1. Introduction

Coherent anti-Stokes Raman scattering (CARS) microscopy is a potent spectroscopic technique for probing vibrational resonances without the need for bio-hazardous contrast materials [1]. Using CARS provides a better resolution and results in lower absorption than direct spectroscopy in the mid-IR. Coupled with high-sensitivity scanning, it may open the way for spectrally resolved real-time in vivo images of living tissue, e.g. during surgery [2].

The CARS process is based on a four-photon interaction. If the frequency difference of the pump and Stokes laser matches a vibrational frequency, simultaneous illumination selectively populates this level and further probe photons are inelastically scattered off this vibrational state generating emission at the CARS frequency (ω_CARS = ω_probe + ω_Pump – ω_Stokes, angular frequency). The ensemble is excited in phase, thus the generation of the CARS signal is coherent to the probe laser. Since the emitted photons have the highest frequency in the process, it can be easily filtered and detected. The intensity of the CARS signal depends on the product of the intensities of the pump, Stokes and probe signals (I_CARS~I_probe·I_pump·I_stokes). Due to this nonlinear interaction only a small focal volume contributes to the CARS signal, providing CARS with its 3-dimensional sectioning capabilities.

Due to the nonlinearity of the process and an anticipated application in the life sciences, a laser source for CARS should possess the following features:

Detecting more than one resonance to identify different substances without changing the laser setup is an additional desirable but challenging feature. This so-called multiplex-CARS (MCARS) demands an increased spectral bandwidth of the pump and/or Stokes laser [3]. In contrast, the probe spectrum has to be spectrally narrow to obtain a high spectral resolution of the CARS signal and so to be able to differentiate between adjacent molecular resonances. This creates a high demand on the CARS laser source.

Today, typical laser sources for CARS employ two actively synchronized mode-locked solid-state lasers or one solid-state laser where the other wavelength is generated via parametric frequency conversion (optical parametric oscillators or amplifiers) in bulk crystals. Therefore, the pulses are inherently synchronized to each other. These bulk systems are very versatile and have nearly ideal parameters for the CARS process. Unfortunately, these sources are quite large, expensive and require trained staff to be maintained and aligned. However, in order to allow for a wider spectrum of researchers to have access to this technology and to enable the medical community to use CARS on a day-to-day basis as a reliable diagnosis/analysis tool, there is a great demand for an easy-to-use, compact and, at the same time, inexpensive and reliable solution.

Fiber lasers can potentially meet those requirements. Through design and integration of the laser system with fiber-based components, the need for robustness, compactness and cost efficiency can be fulfilled. On top of that, new features and possibilities constantly arise from novel fiber designs and concepts [4]. Features like soliton self-frequency shift (SSFS) [5,6], supercontinuum generation (SC) [7] and four wave mixing [8–10] offer frequency conversion schemes unique to fibers. Frequency conversion methods like the optical parametric oscillator rely on free-space propagation with the well-known disadvantages [11].

Without post-processing, the spectra of SSFS are broad, hence reducing the spectral resolution and the contrast between the resonance and the non-resonant background in the CARS signal. SC and SSFS are usually driven by pulses with durations in the femtosecond range [12], whereas usually picosecond pulses are preferable for CARS. To circumvent this issue, spectral compression techniques have been used to convert supercontinuum radiation to picosecond durations via second harmonic generation in a periodically poled lithium niobate (PPLN) crystal [13]. Very recently, a photonic-crystal fiber (PCFs) has been employed to generate the wavelength pair required for CARS by degenerate FWM [13]. By tuning the FWM-pump wavelength by just a few nanometers, a scan over a single vibrational resonance in toluene has been demonstrated in an all-fiber setup, in which both the pump and Stokes signals were delivered from a single fiber end.

In this contribution, we present an approach related to that in [14], however, extended to multiplex-CARS. This approach has the ability to excite and detect resonances between 1200 cm⁻¹ and 3200 cm⁻¹ simultaneously with a single fiber-based laser source. The system is driven by a fiber laser emitting several µJ pulse energy of nearly transform-limited 100 ps pulses at 1036 nm wavelength and 0.5 MHz repetition rate. The output is launched into two different large-mode-area PCFs with appropriate phase-matching characteristics. This way, spectral components are efficiently generated at 648 nm and between 700 nm and 850 nm via degenerated four wave mixing [15]. The generated pulses are later spatially and temporally overlapped to drive the CARS process.

