1. Introduction

Since seminal proposals that envisioned its advantages [1,2], nonlinear optical (NLO) microscopy has progressed firmly in the last decades [3–5], to the point where it is today an indispensable facility in microscopy services of large scientific institutions. Multiphoton excited fluorescence (MPEF) and second harmonic generation (SHG) are among the most relevant NLO microscopy techniques for biological investigations. Common advantages to MPEF and SHG techniques, in contrast to classic single-photon fluorescence microscopy, are deep tissue penetration, 3D-sectioning imaging, out of focus scatter-free imaging and small focal volume (thus high resolution in all three spatial axes). MPEF allows exogenous and endogenous fluorophore excitation and SHG enables observation of the spatial structure of non-centrosymmetric materials and/or interfaces (e.g., changes on the index of refraction) in unlabelled biological specimens [6]. Multimodal NLO microscopy usually combines MPEF and SHG for full exploitation of the advantages of both techniques. These advantages rely on the excitation of the samples by laser pulses that provide very high photon irradiances (typically > 10 ²⁷ photons s^-1 cm^-2), to increase the probability of the rare event of simultaneous absorption of more than one photon by the sample [7,8]. The best trade-off between high photon irradiance and harmless average power levels is offered by lasers delivering pulses with durations in the femtosecond range. Solid-state and optical fiber mode-locked lasers capable of pulse durations of around 100 fs are used regularly in NLO microscopy [3,4,9]. Also, different frequency conversion methods of mode-locked fiber lasers with resulting pulse durations as short as 60 fs have been proposed for NLO microscopy and nonlinear vision studies [10–13]. Less explored are few-cycle sources, which, by delivering pulses with durations in the range of 10 fs, increase the efficiency of the excited nonlinear effects by an order of magnitude, helping to reduce the average power on the samples and maximise their viability. Besides, and inherently to the relationship between time and frequency domains, the spectral composition of few-cycle pulses is extremely broad (bandwidths typically > 200 nm for sources operating in the near-IR), further enabling multispectral (simultaneous, if required) NLO microscopy [14,15]. Currently, commercially available few-cycle sources are provided only by a few bulk solid-state technologies, namely titanium-sapphire (Ti:Sa) oscillators and more recently optical parametric chirped pulsed amplification (OPCPA) systems. These sources are complex, bulky, water-cooled, of cumbersome use and require frequent service and maintenance. They have therefore a high cost of ownership, while their price is often inflated due to lack of competing technologies. This limitation in technology availability is what motivates our present work: the development of fully optical fiber-based few-cycle lasers as an alternative technology to bulk solid-state lasers, to foster a universal adoption of multimodal and multispectral NLO microscopy, making it cost effective, attractive and reliable for many scientific laboratories, industrial laboratories and hospitals facilities. Moreover, the availability of an all-fiber few-cycle source with emission in the 1 µm region is of special interest in biological microscopy, because it will help exploring the performance of nonlinear imaging using few-cycle illumination in a spectral region where there is a good compromise between the two main factors that limit the imaging depth in biological samples: the scattering and the absorption coefficients of the tissues [16–19]. The scattering coefficient decreases exponentially with wavelength [17,18], thus pulses at 1 µm are less affected by scattering than Ti:Sa 800 nm pulses. Pulses in the 1.5 and 2 µm regions [20,21] suffer even less scattering and have been used successfully in three-photon excitation fluorescence (3PEF) microscopy, showing deeper penetration, lower signal to background ratios and reduced out-of-focus background [18,22], compared to using two-photon excitation fluorescence (TPEF) microscopy in the 1 µm region. However, in the 1.5 and 2 µm regions the absorption coefficient in water is up to 2 and 3 orders of magnitude higher compared to the 1 µm band [17,18]. Besides, for a given fluorophore, the two-photon absorption coefficient is several orders of magnitude higher than the three-photon absorption coefficient, thus higher peak powers are required to generate a comparable signal [23]. Because a higher photon density is needed to generate three-photon images, it is more probable to cause damage to the sample, compared to two-photon images.

In this work, we present first a monolithic fiber-optic configuration for generating temporally coherent supercontinuum (SC) that provides transform-limited few-cycle pulses with durations as short as 13.0 fs at a central wavelength of 1060 nm. These are the shortest pulse durations provided to date, to the best of our knowledge, by an all-fiber source in the 1 µm spectral region. Being all-fiber, this type of source is more simple, robust, efficient and cost-effective than previously reported few-cycle temporally coherent SC sources in the 1 µm region [24–27]. Next, we report results on the performance of the few-cycle all-fiber source in multimodal and multispectral NLO microscopy: imaging of different biological specimens is obtained by simultaneous TPEF of multiple fluorophores whose single-photon peak absorption wavelengths are in the band from 480 to 580 nm; TPEF and SHG are combined to obtain images of neurons (via TPEF) and of muscle and pharynx (via SHG) of living C. elegans specimens; the penetration depth capability of the few-cycle source is assessed by TPEF imaging of transparent specimens (zebrafish embryos) and of scattering tissues (Wistar rat retina); an image of the full retina, of ∼ 170 µm depth, is obtained with cellular resolution, showing better imaging resolution and depth than the image of the same sample acquired with a commercial Ti:Sa laser at a central wavelength of 810 nm.

