Multimodal ﬁber source for nonlinear microscopy based on a dissipative soliton laser

: Recent developments in high energy femtosecond ﬁber lasers have enabled robust and lower-cost sources for multiphoton-ﬂuorescence and harmonic-generation imaging. However, picosecond pulses are better suited for Raman scattering microscopy, so the ideal multimodal source for nonlinear microcopy needs to provide both durations. Here we present spectral compression of a high-power femtosecond ﬁber laser as a route to producing transform-limited picosecond pulses. These pulses pump a ﬁber optical parametric oscillator to yield a robust ﬁber source capable of providing the synchronized picosecond pulse trains needed for Raman scattering microscopy. Thus, this system can be used as a multimodal platform for nonlinear microscopy techniques.


Introduction
Multiphoton fluorescence microscopy continues to find new and exciting applications. Secondand third-harmonic generation imaging share some benefits with multiphoton imaging, while offering complementary capabilities. Imaging techniques based on Raman processes, including coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS) microscopy, also continue to proliferate [1]. Raman microscopies typically require two synchronized picosecond pulse trains, with the frequency difference between the two colors tunable to match the vibrational resonances of interest. Despite the technical success of the solid state laser and optical parametric oscillator (OPO) commonly used for CARS and SRS mi-croscopy [2], its cost, size, complexity, and sensitive alignment will limit its use outside of specialized laboratories. This has motivated the development of fiber-based sources of picosecond pulses for Raman imaging. Some studies employ multiple nonlinear-imaging modalities, which generally require that both femtosecond and picosecond lasers be available. Multiphoton and harmonic-generation imaging benefit from maximum peak power, which strongly favors the use of femtosecond-duration pulses. Raman imaging also benefits from high peak power, but picosecond pulses are preferred to match the Raman linewidths. A robust source capable of providing both types of pulses would greatly facilitate such work.
Numerous fiber-based systems have been developed for Raman imaging. In a hybrid approach, fiber lasers are used to pump solid-state OPOs [3,4]. Fiber lasers can also be used with various nonlinear processes in fiber to generate the second color [5][6][7]. Finally, four-wave mixing (FWM) in the normal dispersion region of photonic crystal fiber (PCF) is employed to frequency-convert picosecond pulses from ytterbium-doped fiber lasers [8][9][10][11]. In addition to matching the pulse parameters needed for CARS, sources for SRS microscopy need to have ultra-low intensity noise. To date, no all-fiber source has achieved low enough intensity noise to be suitable for video-rate SRS microscopy with direct detection. Instead, Freudiger and coworkers demonstrated that electronic balanced detection may be used for cancellation of the noise on fiber sources [6], and Rimke et al. showed that a solid-state OPO pumped by an amplified fiber laser can achieve low enough noise on the frequency-shifted pulse train [12].
In addition to sources designed to meet the pulse parameters necessary for Raman microscopies, there is interest in developing sources that are suitable for multiple imaging modalities. Huff and coworkers constructed a multimodal source based on two synchronized picosecond titanium sapphire lasers to provide the desired pulse parameters for CARS [13]. Their system also uses the picosecond pulses to achieve two-photon excitation florescence (TPEF) and sum-frequency generation (SFG) images, which sacrifices performance compared to the use of femtosecond pulses in these imaging techniques. Kumar et al. perform single pulse CARS microscopy in conjunction with multiphoton imaging using a titanium sapphire laser and notch filters scanned by a galvanometer [14]. Fiber-based systems are attractive for these applications since they can reduce alignment sensitivity and cost. Xie et al. demonstrated a partially fiber source based on a femtosecond titanium sapphire laser used to perform CARS, SRS, and photothermal imaging [15]. The second color required for Raman imaging was generated through the soliton self-frequency shift in PCF. Again, this source is limited to very low powers (less than 10 mW) at 75 MHz and does not have the optimal picosecond durations for the Raman techniques. Two other sources demonstrate further fiber integration. One uses FWM of femtosecond pulses in PCF to yield a source capable of performing CARS microscopy and TPEF imaging [16]. However, the femtosecond pulses limit spectral resolution in CARS imaging, and the system is limited to 0.5 nJ at the signal wavelength. Finally, Meyer and coworkers demonstrate an all-fiber source for CARS, TPEF, and second harmonic generation (SHG) microscopy that operates with pulse durations of tens of picoseconds [17]. In summary, to the best of our knowledge, no multimodal source has been demonstrated that is capable of providing both the optimal synchronized picosecond pulses for the Raman microscopies and the high energy femtosecond pulses for multiphoton techniques.
