearity of the gain, is only applicable at 840 nm where the effect is strongest. Very 1 [431] have pointed out that the dynamic loss modulation of recently, Cerullo el a the KLM mechanism may be maximized by employing a symmetrical cavity and have demonstrated self-starting of a KLM laser for 4 . 5W pump power.

2.5. Fourth generation: usefir1 iillrafasl lasers

The extremely rapid development of tunable femtosecond solid-state lasers has brought the field to the point where it is not always necessary to design an experiment around the available laser SOUJCC rather, it is now reasonable to expect that a suitable ultrafast laser will exist for a particular application. Femtosecond solid-state lasers provide direct spectral coverage from -650-1600 nm using the media discussed in seclion 1.3.2. This range is greatly extended by the use of non-linear frequency shifting techniques which can be relatively efficient (>tens of YO)due to the high peak powers available from such lasers. Frequency doubling, tripling and quadrupling has been demonstrated to yield watts of average power tunable down to the ultraviolet. The near infra-red is covered up to -5 p m by picosecond and femtosecond optical parametric oscillators ( 0 ~ 0 ) s . This technology is already commercially available with sophisticated computer control and is remarkably easy to use compared with its primitive antecedents. The principal factors preventing the widespread proliferation of ultrafast technology are cost, size and lack of non-scientific applications. These factors are obviously related and it is

The generalion o f ult?adiort laser pulses

195

these issues which are being actively addressed in the development of fourth generation ultrafast laser technology. There are, i n fact, many applications of ultrafast lasers, aside from time-resolved spectroscopy, which have been proposed. Many more will follow once such lasers are not automatically eliminated from consideration on the grounds of cost, size or complexity. One class of applications requires short pulse durations for metrology e.g. precision optical ranging, three-dimensional imaging, time-gated imaging through turbid media, electro-optic sampling to measure fast electrical signals, seed lasers for high power (terawatt) laser amplifier systems and the generation of short electrical transients. Ultrafast optical signals are useful in their own right in the field of telecommunications and recently there has been considerable excitement concerning the potential to control chemical reactions using precisely defined ultrafast optical signals. A further class of applications is only indirectly concerned with short pulse durations but exploits the relatively high peak power of ultrashort pulses. Non-linear optical processes such as frequency doubling to the blue and to the w are of interest for optical storage, printing and lithography. Parametric generation in the infra-red is important for remote-sensing for pollution monitoring etc. High peak power is required for two-photon microscopy and for applications were non-linear processes are used to monitor e.g. micro-roughness on semiconductor samples using second harmonic generation. As ultrafast lasers become more compact, more robust and cheaper, many new applications will be found. If a versatile ultrafast laser was of comparable size and price to an oscilloscope, it would also become a standard instrument in many laboratories. A reduction in the price of ultrafast technology by a factor of ten will stimulate more new science than a further tenfold decrease in pulse duration or tenfold increase in peak power. Semiconductor diode lasers have long been identified as the lasers of choice for all practical applications and they have a strong tradition in this area. Compact disc players and laser printers etc are the largest markets by far for lasers. Telecommunications is currently a much smaller market but the continuing trend towards optical communication links will increase the demand for this technology. Recently, however, the development of diode laser-pumped fibre lasers and solid-state lasers has suggested that, for ultrafast applications, the use of non-semiconductor laser media may offer many advantages. Nevertheless, semiconductor lasers continue to offer very strong compelition. In general they offer unmatched compactness, limited tunability in a single device (but broad spectral coverage is available from the technology), relatively long pulses (usually picosecond) and relatively low powers. Reference [299] provides a recent review of ultrafast semiconductor lasers. I n their most compact form the gain-switched semiconductor laser can generate pulses as short as tens of picoseconds. This approach is the simplest although the pulses are not transform-limited. Improved performance and subpicosecond pulse operation I3001 may be obtained using active-mode-lockingwith the semiconductor amplifier located in an external resonator. Subpicosecond pulses, however, are obtained in multiple pulse trains and more reliable/reproducible single pulse operation is still limited to >lops, e.g. [301], for the simplest cavities and to -5 ps for cavities with intracavity chirp compensation e.g. [302]. Passive mode-locking using degraded semiconductor lasers e.g. [303], or proton-implanted MQW absorbers in external cavities, e.g. [304], has yielded pulses of the order of I ps. The latter approach is more reliable with long operating lifetimes and has yielded pulses as short as 830 fs with extracavity compression. Recently this approach has been implemented in a monolithic amplifier/absorber structure which has yielded transform-limited pulses as short as 640 fs at 350 GHz [305]. Hybrid mode-locking, in which synchronous electrical

I96

P M 'M Fsenclt

pumping is combined with a saturable absorber, has yielded pulses as short as I .4 ps from similar monolithic structures. In general, the average output powers available from mode-locked semiconductor lasers are limited to a few mW and peak powers lo a few watts. This can be increased by using semiconductor amplifiers and peak powers as high as 160 W have been achieved with pulses as short as 200 fs using an oscillatoramplifier system with a diffraction grating compressor for chirp compensation 12991. Nigher power picosecond and femtosecond pulses may be anticipated for configurations incorporating tapered amplifiers 1351. Diode-pumped fibre lasers offer a reasonably compact source and have been used to generate picosecond and femtosecond pulses, as discussed above and in sections 3 and 4. They can experience stability problems due to environmental fluctuations but these may be minimized by careful design and the use of polarization-preserving fibre e.g. [306]. The major application of ultrafast fibre lasers is to generate signals for optical telecommunications and much work has been wried out on picosecond erbium-doped fibre lasers. Active mode-locking using electro-optic absorption modulators has provided stable sub-10 ps pulse trains with up to 30GHz repetition rate e.g. [307, 3081. Passive mode-locking with resonant M Q W saturable absorbers has provided pulses as short as 320 fs in an Er-doped polarization-maintaining fibre laser [309] and pulses of 260 fs have been obtained from a diode-pumped Nd-doped fibre laser passively modelocked with an anti-resonant Fabry-Perot MQW saturable absorber [310]. The same laser generated pulses as short as 60 fs duration using the optical Kerr effect to provide APM through non-linear polarization rotation with the A.FPSA initiating the modelocking. Femtosecond fibre lasers exploiting the optical Kerr effect for passive modelocking were discussed in the previous section and can provide sub-]00 fs pulses but need careful design to ensure convenient and reliable operation. The approach of Tamura et u l [241,31 I], combined with diode-pumping, appears a promising route to compact and reliable femtosecond fibre lasers. The fundamental limitations of fibre lasers are the relatively low output pulse energies and the lack of tunability. The first of these issues has been addressed by a combination of a gain-switched semiconductor laser and a diode-pumped Er-doped fibre amplifier [312]. Highly chirped pulses of hundreds of ps duration were generated by gain-switching a Bragg reflection diode laser and these were amplified to -2 pJ and compressed to -2 ps using a diffraction grating pair. This approach thus provides microjoule pulses of -0.5 M W peak power at repetition rates of up to 20 kHz and represents a resonably compact and power ultrafast laser system. Extended tunability clearly requires appropriate gain media and these are conveniently provided by diode-pumped vibronic solid-state lasers which show great potential for practical ultrafast lasers. Diode-pumped solid-state lasers can also provide significantly higher pulse energies and average output powers than fibre lasers or semiconductor lasers. The first diode-pumped solid-state lasers were based on GaAs/ AlGaAs diode lasers and Nd3'-doped media. A diode-pumped Nd:YAG laser was demonstrated as early as 1968 [313] but it was not until the 1980s, when long-lived, high power semiconductor lasers became available, that diode-pumped lasers could compete with lamp-pumped systems. The application of established active mode-locking techniques, using amplitude or frequency modulation, to compact diode-pumped Nd:vAo e.g. [314,315], Nd:YLF e.g. [3l6,317], and Nd:glass lasers e.g. [318, 3191, produced surprisingly short (6I O ps) pulses compared to arc-lamp-pumped systems. An explanation for these anomalously short pulses in terms of self-phase modulation (SPM) was suggested in [319]. The focusing requirements of diode-pumping (and other [uses-pumping schemes) often increased the intracavity intensity beyond that

The gerieration of u1trarhor.t laser.pulses

I97

encountered in lamp-pumped systems. This resulted in significant non-linear phase shifts over the interaction length of the laser rods which improved the mode-locking as discussed by Haus and Silberberg [320]. Soon after these experiments, the development of additive pulse mode-locking (APM) provided a technique for generating even shorter pulses from diode-pumped solid-state lasers. The long non-linear interaction length of the fibre made the technique particularly appropriate for the relatively low power diodepumped lasers. Self-starting operation with pulse durations of a few picoseconds was achieved with Nd: YAG [321] and Nd: YLF [322]. Diode-pumped solid-state lasers modelocked using APM with an optical fibre were generally self-starting and the high stability achievable with diode-pumped systems to some extent offset the problems associated with matching the laser and external cavity lengths to interferometric precision. Active stabilization following reference [207] produced useful and reasonably reliable lasers which have been used as seed sources for high power amplifier chains. Diode-pumped Nd :YAG and N ~ : Y Llasers F were also mode-locked using resonant passive mode-locking with MQW A-FPSA devices [323]. This is a promising approach for mode-locking diode-pumped Nd-doped lasers since it offers compact, reliable and self-starting lasers for which the real saturable absorber does not compromise the tunability of the laser media and the minimum pulse durations are limited by the gain linewidth, rather than the absorption recovery times. KLM has also been applied to diode-pumped Nd-doped lasers. In general it is not straightforward due to the limited intracavity power in diode-pumped lasers. Reliable self-starting KLM has yet to be realized and so a supplementary mode-locking technique must be used to generate pulses sufficiently intense to generate significant self-focusing. In the first demonstration [324], a separate non-linear element of s ~ 5 7 glass was used to provide the optical Kerr effect, rather than the Nd: YLF laser rod, and mode-locking was first initiated with an acousto-optic modulator although the system would also start from a mechanical perturbation. 6 ps pulses with 1 W average output power were obtained. A subsequent experiment with a diode-pumped Nd:uac laser used the laser rod as the non-linear medium and delivered 1 W of average output power with pulses of 8.5ps j3251. These systems are extremely impressive in terms of output power although their practical application is limited by the lack of tunability and femtosecond operation is precluded by the narrow gain linewidths. Femtosecond operation is possible with bulk Nd:glass lasers, as has been demonstrated using a Ti:sapphire pump laser e.g. [24&], but diode-pumping of this system has yet to be realized. Nd-doped lasers are particularly attractive for diode-pumping because of the overlap between the Nd absorption band and the high power GaAs/AIGaAs diode lasers which permit lasers to be constructed with sufficient intracavity power to use the optical Kerr effect for mode-locking. There is currently considerable interest in developing mode-locked diodepumped vibronic solid-state lasers which would provide tunable femtosecond operation. One approach is to use a frequency-doubled Nd-doped laser to pump a Ti:sapphire laser. This is necessary because the absorption bands of Ti:sapphire do not overlap sufficiently with any high power diode lasers. Pulses of 110 fs duration have been generated from a KLM Ti:sapphire laser which was synchronously pumped by an intracavity frequency-doubled, diode-pumped Nd :YLF laser [326]. This laser provides a reasonably compact source of tunable femtosecond pulses but there are stability problems associated with intracavity doubling of N ~ : Y Llasers F which, together with the additional ) make it less cost and complexity associated with the intermediate ( N ~ : Y L Flaser, attractive than a direct diode-pumped system. Where no direct diode-pumped option exists, pumping with a diode-pumped minilaser is clearly a useful approach and may

198

P M M/ French

find application with vibronic lasers such as Cr: Forsterite and Cr:unc which can be

pumped by diode-pumped Nd-doped lasers. There is an alternative vibronic laser mcdium to Ti:sapphire, however, wifh similar spectral coverage, which may be directly pumped by InGaAlP semiconductor diode lasers at -670 nm. This is Cr: LisnF which has been demonstrated as the first mode-locked tunable diode-pumped all-solid-state laser 13271 delivering picosecond pulses with active mode-locking [328,329] and passive (RPM) mode-locking [327,328]. Femtosecond operation of a diode-pumped Cr: LisAF laser was demonstrated for the first time using an intracavity MQW fast saturable absorber to generate transform-limited pulses as short as 220 fs [330]. Subsequently an A-FPSA was used to passively (RPM) mode-lock a diode-pumped Cr: LisnF laser to yield pulses as short as 96 fs [331]. These lasers both rely on real saturable absorbers and so compromise the tunability of the Cr:LisAF laser. There is currently intensive work to develop all-solid-state diode-pumped femtosecond Cr: LisaF KLM lasers and the first results will almost certainly be reported in 1994. At present, the chief limitations are the poor beam quality (far from diffraction-limited) and low power available from the 670 nm diode lasers. This may be addressed by the development of 670 nm MOPA devices or high power narrow stripe diode lasers. An altemative approach which has been shown to yield both picosecond [332] and femtosecond [333] pulses is to use a frequency-doubled N ~ : Y Llaser F which operates on the 1319 nm transition. This may be doubled to 659 nm and can provide a diffraction-limited pump beam. As for the case of Ti: sapphire, however, the additional complexity associated with the doubled minilaser make this approach less attractive than direct diode-pumping for practical devices. The future will see several other ultrafast diode-pumped vibronic all-solid-state lasers developed and the potential cost and complexity of ultrafast lasers should be dramatically reduced. There will be considerable emphasis on designing simple and reliable devices for specific applications. Competition between semiconductor lasers, diode-pumped solid-state lasers and diode-pumped fibre lasers will be decided by price, performance trade-offs and all three approaches will find application. The semiconductor laser will only be supplanted in real-world applications when the alternative diodepumped solution is cheaper than fabricating a complex semiconductor structure. For ultrafast lasers to become useful, there must be a move from high-tech to low-tech and from physics to engineering. This will lead to widespread application in chemistry. biology, physics, electronics and telecommunications. Cost will become an increasingly important issue. The development of specialized state-of-the-art femtosecond lasers will continue and probably pulses approaching 5 fs in duration will be generated from laser oscillators. There will be considerable interest in amplifying sub-50 fs pulses and applying them to high intensity physics, particularly x-ray laser generation. Terawatt and petawatt peak powers will be obtained from increasingly compact laser systems. Improved diagnostics and pulse synthesis techniques will lead to more complete descriptions of the amplitude and phase profiles of ultrafast optical signals being routinely used in both the generation and application of ultrafast optical signals.

3. Principles of mode-locking
3.1. Infroductioii There are many excellent review articles and books describing mode-locking. For further references and a more complete picture, see [334, 3351. In general, a laser transition

The geiteiation o f ultiashorl laser pulses

199

lhas a finite Iinewidth over which it can provide optical gain and so laser emission has a finite spectral bandwidth A V , Jn a laser cavity, the radiation is confined to discrete frequencies or modes, vn,,which are separated by 6 v , the mode-spacing. This is represented in figure 3 where TRT is the cavity round-trip time, c is the speed of light and L is the optical length of the cavity. When no attempt is made to control the laser spectrum, the free-running laser modes oscillate independently with random phases. (Even in homogeneously broadened lasers, spatial hole-burning effects mean that more than one mode can oscillate simultaneously.) The resulting laser output is noisy and incoherent with no regular temporal structure. On a very short timescale, give by the inverse of the optical bandwidth, AV, some or all of the modes can radiate in phase and this will produce a sharp feature in the laser output described as a coherence spike. The duration of these spikes or the coherence time of the laser, can thus be as short as 1,Av. The coherence length of the laser is given by cfcoh.In some interferometric experiments the coherence time/ length of a light source can be used to provide femtosecond/micron resolution. This is an inefficient way to use the light, however, sometimes requiring excessive energy to be directed at a sample in order to achieve adequate signal-to-noise levels, and it can be difficult to unambiguously interpret the data obtained. It is usually preferable to work with an optical probe which has a well determined phase profile with all of the energy confined to the time interval corresponding to the resolution desired. This is particularly important if one is concerned to achieve high intensities. Such well defined optical pulses are generated through mode-locking.

Figure 3. Schematic of laser emission

Figure 4. Schematic of mode-locked l a x i output.

If all the modes of the laser can be made to oscillate in phase, i.e. if they can be locked together, then the laser output becomes temporally well defined, with a period corresponding to the cavity round-trip time, as shown in figure 4, where
(3.2) The temporal profile is the Fourier transform of the spectral profile and so, if complete mode-locking is achieved, the duration of the pulses, I,, is related to the full gain linewidth by Heisenbergs uncertainty principle. When all the modes are oscillating

200

P M W French

either in phase or with a constant phase difference between modes, then the laser output is described as 'transfomi-limited' and the product of the full-width-at-half-maximum (FWHM) orthe spectrum of the pulscs (the bandwidth, AV) and of the temporal profile (tp) is equal to a constant (k) which depends only on the shape of the pulses. For Gaussian shaped pulses, k=0.441,for sech' pulses, k =0.315 and for single-sided exponential pulses, k=0.11. In the case of incomplete mode-locking, or in the situation where there is phase structure across the pulse, then the time-bandwidth product will be more than the limiting value:
Avi,,> k.

(3.3)

Although the temporal field profile E ( t ) is always the Fourier transform of the spectral field profile E( v ) , only in the special case of transform-limited pulses can the pulse shape be uniquely determined from the optical spectrum. In all other cases it is necessary to know the phase profile of the pulses in order to calculate the temporal profile from the spectral profile. Techniques for mode-locking lasers fall into two broad categories: active and passive mode-locking. These approaches are schematically represented in figure 5. In the first approach the radiation in the laser cavity is modulated by a signal derived from an external clock source which is matched to the cavity round Lrip time. In the latter case, the laser radiation itself generates a modulation through the action of a non-linear device i n the laser cavity. This modulation is thus automatically synchronized to the cavity round-trip frequency and requires no external clock signal. Passive mode-locking is often referred to as self-moddocking.
la1 ACIIYB Mode-Locklna

tbl PBSSIYBModeLocklng

Figure 5. Schematic of active and passive mode-locking

3.2. Active rnodelocki~ig 3.2.I . Introduction, Active mode-locking is readily explained in the frequency domain.

I t is realized by 'beating' the laser radiation signal with a signal derived from an external oscillator, or clock. An 'active' modulator periodically changes the loss or gain experienced by the intracavity laser radiation and generates sidebands on the laser radiation modes (v,J at a frequency displacement corresponding to the modulation frequency (a). If this is set equal to the laser cavity mode-spacing, then energy is transferred between adjacent laser modes which become coupled together and so oscillate in phase. Figure 6 shows the basic principle. To achieve mode-locking, it is necessary to satisfy the condition Q=c/2L.
(3.4)

Thus the modulation frequency must be matched to the reciprocal of the cavity round-trip time. This is of the order of ns for most bulk lasers which implies that

The generation of irltraslzorl laser pulses

20 I

".

"nw

Figure 6. Principle of active mode-locking.

modulators operating at of the order of hundreds of MHz are required. Such high frequency modulation can be provided by the output of other mode-locked lasers (through gain modulation), or directly by intracavity acousto-optic or electro-optic devices, which can run at rates as high as a few GHz. Although semiconductor diode lasers can clearly be constructed with much smaller cavity lengths, the difficulties involved with the production and delivery of suitable R F signals limit the actively mode-locked repetition rates of these laser systems to tens of GHz-with 1 GHz being more typical. Some semiconductor lasers are actively gainswirched by directly modulating the electrical pump signal with no attempt to match the cavity length to the driving signal. This is something of a special case and is not mode-locking. In general, acousto-optic modulators (AOMS) exhibit a time-varying amplitude (loss) modulation while electro-optic modulators (EOMS) produce a time-varying phase change. Mode-locking achieved through amplitude modulation (gain or loss) is referred to as AM mode-locking. When achieved through phase modulation, mode-locking is often referred to as mi mode-locking. 3.2.2. Loss modulation. Most actively mode-locked lasers employ an intracavity acousto-optic loss modulator which is typically a quartz transducer driven at resonance with a sinusoidal signal of -1 W RF power. This electromagnetic RF wave is converted into an acoustic wave by the transducer and the loss mechanism for the cavity radiation is diffraction at the acoustic wave. The most widely used implementation is to set up an acoustic standing wave at a frequency corresponding to hay the cavity roundtrip frequency c/2L. This diffracts the intracavity radiation, and so introduces loss, throughout each cycle except for the h v occasions ~ when all the acoustic fields are zero. Thus the modulation frequency, 0, is twice the frequency of the applied R F signal. This standing-wave AOM is sometimes called a Bragg modulator. An alternative approach is the travelling wave modulator in which the laser radiation again experiences minimum diffraction loss twice per RF period as it interacts with a transverse propagating acoustic wave. Loss modulation can be pictured'in the time domain as the action of an intracavity 'shutter' which is 'open' twice during each R F cycle-or once per cavity round-trip. The laser radiation in the correct time-window to pass through this 'open shutter' on each

202

P M W French

-t

b (U)

Figure 7. Mechanisms for

temporal and (b) spectml compression of intracavity

radiation.

cavity transit will experience least loss and laser radiation in all other time-windows will be attenuated. Competition in the laser amplifier medium should therefore result in a steady-state with the radiation of the lowest-loss time-window receiving all the gain. Alternatively one can picture a pulse circulating in the laser cavity and being multiplied by the loss modulation function on each round-trip. Since the loss modulation function is peaked, this product will be shorter than the original pulse. In the absence of other pulse shaping mechanisms, this compression mechanism should act on the pulse on every round-trip and so the laser output would eventually consist of a train of infinitely short pulses, spaced at the cavity round-trip frequency. Clearly this is unphysical and in practice the pulses will have a finite duration which is ultimately limited by the gain bandwidth of the laser gain. This may correspond to the spectral width of the gain profile of the laser or to the spectral width of any intracavity frequencyselective elements. In general, a steady-state is reached in a mode-locked laser for which the pulse compression mccltanisrns are balanced by pulse-broadening nrechanisms. This balance can be considered with respect to figure 7. Throughout the evolution of the pulse train, the radiation at the temporal peak of the pulses will experience lower loss, and therefore higher net gain, than the radiation in the temporal wings. This will tend to produce a temporal pulse compression and, by the uncertainty principle, a spectral broadening. Conversely, in the spectral domain, the gain profile of the laser will tend to amplify the spectral peak of the intracavity radiation at the expense of the radiation in the wings of the spectral profile. This tendency will narrow the pulse spectrum and therefore broaden the pulses. In lasers with no modelocking elements, the same spectral compression mechanism produces pulsed or c w operation with a narrow spectral linewidth. For all ultrashort pulse lasers, there will be a balance between pulse compression, achieved through mode-locking, and spectral compression which is a consequence of the finite gain bandwidth. More generally, this can be described as a balance between mode-locking and dispersion. In the early stages of the pulse formation, the spectrum is much narrower than the gain bandwidth and temporal compression (with concomitant spectral broadening) is the dominant process. As the pulse spectral width approaches the bandwidth of the erective cavity filter, the pulse starts to lose energy to the filter and so a steady state is reached for which the pulse energy and its spectral and temporal widths stay constant or periodically evolve about a fixed value. The dynamics of active mode-locking have been studied theoretically by many authors but the seminal work is probably that of Kuizenga and Siegman [336]. For AM mode-locking through loss modulation, the modulation function may be described by: t,,=exp[-A,(I -cos OM)]

(3.5)

The geiiera/ion of ultrashort laser pulses

203

where I,,, is the transmission function of the modulator, 2Am is the peak to peak modulation depth and w,, is the modulation frequency. Using a circulating Gaussian pulse analysis, one can derive the followiug expression for the resulting pulse duration:
(3.6)

One should note that the pulse duration achieved by active mode-locking depends most sensitively on the product of the modulation frequency, fm, and the gain linewidth, AL. The saturated round-trip gain coefficient, a, and the modulation depth, Am, contribute only as the root. For acousto-optic loss modulators the pulse duration varies as the 4 root of the power of the R F modulation signal.

3.2.3. Frequency inodulafion. For FM mode-locking through phase modulation, the pulse duration is still described by equation ( 3 . 6 ) providing that the phase modulation function is given by:

tm=exp[-jAm cosw,t]

(3.7)

where 4Am is defined as the peak-to-peak phase deviation experienced by the light passing through the modulator. To understand how pure phase modulation can result in an amplitude-modulated train of pulses, one can consider the effect of the phase modulation on the intracavity light. In general all the radiation will experience a timevarying (sinusoidal) phase delay in the modulator. This time-varying phase change corresponds to a frequency chirp (d+/dl=dw) and so, on each interaction with the phase modulator, the frequency of the intracavity light will be modified until it is swept out of the gain linewidth of the laser. For the special case of the intracavity radiation which arrives at the peaks and troughs of the sinusoidal phase modulation (when d+/ d! = 0), there will be no net change in carrier frequency of the light although there will be a small quadratic phase modulation that will broaden the spectrum. Thus the frequencychirped intracavity radiation will experience less gain than that light which arrives at the turning points of the phase modulation. This means that there is a timewindow twice in each cycle of the R F modulation signal which corresponds to the lowest loss/ highest gain and so this is an effective amplitude modulation. Electro-optic phase modulators rely on the Pockels effect and so the phase modulation depth varies with the electric field amplitude of the R F signal i.e. with the square root of the RF power. Referring to equation (3.6), it should be seen that the pulse duration therefore scales with the root of the applied RF power.

