We demonstrate an alternative light source for CARS microspectroscopy based on a fiber laser and a photonic-crystal fiber. The light source simultaneously delivers a near-transform-limited picosecond pump pulse at 1033.5 nm and a frequency-shifted, near-transform-limited femtosecond Stokes pulse, tunable from 1033.5 nm to 1400 nm. This corresponds to a range 0 - 2500 cm-1, so that Raman-active vibrations in this frequency range can be probed. The spectral resolution is 5 cm-1, given by the spectral width of the pump pulse. The frequency range that can be probed simultaneously is 200 cm-1-wide, given by the spectral width of the Stokes pulse. The achievable average powers are 50 mW for the pump and 2 mW for the Stokes pulse. The repetition rate is 35 MHz. We demonstrate the capability of this light source by performing CARS microspectroscopy and comparing CARS spectra with Raman spectra.
©2007 Optical Society of America
Microscopic imaging based on coherent anti-Stokes Raman scattering (CARS) allows for chemically specific imaging, because the contrast is derived from molecular vibrations, which are interrogated by two laser pulses (“pump” and “Stokes”). The technique has undergone a major development since its re-discovery in 1999 . The employed pump and Stokes pulse can both be picosecond (CARS microscopy ) or picosecond and femtosecond (CARS microspectroscopy ). The pump and Stokes pulses drive Raman-active vibrations at the pump-Stokes frequency difference, νP – νS, generating a blueshifted anti-Stokes signal. The anti-Stokes signal is enhanced, when νP – νS is equal to the frequency of a Raman active vibration. CARS imaging has proven successful in the non-invasive imaging of living cells  and lipid membranes . As the potential of CARS imaging has unwound, so have efforts to reduce the complexity and cost of the light source. Many current light sources in CARS setups employ either two inter-locked lasers  or a picosecond laser and an optical parametric oscillator.  Although being virtually ideal light sources for the purpose, these options remain costly. Attempts to reduce cost and complexity have been made by developing light sources based on a single femtosecond laser [6, 7] and light sources based on photonic-crystal fibres [8, 9]. An obstacle that typically arises with single-laser approaches is that a femtosecond pulse is desired to generate a Stokes pulse through some nonlinear process with good efficiency, yet a picosecond pump pulse is desired to attain acceptable spectral resolution.
We present here a light source for CARS microspectroscopy capable of delivering simultaneously a near-transform-limited picosecond pulse and a tunable, near-transform-limited femtosecond pulse. The system is kept compact and uncostly through the use of a picosecond fiber laser and a photonic-crystal fiber (PCF) for frequency conversion. Similar tunable, fiber laser-based light sources have been presented previously in Ref.  and Ref. , in which both generated pulses were femtosecond. Our approach differs in that our fiber laser emits picosecond rather than femtosecond pulses.
2. SESAM-modelocked fiber laser
The light source in our setup relies on an ytterbium fiber laser mode-locked by a semiconductor saturable-absorber mirror (SESAM). , which was developed by us with the application for CARS in mind. The laser operates at 1033.5 nm with a repetition rate of 35 MHz. It provides two outputs, “output 1” and “output 2”, at average powers of approximately 4 mW (output 1) and 1 mW (output 2).
A diagram of the laser and the rest of the light source is shown in Fig. 1. The oscillator itself consists of a linear cavity made in polarization-maintaining (PM-) fibers (mode-field diameter (MFD) = 7 μm). A piece of PM-ytterbium-doped fiber (MFD 4.8 μm) pumped by a 300 mW laser diode at 976 nm in the cavity functions as the gain medium. A fiber-Bragg grating (FBG) whose reflected spectrum is 0.5 nm-wide and centered at 1033 nm functions as the first cavity end mirror. The grating stabilizes the laser central wavelength and the spectral width. The second cavity mirror is a semiconductor saturable-absorber mirror (SESAM), which modelocks the laser and assures self-starting of the mode-locking. A 20:80 polarizing tap coupler in the cavity functions as the output coupler and provides two outputs from the laser at average powers of approximately 4 mW (output 1) and 1 mW (output 2). The output wavelength is centered at 1033.5 nm and the spectral width is 0.5 nm.