The manuscript is organized as follows: Section 2 is devoted to presenting the experimental setup. Section 3 discusses the fiber based frequency conversion approach and its spectral and temporal characterization. Finally, section 4 features the experimental results achieved with this source including its application to multiplex-CARS.

2. The ytterbium-doped fiber pump laser system

In order to generate sufficient peak power to drive both fiber-based frequency conversion processes at MHz repetition rates, a master oscillator power amplifier concept has been realized. A passively mode-locked femtosecond all-fiber oscillator serves as seed laser, followed by an acousto-optic modulator (AOM) to reduce the pulse repetition rate. A key component is a narrow-band fiber Bragg grating (FBG) acting as spectral filter, which determines the final pulse duration of the pump source [16]. We have chosen a pulse duration of 100 ps as it falls into an optimal range for fiber FWM based frequency conversion [9]. The system, as seen in Fig. 1 , consists of spliced polarization-maintaining (PM) fibers up to the third amplifier and it is, therefore, alignment-free.

 

Fig. 1 Schematic setup of the fiber laser system employed to drive two four-wave-mixing processes. ISO: optical isolator, AOM: acousto-optic modulator, PM FBG: polarization-maintaining fiber bragg grating, HWP: half-wave plate, CIRC: circulator.

Download Full Size | PDF

The system is driven by a self-starting, passively mode-locked femtosecond fiber oscillator delivering a 25 MHz pulse train of chirped pulses with 7 ps duration. The spectral bandwidth is 11 nm (FWHM), corresponding to a transform-limited duration of about 200 fs. The oscillator setup and performance is similar to the one presented in [17]. To avoid detrimental non-linear effects along the amplifier stages, 20 m of PM fiber are inserted to stretch the pulses to a measured duration of 24 ps. The pulses are amplified in a step-index single-clad ytterbium-doped fiber (referred to as Yb Amp in Fig. 1). A fiber-coupled AOM is used to reduce the repetition rate of the laser to 0.5 MHz. This repetition rate has been chosen to reach peak intensities high enough to efficiently drive the FWM process while still ensuring high fluxes of the CARS signal. Using a PM-circulator and a narrowband fiber Bragg grating (FBG), a narrow spectral part of 17 pm at around 1036 nm wavelength is filtered to create transform-limited 106 ps pulses (see Fig. 2 ). Before and after the spectral filtering, the pulses pass through another Yb-fiber amplifier in double-pass configuration to compensate for the losses introduced by the circulator and the spectral filtering (2nd amplifier) [16]. A PM-polarizer ensures that no spectral side bands are excited, which can potentially arise from light coupled to the wrong polarization axis of the PM fibers. The PM-circulator directs the spectrally filtered and therefore temporally stretched pulses to the 3rd core-pumped ytterbium fiber amplifier. Up to this point, the setup is all-fiber making it very compact and stable. The collimated output of the 3rd core-pumped fiber amplifier is optically isolated and send through a transmission grating (1200 lines/mm) to filter out amplified spontaneous emission (ASE). The main amplifier consists of a double-clad PCF possessing an ytterbium-doped core of 40 µm diameter and a pump cladding of 170 µm diameter. Thanks to the prefiltering of the ASE a spectral contrast of 55 dB signal-to-noise is achieved after amplification of the pulses to up to 14 µJ of pulse energy (see Fig. 2b) at 0.5 MHz repetition rate. At this power level a spectral broadening of the pulses to 98 pm is observed. The pulses possess a peak power in excess of 120 kW, providing sufficient power to drive the fiber based frequency conversion described in the next section. Finally, the output is split by means of a half-wave plate (HWP) and a polarization cube allowing for full control of the power ratio sent to the two frequency-converter fibers (PCF 1 and 2).

