A novel approach for an all-fiber mono-laser source for CARS microscopy is presented. An Yb-fiber laser generates 100 ps pulses, which later undergo narrowband in-fiber frequency conversion based on degenerate four-wave-mixing. The frequency conversion is optimized to access frequency shifts between 900 and 3200cm−1, relevant for vibrational imaging. Inherently synchronized pump and Stokes pulses are available at one fiber end, readily overlapped in space and time. The source is applied to CARS spectroscopy and microscopy experiments in the CH-stretching region around 3000cm−1. Due to its simplicity and maintenance-free operation, the laser scheme holds great potential for bio-medical applications outside laser laboratories.
©2012 Optical Society of America
Coherent anti-Stokes Raman scattering (CARS) microscopy has proven to be a powerful tool in biomedical sciences . By probing intrinsic vibrational molecule resonances of the specimen, chemical selective image contrast is obtained without the use of extrinsic labels . Paired with its high sensitivity, CARS thus allows for microscopy of living cells and is a promising technique for real time in-vivo imaging, e.g. during brain cancer surgery.
As CARS is a four-wave-mixing (FWM) process, the signal generation requires synchronized laser pulses at different wavelengths. This imposes stringent requirements on the light source. At the same time, the frequency difference between the two synchronized pulse trains needs to match the molecular resonance of interest to obtain the desired chemical-selective contrast through the generation of a resonant anti-Stokes signal. Since the signal intensity is proportional to the pump/probe and Stokes intensities ICARS~I2pIs, signal photons are primarily generated in a small focal volume, thus CARS additionally provides 3-dimensional sectioning capabilities.
The common approach to generate the pump and Stokes pulses is to use two synchronized mode-locked solid state lasers or one mode-locked laser in combination with an optical parametric oscillator (OPO), the latter approach thereby obviating the need for electronic synchronization. These systems are very versatile and deliver nearly ideal parameters. On the other hand, they are very expensive, large and require technical staff devoted to their constant alignment and maintenance. For this reason the application of CARS microscopy is still limited to a few specialized laboratories world-wide. A wider use in life sciences and especially in clinical environments critically depends on the development of suitable, easy-to-use, compact and, at the same time, inexpensive and reliable laser sources.
Fiber lasers are ideally suited to fulfill all these requirements, as they can be completely fiber integrated and, therefore, they can be extremely compact and robust. Hence, several sources intended for CARS microscopy have been developed using a fiber based driving laser in combination with some sort of frequency conversion process for the intrinsically synchronized generation of the frequency shifted pulse train. Different approaches for frequency conversion relying either on bulk crystal based OPO  or fiber based processes such as soliton self-frequency shift (SSFS) [4,5] or supercontinuum generation (SC)  have been pursued. Whereas a bulk OPO is clearly not compatible with fiber integration, also the fiber based approaches rely on free-space optics to some extent. One problem is that the frequency shifted solitons tend to be spectrally broad and, hence, they must be adapted to obtain high spectral resolution and a high contrast between the resonance and the non-resonant background in the CARS signal . Furthermore, it is beneficial for the frequency conversion process (SSFS, SC) if the driving laser pulse is in the fs time scale , whereas for CARS mostly ps pulses are preferable. One strategy to solve this issue is to start with a narrowband ps driving laser and convert part of its output to fs durations before it is sent to the frequency conversion stage . This is done by exploiting nonlinear spectral broadening in a fiber and by carrying out a subsequent temporal compression in a grating setup. Other approaches directly start with fs pulses and use second harmonic generation in long crystals to achieve spectral pulse compression at the end , or alternatively, they apply a spectrally focused CARS scheme by chirping the broadband pulses . Also more sophisticated techniques like delayed multiplex CARS with ultrabroadband few cycle pulses have been demonstrated . However, despite considerable progress, none of the presented concepts offers true turn-key and alignment-free operation as they depend on free-space optics and need adjustment of the pulse delay.