Here we demonstrate the use of spectrally-compressed femtosecond pulses from a dissipative-soliton laser to pump a fiber OPO [9]. The OPO generates the second pulse train necessary for CARS and SRS microscopies. By utilizing a high-power femtosecond fiber laser [18,19] for the pump pulse, the same robust platform is capable of producing picosecond and femtosecond pulses near 1040 nm. The high energy femtosecond pulses from this laser have been used for TPEF and SHG microscopy [20][21][22], so the source presented here provides suitable pulses for multiphoton and harmonic-generation imaging with a single exci-tation wavelength, as well as CARS microscopy. As an additional motivation, SRS microscopy stands to benefit from an ultra-low noise fiber source. Since femtosecond lasers tend to have much higher peak powers than picosecond lasers, replacement of the low-power picosecond soliton lasers typically used as the seed laser with a more powerful laser reduces the shot-noise limit of the intensity noise in the source. Thus, the development of high-power femtosecond lasers as a source of quiet picosecond pulses may offer a route to quieter fiber sources. Extensions of the source to provide tunable femtosecond pulses will also be discussed. Dissipative-soliton lasers, which are based on all-normal-dispersion designs, tend to provide the highest-energy pulses for given core size. The performance of these lasers has been well documented [23]. They generate chirped femtosecond pulses; we assess spectral compression of the output pulses to create transform-limited picosecond pulses. Spectral compression occurs when a down-chirped pulse is launched into a nonlinear material [24,25]. Figure 1(a) shows the calculated spectral compression ratio versus impressed anomalous dispersion for 4 nJ transform limited Gaussian pulses with the indicated initial durations. The spectral compression is performed in passive fiber with a 10 µm core diameter up to 100 m long; the length is chosen to minimize the compressed bandwidth. These trends indicate that larger compression ratios can be achieved by impressing larger amounts of negative chirp on the pulse and then compressing the spectrum in correspondingly longer lengths of fiber in order to reach the transform-limited duration. Figure 1(b) shows compression of a simulation of a dissipative soliton. As the compressed pulse becomes longer, its peak power and thus nonlinear phase accumulation is reduced, which results in the saturation of the compression ratio with increasing anomalous dispersion. As indicated in Figure 1(a), the compression ratio will be limited to 10-20 with typical laser parameters. The limit on achievable compression ratio implies a tradeoff in the pulse durations. We targeted roughly 300 fs and 5 ps for the multimodal source. The experimental set-up is shown in Fig. 2. The dissipative soliton laser is based on double-clad Yb-doped fiber with a 10 µm core diameter (Liekki), similar to the cavity presented by Kieu et al. [19]. In standard operation [19], the laser generates pulses with over 20 nm bandwidth and over 10 nJ pulse energy, providing a source of pulses around 100 fs after dechirping with a grating pair. The repetition rate of the cavity is 21.4 MHz, which is selected for ease of producing spectral bandwidths near 10 nm (corresponding to a pulse duration of 300 fs). These bandwidths are ideal for spectral compression to picosecond pulse durations as shown in the simulations described in Fig. 1. The 10-nm-wide spectra are easily obtained with minor adjustment of the waveplates and no other change to the cavity, so the laser can be easily adjusted to provide either the 100 fs pulses or the pulses ideal for spectral compression. The fiber leads on the collimators and combiner are passive fiber with 10 µm core size. Around 5.5 m of HI1060 (Corning) fiber is added before the gain fiber to reduce the repetition rate to 21.4 MHz; the smaller-core fiber is used to reduce possible multimode content in the laser. The HI1060 fiber is spliced on both ends to SMF28e+ (Corning), which is then spliced to the 10 µm fiber to reduce splicing loss [19]. A quartz plate is used for the birefringent filter, which provides a bandwidth around 8 nm.

