1. Introduction

Laser sources delivering high repetition rate ultra-short pulses at the mid-IR (MIR) wavelength of 5 µm – 20 µm (500 cm⁻¹ - 2000 cm⁻¹) have attracted great interest recently. This spectral range, also known as the molecular “fingerprint” region where a majority of chemical bonds signatures reside, is important for a variety of scientific and industrial applications [1, 2]. Due to the limitations of laser gain medium at this spectral region, MIR pulses are usually generated with the aid of near IR (NIR) lasers by means of nonlinear frequency conversion. Among the various frequency down conversion schemes, difference frequency generation (DFG) can be most conveniently implemented due to its simple geometry. The challenging problem of the DFG approach is to find two synchronized light sources with the required spectral spacing as pump and signal wavelengths for generation of MIR pulses at the idler wavelength. Using solid state lasers, mixing of the pump/signal pulses has been realized using the long and short wavelength ends of the Ti:Sapphire laser spectrum [3], using two-color Ti:Sapphire laser outputs [4], using two tightly synchronized Ti:Sapphire lasers [5] and using the signal/idler outputs from an optical parametric oscillator synchronously pumped by a Ti:Sapphire laser [6]. During the preparation of this manuscript, the method of using a dual signal wavelength optical parametric oscillator pumped by a 7.4 W, 42 MHz Yb:KGW oscillator has also been reported based on which up to 4.3 mW (102 pJ) of MIR pulses were generated between 10.5 µm to 16.5 µm [7]. With solid state lasers, both pump pulses and signal pulses can be generated with very high power, from which the most powerful high repetition rate DFG MIR source to date has been reported [7].

Recently, with technology advances in high power femtosecond fiber lasers, new types of MIR systems have been demonstrated using either Er-doped fiber lasers or Yb-doped fiber lasers. Due to the known advantages of fiber lasers such as compactness, low-cost, alignment-free cavity and ease of thermal management, progress in fiber lasers-based MIR systems could lead to low-cost table-top MIR sources. For fiber lasers-based systems, signal pulses at a certain spectral spacing is typically seeded or generated via nonlinear frequency conversion in photonic crystal fibers (PCF). In one approach, a two color Yb chirped pulse amplification (CPA) system is used by which the signal pulses were amplified from the seed pulse selected from a PCF continuum [8–10]. Based on this method, tunable MIR pulses with up to 1.5 mW of average power between 16 µm to 20 µm were reported most recently [10]. This method has technically no limitation on the available signal pulse energy as long as the amplifier power is high enough; however, the MIR wavelength is limited to the longer MIR region since the wavelength of the signal pulse is limited by the Yb gain bandwidth. In another approach, signal pulses are generated as Raman solitons in the anomalous dispersion region (ADR) of a highly nonlinear PCF [11–15]. This method yields widely tunable MIR sources controlled by the soliton self-frequency shift [16], with the limitation of signal pulse energy being constrained to that of the fundamental soliton, which is typically less than 0.2 nJ. Based on this method, widely tunable MIR source at the fingerprint region has been demonstrated: from a 2.2 W, 151 MHz Yb comb, MIR pulses tunable from 3 µm to 10 µm with average power up to 1.5 mW (10 pJ) were generated [15].

For most MIR systems, the power of pump pulses is, in many situations, more than adequate with an upper limit set by the damage threshold of the nonlinear crystal. The capability of scaling up the signal pulse energy is thus critical for improving the frequency mixing efficiency and the generation of energetic MIR pulses. In this paper, we demonstrate a novel approach to generation of milli-watt level femtosecond MIR pulses using intense Stokes pulses excited from a highly nonlinear PCF with two closely spaced zero dispersion wavelengths (ZDW). The Stokes pulses are extracted from the continuum generated in the normal dispersion region (NDR) of the fiber. Compared to the broadband supercontinuum generated in a PCF with one ZDW [17], such NDR continuum has a much narrower bandwidth and was found to be stable, compressible with high spectral intensity [18, 19]. As an efficient and low-noise source, it has been successfully applied to a variety of subjects such as Optical Coherence Tomography [20] and Coherent Anti-Stokes Scattering microscopy [21]. The spectral position of the Stokes pulses depends primarily on the fiber dispersion profile [19, 22, 23], which makes the signal wavelength easy to tune for generation of other MIR wavelengths either by selecting a different PCF or by manipulating the pulse evolution dynamics via longitudinal tapering which yields continuum generation with optimized spectral profile, pulse shape and spectral coherence [24–26].