3. Red-shifted Stokes pulse
Output 1 is used to create the Stokes pulse to be used later for our CARS microspectroscopy experiment (Bottom part of Fig. 1). For the Stokes pulse, we desire a spectrally broad pulse at a longer wavelength than the fundamental laser wavelength. The means of achieving this is as follows. An ytterbium fiber amplifier pumped by a 600 mW laser diode at 974 nm is spliced onto output 1. This amplifies output 1 up to 400 mW after the amplifier. A 5 metre-long piece of standard PM-fiber is spliced onto the amplifier. As the amplified pulse propagates inside this fiber, it undergoes self-phase modulation (SPM), the pulse leaving the fiber has broadened to 15 nm. Average power at this point is 300 mW due to splicing losses. This broadened pulse is then temporally compressed by a pair of transmission gratings (1200 lines/mm) in Littrow configuration separated by 30 mm. The second-order interferometric autocorrelation trace of the temporally compressed pulse depicted in Fig. 2 has a width (FWHM) of 268 fs, which corresponds to a pulse width (FWHM) of 190 fs Assuming a Gaussian pulse envelope. The time-bandwidth product is 1.06. It is not expected that the pulse can be compressed to the transform-limit because the chirp arising from SPM is only linear in the center of the pulse. In addition, the pulse accumulates a significant amount of higher-order dispersion in the 5 m-long fiber, which can not be compensated fully. The temporally compressed pulse is sufficiently short that we can send it through a PCF and generate nonlinear effects. We send the temporally compressed pulse through a 60 cm-long piece of PCF (Crystal Fibre A/S, Denmark), which has zero-dispersion wavelengths (ZDW)s at 770 and 1600 nm (dispersion data can be found in Ref. , where the present fiber is denoted “PCF1”) and a nonlinearity coefficient, γ, of 0.084 (Wm)-1. The laser wavelength of 1033.5 nm is thus in the anomalous dispersion regime of the PCF, where the group-velocity dispersion (GVD), β 2, is negative. Pulses propagating in this regime, or part of them, alter their shapes to converge into pulses known as “solitons”, which are stable under propagation if not perturbed. The fraction of the input pulse that goes into forming a soliton is determined by how well it overlaps with a soliton. A fundamental soliton obeys the relation N 2 = 0.325γP0T2fwhm/|β 2|, where N is the soliton order (N = 1 for a fundamental soliton), γ is the nonlinearity coefficient of the PCF, and P 0 is the peak power. If the spectrum of a soliton is sufficiently wide, it will interact with the Raman-active transitions in the silica fiber and thus be perturbed by intrapulse stimulated Raman scattering, where the blue spectral components act as a Raman pump for the red components. The phenomenon is known as “the soliton self-frequency shift” (SSFS) [14, 15]; the pulse redshifts adiabatically while maintaining its solitonic shape. We will use this redshifted pulse as the Stokes pulse. Although the PCF is not polarization maintaining, we find that the soliton part of the output spectrum is linearly polarized. Examples of spectra of redshifted solitons are shown in Fig. 3. In Fig. 4, we show the result of a measurement of the redshifted soliton center wavelength and average power of the soliton itself vs PCF input power. The coupling efficiency in this case was around 12 %. The redshift is almost linear with respect to PCF input power up until 1400 nm. At this wavelength the soliton power is also seen to drop. These two observations are accompanied by a narrowing of the soliton spectrum. We accredit this to the presence of OH- in the PCF which absorbs strongly at 1400 nm and perturbs the soliton strongly. In the absence of OH-, in principle, redshifts up to 1600 nm, the second ZDW should be possible. The tunability of the Stokes pulse makes it possible to probe vibrations at νP – νS = 0-2500 cm-1 by CARS spectroscopy. Another observation is that the formation of multiple solitons sets in, when the fundamental soliton has redshifted to 1300 nm. This is a consequence of the relation N 2 = (0.325γP0T2fwhm)/(|β 2|); for increased power, the soliton order also increases. The temporal duration of the redshifted soliton is expected to be close to the transform-limit. This is confirmed by second-order autocorrelation; an example trace for 1240 nm is shown in Fig. 3(b).
