1. Introduction

CARS microscopy is a very versatile tool that provides real-time information due to its label-free chemical imaging capabilities of living tissue, which can be used e.g. during brain-cancer surgery [1,2]. However, a widespread use in real-world applications has not been achieved yet due mainly to the lack of suitable laser sources. Today the most widely used CARS sources are based on lasers such as Ti:Sapphire or Nd:Vanadate coupled with parametric frequency conversion in bulk crystals. These sources are expensive and bulky and their operation requires technical staff devoted to their alignment and maintenance. This has restricted the use of CARS microscopy to specialized research laboratories so far. On the other hand, fiber laser sources are known for their reliability and maintenance-free operation and they could potentially play a key role in establishing CARS microscopy as an indispensable tool e.g. in the health-care sector.

There have been so far several fiber-based approaches attempting to develop CARS sources. Some of these approaches are based on self-soliton frequency shift [3] but they offer limited Stokes powers and poor resolution due to the low energy that a soliton can carry and to its broad bandwidth. Other approaches employ spectral compression in periodically poled lithium niobate (PPLN) crystals [4], but these sources all rely on bulk components, thus ruling out any robust all-fiber implementation.

The problems of these first fiber-based sources were solved with the rapid development of a new kind of laser sources for CARS applications based on four-wave mixing in endlessly single-mode (ESM) photonic crystal fibers (PCFs) [5]. These sources rely exclusively on fiber components and can be, therefore, implemented as all-fiber setups. The first all-fiber CARS source emitting the synchronized pulse trains directly from a fiber end was developed in 2012 [6]. Here, the CARS pump was generated from quantum noise, which linked the bandwidth of the generated pulses to the parametric gain bandwidth of the ESM fiber [7]. The superior robustness of this concept should enable CARS microscopy in a clinical environment [8]. Due to the onset spectral broadening, though, the performance of these sources is limited to moderate resolutions. For example the all-fiber source delivered signal bandwidths in the order of 35 cm⁻¹, which is wider than a typical molecular resonance (these being in the order of 5-10 cm⁻¹) [2].

In order to enhance the chemical contrast, the bandwidth of the laser radiation should be narrower than the resonances of the vibrational energy states. This can be done by seeding the four-wave mixing process by an external laser [9]. Using this approach, nearly transform-limited pulses with a spectral bandwidth below 1 cm⁻¹ have been achieved [10]. This narrow bandwidth makes distinguishing between closely spaced molecular resonances of e.g. CH₂ and CH₃ at 2850 cm⁻¹ and 2930 cm⁻¹ possible. However, the drawback of this approach until now is that the external seed sources are typically expensive and not alignment/maintenance-free. Additionally, the conversion efficiency is limited to about 10% in these sources. In order to solve these issues we propose in this work a self-seeding four-wave mixing setup [11–13] with pump pulses with tens of picoseconds duration in order to generate high-energetic, narrowband signal pulses. At this point it is worth mentioning that a similar setup was recently demonstrated but emitting pulses of only a few ps duration. This source was characterized by low conversion efficiencies, relatively broad signal bandwidth and the need for filters in the cavity [14], which limited its performance and usability as a CARS source. Our concept, on the contrary, delivers higher peak powers with higher spectral resolutions without the need for additional filters in the cavity.

In this contribution a high spectral resolution fiber optical parametric oscillator (FOPO) laser source with a highly dispersive cavity and its application to CARS microscopy is demonstrated [13]. The setup directly generates nearly transform-limited spectrally separated, spatially overlapping and temporally synchronized pulses via is a four-wave mixing (FWM). The energy difference between the pump and signal radiation of the FWM process is designed to address resonances around 2850 cm⁻¹ and 2930 cm⁻¹, which allows distinguishing between CH₂ and CH₃ molecular resonances. At a repetition rate of 750 kHz, an output power of 180 mW for the CARS pump and 130 mW for the Stokes signal with a bandwidth as narrow as 1 cm⁻¹ and pulse durations of around 40 ps are demonstrated. Peak powers of 5.6 kW and 2.9 kW for the CARS pump and Stokes pulses, respectively, are obtained with a conversion efficiency of 38%. Due to the relatively long pulse durations, in the order of a few tens of picoseconds, the temporal walk-off between the pulses is negligible, which makes this source ideal for direct fiber delivery, e.g. to a microscope, without complex dichroic delay lines. This greatly enhances the simplicity and the usability of the setup.