3.2.4. Pe@rmance of nctiue mode-locking. For both AM and FM mode-locking, one can drive the modulator at a higher harmonic of the laser cavity. This has the advantage that the modulation frequency is increased and so, by equation (3.6), shorter pulses are produced. The operation of harmonic mode-locking is more complex than conventional modelocking and, depending on the available gain of the laser, one can observe pulse trains a1 the fundamental period of the cavity or at multiples of this period up to that of the applied modulation signal. In general the relative advantage gained is not justified by the additional experimental complexity associated with harmonic mode-locking. The principle disadvantages of active mode-locking by amplitude or frequency modulation are the requirement to match the laser cavity length to the modulator drive frequency and the relative weakness of the pulse compression mechanism which is only

204

P M W French

RFampli6er

Fiter

Figure 8. Schemalic of regenerative moddocking

itself sufficient to generate picosecond pulses. The first disadvantage can be reduced by the technique of regenerative mode-locking [337] which is illustrated in figure 8. Essentially, the R F signal used to drive the modulator is derived from the output of the actively mode-locked laser by detecting the pulse train with a photodiode and filtering and ampliyfing the fundamental frequency component of the electrical signal which will be exactly matched to the cavity round-trip. Any drift in cavity length of the laser will be precisely tracked by an appropriate adjustment of the drive frequency. This technique has recently been successfully applied to Ti: sapphire lasers e.g. 13381. The second disadvantage arises because the modulation function, which is derived from an external source, remains constant as the mode-locking proceeds and the pulses are temporally shortened. Initially the modulation provides a strong compression of the cavity radiation but, as the pulses become much shorter than the width of tbe temporal modulation function, the pulse shaping per round-trip gets relatively weaker. Furthermore, as the pulses compress, the spectral compression mechanism becomes increasingly important. This is the reason why passive mode-locking techniques, in which the pulses induce their own modulation function of lhe same width a s the pulses, usually generate shorter pulses. Actively mode-locked lasers will only generate femtosecond pulses in the presence of gain saturation, i.e. synchronous mode-locking which does provide a sharper modulation function than the sinusoidal drive frequency, or in the presence of some additional intensity-dependent loss, i.e. passive pulse shaping by an effective saturable absorber.
3.2.5. Synchronous mode-locking. Synchronous mode-locking is a form of A M modelocking where gain modulation is the pulse shaping mechanism. It differs from loss modulation in that it requires gain saturation to work efficiently. Figure 9(a) illustrates the basic principle of operation. Essentially, the laser medium is resonantly (or 'synchronously') pumped by a train of pulses from another mode-locked laser. The rate at which the gain is 'switched on' in the slave laser follows the profile of the pump pulse. This provides a much steeper modulation than a sinusoidal signal at the cavity roundtrip frequency. The gain is then 'switched off' by gain saturation which can occur on a timescale faster than the pump pulse duration. This completes the 'fast modulation function' and explains why synchronously mode-locked lasers generate significantly shorter pulses than are obtained with conventional active mode-locking through loss modulation. The length of the synchronously pumped laser, sometimes called the 'slave laser', is set to correspond to the repetition rate of the pump pulse train, derived from the 'master laser'. Thus equation (3.4) is fulfilled when the cavity lengths of the master and slave lasers are equal. Note that strong gain saturation is required to 'close' the modulation function. Synchronous mode-locking is not effective if the individual pulse energies are not suficient to significantly deplete the population inversion in the amplifier. Thus this technique is useful for media such as dyes and colour-centre crystals

The generation of ultrashort laser pulses

(a)
1L -

205

21.

Figure 9. Schematic of synchronous mode-locking.

which exhibit high gain cross sections and low saturation fluences, as opposed to solidstate lasers which are generally not saturated at the power levels available with cw pumping. The dynamics of synchronously pumped dye lasers are extremely complex and have been studied extensively for the case of c w dye lasers e.g. [339-3451. Laser performance depends critically of the precise ratio of the master and slave cavities and, in general, synchronously mode-locked lasers do not evolve to a steady state because of the role of spontaneous emission in the laser. If the master and slave cavities were to be set to the same length, as would initially appear to be appropriate for active mode-locking, then the peaks of the pump and laser pulses would be temporally coincident and there would be no mechanism to suppress the leading edge of the laser pulses for the lypical case where the pump pulse is longer than the laser pulse. The trailing edge would be suppressed by gain saturation, providing that the laser was set up such that the gain inversion was significantly depleted by the amplification of the central portion of the slave laser pnlse. The leading edge, however, wotrld be amplified and raised up and this would cancel out any compression due to suppression of the trailing edge. The net effect of the interaction of the laser pulse with the amplifier would be that the laser pulse would appear L o move forward in its time window with nocompression. Synchronously pumping a dye laser with extremely short pulses, e.g. fibre-optically compressed, frequency-doubled Nd: YAG laser pulses [346], or pulses from a frequency-doubled femtosecond Ti :sapphire laser, does address this problem since both the leading and trailing edges of the dye laser pulses would experience less gain than the peak. For pump pulses of the order of -300 fs, close-to-transform-limitedfemtosecond pulses are readily generated. This is a special case of synchronous mode-locking and is not typical.

206

PM W ' French

A more common scenario for synchronous mode-locking, with pump pulses typically of >50 ps, is as follows. In the time domain, the slave laser pulses are compressed by the combined action of a rapid gain transient, due to ultrashort pulse pumping, and gain saturation as shown i n figure 9(6). The leading edges of the pulses are suppressed because the pump pulse is contrived to arrive at the amplifier medium a little after the slave laser pulse. This is achieved by making the slave laser cavity very slightly shorter than the master laser cavity. The peaks of the pulses are amplified and this depletes enough of the population inversion such that there is insufficient for the trailing edge of the slave laser pulses lo overcome the intracavity losses. The effective modulation depth depends on the pump pulse energy and duration. Shorter and more energetic pump pulses 'switch-on' the gain more rapidly and result in shorter slave laser pulses. The slave pulse duration has been observed to decrease with the square root of the pump pulse duration [342,347]. The mismatch of the pump modulation frequency with the slave laser cavity length is further complicated by the yoopwurd tinre-dip' experienced by the laser pulse in the amplifier, as outlined above. This time-slip must be accommodated in the setting of the relative cavity lengths. Unfortunately, the time-slip occurs to the slave laser pulses but not to the noise in the slave laser. The noise therefore circulates at a different repetition rate to the pulses and fast noise transients can 'move through' the laser pulses. The role of noise, or spontaneous emission, in synchronously mode-locked lasers means that the mode-locking never reaches a true steady-state and this is confirmed experimentallyby the typically observed winged autocorrelation pulses which correspond to time-averages of evolving pulse profiles rather than exotic pulse shapes [344,348]. It is speculated that the spontaneous emission affects the forward time-slip of the slave laser pulses, making the precise amount of time-slip randomly fluctuate from cavity round-trip to round-trip. Certainly the synchronously modelocked lasers exhibit pulse-to-pulse jitter of several picoseconds, even when the autocorrelation traces suggest subpicosecond pulses. There are several approaches to reduce the deleterious effects of noise in synchronously mode-locked lasers. Using a strong intracavity spectral filter, such as a Lyot filter, it is possible to obtain pulses of a few picoseconds duration which are close to transformlimited. The severe filter restricts the bandwidth of the noise to that of the pulses and limits its adverse effects. It reduces the uncertainty in the forward time-slip. Another approach is to seed the spontaneous emission with a weak coherent signal, just before the slave laser pulse is amplified. This can be achieved by using an external cavity, slightly shorter than the slave laser cavity or, equivalently, by using a special mirror with an additional coating layer which reflects a very weak prepulse in front of the main laser pulse. This prepulse greatly reduces the effects of the spontaneous emission. It can be thought of as controlling and making reproducible the precise forward timeslip of the slave laser pulses on each cavity round-trip. This approach reduces pulseto-pulse jitter and permits stable transform-limited pulses to be generated [349]. For synchronous mode-locking to work, it is necessary to find a slave laser medium with the appropriate saturation properties. I n pr'actice, the available pump lasers (usually actively mode-locked argon ion lasers or N ~ : Y Alasers) G can provide up to a few watts o f average mode-locked pump power with pulse energies in the nJ range. The most suitable laser media are organic dyes, colourcentre crystals and semiconductor media. Synchronous mode-locking was first developed with organic dyes [350,351] and became the principal commercial ultrafasl laser for more than a decade. Colourcentre lasers have enjoyed a less widespread application, largely due to the experimental dificulties of working at cryogenic temperatures. Gain modulation in semiconductor

The generation o f ultrashort laser pulses

207

lasers is more commonly achieved through electrical pumping which is closer to loss modulation with a negative loss.

3.3, Passive mode-locking 3.3.1. Introduction. Figure 5(b) shows the simplest model of a passively mode-locked laser which considers just an amplifier and a saturable absorber providing an intensitydependent loss in a cavity. Amplitude modulation is provided by the interaction of the pulse iwdfwith the intracavity elements and no external modulation is required. For this reason, passive mode-locking is often referred to as self mode-locking and the pulse shaping process is described as passive amplitude modulation. There are two distinct types of passive mode-locking according to whether the saturable absorber is considered to be yast or dolo.A saturable absorber may be considered to approximate to an electronic two energy level system with the transition frequency corresponding to that of the laser radiation. Initially all the electrons are in the ground state and so it exhibits absorption. If there is sufficient incident radiation, sufficient electrons will be promoted to the upper state such that the absorption will be saturated. Thus the transinission o f the saturable absorber i s mod$ed by the incident light. The relaxation of the electrons back t9 the ground state is limited by the upper state lifetime of the transition i.e. the absorption recovery time. The degree of saturation of the absorption depends on the amount of energy absorbed over the period of the upper state lifetime. For incident pulses which are much longer than the absorption recovery time, the transmission is an increasing function of the intensily of the pulse profiles and the saturable absorber is said to befast. When incident pulses are shorter than the absorption recovery time, the transmission does not follow the intensity profile of the pulses but depends on the pulse energy. When a passively mode-locked laser operates in this regime, the saturable absorber is said to be slow. In either case there is obviously a requirement to match the absorption transition frequency to the frequency of the laser radiation and so the spectral coverage of passive mode-locking is limited by the availability of suitable saturable absorbers. 3.3.2. Passive inode-locking with a fast saturable absorber. Passive mode-locking with a fast saturable absorber relies on the absorption saturation to generate the amplitude modulation function and do all the pulse-shaping. It was originally used to generate picosecond pulses from solid-state lasers using organic dyes as saturable absorbers and has recently enjoyed a renaissance with solid-state laser media being mode-locked using the optical Kerr effect. Since the transmission of a fast saturable absorber follows the intensity of the incident radiation, the peak of a pulse will experience lower loss than the less intense wings and so will be preferentially amplified, leading to pulse compression. In other words, the induced modulation function will follow the pulse intensity profile (unlike the fixed modulation function of active mode-locking) and so becomes stronger as the pulses compress. Thus a fluctuation in the intensity of a cw laser with a fast saturable absorber will be preferentially amplified and compressed until its duration is of the order of the absorption recovery time. If there is more than one fluctuation present initially in the laser cavity, gain competition will ensure that eventually the most intense fluctuation evolves into the pulse which will take all the gain. Fluctuations can simply arise from noise or can be induced by an external pertubation of the laser. When a pulse is compressed such that its deviation approaches the absorption recovery time,

208

P M W French

the transmission of the saturable absorber ceases to be intensity-dependent, the modulation function becomes fixed and further compression is extremely weak. Although the saturable absorber will continue to suppress the leading edge of the pulse, once the absorption is saturated, the rest of the pulse will be transmitted and there will be no suppression of the trailing edge until the absorption recovers. Thus the final steadystate of the induced modulation function has a fixed width of the order of the absorption recovery time. The laser ouput will therefore evolve to a series of mode-locked pulses at the cavity round-trip frequency whose final pulse duration is limited by this recovery time. The fastest saturable absorber dyes have recovery limes as short as a few ps and have led to the generation of trains of ps pulses e.g. [352,353].Unfortunately, such fast absorbers are relatively dificult to saturate and exhibit high linear loss (i.e. loss at low intensities). For lasers with relatively long upper state gain lifetimes and low gain cross sections, such as Nd: YAG lasers, this leads to Q-switching because initially the loss of the saturable absorber dye is too high to permit cw lasing and the intracavity light is unable to saturate the absorption. Thus the population inversion builds up in the laser amplifier until the gain is high enough to overcome the round-trip losses. When this happens, the iotracavity laser radiation grows very rapidly and is able to saturate the absorption of the dye, producinga mode-locked pulse train. This pulse train quickly extracts all the gain from the amplifier and depletes the population inversion to below the threshold level, thus terminating the mode-locked pulse train. The output therefore takes the form of figure 10. One consequence of the termination of the modelocked pulse train is that the pulses only experience a limited number of round-trips of pulse compression and typically fail to reach the steady-state pulse duration discussed in section 3.1.

Figure IO. Schematic of " M o c k e d

Q-switched pulse train.

In the near infra-red, alternative saturable absorbers to organic dyes can be provided by semiconductor materials, particularly with multiple quantum well (MQW)structures. These exhibit relatively low saturation energies and have been used to generate cw mode-locked pulse trains from solid-state laser media. M Q W structures actually exhibit two distinct absorption recovery times. One is the carrier recombination time (the semiconductor upper state lifetime) which may vary from a few tens of ps to ns, depending on the purity of the material and the precise growth conditions. The second lifetime is of the order of 300 fs and arises from the recovery of the absorption due to the thermdiizatioll of the excitons which also contribute to the absorption. Typically a M Q W absorber with a carrier recombination time of hundreds of ps is readily saturated by the radiation in a cw solid-state laser and cw pulse trains are formed without the pulse energy growing suficient to deplete the gain of the laser amplifier medium below threshold and cause Q-switching. The -1OOps pulses thus formed then have sufficient intensity to saturate the excitonic absorption corresponding to the -300 fs thermalization time and this saturable absorption can then compress the ps pulses down to a few 100 fs or to the limit set by the gain bandwidth. This technique has been successfully

The generatioif of ultrashort laser pulses

209

applied to colour-centre lasers, e.g. 13541, to fibre lasers [355] and to Ti:sapphire [356] and Cr:LisAF 13571 lasers. One interesting variation 011 mode-locking with a fast (semiconductor) saturable absorber is that of resonant passive mode-locking (RPM) which has been used successfully with a number of solid-state laser gain media [358]. This involves using a fast saturable absorber (usually a semiconductor) in a resonant interferometer such as an external cavity or a monolithic thin (sub-wavelength) structure described as an antiresonant Fabry-Perot saturable absorber (A-FPSA) [359]. The mode-locking technique is related to additive pulse mode-locking (see section 3.3.4.2) and has been analysed by Haus et a1 [360]. Although it may be thought of as an interferometric technique based on a resonant non-linearity, the pulse durations achievable are still governed by the absorption recovery time and this is essentially mode-locking with a fast saturable absorber. I t has the advantage that a saturable absorber with a high insertion loss may be used since it is not located in the main laser cavity and only a few % coupling to the resonant absorber cavity is required for successful mode-locking. This weak coupling also has the effect of reducing the thermal loading on the absorber and increasing its effective saturation energy. The latter factor is important because it reduces the tendency of the laser to @switch. Recently there has been much interest in the mode-locking of broadband vibronic solid-state laser media using techniques which siiitulate the action of a fast saturable absorber with an absorption recovery time of <I fs. These techniques exploit the optical Kerr effect to introduce an essentially instantaneously intensity-dependent loss mechanisni in the laser. Thus the modulation function will continue to follow the pulse intensity profile and provide strong compression until the pulse duration reaches its steady-state value which is determined by the gain bandwidth or by other dispersive factors. Such mode-locking techniques can be thought of as exploiting non-rcsonantnon-linearities as opposed to the i.esonant non-linearities associated with the electronic pansitions of conventional saturable absorbers. One important consequence of their non-resonant nature is that they are usefully applicable over almost any desired spectral region. These techniques are discussed in section 3.3.4.

3.3.3. Pussive inode-locking with a sloiv satwable absorber. Passive mode-locking with a doit* saturable absorber does succeed in generating pulses which are shorter than the absorption recovery time of the saturable absorber. This is possible for a special class of lasers in which the pulse shaping is achieved by the combined action of both the saturable absorber arid a saturable aiitplifier. All laser gain media exhibit some degree of saturation in operation-it is this which leads to gain competition and, after many cavity transits, to both pulse compression and linewidth-narrowing. In most lasers, however, the amount of gain depletion per cavity round-trip is relatively small, such that an individual pulse is not able to significantly modify the gain it experiences during a single interaction with the amplifier. The upper state lifetime of the laser amplifier transition (i.e. the gain recovery time) is usually much longer than the cavity roundtrip time and so, once a steady-slate is reached, the gain does not significantly change between each pulse amplification. An important exception to this observation is that of organic dye lasers which have gain recovery times of the order of a few ns, together with high gain cross sections, permitting them to be significantly saturated by single intracavity pulses.

210

P M W French

local tlme

Figure 11. Schematic of passive mode-locking with a slow saturable absorber

Figure I 1 shows schematically how the combined action of saturable absorption and gain can lead to pulse compression. As a pulse circulates in the laser cavity, the leading edge of the pulse is partially absorbed by the saturable absorber before it saturates. I t is contrived, by carefully adjusting the gain and cavity parameters, that after the central portion of the pulse is amplified in the laser active medium, the gain depletion is such that there is insufficient population inversion remaining to amplify the trailing edge of the pulse enough to overcome the intracavity losses. In this manner, the leading and trailing edges of the pulse are suppressed by absorption and gain saturation respectively while the central portion of the pulse is amplified. Thus the induced modulation function is of comparable width to the pulse duration (since the photon flux which saturates both absorber and amplifier is confined within the pulse duration) and the pulses continue to experience strong shaping on each round-trip. This compression mechanism is extremely powerful and has been exploited in cw passively mode-locked dye lasers to generate pulses shorter than 100 fs in duration. The most popular combination of active (gain) and passive (absorber) dyes is that of Rhodamine 6G passively mode-locked using DODCI as the saturable absorber. This was first demonstrated by Ippen el al[361] and has remained the source of the shortest optical pulses ever generated 13621. A refinenlent on passive mode-locking with a slow saturable absorber is to use colliding pulse mode-locking (CPM). This is essentially a technique to increase the effect of the saturable absorber by increasing the degree to which it is saturated. This is normally achieved by adopting a ring configuration so that there are two counter propagating pulses in the laser cavity as shown in figure IZ(a). The pulses collide in the absorber dye jet and saturate it more deeply with the result that the cavity roundtrip loss is reduced. The increased saturation leads to a stronger shaping of the leading edges of the pulses. A saturation enhancement factor of 2 is obtained because each of the counter propagating pulses is amplified in the gain medium and there is a further enhancement o f J2 because the pulses are coherent and it is their resultant intensity that saturates the absorption. There are also thought to be further benefits derived from the population inversion grating which is formed as a consequence of collision of the two coherent pulses [363].

The generation o f ultrasltort laser pulses

21 I

Figure 12. Scliematic of colliding-pulse mode-locking laser cavities.

The CPM technique has been applied to femtosecond ring dye lasers with great success [364]. For some applications, however, a ring geometry is undesirable. In particular it is not trivial to adjust the cavity length of a ring dye laser e.g. for synchronous pumping. An alternative approach is to use the 'anti-resonant ring' (ARR) configuration which was first proposed by Siegman et a1 [365].This is illustrated in figure 3.12(b). The principal difference between the ARR and ring configurations is that only one pulse per cavity round-trip time experiences gain and so the saturation of the absorber is enhanced only by the factor of J2 due to the coherent interaction of the pulses.
3.3.4. Passive inode-locking with /he optical Kerr effect. 3.3.4.I. hrroduction. The optical Kerr effect due to the non-resonant bound electron non-linearity of various media can be exploited in laser cavities to simulate a saturable absorber with a virtually instantaneous absorber recovery. (The response of e.g. silica glass is thought to be < - I fs but this has yet to be determined.) In fact, the picture o f a 'saturable absorber' is not strictly accurate since there is no saturation or depletion of a population difference across a transition. The optical Kerr effect, however, leads to a non-linear intensity-dependent refractive index in the optical elements of the laser cavity, described by
n = no + n2/.
(3.8)

This phenomenon arises from the anharmonic motion of bound electrons in a medium under the influence of an intense propagating electric field. The phase delay experienced by a propagating optical signal is proportional to the refractive index and so the optical Kerr effect causes an optical signal to experience a non-linear intensitydependent phase delay in addition to the linear contribution. Usually this non-linear effect is negligible since nz has the value 3.2x 1 0 ~ ' 6 a n W-' 2 for fused silica and is of the same order for most materials, including gain media, found in laser cavities. For beam waists of -50 jm diameter, a power o f -2.5 MW is required over a propagation

212

P M Mi French
(a) Self Phase Modulation
(b) Self Focussing

A
x

Figure 13. Schcmatic of SPM and self-fKUing

length of I cm to experience a K non-linear phase shift at 800 nm. Inside laser cavities, however, it is possible for ultrashort pulses to reach sufficient peak powers that they experience significant intensity-dependent non-linear phase shifts on each cavity roundtrip. Since the intensity of a pulse will vary across both its temporal and spatial profile, different parts of the protile will experience a different refractive index (i.e. acquire a different non-linear phase shift). If the laser cavity losses are contrived to partially depend on the refractive index experienced by the pulses, either through a (phasesensitive) interferometer or through self-focusing, this can result in an instantaneous irt/oisitl~-f~e~e~inenr loss for the intracavity laser radiation. As discussed above, this corresponds to an almost itfinifelyfas/saturable absorber which provides a modulation function that follows the pulse profile and continues to provide strong pulse shaping on each cavity round-trip until the pulse is compressed to its final steady-state value. The optical Kerr effect is thus a notwesonant non-linearity which therefore has no associated absorption and is essentially wavelength independent. It leads to self-modelocking, or self-locking, through the effects of selfphase rvodulation (sPM)or seF focusing. Both phenomena are a consequence of different parts of an optical signal experiencing different refractive indices and therefore different non-linear phase shifts. srM occurs in lhe temporal domain. If one considers a pulse, it is clear that the peak will be more intense, and therefore experience a higher refractive index than the wings of the pulse. Through propagation, this results in a changing phase, -6 ( I ) , across the pulse profile as showrn in figure 13(a). (Note that -+(I) is written here because the phase, + ( t ) ,is proportional to -kz for a wave propagating in the positive z direction where k = n w / c . ) The differential of phase with respect lo time is a frequency change and so SPM results in pulses with frequency sweeps or chirps. Pulse compression can be obtained by propagating a frequency chirped pulse through a medium with the appropriate group velocity dispersion (GVD)as discussed in section 4. This alone is not strictly mode-locking since there is no effective modulation but the interaction of SPM and GVD is vital for the generation of femtosecond pulses. SPM can lead to modelocking when combined with an interferometer. (An interferometer is a device which essentially converts phase modulation to amplitude modulation.) This approach exploits the intensity dependent phase profile of figure l3(fr) and produces an intensity-dependent loss-the requirement for passive mode-locking. The technique is called coupled

The generation of tiltrashor1 laser pulses

213

L+h 2k

Figure 14. Schematic of additive-pulse mode-locking

cavity mode-locking (CCM) or additive pulse mode-locking (MM) and is described in the next section, 3.3.4.2. SPM always leads to spectral broadening (except in the special case of soliton propagation, and this feature can also be exploited in a weaker modelocking scheme which exploits the Doppler effect from a moving cavity mirror. This is described in section 3.3.4.3. Self-focusing occurs in the spatial domain and exploits the fact that the laser beam profile, usually a Gaussian TEMnomode, experiences a changing refractive index profile across its diameter. This is analogous to propagation in a graded-index lens and results in focusing. If the laser cavity is contrived to have a lower loss (or higher Q) in the presence of this non-linear lens, then this results in an intracavity intensity-dependent loss leading to passive mode-locking. Arising from the optical Kerr effect,this technique is widely known as Kerr lens mode-locking (KLM) and is described in section 3.3.4.4. It is this technique which has produced the shortest pulses to date directly from a laser oscillator.
3.3.4.2. Addiliue pulse inode-locking. Additive pulse mode-locking (APM) exploits the optical Kerr effect to produce an intensity-dependent loss in a laser cavity, thus simulating a fast (instantaneous) saturable absorber. A typical implementation is to use an optical fibre in an external cavity which is coupled to the main laser cavity by a mirror of -80-90% reflectivity. The purpose of the fibre is to provide a non-linear refractive index such that an intensity-dependent phase profile is generated across the pulses from the laser cavity. Other non-linear media may be used but the optical fibre is particularly convenient since it provides a non-resonant (<-1 fs response) non-linearity in a confined geometry (small core area, long interaction length) which produces the necessary non-linear index change for the modest power levels obtainable from c w lasers. Figure 14 shows the essential principle of APM, as originally explained by Mark et al [366]. Each time the pulse in the main laser cavity undergoes reflection at the common mirror (usually an output coupler of -10-20% transmission), it interferes with a pulse from the external cavity with which, if the main and external cavity lengths are matched, it is temporally coincident. If the precise length of the two cavities is set such that the peaks of the two pulses are in phase and so constructively interfere, then, because the pulse from the external cavity has experienced the intensity-dependent refractive index, its wings will be out of phase with those of the pulse from the main laser cavity and so the interference will be destructive (or, at least, less constructive). In this manner, the pulse from the main laser cavity experiences enhancement of its peak and suppression of its wings upon reflection at the common mirror. Thus the external cavity assembly can be considered to provide an intensity-dependent loss which simulates a saturable absorber. It should be noted that, to first order, the GVD of the external cavity is not

214

P M W French

Figure 15. An additive-pulsemode-locked Ti:sapphin laser cavity.

significant since it is only the intensity-dependent phase change which is necessary for compression. Excessive GVD is a problem, however, because strong dispersive broadening in the extemal cavity can limit the final pulse durations achievable. Ironically, although the first A m i laser was the 'soliton laser' [367], for which it was suggested that soliton compression (see next section) was responsible for the mode-locking, it appears that the soliton-like compression that occurs with SPM and negative GVD in a fibre may be unfavourable for APM since the soliton shaping will tend to reduce the non-linear phase change across the pulse profile. APM is an extremely powerful non-resonant mode-locking technique and has been demonstrated with a wide range of laser media including colour-centre lasers 1208, 3681, Ti:sapphire lasers [369,370] and N ~ : Y A G [371], N ~ : Y L [372] F and Nd:glass lasers [373]. Figure 15 shows a typical cavity configuration for an APM Ti:sapphire laser. Initially APM was used in conjunction with active mode-locking to generate femtosecond pulses but it has been demonstrated that APM alone is sufficient for a self-starting, mode-locked femtosecond laser, provided that the laser parameters meet certain requirements [374]. The principal disadvantage of APM is the need to hold a constant linear phase relationship between the pulses in the main laser and external cavities such that the effective reflectivity of the external cavity continues to discriminate in favour of the peaks of the pulses. This constant phase bias necessitates maintaining the cavity lengths to better than a wavelength of light, as should be expected for an interferometric technique, and this generally requires active stabilization. An elegant scheme was demonstrated by Mitschke and Mollenauer [375] who took advantage of the fact that the external cavity is a Fabry-Perot etalon and so its transmission varies sinusoidally with length. By monitoring the output power from the extemal cavity and comparing it to an adjustable reference level, they derived an error singal for an active feedback stabilization scheme which moved a mirror mounted on a piezoelectric transducer. It should be understood that, unlike the situation for active mode-locking, it is not the matching of the absolute cavity lengths which is important but rather the interferometric difference between them. APM can be realized without resorting to an external cavity if two subcavities can be contrived within the laser. One minor extension of the scheme outlined above is to put a beamsplitter in the laser cavity and to terminate it with two mirrors, as shown in figure 3.16(b),with the non-linear element in one arm [376,377]. Theextemal Fabry-

The generation of ultrmhort laser pulses

215

Figure 16. Different cavity configurations for APM.