2. Few-cycle all-fiber temporally coherent supercontinuum

Ultrafast fiber lasers appear as an alternative to Ti:Sa oscillators and OPCPA systems, since they are compact, air-cooled, turn-key, cost-effective and maintenance and alignment-free [9,28]. However, the natural gain spectra of rare-earth elements (the active media in fiber lasers) limit their emission bandwidths and pulse durations to a few tens of nm and a few hundreds of femtoseconds, respectively. Nevertheless, in the last years, research on supercontinuum (SC) generation has demonstrated successful approaches to obtain very high-quality few-cycle pulses using ultrafast fiber lasers as exciting sources of particular nonlinear effects in photonic crystal fibers (PCFs): pumping at a wavelength within the flattened top of the convex dispersion curves of all-normal dispersion (ANDi) PCFs (whose dispersion lies completely in the normal dispersion region), spectral broadening appears due to the action of self-phase modulation (SPM) and optical wave breaking (OWB) [29]. In this way, highly coherent SC emission can be generated, preserving compressible pulses in the temporal domain. Different approaches to generate temporally coherent SC using ANDi PCFs have been proposed. In some configurations the seed stage (or pump stage) of the coherent SC source is built with free-space optics technology: e.g., Ti:Sa lasers, optical parametric oscillators (OPOs) or master oscillator power amplifiers (MOPAs) [24,25,30–34]. In other configurations the seed stage is a fiber laser but light coupling to the ANDi PCF is performed with free-space optics [26,35–37]. Chow et. al. proposed an all-fiber configuration based on ANDi PCFs for the 1.55 µm band [38], but this work lacked a demonstration of transform limited or near-to-transform limited compression of the generated SC, probably due to the long length of the ANDi fiber used in the experiment (64 m). To the best of our knowledge, none of the current state-of-the-art configurations that demonstrate generation of transform-limited or near-to-transform-limited few-cycle temporally coherent supercontinua using ANDi PCFs are an all-fiber configuration (said all-fiber configuration understood as a monolithic fiber-optic configuration where all its stages are fiber based and coupled to each other by a fiber splice or a fiber-based transition). All-fiber configurations to generate few-cycle pulses that do not rely on ANDi PCFs have been reported previously, but with central emission wavelength in the 1.5 µm spectral region, seeded by 1.5 µm gain-switched semiconductor lasers or by 1.5 µm erbium-doped mode-locked lasers. On the one hand, natural dispersion of optical fibers is anomalous in the 1.5 µm region, thus soliton generation and self-compression can be exploited to obtain pulses as short as 20 fs [39] and 14 fs (in a nearly all-fiber configuration [40]). On the other hand, amplifying Er-doped fibers can present normal dispersion in the 1.5 µm region, thus the chirped pulsed amplification (CPA) technique [41] can be exploited (using standard fibers of anomalous dispersion in the CPA compression stage) to obtain pulses as short as 29 fs [42] and 14 fs [43]. Differently, to obtain emission at central wavelength in the 1.0 µm region, amplification with Yb-doped fibers has to be used. Yb-doped fibers and standard passive fibers present both normal dispersion in the 1.0 µm region, so neither soliton generation and self-compression nor CPA with standard fibers can be exploited for the generation of few-cycle pulses with central wavelength at 1.0 µm. Instead, one has to turn to hollow core photonic bandgap (HC-PBG) fibers of anomalous dispersion for CPA compression and to ANDi PCFs for coherent spectral broadening. Emission in the 1.0 µm region can also be obtained by second harmonic generation from pulses in the 2.0 µm region delivered by Er-doped fiber-based sources, although free-space excitation of nonlinear crystals is required [13]. In this section, we describe an all-fiber configuration for generating temporally coherent supercontinuum that provides transform-limited few-cycle pulses with central wavelength at 1.06 µm and durations as short as 13.0 fs. Recently we have reported first evidence of the ability of this all-fiber configuration to deliver few-cycle transform-limited pulses [44]. Here we describe in detail the monolithic architecture of the system, the process of design, manufacture and optimization of ANDi PCFs according to the diagnosis of different emission regimes depending on the fiber geometry and we present results of few-cycle emission of optimised time-domain quality, which are confirmed by independent pulse duration measurement techniques: d-scan, and interferometric autocorrelation.

Temporally coherent spectral broadening in ANDi PCFs is achieved by SPM and OWB using short lengths of ANDi PCFs (few to tens of cm) pumped by input pulses of high peak power (few to tens of kW) and very short duration (few to hundreds of fs). As an example, Heidt et al. [29], showed that a spectral broadening of > 100 nm (with central wavelength at 1060 nm) can be obtained while maintaining perfect coherence by pumping 1 m of ANDi PCF with input pulses of 5 kW peak power and 200 fs duration. In our monolithic configuration, the ANDi PCF is a few tens of cm long and is excited by input pulses of > 15 kW peak power and < 250 fs duration. Ti:Sa lasers, OPOs or MOPAs, with laser generation architectures based on free-space configurations, deliver naturally pulses of this type. However, for an all-fiber laser architecture it is a challenge to offer pulses with these properties to be delivered to an ANDi PCF because the light is completely confined in the cores of the optical fibers, whose core diameters are typically below 10 µm. A peak power of 15 kW inside guiding fibers of 10 µm core diameter yields intensities of > 15 GW/cm² which, for propagation lengths of a few cm, are above the threshold for many undesired nonlinear effects that distort the laser pulses propagating inside the fibers and, particularly, destroy the temporal coherence of the laser pulses. To overcome this problem, we use a chirped pulse amplification (CPA) system [41], with an all-fiber configuration (see Fig. 1), where the temporal compression stage is built with a hollow core photonic bandgap fiber (HC PBGF). Maintaining the temporal coherence throughout its stages, such configuration delivers pulses of > 15 kW peak power and < 250 fs duration, to be used as exciting pulses of the ANDi PCF. The end of the HC PBGF is fusion spliced to the ANDi PCF. To maintain the integrity of the structure of both fibers, a weak splice was performed. We carried out an optimization process to find the arc discharge parameters that maintained the fibers spliced for the best achievable coupling efficiency, which was of 0.4. The current and time of the arc discharge were of ∼ 9 mA and ∼ 0.5 s, respectively (∼ 6 mA and ∼ 1.5 s lower than that of a conventional fusion splice between standard 125 µm single-mode fibers, SM-SM Basic mode of Fujikura FSM 100 P splicer).

