1. Introduction

In recent years intense laser pulses have found application in various industrial and scientific areas. Significant progress has been made over the last few years in improving the performance, meaning the pulse energy and average power, of laser systems emitting ultrashort pulses at high repetition rates. For this purpose, different amplification architectures such as slab [1], thin disk [2,3] and fiber [4,5] have been developed. All of these concepts can deliver average powers in the kW-range and pulse energies in the mJ-range. However, all of them suffer from their specific limitations, e.g. thermal and nonlinear effects, when pushed to even higher performance levels. Spatially separated amplification followed by coherent combination offers a new path to scale the performance of laser systems in spite of the mentioned limitations. Additionally, the amplification does not necessarily have to take place in an active medium, but optical parametric amplifiers can be used instead. Recently the combination of pulses coming from two parametric amplifiers in an OPCPA system has been demonstrated. In this case, the bandwidth of the combined pulses could be increased significantly compared to a single OPCPA system to support single-cycle pulses [6]. Combining a large number of fiber lasers has also been proposed to reach peak-powers in the exawatt level [7].

The viability of the coherent combining approach to scale the pulse energy and average power of laser systems has already been demonstrated by combining two LMA fiber amplifiers, resulting in femtosecond pulses with a pulse energy of 0.5 mJ at an average power of 88 W. A combining efficiency as high as 90% has been achieved [8]. It should be mentioned that a different group carried out successfully a similar experiment using a different combining technique and stabilization method based on a lock-in amplifier [9]. Furthermore, the impact of the B-Integral and of the fiber length mismatch on the combining efficiency in such amplifying interferometers has been studied in [10]. According to this study, the combination of ultrashort pulse amplifiers can be efficiently carried out even under realistic experimental conditions, i.e. at high B-Integrals (~10 rad) and with a mismatch of the dispersive lengths between the amplification channels of the order of a few cm.

The experiment presented in this paper exploits the concept of spatially separated amplification in two rod-type fiber amplifiers. Because of their very good and stable beam quality, these fiber amplifiers allow for a highly efficient coherent overlap of the beam profiles. Using this technique together with established technologies such as chirped-pulse amplification (CPA) and active pulse shaping to mitigate the detrimental impact of nonlinear effects [11], it was possible to obtain femtosecond pulses with an energy of 3 mJ and a peak power as high as 5.4 GW. This is, to the best of our knowledge, the highest energy and peak power achieved to date using fiber based femtosecond laser systems. The improvements compared to the previous results in [8] could be achieved with a longer stretched pulse duration, a more powerful preamplifier system and the usage of a pulse shaper.

2. Experimental setup

The experimental setup is shown in Fig. 1 . It is based on a strongly modified version of the experimental setup presented in [12]. A mode-locked oscillator delivers pulses with a bandwidth of about 7 nm at a repetition rate of 40 MHz and an average power of 150 mW. The center wavelength is 1030 nm. The pulses are stretched in a fiber stretcher consisting of 100 m single-mode fiber with a core diameter of 6 µm. After amplification in the first preamplifier, the pulses are stretched in a grating stretcher with 1740 lines/mm gratings and a spectral hard-cut of 7 nm to a pulse length of about 2 ns. Then, a spatial light modulator (SLM) is used to compensate for any residual spectral phase (including accumulated nonlinear phase due to SPM in the amplifiers) that may still exist behind the compressor. This SLM is controlled by an active pulse shaping algorithm based on MIIPS [13]. With a combination of two acousto-optic modulators (AOM) and a second single-mode fiber preamplifier, the repetition rate can be reduced to 10 kHz, while still maintaining sufficient average power to seed the third preamplifier. This one uses an ytterbium-doped photonic crystal fiber with a core-diameter of 40 µm and a high numerical aperture pump-cladding of 200 µm. It can provide average powers of about 500 mW while still withstanding output pulse energies of up to 50 µJ.