In this paper, we present a novel approach for all-fiber CARS laser sources which allows for complete fiber integration. Our system comprises a spectrally filtered fiber oscillator and a fiber optical parametric frequency conversion stage using degenerate four-wave-mixing (FWM) in an endlessly single-mode fiber. The frequency shift that can be obtained is determined by the fiber dispersion, which in turn influences the phase-matching condition of the parametric process. Moreover this frequency shift can be changed by tuning the wavelength of the driving fiber laser. Furthermore the FWM process inherently ensures that the pump laser pulse and the shifted component are both temporally and spatially overlapped at the fiber output. Thus they are readily available for injection into a CARS microscope without the need of any combining optics or delay lines. This reduces overall system complexity significantly and makes it particularly easy to use. Besides, the single-fiber-end output is directly compatible with fiber-delivered probes  for in-vivo imaging. This holds true, since long ps pulses have been used, for which the additional dispersion within the microscopic objective and the tissue sample and even within another short piece of delivery fiber is negligible.
Our experimental scheme is based on an alignment-free ytterbium fiber laser delivering 118 ps pulses at an average power of up to 2 W and with variable repetition rates in the lower MHz range. The FWM fiber was chosen to attain frequency shifts from more than 3500 cm−1 to below 1000 cm−1 by tuning the driving laser by +/−20nm within the Yb-gain bandwidth. In the following sections the pump laser (section 2) and the frequency conversion stage (section 3) will be described. Afterwards experimental results from CARS spectroscopy of the aromatic CH-stretching vibration of toluene and CARS microscopic images are presented in section 4.
2. The all-fiber pump laser system
To drive the nonlinear frequency conversion process, a reliable high-power pump laser is required. Hence, a fiber master oscillator power amplifier (MOPA) system, completely based on commercially available single-mode fiber components was developed. The exclusive use of polarization maintaining fibers results in an environmentally stable system. Furthermore, all components in the setup are fusion spliced to obtain alignment-free operation.
The fiber MOPA system is schematically depicted in Fig. 1 . We begin with a self-starting, fully fiber-integrated femtosecond oscillator, passively mode-locked by a semi-conductor saturable-absorber. The oscillator setup and performance is analog to the one presented in . Using a circulator and a narrowband (17pm) fiber Bragg grating (FBG), a small part of the broad oscillator spectrum is carved out, and amplified in two core pumped pre-amplifiers (Nufern PM-YSF-HI). A fiberized acousto-opical modulator then reduces the repetition rate to 1 MHz just before the cladding pumped main amplifier (Liekki Yb1200-10/125DC-PM), which can provide up to 2 W of output power. Due to the narrow spectral bandwidth, the pulses reflected by the grating possess a long duration in the temporal domain. At the output of the main amplifier we measure a 0.07 nm spectral width (Fig. 2(b) ) and a corresponding pulse duration of 118 ps (Fig. 2(a)), using an optical spectrum analyzer with 0.02 nm resolution and a 19 ps response time photodiode. These pulses, possessing μJ-level energies and peak powers in the several-kW range, are used for frequency conversion in the next section. They can, moreover, be wavelength-tuned by variation of the filter element, e.g. by stretching the FBG. In addition, simultaneous multiple wavelength operation or selectable discrete wavelengths using a fiber switch could be realized by application of several FBGs. Furthermore, we obtain chirped fs pulses at the transmission port of the FBG, which are not used further in this contribution, but which could be compressed as demonstrated in  and used for different nonlinear microscopy techniques such as two-photon fluorescence or second harmonic generation as additional methods providing image contrast.
3. In-fiber frequency conversion by degenerate four-wave-mixing
The wavelength range from 780 to 980 nm for pumping the CARS process is beneficial for microscopy applications as it provides a good trade-off in terms of spatial resolution, penetration depth, non-resonant background and photodamage . This wavelength range is therefore, the most widely used. The corresponding Stokes wavelength is, consequently, above 1000 nm. In our case, in order to exploit the advantages of Yb-fiber laser technology, the Stokes wavelength is designed to fall within the Yb-gain bandwidth between 1020 and 1080 nm. The all-fiber single-laser approach pursued here requires then, the frequency conversion of this Stokes wavelength to shorter wavelengths. The frequency-converted signal should be strong enough to pump the CARS process. It has been shown that degenerate FWM in photonic-crystal fibers can generate distinct new frequencies with high spectral densities [11–13]. Endlessly single-mode PCF designs allow for high conversion efficiencies as they ensure a good mode overlap even for widely separated wavelengths. The use of long picosecond pulses minimizes spectral broadening of the driving pulses induced by self-phase modulation (SPM) and thus allows for narrowband signal generation.