 

Fig. 1. (a) All-fiber configuration of a temporally coherent supercontinuum source of few-cycle pulses. GDD: group delay dispersion; GVD: group velocity dispersion. (b) Qualitative representation of the spectral and temporal properties of the pulsed optical signal as it evolves throughout the stages of the all-fiber source (stages 1 to 5) and at the output of the temporal compressor (stage 6) used to compress the pulse down to its Fourier limited duration. Δλ_i and Δτ_i: spectral bandwidth and temporal duration of the pulses, respectively, at the output of i-th stage. (c) Properties of the pulsed signal at the output of each stage, for an example of implementation of the all-fiber configuration where the active fiber is Yb-doped, thus with laser emission in the 1 µm band. MFD: mode field diameter, λ_c: central wavelength, PRR: pulse repetition rate, Δλ_FWHM: spectral bandwidth at full width at half maximum, Δτ_FWHM: pulse duration at full width at half maximum, P_avg: average power, P_p: pulse peak power, I_p: pulse peak intensity. Values are given for a pump laser diode wavelength and average power of 976 nm and 3.75 W, respectively.

Download Full Size | PDF

Figure 1 illustrates our approach to obtaining an all-fiber configuration of femtosecond temporally coherent supercontinuum sources, based on a sequential structure of fiber-optic stages, each of them connected to the next by a fused fiber splice. The table in the bottom of Fig. 1 shows data of an example of implementation of this configuration, including measured values of the spectral and temporal domain properties of the pulsed signal at the output of each stage.

The first stage (stage 1) corresponds to the seed of the system, a passively mode-locked all-fiber oscillator that delivers transform limited pulses of hundreds of femtoseconds duration and MHz range pulse repetition rate (PRR) with a central emission wavelength λ_c. PRR and λ_c remain unchanged throughout all stages of the system. In the implementation example of Fig. 1(c) the oscillator is an Yb-doped fiber linear cavity with a semiconductor saturable absorber mirror (SESAM) in one end of the cavity and a chirped fiber Bragg grating (CFBG) at the other end, acting as output coupler. The net group delay dispersion in the cavity is anomalous, tailored to obtain solitonic emission regime. Stage 2 is composed of a polarization maintaining (PM) single mode optical fiber and a PM fused fiber combiner. The fibers of this stage have a normal group velocity dispersion (GVD > 0). The function of stage 2 is to stretch the pulses temporally so they can be amplified in the next stage without generating nonlinear effects (the peak power remains below the threshold of generation of nonlinear effects in the optical fiber core). Also, it pre-compensates the anomalous dispersion that the pulses will experience in stage 4. The fused fiber combiner launches light from a laser diode into the active fiber of the next stage. Stage 3 is composed of a double clad PM rare-earth doped active fiber (an Yb-doped fiber in the implementation example of Fig. 1(c)), with normal group velocity dispersion (GVD > 0). The function of stage 3 is to amplify the pulses to the maximum possible peak power without biasing nonlinear effects that would distort the temporal and spectral shape of the pulses. The amplification is produced progressively in the fiber by stimulated radiation in the active rare earth ions of the fiber core, which are pumped to excited states by the light coming from the laser diode of the previous stage through the first clad of the fiber (this clad is passive, i.e., undoped). Stage 4 is composed of a hollow core photonic bandgap (HC PBG) microstructured fiber (model HC-1060 from Thorlabs in the implementation example of Fig. 1(c)) with anomalous group velocity dispersion (GVD < 0). The function of stage 4 is to compress the pulses to achieve the peak power required at the input of the ANDi PCF to efficiently generate temporally coherent spectral broadening by SPM and OWB. Also, its length is chosen so that the net group delay dispersion (GDD) suffered by the pulses from the oscillator output is slightly anomalous (- 0.015 ps² in the example of implementation of Fig. 1). This way, the pulse still undergoes compression in the first segment of the ANDi PCF and SPM efficiency is optimised by having the maximum peak power achievable inside the ANDi PCF. Since the material of the HC PBG fiber core is air, nonlinear effects due to high peak power are avoided, hence the pulse does not experience additional spectral broadening. Stage 5 is composed of a highly nonlinear ANDi PCF a few tens of cm long. GVD in this fiber is normal within a very broad spectral bandwidth (broader than 300 nm, centred at 1060 nm, in the implementation example of Fig. 1), with a quasi-flat and symmetric shape (see Fig. 2(d2)). The function of this stage is to spectrally broaden the spectrum of the pulsed signal by SPM under near-zero-dispersion conditions, which preserves the temporal coherence of the pulses so that they remain compressible down to pulse durations corresponding to the Fourier limit of their spectrum. At the end of stage 5, the pulsed signal is delivered to free-space and collimated. The output of the fiber is a home-made endcap, made by collapsing the holes of the ANDi PCF with an arc discharge of the fusion splicer after cleaving the end of the fiber in angle. The output beam from the endcap is collimated with an achromat lens from Schäfter & Kirchhoff, model 60FC-4-M12-08. Stage 6 is composed of a free-space temporal compressor of fixed anomalous dispersion (GDD < 0) and variable normal dispersion (GDD > 0). It compresses the temporally coherent pulses from the output of stage 5 down to close to their Fourier limited durations (12.2 fs in the example of implementation of Fig. 1). The variable compression is introduced by a pair of glass wedges placed in the path of the optical signal. Dispersion is varied by changing their relative position (insertion length), therefore changing the amount of glass material effectively traversed by the optical signal. Note that stage 6 is not part of the all-fiber configuration of the system. By few-cycle all-fiber supercontinuum, we refer to an all-fiber configuration that generates temporally coherent pulses compressible to transform limited few-cycle duration, which does not imply that few-cycle duration is necessarily happening at the very output of the all-fiber configuration. Indeed, it is not especially interesting to get this duration at the output of the fiber architecture, since such short pulses will always require dispersion pre-compensation when being applied (either with free-space or with fiber compression means).