 

Fig. 1 Schematic setup of the coherently combined chirped-pulse fiber amplifiers. SLM: spatial-light modulator, AOM: Acousto-optic modulator, HC: Hänsch-Couillaud detector

Download Full Size | PDF

The two main-amplifiers are coherently combined in a Mach-Zehnder type interferometer. The splitting and combining of the beams is realized using polarization dependent beam splitter cubes, i.e. employing the polarization combining technique described in [14]. A half-wave plate in front of the input cube is used to split the incoming beam into two with an intensity ratio of 1:1 to get the same seed power for both main amplifiers. An actively controlled delay line is used to match the optical path lengths of the two channels to one another. This delay line is realized with a piezo-mounted mirror set on a manual translation stage. While the translation stage helps to achieve a coarse path length match, the piezoelectric actuator is responsible for the fine corrections of the path length mismatch in a range of just several wavelengths. With a double pass through a quarter-wave plate positioned at an angle of 45°, the polarization of the reflected beam at the splitting cube, i.e. of the beam going through the delay line, is rotated by 90° and the beam is then transmitted straight through the cube.

The main amplifiers are two 80 cm long polarizing rod-type PCF fibers with a mode-field diameter of 75 µm. Up to a power level of 20 W the emitted beam quality is very close to diffraction-limited with a very high pointing stability. Higher average powers lead to slight mode deformations which are detrimental for the coherent addition. Both amplifiers are pumped by 915 nm laser diodes. This pump wavelength is chosen to obtain a higher inversion level compared to the standard 976 nm pumping of ytterbium-doped fibers. As a result, higher pulse energies can be extracted before saturation-induced pulse shaping reduces the temporal and spectral width of the amplified chirped pulse.

As mentioned before, the coherent addition is achieved using the polarization combining technique [14]. Assuming that a polarizer cube is simultaneously illuminated by s- and p- polarized light with equal power and that there is no phase difference between those two components, then the emitted beam is linearly polarized with its polarization orientation tilted by 45°. Any phase difference between the two incoming fields results in a change of this output polarization state. Consequently, the polarization state provides direct access to the path length mismatch between the two branches of the amplifying Mach-Zehnder interferometer. For that purpose, the reflection of an anti-reflection coated laser window directly behind the combining cube is used as a signal for a Hänsch-Couillaud (HC) detector [15]. This detector can detect the polarization state of the combined beam and provides a feedback signal for the piezoelectric actuator in the delay line using some control electronics. The system is setup to optimize the combined beam for linear polarization. An earlier analysis has shown that the path length stabilization scheme employed, including the electronics and piezo-stage, has a bandwidth of approximately 1 kHz, which is sufficient to correct the majority of perturbations that can be found in a laboratory. Finally, the combined pulses are recompressed in a grating compressor that shares some of its components with the stretcher and reaches a throughput efficiency of 80%. The characterization of the combining process has been done by estimating the degree of linear polarization (DOLP) and the system efficiency, as described in a previous paper [8]:

One should note that this definition of the system efficiency includes loses at the combining cube due to non-perfect polarization of the amplifier emission. To make sure that the polarization of the combined beam is not just optimized at one particular distance from the combining cube, the polarization dependency of the compressor gratings is exploited in using it as the analyzer. This ensures an excellent overlap of the beams over a distance of at least 10 m.

3. Experimental results

The system was operated at a repetition rate of 15 kHz to keep the pulse energies at moderate levels. In this configuration, the two amplifiers delivered compressed average powers of 17.7 W each. After combination, an average power of 32.6 W was obtained with a DOLP of 87%. At the dark port of the combining cube, a power of 2.3 W was measured, caused by a certain degree of depolarization in the amplifiers. Hence, the system efficiency was estimated to be 89%, which is comparable to the values of around 90% shown in previous experiments at significantly lower power levels.

The spectrum in Fig. 2 a ) shows the spectrum of the signal after amplification. The spectral bandwidth is slightly narrowed from an initial value of 4 nm, delivered by the pre-amp system, to 3 nm at the output. This was mainly due to saturation effects which cause the leading edge of the pulse (red) being more amplified than the tail (blue). Nevertheless, a good match of the amplified spectra coming from the two channels can be observed in Fig. 2, which allowed obtaining a high system efficiency. After compression, a duration of the autocorrelation trace of 780 fs (Fig. 2 b) was measured which, using a deconvolution factor of 1.68 calculated with the transform-limited pulse, corresponds to an estimated pulse duration of 465 fs. Again, a very good match between the characteristics of the pulses coming from the channels and the combined pulses could be observed. The combined pulse energy was 2.6 mJ directly after the output polarization cube and 2.1 mJ behind the compressor. The B-Integral in this configuration for the two combined amplifiers was calculated to have a value of 7 rad, whereby the SLM reduced the pulse quality degradation due to the presence of nonlinearity.