 

Fig. 2 Output characteristics of the amplified ps pulses. (a) Temporal pulse shape measured with a cross-correlation setup (black) together with a Gaussian fit (red). (b) Optical spectrum at 14µJ pulse energy and 0.5MHz repetition rate (black, resolution of the optical spectrum analyzer: 0.02nm) and the output spectrum of the femtosecond oscillator (gray).

Download Full Size | PDF

3. In-fiber frequency conversion by degenerate four-wave-mixing (DFWM)

In previous work degenerate FWM in photonic-crystal fibers has been used to generate distinct new frequencies with high spectral densities [8–10]. Two pump photons are converted into a signal and an idler photon obeying energy conservation (2ω_pump = ω_idler + ω_signal), hence, a distinct pair of new frequencies can be created from quantum noise. The precondition for an efficient conversion is momentum conservation (also referred to as phase-matching) 2k_pump = k_signal + k_idler + 2γP_pump where γ is the nonlinearity parameter, P_pump the input (pump) peak power and k denotes the absolute value of the corresponding wave vectors. In order to reach high conversion efficiencies, a good overlap of the propagating modes at all interacting wavelengths has to be ensured. This condition is satisfied in so-called endlessly single-mode photonic crystal fibers, even in large-mode area configurations [9].

As mentioned above, a pulse duration in the order of 100 ps has been chosen in the described experiments. This pulse duration appears as a good compromise to minimize the effects of self-phase modulation (SPM), which is dominant at shorter pulse durations and Raman/Brillion scattering, known to become severe at longer pulse durations [12]. Since the spectral resolution of the CARS signal depends on the spectral width of the probe, it is also favorable to use narrow band ps-pulses, if their peak power is high enough. Due to the dispersion in the fiber, which can be tailored by means of the photonic structure, different phase-matched wavelength pairs can be expected in different PCFs. Moreover, the slope of the mismatch at the phase-matched frequencies can be different, thus, resulting in different FWM-gain bandwidths. Figure 3 summarizes the properties of the considered PCFs (nomenclature by NKT photonics, RD5: custom design) as well as their expected phase-matched FWM-signal wavelengths. Furthermore, the calculated spectral characteristics of the parametric gain in those fibers are shown. The simulation was performed by using the poor man’s approach [18] to calculate the effective index of the fiber. Afterwards the phase matching curve with the specific pitch Λ and hole diameter d of the fiber was calculated.

 

Fig. 3 The spectral characteristics of the parametric gain in different LMA-PCF fibers when pumped with linear polarized 1036 nm radiation with 20 kW of peak power. The table summarizes the central FWM-signal wavelengths and, furthermore, the group velocity difference between the signal and the pump pulse while propagating in the specific fiber.

Download Full Size | PDF

A tradeoff has to be met for the optimal PCF to create the pump radiation for CARS. First, the fiber should possess a narrow gain bandwidth, on the other hand, the generated wavelength should not be too close to the near-infrared spectral region to avoid a detrimental increase of the pump/probe bandwidth. In this case the contrast of the CARS signal to the non-resonant FWM signal will decrease if the Raman signal linewidth is surpassed.

As the phase-matching is strongly related to the fiber parameters, any geometrical variations along the fiber length will reduce the efficiency and the spectral purity of the generated signal. Therefore, it is beneficial to use short fibers to favor the FWM process with respect to other competing nonlinear effects such as stimulated Raman scattering [12]. On the other hand, a very short fiber would require high pump peak powers to obtain reasonable conversion efficiencies. In the herein presented experiments we have chosen a length of 0.5m of the LMA-8 (PCF 1) fiber mentioned above for narrowband CARS-pump generation. At this fiber length a pulse energy of 320 nJ at 648 nm has been obtained at an input pulse energy of 1.8 µJ with a FWHM-bandwidth of less than 10 cm⁻¹ (0.42 nm). The corresponding conversion efficiency is as high as 16%. Figure 4 shows the measured output pulse energy in the visible and the spectral bandwidth as a function of the overall transmitted energy.

 

Fig. 4 Conversion of the pulse energy to 648 nm in 50 cm LMA-PCF and the spectral bandwidth of the FWM-signal.