2. Generation and characterization of intense Stokes pulses

The source laser used in our study is a custom-built Yb fiber Chirped Pulse Amplifier (CPA) [27], the schematic of which is shown in the upper part of Fig. 1 . The seeding pulse is provided by a diode-pumped passively mode-locked Yb fiber oscillator, which employs nonlinear polarization evolution (NPE) as mode locking mechanism and a uni-directional ring cavity for stable and self-starting operation [28]. The oscillator runs at a 40 MHz repetition rate with its central wavelength tunable from 1025 nm to 1040 nm and a typical bandwidth of 15 nm. A core-pumped fiber pre-amplifier using 60 cm of Yb-doped fiber (23,600 ppm doping, 0.12 NA, 6 µm core diameter) amplifies the oscillator output to 5 nJ after pulse stretching with 20 meters of single mode fiber (Lucent 980). Cladding pumping is employed for the power amplifier which consists of a tapered fiber bundle (TFB) for pump/signal mixing and 4 meters of double clad Yb-doped fiber (Nufern fiber amplifier with PLMA-YDF-15/130 fiber) as gain medium. 4.5W of laser power can be extracted directly from the amplifier output after filtering out the residual pump. Pulses from the power amplifier are compressed by an external grating pair (600 l/mm, gold coated) in a double pass configuration. Between each stage, an optical isolator is implemented to prevent optical feedback. The CPA system delivers 34 nJ of pulse energy (1.36 W of average power) with a pulse width of ~320 fs, a center wavelength of 1035 nm at 40 MHz repetition rate. The spectrum and autocorrelation trace of the CPA output is shown in Fig. 2(a) .

 

Fig. 1 Experimental set-up of the MIR source. Upper part: schematic of the Yb fiber CPA system as pump source. Lower part: layout of the DFG MIR system.

Download Full Size | PDF

 

Fig. 2 (a) Output spectrum of the Yb CPA system (inset: autocorrelation trace); (b) Spectrum of the NDR continuum with 150 mW, 177 mW and 244 mW of coupled-in power; (c) Typical spectrum of the Stokes pulse; (d) Measured power of the Stokes pulse and its SHG from the BBO crystal.

Download Full Size | PDF

The photonic crystal fiber for NDR continuum generation is supplied by Crystal Fibre [29]. It has a parabolic dispersion profile and two closely spaced zero dispersion wavelengths nominally at ~1022 nm and ~1075 nm with a parabolic maximum of ~0.6 ps/nm/km and monotonically decreasing dispersion at wavelengths below and above the zero dispersion points. The fiber has a core diameter of ~2.2 µm, a high numerical aperture of 0.37 and a nonlinearity of ~0.37 (W∙km)⁻¹. With a silica fiber core and microstructured air holes cladding, single mode guiding is realized by the mechanism of modified total internal reflection [30, 31]. According to the fiber specification, the spectral bandwidth of our source laser lies in the slightly anomalous dispersion region of the fiber with the peak wavelength closer to the lower ZDW.