Perot interferometer is thus replaced with a Michaelson interferometer and the coherent pulse addition then takes place at the heamsplitter. This has the advantage that it is more compact than the conventional system of figure 16(a)and although interferometric stabilization is still required to match the lengths of the two arms, it can be more stable e.g. [378]. A more elegant approach is to replace the Fabry-Perot interferometer with a nonlinear Sagnac interferometer or anti-resonant ring interferometer [379], as shown in figure 16(c). In this configuration the two counter propagating beams in the ring interferometer experience different intensity-dependent phase changes since they experience the gain/loss and the non-linearity in a different sequence. For a Sagnac (anti-resonant ring) interferometer, the reflectivity is given by: (3.9) where &w-&w is the phase difference between the two counter propagating beams. In the absence of non-linearity, the two beams propagate along exactly the same optical path in the same environment and so experience the same linear phase shift. If the counter propagating beams can be contrived to experience different non-linear phase shifts, however, the reflectivity of the Sagnac interferometer will be intensity-dependent according to equation (3.9). This is achieved by placing asymmetrically a non-linear medium and either gain or loss in the interferometer loop. The final amplitude of each beam upon returning to the beamsplitter is thus the same but not the acquired nonlinear phase change since one beam will propagate through the non-linear medium before gain or loss has been applied while the other will encounter it after having its amplitude modified. Thus the non-linear Sagnac interferometer has an effective intensity-dependent reflectivity which can be exploited for mode-locking. The beauty of this approach is that there is no requirement for active stabilization since each beam propagates along exactly the same path and any pathlength changes arising from

216

P M W French

lsdatw

cmtmier

Pdatiuilon

miadsum
CMtlOlW

Figure 17. Linear ( a ) and ring (b) APM fibre lasers exploiting Sagnac interferometers.

perturbations which are slow compared to the transit time of the interferometer loop will affect each beam equally and cancel out. Although first demonstrated in a bulk laser system [380], this approach has been implemented more successfully in fibre lasers using fibre couplers instead of beamsplitters [381] and has produced pulses of hundreds of fs duration e.g. [382,383]. As a further refinement, the linear phase difference between the counter propagating beams (the phase bias) can be conveniently adjusted by varying the polarization state of the beams and exploiting the linear birefringence of the (fibre) interferometer media. In this manner, the fibre Sagnac interferometer can be set to transmit or reflect the low intensity light. This device is sometimes called a non-linear loop mirror (NOLM) and has been demonstrated as an optical switch e.g. [384]. When employed for passive mode-locking, the asymmetry of the non-linear interferometer has usually been achieved using gain from an active fibre amplifier in a configuration described as a non-linear amplifying loop mirror (NALM) [385]. Figure 17 shows two typical cavityconfigurations to implement this type of APM in fibre lasers. The active amplifier fibre ensures that the counter propagating beams in the fibre loop experience different non-linearity before recombining at the coupler with similar amplitudes. The polarization controllers together with the birefringence of the fibre, permit the linear reflectivity of the fibre loop to be adjusted to optimize the mode-locking. I n the linear cavity of figure 17(a), the linear phase difference is set such that the NALM reflects the high intensity light while for the cavity of figure 17(b), the high intensity light is transmitted by the NALM. The purpose of the isolator in figure 17(6) is to absorb the unwanted reflected low intensity light. Another simple approach to APM is to employ two polarization axes inside a laser to define the subcavities, as indicated in figure 16(d). In this case the intensity-dependent non-linear phase change produces polarization rotation which is converted into intensity-dependent loss using a polarization-selective element e.g. [386, 3871. This amounts to replacing the external cavity Fabry-Perot with a Mach-Zehnder interferometer. The technique is particularly useful with fibre lasers and has yielded pulses

The generation of ultrashort laser pulses

217

Amdilkr libre

Pump libreCwple(

Figure 18. Linear ( a ) and ring (b) APM fibre lasers exploiting polarization-rotation,

as short as 38 fs from a Nd-doped fibre laser [388]. As with the previous technique, the phase bias can be adjusted by varying the polarization state of the intracavity light in the intracavity birefringent medium (i.e. fibre). Figure 18 shows two typical cavity configurations for this mode-locking scheme. The linear cavity shown in figure 18(a) incorporates a dispersive delay line to ensure that the overall dispersion of the cavity takes the appropriate value (see section 4). This approach has been adopted for Nd-doped fibre lasers operating at 1.064 pm where the CVD of the fibre is strongly positive and has been extremely successful 1471. With Er-doped fibre lasers operating around 1.55pm, the fibre CVD is typically small and negative and so no external dispersive delay line is necessary. The approach shown in figure 18(b) has been used with Er-doped fibre lasers e.g. [389,390] and has yielded pulses as short as 77 fs [241]. I t should be noted that for fibre lasers mode-locked using APM, any change in the birefringence properties of the fibre will lead to a change in the bias phase and therefore could degrade the mode-locking. Thus environmental changes, e.g. temperature, pressure, stress, can mean that the polarization controllers must be reset to restore the optimum phase bias. T h i s problem is more significant for longer fibre lengths and can be reduced by employing polarization-preserving fibre [392]. APM fibre lasers are usually started by adjusting the polarization controllers until mode-locking is observed, indicating that the correct linear phase bias has been reached. Usually such lasers are then stable and may be self-starting as long as environmental changes do not cause the linear phase bias to drift too far-typically for periods of hours. As no active stabilization is required, this represents a considerable improvement on bulk APM lasers with external cavities. APM, because it is based on the intensity-dependent non-linear refractive index due to the optical Kerr effect, provides a means of passive mode-locking any c w laser, provided that a suitable non-linear medium is used. This can be the laser medium itself or a separate element which is often an optical fibre for the reasons discussed above. Provided that a non-resonant or reactive non-linearity is used, i.e. the non-linear refr'active index is due to deformation of the electronic structures of the medium, rather

218

P M W French

than electronic transitions between energy levels, the effective saturable absorber, or intensity dependent loss, resulting from APM has a virtually instantaneous response time and may be used at any wavelength. The principal disadvantage of the nonresonant non-linearity is that it requires a relatively high intensity in order to induce a significant change in the refractive index and therefore a significant phase change. This is an important issue when considering the self-starting of APM lasers. For passive mode-locking with real (i.e. resonant) saturable absorbers, the non-linearity is usually strong enough to ensure that the passive mode-locking is self-starting i.e. initial fluctuations in the intracavity flux arising from noise or mode-beating are able to induce sufficient non-linear amplitude modulation to evolve into mode-locked pulse trains. For non-resonant non-linear media, the intensity required to achieve the same degree of amplitude modulation is orders of magnitude higher. Nevertheless, self-starting has been observed in many APM lasers (eg. Ti:sapphire, Nd:unc) but not in all (e.g. colour-centre lasers). An important advantage of using an optical fibre as the nonlinear element in an APM scheme is that the induced non-linear amplitude modulation can be readily increased by simply increasing the non-linear interaction length i.e. the length of the fibre. Achieving a threshold of induced non-linear amplitude modulation is a necessary but not sufficient condition for self-starting of APM lasers. lppen et a1 [391] have indicated that, for the initial cavity flux perturbation to grow, it is necessary to avoid gain saturation and they have derived the following condition for self-starting in APM lasers:
K/gPaTp.

(3.10)

This equation considers whether a pulse @ertubation) will experience compression and growth. K is the proportionality constant which relates the change in intracavity flux to the resulting change in gain. It depends on the non-linearity, the losses and the degree of coupling between the two cavities. g is the saturated gain prior to the perturbation and CT is the gain cross section. P is a numerical factor of order unity which depends on the precise pulse shape and T~ is the duration of the pulse or perturbation. It will be seen that a high gain cross section is not favourable for self-starting and this explains why APM colour-centre lasers usually need some active modulation, such as synchronous pumping or loss modutdtion, while Ti:sapphire and other solid-state APM lasers will self-start. For lasers with high gain cross sections, therefore, an additional modulation mechanism is required to initiate the APM. This is usually an active mode-locking technique which can generate pulse trains as short as 100 ps. APM can then compress these pulses to the femtosecond regime. Such lasers can be described as hybridiy modelocked. For synchronization purposes it is often convenient to include some active modulation, clocked to an extemal driving signal, even in lasers which will self-start. I n these cases the modulation depth may be extremely weak since APM is doing the work of mode-locking and the active modulation merely provides a seed. APM is a particularly attractive technique to mode-lock fibre lasers since the long non-linear interaction length ensures that significant phase changes are induced at modest intracavity power levels. The same consideration ensures that for bulk lasers, a fibre is the most attractive medium for the non-linear element. Increasing the fibre length does have some disadvantages, however, since it increases the GVD in the system and also it can lead to excessive induced phase changes. For lasers with fibres in external cavities this is not such a severe problem although it can limit the minimum pulse duration achievable. The success of APM depends on achieving an intensity-dependent loss (reflectivity) which favours high intensities and discriminates against low intensities.

The geireration of ultrasfiort laser pulses

219

Remembering that A m i is an inlerferometric technique, it should be clear that if the relative induced peak non-linear phase change exceeds n, then the pulse wings can experience similar loss/gain to the peaks and the mbde-locking will be less effective. I n general, as the pulses evolve in an APM laser, they compress until the non-linear phase change, integrated over the fibre length, reaches some unfavourable value. The pulse duration at which this happens is related to the GVD of the fibre/non-linear cavity by similar scaling rules to solitons (see section 4) derived from the non-linear Schrodinger equation (NLSE). With excessive non-linear phase change, the pulses can fragment as different parts of their intensity profiles experience the same interferometric attenuation. This is often seen i n erbiumdoped fibre lasers when the pump power is increased above threshold as indicated by the observation of many pulses per cavity round trip at apparently arbitrary time intervals. For a given pump power, and therefore average intracavity power, the pulse energy of the laser is fixed by the number of pulses per cavity transit and the non-linear intensity-induced pbase change for a given length of fibre will be determined by the pulse duration. If this phase change becomes too large, the pulse may break up into several pulses, each with a lower pulse energy. This behaviour occurs in erbium fibre lasers where the long upper-state lifetime (-8 ms) leads to negligible gain saturation by each pulse amplification and therefore there is almost no gain competition between pulses. With gain media of shorter upper-state lifetimes, increased gain saturation and competition produce lasers which tend to operate with a single pulse per round trip or with multiple equally spaced pulses. Under these conditions, excessive non-linearity can limit the achievable pulse duration by causing the laser to become unstable if the pulse becomes too short. A large non-linear phase accumulation is accompanied by spectral broadening. For lasers with a spectral filter, this becomes a source of loss and so the spectral width and pulse duration are clamped to a final steady-state value. With no spectral limitation, however, the pulses will eventually break up in a manner analogous to that of solitons undergoing periodic perturbations [393]. This endgame of APM lasers (and other lasers mode-locked with the optical Kerr effect) is not h l l y understood and is the subject of much current work. It is discussed further in section 4. One can summarize by saying that increasing the fibre length will make an APM laser easier to self-start and mode-locking will be achieved at lower pump powers. On the other hand, to generate the shortest possible pulses, it is necessary to maintain stability and avoid excessive non-linear phase modulation once the pulse durations approach their final value by minimizing the length of fibre.

3.3.4.3. Moving mirror. ntode-locking. It is possible to exploit the self-phase-modulation (SPM) resulting from the optical Kerr effect in a laser and achieve mode-locking without employing interferometry. Using a moving mirror, in an external cavity or in a single laser cavity, can generate short pulses using a combination of the Doppler shift and SPM. This mode-locking mechanism is relatively weak but is useful as a simple means to generate picosecond pulses [394] and to initiate Kerr lens mode-locking in solidstate lasers. Figure 19 shows the two common cavity configurations. Although at first sight, the configuration of figure 19(b) looks similar to APM, the fact that one of the mirrors is moving means that it is not possible to maintain the interferometric phase difference between the cavities. When none of the mirrors are moving, mode-locking is not observed. Furthermore, in contrast to APM, mode-locking is still obtained for significant differences between the two cavity lengths (up to several mm) and no modelocking is observed when the cavities are exactly matched.

220

P M W French

4-c

Figure 19. Typical cavity wnfiguralions for moving mirror mode-locking

For each of the cases shown in figure 1 9 the moving mirror imparts a Doppler shift to the incident radiation. The mirror velocities are typically of the order of 0.1-10 mm s-' and so the corrzsponding Doppler shift is extremely small relative to the laser mode spacing and gain linewidth. Nevertheless, the action of the Doppler shift over many cavity transits is such that the lowest loss cavity modes are constantly being swept out of tlie gain window and new lower loss modes are growing in their place, such that a steady state is never reached. In the absence of non-linearity, this mechanism prevents the laser from evolving to cw operation on a single mode [395]. When nonlinearity such as SPM is present in the laser, however, the radiation can reach a modelocked steady state. This is because the spectral broadening due to SPM can act with the intracavity spectral filter to provide a frequency-pulling eRect that balances the Doppler shift and permits a pulse spectrum to stay in the gain window. srhi is an intensity-dependent process and so this pulse shaping mechanism provides the inlensitydependent loss/gain required for passive mode-locking. Mode-locking using a moving mirror was first demonstrated in a single cavity with a helium neon laser [396, 3971 and later, using a linear external cavity, with a Ti:sapphire laser [394]and Nd:fibre lasers [398,3991. Although pulses as short as a it is only recently that the above few picoseconds duration have been generated [400], explanation has been proposed for this mode-locking mechanism [401]. When a linear external cavity with a moving mirror is employed, the situation is complicated by the fact that the external cavity acts as a Fabry-Perot structure whose length, and therefore mode-spacing, is changing on a slow time scale. This linear Fabry-Perot acts as a complex output coupler with an effective reflectivity in a manner analogous to the nonlinear external cavity in A r r d . Typically, for the mirror velocities used, the external cavity changes its length by a wavelength in a few hundred cavity round-trips. This produces a laser whose wavelength changes on the same timescale, as experimentally verified in [400].Sargsjan et nl 13991 suggest that the external cavity enhances the frequency shift on each transit (originating from the Doppler effect) by a factor of 10100 and therefore each laser mode i s frequency-shifted to the next laser mode in a few tens of cavity round-trips, thus producing a weak mode-locking mechanism. They go on to suggest that tlie action of SPM in the main laser cavity can adjust the optical length of the main cavity and therefore its mode-spacing, such that the laser radiation changes its wavelength to one that experiences a higher reflectivity a1 the external cavity. This therefore would result in an intensilydependent loss and passive mode-locking. It should be understood that although this moving mirror technique requires an external signal to drive the mirror motion, it still is a passiue mode-locking technique. This is clear when one considers that it is possible to change the cavity length and still achieve mode-locking with the same driving signal. It was found experimentally that

The generation of ultrashort laser pulses

22 1

by changing the lengths of the single laser cavity (or, in the case of the coupled cavity system, the lengths of the main and external cavities), it was possible to generate picosecond pulse trains with G H z repetition rates-something which is not easy to do with active mode-locking for which there are problems associated with generating RF signals of sufficient power to drive modulators, with transmitting these R F signals and with coupling the energy efficiently into the modulators. The principal disadvantage of moving mirror mode-locking is that the output pulse trains are not truly continuous since the moving mirror must oscillate about some position and therefore come to rest during its cycle of motion. At these points, the mode-locked pulse trains disappear. Also, because the mirror velocity is not constant throughout its cycle, the pulses evolve in amplitude, duration and degree of coherence. Most c w laser diagnostics integrate over at least several ms and so the resulting data represent a time-average of the laser output. This makes if difficult to determine the precise properties of the laser pulse trains. In a practical system, the moving mirror is mounted on a loudspeaker, mechanical shaker or piezoelectric transducer and oscillated at frequencies of u p to 100 Hz over

Figure 20. Typical cavity for a Ti:sapphire mode-locked with a moving mirror.

amplitudes up to a Few mm. Figure 20 shows a typical configuration for a Ti:sapphire laser mode-locked in this manner. This mode-locking technique is useful as a mechanism to initiate Kerr lens mode-locking [402] by generating picosecond pulses which will then have sufficient intensity to produce a non-linear lens as described in the next section.

3.3.4.4. Kerr ler7s nzodc-locking. The shortest pulses directly generated from any laser oscillator have been produced by Kerr lens mode-locking (KLM)in a Ti:sapphire laser with sub-20 fs pulse durations being routinely achieved in several laboratories e.g. [403-4061. The mode-locking mechanism is again a fast saturable absorber based on the optical Kerr effect but, unlike APM, it is not an interferometric technique and so there is a much relaxed requirement for stabilization of the laser cavity. For this technique, the intensity-dependent loss arises from self-focusing in a non-linear medium. Thus the non-linearity is still SPM but in this case it is the transverse non-linear phase profile of the beam, rather than the temporal profile, that produces pulse compression. The temporal phase profile of the laser signal is, of course, also modified by SPM in a manner appropriate for soliton-shaping (see section 4) but this is not the primary pulse compression mechanism. Passive amplitude modulation arises from the fact that intense intracavity radiation can induce a non-linear, or Kerr, lens and the cavity can be contrived such that the net round-trip loss is lower in the presence of this lens than without it. Thus a pulse circulating in the cavity will experience a lower loss, or a higher gain, than a less intense pulse or a cw signal. KLM was first experimentally observed in a Ti:sapphire laser [407] and first explained in terms of self focusing by Piche [408] and

222

P M W French

Fiyrc 21. Schematic of Kerr lens mode-locking. Solid output beam is high intensity pulsed radiation, dashed beam is low intensity radiation.

Negus el a1 [409]. Self focusing had previously been proposed as a mechanism for Q-switching and pulse compressions e.g. [410,41I]. Figure 21 shows the basic principle of Kerr lens mode-locking using lenses as the focusing and collimating elements rather than curved mirrors which would be found in a typical laser cavity. I t should be seen that for high powers (pulsed radiation), the action of self focusing is to produce a narrower beam waist at the focus of L, than would otherwise be obtained and, for the asymmetric configuration shown above, to partially collimate the output beam (shown by the solid line). The second lens, L2,is set to collimate this high power beam. The lower power (cw) radiation is not focused to such a small beam waist in the laser rod and is not fully collimated after passing through the second lens, LI. Differential gain for the high and low power radiation can be obtained in two ways. For the case where the pump waist is smaller than the c w laser radiation waist, as shown in figure 21, then self focusing improves the overlap between the laser cavity and pump modes and so increases the gain extraction. This is sometimes described as gain-guidirig.The presence of the aperture after the second lens increases the relative loss for the less collimated low power beam. This is described as KLM with a hard aperture-as opposed to gain-guidingwhich may be called a soft aperture. The optimization of KLM laser cavities has been the subject of much investigation e.g. [412-4151 and is now reasonably well understood. Usually the self focusing occurs in the laser rod and so both of the above mechanisms provide an intensity-dependent gain/loss modulation. For lasers in which this is not convenient, perhaps because the gain media are too short for significant self focusing to occur or because the gain media is not solid-state. il is possible to use an additional non-linear element in the laser cavity to provide self-focusing [416,417]. I n this case it is usually necessary to employ a hard aperture to discriminate against the lower intensity light since gain-guiding will not be available in the self focusing medium. Figure 22 shows a schematic of a typical K L M Ti:sapphire laser in which the laser rod acts as the self-focusing element. The prism pair is to provide adjustable intracavity GVD as discussed in section 4. KLM is clearly a mode-locking mechanism for bulk lasers since it is not possible to exploit self focusing in a fibre. Compared to APM, the relatively short interaction length means that relatively high powers are required to generate sufficient non-linear phase change for a significant amplitude modulation and so KLM lasers require higher pump powers and mode-locking is harder to start. Usually, KLM lasers are not self-starting since the initial intensity fluctuations in a laser cavity are not strong enough generate sufficient non-linearity to evolve into stable pulse trains. An additional mechanism is therefore required to initiate mode-locking and form pulses which ace then sufficiently short and energetic to access the optical Kerr effect non-linearity and induce K L M . Many different modulation schemes have been used to start the mode-locking in lasers

The generution of ultrashort laser pulses

Aperture

223

,1

M d

Figure U. Schematic of Kerr lens mode-locked Ti:sapphire laser. Solid b e m is high intensity pulsed radiation, dashed beam is low intensity radiation.

with KLM including acousto-optic modulators e.g. [418,419], regenerative mode-locking e.g. [420], synchronous pumping e.g. [421], saturable absorption using a dye e.g. [422], or a semiconductor MOW absorber in a resonant external cavity (RPM)e.g. [423], or in the main laser cavity e.g [424], or moving-mirror mode-locking with an external cavity e.g. [425]. Moving a mirror in the single laser cavity systematically using a piezoelectric transducer or simply tapping a mirror, translating an intracavity prism or scanning the laser cavity length using tilt plates have all been found to initiate mode-locking. To date there is one report of a self-starting KLM femtosecond Ti:sapphire laser [426] which exploits a self-focusing lens due to gain saturation. This works because the threshold power to achieve self-starting (see next section) is much lower for this resonanl non-linearity than it is for the optical Kerr effect. Unfortunately, this has only been achieved near the optimum wavelength of 840 nm.
3.3.4.5, Se(f-srurfing ofpassively inode-locked lnseus. Lasers which are passively modelocked with fast saturable absorbers rely on intensity-dependentloss to provide a passive amplitude modulation or pulse compression. Once a laser is switched on and freerunning, it is necessary for there to be a fluctuation of the intracavity radiation, i.e. a noise spike, of sufficientintensity to generate a sufficiently strong modulation for modelocking to commence. In principle, the action of gain competition should mean that any intensity-dependent loss will result in the most intense fluctuation winning all of the gain and evolving into a mode-locked pulse train. In practice, however, this does not always happen due to competing mechanisms which tend to disperse the pulses, and to dynamic gain saturation which can suppress the growth of pulses [391]. It has been proposed and demonstrated [427] that if a fluctuation does not experience significant compression within a cavity mode coherence time, then it will not survive and grow but will be dispersed. This sets a power threshold for self-starting which depends on the strength of intensity-dependent modulation. For lasers passively mode-locked with the optical Kerr effect, the low value of the non-linear coefficient means that a relatively high power is required, compared to that needed for real saturable absorbers, to induce a sufficient passive amplitude modulation for mode-locking. For this reason, many such passively mode-locked lasers are not self-starting although the mode-locking mechanism is self-sustaining since the passive amplitude modulation is strong for an ultrashort (and therefore intense) pulse. It is relatively easy to achieve self-starting in APM lasers because of the relatively long non-linear interaction length of the optical

224

P M IVFrench

fibre and so gain saturation is the significant factor for the self-starting of APM lasers, as was discussed in section 3.3.4.2. KLM is generally applied lo solid-state lasers with low gain CI-oss sections for which gain saturation is not a problem. To achieve selfstarting, it is therefore desirable to maximize the mode coherence time. It has been shown that this coherence time is reduced by the presence of spurious reflections in the laser which set up subcavities with competing unevenly spaced mode structures [428]. This effect may be greatly reduced by adopting unidirectional ring cavities as has been demonstrated for both a low power APM fibre laser and a picosecond KLM Ti:sapphire laser [429]. Spalial hole burning can also decrease the coherence time [430] and so should be avoided. In practice, however, there is not yet any simple scheme to achieve self-starting KLM without compromising the performance of the laser in terms of pump power, output power and/or tunability. This remains one of the challenges for the next generation of usefulultrafast lasers for real world applications. As mentioned in section 2.4, recent work [431] on symmetric cavities indicates that self-starting may be achieved at modest pump powers, If this proves feasible with diode-pumped KLM lasers, simple, low cost femtosecond sources may be anticipated.