2. 2. Fiber optical parametric oscillator for CARS

As schematically shown in Fig. 1, an Ytterbium- doped fiber-based MOPA system [15] delivers µJ-level pump pulses to the FOPO with a repetition rate of 750 kHz, a duration of 65 ps and central wavelength of 1032 nm. The FOPO cavity consists of a 13 cm long piece of endlessly single-mode PCF (which is a one-hole-missing design with a hole-to-hole distance of 3.25 µm and a relative hole size of 0.5) where the parametric light conversion takes place, an output coupler, a 275 m long polarization-maintaining feedback fiber and a delay stage. In contrast to bulk OPOs, the PCF provides gain via degenerate four-wave mixing (FWM) [5].

 

Fig. 1 Schematic setup of the FWM-based CARS Source. HWP #1 half-wave plate (zero order, 1030nm), HWP #2 and 3 (zero order, 808nm).

Download Full Size | PDF

Without feedback the gain bandwidth of the FWM process results in a broadband signal of about 5 nm (i.e. 75 cm⁻¹ at 795nm) when pumping the PCF at 1032 nm, similar to the free-running system reported in [5]. The PCF fiber is polarization maintaining, which results in a different dispersion for each of the polarization axis. This leads to slightly different phase-matching conditions and, consequently, to a signal and idler wavelength shifts when switching between the input polarization states of the pump light. Altering the orientation of HWP #1 by 45° leads to a relative wavelength shift between signal and idler of 80 cm⁻¹ (5 nm at 795 nm for the signal). The output coupler of the cavity consists of a zero-order half-wave plate (HWP) at 808 nm and a broadband polarizing beam splitter. Considering the losses in the successive elements of the cavity, about 0.1% of the signal emitted from the PCF is fed back to its input. This feedback is delayed via a polarization-maintaining step-index single mode fiber with a core diameter of 6 µm and an additional variable delay line with a maximum travel of 20 cm.

The chromatic dispersion D in the feedback fiber ($D·l_{fiber}=32\frac{ps}{nm}$ at 800nm) leads to temporal broadening of the feedback signal to a point at which it exceeds the pulse duration of the pump light emitted by the fiber MOPA. Without the feedback, as mentioned before, the optical parametric generation (OPG) spectrum is about 5 nm wide, which roughly corresponds to the phase-matched gain bandwidth provided by the conversion fiber (see Fig. 2). On the other hand, the OPG pulses (i.e. pulses generated without feedback) have pulse durations of roughly 40 ps. Accordingly, the initial OPG signal (the one generated by the first pump pulse) fed-back into the delay line will be stretched to 160 ps as it acquires chirp. Now, if the cavity round-trip time matches the repetition period of the pump laser (or a multiple of it), only a part of the feedback signal pulse will overlap with the incoming 65 ps pump. Crucially, due to the chirp imprinted by dispersion on the feed-back pulse, a partial temporal overlap also implies that only some of the spectral components of the signal pulse will interact with the new incoming pump pulse. Consequently, this temporal gain narrowing results in a progressive decrease of the bandwidth of the signal pulses without the use of any narrow-band spectral filtering device in the cavity. Additionally, the chirp of the feed-back pulse provides a means to tune the output wavelength of the generated signal and idler pulses. By changing the cavity length of the resonator (with the delay stage) the successive pump pulses overlap with different spectral components of the signal pulses, which will shift the central wavelengths of the signal and idler waves (see Fig. 3(a)) [16]. In this particular implementation the cavity length has to be changed at a rate $x=D·l_{Fiber}·c=9.6\frac{mm}{nm}$ in order to tune the signal wavelength (note that this rate is determined by the dispersion of the delay line). If narrow signal and idler bandwidths are important for the application, as in the case of CARS microscopy, the dispersion has to be high to reduce the effective chromatic filter bandwidth in the FOPO cavity.