The temporal and spectral stability of the redshifted Stokes pulse should be considered. As is apparent from Fig. 4, the center wavelength of the redshifted pulses is coupled to the input power and - because of dispersion in the fiber - so is the delay of the pulses. A fluctuation in laser power therefore manifests itself as a fluctuation in wavelength and delay of the Stokes pulse. From the dispersion data , a fit to the wavelengths vs. input power data in Fig. 4 and the law of propagation of errors, we relate the standard deviation on the soliton center wavelength, σ(λS), and delay, σ(τS), to the standard deviation on the intrafiber average laser power, σ(P). We find
For typical parameters, λS= 1240 nm and σ(P)/P = 0.01, these numbers are σ(τS) = 100 fs and σ(λS) = 1 nm. The temporal standard deviation comes close to what can be achieved with inter-locked lasers - in Ref. , 21 fs was achieved. From this, we conclude that fast fluctuations are of little importance. There is still some long-term drift in the laser power, though. But this can in principle be compensated for either manually or by a feedback loop that adjusts the PCF input power. Employing SSFS allows us to generate a Stokes pulse that is spectrally separated from other contaminating spectral components in the PCF output. Although not done here, this allows for filtering out just the Stokes pulse with a longpass-filter so as to minimize the power load on the sample.
4. Pump pulse
Output 2 is employed to create the pump pulse for our CARS experiment. The spectral width and duration of output 2 suit our requirements for a pump pulse, so the task at hand is to amplify the pulse to a useful power. To this end, a non-PM ytterbium amplifier pumped by a 300 mW laser diode at 976 nm is employed (top portion of Fig 1). The unamplified power of 1 mW makes it possible to achieve amplification to 50 mW average power. A piece of non-PM standard single-mode fiber is spliced onto the amplifier, which serves only as a coarse means of adjusting the interpulse delay between the pump and Stokes pulses. The duration of the pump pulse is measured by second-order interferometric autocorrelation as shown in Fig. 5. FWHM of the trace is determined to be 5.8 ps, corresponding to a pulse FWHM of 4.1 ps assuming Gaussian pulse envelope. With a spectral FWHM of 0.5 nm, this gives a time-bandwidth product of 0.57. The pulse is thus slightly chirped due to dispersion in the fiber.
5. CARS microspectroscopy
The setup for CARS microspectroscopy is depicted in Fig. 6. The pump and Stokes pulses derived from the fiber laser-based light source are combined on a dichroic mirror and spatially overlapped. A delay stage in the pump arm is used to adjust the temporal overlap. The pulses are filtered by a 1000 nm-longpass filter and are focussed onto the sample by a 40x 0.65 NA microscope objective. The transmitted light is collected by another 20x 0.5 NA objective and sent into a polychromator (McPherson, model 218) after having passed through a 900 nm-shortpass filter, which separates out the CARS signal. A cooled CCD camera (Andor iDus) detects the CARS spectrum.
As a measure of the resolution of our spectrometer, we use the measured FWHM of a 632.8 nm He-Ne laser, which was determined to be 0.1 nm or 2.5 cm-1.
The CARS field, ECARS, generated in a sample with incident pump and Stokes fields, EP and ES, is composed of a resonant and a nonresonant part, Er and Enr.
Where χr (3) and χnr (3) are the resonant and nonresonant susceptibilities. The measured CARS intensity is
By measuring the nonresonant signal, Inr, separately, ICARS can be normalized,
And the resonant third-order susceptibility can be modelled as
χnr (3) can be modelled as a real number. 
