1. Introduction

High peak-power ultrafast lasers operating in the mid-infrared are uniquely useful for implementing tabletop sources of soft X-rays and electrons, due to favorable scaling of electron ponderomotive energy with longer wavelength, as well as phase matching for high-order harmonic generation (HHG) [1–6]. HHG sources in the extreme-UV spectral region driven by near-IR ultrafast lasers have matured to the extent that they are now used extensively to uncover new material behaviors and enable powerful new imaging metrologies [7–12]. Specifically, the single-atom photon energy cut-off and phase matching cutoffs for HHG scale as $\lambda ^2$ and $\lambda ^{1.7}$ respectively, which has been experimentally validated through the generation of bright, spatially coherent soft X-ray beams at keV photons energies using a 4 $\mu$m wavelength driving laser [13]. Similarly, laser wakefield accelerators (LWFA) would benefit from high-power mid-IR lasers due to the normalized vector potential scaling as $\lambda ^2$ and the plasma density as 1/$\lambda ^2$. An early experimental demonstration using a 4 $\mu$m wavelength driving laser with 25 mJ energy and $\sim$100 fs duration pulses has demonstrated the acceleration of electrons in a LWFA to 12 MeV in a tabletop setup [6]. However, despite the clear motivation, ultrafast laser technology in the mid-IR has been too immature allow for routine operation of these sources.

Over the past two decades, progress in the generation of ultrafast, high energy, high average power mid-IR lasers has followed four approaches. First, the development of transition metal (Cr$^{2+}$, Fe$^{2+}$) doped chalcogenide (ZnSe, ZnS) lasers has allowed for routine modelocking of ultrafast oscillators around 2.5 $\mu$m [14]. Chirped pulse amplification using the same gain media—Cr$^{2+}$:ZnSe in particular—has recently allowed the generation of multi-mJ optical pulses with sub-100 fs duration at kHz repetition rate [15–17]. Scaling these systems to higher average power is, however, challenging. Second, Tm-doped fiber lasers have reached >kW average power in the continuous wave regime, and up to 1.8 mJ energy with sub-100 fs duration at 100 kHz repetition rate [18]. These systems are currently limited in energy scaling by the accumulation of nonlinearity—demanding complex multiplexing architectures—and in average power scaling by transverse mode instabilities—demanding the development of high average power pump lasers in the 1.6-1.8 $\mu$m spectral range [19–21]. Third, Ho-doped gain media were first developed as early as the 1960s, but lacked suitable pump lasers. The recent availability of industrial grade high average power CW Tm:fiber lasers as pump sources has triggered a renewed interest in Ho:YLF [22–25], Ho:YAG [26], Ho:CaF$_{2}$ [27–31] and Ho:CALGO [32–34] resulting in amplified systems delivering tens to hundreds of mJ energy at 2 $\mu$m wavelength at tens of Hz to >kHz repetition rate. While the bandwidth of Ho-doped systems currently only supports ps duration pulses, they are excellent pump lasers for nonlinear parametric down conversion to the mid-IR. Finally, optical parametric chirped pulse amplification (OPCPA) can deliver mJ-level ultrafast pulses at wavelengths $\geq$3 $\mu$m, pumped either using NIR or 2 $\mu$m wavelengths. These systems have delivered up to 130 $\mu$J energy at 160 kHz repetition rate [35] or 3.2 mJ energy at 1 kHz repetition rate [36] at 3 $\mu$m wavelength, 3.4 mJ energy at 1 kHz repetition rate at 5 $\mu$m wavelength [37], 0.7 mJ energy at 100 Hz repetition rate at 7 $\mu$m wavelength [38] and even up to 65 $\mu$J at 1 kHz at 11.4 $\mu$m wavelength [39]. The current challenge for these systems lies in improving long-term stability, likely through reducing their complexity.

In this paper, we report a fiber-based seed laser for mid-IR OPCPA that can operate hands-off for weeks without noticeable drift in performance. This dual-output, 100 MHz source can seed a Ho-doped 2 $\mu$m high energy amplifier using >1 nJ pulse energy and a TEM$_{00}$ spatial profile, as well as synchronously provide pulses at 3 $\mu$m wavelength spanning the 2.5–3.5 $\mu$m spectral range with up to 39 pJ energy and an excellent spatial profile. The two outputs show power instabilities of <0.2% over extended operation.