3.4. Iipbvid inode-locking

It is possible to combine the action of a saturable absorber and loss or gain modulation in tlie same laser system. This is referred to as hybrid mode-locking and represents a compromise between active and passive mode-locking, enjoying some of the advantages of both approaches but sharing some of the difficulties-notably the need to match the cavity length to the driving signal frequency according to equation (3.4). One advantage is that the use of a synchronous pump or modulation often provides a convenient means to synchronize the laser and other devices to a master clock. The most common examples of hybrid mode-locking are synchronously-pumpeddye lasers which incorporate saturable absorber dyes, syncluonously-pumped (electrically) semiconductor lasers which incorporate semiconductor saturable absorbers and solid-state lasers which employ a secondary mode-locking mechanism to initiate passive mode-locking which exploits the optical Kerr elfect (e.g. A P M , K L M ) . These are all discussed in section 2. In most cases it does not represent a distinct class of mode-locking and will not generate shorter pulses than are obtainable with passive mode-locking. For c w dye lasers, however, as discussed in section 2.3, the synchronouspumping of a dye laser containing bolh a saturable amplifier and saturable absorber dye enabled various dye laser systems to be mode-locked by the action of gain and absorption saturation which would not otherwise be possible. The synchronous pumping produced a reasonably intense pulse which could more strongly saturate the amplifier/absorber media than the intensity fluctuations of the free-running laser. This permitted the use of dye combinations whose saturation parameters precluded their operation in a standard laser passively mode-locked with a slow saturable absorber. This technique therefore extended the spectral coverage of cw mode-locked dye lasers. I t also tended to produce higher pulse energies than pure passive mode-locking. See [94] for a review of hybridly mode-locked dye lasers. J u general, the performance of these lasers was not as good as purely passively mode-locked dye lasers owing to the instabilities i n the matching of the pump and slave laser cavity lengths and instabilities in tlie actively mode-locked pump lasers. The presence o f the saturable absorber, however, did tend to suppress the spontaneous emission background which is a source of noise and

The generation o f ultrashort laser pulses

22s

instability in the purely synchronously mode-locked dye laser. The best hybridly modelocked lasers functioned much like standard passively mode-locked CPM dye lasers and yielded pulses as short as -30 fs e.g. 14321. More often, however, the pulses were longer and often the autocorrelation traces exhibited exponential wings, suggesting fluctuations in the pulse profiles. This may have been partly due to the large amounts of non-linear frequency chirp arising from self-phase modulation of the dye laser pulses (which tended to be significantly more energetic than their purely passively mode-locked counterparts) and cross-phase modulation due to the intense pump pulses. Hybridly mode-locked dye lasers were not significantly developed after the arrival of femtosecond solid-state lasers and their operation has not been fully explained. I n the context of ultrafast solid-state lasers, hybrid mode-locking usually refers to the presence of an active mode-locking mechanism which is used to initially generate pulses of sufficient intensity to access the optical Kerr effect and induce a significant passive amplitude modulation in the laser. As discussed in the previous sections, KLM is not generally self-starting and so a variety of supplementary active (and passive) modulation techniques have been employed. The regenerative mode-locking approach with an acousto-optic modulator is a particularly elegant way to initiate KLM and maintain mode-locked operation in the presence of significant perturbations.

4. Role of group velocity dispersion and self-phase modulation in ultrafast lasers

4.1. Group uelocity dispersion

4.1.1. Introduction. The control of group velocity dispersion (GVD)inside the laser cavity was the crucial step in the development of femtosecond lasers. 14331 gives an elegant discussion of the role of GYD in ultrafast lasers as does [434]. Ultrashort pulses have correspondingly broad spectral profiles and are tberefore sensitive to dispersive phenomena. One observable consequence of GYD is that the different frequency components of an optical signal will propagate at different speeds through a dispersive medium and this wilt typically lead to temporal broadening. There will be a resulting distribution of frequency components throughout the temporal envelope which is described as chirped. A positive chirp corresponds to an observer experiencing an increase in frequency with time and a negative chirp corresponds to a decrease in frequency with time. Figure 23(a) shows the electric field of a pulse with a positive chirp or upchirp and figure 23(b) shows a negatively chirped or downchirped pulse. It should be understood that while GVD changes the temporal profile of an optical signal, e.g. an ultrashort pulse, it does not change the spectral profile. It merely adjusts the relative temporal positions of the spectral components. Silica glass, the standard optical medium, is said to exhibit normal dispersion for visible light. This corresponds to the situation where higher frequencies travel slower than lower frequencies. An initially unchirped (i.e. transform-limited) optical pulse will therefore becomepositioely chirped (upchirped) after propagating through a medium with norrmd dispersion. Conversely, the situation where higher frequencies travel faster than lower frequencies is described as anomalous dispersion. Propagation of a unchirped pulse though such a medium will produce a negative frequency chirp (i.e. dolo,vnchiry). Note that the effect of propagation through a dispersive medium can, in principle, be reversed by propagation through a medium of the opposite dispersion.

226

P M W French

Figure W . Schematic of the electric field of a pulse with

(U)

positive frequency chirp and

( b ) negative frequency chirp.

CVD can arise from the fact that the refractive index of a medium n ( m ) , varies as a function of frequency. Thus the wavevector k , the phase velocity ( m / k ) , and the group velocity (do/&), are also functions of frequency. This is a material property of the medium and may be described as material or chromatic dispersion. It can be shown (see for example 14351) that for every absorption (emission) transition in a medium, there is a corresponding dispersion profile. These correspond to the real and imaginary parts of the complex wavevector and are related by the Kramers-Kronig transformation. In most real media, there are a large number of transitions and the refractive index at any frequency results from the superposition of the dispersion profiles of all the absorption transitions. In particular, for silica glass, the spectral properties of the medium throughout the window of transparency are such that the material dispersion is trorn7al below -1.28 pm and anorrialous above this wavelength. Most other optical materials exhibit similar dispersive properties. Figure 24 shows the change in the temporal and spectral profiles of a Gaussian pulse after propagating through a medium of length Land refractive index n(m). Notice that the temporal profile changes width but the spectral profile is unchanged. Initially the pulse is assumed to be transform-limited (unchirped) and so the time-bandwidth product is given by 0.441 (for the case of Gaussian pulses). After propagation through the dispersive medium, the pulse remains Gaussian but becomes chirped and temporally broadened. The time-bandwidth product therefore exceeds the limiting value of 0.441.

qn .AV = 0.441

qn . AV w 0.441

Figure 24. Scliematic orthe broadening of a Gaussian pulse due to dispersion

The generation of ultrasliorl laser pulses

221

The effect of GVD on a light pulse can be simply considered in terms of the dispersion of the group delay o f the different frequency components. The group delay of the light as a function of angular frequency after propagation through the medium of length L is given by T(w)where

u,=dw/dk is the group velocity and k is the real part of the complex wavevector which is related to the refractive index by:

Note that the phase change experienced by an electromagnetic wave of the form on propagating through a medium of length L is given by #(a) = -wn(w)L/c. The group delay is thus given by -d4/do. For a pulse of bandwidth Aw=2aAv, the spread in propagation time of the frequency components of the light pulse is given by:
eil~~-h:)

d A t = L -( T ( o ) ) A o . do

(4.3)

The pulse broadening/unit length/unit bandwidth is therefore given by:

Dispersion is often defined as pulse broadeninglunit bandwidth and is given by: At - L d ( T ( w ) ) = L-=-d2k -Aw dw do2 d2f#J da2

(4.5)

d2k,dw2>0 (and d2f#J/do2<O)for normal dispersion, which produces a positive frequency chirp. Normal dispersion is often (but not always) described as positive dispersion. Anomalous dispersion therefore corresponds to the case when d2k/do21 0 (and d2f#J/dw2>0). GVD is often defined with respect to wavelength and in particular is characterized using the dispersion parameter, D (GvD,unil length). For optical fibres it is common to quote dispersion (i.e. D ) in units of ps/(nm km). Thus the D parameter is positive for normal dispersion (i.e. d2k/dw2>0).

One can thus use equalions (4.5) and (4.6) for a rough estimate of the degree of broadening experienced by a pulse once the dispersion of a medium is known. For a Gaussian pulse, however, it is straightforward to calculate the change in pulse profile more precisely. The frequency-dependent phase shift, f#J (w),experienced by the light propagating through the medium may be expanded about the central frequency w o using the Taylor expansion:
f # J

(0) =

1d2f#J $ (w - W O ) 3 + . . . 4 (WO)+-ddJ (a,- wo)+ (w --Wu)2+- I #-

dw

2 dw2

6 do

(4.7)

228

P M WFre,rcl1

For most situations, it is only necessary to consider terms up to dz$/dw2 and that is what is generally meant by 'dispersion' in the context of ultrafast lasers. For extremely short pulses (<-SO fs) with correspondingly broad bandwidths, the third order dispersion term may be significant. Also, at the wavelength for which d 2 $ / d d = 0 , the third order term is significant. This is particularly relevant for thecase of optical transmission in silica glass fibres near -1.3 pm. A Gaussian pulse of width rin and central frequency @~,=27rv,, can be described by:
&n In 2iZ,~GL eimor

(4.8)

To calculate the effect of dispersion, equation (4.7) should be Fourier-transformed to the frequency domain and the appropriate contribution from equation (4.6) should be added to the phase. Transforming back to the time domain gives the resulting output (dispersed) profile which, as shown in [433] produces a frequency chirp of opposite sign to d 2 r $ / d d and a pulse broadening given by:
If2

where This simplifies to

d2r$ @'=(s) and p=8 In 2

1:.
L

(4.9)

(4.10)

which may be rewritten as:

(4.1 1)

where
(4.12)
LO i s the characteristic 'dispersion length', defined as the length over which the pulse duration increases by a factor of ,12. It can be considered as the propagation length of a system over which dispersion plays a significant role. Thus it should be seen that the typical effect of propagation through a dispersive medium is to temporally broaden pulses. h i d e a laser cavity, dispersion can impose a limit on the minimum pulse duration achievable. In section 3, the final steady-state mode-locked pulse duration was said to be fixed by a balance between temporal compression, due to the mode-locking process, and spectral compression, due to the effective intracavity filter. In the presence of significant intracavily dispersion, the effective intracavity spectral filter will still limit the steady-state spectral width of the pulse but the corresponding pulse duration will be determined by the intracavity dispersion. For the optimum value of intracavity GVD (close to zero), the ultrashort laser pulses will be unchirped and the duration will be the minimum (i.e. transform-limited) value which can be supported by the spectral width. The spectral width itself may be determined by the balance between the intracavity filter and the mode-locking process. It should be noted, however, that for vibronic solid-state lasers such as Ti:sapphire, the gain bandwidth is so broad (> 100 nni) that the pulse spectrum never reaches the limiting

The generution of ultrashort laser pulses

Table 2.
GVD

229

conlribulions of some common ultrafast laser elements

dQ/dnr (fs)

Dispersive element
I mm fused silica at 600 nm I mm flint glass 600 nm 20 mm Ti:sapphire at 800 nm

Single-stack dielectric mirror Double-stack dielectric mirror Four prism sequence at 600 nm ( I =25 cm) Grating pair at 600 nm (b= IOcm)

- 50 -1200 Up to -300 Up lo -8000 u p IO -350

-60,000

- 10

value imposed by the cavity filter. Instead, the minimum pulse duration is fixed by the onset of instabilities which prevent further pulse compression in the laser. These instabilities arise from non-linear optical processes which become increasingly important as the duration reduces and therefore the peak power increases. High order dispersion (d34/dw3 and even d44/do4) also become important for pulses of such short duration (and therefore such broad bandwidths). Nevertheless, by careful optimization of the intracavity dispersion, it is possible to generate pulses of less than IO fs duration in lasers with very large gain linewidths. To generate such short pulses, it is vital to be able to set the intracavity GVD to its optimum value. This may be done by employing a combination of media which produce the required net value of intracavity GVD. In practice this is usually not possible and it is necessary to employ optical structures whose GVD may be adjusted.
4.1.2. Sources of
GVD

4.1.2.1. Introduction. Any element through which light propagates will be a source of material dispersion. In a laser cavity this typically includes the gain medium, any

saturable absorber medium, any transparent elements such as tuning wedges, prisms, modulators and even the air path between components. The resulting dispersion may therefore be calculated as:

(4.13)
CVD may also result from interferometric components such as dielectric mirrors and Fabry-Perot filters when encountered off-resonance. Table 2 gives very approximate estimates of d24/do2 in fs2 for several common (intracavity) laser elements. This quantity can be readily calculated from tabulated values of dzn/dk2 using equation (4.13). GVD only becomes a probelm when the pulse spectrum is of the order of several nm (or the pulse duration is of the order of -100 fs). For picosecond lasers CVD is usually therefore not an issue and pulses shorter than 200 fs were frequently generated in mode-locked dye lasers before the role of GVD was well understood-simply because the laser pulses did not propagate through extended media. As a very rough rule of thumb, a value of d2q5/do2=-200 fs2 is sufficient to prevent a laser generating pulses much shorter than -100 fs duration. Equation (4.8) (following [I]) indicates that the strength of the pulse broadening tendency due to GVD decreases with the fourth power of the pulse duration.

4.1.2.2. Dielectric ntirrors. Multilayer dielectric mirrors achieve high reflectivities

through the interference of multiple reflections within the mirror structure. There is an

230

P M WFrcncii

accompanying phase shift, @ (a), which is a source of GVD. The realization of the crucial role of GVD in femtosecond dye lasers more or less coincided with the identification of the dielectric mirrors as the major sources of intracavity dispersion [433, 436, 4371. A single-stack high-reflecting (HR) dielectric mirror used on resonance contributes negligible CVD but as soon as it is used away from resonance there is a significant contribution. When the incident light is within -5% of the resonance frequency, then the reflectivity is still close to l O W ?and the dispersion contribution will be less than -50 fs'. Thus standard single-stack HR dielectric mirrors employed close to normal incidence are usually sufficient for lasers generating pulses as short as -50 fs. When such mirrors are used such that they are partially transmissive (e.g. a few %) they can contribute dispersion values of >200 fs'. This can occur when these mirrors are used either at significant angles of incidence, or with light at frequencies significantly detuned from resonance, to facilitate spectral control of lasers in order to adjust their filter width or centre wavelength [438,439]. Output couplers often exhibit higher dispersion than HR mirrors and must be carefully designed to minimize their GVD contribution. Multiplestack mirrors can contribute very high dispersion if the light must pass through the upper stack before being reflected at the lower one. The dispersive properties of dielectric mirrors may be advantageously employed to control the net GVD of a laser cavity. For visible lasers, most of the media contribute positive (normal) GVD and so a mirror which contributes negative (anomalous) GVD may be used to bring the net value close to zero. A single-stack dielectric mirror used at frequencies below resonance typically exhibits negative GVD. Thus a dielectric mirror coating may be designed for a specific laser system or a standard dielectric mirror may be used away from normal incidence so that the effective layer thickness is decreased and therefore the effective resonance frequency is increased. In this manner an appropriate net value for d2q5/doz may be achieved [437,438,440,441]. A further level of sophistication may be achieved with the dielectric mirror coatings being designed to compensate for both the dz@/da2 and the d3@/dw3 contributions of the other intracavily components, e.g. [296, 4411.
4.1.2.3. Gires-Tournois interferometer. Any interferometer used off resonance will be dispersive and so may be employed to control GVD. A dielectric mirror is an example of particular relevance to ultrafast lasers. Fabry-Perot interferometers may also be used but the requirement that they operate off resonance necessarily makes them lossy. The Gires-Tournois interferometer (on) is a special case of a Fabry-Perot interferometer in which one mirror is of 100% reflectivity [442]. A schematic is shown in figure 2S(n).

(a) (b) Figure 25. Schematic of ( 0 ) the structure of a Gires-Toournois interferometer and (6) group delay as a runction of frequency.

The transmission of the CTI is independent of wavelength but the group delay, d@/do, which is illustrated in figure 25(b), resembles the transmission function of the

The generation of ultrashort laser pulses

23 I

conventional interferometer. The quantity to is the transit time of the interferometer given by to= 2n dcos S/c. The dispersion of the GTI, d2$/dw2, is obtained by differentiating the group delay with respect to frequency and is given by: d2$ - - 2ti ( I - rz)r sin at0 do2 (1 + 2 - 2r COS ~ t o )
--

(4.14)

I t should be clear that, for a given frequency CO, it is possible to obtain either positive or negative GVD by varying the value of t o . This can be conveniently adjusted by changing either the angle of incidence or the interferometer spacing [433,444]. Increasing to increases the dispersion but it also increases the finesse of the group delay function and so reduces the frequency range over which d2b/dw is constant. For femtosecond pulses this restriction makes GTIS less attractive than dielectric mirrors or prism sequences but they have found application in picosecond lasers 14451 and are now used in commercial Ti:sapphire lasers. In principle it is possible to exploit the non-linear frequency dependence of the group delay function of the GTI to compensate for higher order dispersion, as suggested in [444] but this approach has not yet yielded shorter pulses than those obtained with other techniques.
4.1.2.4. Prism sequences. Once intracavity GVD had been identified as the crucial parameter in sub-100 fs lasers, the development of the prism sequence providing adjustable dispersion permitted systematic analysis of the phase modulation in ultrafast lasers and the generation of pulses of less than 30 fs. Previously the role of dispersion had been considered in mode-locked lasers e.g. [446], and the diffraction grating had been proposed as a means ofproviding negative GVD [447]. This work indicated the relationship between angular dispersion and negative GVD which was later generalized [448] and applied to prism pairs [449]. In general, it can be shown, e.g. [434], that angular dispersion produces a GVD contribution:

(4.15) where L is the distance from the input to the output of the dispersive system. For pairs of elements (prisms, gratings) the first element provides the angular dispersion and therefore the GVD. The second element recollimates the spectral components. Using two pairs of elements permits the lateral displacement of the spectral components to be cancelled out and recovers the original beam profile. Figure 26 shows how pairs of prisms may be employed in a laser cavity to provide adjustable GVD. The first prism angularly disperses the incident beam and the second prism acts to collimate the rays. The angular dispersion produces a negative GVD contribution which is proportional to the prism separation, 1 and to (dnldw). There is also a positive GVD contribution due to the material dispersion of the actual glasspath the laser beam takes through the prism sequence. This contribution is proportional to dnldo and to the length of glass-path. By translating one of the prisms along its axis of symmetry, it is possible to change the amount of glass-path, and hence the positive GVD contribution, without changing the beam direction. In this manner, the total GVD of the prism pair may be conveniently adjusted. For lasers with linear cavities, a single prism pair may be used with a retroreflecting mirror, as shown in figure 26(a). This has the disadvantage that any output beam through the retroreflecting mirror will be spatially chirped due to lateral beam walk-off i.e. there will be a spatial distribution

232

P A4 I Y Frencli

Figure 26. Schematic o f prism pairs for variable

CVD

of frequency components and the beam will no longer have a circular profile. Alternatively, a sequence of two prism pairs may be employed which will remove all spatial chirp and leave the laser beam undeviated. Such a prism sequence may be inserted in any part of a laser provided that there is sufficient length, The spatial distribution of the frequency components may be exploited by inserting appropriate phase or amplitude masks in the region between the second prism and the mirror of figure 26(a), or between the middle two prisms e.g. figure 26(h). A simple aperture provides a frequency filter which may be used to tune the laser or to restrict the bandwidth. More complicated masks may be used to manipulate the phase of the laser pulses to synthesize complex pulse shapes or to correct higher order dispersion. For the case of four Brewster angled prisms, the total GVD arising from both the angular and material dispersion is given by [449] :

where p is the angular deviation of the dispersed light from a reference ray drawn from the apices of each prism. The first term is a positive (D>O) contribution to the GVD and the second is negative. Usually the angle p is small and so cos p 1. The quantity /sin p determines the amount of glass-path in the prisms and must be of the order of twice the beam diameter (-2 mm) io contain all of the laser beam. In practice, for fused silica prisms, a spacing of -13cm is sufficient for the negative GVD term to balance the positive contribution, With a separation of -25 cm, it is straightforward to translate one of the prisms and adjust the net G V D of the prism sequence from positive fo negative values corresponding to propagation through several mm of glass. Note that the negative GVD contribution of equation (4.13) may be readily obtained [434] using equation (4.12) and the relation:

(4.17)

The generation of ultrashort laser pulses

233

I n principle one can obtain almost any amount of negative GVD using this technique and so set the net cavity GVD to zero. In practice, there are restrictions on the cavity lengths of most lasers and prism separations exceeding -50 cm are impractical. For solid-state lasers, where the G V D arising from material dispersion of the gain medium can exceed that o f 20 mm of glass, this difficulty can be overcome by using prisms constructed from dense glasses such as SFIO, F2, BK7 [450,451]. This works well for pulse durations down to -30 fs but for shorter pulses, the relatively higher third order (d#/dw3) GVD contributions prevent shorter pulses from being generated. The best remedy is to design the laser to have the smallest possible G V D contributions from the components such as the gain medium etc, and to use a fused silica prism sequence to fine tune the dispersion. This approach necessitates using gain media of only a few mm in length.

4.1.2.4. Dtfiaction gratings. Diffraction gratings provide CVD [447] in a similar manner to prisms. They have the advantage that they are much more dispersive and d o not exhibit material dispersion. It will be seen from table 2 that for even a modest grating separation, the amount of GVD provided is many orders of magnitude greater than for prisms. Their disadvantages include higher insertion loss which makes them suitable only for very high gain lasers such as fibre lasers e.g. [452]. They are widely used outside laser cavities to compensate for dispersion in optical fibres and to achieve pulse
I

I I

Figure 21. Schematic of a diffraction grating pair for variable negative CVD.

compression. Figure 27 shows a typical grating pair. As with the prism pair, the device may be double-passed or used with a second pair in order to spatially recombine all the spectral components and recover the beam profile, thereby avoiding the non-circular beam profile and transverse chirp caused by lateral beam walk-off. In some compact arrangements, only one grating is used and is multi-passed using mirrors retroretlecting corner cubes or prisms. The dispersion may be simply calculated [434] using equation (4.12) and replacing the parameter I with b/cos a. The grating equation for the first diffracted order (sin a + s i n p = 2 a c / o d ) may be used to calculate (da/dw) yielding:
_ dz= 4_

do2

4lrcb msd2cos3 a

where

(4.18)

This simple expression is for plane waves. A more complete treatment which considers the case of finite beam size for both prism and grating pairs is given in [453]. Diffraction gratings are ideally employed in the Littrow configuration for maximum transmission efficiency (i.e. lowest loss) and in principle blazed gratings can achieve 90% transmission. I n practice it is often convenient to adjust the CVD by fine tuning the angle of the gratings and so the efficiency is usually rather lower.

234

P M W French

*
fl

_fl

-_
f2

-_

fz

Figure 28. Schematic of a diffraction grating pair with telescope for variable positive CYD.

It is also possible to obtain positive CVD from grating pairs (and prism pairs) by placing a telescope between the gratings (prisms) [454] as shown in figure 28. This + Mz,)where M=fi/L elegant concept produces a CVD which is proportional to (il is the magnification o f the telescope. zI and z2 are the distances from the gratings to the focal planes of the input and output lenses respectively. The retroreflector is used to double pass the grating pair to cancel the lateral beam walk-offand transverse chirp. In some configurations the lenses are replaced by curved mirrors and additional folding mirrors may be employed to achieve a compact design. This device was developed to compensate the negative dispersion of tens of km of optical fibre at wavelengths above the zero dispersion wavelength and also to temporally expand pulses by factors of several 1000 in order to reduce the peak power for chirped pulse amplification (CPA) schemes [455].
4.2. Self phase modulation

Optical phase modulation occurs when the refractive index o f a medium is modified by an optical signal which is propagating through it. If the refractive index is modified on a timescale comparable to or faster than the optical signal, then it will acquire a temporal phase profile. When the refractive index is modified by the signal w,hich experiences the phase modulation, it is described as selfphase modiilufion (SPM). If the refractive index changes are caused by a separate optical signal then the phenomenon is described as cross-phase rnodulution (CPM or XPM). Optical phase modulation is a non-linear effect and so requires relatively strong optical signals. I t can arise due to gain or absorption saturation where the refractive index of a gain/absorber medium changes as the population inversion changes, according to the Kramers-Kronig relationship, or it can arise as a consequence o f the optical Kerr effect e.g. [2,434]. For the (most common) situation when the optical signal is changing much more slowly than the response time of the non-linearity, one can think in terms of an intensity-dependent refractive index: P n = no f n 2I=no +n2--. (4.19) Aeir Where II is the linear refractive index, P is the optical power and A,F is the crosssectional area. This is the case for the optical Kerr effect, for which the electronic non-linearity has a response time of the,order of - 1 0 ~ s . It is also the case for lasers mode-locked with fast saturable absorbers i.e. absorbers whose upper state lifetime is much shorter propagating through a than the laser pulses. Consider a pulse of intensity profile I(r),

The generation of ultrashort laser pulses

235

non-linear medium of length L. As well as accumulating the usual linear phase @ o , the light will also experience an additional non-linear phase change which will be proportional to the intensity profile of the pulse i.e.:

277 ,$= @o+A@= -n -L=

, I

-(no+nJ(t))

- L.

2R d

(4.20)

Thus the pulse will acquire a phase profile A@( I ) = -(2n/d)n2Lf ( I ) . Since the pulse has a time-varying intensity profile, this will produce, by the action of SPM, a time-varying phase profile. The first derivative of this phase profile, d@/dt, will correspond to a local change in frequency:

271 d @ --d A@(t)=-;lz-L-dr di

d

df(t) - 6w(r). di

(4.21)

Thus the non-linear intensity-dependent refractive index results in a frequency chirp. It should be understood that while the action of CVD is to produce a linear frequency chirp by effectively redistributing in time the various frequency components of an optical signal, the action of SPM (and XPM) is to produce a non-/ineur frequency chirp by shifting some of the frequency components to new frequencies. Thus photons at new frequencies are created and so the spectral width of the optical signal is increased. For the case of pure SPM in the absence of any GVD, there is no change in the temporal profile of a pulse. The action of SPM is illustrated in figure 29.

c
Figure 29. Schematic of the action

of SPM on an optical pulse.