 

Fig. 2. Summary of properties of three representative microstructured fibers manufactured in this work, numbered 1, 2, and 3. (a) SEM image of the cross section of an ANDi PCF (fiber 2), with parameters d = 0.596 µm and Λ = 1.615 µm (detail of fiber core region in Fig. 2(c2)). (b) Positions of fibers 1, 2, and 3, according to their corresponding experimental values of d and Λ, on the map of calculated values of d and Λ that provide acceptable dispersion curves for temporally coherent spectral broadening. (c1)-(c3) Detailed SEM images of the core region of fibers 1, 2, and 3. (d1)-(d3) Calculated dispersion curves for fibers 1 to 3, respectively. (e1)-(e3) Optical spectra at the output of the ANDi PCF (stage 5 in Fig. 1) obtained with fibers 1 to 3, respectively, for same fiber length (20 cm) and pump diode (stage 2 of Fig. 1) average output powers of 1.3 W (black line), 3.3 W (red line), and 3.75 W (blue line).

Download Full Size | PDF

2.1 All-normal dispersion photonic crystal fiber (ANDi-PCF): design and fabrication

The ANDi fiber of section 5 was manufactured with F300 silica and exhibits a standard PCF solid core design: the solid-core results from a missing hole that is surrounded by N rings of holes running along the fiber longitudinal axis. Holes are displayed in a periodic manner with a triangular lattice. The air holes have a diameter d. The distance between hole centres, called pitch, is Λ. The microstructure of air holes is surrounded by a jacket of uniform silica that confers the fiber a typical diameter of 125 µm. Efficient confinement of the light inside the core is obtained for a value of N equal to 7. The required properties of the GVD curve are calculated using the semi-empirical model proposed by Saitoh et al. [45]. Considering a central wavelength of 1060 nm, a set of potentially valid dispersion curves is obtained combining the values of Λ and d within the following ranges: Λ = [0.50 ; 0.64] µm ; d = [1.50 ; 1.70] µm. ANDi PCFs with optimum values of Λ and d have been manufactured through an optimization iterative process (see subsection Design of ANDi PCF). Figure 2 summarizes the results of the optimization process of the ANDi PCF manufacture, from design to performance on temporally coherent spectral broadening. Figure 2(a)) shows a scanning electron microscope (SEM) image of the cross section of the manufactured ANDi PCF that presents the largest temporally coherent spectral broadening (fiber 2). Figure 2(b)) shows a map of calculated values of Λ and d that provide adequate dispersion curves for temporally coherent spectral broadening. The red region denotes the condition of the dispersion maximum belonging to the exciting laser bandwidth (1060 ± 15 nm). The blue region extends the condition to a broader bandwidth (1060 ± 30 nm). Subsection Design of ANDi PCF offers an extended explanation of the elaboration of this map. Figures 2(c1)-(c3)) are detailed SEM images of the core region of representative ANDi PCFs manufactured during the optimization process, with different values Λ and d (measured from the corresponding SEM images) and numbered from 1 to 3. Figures 2(d1)-(d3)) show the calculated dispersion curves for fibers 1 to 3, respectively. Finally, Figs. 2(e1)-(e3)) show the spectra at the output of the ANDi PCF (stage 5 in Fig. 1) obtained with fibers 1 to 3, respectively, for the same exciting conditions (output signal of stage 4 in Fig. 1) and same length (20 cm), at 3 different driving currents of the laser diode that pumps the amplifying stage (stage 3 in Fig. 1). Fiber 1 presents a dispersion curve out of the target of ANDi condition, being the dispersion parameter D ∈ [0; + 5 ps/nm km] within the exciting laser bandwidth. Consequently, an asymmetric spectral broadening is observed, governed by temporally incoherent dynamics of first stages of SC construction in an anomalous GVD regime: modulation instability, soliton fission and Raman self-frequency shift [46,47]. Contrarily, fiber 2 presents an ANDi curve, being the dispersion parameter D ∈ [-5; -25 ps/nm km] within a full bandwidth of 300 nm, centred at a maximum dispersion wavelength of 1060 nm. SPM governs a symmetric temporally coherent broadening. The near-zero dispersion of the ANDi fiber limits pulse temporal stretching, thus driving high SPM efficiency. The evolution of spectral broadening in fiber 2 is shown in Fig. 2(e2)). The average pump power of 3.75 W is slightly above the average pump power limit below which spectral broadening is due to SPM only. The shoulder peak at 930 nm in the short-wavelength edge of the spectrum indicates the beginning of the appearance of OWB effects, which are to be avoided to maintain “pure” temporally coherent spectral broadening by SPM [29]. Fiber 3 presents an ANDi curve as well, being D ∈ [-25; -40 ps/nm km] within a full bandwidth of 300 nm, centred at a maximum dispersion wavelength of 1045 nm. These values are more distant from zero dispersion than those of fiber 2. Consequently, SPM still broadens the spectrum while preserving temporal coherence, but with less efficiency (compared to fiber 2), because the pulse suffers larger temporal stretching.