2. Front-end design

A commercial erbium oscillator (Menlo Systems, GmbH, ELMO) with two identical outputs delivering $\sim$50 pJ pulses at 100 MHz repetition rate is used to seed the front-end (Fig. 1). The pulses have approximately 40 nm of bandwidth centered at 1550 nm and are negatively chirped to $\sim$700 fs. Both outputs are amplified to approximately 2 nJ energy and compressed to <40 fs pulse duration in two parallel Er-doped fiber amplifiers (EDFAs). Each EDFA uses 1.7 m of doped polarization maintaining fiber, core-pumped in the forward and reverse directions by single-mode diodes locked to 976 nm wavelength by a fiber Bragg grating with up to 1 W of power. Self-phase modulation throughout the amplifier broadens the spectrum, with compression occuring in anomalous dispersion fiber (PM1550). The degree of spectral broadening and compression at the output of the amplifiers is controlled by adjusting the length of passive fiber before and after the amplifier, as well as the diode currents. Second harmonic generation frequency resolved optical gating (SHG-FROG) is used as feedback to find the configuration that provides the shortest, highest power pulses.

 

Fig. 1. Schematic diagram of fiber-based mid-IR ultrafast seed laser system. EDFA: erbium-doped fiber amplifier, HNLF: highly nonlinear fiber, CFBG: chirped fiber Bragg grating, Tm:HoDFA: thulium/holmium co-doped fiber amplifier, MgO:PPLN: MgO-doped periodically poled lithium niobate.

Download Full Size | PDF

The spectrum of each EDFA output is further broadened in an anomalously dispersive highly nonlinear fiber (HNLF) spliced onto a short piece of passive fiber. The compression of the pulses enables a strong and tunable response in the HNLF. The parameters of the 2 $\mu$m arm are chosen such that the red-shifted Raman wave is centered on the Ho:YLF emission wavelength of 2052 nm. This spectral band is then selected and temporally stretched to seed a Tm,Ho co-doped fiber amplifier (see section 3 for details).

The HNLF parameters for the other EDFA arm target generation of an octave spanning supercontinuum as well as temporal compression of the fiber output. This output is then focused into an MgO-doped periodically poled lithium niobate (MgO:PPLN) crystal to induce a $\chi ^{(2)}$ nonlinear response and produce an OPCPA seed beam with a 3 $\mu$m center wavelength (see section 4 for details).

3. 2 $\mu$m amplification in cooled Tm:Ho fiber

Following earlier designs [40], we determined that 12 cm of HNLF with a dispersion of 1 ps/nm$\cdot$km allowed for production of a few tens of mW power in the Raman wave at >2 $\mu$m wavelength. The wavelength can be tuned to the Ho:YLF emission by adjusting the EDFA diode powers. The total power at the output of the HNLF is measured to be 108 mW and we estimate—by numerically normalizing the measured HNLF output spectrum to the pulse energy and integrating over a spectral band of interest—that the power contained in the 2049-2055.5 nm spectral band is $\sim$2 mW. The output of the HNLF is sent into a chirped fiber Bragg grating (CFBG) by an optical circulator where it undergoes temporal stretching and spectral narrowing. The CFBG employs PM1950 fiber, operates at a center wavelength of 2052 nm with a reflection bandwidth of $\sim$8 nm (Fig. 2(a)) and targets group delay dispersion (GDD) and third order dispersion (TOD) values of 351 ps$^2$ (157 ps/nm) and -41 ps$^3$, respectively. These values of GDD and TOD were selected to match those of a high efficiency grating compressor that will be implemented after regenerative amplification, and to ensure that the pulse duration remains >300 ps after spectral narrowing in the amplifier. The minimum reflectivity of the CFBG over 80% of the full width at half maximum (FWHM) bandwidth is 35%. Upon the exit of the circulator, we measure 200 $\mu$W average power in the 2048-2056 nm spectral band, in agreement with the expected power loss from the specifications of both the circulator and the CFBG. While the duration could not be directly measured—owing to both the low output power and the unavailability of a high bandwidth photodiode—we anticipate a stretched pulse duration of $\sim$1.2 ns, based on the chirp rate imposed by the CFBG and the measured reflected bandwidth.

 

Fig. 2. (a) Measured output spectra of 2 $\mu$m HNLF and Tm,Ho amplifier arm of system. (b) Measured output power versus pump power for a coil temperature of 5$^{\circ }$C (orange line) and 20$^{\circ }$C (dark line). Inset: photograph of the fiber coiled onto the water-cooled spool during amplification. (c) Log-scale normalized amplifier spectra, showing the onset of ASE as the gain approaches 30 dB. The ASE is below the noise floor for lower powers. (d) Power stability data of Tm,Ho amplifier output, exhibiting an RMS stability of 0.18% over 70 hours, and laboratory humidity recorded over the same time period (right axis).