If the spectral width of a pulse in increased, it has the potential to sustain a shorter pulse duration. This is the basis of fibre-optic pulse compressors and soliton pulse shaping. Note that energy is conserved and so the pulse energy is unchanged by SPM alone. The increased spectral width, however, does increase the loss experienced by a pulse interacting with a frequency filter. The most common cause of SPM is the nonlinear refractive index due to the optical Kerr effect. I n fused silica and most solid-state cmz W-') and so appreciable nonlaser media, the value of n2 is very low (3.2 x linear phase changes are only experienced by retatively intense optical signals (e.g. > I kW peak power) unless the interaction lengths are extremely long. Two special cases where this is the case are propagation in laser cavities, where a pulse may interact with a medium of a few mm length on each of millions of cavity round-trips, and propagation in optical fibres where single pass km interaction lengths are possible. Thus SPM is extremely important for the generation and propagation of ultrashort pulses in lasers and in optical fibres. For (e.g. fibre) geometries which maintain a constant beam size (and therefore intensity) over the interaction length, one can estimate the increase in

236

PM

M/

Erend

spectral width of an initially unchirped optical signal according to:

(4.22)

where Awo is the initial spectral width which is inversely related to the pulse duration. Thus one can define a characteristic non-linear length LNL,over which the spectral width of a pulse increases by a factor of J2 due to SPM:
(4.23)

When a pulse propagates through a length L of non-linear medium with an intensity such that L>>LNL, then it will exhibit strong nonlinear spectral broadening due to SPM.

4.3. Pulse coinpression using SPM and GVD

Consider an optical pulse propagating in a medium which exhibits an intensitydependent refractive index due to the optical Kerr effect. From equation (4.18), the instantaneous frequency chirp, 6w(t), will be proportional to -dI(t)/d/. Thus for the front of the pulse where the medium experiences an increase of intensity with time, 6 w will be negative and so the light will be shifted to lower frequencies. At the peak of the pulse dI(c)/dt=O and so there will be no frequency shift. At the back of the pulse the medium will experience a decrease of intensity with time and so 6 w will be positive. Thus there will be a sweep of frequencies from low to high across the temporal profile of the pulse. This is described as an upchirp or positive chirp and occurs for media which exhibit positive n2. If this positively chirped signal then propagates in a medium which exhibits negative G V D (i.e. anomalous, d2k/do2<0) such that high frequencies travel faster than low frequencies, then the back of the pulse will travel faster than the front and will tend to catch up. This results in pulse compression and is schematically

Figure 3 0 . Schematic of non-linear optical pulse compression.

illustrated in figure 30. Pulse compression thus arises when the non-linear frequency chirp and CVD have opposite signs (i.e. when n2 and d2k/do2 are of opposite sign). For the case when these quantities have the same sign, the back of the pulse will travel slower than the front and so the dispersive pulse broadening and the concurrent frequency chirp will be enhanced by the SPM. Non-linear pulse compression/broadening can occur inside or outside laser cavities. In mode-locked laser cavities there is usually sufficient peak power to induce significant SPM and the interaction of SPM and GVD is what determines the final steady-state of the laser. Outside a laser cavity, this interaction provides a powerful means of pulse compression and is especially convenient when a single mode optical fibre is used as the non-linear medium. When a powerful ultrashort pulse, e.g. from a femtosecond dye laser or solid-state laser, propagates in an optical fibre, for wavelengths below -1.3 pm,the CVD is positive and this, combined with SPM, produces strong non-linear broadening of both the temporal and spectral profiles with a high degree of positive frequency chirp. If this

Tlie generation of ultrashort laser pulses

237

Input Pulse

Spectrally and Temporally Broadened Pulse

Compressed Pulse

Figure 31. Schematic of fibre-optic pulse compression with a diffraction grating pair

optical signal is subsequently propagated through a linear medium which exhibits negative GVD, then the high frequency components towards the rear of the pulse catch up with the lower frequency components at the front and compression is realized. An adjustable dispersive delay line, such as a diffraction grating pair, may be used to optimize the negative GVD and compress the pulse to its new transform-limited duration as set by the broadened spectral profile. This technique, which is illustrated in figure 31, is extensively discussed in the literature. Reference [456]provides a comprehensive discussion of the theory and [457]is a tutorial review. It was first demonstrated using optical fibres to provide the SPM with sodium vapour providing the negative CVD [458]. The conventional fibre-grating compressor was first demonstrated by Shank et a1 [459] to compress the pulses from a femtosecond dye laser system. In principle this technique may be applied to almost any laser, given sufficient pulse energy. There are, however, some disadvantages in extracavity pulse compression as a means of obtaining high quality femtosecond pulses. The diffraction gratings provide the linear CVD to compensate linear chirp arising from SPM in the fibre. In practice neither the chirp produced by the SPM nor by the diffraction gratings is perfectly linear. The positive frequency chirp produced by the interaction of positive SPM and positive CVD is considerably more linear, however, than the chirp arising from a medium which exhibits SPM with no positive GVD. This happily is the case for silica fibre at wavelengths below -1.3 p m , The non-linear negative GVD of diffraction gratings arising from third order dispersion can be compensated by the third order GVD contributions from other dispersive delay lines such as a prism sequence and this approach has produced the shortest optical pulses yet generated from a fibre-optically compressed, amplified CPM dye laser [460]. For pulses at wavelengths above the zero GVD wavelength of optical fibres (i.e. -1.3 pm) positive chirp from SPM and negative chirp from GVD is obtained in the same fibre. Thus pulse compression is possible with no external source of CVD as shown in figure 32. This may be recognized as high order solifon compression as described in section 4.4. It is a very simple and powerful technique [461]which has been used to
Input Pulse Spectrally Broadened and Temporally Compressed Pulse

Figure 32. Schematic of fibre-optic pulse compression for fibre with negative GVD

238

PM

w l'r.crrcl1

achieve the highest ever pulse compression ratio of 2700x in conjunction with a separate conventional fibre-grating compressor [462], generating pulses as short as 33 fs at
1.32 pm.

4.4. Soliton propagation and solilon shaping

Consider the situation shown in figure 32 when a pulse is propagating in a medium which exhibits positive SPM due to a positive n2 as well as negative (anomalous) GVD. If the pulse intensity is very low then the w h i will be very small (L<cLNL) and if there is significant dispersion (LKL,,) then the pulse will broaden in time. Conversely, if the pulse intensity is very high such that the SPM produces a large amount of spectral broadening and frequency chirp (L>>LNL), then the non-linear pulse compression will be strong and the pulse ?ill shorten. For some particular intensity the tendency to broaden arising from CVD, and the tendency towards non-linear compression, arising from the S P M ~ C V Dinteraction, will balance and the pulse will propagate with no change ill its duration. The power at which this happens is called the soliton power. and the pulse which propagates without changing its duration is called a soliton. Soliton propagation thus occurs when the non-linearity in a medium is balanced by the dispersion. This can be represented in simple mathematical terms as being when the characteristic dispersion length of equation (4.12) is equal to the characleristic non-linear length of equation (4.23), i.e. Ln L N L .Equating these lengths provides a useful estimate of the parameters of lhe soliton pulse propagation Po and Zo (see equation (4.24)).

A , ~ D ~ ~- A d Po = 2n2cn2ri I I ~ W Z 2nn2Zo' ~

(4.24)

Po is the fundamental soliton power and 20is the soliton period, which is the scaling length of the interaction between the SPM and the GVD. Thus pulses with peak powers lower than Po will dispersively broaden over the soliton length Z,, which is determined by the dispersion parameter D and by the pulse duration t oThere . will be some spectral broadening/temporal compression mechanism which will act against the dispersive broadening but it will not be suRcient to cancel it. Pulses with peak equal to Po will exactly balance non-linear compression and dispersive broadening and so will propagate without changing either their temporal or spectral width. For pulses with peak powers higher than Po, the non-linear compression will be stronger than the dispersive broadening and the pulses will be spectrally broadened and temporally compressed. The propagation of such pulses is described by the non-linear Schrodinger equation (NLSE). A pulse with a sech' intensity profile of peak power Po and duration ro is an exact stable solution of this non-linear wave equalion. I t is the fundamental soliton solution but there is in principle an infinite set of higher order solutions for pulses whose peak power scales as N 2 P uwhere Nis an integer. Figure 33 shows the propagation of an ideal N = 2 soliton and figure 34 shows the propagation of an ideal N = 3 soliton. I t will be seen that initially these N = 2 soliton pulses compress since P > POand so nonlinear compression dominates dispersive broadening. As these pulses propagate and compress, however, they also undergo significant spectral broadening and this makes the dispersive broadening mechanism stronger. At some point the dispersive broadening dominates the non-linear compression and the pulses broaden until the original pulse

The generation o f ultrasliorl laser pulses

239

-l o r d time

Figwe 33. SchemdtiC of N = 2 soliton pulse propagation.

duration, zo, is recovered after propagating a distance Zo (one soliton period). For the N = 3 soliton pulse, the initial non-linear compression mechanism is very strong and the pulses compress to a minimum duration and then split into two pulses as a consequence of the complex interaction of the SPM and GVD. Further propagation reverses this sequence of events until, after propagating a distance 20, the pulse recovers its original shape and duration. Yet higher order soliton pulses will have more complicated evolutions as they propagate but they should all recover their original shape and duration after each soliton period i.e. after each propagation of Zo. In practice this is difficult to observe because real systems do not exactly obey the NLSE and so the soliton propagation suffers perturbations. As the order of soliton increases, the solitons become more sensitive to perturbations which usually cause them to break up into conventional dispersive waves and (sometimes) bunches of fundamental (i.e. N = 1) solitons. The physics and application of optical solitons is beyond the scope of this article. I t is covered in an extensive literature which may be introduced by [463]. Further detailed discussion may be found in the textbooks [464-4661 and references therein. For the purposes of ultrashort pulse generation, the compression arising from a medium which exhibits SPM and negative GVD may be described as sofiton shaping or soliron cornpression. Soliton shaping, i.e. the interaction of G V D and SPM, may occur inside lasers and influence the pulse evolution or it may be exploited extracavity to achieve pulse compression as discussed in the previous section. When encountered in ultrafast lasers, fundamental solilon pulses are reasonably stable to perturbations in energy and to variations in dispersion and non-linearity. Higher order solitons are generally not stable. All soliton pulses are particularly unstable in the presence of periodic perturbations whose period approaches the soliton period Zo. This phenomenon is often the limiting factor in the achievement of very short femtosecond pulses in laser oscillators.

- loco1 time

Figure 34. Schematic of N=3 soliton pulse propagation,

240

P M W French

4.4.2. Soliton sliapifrg in fenrtosecomf dye lasers. Soliton shaping was identified as a pulse compression mechanism in CPM dye lasers where it was demonstrated that the combined action of G V D and SPM could reduce the pulse duration [467]+vcn to valucs below that achieved with no SPM in the laser [468]. Haus and Silberbergcalculated that a factor of 2 improvement in pulse duration could be obtained by increasing the SPM in a laser 14691. It was theoretically 14701 and experimentally observed that when both positive and negative SPM were present, the laser was not able to generate such short pulses as for the case of pure positive SPM arising from the optical Kerr effect. By operating a CPM laser at a wavelength where the SPM contribution from absorber saturation was minimized, pulses as short as 27 fs were generated [471].

I
Figure 35. Schematic of component functions of a mode-locked laser.

Figure 35 shows a functional schematic of the various processes found in a typical mode-locked ultrafast dye laser. The precise interplay of all these processes fixes the final laser dynamics which greatly vary according to which specific system is under consideration. The left of the figure shows the components found in any laser. Those shown on the right are usually present in a passively mode-locked laser but are not always significant. Before the development of femtosecond lasers and the optimization of the cavity dispersion, the final mode-locked steady-state was fixed only by the interaction between the modulation function and the spectral filter. For the case of active mode-locking, as discussed i n section 3, the modulation function is derived from a modulator and remains fixed during the evolution of the mode-locked pulse. For passive mode-locking using a real (resonant) saturable absorber, the modulation function was related to the saturation of the absorber and amplifier population inversions. For the idealized case of a laser with only amplitude modulation (i.e. no SPM), then the shortest pulses would be obtained for zero intracavity CVD. In any real ultrafast laser, however, the high intracavity peak powers mean that there is always some non-linear pulse propagation and so SPM will play a role. The development of the femtosecond dye laser and the parallel development of fibreoptic pulse compression highlighted the issue of frequency chirp on ultrashort pulses.

The generation o f uitru3ifor1 laser pulses

24 1

Subsequently soliton-like pulse propagation dynamics were observed in a CPM dye laser [472]. The authors observed the characteristic periodic compression and pulse splitting of N = 3 soliton pulses occurring over thousands ofcavity round-trips. One can consider the propagation of pulses in a periodic structure, such as a laser cavity, containing discrete elements of SPM and GVD as an approximation to the propagation of pulses in an optical fibre, where the SPM and GVD is continuously distributed, provided that the changes on each cavity round-trip are small. Thus one soliton period, Z,,corresponds to thousands of cavity round-trips. This is analogous to using a step-propagation model for the numerical simulation of the pulse propagation in a fibre as described by the NLSE. Other soliton propagation phenomena in femtosecond lasers were subsequently observed e.g. [473,474]. It was not possible to achieve detailed agreement between the experimental observations of real ultrafast laser systems and the predictions of an NLSE model which mainly considered accounts for the SPM from the optical Kerr effect and negative G V D (the other components of figure 35 being introduced only as perturbations e.g. [475]). A more complete numerical model [476], which successfully reproduced almost all of the experimental observations of a CPM dye laser, included all the elements of figure 35the modulator being provided by the combined action of saturable absorption and saturable gain. It was shown that the presence of the spectral filter (due to the linewidth of the dye media) strongly perturbed the NLSE description of the laser for pulses whose bandwidths were comparable to the filter linewidth. Only for the special case of weak mode-locking (pulse bandwidth<<filterlinewidth) and large positive non-linearity (SPM) was it possible to observe the soliton propagation phenomena described above. This special case for the numerical model agreed well with the reported experimental parameters. The roles of positive and negative non-linear frequency chirp, due to the optical Kerr effect in the dye solvents and the time-dependent saturation of the absorber dye respectively, were also investigated numerically. It was confirmed that the strong pulseshaping mechanism (passive mode-locking with a slow saturable absorber) could be counteracted by unfavourable phase modulation (i.e. when nz and d2k/dw2 are of the same sign) which produced an unstable laser output or very long pulses. Stable operation with the minimum achievable pulse duration (which is aiei,a)is ultimately limited by the filter linewidth) was best approached for pulses with favourable soliton shaping (i.e. n2 and d 2 k j d d of opposite signs). Operation at or near zero intracavity GVD produced satellite pulse formation due to spectral walk-off: the net effect of the intracavity non-linear frequency chirp contributions in the absence of GVD was to change the mean carrier frequency 011 each cavity round-trip. This resulted in the pulse spectrum walking out of the gain window of the laser and permitting a second satellite pulse to grow using the undepleted gain. In general, if the mode-locked pulses fail to saturate the gain i n a laser, then other features (e.g. noise) may be amplified, producing instabilities. Even in the favourable soliton shaping regime, excessive SPM was shown to lead to instability. This was partly due to strong spectral broadening and consequent losses at the spectral filter, as well as other non-linear mechanisms for instability. The numerical simulations confinned and helped to explain the prescription that to produce the shortest possible pulses, it was first necessary to have the broadest possible filter linewidth and then to have (positive) SPM as close as possible to the ideal case such that it could be exactly balanced by optimizing the (negative) GVD of the cavity. In essence, the closer pulse propagation resembles that of a fundamental soliton governed by the unperturbed NLSE, the shorter will be the minimum achievable pulse

duration. Any other non-linear pulse propagation will result in spectral broadening leading to (periodic) loss at the spectral filter and to other non-linear instabilities. Thus thc dwation of the shortest, stable, steady-state pulses ii'ill be close to that of llie sltortesl ,fuitdaniental snlilon pulscs wiiich can be sustained in the laser. In the absence of any limiting instability, this duration will be approximately half the value expected in the absence of SPM and CVD, given the width of the net spectral filter of the laser. Often it is not possible to achieve this pulse duration due to instabilities arising from higher order CVD, non-linear scattering and periodic perturbations to the soliton propagation, as discussed in the next section. For the linewidths available in dye lasers, however, the relatively low non-linear interaction lengths in the cavity and the higher order CVD contributions and non-linear sources of instability are not significant. Furthermore, the strong amplitude modulation function of the slow saturable absorber passive modelocking tends to suppress the growth of any noise features in the laser. Thus feintosecond dye lasers cannot be expected to yield significantly shorter pulses than already achieved since they are gain linewidth-limited, To directly generate pulses of IO fs duration and lower, it is necessary to use gain media with much broader linewidths, such as vibronic solid-state lasers, or to compress the pulses extracavity e.g. using a fibre-grating compressor.

4.4.3. Soliton shaping i n feinlosecond solid-state lasers. Femtosecond vibronic solidstate lasers such as Ti:sapphire lasers differ from dye lasers in three imporpant respects. First, the linewidth of llie amplifier is many times greater and so spectral filtering is not usually the factor that limits the minimum achievable pulse duration. Second, the intracavity power levels may be up to 10 times higher in solid-state lasers. Third, the passive mode-locking techniques based on the optical Kerr effect have an almost instantaneous response and so, as well as not limiting the minimum pulse duration achievable, they do not contribute any unwelcome frequency chirp due to absorption saturation. These factors have resulted in the routine generation of sub-20 f s pulses as discussed in section 3. The generation of shorter pulses is generally limited by the interaction of SPM and CVD which must include higher order terms that become significant for the extremely broad bandwidths of such pulses. References 14771 and [478] provide excellent discussions of the issues underlying the generation of the shortest possible pulses from femtosecond solid-state lasers. A further difference between femtosecond solid-state lasers and dye lasers is that the relatively long upper state lifetimes, and therefore saturation Ruences, of vibronic solid-state media compared to dyes mean that the amount of pulse shaping due to gain saturation is insignificant. In the steady-state, the dominant pulse compression mechanisms are therefore passive amplitude modulation (e.g. KLM/APM) and the interaction between SPM and CVD. In the absence of higher order CVD and strong perturbations to the pulses, the final pulse would be expected to be a fundamental soliton whose duration was proportional to the CVD divided by the pulse energy and the non-linear phase shift/unit power per cavity round-trip, e.g. [478,288]. In this situation, the pulse shaping would be described by the NLSE which assumes that the pulses experiences SPM and CVD sitiiullaileorrsly, as for the case of propagation in an optical fibre. In femtosecond solid-state lasers, however, the SPM and CVD occur discretely in various of the intracavity components. SPM is confined to the laser rod and contributions to the CVD are predominantly from the laser rod and the prism sequence. For this reason, such lasers have been described as 'solitary laseis' 12881 rather than 'soliton lasers'. On each round-trip, the pulses can undergo significant changes in duration as they experience

Tlre gcncrufiori of ultrashort laser pulses

243

predominantly c v v or srM i n particular components. This has been experimentally confirmed by measuring the pulse duration at different positions in the cavity [480]. The NLSE is no longer valid and the steady-state pulse duration has been described as [288, 478, 4791.
3.53101 rp=+abW (4.25) bW where ID/ is the total cavity G V D (neglecting higher order contributions), Wis the pulse energy, r $ is the round-trip non-linear phase shift/unit power experienced by the pulses and a is a numerical factor obtained from simulations which varies with the position in the cavity. In general, the pulses are spectrally broadest at the midpoint of the prism sequence (i.e. after a single pass) but if the output is taken from this dispersive end of the cavity, it is necessary to use an extracavity prism pair to reach the minimum pulse duration and to remove the spatial transverse frequency chirp which arises from the single pass through the cavity prisms. Equation (4.25) implies that the pulse duration may be arbitrarily reduced by reducing the intracavity dispersion. In practice 1 0 1 cannot be reduced indefinitely before the GVD fails to satisfactorily compensate for the SPM and instability sets in. The minimum value of 1 0 1 for which stable pulses are obtained decreases as the strength of the passive amplitude modulation increases. Thus improving the strength of the KLM amplitude modulation improves the quality of the pulses and permits shorter pulses to be generated. The presence of higher, i.e. third order, GVD limits the minimum duration achievable i n two ways: it obviously contributes to the value of the net cavity dispersion and it is also responsible for the onset of some instabilities, e.g. [479]. It was the reduction of intracavity third order dispersion (TOD) that led to the decrease of the state-of-the-art pulse durations from these lasers from -50 fs to <20 fs. Initially the dense glass prisms (e.g. SFIO) which had been required to compensate for GVV of the 2cm long Ti :sapphire laser rods were identified as contributing significantly to the TOD and they were replaced by prisms made from glasses such as F2 [481] or LaKL21 [482] which exhibit significantly lower TOD at -800 nm. Further improvements were made by replacing these glasses with fused silica The angular dispersion, and therefore CVD available for a given prism separation, is much lower for fused silica prisms than denser glasses and so it was necessary to reduce the length of the laser rod in order to keep the prism separation required for GVD compensation to a reasonable value. For the case of Ti:sapphire. it has recently become possible to obtain crystals with higher doping levels (up to -5%), and therefore shorter absorption lengths, compared to the conventional 2 cm long, 0.3% doped crystals. The TOD of the prism sequence contains a term proportional to the separation as well as a material contribution so any reduction in the required (second order) GVV. and consequent reduction of the prism separation, also reduces the net .rev. Using laser rods as short as 4 mm, pulses of -IO fs have been generated in cavities designed to minimize TOC e.g. [483,484]. A further step in this direction has been to dispense with the prisms altogether and use carefully designed dielectric mirrors to provide the CVD compensation. This approach has resulted in an extremely stable laser which generates pulses as short as -1 1 fs with a Ti:sapphire rod of 1.8 mm length [485]. TOD does not simply broaden the pulses in femtosecond solid-state lasers. In all femtosecond lasers there exist mechanisms for energy transfer from soliton-like pulses to dispersive background radiation. Essentially, when a soliton is perturbed nonadiabatically (e.g. abrupt loss or gain), then it will shed energy in the form of dispersive

244

P M W French

radiation to recover its soliton properties, albeit with a different set of parameters (wavelength, energy, duration etc). If the perturbations occur with a periodicity close to the soliton period, then this effect will cause instabilities and ultimately the breakup of the soliton. More precisely, this energy transfer from a soliton to a dispersive wave can be resonantly enhanced at particular wavelengths for which a phase-matching process occurs. Phase-matching may be due to periodic perturbations to the soliton which are resonant with the soliton period e.g. [486,487] or, for close to zero GVD,it may be due to TOD e.g. [488-4901. The latter effect is more important for solid-state lasers while the former is important for femtosecond fibre lasers. Following equation (4.25), as the (second order) GVD in solid-state femtosecond lasers is reduced (e.g. by reducing the intracavity glass-pall]), the pulse duration decreases and the spectrum broadens until the optimum value is reached. Further reduction of the GVD rapidly results in the appearance of a spectral peak corresponding to the point at which this phase-matching occurs and there is a resonant energy transfer to the dispersive radiation continuum. Decreasing the second order FVD further results in the break up of the mode-locked pulses. This value o f the GVD at which the resonance appears is determined by the TOD. Reference [293] estimates that the minimum pulse duration achievable is proportional to the square root of its magnitude. If yet shorter pulses are to be generated from such lasers, the TOO must be further reduced.
4.4.4. Solifori slrapirrg in feriitosecond j h r e h e r s . Optical fibres provide an important class of lasers which have been the subject of intensive research, partly driven by their potential application to ultrahigh bit-rate telecommunications. Reference [49 I ] contains an excellent review of the state-of-the-art in this field. Mode-locking is typically achieved using APM as described in section 3.3.4.2. Unfortunately, such fibre lasers generally do not reliably produce stable trains of high quality femtosecond pulses. I n part this is d u e to the sensitivity of the birefringence and other properties of optical fibres to changes in their environment e.g. temperature, mechanical stress, vibrations etc. This may be partly accommodated by using carcully designed lasers with polarizationpreserving fibre e.g. [492]. A further problem associated with fibre lasers is that they tend to become unstable in the presence of excessive SPM which can occur at relatively modest pulse energies due to the long non-linear interaction length available in fibre geometries. Fibre lasers may be classified as soliton lasers or solitary lasers, depending on whether they operate at wavelengths for which the fibre GVD is normal (positive) or anomalous (negative). Erbium-doped fibre lasers operate at wavelengths where standard fibre GVD is negative and so they support the propagation of soliton pulses. Even in lasers where there are various components with different magnitudes and signs of G V D and where there are large perturbations due to loss and amplification, it has been shown that, as long as the length scale for these perturbations is much less than the average soliton period (calculated from the average pulse duration and energy), then the pulses may be considered as fundamental soliton pulses [493]. Such behaviour is described by the average soliton picture [494]. Erbium fibre soliton lasers have generated pulses as short as a few hundred femtoseconds in many configurations. In general it is difficult to stably generate much shorter pulses due to the resonant energy shedding mentioned in the previous section. As the pulse duration is decreased, the soliton period also decreases and approaches the cavity length. Thus the perturbations which the pulse experiences on each cavity transit come to have a similar periodicity to the soliton propagation and at some point the phase-matching condition is fulfilled. This energy transfer resonance

The gcncrafion of ultvasliorl laser pulses

245

will show up i n the spectrum of the laser output which will show sidebands growing at those wavelengths for which the phase matching conditions are fulfilled. The positions of these spectral peaks have been predicted [486, 4871 and confirmed experimentally, e.g. [49S]. To avoid the instabilities associated with these sidebands and the underlying resonant energy transfer, it is necessary that the cavity length of soliton lasers should be much shorter than the soliton period. A limiting condition has been derived which indicates that the minimum soliton pulse duration achievable scales with the square root of the dispersion-length product [496,497]. By designing such lasers with low GVD components and keeping the cavity length to a minimum, pulses as short as -100 fs have been obtained [498]. An elegant way to side-step the instabilities associated with the resonant perturbations in soliton fibre lasers is to use lasers with large amounts of positive and negative dispersion in discrete components. These fibre lasers are essentially similar to femtosecond solid-state lasers, i.e. they may be considered more or less as solitary lasers. Originally this approach was encountered with Nd:doped fibre lasers which operated at wavelengths for which the fibre exhibited strong positive (normal) GVD which was compensated for by diffraction grating pair,, e.g. [499]. The shortest pulses to date from a fibre laser, of 33 fs duration, were obtained from a solitary Nd:doped fibre laser [SOO]. An all-fibrejmplementation with Er:doped fibre was recently demonstrated using both positive and negative GVD fibre to ensure a small net positive GVD [Sol]. This laser generated pulses as short as 77 fs. As well as avoiding the resonant perturbations of soliton lasers, these solitary lasers also reduce the intracavity peak powers since the pulses are strongly dispersively broadened in the fibre cavities. This correspondingly reduces the SPM in the cavity and so permits pulses with higher energies (up to - 1 nJ) to be obtained before the onset of instability than is possible for soliton lasers. As the pulses get shorter, fibre lasers must also address the issues associated with TOD that occur for femtosecond solid-state lasers. In general, however, the narrower gain profiles of fibre lasers will preclude pulses much shorter than -30 fs from being generated. Nevertheless, for many real-world applications, and for telecommunications in particular, fibre lasers provide a compact source of femtosecond pulses.