In degenerate FWM two pump photons are annihilated and one signal and one idler photon are generated at shorter and longer wavelengths, respectively. To avoid confusion, the terms pump and signal are written in italic in the following, wherever they refer to the FWM process. This should distinguish them from the pump and signal of the CARS process. (Note that the signal generated by FWM will be used as the pump for the CARS process. Moreover, the residual FWM pump pulse will serve as the CARS Stokes (see upper inset Fig. 4). The parametric process of FWM can only take place if both energy 2ωpump=ωsignal+ωidler and momentum conservation 2kpump= ksignal+ kidler + 2γPpump are fulfilled . The simultaneous fulfillment of these two conditions is known as phase-matching. Besides the product of nonlinearity parameter γ and pump power, the phase matching is only dependent on the dispersion of the fiber. Hence, the signal wavelength for a given pump can be conveniently selected by adjusting the structure parameters of the fiber. To select a suitable fiber we calculated the dispersion of the effective refractive index for different available PCFs (see legend in Fig. 3 ) using a numerical mode solver. Together with the equations above, the phase matched wavelength sets are easily determined for each fiber. Figure 4 displays the calculated mode profile for PCF#2 and the corresponding signal (orange) and idler (black) wavelength pairs for pump wavelengths between 1 and 1.1 μm. Additionally, the frequency difference (green) between signal and pump is plotted. This shows whether the generated signal could be used together with the pump laser to drive any CARS resonance of interest. In fact, for PCF#2 tuning the pump between 1022 and 1055 nm generates the desired signal wavelengths between 770 and 964 nm. This corresponds to frequency shifts between 3200 and 900 cm−1. Thus tuning the pump by only 33 nm gives access to the whole frequency range important for CARS microscopy. Of course, the same frequency difference is obtained with the idler, which could alternatively be used whenever longer wavelengths are of interest.
The investigated PCFs are one-hole-missing designs consisting of 7 rings of air holes. A microscope image of the fiber cross section is shown in Fig. 3(b), the design parameters (hole-to-hole distance and size) are given in the legend. The simulated frequency shifts obtained for the different fibers assuming an Yb-doped pump laser are shown in Fig. 3(a). At fixed pump wavelength of 1030 nm, the frequency shifts span from 1528 to 4340 cm−1 with the presented fiber set, demonstrating the flexibility of the DFWM approach. We decided to use PCF#2 for our CARS experiments as it shifts across the important CH-stretching bands around 3000 cm−1 when using the pump laser around 1030 nm. Furthermore, as discussed above, this fiber perfectly maps the CARS relevant frequency range into a tuning range easily accessible with Yb-fiber lasers. The fiber’s mode-field area is around 14 μm2, which gives rise to sufficiently high intensities already at moderate peak powers of some kW. For comparison, the corresponding curve for a commercially available fiber (LMA-5 NKT Photonics) is plotted (gray). It shows nearly the same characteristic and, thus, it could readily be used for the development of a low cost commercial CARS laser source.
As the phase matching is strongly related to the fiber parameters, any geometrical variations along the fiber length will reduce the efficiency and spectral signal purity. It is therefore beneficial to use short fibers to favor the FWM process over other competing nonlinear effects such as spontaneous Raman scattering . A very short fiber on the other hand, would require high pump powers to obtain moderate conversion efficiencies. We determined that the optimum fiber length was 0.7 m for our purposes. This length allows for a clean signal and idler generation as shown in Fig. 5(a) for a pump power of 0.37 W. The fiber is pumped at 1033 nm generating a signal at 785 nm, which corresponds to a frequency shift of ~3050 cm−1, matching the aromatic CH-stretching vibrational resonance.