Experimental results
To generate the picosecond pulses, the chirp on the output pulse is reversed with a grating pair (LightSmyth Technologies, 1600 lines/mm) that provides -3.3 ps 2 of anomalous dispersion, and then spectral compression is performed in around 55 m of passive fiber with 10 µm core diameter. A typical result is presented in Fig. 1(c) to show the excellent agreement with simulation. A second compressed spectrum and the corresponding autocorrelation are shown in Figs. 3(a) and (b) to highlight the fact that spectra with different shapes, but similar bandwidths near the base of the spectrum, yield similar results. Although the initial pulse energies are over 10 nJ, only 1-1.5 nJ remains after compression. Most of the loss occurs in coupling the beam back into fiber for the compression, owing to the use of sub-optimal components in this proofof-concept experiment. Given the need to amplify the pulses, we did not worry about the low efficiency of the compression, which can be increased easily in the future. The 7 ps pulses are amplified in a divided-pulse amplifier (DPA). [26] The combination of 10 µm core fiber and division by 16 corresponds to direct amplification in a fiber with 40 µm mode-field diameter. However, fibers with modes this large are not compatible with standard splicing techniques. We find the DPA, which was available in our lab, easy to use and reliable; with it, amplification to 40 nJ can be achieved using standard single-mode fiber compatible with fiber-format combiners and collimators. Figure 3(b) shows a triangular autocorrelation trace, which indicates that the spectral compression is creating square-shaped pulses. These pulses are used to pump the OPO described previously [9] and shown in Fig. 2. A Fabry-Perot filter (OZ optics) replaces a grating-based filter [9] in the feedback loop; this serves to decouple the OPO path length from the resonant wavelength. The long-wavelength FWM product is resonated in the cavity. With 15-20 nJ coupled into 28 cm of PCF (5 µm core, zero dispersion wavelength near 1051 nm), 3 nJ of signal pulses can be generated around 800 nm. This corresponds to conversion efficiencies of 15-20%. Over 10 nJ of the 1040 nm pulses can be picked-off before the OPO to serve as the Stokes light for CARS or SRS microscopy while maintaining similar performance from the OPO. Signal pulse energies up to 4-5 nJ can be generated from the OPO at higher pump powers, but the spectra become more structured and the intensity fluctuations increase. If desired, higher pulse energies can be achieved by optimizing the PCF length and the pump pulse duration. The pulse parameters for both beams are similar to the previouslyreported performance [9], so this source is well suited for CARS microscopy in the popular C-H stretch region of the spectrum. The tuning range of the OPO signal wavelength is around 790-820 nm, and is currently limited by the 1450-1500 nm tuning range of the Fabry-Perot filter in the feedback loop. In this experiment, free-space coupling of light into and out of the PCF was used for ease of optimization; an all-fiber version of this OPO was recently demonstrated, and it exhibits similar performance [27].
Greater wavelength tunability will increase the usefulness of this source. Currently, the femtosecond pulses are tunable over the bandwidth of the ytterbium gain medium. Fiber OPOs have been designed to achieve tunable femtosecond pulses with up to 2 nJ of pulse energy, as demonstrated in references [28,29]. Recent results from Kieu and coworkers show that high energy femtosecond pulses can be created from a picosecond-pumped fiber OPO by exploiting dissipative-soliton formation in the OPO cavity [30]. Since the pulse duration from the spectrally compressed dissipative-soliton laser presented here is tunable over a wide range through mode-locking and compressing different spectra, it is capable of providing pump pulses for various fiber OPOs optimized for femtosecond pulse generation. Thus, the addition of a femtosecond fiber OPO to the set-up demonstrated here would provide a natural route for extending the wavelength tunability of the femtosecond pulses.
One motivation for replacing the low-power soliton laser typically used to generate the seed pulses in fiber-based systems for Raman imaging is the potential to start with a higher energy pulse, which has a lower shot noise limit. This is particularly important for SRS microscopy, which requires a nearly shot-noise limited source in the 10-20 MHz frequency range used for lock-in detection. Prior studies indicate that fiber OPOs provide a route to low-noise pulses that is superior to filtering a supercontinuum [31], so optimization of the pump laser is important to increasing the RIN performance of fiber systems for Raman microscopy. In the source presented here, the spectrally compressed pulses have 20-30 mW of average power. After amplification, the 1040 nm pulses have relative intensity noise (RIN) around −150 dBc/Hz (Fig. 4). The OPO itself adds around 10 dB of noise, yielding a RIN around −140 dBc/Hz at 800 nm. This is comparable to the RIN achieved using the soliton seed pulse [9], and would be suitable for SRS microscopy with balanced detection [6]. However, the higher starting average power in this system makes it possible to improve the noise performance of this source by optimization of the laser. Eventually, it may be advantageous to construct a dissipative-soliton laser with large- core fiber, which can directly produce 40 nJ pulses for pumping the OPO without amplification [32,33]. This would reduce the shot noise limit of the pump pulses even further, and eliminate the RIN added during amplification. As a practical benefit, elimination of the amplifier would also reduce the number of parts and complexity of the set-up.

Conclusion
In conclusion, we have demonstrated a fiber source based on a spectrally-compressed dissipative-soliton laser and a fiber OPO capable of producing the two picosecond pulse trains desired for CARS and SRS microscopy. This is the first demonstration that dissipative soliton lasers can provide a route to a Raman scattering microscopy source. In addition, the dissipative soliton laser can produce high energy femtosecond pulses at 1040 nm that are suitable for other imaging modalities, such as TPEF and SHG microscopy [20][21][22], and the wavelength tunability could be extended by using the pulses to pump a fiber OPO optimized for femtosecond pulse generation. Although the source is not yet quiet enough for SRS microscopy with direct detection, this device provides RIN levels comparable to the best achieved by fiber sources to date and could be further optimized for low-noise operation through the design of the laser and through energy scaling with large core fibers. A fiber source able to provide the correct pulse parameters for Raman and multiphoton microscopies could provide a tremendous cost advantage and extend the application of these techniques.