5. Measurement of ultrashort laser pulses 5.1. Introduction

One of the most important issues in generating ultrashort light pulses is knowing to what extent success has been achieved. Many diagnostics have been developed to monitor ultrashort pulse laser experiments but each tool has significant disadvantages which preclude it from being an exclusive choice. I n principle, if the amplitude and phase of the optical signals could be monitored in either the spectral or temporal domain, a Fourier transform calculation should yield all the desired information. In practice i t is rarely possible to uniquely determine both the amplitude and phase of optical signals in real-time and several diagnostic tools are used simultaneously.Typically it is desirable to monitor the pulse energy using a power meter; the pulse train [pulse-to-pulse stability) using a fast photodetector and oscilloscope; the pulse spectrum using a realtime [scanning/ccn read-out) spectrograph and the pulse duration using an autocorrelator. There are many other specialist diagnostics which have been developed for particular experimental requirements. 1 know of no recent review of ultrafast measurement

246

P M W Fvcncli

DeIlMion R a w SlCl"d

Figure 36. Schematic of electro-optic streak camera.

techniques but the basic principles of most are to be found in [502,.503].I t is beyond the scope of this article to provide a comprehensive review of ultrafast measurement techniques and only a fraction of relevant work has been acknowledged here. I t is not possible, however, to operate even the most reliable ultrashort pulse lasers without some ultrafast diagnostic and so the aim of this section is to outline the basic principles of ultrashort pulse measurement and to describe the most common techniques for monitoring state-of-the-art ultrafast lasers. In general, ultrafast measurement techniques can be classed as linear or non-linear. The former are usually based on electronic sampling and the latter on interferometry.
5.2, Linear lecltniques

The most convenient way to measure trains of optical pulses is probably to use an ultrafast photodiode with an oscilloscope. Unfortunately, the fastest photodiodes commercially available have a rise time of the order of 10 ps and the best (sampling) oscilloscopes are lim'ted to the same order of magnitude. The linear response of photodiodes makes them very attractive when their resolution is sufficient.One way to exploit very fast photodetectors with response times down to -300 fs is to use the technique of elccfvo-opricsuiitpling, a review of which is given in [504]. An alternative, and rather simpler, linear technique is the electro-optic streak cai17eruwhich was first proposed in 1956 [SOS]. This can operate in both single-shot mode, with a resolution down to -300 fs but typically > I ps, or in synchronous mode with resolutions typically >5 ps although -I ps resolution has been achieved [506].Figure 36 is a schematic of the operation of a streak camera. The incident ultrafast optical signal is converted into a stream of electrons by the photocathode. This electron stream is accelerated towards the anode where it passes through a central aperture. It then is directed through a pair of electrostatic deflection plates and onto a scintillator screen where it is converted back into an optical signal which may be imaged onto a detector such as a multi-channel analyser. In single-shot mode, the signal applied to the deflection plates may be a single voltage ramp which is synchronized to the incident optical signal (usually using a photoconducting switch). For a pulse input, the photon stream will be converted into an electron stream of similar length and with the transverse dimensions of the slit. This electron stream will be swept across the scintillator screen by the deflection field and so, the longer the incident pulse (photon stream), the longer will be the corresponding 'streak'. In synchronous mode, there is usually a well defined train of pulses incident at the photocathode and the

Tlie generation o f ultrashort laser pulses

241

+A

Figure 31. Schematic of second harmonic generation autocorrelator.

continuous synchronous deflection signal is sinusoidal at the period of the pulse repetition rate. This signal may be derived from the optical signal using a photodiode and a low pass filter or a tunnel diode oscillator. The electron streaks from each of the periodic pulses are superimposed on the scintillating screen which typically has a response time of the order of ms. Thus measurements are integrated over thousands of pulses resulting in high sensitivity but requiring accurate synchronization and low pulse-to-pulse jitter in order to preserve the temporal resolution. It should be understood that although the conversion of optical signals to electronic and back to optical is rather convoluted, the streak camera gives the most faithful picture of ultrafast optical signals and is linear over a huge dynamic range. The synchronously operating streak camera is an extremely powerful instrument-especially when used in conjunction with a mode-locked laser. The sensitivity can be equivalent to single-photon counting and, with highly stable passively mode-locked lasers, resolutions of -1 ps have been achieved. It is particularly convenient to use mode-locked lasers to repetitively excite a sample and to monitor fluorescence/absorption changes using a streak camera.

5.3, Seeond harmonic auloconelation

For femtosecond pulses, the resolution of streak cameras is not sufficient. Single-shot streak cameras have achieved sub-ps results but these measurements are not routine and require a significantly higher pulse energy than is available from most unamplified femtosecond lasers. Instead, non-linear correlation techniques are widely used to study lasers and as diagnostics i n femtosecond spectroscopy. The most widely used technique, which exploits second harmonic generation (SHG) in non-linear crystals, is based on a second order autocorrelation function and was first demonstrated in 1966 [507]. Figure 37 is a schematic illustrating the basic principle of operation of a SHG autocorrelation measurement [508] which is typically used to measure the duration of ultrashort optical pulses. An incident optical pulse is split into two sub-pulses by a 50/50 beam-splitter at the centre of a Michelson interferometer. The sub-pulse trains ( E , and Ez) are reflected in each arm and recombined at the beam-splitter before being focused onto a nonlinear frequency doubling crystal. The resulting second harmonic signal generated in

248

P M W Frcnch

the propagation direction of El and E2, is passed by the filter and detected by the photomultiplier, the fundamental radiation being blocked by the filter. If the length o f one arm of the Micliaelsoo interferometer is scanned through the point where il matches the length of the other ann, the detected SHG signal will increase and then decrease, following the coherent addition of the electric fields of the two sub-pulse trains as they are superimposed at the non-linear crystal. The efficiency of the second harmonic generation is proportional to the intensity of the fundamental radiation and so the intensity of the second harmonic radiation is proportional to the square of the intensity of the fundamental radiation. It is this nonlinear dependency of the SHG signal on the incident (fundamental) intensity that permits the measurement of arbitrarily fast coherent optical signals using an interferometer. The precise form of the SHG signal changes according to the particular configuration used. Type I phase-matching (parallel 0-0 subtrain polarizations) or type TI phasematching (perpendicular o-e subtrain polarizations) may be employed in the birefringent non-linear SHG crystal and the sub-pulse train beams may be collinear or parallel and recombined at the focus o f a lens. In the case of an autocorrelation measurement, E2 is the same signal as El but delayed by the mismatch in the interferometer arms ( 7 ) . In an alternative configuration, essentially the same apparatus may be used to perform a cross-correlation measurement between different pulse trains E , and E2. These may be of different wavelengths in which case sum-frequency generation, rather than SHG, would be used. The physical principles are identical : the combination of non-linear frequency conversion with interferometry to obtain temporal information concerning ultrafast optical signals. For the case of collinear beams with type I phase-matching, the SMG measured signal is proportional to I I 2 ( r )which is given by I(El+ E2)212where El and & are the electric fields of the two sub-pulse trains. For autocorrelation measurements, E2(t)= E,(r- 7 ) . so
f12(t)

=-

+ ,

2Tj-,

I { E l ( / )+ Ej(t- TI} 212 d/.

(5.1)

For the simple case of chirp-free transform-limited pulses, E, = &(t)cos(wt) where & ( / ) = E * ( / ) and
fI2(r)=-

J,
1
lr

is2(() cos2(wr)+&(t) cos(wt).c(t-r) cos(w[r-r])

+2(t-7)

COS2(o[t-T])12dt.

Expanding the integrand yields the interferometric autocorrelation profile

,12( T ) =

1 5

j-,
+ ,

E4( /) C0S4(Oi)

+42(1) COS(ot)&( I-5 ) COS( (U[/-

T])

+6$(t) c o ~ * ( w t ) & ~7) ( t -cos2(o[r-r]) dt

+ 4 & ( t )cos(wt)El(t- r ) cos3(w[t- r ] )

+E4(t-T)

COS*(W[l-T]).

(5.3)

The generation of ulfraslror.1laser pulses

249

This can be simplified, following [509] to: II~(T)=~-.z4(f)dt+4: 8 2Tsr 8

3 1
+r

+r~(t)&2(t-~)dt

3 1 +T +2-cos(2or)c2(f)$(t-r)dt 8 2TSr

(5.4)
where the slowly varying envelope assumption is made and the carrier wave is averaged over a period 2To=2nm where ni is an integer. Figure 38 shows an experimental interferometric autocorrelation trace. It will be seen that the peak-to-background ratio of the experimental trace in figure 38 is 8: 1. This can be seen from equation (5.4)where the intensity of the peak, which corresponds to when the two subpulse trains are exactly overlapped (at z = 0), is given by :

and the intensity of the background is given by:

This background signal corresponds to when the pulses in the two subtrains are not overlapping at all (r>>tP) and so the SHG signal contribution from each subtrain is independent and all the cross terms in equation (5.3) vanish. It is important to achieve the correct peak-to-background ratio in an autocorrelation measurement as this confirms that the Michelson interferometer is correctly aligned. In the case of chirped pulses, the field envelope ~ ( f would ) be complex and equations (5.3) and (5.4) would be rewritten, with the latter, following [YO], as:

(5.7)

250

P M W French

Figure 38. Expcrimental interfcrometric autocorrelation trace.

The same peak-to-background contrast ratio would apply for a chirped pulse. A useful indication of the chirp (and therefore coherence) of a pulse train is the visibility of the fringes of an interferometric autocorrelation trace. If there was a significant chirp on the pulse (i.e. a variation of wavelength with time across the pulse profile), then the leading edges of one pulse subtrain would not be coherent with the trailing edges of the other subtrain and so the interference fringes would not be seen iii the wings of the autocorrelation trace. The experimental interferometric autocorrelation trace of figure 38 corresponds to a reasonably well mode-locked pulse train (i.e. close to transformlimited), with a measured time-bandwidth product of 0.39 assuming a sech pulse profile, and so the fringes are visible right across the trace. I t should be noted that the second order autocorrelation function is always symmetric and so can only give limited information concerning the pulse shape and chirp. Additional information concerning the amplitude and phase profile of the pulses can be obtained from iteratively fitting a trial pulse to the interferometric autocorrelation profile and the spectral profile, e.g. [510, 51 I]. Alternatively, interferometric correlation measurements may be employed whereby the pulse to be measured is correlated with an appropriate, relatively well characterized, probe pulse which may be derived from the pulse under measurement, e.g. [512,513]. Although the interferometric autocorrelation yields information about the pulse duration and phase structure of an optical pulse, it is often experimentally difficult to measure. For a 100 fs FWHM pulse at 800 nm, a sampling of six points per fringe entails the collection of 1400 data points for each trace. This requires relatively stable lasers and opto-mechanics and/or a fast detection system. In practice, the autocorrelation trace is often recorded such that the detector response integrates the SHG signal and the cos(o1r) carrier fringes are averaged out. From equation (5.4) it can be seen that such averaging over a period 2To=2nm yields:

I
+T

I2(t)dt 1+2

/
\

[(l/2T) [ + I ( t ) I ( t - r ) df
J-T

1 I

(1/2T) [I2(t)df
J -T

The same equation is obtained by averaging out the carrier fringes in equation (5.7). It should be noted that the term in square brackets is the second order normalized intensity aufocovrdulionfunction of the pulse. The peak-to-background contrast ratio

The generation of ultrashort laser pulses

25 1

(4

(b)

(e)

Figure 39. Schematic autocorrelation traces for different degrees of mode-locking.

Table 3. Autocorrelation deconvolution values and k values.

Pulse shaw Rectangulai Gaussian Sech Single-sided exponential

AfIk

k<Avt.

I
1.414 1.543 2

0.886
0.441

0.315
0.110

for this autocorrelation measurement is readily seen to be 3 :1. Typical experimental traces obtained are represented in figure 39 and an ex?mple of an experimental autocorrelation trace, Corresponding to a well mode-locked pulse train, is shown in figure 40. As with the interferometric autocorrelation function, there is no information about pulse asymmetry and, in this case, no information about tfie phase structure of the pulse. Furthermore, to establish the pulse duration, it is not possible to count carrier fringes but one must make an assumption about the pulse shape and divide the autocorrelation FWHM, At, by the appropriate deconvolution factor to obtain the pulse duration. Some indication of the pulse shape can be obtained by comparing calculated autocorrelation profiles, corresponding to various pulse shapes, and observing the best fit. Since the intensity autocorrelation trace gives no information about pulse chirp, it is helpful to measure the associated spectral pulse profile and to compare the experimental timebandwidth product Avfp with the minimum value, k, corresponding to the case of chirp-free, transform-limited pulses. Table 3 shows the autocorrelation deconvolution factors and k values for the most commonly assumed pulse shapes 15141. An excessive time-bandwidth product may indicate either frequency chirp of the pulse or incomplete mode-locking. In the latter case, the measured spectra1 width is the noise bandwidth i.e. the spectral width of the shortest noise structure or spikes which occur within the pulse envelope. It is possible to dislinguish between these two cases and

U
t<,,jb

= 2.0 psec

Figure 40. A typical experimental collinear SHG autocorrelation trace (calibration/baseline

is obtained by blocking input to autocorrelator).

252

P M W French

establish the degree of mode-locking by examining the collinear intensity autocorrelation trace. Figure 39 shows schematics of the expected autocorrelation traces of (a) a free-running laser; ( b ) a partially mode-locked laser and ( c ) a fully mode-locked laser. Figure 39(a) represents the case for no mode-locking. The width of the coherence spike is the coherence time and therefore the inverse of the noise bandwidth. The contrast ratio is 3:2. Figure 39(b) corresponds to partial mode-locking where the central coherence spike again indicates the noise bandwidth and the broader feature indicates the width of the pulse envelope or noise burst. The autocorrelation of a noise burst has the characteristic contrast ratio of 3 : 2 : 1. Figure 39(c) represents the autocorrelation trace of a pulse from a fully mode-locked laser which may be chirped, and therefore have excess bandwidth, but which shows the contrast ratio of 3 : 1. The correct contrast ratio is again a useful check of the alignment of the Michelson interferometer. If type I1 phase-matching is employed with collinear subpulse trains [SOS], or if type I phasematching is used with non-collinear subpulse trains which are overlapped in the SHG crystal [6, 5151, then a background-flee U Z ~ O C O F ~ C ~ trace U~~O is?obtained ~ since SHG only takes place when the two subpulses are coincident in the SHG crystal. The contrast ratio is therefore I : 0 and the s m intensity is simply proportional to the second order intensity autocorrelation function. Although this measurement gives no information about the alignment of the autocorrelator, it is preferred in many applications because it clearly indicates the presence of any low-lying pedestal or extended wings of the pulse. There are many other possible schemes to perform non-linear (auto)correlation measurements but there is not space here to describe them. One important variation involves introducing a group velocity delay across the wavefronts of the subpulse trains and detecting all the SHG signal in the plane perpendicular to the propagation direction, e.g. [516, 5171. In this manner, single-shot measurements can be made which can also be interferometric, e.g. ISIS]. Another single shot approach is to use the non-linear phenomena of two photon fluorescence (TPF) instead of second harmonic generation, e.g. 1519, 5201. This can be extended further to multiple-order fluorescence, e.g. [521]. Recently a novel single-shot technique for simultaneously recording both the temporal and spectral profiles to provide a complete description of an optical pulse has been demonstrated [522]. This approach is reminiscent of that described in [50] but uses optical Kerr gating rather than SHG autocorrelation and involves a two-dimensional phase retrieval algorithm to reconstruct the pulse profiles. With the rapidly increasing power of personal computers and the improvements in laser and detector technology, it s e e m likely that we will soon have a single instrument which can provide a complete description of the amplitude and phase profiles of an ultrashort optical pulse. At present, however, most researchers rely on autocorrelators and care must still be exercised in the interpretation of such data, particularly when the pulses are chirped or when there is some uncertainly about their shape. In most instances, the autocorrelation profile is recorded using a slow detector which integrates over thousands of pulses in a train. Changes in the average output power of an unstable laser can lead to changes in the recorded autocorrelation profile. A further cause of ambiguity arises when the pulse train under examination is fluctuating in pulse shape, duration or liming. The non-linear correlation processes are weighted towards shorter, inore intense pulses and an unstable laser can appear to be delivering shorter pulses than it consistently does. When an autocorrelation trace does represent a time average of an evolving pulse train, it often exhibits broad exponential wings [523, 5241 which can lead to erroneous assumptions about the pulse shape.

The g e n e r a t i o n of ulfrasltort laser pulses

253

References
[ I ] Svelto 0 1989 Principlm o / L o e r s ed and transl. D C Hanna (Ncw York: Plenum) [2] Sicgman A E 1986 h e r s (Mill Vellcy, CA: University Science Books) [3] Yariv A 1989 Quawtuni Electronics 3rd ed (New York: Wiley) [4] Hcrrinan J and Wilhelmi B 1987 Losers for. Ullrasliort Light Pulses (Berlin: Akddcmie) [5] S a t m a r i S, Schafer F P, Moller-Horsche E and Muckenhcim W 1987 Opt. Corruiiun. 63 305 (61 Hargrove L E , Fork R L a n d Pollack M A 1964 Appl, Phys. Lett. 5 4 [7] Liu K X, Flood C I, Walker D R and van Driel H M 1992 Opt. Lett. 17 1361 [SI Malcolm G P A and Ferguson A I 1991 Opt. Lett. 16 1967 191 Ramaswamy M, Gouveia-Net0 A S, Negus D K. lzatt J A and Fujimoto J G 1993 Opt. Lett. 18 1825 [ I O ] Ober M H, Hofer M , Fermann M E 1993 Opt. Lett. 18 367 [ I I] Tamura K, lppen E P, Haus H A and Nelson L E 1993 Opt. LetI. 18, 1080. [I21 Fork R L, BritoCruz C H, Beckcr P C and Shank C V 1987 Opr. Lett. 12 483 [I31 Sorokin P P and Lankard J R 1966 lBM J. Res. Dev. 10 162 [I41 Zhangand F G and Schafer F P 1981 Appl. Phys B 7 6 . 21 I [I51 Polland H I. Elasaesser T, Seilmeicr A, Kussler W, Marx N 1, Sens B and Drexhage K H 1983 Appl. Phys. B 32 53 [I61 Sclmidt W a n d Schafer F P 1968 P1ry.s. Lett. 26A 558 [I?] Brackmann U 1986 Lorndackronie Loser Dyes Cafalogw (Gottingen: Lambda Physik) [IS{ Schdcr F P 1977 Dye Losers 2nd cd (Topics in Applied Physics 1) (Berlin: Springer) [I91 Duarte F J and Hillman L W ed 1990 Dye Laser Principles: With Applications Quantum ElectrodcyPrinciplm nnd Applicalioras (San Diego, CA: Academic) 1201 Kocchner W 1977 Solid-state Loser Engineerirg 2nd ed (Springer series in optical sciences 1) (Berlin: Springer) 1211 Paync S A and Albrccht G F 1991 Encyclopedia o/Lasers arid Opl/CQ/ Tecltrzology (San Diego, CA: Acadcmic) pp 603-24 [22] Moulton P F 1992 Pror. IEEE 80 348 [2S] Moulton P F 1986 J . Opt. Soc. A m B 3 125 [24] Alfrey A J 1989 I E E E J. Qrmanfwn Elcetron. Q E 2 5 760 [25] Rizvi N, French P M W a n d Taylor J R 1992 Opt. Lett. 17 1605 [26] Mcllish R M, Frencb P M W, Taylor J I<, Delfyett P J and n o r u L T 1994 Electron. Lett. 30 223 [27] Yanovsky V, Pang Y ,Wise F and Minkov B I 1993 Opt. Lett. 18 1541 [ZS] Conlon P J, Tong Y P, French P M W, Taylor J R and Shestakov A V 1994 Opt. Lett. 19 [?9] Johnson B C, Moulton P F and Mooradian A 1984 Opt. Lett. 9 I16 1301 Pollack C R 1991 Eficyelopedio o/ Losers and Optical Tccltnolog? (San Diego, CA: Academic) pp 9-26 [31] Mollenaucr L F and Stolen R H 1984 Opt. Lett. 9 13 [32] Chcn Y-K Chen and W u M C 1992 IEEE J . Quantzini Electron QE-28 2176 [33] Corzinc S W, Bowers J E, Przbylek G, Koren U, Miller B 1 and Soccolich C E 1988 Appl. Phys. Lett. 52 348 [34] Dclfyett P J, Florez L T, Stofcl N, Gmitter T, Andreadakis N C, Silberberg Y, Heritage J P and Alphonse G A 1992 IEEE J. Quuntmn Electfon. Q E Z s 2203 [35] Mehuys D, Welch D F and Goldberg L 1992 Electron. Lett. 28 1944 [36] Hargrove L E , Fork R L a n d Pollack M A 1964 Appl. Plryx Lett. 5 4 1371 Smith P W 1970 Proe. IEEE 58 1342 1381 New G H C 1983 Rei,. frog. Pliys. 46 877 (391 Krausz F, Fermann M E, Brabec T, Curley P F, Hofer M, Ober M H,Spielmann Ch, Wintner E and Schmidt A J 1992 IEEE J . Quantum Eleclron Q E 2 8 2097 I401 Peterson 0 G, Yuccio S A and Snavely B B 1970 Appl. Pliys. Lett. 17 245 1411 Ippen E P, Shank C V and Dienes A 1972 Appl. Plzjs. Lett. 21 348 [42] Mollenaucr L F and Stolen R H 1984 Opt. Lett. 9 13 [43] DiDomcnico M 1964 J . Appl. Pliys. 35 2870 [44]Yariv A 1965 Appl. Plrys. 36 388 [4S] Ammann E 0, McMurtry B J and Oshman M K 1965 IEEE J. Qriaantum Electron. QEl 263 1461 DeMaria A J and Stetser D A 1965 Appl. Phys Lftt. 7 71 [47] Deutsch T 1965 Appl. Plzy,!. Lett. 7 80 1481 Mocker H W a n d Collins R J 1965 Appl. Pliys. Lett. 7 270