Download Full Size | PDF

In [9] it has been shown that the temporal pulse shape becomes more and more complex due to the occurrence of successive back-conversion processes in longer fiber sections or at higher input peak powers.

On the other hand, using a short conversion fiber or moderate pulse energies, clean signal pulses can be generated, at the price of sacrificing conversion efficiency. These signal pulses can be significantly shorter than the pump pulses, as the conversion is only efficient around the pulse peak. Figure 5b illustrates this behavior by showing cross-correlation measurements of the FWM-pump and –signal pulses at different energy levels. Due to a strong walk-off between the pump pulse and the created signal pulse (see the table in Fig. 5b) a shift of the conversion to the trailing edge of the pump pulse is revealed. According to the group velocity difference of the pulses involved during the FWM process in the LMA-8 fiber the walk-off is as high as 35 ps/m. Depending on this value and the input pulse duration of the pump, a shortening of the signal pulse duration is possible. At 1.8 µJ of pump energy the signal pulses have a duration of 65 ps at 320 nJ of pulse energy which translates to a peak power of 5.5 kW.

 

Fig. 5 (a) Trace-by-trace normalized cross-correlation time traces of the pump pulses of the FWM process (at 1036nm) with increasing output energies, showing the depletion of the pump pulses. (b) Trace-by-trace normalized time traces of the degenerated FWM signal (at 648 nm) with increasing output energies, showing the evolution caused by the temporal walk-off between the driving pump pulses and the converted signal pulses.

Download Full Size | PDF

To create the broadband multiplex-CARS-Stokes wave, 1m of PCF with a hole-to-hole distance of Λ = 4 µm and a hole diameter of d = 2 µm, referred to as RD5 (PCF 2), has been employed. A narrow-band conversion to 714 nm is obtained at 0.8 µJ. Above this energy a Raman-seeded parametric conversion shifts a portion of the energy to longer wavelengths filling the spectral gap between the pump wavelength at 1036 nm and the phase-matched FWM-signal wavelength. The continuum (see Fig. 6 ) has a high spectral energy density of 4 nJ/nm (at 800 nm wavelength, 2.4 µJ FWM-pump energy), which corresponds to a peak power density of >40 W/nm assuming a pulse duration below 100 ps in this spectral band. This assumption can be done since the conversion mechanisms in play are on a much shorter time scale than the pulse length.

 

Fig. 6 SC generation based on FWM in 1m RD5 fiber as a function of input pulse energy.

Download Full Size | PDF

4. Application of the fiber source to CARS spectroscopy

To demonstrate the multiplex CARS capabilities of the setup a preliminary not optimized experiment was conducted. A CARS spectrum was measured by placing a sample of toluene into the overlapped and focused beams of the pump/probe and Stokes signal (Fig. 7 ). Highly reflective (HR) mirrors in both branches filtered the residual pump from the frequencies generated via FWM.

 

Fig. 7 Super continuum generation based on FWM as a function of the overall output pulse energy. HR 1036 nm: highly reflective mirror at 1036 nm, LP 650 nm: long pass filter at 650nm, SP 650 nm/SP 600 nm: short pass filter at 650 nm/600 nm, OSA: optical spectrum analyzer.

Download Full Size | PDF

A HR mirror at 650 nm wavelength was used to overlap the pump/probe beam with the Stokes beam. A long-pass filter at 650 nm removes undesired frequencies originating from the SC generation. A delay stage was used to overlap the pulses in time. Additionally, both beams were focused with 10 × to ensure that every frequency component of the SC overlaps with the pump/probe beam in the focus where the test sample was placed. The generated CARS signal was selected by two short-pass filters at 650 nm and 600 nm. The CARS spectrum ranged from 540 to 605 nm corresponding to a frequency difference between pump and Stokes from 1200 cm⁻¹ to 3200 cm⁻¹, as depicted in Fig. 8 . The experiment was performed at 0.5 MHz using an overall average power in the super-continuum of 1.7 W and 150 mW of average pump power at 648 nm. The CARS signal was then filtered and focused into a multimode fiber to deliver the light to an optical spectrum analyzer.