In reference  it has been shown that the temporal pulse shape gets increasingly complex due to the occurrence of successive back-conversion processes in longer fiber sections. On the other hand, using a short conversion fiber, clean signal pulses can be generated, at the price of sacrificing some conversion efficiency. These signal pulses can be significantly shorter than the pump pulses, as conversion starts only around the pulse peak. As we are interested in clean signal generation, we chose such a working point. The temporal trace of the signal recorded with a photodiode (18.5 ps) and a sampling oscilloscope (70 GHz) is shown in Fig. 6(b) . It has a full width at half maximum of 43 ps, which is about 3 times shorter than that of the pump pulse. More important, however, is the spectral width of the generated signal, as it influences the signal to noise ratio of the CARS signal due to the nonresonant background . Figure 5(b) shows the signal spectrum for different signal powers (obtained by increasing the pump power). With increasing power the spectral power density increases up to 15 mW/nm for 48 mW total signal power. However, as the signal width gets broader, the power density does not increase further when going to higher power. For our CARS experiments we worked with 20 to 35 mW frequency converted power. If higher power would be required, one would have to use higher repetition rates or shorter conversion fiber lengths to remain in the narrow linewidth operation point.
As mentioned above, by tuning the wavelength of the pump laser the frequency shift can be varied over a large range, whereby both pump and signal wavelengths change. A continuous tunability over a few nanometers can be obtained from our pump laser by stretching the FBG. The resulting FWM signal spectra in terms of frequency shift are shown in Fig. 6(a) for PCF#2. The signal wavelength changes from 775 to 788 nm when the pump is tuned from 1030.5 to 1033.5 nm. So the frequency shift changes by ~200 cm−1 with this 3 nm pump tuning.
4. Application of the two-color fiber source to CARS spectroscopy and microscopy
To demonstrate the suitability of the above concept for CARS applications, we performed a measurement of the aromatic CH-stretching vibrational resonance of toluene, using the simple scheme depicted in Fig. 7 . After the pump laser being frequency shifted by DFWM in PCF#2, the output of the PCF is simply collimated and focused into the sample. No additional delay line is required as the pulses already overlap in space and time at the output of the fiber, which simplifies both the setup and the experimental effort, something of particular interest for the user. Behind the sample, the light is collected by a lens and coupled into a multimode fiber, which is connected to the detector. A 770nm short pass filter is inserted behind the sample to block any residual pump light. Another long pass filter (650nm) is used in front of the sample to block any small amount of light that might be present at the CARS signal wavelength, generated by non-phase-matched FWM in the PCF .
About 26 mW of CARS pump is sent to a 10x microscope objective (Olympus) to be focused into the sample. The signal is detected with a CCD spectrometer, but it could of course be obtained by any non-spectrally sensitive detector as well. As the signal is strong, no special care had to be taken regarding efficient photon collection or stray light. Figure 8(a) shows the CARS signal spectra for different frequency shifts between pump and Stokes. The width of each single spectrum is about 50 cm−1. Integration yields the total signal power, which is plotted in Fig. 8(b) together with a fit of the resonance line. The resulting linewidth is 60 cm−1 as a result of the convolution of the laser linewidth with the resonance.
In order to test the imaging capabilities of the laser source, we attached it to a home-built laser scanning microscope. Using a 20x objective (Mitutoyo Plan Apo NIR), images of glass spheres in toluene were taken (Fig. 9 ). Tuning to the resonance (left hand side of Fig. 9) the surrounding toluene generates a strong resonant CARS signal and hence appears bright, whereas the glass spheres remain dark creating a strong contrast in the picture. The raw images in Fig. 9 contain 970x970 pixels and were acquired in less than 2 seconds without averaging. The series in Fig. 9 shows how the signal vanishes when the laser is tuned out of resonance. The nonresonant background is in comparison to the resonant CARS signal almost negligible (compare Fig. 9 left-right).