Light from the source laser is split into two arms (pump and signal) by a polarizing beam splitter and a half-wave plate. On the signal arm, light is coupled into 12 cm of PCF using an aspheric lens with 4.5 mm of focal length; coupling efficiency of ~35% is typically achieved. A half-wave plate and a variable attenuator are implemented prior to fiber coupling for fine control of power and polarization state of the coupled light. The output spectrum is collected and analyzed by an optical spectrum analyzer (OSA, Hewlett-Packard, 86142A). Figure 2(b) shows the collected spectrums with coupled-in power of 150 mW, 177 mW and 244 mW. It is found that with increasing coupled-in power, an anti-Stokes side peak appears first at a wavelength around 900 nm following an initial spectrum broadening due to self-phase modulation. This is because the pump wavelength is closer to the lower ZDW of the fiber. At a “threshold” power of 86 mW, the Stokes side continuum becomes observable and further increasing in power leads to rapid scaling up of the continuum power which eventually leads to the dual-peak spectrum structure typically found in this kind of PCFs. The Stokes side continuum (Stokes pulse) acts as the signal wavelength for subsequent frequency mixing and our following discussion will be focused only on this side.

The spectrum of the Stokes pulse is structured, with most of its power residing in a sharp primary peak on top of a broadband baseline. Figure 2(c) shows the spectrum of the primary peak in the region of 1250 nm, with its bandwidth measured to be around 25 nm. We note that the small spectral ripples in the spectrum are a minor artifact possibly due to a loose fiber coupling. Spectral position of the peak can be slightly adjusted on top of the baseline by varying the coupled-in power. The power of the Stokes pulses is measured using a germanium detector (LM-2 IR with Fieldmaster GS, Coherent) after filtering out the lower wavelength signal using a long pass dichroic mirror (DMLP 1180, Thorlabs). As is shown in Fig. 2(d), the average power of the Stokes pulse ranges from 43 mW to 58 mW as its peak is tuned between 1250 nm and 1285 nm, this corresponds to 1.075 nJ to 1.45 nJ of pulse energy which is 5 to 10 times larger than that of the fundamental Raman soliton generated in a typical highly nonlinear PCF. On the same plot, we also show the measured power (corrected for filter transmission) of Second Harmonic Generation (SHG) from the Stokes pulse by focusing it inside a 3 mm type-I beta-barium borate (BBO) crystal. The increase in SHG conversion efficiency confirmed that pulse energy of the primary peak becomes stronger with power increase of the Stokes side continuum. The SHG wavelength has been checked with a spectrometer (USB2000, Ocean Optics) to confirm the doubled optical frequency.

The Stokes pulse spectrum with the sharp primary peak structure is desirable for frequency mixing due to its high spectral intensity and is selectively excited by using a short length of PCF. The fiber length dependence of the NDR continuum generation has been studied both theoretically [20] and experimentally [22] and it was found in good agreement that a short length of PCF yields structured NDR continuum with a sharp peak. Increasing the fiber length to 30 to 50 cm resulted in a smoothened broadband continuum which was basically unchanged with further increasing in fiber length. The experimental results in [22] also showed that with the short fiber in use, the spectral position of the primary peak changed with variation of coupled-in power or fiber length, comparing to which our conclusion is in good agreement. By using a short PCF, we also avoid pulse broadening from excessive dispersion. Compression of the Stokes pulse is possible [32] at the expense of adding an additional dispersion compensation stage which would increase the complexity of the subsequent MIR system.

The Stokes pulse wavelength primarily depends on the fiber dispersion profile, which results in a limited tuning range compared to the soliton self-frequency shift. Extension of its spectral range can be realized by using PCFs with different dispersion profiles [19, 22, 23]. As has been shown in [22], Stokes pulses spanning nearly 100 THz can be generated by using a series of PCFs with similar 1st ZDWs and a set of 2nd ZDWs ranging from 867 nm to 1138 nm (plus one PCF with vanishing ZDWs) pumped by a 795 nm laser. Even finer control of the Stokes pulse profile can in principle be realized via longitudinal tapering, where evolution dynamics of the pulse can be manipulated by dispersion micro-management along its propagation length [24–26].