 

Fig. 2 (a) Spectra and (b) autocorrelation traces of the single amplifiers and of the combined beam at a combined pulse energy of 2.1 mJ.

Download Full Size | PDF

In the next step, the repetition rate was reduced to 10 kHz, leading to a B-Integral in the complete preamplifier chain of about 1.2 rad. Compressed output powers of 14.6 W (channel 1), 14.3 W (channel 2) and a combined average power of 28 W have been achieved. This corresponds to a compressed pulse energy of 2.8 mJ. In this configuration, a DOLP of 88% and a system efficiency of 89% were obtained. The autocorrelation traces with a duration of 840 fs showed a slightly degraded pulse quality compared to the previous one due to the accumulation of more spectral phase imposed by SPM which could not be fully compensated for by the pulse shaping system. However, the FWHM pulse duration could be estimated to be as short as 470 fs, corresponding to a peak power of ~5 GW (assuming that a 10% of the energy was contained in the pre- or post-pulse features).

Finally, the output power of the channels was raised again to a compressed 16.3 W per channel, resulting in a combined and compressed pulse energy of 3 mJ (Fig. 3 ). The B-Integral in this case was calculated to be 9 rad. Again, this value is for the two combined amplifiers only. However, in this configuration, the DOLP dropped to 79% and the system efficiency to 84%. Difficulties with the excitation of the fundamental mode in the few-mode fibers of the main amplifier prevented obtaining better values and/or higher power levels. Nevertheless, pulse durations as short as 470 fs could be achieved, corresponding to peak powers as high as ~5.4 GW.

 

Fig. 3 Autocorrelation traces of the single amplifiers and of the combined beam at combined pulse energies of (a) 2.8 mJ and (b) 3 mJ.

Download Full Size | PDF

The characterization of the stability of the setup was done by recording the voltage applied to the piezo in the delay line. With an additional calibration, the compensated optical path length difference (OPD) can be calculated. Two of these traces are shown in Fig. 4 for a timescale of 450 s. The first one was recorded right after switching on the amplifiers while, in the second case, the system was already running for a couple of minutes. Hence, a drift of the OPD can clearly be seen at the beginning of the first trace, before the system reaches a stable state. This drift can be explained by thermal effects that have an influence on the optical path lengths of the channels. In the second trace a peak-to-peak fluctuation of 2.3 rad could be measured. This high stability of the setup was achieved by placing the whole setup into a housing and using water cooled modules for the two main amplifiers, therefore reducing thermal drifts.

 

Fig. 4 Compensated OPD (a) just after the system was switched on at 2.1 mJ compressed pulse energy and (b) after a couple of minutes running at a compressed pulse energy of 3 mJ.

Download Full Size | PDF

4. Conclusion

In conclusion, it has been shown that with two coherently combined amplifiers in a femtosecond fiber CPA system, pulse energies can be achieved that are higher than the currently reported record value of 2.2 mJ for a single amplifier [5]. Furthermore, the extraction of the presented results from a single amplifier stage would overstress current fiber laser technology. Thus, the 3 mJ of compressed energy reported here would require 3.8 mJ out of the single fiber amplifier. This, assuming a stretched pulse duration of 2 ns, an active fiber as used in the setup (75 µm MFD, length = 80 cm), would lead to a B-integral as high as 12 rad, peak powers as high as 2 MW, a fluence of 86 J/cm² and a peak intensity of 45 GW/cm². Hence, the extraction of the energy levels presented herein would be very close to or even beyond the well-known limitations of fiber based amplification of ultrashort pulses.

The coherent combining approach makes it possible to circumvent those limitations. At the same time, total combining efficiencies of up to 89% could be achieved, which are close to the values previously reported at significantly lower pulse energies. The stability tests also revealed that a carefully setup system lowers the requirements on the compensation of OPD fluctuations. These results make us confident that further performance scaling to pulse energies beyond 10 mJ is possible by using a larger number of amplifiers in a similar configuration. It should also be emphasized that this concept suits any amplification architecture of ultrashort laser pulses, including parametric amplifiers.