Download Full Size | PDF

Following temporal stretching, the pulses are directed to a single stage, cladding-pumped fiber amplifier. The fiber employed for amplification is a 3 m long, double-clad (6/130 $\mu$m), polarization maintaining, Tm,Ho co-doped fiber. This fiber is pumped by an 8 W, 793 nm, multi-mode fiber coupled diode with a 105 $\mu$m core and 0.22 NA. The diode is mounted onto a thermo-electric cooler (TEC) set to 15$^{\circ }$C. The gain fiber is coiled onto a water-cooled spool and can be cooled as low as 5$^{\circ }$C before the onset of condensation. Fiber cooling was motivated by the low optical-to-optical efficiency in Tm,Ho gain fiber as well as the three-level nature of the Tm$^{3+}$ and Ho$^{3+}$ ions. Upon amplification we obtain up to 118 mW average output power at 3.85 W pump power when the fiber is maintained at 20$^{\circ }$C temperature. Cooling to 5$^{\circ }$C allows a 17% boost in average power, reaching up to 138 mW at the same pump power (Fig. 2(b)).

Note that we limit the gain of this amplifier to <30 dB to prevent amplified spontaneous emission (ASE) buildup, which can cause catastrophic damage to components via amplifier self-lasing or gain switching. Observed through a spectrometer with 16-bit dynamic range, we see a gently growing pedestal in the 1.9-2.04 $\mu$m region due to ASE (Fig. 2(c)). At the full power of 118 mW, we estimate about 2% of the power to be ASE based on the measured spectrum. Despite the relatively large gain of the amplifier, we observe no significant spectral narrowing or spectral reshaping.

Finally, we assess the long term stability of the output. The average power over 70 hours was stable to <0.2%, in part correlated to humidity variations in our laboratory which is apparent in the final 17 hours of the data set (Fig. 2(d)). We also characterized the pulse-to-pulse instability using an InGaAs photodiode attached to a 2 GHz oscilloscope and found instabilities to be <0.35%. Further, this amplifier has successfully been left on for $\sim$200 hr showing no noticeable drift in performance without any readjustment or active feedback control.

4. 3 $\mu$m generation in MgO:PPLN

Intra-pulse difference frequency generation (IP-DFG) driven by ultrafast fiber lasers has proven to be a simple method to produce stable mid-IR frequency combs, with conversion efficiencies of <1% and average powers up to $\sim$1 mW [41–47]. Multi-watt fiber lasers using free-space compressors and/or external nonlinear compression have scaled fiber-based IP-DFG to GHz repetition rates [48,49] and higher average powers with conversion efficiencies up to 2% [50–55]. In this work, our goal was to maximize power in the 3 $\mu$m band via IP-DFG in a simple configuration without the need for multi-stage amplification or free-space compression.

In spectral broadening of this EDFA arm in HNLF, a blue-shifted dispersive wave develops along with the Raman wave, and these two spectral bands briefly temporally overlap in the fiber. The spectral content of the dispersive and Raman waves may be optimized to produce the desired mid-IR spectrum. We performed pyNLO [56,57] simulations of how the EDFA pulses retrieved by FROG (Fig. 3(a–b)) propagate in the HNLF. PyNLO solves for pulse profiles and spectra in the presence of SPM, Raman scattering and self-steepening. In simulation, we vary the length of passive pre-fiber, the length of HNLF and the value of dispersion in the HNLF in order to find an optimum for 3 $\mu$m generation. With our initial goal being direct DFG between the dispersive and Raman waves, we found that a minimal amount of pre-fiber and 3 cm of HNLF with 1 ps/nm$\cdot$km provides the best spectrum and compression based on our measured EDFA output. Simulations of the pulse profile and simulated and measured spectra are shown in Fig. 3(c–d). The temporal beating in the simulated pulse envelope stems from interference between the high frequency dispersive wave and low frequency Raman wave, and indicates that they are overlapped in time.

 

Fig. 3. (a) SHG-FROG pulse retrieval of EDFA output used for spectral broadening in HNLF. The retrieval error is 0.44% on a 256x256 grid. (b) Measured and retrieved spectra and spectral phase for the same EDFA output. (c) Simulated pulses after HNLF tuned to generate 3 $\mu$m via $\chi ^{(2)}$ interaction. The beating in the pulse envelope indicates temporal overlap between the dispersive and Raman shifted waves. (d) Simulated and measured spectra out of the HNLF.