3. Generation and characterization of MIR pulses

The experimental set-up for the generation of MIR pulses is shown in the lower half of Fig. 1. The polarizing beam splitter splits as much as 440 mW to 650 mW power into the pump arm. After passing through a delay line for timing control, its beam diameter is adjusted by a telescope system for mode matching with the signal beam. The resulting beam size is measured to be ~5 mm in diameter. On the signal arm, the Stokes pulses are extracted from the highly nonlinear PCF, a fraction of which is split off a pellicle beam splitter to the OSA for synchronous monitoring of the signal pulse spectrum. Polarization state of the rest of the Stokes pulses is adjusted to vertical by a half wave plate. The pump/signal pulses are collinearly combined by a dichroic mirror (DMLP 1180, Thorlabs). Typically about 330 mW to 480 mW of pump power and 43 mW to 58 mW of signal power are measured after the dichroic mirror. The beams are focused inside the nonlinear crystal by a plano-convex lens with 100 mm focal length, which yields a focal spot diameter of ~30 µm.

The nonlinear crystal used for frequency mixing is a 2 mm thick, type-II AgGaS₂ crystal cut at θ = 50° (AGS-402H, Eksma Optics). It has an anti-reflection coating from 1.1 µm to 2.6 µm for the front surface and 2.6 µm to 11 µm for the back surface. Although the GaSe crystal has been routinely employed for high power MIR pulses generation, which, compared to AgGaS₂, possesses a higher nonlinearity (63 pm/V versus 31 pm/V), a larger MIR transparency range (0.65 µm to 18 µm versus 0.53 µm to 12 µm) and a higher damage threshold (28 MW/cm² versus 10 MW/cm²), we chose the type-II AgGaS₂ crystal for a number of reasons. First, AgGaS₂ can be polished and AR-coated, while GaSe cannot. With a high index of refraction (n_e = 2.908, n_o = 2.568 at 1.06 µm), using GaSe results in a significant power loss from Fresnel reflection which can be as high as 50%. Such loss is un-tolerable for the power level of our system. Second, AgGaS₂ has much smaller birefringence which results in less spatial walk-off between ordinary (signal) and extraordinary (pump and idler) beams. This means that the beams can be focused tighter before spatial walk-off effect starts to degrade the mixing efficiency. More importantly, the output from AgGaS₂ is expected to have a more circular shape which is desirable for one of our intended applications (stand-off trace gas detection) where the MIR output is expected to maintain a good spatial profile during its propagation in meters of path length. Finally, the Group Velocity Mismatch (GVM) is also smaller in AgGaS₂ which not only increases the effective interaction length but also limits the GVM induced temporal broadening of the MIR pulse. A comparison of the spatial/temporal walk-off between the two crystals is shown in Fig. 3 . A detailed summary of GaSe is described in [15] from which a more complete picture of MIR crystal selection can be found. Based on the pump power level we have as well as our intended application, we chose AgGaS₂ as a power-efficient DFG crystal.

 

Fig. 3 Comparison of GaSe (type I) and AgGaS₂ (type II) on (a) Spatial walk off angle; (b) pump – signal GVM and pump – idler GVM.

Download Full Size | PDF

After the crystal, a 2 mm thick AR coated Germanium filter is used to block pulses other than the MIR wavelength. The power level is measured directly after the filter with a thermo-pile detector (PS-19 sensor with FieldMax 2 detector, Coherent). The measured MIR power ranges from 2.4 mW to 3 mW as the signal wavelength is tuned between 1250 nm to 1285 nm with the crystal phase matching angle adjusted from 47.9° to 51.6° (calculated), respectively. The maximum MIR power is obtained from the mixing of 53 mW (1.325 nJ) of signal pulses centered at 1275 nm and 480 mW (12 nJ) of pump pulses centered at 1035 nm. The resulting MIR wavelength is thus tunable from 5.3 µm to 6 µm with the maximum power obtained with peak wavelength centered around 5.5 µm, which is confirmed by the follow-up spectrum measurement. We found that at the current focusing diameter (~30 µm), the crystal is susceptible to damage as the pump power exceeds 500 mW so we kept a constant pump power of 480 mW when the signal wavelength is tuned between 1250 nm to 1275 nm. As the signal wavelength is tuned above 1275 nm, this power level cannot be maintained as the power used for PCF coupling exceeds 550 mW. During the experiments, we observed that the long term system stability is primarily affected by the instability in fiber coupling. In our current set-up, the distance between the fiber amplifier and the PCF is longer than 5 meters so that the stability of the Stokes pulses generation becomes deteriorated over time considering that sub-micron precision is required for fiber coupling.