2.1.1 Design of ANDi PCF

Design of PCFs for this SC source is governed by the trade-off between technological limits and design requirements. The SC shall be temporally coherent and be seeded by a femtosecond Yb-doped fiber laser. Thus, we need to use PCFs having all-normal dispersion (ANDi) within the laser line as well as within all the spectral range where SC is expected to be generated. As the nominal value of the central wavelength is 1060 nm, the natural choice of the material is silica. Therefore, as explained above, we have chosen a geometry which consists of a triangular array of the air holes (of diameter d and pitch Λ) in the silica glass matrix, where the core is formed by omitting one hole in the centre of the structure. The main issue in the PCF drawing process is controlling the shape and homogeneity of the diameters of the air holes. Not all technologies allow fabricating PCFs with air holes of very small size (diameter) and, more importantly, guarantee the stability of the structure along hundreds of meters and the reproducibility of such designs. Therefore, the search of the designs interesting for us are implemented using the hole diameter as the independent parameter. The minimum diameter of the holes is chosen to be d_min = 500 nm. The maximum diameter is d_max = 4000 nm. The distance between the holes Λ is defined within the range d / Λ = 0.2 - 0.8. This interval includes the values which guarantee the PCFs to be single mode (d / Λ ≤ 0.43) [48,49]. The PCFs having d / Λ > 0.43 could be either single mode or multimode. The spectral bandwidth of interest is 400 nm at a central wavelength λ₀ = 1060 nm. In order to have relatively symmetric spectral broadening it is desirable to have a dispersion profile D(λ) as symmetric as possible relative to the excitation wavelength λ₀ [50]. Another requirement is to have a dispersion as small as possible in absolute value to guarantee that self-phase modulation will be the dominating spectral broadening mechanism.

To calculate the dispersion curves, we used the semi-empirical model proposed by Saitoh et. al. [45]. Within this model, the theory of single-mode optical fibers is applied to describe the dispersion properties of index-guided PCFs. Being semi-empirical, this model allows iterating over a large volume of parameters to filter out designs fitted to technical requirements. Using this model, we scanned the (d, Λ) - space within the limits mentioned above and applying the condition that the dispersion maximum occurs within the laser bandwidth, D_max ∈ Δλ, or to a slightly more extended range, D_max ∈ Δλ ± 15 nm. The resulting parameters’ map is shown in Fig. 3. The red area corresponds to the stricter condition D_max ∈ Δλ. The variety of enumerated points corresponds to different pairs (d, Λ) whose dispersion curves are shown in Fig. 4. The yellow cross on the red area corresponds to the target design.

 

Fig. 3. Acceptable values of geometrical parameters. Δλ is the exciting laser bandwidth, 30 nm, which is the maximum bandwidth obtained in experimental measurements.

Download Full Size | PDF

 

Fig. 4. Corresponding dispersion curves of acceptable ANDi PCFs, according to the map of Fig. 3.

Download Full Size | PDF

2.1.2 Fabrication method

The fiber has been fabricated using the standard stack and draw method, where 168 capillaries of 1 mm outer diameter are stacked together around a solid rod of 1 mm diameter. The stack has been introduced and chocked in an outer tube that will be part of the outer cladding of the final fiber. This preform, called primary preform, has been drawn into several 1 m long microstructured capillaries (or canes) by applying vacuum into interstitial holes while keeping the 168 inner holes of capillaries at atmospheric pressure. One of these microstructured canes is then introduced into a jacket tube to reach the expected outer diameter, pitch and hole diameters ranges during the final drawing. Finally, the fine control of hole diameter and pitch is obtained by controlling a quadruple of parameters which are the furnace temperature, the preform and drawing speeds, and the pressure applied to the holes. Despite the high quality of the control devices and of the regulation of the electronic systems of the drawing tower, stabilising and controlling the diameter of the small air holes (500-600 nm) with an expected accuracy of 50 nm is challenging, especially achieving repeatability from a drawing to another. Indeed, many factors influence the repeatability and the fine control of the air hole diameter, among which, for example, micro-geometrical fluctuations along one tube or between external tubes (second drawing step) or residual material stress and geometrical fluctuations between microstructure capillaries (first drawing step). In order to obtain fiber samples with stabilized and controlled dimensions, we have performed live tests during the drawing. These consists in (i) periodically collecting 1 m long fiber samples (every 500 m with fixed drawing parameters) and (ii) estimating the chromatic dispersion curve by launching kW-range peak power pulses in the samples in order to observe the signature of nonlinear mechanisms. Indeed, an anomalous dispersion will lead to asymmetric spectral broadening governed by modulation instability, soliton fission and Raman self-frequency shift. This occurs when hole diameters are too large (e.g., larger than 630 nm, for a pitch of 1650 nm). On the other hand, when the chromatic dispersion is all-normal but far from zero and exhibits fast variations of group velocity over the spectral band of interest (e.g., holes diameter smaller than 520 nm, for a pitch of 1580 nm), the shape of the spectrum is triangular and broadening efficiency is low. Furthermore, despite the 7-layer structure chosen to minimise confinement losses, for such small d / Λ ≤ 0.33, the fundamental mode experiences larger confinement/bending losses that compromise the achievement of good results. The pressure applied in the preform was consequently tuned in order to reach the moment when we observed the development of a symmetric efficient broadening that is the signature of the expected single contribution of SPM (e.g., for holes diameter of 596 nm and pitch of 1615 nm). The inertia of the process requires to waiting for several minutes (i.e., several hundred meters of drawn fiber) in order to obtain a sample with stabilized geometrical parameters although the applied changes in pressure are as small as 0.1 kPa for a total applied pressure from 12 to 15 kPa depending on the drawn preform and samples.