Download Full Size | PDF

The HNLF output is focused into a 1 mm thick MgO:PPLN crystal to produce pulses at 3 $\mu$m wavelength. Gold-coated parabolic mirrors are used to collimate and refocus the beam to an $\sim$11 $\mu$m $1/\mathrm {e}^2$ spot diameter. Fine adjustments to the output pulse are again made by tuning the EDFA pump diode powers. The crystal is mounted in a temperature controlled oven and features 16 separate domain apertures, each with area 0.5x0.5 mm$^2$ and a fixed poling period. The range of periods spans 24.06-36.95 $\mu$m to cover a wide range of phase-matching conditions. Longer poling periods (>30 $\mu$m) directly phase-match DFG between the dispersive and Raman waves, and the mid-IR spectrum can be tuned by choice of poling period. We obtain $\sim$pJ pulse energy by this method, as is consistent with similar work [42–44].

Measuring through a 2.4 $\mu$m long pass filter and correcting for its transmission loss, we achieve a much higher pulse energy of 39 pJ on a 25.86 $\mu$m poling period at 100$^\circ$C with evidence of cascaded $\chi ^{(2)}$ nonlinearity. Direct DFG between the dispersive and Raman waves would cause a decrease of energy in the dispersive wave and an increase of energy in the Raman wave at the MgO:PPLN output. However, we observe the opposite effect, as well as the presence of a strong spectral peak centered at 926 nm (Fig. 4(a)). An analysis of processes that are phase-matched at this poling period (Fig. 4(b)) indicates that second-harmonic/sum-frequency generation of the Raman wave at $\sim$1.8 $\mu$m wavelength produces the 926 nm peak, which then pumps DFG seeded by the dispersive wave at $\sim$1.3 $\mu$m wavelength to produce the measured 3 $\mu$m spectrum. Furthermore, once the 3 $\mu$m beam is generated the phase-matching conditions allow the Raman wave to continue adding energy to it (with 3 $\mu$m now as the signal wavelength) via direct DFG, though we did not attempt to measure the $\sim$4.5 $\mu$m idler beam for this process. The measured 3 $\mu$m spectrum spans the 2.5–3.5 $\mu$m spectral range, ideally matching the amplification bandwidth of a 2 $\mu$m pumped ZGP-based optical parametric amplifier. It should be noted that the benefit of passive carrier envelope phase stability from direct IP-DFG is lost in this configuration. With the amplifier pump diode currents set to maximize the 3 $\mu$m power, the power incident on the crystal is 171 mW, which corresponds to a conversion efficiency of 2.3%. To the best of our knowledge, this is the highest reported mid-IR power produced by parametric down conversion of a single fiber output without external compression, and the efficiency rivals that of $\mu$J pump sources [50–52,55].

 

Fig. 4. (a) Measured spectra out of HNLF and MgO:PPLN crystal for 3 $\mu$m generation. The spectral intensity of the mid-IR is enhanced for visibility. Inset: in-focus spatial profile of mid-IR beam. (b) Pump wavelength as a function of poling period for second harmonic generation (SHG) and DFG in MgO:PPLN at 100$^\circ$C, with one of the wavelengths in the DFG process fixed to 3 $\mu$m. The dotted black line indicates the poling period which experimentally yields the highest average power at 3 $\mu$m. The spectral peak at 1.8 $\mu$m produces second harmonic (step 1), which then parametrically amplifies the spectral peak at 1.3 $\mu$m to produce 3 $\mu$m (step 2). Calculations were made using the Sellmeier equations provided by Covesion. (c) Measured power stability data of 3 $\mu$m output, exhibiting an RMS stability of 0.09% over 68 hours, limited by lab humidity fluctuations (right axis).

Download Full Size | PDF

Collimating the mid-IR beam with a third off-axis parabolic mirror and refocusing with a lens produces an excellent in-focus spatial mode for seeding an OPCPA (Fig. 4(a) inset). The pulse energy is also more than sufficient—mJ mid-IR OPCPAs have been seeded with energies as low as 10 pJ [37] and 7 pJ [38,58]. The power instability was measured over 68 hours to be less than 0.1% RMS (Fig. 4(c)), which remarkably is better than that of the EDFA. We attribute the improvement to the presence of parasitic processes in the MgO:PPLN. Several mW of visible light is produced though a cascaded sum-frequency generation, which is known to dampen power instabilities in systems such as Cr:ZnS lasers [59].

5. Conclusion

A simple, stable front-end for a mid-IR OPCPA has been successfully implemented using spectral broadening and parametric down conversion of pulses from a reliable Er-fiber oscillator. This system produces pulses to seed a high-energy Ho:YLF pump laser, as well as broadband mid-IR light to seed an OPCPA though a novel mechanism that exhibits unprecedented efficiency and stability. The compact and robust architecture provides a reliable foundation for a high energy and repetition rate mid-IR source to power next-generation LWFA and soft X-ray HHG systems.