 

Fig. 8 Measured CARS spectrum of toluene with identified resonances.

Download Full Size | PDF

The poor contrast of the CARS spectrum is due to the presence of nonresonant background radiation. Nevertheless the aromatic C-H vibrational resonances near 3000 cm⁻¹ generate a high response and possess a good contrast, while weaker resonances in the so called fingerprint region show a reduced visibility. Thus, in order to detect certain substances in biological samples it is necessary to apply numerical algorithms to extract the composition of the specimen from the MCARS spectrum (e.g. Kramers-Kronig or maximum entropy [19,20]). To further improve the capabilities of the CARS-source further and to increase the visibility of the Raman resonances different measures may be taken. Other samples of acetone and mixtures of acetone and toluene were measured showing the main resonances of both substances in the region of 3000cm⁻¹. These results are not presented in this paper since these measurements were taken with a different conversion fiber to create the Stokes wavelength.

To address smaller molecular vibrational energies and to cover the whole fingerprint region, a different combination of fibers could be necessary. The first possibility is to use two LMA-8 fibers. One to create the narrow band pump/probe radiation and the other one for supercontiuum generation ranging from 648 nm to >800 nm. This would require much more pump energy and/or fiber lengths in the supercontinuum part to achieve sufficient broadening. Another possibility is to use two LMA-5 fibers in the same configuration which would relax the demand for pump peak power but also decrease the spectral resolution of the CARS process since the probe pulses would become wider. A custom LMA fiber design with a DFWM phase matching wavelength at 690 nm (pumped with 1036 nm radiation) would enable excitation of molecular vibrational energies from 940 cm⁻¹ to 3200 cm⁻¹. To reduce the influence of the nonresonant background radiation while maintaining the multiplex capabilities of the source, the pump and Stokes beams could be created at longer wavelengths. Additionally, in order to maintain a reasonable spectral resolution a spectrally narrow probe beam could be generated by seeding the FWM process by a narrowband diode laser source located spectrally at the high frequency gain peak of the fiber [21]. Combining the seeding concept with LMA-5 fibers for continuum and pump generation, a multiplex-CARS source covering the whole resonance spectrum below 3200 cm⁻¹ could be realized with high spectral resolution.

5. Conclusion

To the best of our knowledge, we have demonstrated a multiplex-CARS source based on parametric in-fiber frequency conversion by four wave mixing for the first time. We were able to use the same fiber concept with different parameters to create an intense, narrowband pump and probe beam to ensure very high spectral resolution (below 10 cm⁻¹) and a smooth continuum with very high spectral densities. FWM induced supercontinuum generation allows for a controlled energy conversion through the design parameters of the PCF enabling a high power density supercontinuum.

An Yb-fiber laser generates 106 ps pulses with 14 µJ of pulse energy at 1036 nm, 0.5 MHz, which are then frequency converted in a PCF to obtain high-power, passively synchronized CARS pump pulses with a duration of 65 ps. The generated wavelength pair of 648 nm and 700-820 nm was used to probe vibrational resonances of toluene between 1200 and 3200 cm⁻¹.

In contrast to other fiber based frequency conversion concepts on the basis of supercontinuum generation or Raman-scattering based soliton shifts, FWM gives the opportunity for narrowband frequency conversion to shorter wavelengths. Using the emission of FWM induced SC for the CARS Stokes, a multiplex CARS source with synchronized pump, Stokes and probe beams and high spectral resolution of ~10 cm⁻¹ has been created.

This unique light source works in the same wavelength range as titanium-sapphire based systems but with superior stability and usability. Furthermore, the laboratory setup of the current system has a footprint of only 1 m² and consists of commercially available fiber components, securing robust and low-cost implementation. Replacing the fs- by a ps-oscillator with a lower repetition rate would simplify the system even further to a point where only one amplifier and the conversion PCFs are required. As an alternative, a high-repetition-rate fiber-coupled microchip laser [22] might be employed as a seed source. Hence, the presented approach represents a true alternative to much bulkier systems and it has the potential to achieve a wider use of CARS microscopy technology in real-world applications.