1 Introduction

Nonlinear optical phenomena such as self phase modulation, four wave mixing and Raman processes in various fibers and bulk media can be efficiently exploited for the compression of femtosecond pulses. These techniques are particularly important in few-cycle and single-cycle laser pulse generation [1–4], attosecond experiments [5] and other cases where shorter pulses are required than the gain and/or dispersion bandwidth allowed by laser oscillator or amplifier architectures. Therefore, one has to perform extra cavity pulse compression which necessitates (i) nonlinearities to increase the spectral content and (ii) proper spectral phase management to end up with a pulse which is as close to the transform limit as possible. In some cases, elements (i) and (ii) are realized in the same medium making use of the strong interplay of various linear and nonlinear phenomena during the compression process. Examples include self-compression in filaments [6] or in the soliton regime [7].

For most input pulses, however, more robust and broadband solutions are offered by schemes where the spectral broadening and pulse compression stages are separated [8–14]. In these cases, the spectral broadening stage usually consists of single-mode solid-core [8–10], multi-mode [12], rod-type [13] or gas-filled hollow-core [14] fibers. The spectrum of femtosecond oscillator pulses with up to ~30 nJ energy can be efficiently broadened in short single-mode glass fibers with 2–5 μm core diameter [8, 9], whereas the ~mJ pulses of femtosecond amplifiers can be efficiently guided and broadened in gas-filled capillaries [14]. With the recent advent of femtosecond technology resulting in high-energy, 100 nJ…several μJ pulses with >1 MHz repetition rate [15–18] (in the Ti:sapphire case, directly from the oscillator), new methods for pulse compression have to be devised for these novel parameter regimes. This has to take into account the desired spectral broadening and the damage threshold of nonlinear optical elements. A particular challenge for Ti:sapphire oscillators is also represented by the large bandwidth of both the source and the required spectral broadening, therefore, schemes devised for the compression of multi-100-fs pulses [10–13] can not be implemented.

Large mode area (LMA) photonic crystal fibers (PCFs) offer endlessly single-mode guiding combined [19] with a mode field diameter of up to around 100 μm and an M2 value of 1.05–1.2 at the fiber output [20, 21]. These features make them suitable for long-cavity Ti:sapphire oscillator pulse compression. The large mode area prohibits optical damage at the fiber entrance and fiber guiding allows sufficient nonlinearity for pulse compression at the same time, as indicated by supercontinuum (SC) generation experiments [22–24]. The potential offered by LMA PCFs was shown recently in compression experiments with relatively long and/or heavily chirped pulses [25, 26], based on the principle of chirped-pulse supercontinuum generation [27]. In those cases, however, sub-ps input pulses were used to avoid the optical damage of fibers. Because of this, long fiber pieces of several centimeters have to be applied to reach the desired spectral broadening. As opposed to this, our approach is that we demonstrate efficient femtosecond pulse compression in LMA PCFs with transform-limited input pulses. This concept enables the usage of short (<9 mm) LMA fibers allowing the pulse compression stage to be performed with broadband, low-dispersion chirped mirrors (CMs), in contrast with the scheme based on chirped input pulses [26]. Our simple setup results in 15…18-fs compressed pulses with up to 100 nJ pulse energy. This means that dispersion compensation over a significantly higher bandwidth can be performed correctly, implying the bandwidth scalability of our technique, opening the pathway for few-cycle pulse generation in this parameter regime.

2 Spectral broadening in large-mode-area fibers

As seed pulse for our experiments depicted in Fig. 1, we used a self-built long-cavity, chirped-pulse Ti:Sapphire oscillator operating in the positive dispersion regime [28] pumped by 8 W from a continuous-wave solid-state laser at 532 nm (“finesse” of Laser Quantum). It produces heavily chirped, 220 nJ, picosecond pulses with 3.6 MHz repetition rate that are compressible to 70 fs by prisms.

Fig. 1
figure 1

Scheme of the experimental setup. AC interferometric autocorrelator, LMA PCF large-mode-area photonic crystal fiber

Full size image

In order to suppress pedestals induced by the third-order dispersion, we used a four-prism compressor for the oscillator output (Fig. 1). This way, pulses entering the fiber are 1.4-times their transform limit which enables efficient spectral broadening and minimizes the effect of fiber dispersion, too. We used different achromatic lenses for input coupling. Substantial broadening could be achieved this way in 7…9 mm fiber pieces. As a result, dispersion compensation is also simple and can be performed with commercially available chirped mirrors. To this end, we used a single CM pair with alternating group delay dispersion (GDD) oscillations of the mirrors.

Broadened spectra and power output from different LMA fiber types are shown in Fig. 2a, b on logarithmic and linear scales, respectively. The oscillator spectrum is also shown for reference in both figures. It can be seen that efficient spectral broadening can be achieved for a variety of fibers and supercontinua covering a ~280 nm band can be generated especially with LMA PCFs with smaller core diameters, as shown in Fig. 2a.