 

Fig. 2 Demonstration of the reduction of the output signal bandwidth with (blue lines) and without (black line) signal feedback.

Download Full Size | PDF

 

Fig. 3 (a) Tuning of the output spectrum of the FOPO. (b) Output power and conversion efficiency of the FOPO as a function of the emission wavelength.

Download Full Size | PDF

Instead of tuning the signal and idler wavelengths by changing the FOPO cavity length, one may alternatively slightly change the repetition rate of the pump laser [11]. However, for simplicity, a translation stage in the FOPO delay line has been inserted in our setup for the time being (see Fig. 1).

Figure 3(a) shows the tuning of the signal wavelength obtained by moving the translation stage. As can be observed, signal wavelengths are generated between 780 nm and 795 nm with a bandwidth of 60 pm FWHM (see Fig. 2). This implies bandwidth reduction of almost two orders of magnitude with respect to that obtained in the OPG signal (i.e. without feed-back) (see Fig. 2). The remaining pump light leaves the cavity with a bandwidth of <40 pm FWHM.

With a pump power of 450 mW at 0.75 MHz, an output power of 170 mW at 795 nm (see Fig. 3(b)) has been achieved. As the pulses are as short as 40 ps, the resulting peak power is 5.6 kW. The remaining out-coupled pump light amounts to 135 mW, which corresponds to a peak power of 2.9 kW. Additionally, Fig. 3(b) illustrates the evolution of the output powers while tuning the wavelength of the signal.

The enhancement of the pulse stability has been measured using a 30 GHz real-time oscilloscope. We compare the FOPO signal with the signal generated from noise. Both signals were measured at the respective pump power, which leads to the maximum spectral power density. Without feedback the peak-to-peak signal fluctuations are 13.4% RMS. With feedback the stability is enhanced to 3.2% RMS. The FWM pump source stability was measured with 1.5% RMS.

The most recent result on narrowband FWM sources for CARS [14] reports a conversion efficiency of up to 20%. The presented source delivers 38% which is, to the best of our knowledge, the highest total conversion efficiency of any short pulse narrowband four-wave mixing source to date. At higher pump powers the conversion efficiency increases, but this is accompanied by the broadening of the bandwidth beyond 1 nm. In this operating regime a signal conversion efficiency of 50% has been measured.

The asymmetrical conversion behavior within the gain region has been investigated in detail in [12] and [17]. With high conversion efficiencies the phase-matching condition is altered by the power of the signal and idler waves, which results in the gain maximum being shifted towards higher wavelengths within the phase-matched gain region.

At any given pump wavelength it is possible to change the central wavelength of the gain region by using the second polarization axis. Hereby the phase-matching condition changes slightly, which results in a wavelength shift of the gain from 792 to 798 nm when pumping at 1032 nm. This can be easily done by rotating HWPs 1 to 3 by 45°.

3. Application to CARS microscopy

A laser scanning microscope similar to the one described in [2,5] has been used to demonstrate the benefits of the presented FOPO in CARS spectroscopy. Please note that in this section the designations of the different waves have been changed to make them compatible with the CARS microscopy nomenclature. Hence, the FWM signal at around 795 nm will be from now on called the CARS pump and the pump of the FWM process will be called the Stokes signal.

In order to avoid photo damage in the sample, we limited ourselves to an overall CARS pump and Stokes power of 80 mW at the sample for imaging thin tissue sections. Up to date there has been little research on the long-term effects of intense light on living cells. For many biomedical applications, however, higher laser powers are required in order to allow for deep imaging of native tissues and organs. In this respect previous all-fiber laser concepts do not allow for upscaling of the total output power comparable to solid state lasers, e.g., Titanium sapphire lasers. The presented fiber laser enables increasing the average laser power by adjusting the repetition rate. In this case, multiple pulses are oscillating in the FOPO cavity. For imaging 200 µm deep in a native skin tissue of a bovine animal model the laser repetition rate was increased by a factor of 3 to 175 mW of pump and 160 mW of Stokes laser power at the laser focus.