254

P M W French

[49] DeMaria A J SteUer D A and Heynan I.1 1966 Appl. P h y ~Letr. . 8 174 [SO] Schmidt W and Schafer F P 1968 P ~ J Leu. X 26A 558 [51] Bradley D J, Durrant A J F, ONeill F and Sulherland B 1969 Pltys. Lett. 30A 535 1.521 Bradley D J 1974 Oplo-eleelroii G 25 1531 Glen W H. Brienza M J and DeMaria A J 1968 Appl. Plfys. Lett. 12 54 [54] Bradley D J and Durrant A J F 1968 Plys. Lett. 27.4 73 [ 5 5 ] SoRer B H and Linn J W 1968 J. Appl. Plfys. 39 5859 [56] Armstrong 1 A 1967 Appl. Plrys. Lelt. 10 16 [57] Giordmaine J A, Renlzepis P M, Shapiro S L and Wecht K W 1967 Appl. Phys. Lett. I I 216 1581 Garmire E M and Yariv A 1967 IEEE . I Olrenlurn . Electron. Q E 3 222 i59j Weber H P 1968 N i p . Letr. 27A 321 1601 Bradlev D J. Liddv B. Roddie A G. Sibbett W and Slat W E 1971 Ont. CO~JJWJUIJ. 3 426 i61i Letokdov V'S 1969 Sou. Phys.-JfP Ul 562 [62] Fleck J A 1968 Appl. P l y Lelf. 12 178 (631 Penney A W and Heynau U A 1966 Appl. Plxys. Lett. 9 57 [64] Kachen G, Steinmetz L a n d Kysilka J 1968 dppl. Phys. Lett 13 229 [65] Burneika K,Grigonis R, Piskarskas A, Sinkyavichyus G and Sirutkaitis V 1988 Kwnr. Elecrron. 15 1658 [6h] DiDomenico M, Marcos H M, Geusic J E and Smith R E 1966 Appl. Phys. Lett. 8 I80 [671 Osterink 1 M and Foster J D 1968 J . Appl, Pliys. 39 4163 [68] Kuizenga D J and Siegman A E 1970 J , IEEE Quanrum Elecrron. Q E 6 694, 709 [69] Crowell M I i 1965 IEEE J. Quanrrmt Electron. QE-1 12 [70] Statz H and Tang C L 1965 J. Appl. Phyx 36 3923 [71] Duguay M A, Shapiro S L and Rentzepis P M 1967 Pltj~. Rev. Lerr. 19 1014 [72] Gaddy 0Land Schaefer E M 1966 dppl. Pljys. Lett. 9 281 [73] Lamb W E Jr 1964 Phyx Reo. I 3 4 A1429 [74] Fox A G and Smith P W 1967 P h p Reo. Lett. 18 826 [75] Laussade J P and Yariv A 1968 Appl Phys. Len. 13 65 [76] Comly J. Yariv A and Garmire E 1969 Appl, Plijs. Lell. 15 148 [77] Sicbert L D and Montry G R 1978 Appl. P h p . Lett. 33 645 [78] Garmire E M and Yariv A 1967 IEEE 1. Qrxuitwn Elecfron. Q E 3 222, 371 (erratum) [79] Cubbcda R and Svelto 0 1969 P h p Lett. 29A 78 [SO] Smith A W 1969 Appl, Phys. LeN. 15 194 [81] Duguay M A, Hanson J W a n d Shapiro S L 1970 IEEE J. Quantwi Electron, Q E 6 725 [82] Gires F and Toumois P 1964 C. R. Acad. Sci., Paf'is 258 61 I2 [83] Giodmainc J A, Dvguay M A and Hansen 1 W 1968 IEEE J. Quantum Elecrron. Q E 4 252 [84] Duguay M A and Hansen J W 1969 Appl. Pliys. Lert. 15 194 [SS] Treacy E B 1968 Plt)ir. Lerr. %A 34 [86] Treacy E B 1969 IEEE J. Qimtum Elecrro,i. Q E 5 454 [87] Fischer R A. Kclley P L a n d Gustafson T K 1969 Appl. Pltys. Lett. 14 140 [88] Laulrereau A 1969 Phys. Lerf. 29A 539 [89] Lariontsev E G and Serkin V N 1975 Sou. J. Quanlum E/eclrvlr. 5 796 [90] Islam M N,Sunderman E R, Soccolich C E, Bar-Joseph I, Sauer N, Chang T Y and Miller B I 1989 J. IEEE Qiron1zmr Elcelron. QE-2.5 2454 1911 Silberberg Y,Smith P W,Eilenberger D J, Miller D A B, Gossard A S and Wiegmann W 1984 Opt. Lett. 9 507 [92] Silberberg Y and Smith P W 1986 lEEE J. Quonriiln Electrarr. QEZZ 759 [93] Cotzinc, S W. Rowers J E, Prrbylek G, Korcn U, Miller B I and Soccolich C E 1988 Appl. Phys. Lett. 52 348 1941 Dieis J-C1990 Femtosecond dye lasers Dye Loser Principles: Wilh Applicarions. Quantunt ElectronicsPri,J<'i/de,S mtd Applicolions ed F J Duarte and L L Hillman (San Diego, CA: Academic) pp 41 -1 32 [95] lppen E P, Shank C V and Dienes A 1972 Appl Pliys. Lerr. 21 348 [96] New G H C 1972 Opr, Comrrnm. 6 188 [97] New G H C 1974 IEEE J. Qtiaiztfnil Electron, QE10 I15 [98] Letouzey J P and Sari S 0 1973 dppl. Pliys. Len. 23 311 [99] Shank C V and lppen E P 1974 Appl. P l y Lett. 24 373 [loo] lppen E P and Shank C V 1975 dppl. Pliy.s. Lpll, 27 488 [loll Haus H A 1975 IEEE J . Qimiluin Electron. Q E l l 736

The getteration of ulrrmhort laser pulses

255

[IO21 Ruddock I S and Bradley D J 1976 Appl. Pliys. Lett. 29 296 [I031 Diels J-C, Van Stryland E and Benedict G 1978 Opt. Cornfltun. W 93 [IO41 Fork R I, Grecnc B 1 and Shank C V 1981 Appl. Plrw. Letf.38 671 [IOS] Halboul J M and Tang C L 1983 IEEE J. Quantum Electron. Q E 1 9 487 [IO61 Stix M S and Ippen E P 1983 IEEE J . Quantum Elecrron. Q G l 9 520 [I071 Siegman A E 1981 Opt. Len. 6 33 [IO81 Vanherzeele H, D i e l s J-C and Torti R 1984 Opt. Let,. 9 549 [IO91 Dietel W, Dopel E, Kuhlke D and Wihelmi B 1982 Opt. Commun. 43 433 [I IO] Dietel W, Fontaine J J and Diels J-C 1983 Opt. Lett. 8 4 [I I l l Fontaine J J, Dietel W and Diels J-C 1983 IEEE J. Qrruntunr Electron. QE-I9 1469 [I 121 Dicls J-C. Fontaine J J. McMichael I C, Wihelmi B, Dietel W and Kuhlke D 1983 Sou. J. Qtnnfuni Electron. 13 1562 [I 1.31 Rudolph W a n d Wilhelmi B 1984 Opf. COWJWI. 49 371 [I 141 Fork R L, Gordon J P and Martinez 0 E 1984 Opt. Lett. 9 I50 [I151 Silvestri de S, Laporta P and Svelto 0 1984 IEEE J. Qrruntunt E/ecfron.QE-20 533 [ I 161 Dietel W. Dopel E,Hehl K, Rudolph W a n d Schmidt E 1984 Opt. Comfnun. 50 179 [ I I71 Martinez 0 E, Fork R L and Gordon J P 1984 Opt. Lett. 9 156 [ I 181 Diels J-C, Dietel W, Fontaine J J, Rudolph W and Wilhelmi B I985 J. Opt. Soc. Am. B 2 680 [I 191 Martinez 0 E, Fork R L and Gordon J P 1985 J . Opt. Soc. Am. B 2 753 [I201 Valdmanis J A, Fork R L and Gordon J P 1985 Opt. Lett. 10 131 [ 1211 Valdmanis J A and Fork R L 1986 IEEE J. Quuntuin Elecrron. Q K 2 2 112 [I221 Salin F, Grangier P, Roger G and Brun A 1986 Pliys. Rev. Lett. 56 1132 [I231 Haus H A and Silberberg Y 1986 IEEE J . Quantum Electrorr. Q E 2 2 325 [ 124) Avramopoulos H, French P M W,Williams J A R, New G H C and Taylor J R 1988 IEEE J. Quantum Electron. Q E W 1884 [I251 French P M W, Williams J A R, New G H C and Taylor J R 1987 Reu. Pl?)s, Appl. 22 I651 1126) French P M W and Taylor J R 1988 Opt. Lett 13 470 [I271 Williams J A R, French P M W, Taylor J R and Goldsmith H 1988 Opt. Letr. 13 811 [I281 Sibbett W, Taylor J R a n d Welford D 1981 IEEE J . Quunturn Electron. QE-17 500 [I291 New G H C and Rea D H 1976 Appl. Phys. 47 3107 [I301 Heppner J and Kuhl J 1985 Appl. P h j ~ Left. 47 453 [I311 French P M W. Chen G F and Sibbett W 1986 Opt. C o w " . 57 263 (1321 Kuhlke D, BonkhoRer T and von der Linde D 1986 Opt. Comfnuit. 59 208 [I331 Yamashita M, Ishikawa M, Tonzuka K and Sato T 1986 Opt. Lett. 11 504 [I341 French P M W, Gomes A S L, Gouveia-Neto A S and Taylor J R 1986 Opt. Coniinrm. 59 366 [I351 Smith K, Langford N, Sibbett W and Taylor I R 1985 Opf, Lett. 10 559 [I361 Chan C K and Sari S 0 1974 Appl. Phys. Letf.25 403 [I371 Eckstein J N, Ferguson A I, Hansch T W, Minard C A and Chan C K 1978 Opt. Coiiaiun. 27 466 [I381 Kuhl K, Klingenberg H and von der Linde D 1979 Appl. P l i j ~ 18 279 [I391 Seilmeier A, Kaiser W, Seus B a n d Drexhage K 1983 Opt. Let!. 8 205 [I401 Mollenauer L F and Bloom D 1979 Opt. Letr. 4 427 [I411 Mollenauer L F, Vieria N D and Szeto L 1982 Opt. Let!. 7 414 [I421 Heritage J P and lain R K 1978 Appl. P h s . Len. 32 101 [I431 Ausschnitt C P and lain R K 1978 Appl. Pltys. Left. 32 727 11441 Ausschnitt C P, Jain R K and Heritage J P 1979 lEEE J . Quontunr Electron. QE-15 912 [I451 Johnson A M and Simpson W M 1985 J . Opt. Soc. Am. B 2 619 [I461 Jain R K and Ausschnitt C P 1978 Opt. Lett. 2 117 18 279 [I471 Kuhl J, Klingenerg H a n d von dcr Linde D 1979 Appl. P h j ~ [I481 Johnson A M and Simpson W M 1986 IEEE J . Quanrwn Electron. Q E 2 2 133 [I491 Kafka J D and Baer T 1985 Ultrashort Pulse Speetroucopy and Applications (SPIE533) [I501 McDonald D B, Waldeck D and Fleming G R 1980 Opt. Coiimtun. 34 127 [I511 May P G, Sibbett W, Smith K, Taylor J R and Willson J P 1982 Opt. Cornniun. 42 285 [I521 Scavannec A 1976 Opt. Conmin. 17 14 [I531 Frigo N 1, Daly T and Mahr H 1977 IEEE J . Quunturri Electron. Q E 1 3 101 [I541 Catherall 1, New G H C and Radmore P M 1982 Opl. Leu. 7 319 [I551 Kim D M. Kuhl J, Lambrich R and yon der Linde D 1978 Opf. Coiiiniun. 27 123 [ I 561 Smith K, New G H C and Catherall J 1986 Opt. Commn. 58 I18 [I571 Catherall J and New G H C 1986 IEEE J. Q i i ~ i r t ~ fElectron. li QE22 1593 [I581 Van Stryland E W 1979 Opt. Conunun. 31 93

256

P M W French

[159] Stamm U 1988 Appl. Plrys. B45 101 [I601 Beaud P, Zyssett B, Schwarrenbach A P and Weber H P 1986 Opt. Lett. I I 24 [I611 Choa F Sand Liu P L 1988 Opt. Letl. 13 743 [I621 N e w G H C 1990 Opt. Lett. 15 1306 [I631 Peter D S, Beaud P, Hodel W and Weber H P 1991 Opt. Lett. 16 405 [I641 Shank C V and lppen E P 1974 Appl. Phjs. Lett. 7 . 4 373 [I651 Ryan J P, Goldberg L S and Bradley D J 1978 Opt. Cominitn. 27 127 [I661 Isltida Y, Yajima T and Naganuma K 1980 Jupun J. Appl. Phys. 19 717 11671 Fehrenbach G W. Gruntr K J and Ulbrich R G 1978 Appl, Phys. Left. 33 159 [I681 Smith K, Sibbctt W and Taylor J R 1984 Opt. Cornntufl. 49 359 [I691 Mourou G A and Sizcr T II 1982 Opt. Commun. 41 47 [I701 Johnson A M and Simpson W M 1983 Opf.Lett. 8 554 [I711 Vanhemele H,Torti R and Diels J-C 1984 Appl. Opt. 23 4182 [I721 Norris T,Suer T and Mourou G 1985 J . Opt. Soc. Ani. B 2 613 [I731 Dobler J, Scliulz H H and Zinth W 1986 Opt. Coinmm 57 407 [I741 Dawson M D, Boggess T F, Garvey D W and Smirl A L 1986 Opf. Commun. 60 79 [I751 Dawson M D, Boggess T F and Smirl A L 1987 Opf. L e u 12 254 [I761 Knon W H 1987 J. Opt. Soc. Ant. B 4 1771 [I771 Chesnoy J and Fini L 1986 Opt. Letr. 1 1 635 [I781 Dawson M D, Mason D, Boggess T F and Smirl A L 1988 Opl. Left. 13 126 [I791 Kubota I+, Kurokawa K and Nakazawa M 1988 Opl. Lptt. 13 749 [I801 Angel G, Gagcl R and Laubereau A 1989 Opt. Lett. 14 IO05 [ISl] Agrawal G P 1989 Nonheor Fiber Opftcs (New York: Academic) [I821 Gomes A S L, Gouveia-Nelo A Sand Taylor J R 1988 Opt. Qunntsin Electron. 20 95 [I831 lppen E P, Shank C V and Gustafson T K 1974 Appl. Phys. LeIt. 24 190 [I841 Miya T,Terunuma Y, Hosaka T and Myashila T 1979 Electron. Lett. 15 106 [I851 Slolen R Hand Lin C 1978 Phys. &U. A 17 17 [I861 Nakatsuka H, Grischkowsky D and Balant A C 1981 Phys. Rev. Left. 47 1910 [I871 Shank C V, Fork R L,Yen R, Stolen R H and Tomlinson W J 1982 Appl. P / i p . Lett. 40 761 I1881 Grischkowsky D and Balant A C 1982 Appl, Phys. Lctt. 41 I [I891 Nikolaus Band Grischkowsky D 1983 Appl. Pl~js. Left. 42 I [I901 Nikolaus B dnd Grischkowsky D 1983 Appl. Phjs. Lett. 43 228 [I911 Tomlinson W J, Stolen R H and Shank C V 1984 3. Opt, Soc. Am. B 1 139 11921 Fujimoto, J G. Weiner A M and lppen E P 1984 Appl. P h p . Len. 44 832 [I931 Halbout J M and Griscltkowsky D 1984 Appl. Phys. Lett. 45 832 [I941 Kaka J D, Kolner B H,mer T and Bloom D M 1984 Opt, Lelt. 9 505 [ 1951 Gomes A S L, Sibbett W and Taylor J R 1985 IEEE J. Q U U I ~E ~/ c Uf r ~oIn .QG2l 1157 [I961 Gouveia-Ncto AS, Gomes A S Land Taylor J R 1987 Opt. Lett. I2 395 [I971 Palfrey S L and Grischkowsky D 1985 Opl. Left. 10 562 [I981 Fork R L, Shank C V, Hirlimann C, Yen R and Tomlinson W J 1983 Opt. Lett. 8 I [ 1991 Migus A, Antonetti A, Etcbcparet J, Hulin D and Orszag A 1985 J. Opt. Soc. Ani. B 2 584 (2001 Knox W H,Fork R L. Downer M C. Stolen R H. Shank C V and Valdmanis J A 1985 Appl. Plzys. Lett. 46 I120 [ZOI] Fork R L. Brito-Cruz C H, Becker P C and Shank C V 1987 Opt. Left. 12 483 [202] Kapteyn H C and Mumane M M 1994 Opt. Plio1onic.f N e w 5 20 I2031 Delfyett P J, Florn L T. Stofel N, Gmittcr T, Andreadakis N C, Silberberg Y, Heritage J P and Alphonse G A 1992 IEEE J. Qtmfum Electron. QED 2203 I2041 Dahlstrom L 1972 Opt, Conuiiun. 5 157 [205] Comly J, Garmire E, Laussadc J P and Yariv A 1968 App/. P h p . Lett. 13 176 [206] Mollcnauer L F and Stolen R H 1984 Opt. Lett. 9 13 [207] Mitsclikc F M and Mollcnauer L F 1986 IEEE 3. Quonnnn Elcclron. QEU 2242 [208] Blow K J and Wood D 1988 J. Opt. Soc. Am. B 5 629 [209] Pinto J F, Yakymyshyn C P and Pollack C R 1988 Opt. Lett. 13 383 [ZlO] Blow K J and Nelson B P 1988 Opt, Lett. 13 1026 [Z! I ) Kcan P N. Zhu X, Crust D W, Grant R S, Langlord N and Sibbett W 1989 Opt. Letr. 14 39 [212] Piche M and Oucllette F 1986 Opt, Lett. I I I5 [213] Quellelte F and Piche M 1986 Opt. Consnun.60 99 I2141 Mark J, Liu L Y, Hall K L, Haus H A and Ippen E P 1989 Opt. Lett. 14 48 [215] F r e d P M W, Williams J A R and Taylor J R 1989 Opt. Lett. 14 668

The genevation o f ullrashorl laser pulses

251

[216] Goodberlet J, Wang J, Fujimoto J G and Schulz P A 1989 Opt. Lett. 14 1125 [217] Goodberlel J, Jacobson J, Fujimoto J G, Schulr P A and Fan T Y 1990 Opt. Leff.15 504 [218] Liu L Y , Huxley J M, lppen E P and Haus H A 1990 Opt. Len. 15 553 [219] Krausz F, Spielmann Ch, Brabec T, Wintner E and Schmidt A J 1990 Opf. Len. 15 737 [220] Liu J M and Chee J K 1990 Opt. Lett. 15 685 [221] Malcolm G P A , Curley P F and Ferguson A I 1990 Opt. Lett. 15 1303 (2221 Sennaroglu A, Carrig T J and Pollock C R 1992 Opt. L e x 17 1216 12231 Carruthers T F and Duling I N Ill 1990 Opl. Left. 15 804 [224] Doran N J and Wood D 1988 Opt. Letr. 13 56 I2251 Femann M E, Haberl F, Hofer M and Hochreiter H 1990 Opt. Lett. 15 752 [226] F e m a n n M E, Hofer M, Haberl F, Schmidt A J and Tun L 1991 Opt. Lett. 16 244 [227] Bulushev A G, Dianov E M and Okhotnikov 0 G 1991 Opt. Lett. 16 88 [228] Duling I N I11 1990 All-firepassiuelp mode-lockedfrgrae 8 laser Proc. OSA Annual aeeting, (Boston, 1990) PDP4 (Washington, DC: Optical Society or America) pp 9-13 12291 Richardson D 1, Laming R 1, Payne D N,Matsas V and Phillips M W 1991 Electron. Len. 27 543 (2301 Ndkazawa M N, Yoshida E and Kimura Y 1991 Appl, Phys. Lett. 59 2073 12311 Dennis M L and Duling I N 111 1992 Elecrvon. Lett. 28 1894 [232] Pandit N, Noske, D U, Kelly S M J and Taylor J R 1992 Eleclron. Lett. 28 455 [233] Nakazawa M N.Yoshida E and Kimura Y 1993 Electron. Left.29 63 [234] Stolen R H, Bolineau J and Ashkin A 1982 Opt. Lett. 7 512 [235] LaGasse M J, Wong D L, Fujimoto J G and Haus H A 1989 Opt, Lett. 14 311 [236] Haus H A, Fujimoto J G and lppen E P 1991 J. Opt. Soc. Am. B 8 2068 [237] Hofer M, Fermann M E, Haberl F, Ober M H and Schmidt A J 1991 Opf. Letr. 16 502 [238] Matsas V 1, Newson T P, Richardson D J and Payne D N 1992 Electron. Left. 28 1391 [239j Noske D U, Pandit N and Taylor J R 1992 Electron. Lett. 28 2185 [240] F e n " M E, Andrejo M 1, Silberberg Y and Weiner A M 1993 Opt. Left. 18 48 [241] Tamura K, lppen E P, Haus H A and Nelson 1,E 1993 Opt. Lett. IS 1080 [242] Ober M H, Hofer M and Fermann M E 1993 Opf. Lett. I8 367 [243] Kcller U, Knox W H a i d Roskos H 1990 Opt. Lett. 15 1377 [244] Haus H A, Keller U and Knox W H 1991 J . Opt. Soc. Ani. B 8 1252 [245] Jambovitz G R, Keller U and Asom M T 1992 Opt. Lett. 17 1791 12461 Mellish R, French P M W. Taylor J R, Delryett P J and Florez L T 1993 Election. Lett. 29 894 (2471 Keller U, Miller D A B, Boyd G D, Chiu T H, Ferguson J F and Asom M T 1992 Opt. Lett. 17 505 [248] Keller U, Cliiu T H and Ferguson J F 1993 Opf. Letf. 18 1077 [249] Keller U, Knox W H and 'IHooit G W 1992 IEEE J . Quantuin Electron. QE-28 2123 [250] French P M W, Kelly S M J and Taylor J R 1990 Opt. Lett. 15 378 [251] Wigley P G J, French P M W a n d Taylor J R 1990 Gleclrorr. Leu. 26 1238 [252] French P M W, Noske D U, Rizvi N H, Williams J A R and Taylor J R 1991 Opt. Commun. 83 185 [253] Smith P W 1967 Appl. Plij~. Lett. IO 51 [254] Sargsjan G, Stamm U, Unger C, Zschocke W a n d Ledig M 1991 Opf. Commun. 86 480 !255] Cu!lcr C C 1992 IEEE J . Quuntum Electron. QE28 282 [256] Wu Q, Zhou J Y, Huang X G,Li Z X and Li Q X 1993 J . Opt. Soc. A m . B IO 2080 12571 Spence D E, Kean P N and Sibbett W 1990 Sub-lOO/s pulse generutloti form U self-moddocked Ti:sapplrire lascr Conf on LosersmidEiecI~o-Optics (Anaheim, 1990) paper CPDPIO-I (Washington, DC: Optical Society o i America) pp 619-20 [258] Spence D E, Kean P N and Sibbett W 1991 Opt. Lett. 16 42 [259] Sarukura N, lshida Y, Nakano H and Yamamoto Y 1990 Appl. Phps. Letf. 56 814 [260] Sarukura N and lshida Y 1992 IEEE J . Quantum Electron. QE28 2134 [261] Squier J, Salin F, Coe S, Bad0 P and Mourou G 1991 Opt. Lett 16 85 [262] Piche M 1991 Opt. Co,nmun. 86 156 [263] Marconi M C, Martinez 0 E and Biodati F P 1987 Opt, Contntrni. 63 21 1 [264] Negus D K, Spinelli L, Goldblatt N and Feuget G 1991 Sub-100 fs pulse generation by Kerr Lens Mode-locking in Ti:AI20, OSA Proc. on Aduunced Solid-Stute Lasers (Hilton Head, 1991) paper PDP4 (Washington, DC: Optical Society of America) pp 120-4 [265] Squier J, Salin F, Mourou G and Harler D 1991 Opf. Lett. 16 324 [266] Likforman J P, Grillon G, JoRre M, Le Blanc C, Migus A and Antonetti A 1991 Appl. Phys. Lett. 58 2061 [267] Curley P F and Ferguson A 1 1991 Opt, Lett. 16 1016 I2681 Keller U, 'tHoof1 G W, Knox W H and Cunningham J E 1991 Opt. Left. 16 I022

25 8
[269] [U01 [271] [272]

P M W French

Spielmann Ch, Krausz F, Brabec T, Wintner E and Schmidt A J 1991 Opt. Letf. 16 1180 Rizvi N H, French P M W, Taylor J R, Delfyett P J and Florez L T 1993 Opt. Lett, 18 983 Rimi N H, French P M W a n d Taylor J R 1992 Opt. Leu. I1 279 Gabetta G, Huang D,Jacobson J, Ramaswamy M, lppen E P and Fujimoto J G 1991 Opt. Lett. 16 1756 [273] Malcolm G P A and Ferguson A I 1991 Opt. Left.16 1967 [274] Ramaswamy M, Gouveia-Net0 A S,Negus D K, lzatt J A and Fujimoto J G 1993 Opt. Lett, 18 1825 [2751 Liu K X. Flood C J, Walker D R and van Driel H M 1992 Opt. Lett. 11 1361 [276] Seas A, Petricevic V and Alfano R R 1992 Opt. Lett. 17 937 I2771 Sennaroglu A, Pollack C R and Nathel H 1993 Opf.Lett. 18 826 [278] Yanovsky V, Pang Y,Wise F and Minkov B I 1993 Opt. Lett. IS 1541 [279] Li Kam Wa P, Chai B H T and Miller A 1991 Electron. Lett, 27 2351 [280] Rizvi N H.French P M W a n d Taylor J R 1992 Upt, Letf. 17 879 [281] Evans J hl, Spence D E, Sibbett W, Chai B H T and Miller A 1992 Opt. Left. 17 1447 [282] Rizvi N H, French P M W and Taylor J R 1992 Opt Lett. 17 1607 [283] Li Kam Wa P, Chai B H T and Miller A 1992 Opt. Lett. 11 1438 [284] Sennaroglu A, Pollack C R and Nathel H 1994 Opt. Left,19 390 [2851 Conlon P, Tong Y P, French P M W, Taylor J R and Shestakov A V 1994 Upf. k i t . 19 1468 [286] Salin F, Squier J and Piche M 1991 Opl. Lett. 16 1674 [287] Clien S a n d Wang J 1991 Opt. Lett. 16 1689 [288] Brabec T, Spielmann Ch and Krausz F 1991 Opt. Len. 16 1961 [289] Huang D. Ulman M, Acioli L H, Haus H A and Fujimoto J G 1992 Opt. Lett, 17 jl I [2901 Huang C-P, Kapteyn H C, McIntosh J W a n d Murnane M M 1992 Opt. Lett. 17 139 [291] Krausz F.Spielmann Ch, B r a b s T, Wintner E and Schmidt A J 1992 Opf. Lett. 11 204 [292] Spielmann Ch. Curley P F. Brabec T. Wintner E and Krausz F 1992 Electron. L m , 28 1532 [2931 Iiuang G P , Asaki M T, Backus S, Murnane M M and Kapteyn H C 1992 Opt. Leu, 17 1289 [294] LemotTB E and Barty C P J 1992 Opt. Lett. 11 1367 I2951 Asaki M T, Huang C-P. Garvey D, Zhou J, Kapteyn H C and Murnane M M 1993 Opt. L m 18 977 [296] Sting1 A. Spielmann Ch, Krausz F and Szipocs R 1994 Opt. Left. 19 204 I2971 Tamura K, Jacobson J. lppen E P, H a u s H A and Fujimoto J G 1993 Opt, Lett. 18 220 [298] Lai Ming 1994 Opt. Len. I9 722 12991 Delfyett P J, Florez L 7, Stofel N, Gmitter T, Andreadakis N C, Silberberg Y, Heritage J P and Alphonse G A 1992 IEEE J. Quantum Electron. QEZ8 2203 [300] Corrine S W, Bowers J E, Przbylek G, Koren U. Miller B 1 and Soccolich C E 1988 AppL P l r p Lett. 52 348 I3011 Ho P-T, 1979 Electron. Lett. I 5 526 [302] Kuhl 1. Serenyi M and Gobel E 0 I987 Opf. Lett. 12 334 I3031 Yokoyama H, It0 H and lnaba H I982 Appl. Phys. Lett, 40 105 13041 Silberberg Y,Smith P W, Eilenberger D A B, Miller D A B, Gossard A and Wiegman W 1984 Opi. Left. 9 507 [30S] Chen Y-R and W u M C 1992 IEEEJ. Quannnn Efemon, QED 2176 13061 Fcrman M E, Andrejco M 1, Silberberg Y and Stock M L 1993 Opt. Lett. 18 894 13071 Takada A, Miyazawa H 1990 Electron. Lett. 26 217 I3081 Takara H, Kawanishi S, Saruwatari M, Noguchi K 1992 Electron. Left.28 2075 e Souza E A, Soccolich C E, Pliebel W. Stolen R H, Simpson J R and DiGiovanni D J 1993 Electron. 13091 D Lett. 29 447 [3101 Ober M H, Hofer M, Chiu T H and Keller U 1993 Opt. Leu. 18 1532 [31 I] Tamura K, Haus H A and lppen E P 1992 Electron. Lett. 28 2226 M E, Blixt P, Tellefsen Jr J A and Harter D 1994 Opt, Lett. 19 1043 [3121 Galvanauskas A, Fer" [313] Ross M 1968 Proc. IEEE56 196 [314] Maker G T, Keen S J and Ferguson A I 1988 Appl. Phys. Lett. U I675 (3151 Maker G T and Fcrguson A I 1989 Opt, Left. 14 788 [316] Maker G T and Ferguson A I 1989 Electron. Left.25 1025 [317] Weingarten K 1, Shannon D C, Wallace R W and Keller U 1990 Opt. Lett. 15 962 [318] Bas" S and Byer R L 1988 Opt. Lett. 18 458 [319] Krausz F, Brabec T, Wintner E and Schmidt A J 1989 Appl, P h ~ s Left. . 55 2386 [320] Haus H A and Silberberg Y 1986 IEEE J. Quanttmt Electron. QE22 325 [321] Goodberlet J, Jacobson J, Fujimoto J G, Schulz P A and Fan T Y 1990 Opt, Left, 15 504 [322] Malcolm G P A , Curley P F and Ferguson A I 1990 Opt. Lett. 15 1303