Fig. 2
figure 2

a Supercontinua in different LMA fibers with 12, 10, 8.5, and 5 μm core diameters on logarithmic scale in arbitrary units. b Laser spectrum, and broadened spectra in ESM-12 and LMA-10 fibers with 70 and 100 nJ pulse energies, respectively. The inset shows the output beam profile. c Group delay dispersion of the pulse compression chirped mirrors

Full size image

The efficiency of the supercontinuum generation process is between 40 and 60 % in every case, limited by fiber coupling and propagation losses. This means that 80–100 nJ pulse energy can be easily achieved except for fiber LMA-5 (mode field diameter, MFD = 4.2 μm) where optical damage at the entrance face starts to play a role at this pulse energy level. This scheme provides substantially increased spectral content with lower chirp representing more favorable conditions for pulse recompression.

It is also interesting to note that the broadest spectrum can be produced in LMA-8 (MFD ~6 μm). For smaller core diameters, damage prohibits coupling of more than 36 nJ into the fiber. In contrast, for larger cores the available ~200 nJ from the laser safely does not cause damage at the fiber entrance but these fibers result in less spectral broadening.

3 Pulse compression experiments

In order to have a stable, short-pulse light source without even any stochastic damage due to pulse energy fluctuations, we set out to compress the somewhat narrower outputs of LMA-10 (MFD ~7.5 μm) and ESM-12 (MFD ~6 μm). (The transform limit of the spectra from LMA-5, LMA-8, LMA-10, and ESM-12 are 11, 6, 11, and 12 fs, respectively). We used a pair of matched CMs resulting in a flat dispersion profile over a broad spectral range, see Fig. 2c. The typical total number of bounces was 20–22, giving ~−1,000 fs2 GDD in total. This compensates the linear dispersion of the short fiber piece (~290 fs2, following closely the material dispersion of fused silica [19]) and the accumulated positive chirp resulting from the self phase modulation process in the fiber.

Compressed pulses after the CMs were characterized by a short-pulse interferometric autocorrelator (APE GmbH), the results of which are plotted in Fig. 3. It can be seen that pulses at close to the transform limit can be generated by standard CM compression. Approximate pulse shape reconstruction was carried out by fitting the measured autocorrelation (AC) traces by approximating the (originally unknown) spectral phase function with a polynomial, as done in PICASO pulse reconstruction [29]. Best computed AC curves resulting from this fitting are also plotted in Fig. 3 (see black solid AC envelopes). It can be seen that reasonably good AC reconstruction can be achieved this way.

Fig. 3
figure 3

Measured interferometric autocorrelation traces after compression in a LMA-10 and b in ESM-12 fibers (red curves). The envelopes of computed autocorrelation traces (from measured spectra) for transform-limited (blue dotted curves) and chirped cases (black solid curves) are also shown in arbitrary units. The insets show transform-limited (blue dotted curves) and reconstructed (black solid curves) pulse shapes

Full size image

Reconstruction results show that compressed pulses have practically only uncompensated higher order chirp. For best-fit pulses the residual third-order dispersion was found to be −800 and −1,600 fs3 for pulses in Fig. 3a, b, respectively. This alone can explain the observed pulse lengthening to ~1.4-times the transform limit and the presence of pre- or post-pulses. Residual (uncompensated) GDD was negligible, 20 and −20 fs2, respectively. The third-order effect can be largely compensated for in a more advanced setup by the usage of third-order dispersion compensating CMs that were not at our disposal during the experiments. It can be stated, however, that ~1.4-times transform-limited pulse lengths between 15 and 18 fs can be easily achieved without any active phase control with simple, broadband chirped mirrors. This corresponds to a compression factor of >4 in each case.

4 Summary

In summary, our approach to high-pulse-energy femtosecond supercontinuum generation and pulse compression in LMA PCF fibers delivered broad spectra and short laser pulses with >4 pulse compression factors. The method presents an alternative to bulk compressors [30] with similar throughput and spectral content. Further scalability is offered by the commercial availability of LMA PCFs with 35 μm core diameter [19], thus we estimate that high repetition rate pulses with up to 1 μJ pulse energy can be delivered by this scheme (with somewhat longer input pulses to stay below the ~4 MW critical power for self focusing). The investigated pulse parameter regime involving high-energy high-repetition-rate femtosecond oscillator pulses corresponds not only to long-cavity Ti:sapphire, but also to several other novel laser sources including Yb-based thin-disk lasers as well as fiber laser sources [16–18]. We will apply this light source where relatively high-energy, ultrashort pulses are required such as high repetition rate strong-field experiments [31, 32] and ultrafast plasmonics [32, 33].