Against the common notion that pulses in the short ps range are optimal for CARS, longer pulses in the range of tens of picoseconds have many practical advantages. Obviously the temporal overlap between pulses becomes much less critical and, thus, passing several optical elements or even a certain fiber segment does not require for any readjustments. Moreover, very narrow bandwidths and high spectral resolution is obtained for transform limited pulses at durations of several tens of ps, which, at the same time results in higher signal to noise ratios as the nonresonant background coming from besides the excited vibrational line is suppressed. Finally, and even more important, when it comes to fiber delivered probes, the impact of undesired propagation effects is drastically reduced. This relates to both dispersion and self-phase-modulation, when comparing pulses with equal peak power. At the same time longer pulses generate the same amount of CARS signal compared to shorter ones, as long as the peak and average power are kept constant. This implies that the product of the repetition rate and the pulse duration remains constant as illustrated in Fig. 10 . Thus, when the pulse duration is increased, the pulse energy needs to be increased by the same factor to keep the peak power constant. In order to remain at the same average power, consequently, the repetition rate has to be decreased by that factor. This increase in pulse energy is inherently happening in the amplifier if the repetition rate is decreased. Under this prerequisite, each long pulse generates more CARS signal even though the total CARS average power remains constant. Hence, integration over several pulses for each pixel is not required anymore. On the other hand, fast acquisition times and real-time imaging require high repetition rates, thus repetition rates around few MHz represent a good trade-off. For example, a 1-MHz system would allow for an image of 300x300pixels at 11 frames per second.
We have demonstrated a novel all-fiber laser source approach for CARS microscopy based on degenerate FWM. An environmentally stable Yb-fiber laser generates 118 ps pulses around 1030 nm, which are then frequency converted in a PCF to obtain high-power, inherently synchronized CARS pump pulses with a duration of 43 ps. The generated wavelength pair around 780 nm and 1030 nm was used to probe the aromatic CH-stretch vibrational resonance of toluene at 3050 cm−1. Reproduction of the CARS resonance bandshape by spectral tuning as well as microscopic imaging within the aromatic CH-stretching region demonstrates the CARS capabilities of the laser.
In contrast to other fiber based frequency conversion arrangements on the basis of supercontinuum generation or Raman scattering based soliton shifts, FWM gives the opportunity for narrowband frequency conversion directly to shorter wavelengths. Using the emission of an Yb-fiber laser as CARS Stokes and as the field driving the FWM process, high-power CARS pump pulses can be generated in the preferred wavelength region around 800 nm. Alternatively the FWM idler around 1500 nm could be used as Stokes, whenever longer wavelengths would be required. Furthermore the FWM process is driven by long ps pulses and is consequently directly compatible with CARS, without the need for pulse post-processing like in other fs-based frequency conversion schemes.
The laser operates in the near infrared wavelength range offering several advantages as a maximized depth penetration and minimal sample absorption and photo damage in biological tissue. The long ps pulses minimize the influence of dispersion and do not loose temporal overlap. Furthermore the spectral resolution of the source with an excitation bandwidth of ~26 cm−1 as demonstrated corresponds to the linewidth of a vibrational resonance in the CH-stretching region (~20cm−1) for imaging lipids, thus, yielding a high ratio of resonant signal to nonresonant background. The resolution could be further improved by cw-seeding of the FWM process , which would condense the bandwidth closer to the transform limit. This would reduce the bandwidth to few cm−1.
The main advantage of the proposed approach, however, is its simplicity of operation due to its full fiber integration. It provides pump and Stokes pulses temporally and spatially overlapped from one fiber end in combination with a compact and maintenance-free overall system. Hence, it paves the way to fiber-delivered in situ imaging systems.
By replacing the fs- by a ps-oscillator with low repetition rate the system could be simplified even further. Only one core pumped main amplifier would be required which could be directly spliced between the oscillator and the conversion PCF. Also a high repetition rate fiber coupled microchip laser  might be an alternative. With the whole system fitting into a small box, it could directly be attached to the microscope, thus becoming as easy to use as any halogen lamp. It could be made tunable by tuning the oscillator, but it could also be designed with a fixed frequency difference to serve one specific application as e.g. the imaging of lipids. As such, it could be a simple extension to existing fluorescence microscopes or a compact and reliable device for in situ imaging, thus being a significant step towards establishing CARS microscopy in real-world and, in particular, in bio-medical applications.
This work was partly supported by the German Federal Ministry of Education and Research (BMBF) under contract 13N10773 and 13N10774 as well as the Inter Carnot & Fraunhofer program under the project APUS. M. Baumgartl acknowledges support from the Carl-Zeiss-Stiftung. The authors thank T. V. Andersen from NKT Photonics for providing the photonic-crystal fibers.