As an example of CARS microspectroscopy with the present light source, we acquire a CARS spectrum of benzonitrile, which has a narrow, isolated resonance at 1599 cm-1 owing to the ring stretch vibration. The CARS signal from benzonitrile is acquired for 100 s, and the nonresonant signal is subsequently measured as the CARS signal from a glass plate with similar acquisition time. The results of the two measurements are shown in Fig. 7(a). Figure 7b shows the CARS spectrum normalized according to Eq. 7. The Raman spectrum of benzonitrile was acquired separately and fitted to Im[χ (3)] as shown in the inset of Fig. 7(b). The normalized CARS spectrum was fitted to a|χ (3)|2 + bRe[χ (3)]. A χ (3) with one resonance used in both cases. The result of the fit to the Raman spectrum was ν 0 = 1599 ± 2 cm-1 and Γ0 =4±2cm-1, and of the fit to the normalized CARS spectrum ν 0 = 1592 ± 3 cm-1 and Γ = 5 ± 1 cm-1. The CARS and Raman spectra agree within the uncertainties. We note the long acquisition time employed above and concede that, as such, our setup does not rival Raman spectrometers in terms of speed. Four partial explanations to the long acquisition time can be given. First, our polychromator and CCD are not optimized for the CARS wavelength at 880 nm, the overall efficiency is about 5 %. Second, the pump and Stokes wavelength employed here (1033 nm-1400 nm) are longer than the ones employed in previous studies (700 nm-1064 nm). It is well known from Raman spectroscopy that signal yield increases with shorter wavelength, the same applies for CARS. Third, due to nonresonant background signal in CARS, normalization of CARS spectra to the nonresonant signal is required in order to do a meaningful interpretation. For this, a good signal-to-noise level is nescessary, to obtain that, we have used a long acquisition time. Decent spectra can be obtained quicker as evidenced by the inset in Fig. 7(a). Last, pump and Stokes powers are limited in this study. Possible paths for improvements of powers will be discussed in the next section.
For our approach to be practical as a light source for CARS microspectroscopy, the pump and Stokes power need to be increased, and the Stokes frequency shift must be extended to permit imaging above 2500 cm-1 in the C-H stretch region of the spectrum.
The pump power can relatively easy be scaled up using picosecond fiber amplifiers. In  amplification of picosecond pulses at 33 MHz-repetition rate up to 700 mW was achieved using a pump-cladding fiber. A narrower pump spectrum than the one we have employed may be desired for some applications. In that case, pulse amplification combined with nonlinear spectral compression may prove useful . Using a fiber Bragg grating of higher quality than the one used here might also produce a smoother laser spectrum.
The power of the redshifted soliton is governed by the expression N 2 = (0.325γP0T2fwhm)/|β 2|, with N = 1 for the fundamental soliton being relevant here. It is seen that to achieve large P 0, the ratio γ/|β 2| should be small. We have attempted to produce the Stokes pulse in a PCF with larger core (NL-3-850, Crystal Fibre, Denmark) and hence smaller γ = 0.040 (Wm)-1. This yielded approximately a factor of 2 improvement in Stokes power, but the redshift was limited to 1500 cm-1 due to our limited laser power. To our knowledge, PCFs are not available with much lower γ/|β 2|-ratios than the ones we have employed. So, development in PCF technology is needed to reach higher Stokes powers. We are aware of a slightly different approach for generating high-power solitons - using hollow-core photonic-bandgap fibers (PBG)s filled with liquid, solitons with peak powers in the megawatt-range can be generated and redshifted . But in this approach, the redshift is limited by the transmission range of the hollow-core PBG.
The redshift of the Stokes pulse can be extended to beyond 2500 cm-1 provided the OH- absorption can be reduced, which is possible in newer PCFs .
We have demonstrated a light source for CARS microspectroscopy based on a fiber laser and a PCF, and we have demonstrated spectral resolution and stability in a CARS microspectroscopy experiment. This light source represents a major reduction of the cost of a light source for CARS microspectroscopy and may with further improvements develop into a useful alternative to bulk laser-based light sources.
This work was funded by the Danish Research Council for Technologies and Production Sciences and the Carlsberg Foundation. We thank Crystal Fibre A/S and Thomas Vestergaard Andersen for supplying the PCF and helpful discussions.