Since not all of the generated CARS pump and Stokes radiation is used for CARS measurements, the length of the conversion fiber was changed to 19 cm to reduce the pump power required for the parametric conversion. This way the overall output power was reduced to provide 80 mW of overall average power without further attenuation at the sample. The downside of this adjustment is that it also leads to the tuning bandwidth being reduced.

As mentioned before, to address molecular resonances around 2850 cm⁻¹ and 2930 cm⁻¹, we make use of both polarization axes of the conversion fiber. Figure 4(a) depicts the central wavelength of the CARS pump and the evolution of the output powers of the CARS pump and Stokes waves while tuning their central emission wavelength. Figure 4(b) shows the average powers on target and the measured bandwidth of the CARS pump. As can be seen, with this approach it is possible to obtain high conversion efficiencies around 35% at both design resonances without changing the pump wavelength of the FWM process.

 

Fig. 4 (a) Output wavelengths of the CARS Pump and average power of CARS pump and Stokes with respect to the delay of the delay stage. (b) Average power of CARS pump and Stokes on target and CARS pump bandwidth in respect of the resonance frequency.

Download Full Size | PDF

Since the CARS pump and Stokes waves emerge from the same fiber, both pulses are passively synchronized and overlap in space (due to the fiber) and time (due to the FWM process and the low walk-off in the fiber). Therefore, the fiber output of the source can be directly used for the laser scanning microscopy. In order to filter any unwanted radiation emerging from the FOPO, a long-pass filter at 750 nm is placed into the laser beam. Additionally, a 20x objective optimized for NIR is used to focus the light onto the sample. After the condenser lens, two short-pass filters at 780 nm are used to block the residual pump and Stokes radiation. Moreover, to avoid signals by two-photon exited fluorescence (TPEF) or second harmonic generation (SHG) of the Stokes radiation, a long-pass filter with an edge wavelength of 530 nm is also introduced in the setup. We detect the anti-Stokes radiation around 645nm without averaging with a pixel dwell time of 2µs, which results in an acquisition time of 32 seconds for a picture with a resolution of 4096x4098 pixels. Figure 5 depicts high spectral resolution CARS images at 2850 cm⁻¹ (CH₂, blue) and 2930 cm⁻¹ (CH₃, green) of a human perivascular tissue sample. Lipid filled adipocytes are contrasted in light blue against the greenish protein fibers of the connective tissue demonstrating discrimination of lipids and proteins by CARS imaging at two wavenumber positions. Furthermore, by slightly changing the resonator cavity length with the translation stage, the CARS pump wavelengths can be tuned across these resonances to maximize the contrast between the two CARS signals.

 

Fig. 5 Superposed and color-coded CARS signals probing CH₂ at 2850 cm⁻¹ (blue) and CH₃ at 2930 cm⁻¹ (green). The scale bar corresponds to 200µm.

Download Full Size | PDF

4. Conclusion

In conclusion we have presented a narrowband FWM-based fiber OPO designed for its application in CARS microscopy. This FOPO has the highest conversion efficiency demonstrated to date in a picosecond pulsed fiber-based OPO. The FOPO cavity uses a passive dispersive fiber for wavelength selection and does not need any additional filtering components. The setup is able to excite two of the most important molecular resonances in the region around 2850 cm⁻¹ and 2930 cm⁻¹. It can be easily switched between both resonances by changing the polarization state of the pump launched in the OPO cavity. The setup also provides means to fine-tune the energy difference of the CARS pump und Stokes signal by simple adjustment of the cavity length without the need to change the pump wavelength of the FWM process.

The enhanced stability, the unprecedented high peak-powers and the high spectral resolution leads to high contrast chemically resolved pictures. In this respect this source outperforms any other fiber-based CARS source presented to date. In the near future it is planned to integrate this concept into an all-fiber setup, thus making it the ideal source for CARS microscopy in a biomedical environment, e.g. for intra-surgical imaging of the brain.