The optimal focusing spot size of ~30 µm is determined with considerations of a set of frequency mixing parameters. The first one to consider is the pulse splitting length $L_{ps}$($L_{ps}={\tau }_p/{\delta }_{wo}$, with ${\tau }_p$ being the pump pulse width and ${\delta }_{wo}$ being the temporal walk-off resulted from GVM as depicted in Fig. 3(b)) which is calculated to be 0.8 mm based on the pump source and the crystal used. As the effective interaction length, this sets a lower limit of 32 µm for the focusing spot size in order to prevent spatial walk-off, which is calculated from $D_{wo}=L\tan {\theta }_{wo}$, with ${\theta }_{wo}$being the spatial walk-off angle as plotted in Fig. 3(a). The confocal parameter at this spot size is calculated to be 1.5 mm which is longer than the effective interaction length. This ensures efficient amplification of the idler pulse as long as it stays temporally with the pump pulse. Tighter focusing leads to higher peak intensity with the trade-off of degradation in mixing efficiency due to the spatial walk-off effect.

After the Ge filter, the MIR beam is collimated with a CaF₂ lens and its spectrum is measured using a MIR wave-meter (721B-MIR, Bristol instrument) which employs a scanning Michelson interferometer for Fourier transform infrared (FTIR) analysis of the MIR spectrum. Figure 4(a) shows a typical spectrum with peak wavelength around 5.5 µm; a bandwidth of 133 nm is measured. With the MIR system set up and running, the cross-correlation trace is readily obtained by translating the pump arm delay line and measuring the MIR power at respective time delays. Figure 4(b) shows the DFG cross-correlation trace using a lower pump power, the cross-correlation width is found to be 220 fs, which confirms the ultra-short pulse widths of the mixed fields. The MIR pulse width is expected to be on a similar scale.

 

Fig. 4 (a) Typical spectrum of the MIR pulse at ~5.5 µm; (b) Cross-correlation trace measured by difference frequency generation.

Download Full Size | PDF

4. Conclusion

In conclusion, we have demonstrated a novel method of femtosecond MIR pulses generation using intense Stokes pulses from a PCF with two closely spaced ZDWs as signal pulses. Up to 1.45 nJ of Stokes pulse energy is obtained between 1250 nm and 1285 nm via continuum generation in the normal dispersion region of the PCF. Mixing the intense Stokes pulses with the rest of the pump pulses inside a power-efficient 2 mm thick type-II AgGaS₂ crystal directly yields MIR pulses around 5.5 µm with up to 3 mW of average power which corresponds to 75 pJ of pulse energy.

Although tunability of the MIR wavelength is at present small compared to the sources based on the Raman solitons and the two wavelength optical parametric oscillators, extension of the MIR wavelength can be realized by using PCFs with different dispersion profiles. Optimization of the frequency mixing process is also expected to be facilitated via finer control of the Stokes pulses profile through longitudinal tapering. The current output wavelengths cover several important chemical bonds such as C = O and C = C and our MIR system can readily be used for applications such as time resolved vibrational spectroscopy of trace gas and femtosecond infrared 2D spectroscopy on protein and amino acids [33]. Besides the high MIR power, the approach we adopt is power-efficient in both the Stokes pulses generation scheme and the frequency mixing process, which we believe to be a promising solution of developing high power table top MIR sources at a relatively low cost.