References and links
1. A. Zumbusch, G. R. Holtom, and X. Sunney Xie, “Three-dimensional vibrational imaging by Coherent Anti-Stokes Raman Scattering,” Phys. Rev. Lett. 82, 4142–4145 (1999). [CrossRef]
2. E. O. Potma, D. J. Jones, J.-X. Cheng, X. S. Xie, and J. Ye, “High-sensitivity coherent anti-Stokes Raman scattering microscopy with two tightly synchronized picosecond lasers,” Opt. Lett. 27, 1168–1170 (2002). [CrossRef]
3. M. Müller and J. M. Schins, “Imaging the thermodynamic state of Lipid membranes with Multiplex CARS Microscopy,” J. Phys. Chem. B 106, 3715–3723 (2002). [CrossRef]
4. J.-X. Cheng and X. Sunney Xie, “Coherent anti-stokes Raman Scattering Microscopy: Instrumentation, Theory, and Applications,” J. Phys. Chem. 108, 827–840 (2004). [CrossRef]
5. F. Ganikhanov, S. Carrasco, X. Sunney Xie, M. Katz, W. Seitz, and D. Kopf, “Broadly tunable dual-wavelength light source for coherent anti-Stokes Raman scattering microscopy,” Opt. Lett. 31, 1292–1294 (2006). [CrossRef] [PubMed]
6. T. Hellerer, A. M. K. Enejder, and A. Zumbusch, Appl. Phys. Lett. 85, 25–27 (2004). [CrossRef]
8. H. N. Paulsen, K. M. Hilligsøe, J. Thøgersen, Søren Rud Keiding, and J. J. Larsen, “Coherent anti-Stokes Raman scattering microscopy with a photonic crystal fiber based light source,” Opt. Lett. 28, 1123–1125 (2003). [CrossRef] [PubMed]
10. H. Lim, J. Buckley, A. Chong, and F. W. Wise, “Fibre-based source of femtosecond pulses tunable from 1.0 to 1.3 μm,” El. Lett. 40, 1523–1525 (2004). [CrossRef]
11. J. Takayanagi, T. Sugiura, M. Yoshida, and N. Nishizawa, “1.0-1.7-μm wavelength-tunable ultrashort-pulse generation using Femtosecond Yb-Doped Fiber Laser and Photonic Crystal Fiber,” IEEE Photon. Technol. Lett. 18, 2284–2286 (2006). [CrossRef]
12. R. Herda and O. G. Okhotnikov, “Dispersion compensation-free fiber laser mode-locked and stabilized by high-contrast saturable absorber mirror,” IEEE J. Quantum Electron. 40, 893–899 (2004). [CrossRef]
13. T. Schreiber, T. V. Andersen, D. Schimpf, J. Limpert, and A. Tünnermann, “Supercontinuum generation by femtosecond single and dual wavelength pumping in photonic crystal fibers with two zero dispersion wavelengths,” Opt. Express 13, 9556–9569 (2005). [CrossRef] [PubMed]
16. M. D. Levenson and S. S. Kano, Introduction to nonlinear laser spectroscopy (Academic Press, inc. 1987).
17. L. A. Gomes, L. Orsila, T. Jouhti, and O. G. Okhotnikov, “Picosecond SESAM-Based Ytterbium Mode-Locked Fiber Lasers,”, IEEE J. Quantum Electron. 10, 129–136 (2004). [CrossRef]
18. J. Limpert, N. Deguil-Robin, I. Manek-Hönninger, F. Salin, T. Schreiber, A. Liem, F. Röser, H. Zellmer, A. Tünnerman, A. Courjaud, C. Hönninger, and E. Mottay, “High-power picosecond fiber amplifier based on nonlinear spectral compression,” Opt. Lett. 30, 714–716 (2005). [CrossRef] [PubMed]
19. D. G. Ouzounov, F. R. Ahmad, D. Müller, N. Venkataraman, J. Silcox, K. W. Koch, and A. L. Gaeta, “Generation of Megawatt Optical Solitons in Hollow-Core Photonic Band-Gap Fibers,” Science 301, 1702–1704 (2003). [CrossRef] [PubMed]