The generation of ultrashort laser pulses

[323] (3241 I3251 13261

259

Wcingarten K J, Keller U, Chiu T H and Ferguson J F 1993 Opt. Lett. 18 640 Malcolm G P A and Ferguson A I 1991 Opl. Lett. 16 1967 Liu K X,Flood C J. Walker D R and van Driel H M 1992 Opl. Lett. 17 1361 Lamb K, Spence D E, Hong J, Yelland C and Sibbett W 1994 A scif-ntode-locked 7i:sapphire loser pwnped by U frequency-doubled Nd: YLF miniluser OSA Topicul Meeting O ~ IUltrafust Phenomenu (Dunu Point. 1994) paper MA4 (Washington, DC:Optical Socicty of America) pp 10-2 [327] French P M W, Mellish R, Taylor J R, Delfyett P I and Florez L T 1993 El@-tron. Lett. 29 1263 [328] Balembois F, Georges P and Brun A 1993 Opt. Lett. 18 1730 [329] Mellish R, French P M W, Taylor J R, Delfyett P J and Florez L T 1993 Opt. Lett. 18 1934 [330] Mellish R, French P M W, Taylor J R, Delfyctt P J and Florez L T 1994 Electron. Lett. 30 223 [331] Kopf D, Weingarten K 1, Brovelli ?, Kamp M and Keller U 1994 Diode-pumpeda~h-!oOfsposriwly modelocked Cr: LiSAF using un A-FPSA C o n ! O H Lo,wrsand Electro-Optic3 (Anahcim. 1994) paper CPD22 (Washington, DC: Optical Society o f America) pp 49-50 (3321 Dymott M 1 P, Botheroyd I M, Hall G I , Lincoln J R and Ferguson A I 1994 Opt. Lett. I9 634 [333] Lincoln J R, Botheroyd 1 M, Dymott M J P and Ferguson A I 1994 All-rolid-slare/eitrt~sec~~d Kerr lens nrodc-locked Cr: LiSAF laser Cor6 on Losers and Electro-Optics (Anaheim, 1994) paper CW12 (Washington, DC: Optical Society of America) pp 234-6 I3341 Yariv A 1989 Qrronturn Electronics 3rd ed (New York: Wiley) [335] Hemnan J and Wilhelmi B 1987 Lusemfor Ulrmslror.r Light Pulses (Berlin: Akademie) [336] Kuizenga D J and Siegman A E 1970 IEEE J. Quonturn Electron. Q E 6 694, 709 [337] Huggett G R 1968 Appl. Plzys. Lett. I3 186 [338] Spence D E, Evans J M, S l a t W E and Sibbett W 1991 Opt. Lett. 16 I762 13391 Scavannec A 1976 Opt. Conunun. 17 14 [340] Frigo N J, Daly T and Mahr H 1977 IEEE J. Qumtum Electron Q E 1 3 101 [341] Auschnitt C P and Jain R K 1978 Appl. Phys. Letf.32 727 [342] Auschnitt C P, Jain R K and Heritage J P 1979 IEEE J . Quarzrurn Electron. QEl5 912 [343] Catherall J M, New G H C and Radmore P M 1982 Opt. Lett. 7 319 [344] Catherall J M and New G H C 1986 IEEE J. Quontum Electron. Q E 2 2 1593 13451 Stamm U 1988 Appl. Phjv. B 45 101 [346] Johnson A M and Simpson W M 1986 IEEE J. Qniantum Electron. Q E 2 2 133 13471 Johnson A M and Simpson W hi 1985 J . Opt. Soc. Am. B 2 619 [348] Van Stryland E W 1979 Opt. Convnun. 31 93 [349] Peter D S, Beaud P, Hodel W a n d Weber H P 1991 Opt. Lett. 16 405 [350] Runge P K 1971 Opt. Cornman. 4 195 [351] Clian C K and Sari S 0 1974 Appl. Phys. Lett. 25 403 [352] DeMaria A J, Stetser D A and Heynan H 1966 Appl. Piys. Lett. 8 174 [353] Bradley D J, Durrant A J F, ONeill F and Sutherland B 1969 Phys. Leu. 30A 535 13541 Islam M N. Sunderman E R. Soccolich C E. Bar-JoseDh . I. . Sauer N. Chane T Y and Miller B I 1989 IEEE J . Quamnn Electron. Q E Z S 2454 [355] DeSoura E A, Islam M N. Soccolich C E, Pleibel W, Stolen R H,Simpson J R and DiGiovanni D J 1993 Elecrrorr. Lett. 29 447 13561 Mellish R, French P M W, Taylor J R, Delfyett P J and Florez L T 1993 Eicctron. Lett. 29 894 [357j Mellish R, French P M W. Taylor J R, Delfyett P J and F l o r a L T 1994 Electron. Lett. 30 223 [358] Keller U, Knox W H and 'tHooft G W 1992 IEEE J. Qifanturn ElectroB. QE-28 2123 [359] Keller U, Miller D A B. Boyd G D, Chiu T H, Ferguson J F and Asom M T 1992 Opf. Lett. 17 505 [360] Haus H A, Keller U and Knox W H 1991 J . Opt. Soc. Ant. B 8 1252 [361] Ippen E P, Shank C V and Dienes A 1972 Appl. Phys. Lett. 21 348 [362] Fork R L, Brito-Cruz C H, Becker P C and Shank C V 1987 Opt. Lett. 12 483 [363] Diels J-C, Fontaine J J, McMichael I C, Wilhelmi B, Dietel W and Kuhlke D 1983 Sou.J. Quontunt Elecfron. 13 1562 13641 Fork R L, Greene B I and Shank C V 1981 Appl. Phjrr. Lett 38 671 [365] Siegman A E 1981 Opt. Lett. 6 33 [366] Mark J, Liu L Y, Hall K L, Haus H A and lppen E P 1989 Opt. Letf. 14 48 13671 Mollenauer L F and Stolen R H 1984 Opt, Leu. 9 13 [368] Kean P N, Zhu X, Crust D W, Grant R S, Langford N and Sibbett W 1988 Opt. Lett. 13 IO26 [369] French F M W, Williams J A R and Taylor J R 1989 Opt. Lett. 14 668 [370] Goodberlet I, Fujimoto J G . Schulz P A and Fan T Y 1990 Opt. Lett. 15 504 [371] Liu J M and Chec J K 1990 Opt. Lett. 15 685 [372] Goodberlet J, Jacobson J, Fujimoto J G and Schulz P A 1989 Opf. Lett. 14 I125

. .

260

P M W French

13731 Krausz F, Sliielmann Ch, Brabec T, Wintner E and Schmidt A J 1990 Opt. Lett. I 5 737 (3741 lppen E P. Yiu L Y and Haus H A 1990 Opt. Leu. 15 183 (37.51 Milschke F M and Mollenauer L F 1986 IEEE J. Quarrtrrm E/@ctron. QE22 2242 13761 Ouellett F and Piche M 1986 Opt. Corniiiim. 60 99 [377] Morin M and Piche M 1989 Opt. Lett. 14 I I19 [378] Spielmann Ch. Krausz F, Brabec T, Wintner E and Schmidt A 1 1991 App/. P h s . Lett. 58 2470 [379] Otsuka K 1983 Opt. Lett. 8 471 [380] Carruthers T F and Duling I N Ill 1991 Opt. Lett. 1 6 539 [381] Doran N J and Wood D 1988 Opi.Lett. 13 56 [382] Duling I N 111 1991 Opt. Left. 16 539 13831 Nakazawa M, Yoshida E aud Kimura Y 1993 Elemon. Lftr. 29 63 [384] Blow K J, Doran N J and Nayar B K 1989 Opt. Lett. 14 754 [385] Fermann M E, Haberl F, Holm M and Hochreiter H 1990 Opt. Left. ?5 752 (3861 Dahlstrom L 1972 Opt. Conusun. 5 157 6 502 (3871 Hofer M, Fermann M E, Haberl F, Ober M H and Schmidt A I 1991 Opt. Lett I (3881 Hofer M. Ober M H, Haberl F and Fermann M E 1992 IEEE J. Quantum Electron. Q E 2 8 720 I3891 Malsas V J, Newson T P, Richardson D J and Payne D N 1992 Elccfron. Left, 28 1391 I3901 Tamura K. Haus H A and lppen E P 1992 Electron. Left. 28 2226 E P, Yiu L Y and Haus H A 1990 Opt. Left, I 5 183 [3911 I ~ p e n [392j Femann M E, Andrejco M J, Silberberg Y a n d Stock M L 1993 Opt. Lett. 18 894 [3931 Kelly S M J 1992 Elcetr.on. Letf. 28 806 [3943 French P M W, Kelly S M J and Taylor J R 1990 Opt, Leu. 15 378 [395] Cutler C C 1992 IEEE J . Quantum Electroii. QE28 282 (3961 Smith P W 1967 App/. Phys, Lett. I 0 51 (3971 Bambini A and Burlamacchi P 1968 J. Appl. Phys. 39 4864 (3981 Wigley P G J, French P M W a n d Taylor J R 1990 Electron. Lcff.26 1238 [399] Sargsjan G. S t a " U, Unger C. Zschocke W and Ledig M 1991 Opt. Corrv" 86 480 (4001 French P M W, Noske D U, Rirvi N H, Williams J A R and Taylor J R 1991 Opf. Co~ninrm. 83 185 (4011 Wu Q, Zhou J Y, Huang X G, Li 2 X and Li Q X 1993 J . Opf. Soc. Am. B 10 2080 (4021 Rizvi N H, French P M W a n d Taylor J R 1992 Opr. Lett. 17 281 14031 Proctor B a n d Wise F 1992 Opt. Left. 17 1295 (4041 Lemoii B and Barty C 1992 Opt. Len. 17 1367 (4051 Curley P, Spielmann Ch, Brabec 7, Krausz F, Wintner E and Schmidt A J 1993 Opt. LPII.18 54 Garvey D, Zhou J, Kapfeyn H C and Mumane M M 1993 Opt. Lett. 18 I4061 Asaki M T,Huang, C-P, 977 [407]Spence D E, Kean P N and Sibbctt W 1991 Opf. Lert. 16 42 [408] Piche M 1991 Opt. Cornniun. 86 156 [409] Negus D K, Spinclli L, Goldblatt N and Feugnet G 1991 Digirnl Solid-Stote LDsers (Hilfon Head, SC 1991) Paper PDP4 (Washington. DC: Optical Society of America) pp 120-4 [410] Lariontsev E G and Serkin V N 1975 Sou, J. Quomitin Electron. 5 796 [41 I] Marconi M C, Martinez 0 E and Diodati F P 1987 Opt. Coimiiun. 63 211 (4121 Satin F, Squicr J and Piche M 1991 Opr. Lett. 16 1674 (4131 Huang D, Ulman M. Acioli L H, Haus H A and Fujimoto J G 1992 Opt. Lett. 17 51 I 1414) Martinez 0 E and Chilla J L A 1992 Opt, Le//. 17 1210 [415] BrabccT. Curley P, Spielmann Ch, Wintner E and khmidt A I 1993 J. Opt. Soc. Am. B 10 1029 [416] Malcolm G P A and Ferguson A I 1991 Opt. Lett. 1 6 1967 (4171 Gabetta G.Huang D, Jacobson J, Ramaswamy M, lppen E P and Fujimoto J G 1991 Opt. Lett. 16 1756 14151 Curley P F and Ferguson A I 1991 Opt. Lett. 16 1016 (4191 Li Kam Wa P, Chai B H T and Miller A 1991 Elcctr-on. Lett. 27 2351 14201 Kalka J D, Watts M L and Pieterse I-W J 1992 IEEE J . Q~ruiitrintElecwon. QE-28 2151 14211 Spielmann Ch, Krausz F, Brabec T, Wintiier E and Schmidt A J 1991 Opt, Left. 16 I180 (4221 Sarukura N a n d Ishida Y 1992 IEEE J . Quantum E/ectron. QE-28 2134 I4231 Keller U. 'tHoofl G W, b o x W H a n d Cunningham J E 1991 Opt. Lett. 16 I022 (4241 Rirvi N H, French P M W, Taylor J R, Delfyett P 1 and Flarez L T 1993 Opf. Leu. 18 983 (4251 Rirvi N H, French P M W and Taylor J R 1992 Opt. Lett. 17 279 (4261 Lai Ming 1994 Opf. Lett. 19 722 [427] Kraus F, Brabec T and Spielmann ch 1991 Opf. Lfff. 16 235 14281 Haus H A and lppen E P 1991 Opt. Lett. 16 1331

The generation o f ultrashort laser pulses

26 1

[429] Tamura K, Jacobson J, lppen E P, Haus H A and Fujimoto J G 1992 Opt. Lett. 17 220 [430] Krdusr F and Brabec T 1993 Opt. Lett. 18 888 [431J Cerullo 0 , De Silvestri S and Magni V 1994 Opt. Lett. 19 1040 [432] Kubota H, Kurokawa K and Nakazawa M 1988 Opt. Letl. 13 749 [433] De Silvestri de S, Laporta P and Svelto 0 1984 IEEE J . Quantum Electron. Q E 2 0 533 [434] Rudolph W and Wilhelmi B (ed) 1989 Light Pulse Compression (Char: Harwocdj [435] Yariv A 1985 Optical Electronics Orlando, FL: Holt, Rinehart and Winston) I4361 Dietal W, Dopel E, Hehl K, Rudolph W and Schmidt E 1984 Opt. Commun. 50 179 [437] De Silvestri S. Laporta P and Svelto 0 1984 Opt. Lett. 9 335 [438] French P M W, Gomes A S L, Gouveia-Neto A S and Taylor J R 1986 Opf. Coiiurtun. 60 389 [439] French P M Wand Taylor J R 1987,Opt. Commun. 61 224 [440] Yamahita M, lshikawa M, Torizuka:K and Sato T 1986 Opt. Lett. 11 504 [441] Yamahita M, Toriruka K and Sato T 1987 1EEE 1. Quantum Electrort. Q E 2 32005 [442] Gires F and Tournois P 1964 C. R. Acad. Sci. P u i s 258 61 I 2 14431 Heppner J and Kuhl J 1985 Appl. Phys. Lett. 47 453 [444] French P M W, Chen G F and Sibbett W 1986 Op. Commun. 57 263 [445] French P M W, Gomes A S L, Gouveia-Neto A S and Taylor J R 1986 Opt. Qiiuntum Electwn. 18 171 [446] Smith A W 1969 Appl. Phys. Lett. 15 194 [447] Treacy E B 1969 IEEE J . Quantum Eiectroit. Q E 5 454 [448] Martinez 0 E, Gordon J P and Fork R L 1984 J. Opt. Soc. Ant A I 1003 [449] Fork R L, Martinez 0 E and Gordon J P 1984 Opt. Lett. 9 150 14501 Kaka J D and Baer T 1987 Opt. Lett. 12 401 [451] Bor 2 S and Racz B 1985 Opt. Comniwz. 54, 165 14521 Kaka J D,Head D F and Baer T 1986 Proc. Liltrafmf Plienomenn V (Springer Series in Chemical Physics 46)(Berlin: Springer) pp 51-3 I4531 Martinez 0 E 1986 J . Opl. Soc. Am. B 3 929 [454] Martinez 0 E 1987 IEEE J . Quantum Electron. Q E 2 3 59 [455] Pessot M, Main P and Mourou G 1987 Opt. Commun. 6 2 419 I4561 Tomiinson W 1, Stolen R H and Shank C V 1984 J. Opt. Soc. Am. B 1 139 Eiectron. 20 95 I4571 Gomes A S L, Gouveia-Neta A S and Taylor J R 1988 Opl. Q~rantum I4581 Nakatsuka H, Gnschkowsky D and Balant A C 1981 Phys, Rev. Lett. 47 1910 [459] Shank C V, Fork R L, Yen R and Stolen R H 1982 Appl. Pltys. Lett. 40 761 [460] Fork R L, Brito-Cruz C H, Becker P C and Shank C V 1987 Opt. Lell. 12 483 14611 Tai K and Tomita A 1986 Appi. Phys. Lett. 48 1033 [462] Taylor J R? [463] Mollenauer L F and Stolen R H 1982 Lam Focus (April) 193-8 [464] Agrawal G P 1989 Nonlinear Fiber Optics (New York: Academic) [465] Hasagawa A 1990 Optical Solitons irt Fibres (Berlin: Springer) [466] Taylor J R (edj 1992 Optical Solitons-77reory and Experiment (Cambridge: Cambridge University

Press)
[467] Diels J-C, Fontaine J J. McMichael I C, Woilhelmi B, Dietel W and Kuhlke D 1983 Sou. J . Qtrantum Eiectron. I 3 1562 [468] Martinez 0 E, Gordon J P and Fork R L 1984 Opt. Lett. 9 156 [469] Haus H A and Silberberg Y 1986 IEEE J. @antunr Electron. Q E 2 Z 325 [470] Martinez 0 E, Gordon J P and Fork R L 1985 J , Opt. Soc. Am. B 2 753 [471] Valdmanis J A, Fork R L and Gordon J P 1985 Opt. Lett. 10 131 [472] Salin F, Grangier P, Roger G and Brun A 1986 P h p Reu. Left. 56 1132 [473] Wise F, Walmsley 1 A and Tang C L 1988 Opt. Lett. 13 129 14741 Nighan W L Jr, Gong T and Fauchet P M 1989 Opt. Lett. 12 447 [475] Salin F, Grangier P, Roger G and Brun A 1988 Phys. Reu. Lett. 60 569 [476] Avramopoulos H, French P M W, Williams J A R, New G H C and Taylor J R 1988 lEEE J. Quatttvm Electron. QE7A 1884 14771 Krausz F, F m a n n M E,Brabec T, Curley P F,Hofer M,Ober M H, Spielmann Ch, Wintner E and Schmidt A J 1992 IEEE J. Quantum Electron. Q E 2 8 2097 14781 Soielmann Ch. Curlev P F, Brabec T and Krausz F 1994 IEEE J. Ouontum Electron. OE-30 I100 i479j Brabec T, Spielmann-Ch A d Krausz F 1992 Opt. Lett. 17 748 [480] Curley P F, Spielmann, Ch,Brabec T,Wintner E and Krausz F 1993 J. Opt. Soc. ANI.B 10 1025 [481] Krausz F, Spielinarm Ch, Brabec T, Wintner E and Schmidt A J 1992 Opt. Leu. 17 204

262

P M W Freticli

(4821 Huang C P, Asaki M T, Backus S, Murnane M M. Kapteyn H C and Nathel H 1992 Opr. Let/. 17 (4831 Curley P F, Spcilmann Ch. Brabec T, Krausz F, Wintner E and Schmidt A J 1993 Opt. Lett. 18 (1993) 54 [484] Asaki M T,Huang C P,Garvey D, Zhou 1, Kapteyn H C and Murnane M M Opt Lett. 18 977 [485] Sting1 A, Spielmann Ch, Krausz and Szipocs R 1994 Opl.Lell. 19 204 [486] Gordon J P 1992 J. Opt, Soc. Ani. B 9 91 [487] Kelly S M J 1992 Electron. L e n 28 806 14881 Wai P K A, Menyuk C R, Lee Y C and Chen H H 1986 Opt. Lefl, 11 464 [489] Elgin J N 1992 Opt. Lelt. 17 1409 [490] Brabec T and Kelly S M J 1993 Opt. Lelr. IS 2002 (4911 F m a n n M E 1994 Appl. Plys. B 58 197 [492] Fermann M E, Andrejco M J, Silberberg Y and Stock M L 1993 Opt. Lefr. 18 894 (4931 Smith K. Armitage J R, Wyatt R. Doran N J and Kelly S M J 1990 Electron. Len. 26 1149 14941 Kelly S M J, Smith K, Blow K J and Doran N J 1991 Opt. Left. 16 1337 14951 Noske D U, Pandit N and Taylor J R 1992 Opf. Left. 17 1515 (4961 Dennis M L and Duling I R 111 1993 Appl. Phys. Lelf. 62 291 I (4971 Matsas V J, Loh W H and Richardson D J 1993 Phofon. Tech Lett, 5 492 [498] Nakazawa M, Yoshida E and Kimura Y 1993 Elecfron. Leu. 29 63 [499] Femann M E, Hofer M, Haberl F, Schmidt A J and Tun L 1991 Opt. Lett. 16 244 [500] Obcr M H. Hofer M, Fermann M E 1993 Opt. Lett 18 367 (5011 Tamura K, lppen E P,Haus H A and Nelson L E 1993 Opl. Letl. IS 1080 (5021 lppen E P and Shank C V 1977 Tecltnmiquesfor Mensurements Ultrarkorl Light Pulses ed S L Shapiro (Berlin: Springer) [503] Bradley D J and New G H C 1974 Proc. IEEE 62 313 [504] Valdmanis J A and Mourou G 1986 IEEE J . Quanrutn Eleclron. QE22 69 (5051 Zavoiski E K and Fanchenko S D 1956 Sou. Phys. Dokl. 1 285 [506] Willson J P, Sibben Wand SIeat W E 1982 Opt C o n 42 208 (5071 Maier M, Kaiser W, Giordimaine J A 1966 Phys. Rev. Lefr. 17 1275 [SOS] Weber H P 1967 J . Appr Phys. 38 2231 (5091 Mind1 T, HeAerle P, Schneider S and Dorr F 1983 App/. Phys, B 31 201 (5101 Naganuma K, Mogi ak and Yamada H 1989 IEEE J. Quantum Elecfron. QE25 1225 [51l] Diels J-C, Fontaine J J, McMichael I C and Simoni F 1985 Appl. Opl. 7 . 4 1270 (5121 Diels I-C, Fontaine J J and Rudolph W 1987 Rev. Pliys. Appl. 22 1605 [513] Chilla J L A and Martinez 0 E 1991 Opt. Lell. 16 39 (5141 Sala K L, Kenny-Wallace G A and Hall G E 1991 IEEE J . Qiruntum Ekcfrort. Q E l 6 990 15151 lppen E P and Shailk C V 1975 Appl. Plrys. Leu 27 488 [516] Jansky J. Conadi G and Gyuzalian R N 1977 Opf.Commun. 23 293 (5171 Salin F, Georges P and Brun A 1987 AppL Opr. 26 4528 (5181 Szabo G, Bor 2 and Muller A 1988 Opt. Lell. 13 746 (5191 Giodmaine J A, Rentzepis P M, Shapiro S L and Wecht K W 1967 Appl. Phyx Left. 11 216 [520] Le Blanc C P, Szabo G and Sauerbrey R 1991 Opt. Lett. 16 1508 (5211 Sarukura N, Watanabe M, Endoh A and Watanabe S 1988 Opl. Lefl. 13 996 (5221 Trebino R and Kane D J 1993 J . Opf. Soc. Am. A 10 I101 15231 Van Stryland E W 1979 Opt. Co. 31 93 (5241 Catherall J M and New G H C I986 IEEE J. Qua~tumE h r o n . QE22 1593