2.2 Pulse compression and optimization

To confirm that our pulses at the output of the ANDi PCF are temporally coherent (thus compressible to close to their Fourier transform limit), we use the free-space temporal compressor of stage 6 (Fig. 1) and compare results of two different methods of pulse characterization in the time domain: d-scan technique and interferometric autocorrelation. The d-scan technique is based on performing a dispersion scan to the pulses with a pulse compressor while measuring the spectrum of the resulting second-harmonic (SH) signal. From this measurement, the phase and the amplitude properties of the electric field of the pulses are retrieved [51]. Figures 5(a1)-(a4)) show the properties of the pulse measured with the d-scan technique, for the case of ANDi fiber 2 and a pump diode average output power of 3.75 W. The compressor is designed to provide a range of net GDD to the pulse from - 1000 fs² (0 mm insertion length, Figs. 5(a1),(a2)) to + 1500 fs² (17 mm insertion length, Figs. 5(a1),(a2)). Details on the design of this compressor and corresponding application of the d-scan technique have been reported previously by us [52]. The compressor is adjusted to achieve the best quality pulse (shortest FWHM pulse width and highest peak power ratio between main peak and side-lobes). In these conditions, the pulses present very low GDD, third order dispersion (TOD) and fourth order dispersion (FOD) values (+ 276 fs², - 5162 fs³ and - 1621 fs⁴, respectively) with 56 ± 4% of the energy in the main peak and with peak power ratio between the main peak and the side-lobes greater than 8. The measured pulse width at FWHM of the compressed pulse is 13.0 fs (3.7 optical cycles). This result demonstrates the high degree of temporal coherence of the pulsed signal, since it is very close to the Fourier limited duration supported by its optical spectrum (12.2 fs). The fact that the pulses are properly compressed by a compressor of negligible TOD proves that the pulses suffer very low TOD (comparable to that of the compressor) while being spectrally broadened at the ANDi fiber. This effect results from the flatness of the dispersion curve of the ANDi fiber (Fig. 2(d2)). Average and peak power of the beam at the output of the compressor are P_avg = 160 mW and P_p =169.2 kW, respectively. Focused on a sample area of e.g., 10 µm², the photon irradiance of the beam during pulse propagation on the sample would be 8.5 × 10 ³⁰ photons s ^-1 cm ^-2 (neglecting losses in the optics), a figure well above typical multiphoton excitation thresholds. Figure 5(b)) shows an autocorrelation trace of the pulses obtained at the output of stage 6 with an interferometric autocorrelator. This trace corresponds to the pulsed signal compressed to its minimum duration by the same variable dispersion compressor employed for the d-scan technique. From this trace, an estimation of the FWHM of the pulse intensity profile of 12.6 fs is obtained. Despite not providing unambiguously the real form of the pulse intensity profile [53,54], the autocorrelation method is widely known and trusted by microscopists. Hence, the fact that it produces, independently, an estimated result of the pulse width comparable to the accurate result of the d-scan method, is relevant for a straightforward use of the source in current NLO microscopy setups. To the best of our knowledge, with the measurement of a temporal pulse width of 13.0 fs (3.7 optical cycles) we report the shortest pulses obtained to date from an all-fiber source in the 1 µm spectral region, being > 3 times shorter than the shortest pulses delivered by previously reported all-fiber sources in this region, which, limited mainly by the gain spectrum of ytterbium and by fiber dispersion and nonlinearity management constraints, do not support durations below 42 fs [55–58].

 

Fig. 5. Independent measurements of the properties of the pulsed signal after the stage of free-space compression using two different techniques: SHG d-scan technique and interferometric autocorrelation. (a1)-(a5) SHG d-scan retrieval results and resulting pulse in the spectral and time domain. (a1) Measured, calibrated d-scan trace. (a2) Retrieved d-scan trace. (a3) Red line: measured linear spectrum. Blue line: retrieved spectral phase. Orange line: 4th order polynomial fit of the spectral phase. (a4) Dashed yellow line: temporal intensity profile of the transform limited pulse. Light green: temporal intensity profile of the measured pulse. (a5) Characterizing parameters of the pulse, obtained from the d-scan retrieval. (b) Interferometric autocorrelation trace of the pulse, compressed to its minimum achievable duration. A pulse width of 12.6 fs is estimated assuming a deconvolution factor 0.71 for a Gaussian shaped pulse.

Download Full Size | PDF

2.2.1 Instrumentation

Fiber splices in the example of implementation shown in Fig. 1 were performed with a Fujikura FSM 100 P fiber fusion splicer. Microscope images of PCFs sections shown in Fig. 2 were obtained with a Zeiss Ultra 55 scanning electron microscope. Image postprocessing and statistical analysis followed, to obtain d and Λ parameters for each sample shown in Figs. 2(c1)-(c3)). Spectra of Fig. 2 were taken with 50 pm spectral resolution using a Yokogawa AQ6373B optical spectrum analyser. D-scan measurements in Fig. 5 were taken with a Sphere Ultrafast Photonics d-micro system, which includes an optimised compressor for microscopy. Autocorrelation traces in Fig. 5 and 6 were obtained with 1 fs resolution using a Femtochrome Research FR-103TPM interferometric autocorrelator.

 

Fig. 6. (a) Setup diagram. The excitation laser (red) is focused with a microscope objective on the sample. The TPEF signal (green) is collected in the backward direction. In the forward direction, the excitation laser and the SHG signal (blue) are collected with a microscope objective. A dichroic mirror is used to separate the laser light from the nonlinear signals. The TPEF signal is filtered from the SHG with a narrow bandpass filter (filter 2). The TPEF and SHG signal are detected using PMT detectors: the transmitted light from the excitation laser is detected with a photodiode. (b1)-(b3) Mean intensity of the generated TPEF signal as function of insertion length of the variable compressor glass wedges, for different objectives and sample specimens (Convallaria and pollen). (c) Interferometric autocorrelation trace of the pulse, measured at the sample plane, compressed to its minimum achievable duration, for the case of XL Plan N 25x objective. An estimated duration of 15.6 fs is calculated from the autocorrelation trace (assuming a deconvolution factor of 0.71 for a Gaussian shaped pulse), which is only slightly longer than the duration at the output of the source (12.6 fs, estimated with the same Gaussian-shaped approximation, in Fig. 5(b)). This difference is mostly due to residual uncompensated higher order dispersion (TOD and above) of the optics in the path of the beam, and not due to loss of spectral components.