References and links
1. C. L. Evans and X. S. Xie, “Coherent anti-stokes Raman scattering microscopy: chemical imaging for biology and medicine,” Annu Rev Anal Chem (Palo Alto Calif) 1(1), 883–909 (2008). [CrossRef] [PubMed]
2. A. Zumbusch, G. R. Holtom, and X. S. Xie, “Three-Dimensional Vibrational Imaging by Coherent Anti-Stokes Raman Scattering,” Phys. Rev. Lett. 82(20), 4142–4145 (1999). [CrossRef]
4. E. R. Andresen, C. K. Nielsen, J. Thøgersen, and S. R. Keiding, “Fiber laser-based light source for coherent anti-Stokes Raman scattering microspectroscopy,” Opt. Express 15(8), 4848–4856 (2007). [CrossRef] [PubMed]
5. A. F. Pegoraro, A. Ridsdale, D. J. Moffatt, J. P. Pezacki, B. K. Thomas, L. Fu, L. Dong, M. E. Fermann, and A. Stolow, “All-fiber CARS microscopy of live cells,” Opt. Express 17(23), 20700–20706 (2009). [CrossRef] [PubMed]
6. M. Marangoni, A. Gambetta, C. Manzoni, V. Kumar, R. Ramponi, and G. Cerullo, “Fiber-format CARS spectroscopy by spectral compression of femtosecond pulses from a single laser oscillator,” Opt. Lett. 34(21), 3262–3264 (2009). [CrossRef] [PubMed]
7. K. L. Corwin, N. R. Newbury, J. M. Dudley, S. Coen, S. A. Diddams, K. Weber, and R. S. Windeler, “Fundamental noise limitations to supercontinuum generation in microstructure fiber,” Phys. Rev. Lett. 90(11), 113904 (2003). [CrossRef] [PubMed]
8. R. Selm, M. Winterhalder, A. Zumbusch, G. Krauss, T. Hanke, A. Sell, and A. Leitenstorfer, “Ultrabroadband background-free coherent anti-Stokes Raman scattering microscopy based on a compact Er:fiber laser system,” Opt. Lett. 35(19), 3282–3284 (2010). [CrossRef] [PubMed]
10. B. Ortaς, M. Plötner, T. Schreiber, J. Limpert, and A. Tünnermann, “Experimental and numerical study of pulse dynamics in positive net-cavity dispersion modelocked Yb-doped fiber lasers,” Opt. Express 15, 15595–15602 (2007).
11. W. Wadsworth, N. Joly, J. Knight, T. Birks, F. Biancalana, and P. Russell, “Supercontinuum and four-wave mixing with Q-switched pulses in endlessly single-mode photonic crystal fibres,” Opt. Express 12(2), 299–309 (2004). [CrossRef] [PubMed]
12. D. Nodop, C. Jauregui, D. Schimpf, J. Limpert, and A. Tünnermann, “Efficient high-power generation of visible and mid-infrared light by degenerate four-wave-mixing in a large-mode-area photonic-crystal fiber,” Opt. Lett. 34(22), 3499–3501 (2009). [CrossRef] [PubMed]
13. L. Lavoute, J. C. Knight, P. Dupriez, and W. J. Wadsworth, “High power red and near-IR generation using four wave mixing in all integrated fibre laser systems,” Opt. Express 18(15), 16193–16205 (2010). [CrossRef] [PubMed]
14. G. P. Agrawal, Nonlinear Fiber Optics (Academic, 2001).
15. P. J. Mosley, S. A. Bateman, L. Lavoute, and W. J. Wadsworth, “Low-noise, high-brightness, tunable source of picosecond pulsed light in the near-infrared and visible,” Opt. Express 19(25), 25337–25345 (2011). [CrossRef]
16. A. Steinmetz, D. Nodop, A. Martin, J. Limpert, and A. Tünnermann, “Reduction of timing jitter in passively Q-switched microchip lasers using self-injection seeding,” Opt. Lett. 35(17), 2885–2887 (2010). [CrossRef] [PubMed]