Download Full Size | PDF

3. Broadband multispectral and multimodal nonlinear imaging

The output from stage 6 was coupled to an adapted inverted confocal microscope (Eclipse TE2000-U, Nikon) (Fig. 6(a)) modified for nonlinear imaging experiments. The variable dispersion compressor described above was used to pre-compensate the dispersion of the optical elements in the path of the microscope towards the sample, where the pulse duration was maintained below an estimated duration of 16 fs (Fig. 6(c)). The coupling system of the external illumination source included two galvanometric mirrors (Cambridge Technology, UK) and a telescope arrangement. We used a dichroic mirror (FF825-SDi01, Semrock) for sending the pulses to the illumination objective. The generated fluorescent light captured using this objective was collimated and sent through the same dichroic mirror, in a non-descanned configuration, for detection using a photomultiplier tube (PMT) detector (H9305-04, Hamamatsu). Three different microscope objectives, with different refractive index immersion media, numerical apertures (NA) and magnifications were used (see table with properties in Fig. 6). Dispersion pre-compensation was performed for every objective. The few-cycle all-fiber source described above was used as the illumination source for all image acquisitions, except for images in Figs. 7(F),(H) that were generated with a pulsed Ti:Sa laser (MIRA 900-F, Coherent, 200 fs nominal pulse width) operating at a central wavelength of 810 nm.

 

Fig. 7. (A)-(D) TPEF images of the tail of a 2-days-old transgenic line zebrafish embryo (Caax-GFP) expressing GFP in all cell membranes. (A)-(C) Intensity-normalized images corresponding to 26, 71, 150 µm depth. Scalebar: 40 µm. (D) Lateral re-slice of a Z-stack composed of 300 images (0.71 µm step spacing). Scalebar: 20 µm. (E)-(H) Comparison of TPEF imaging performance between our few-cycle all-fiber temporally coherent supercontinuum source (E),(G) and a Coherent MIRA 900-F laser as illuminating sources (F),(H), for TPEF imaging of an excised rat retina (retinal ganglion cells side up) stained with Alexa Fluor 647-phalloidin and Alexa Fluor 405-phalloidin, respectively. (E) Lateral re-slice of a Z-stack composed of 376 images (0.52 µm step spacing) acquired with the few-cycle all-fiber source. (F) Lateral re-slice of a Z-stack composed of 404 images (0.50 µm step spacing) acquired with the Coherent MIRA 900-F laser. Scalebar: 15 µm. CG: Ganglion cells; IPL: Inner plexiform layer; INL: Inner nuclear layer; OPL: Outer plexiform layer; ONL: Outer nuclear layer; OS: Outer segment photoreceptors.

Download Full Size | PDF

Possible leakage of fundamental laser light was blocked with a 350 - 700 nm bandpass BG40 filter (FGB37-A, Thorlabs, filter 1 in Fig. 6(a)). The two-photon excited fluorescence (TPEF) signal was filtered with fluorescence filter cubes (DAPI, FITC and TRITC: Standard series, Nikon) and collected in the backward direction. We used a 25x NA1.10 water immersion objective (Apo LWD, Nikon) for the second harmonic generation (SHG) signal collection in the forward direction. A dichroic mirror (FF665-Di02, Semrock) was used to separate the laser light from the nonlinear signals. The SHG signal was filtered with a bandpass filter (FF01-542/27-25, Semrock, filter 2 in Fig. 6(a)).

The dispersion pre-compensation performance of the variable compressor was evaluated by varying the glass wedges insertion to adjust the duration of the pulse at the sample plane. For each image, we measured the mean intensity of the generated TPEF signal as the function of glass insertion. The results can be seen in Figs. 6(b1)-(b3)). The induced dispersion per glass insertion length was + 147 fs² / mm. We see that the pre-compensation system can efficiently be used to maximise the fluorescent signal from the sample. A maximum TPEF intensity depending on the insertion of the glass was observed. This indicates that the system is capable of pre-compensating the dispersion introduced by the different microscope objectives. The discrepancies found in the optimum insertion length when imaging different samples can be attributed to changes in the refractive index of the samples.

Using the 25x objective under the optimised GVD settings of the compressor, we have successfully imaged several samples. Importantly, good fluorescence signal and depth were achieved using an average power of the pulsed beam of ∼ 4 mW, measured at the sample plane (the average power of the beam at the output of the variable compressor was 160 mW; it was attenuated by a variable neutral density filter and by intrinsic losses of the optical components through the microscope optical path). The maximum penetration achieved corresponds to 220 µm (Figs. 7(A)-(D)) in depth of the tail of a transgenic line zebrafish embryo (Caax-GFP) expressing GFP in all cell membranes. Zebrafish embryos are transparent, so they allow imaging at these large penetration depths. To test the penetration capabilities of the laser within a scattering tissue two excised retinas were stained with the same cytoskeleton marker (phalloidin), each conjugated with a different fluorescent dye: AlexaFluor 405 and 647, to be excited with a Ti:Sa Coherent MIRA 900-F laser (Δλ_FWHM = 10 nm, λ_c = 810 nm) and with our few-cycle all-fiber source (Δλ_FWHM = 150 nm, λ_c = 1060 nm), respectively. We proceeded to record the full retina (∼ 170 µm) with cellular resolution, acquiring z-stack images (Figs. 7(G),(H)). For both lasers, we used the same laser power at the sample plane and similar step spacing for constructing the z-stacks. Then lateral re-slices of the z-stack images were performed. Figures 7(E),(F) show the comparison of the lateral re-slice TPEF images acquired with our few-cycle all-fiber source and with a Coherent MIRA 900-F laser. In the image acquired with our few-cycle all-fiber source, all synaptic (bright regions) and nuclear (gap regions) layers that characterize the tissue can be distinguished (Fig. 7(E)). However, in the images acquired with the Coherent MIRA 900-F laser it was only possible to distinguish four layers (Fig. 7(F)). Larger imaging depth was achieved using the few-cycle all-fiber source, with which properly resolved images of the retina deeper layers ONL and OS (Figs. 7(E),(G)) were obtained. It is interesting to mention that the rat retina is highly autofluorescent to light in the blue-green region of the spectrum. In addition, the external segment of the photoreceptor cells where opsins (photopigments) are packaged, is highly absorbing to visible light. Therefore, illumination sources in the IR spectrum combined with red fluorescent dyes are ideal for depth imaging to prevent the autofluorescence generation/distortions in this tissue.

To test the capability of our few-cycle all-fiber source to nonlinearly excite multiple markers we proceeded to acquire TPEF images of multiple fluorophores: GFP, SYTOX Green, Alexa Fluor 568, tagRFP and Alexa Fluor 647. We also acquired SHG images of unlabelled tissues. Care was taken to use the corresponding filters for acquiring the TPEF signals. Figure 8(A) shows an image of a mouse intestine section stained with SYTOX Green (FITC filter) labelling the nuclei shown in yellow, and Alexa Fluor 568 phalloidin (TRITC filter) labelling the actin filaments shown in blue. Figure 8(B) shows the rhizome of Convallaria majalis stained with Fast Green and Safranin. Chloroplasts are shown in green (FITC filter), and cell walls are shown in red (TRITC filter). In Fig. 8(C), it is also possible to see the autofluorescence of a pollen grain. The large emitted autofluorescence spectrum was detected using two different fluorescence filters, FITC filter shown in green and TRITC filter shown in red. We have also been able to image in vivo samples with cellular resolution. In particular, paralysed C.elegans specimens. In Fig. 8(D), we can see a C. elegans OH15500 strain expressing tagRFP in all neurons. The SHG signal revealed all the different structures of the pharynx of the animal: corpus, isthmus and posterior bulb. Moreover, we can see the muscle fibers and their contraction during swallowing over time. These structures are a high-priced reference while imaging the neurons with TPEF. By using different emission filters, we have been able to split the fluorescence from the different fluorophores to visualize multiple structures with a single illumination source.

 

Fig. 8. (A) MAX projection of TPEF images of a mouse intestine section. Nuclei (yellow) and actin filaments (blue) are shown. Scalebar: 10µm. (B) MAX projection of a Z-stack of TPEF images of a rhizome of Convallaria majalis. Chloroplasts (green) and cell walls (red) are shown. Scalebar: 5µm. (C) MAX projection of a Z-stack corresponding to 37 images (1.95 µm step spacing) of autofluorescence TPEF images of a pollen grain. FITC (green) and TRITC (red) filtered signals. Yellow corresponds to the overlap of autofluorescence signal filtered with the FITC and TRITC filters. Scalebar: 10µm. Selected frames from the same Z-stack. (D) SUM projection of a Z-stack corresponding to 56 images (0.65 µm step spacing) of SHG signal of the muscle and pharynx (grey) and TPEF from the neurons (red) of a living C. elegans (OH15500 strain). Scalebar: 25 µm.

Download Full Size | PDF

4. Conclusion

We have presented a monolithic fiber-optic source of transform-limited few-cycle pulses. The simplicity, robustness and cost-effectiveness of its configuration are powerful factors in favour of this technology to replace traditional solid-state sources of few-cycle pulses in various applications. The successful performance of our source to provide high quality TPEF and SHG microscopy images anticipates its utility in three-photon excited fluorescence (3PEF) and third harmonic generation (THG) microscopy, as well as in other NLO microscopy techniques, such as coherent anti-Stokes Raman scattering (CARS) [59], multiphoton fluorescence lifetime imaging microscopy (MP-FLIM) [60,61] and, consequently, in multimodal and multispectral combinations between all mentioned techniques. The central wavelength of ∼ 1 µm is also favourable for microscopy compared to the ∼ 800 nm central wavelengths of Ti:Sa lasers, as it results in larger penetration depth in biological tissues and enables performing third-harmonic microscopy without going into the deep-UV, as would be the case of a central wavelength of ∼ 800 nm. Utility of this type of source is anticipated also in applications of other disciplines that rely on the inherent properties of few-cycle pulses, such as ultrafast spectroscopy, optical frequency comb generation and frequency metrology [62–64]. We have presented pulse durations as short as 13.0 fs (3.7 cycles), with a central wavelength of 1060 nm, but this is a non-fundamental limit that we expect to overcome with enhanced fiber manufacture precision to obtain flatter and nearer to zero dispersion curves of the ANDi PCFs. Based on the all-fiber configuration reported here, our future work aims at increasing the output average power to the few-watts level (by increasing the pulse repetition rate of the fiber oscillator seed to the GHz range [65]) and at obtaining few-cycle emission at central wavelengths of 1.5 µm and 2.0 µm, particularly useful for localized nonlinear excitation of semiconductor materials transparent to these wavelengths [66].