1. Introduction

Two photon fluorescence microscopy is an established biological imaging technique, used widely in conjunction with tunable ultrashort pulse Ti:Sapphire lasers, which typically operate in the range 700 - 1000 nm. There is however an increasing demand for ultrashort-pulse lasers that can be operated at shorter wavelengths, particularly between 500 and 600 nm, as these would broaden the application of two-photon fluorescence to a much wider range of biological molecules [1–3]. High peak powers are required to excite the fluorescence through the two-photon absorption process, while low average powers are desirable so not to thermally compromise the biological sample. Generally two-photon absorption spectra tend to be fairly broad (20 - 30 nm) and therefore continuous tunability of the excitation laser is not required. Beam quality must be high to achieve high resolution, and high repetition rates are required for rapid scanning of the sample.

Generating ultrashort pulsed output at 500 - 700 nm has been addressed in a variety of ways. For example, a 1047 nm-pumped optical parametric oscillator (OPO) with intracavity sum-frequency mixing of the pump and signal was demonstrated to yield tunable femtosecond output in the 608 - 641 nm range [2]. Another approach was to pump a photonic crystal fiber with the output of a femtosecond Ti:sapphire laser to generate broadband visible radiation concentrated around 500 - 600 nm [3]. While it was demonstrated that both these systems could be used as sources for two-photon microscopy, it is desirable to explore alternative approaches that might offer increased simplicity and efficiency.

Raman shifting of conventional lasers to access new wavelengths is a well established technique. Stimulated Raman scattering (SRS) in crystalline media has been employed in a wide variety of configurations to efficiently generate visible and UV output [4, 5]. SRS can operate very efficiently using just a single or double pass through a Raman medium for pulses with high peak power [6]. Placing a cavity around the Raman medium to resonate the Stokes wavelength(s) has several significant advantages [7, 8]: it allows conversion of lower-power pulses; it improves beam quality; and it allows effective control over the conversion and cascading of the SRS process to second and higher Stokes orders, so that any desired order can be selectively output, or alternatively multiple wavelengths can be output simultaneously.

For pump pulses with durations of nanoseconds or longer, a short Raman resonator can allow effective SRS conversion of a single pump pulse. For picosecond pulses that are shorter than the transit time through the Raman medium, a resonator can still effectively be employed, but now a train of pulses must be used to synchronously pump an external resonator with a cavity length matched to that of the mode-locked pump laser. Several groups have reported crystalline and gaseous picosecond Raman oscillators synchronously pumped by finite pulse trains from a Q-switched mode-locked laser [9–13], enabling the generation of a range of wavelengths in the visible and IR regions. However, all of these schemes employed pulse energies of the order of μJ or even mJ.

By using a CW mode-locked laser to synchronously pump a crystalline Raman laser, we can generate output in the yellow-orange spectral region that is more suited to applications requiring nJ pulse energies and CW pulse trains. In this paper, we describe such a synchronously pumped Raman laser operating at 559 nm, pumped by a frequency-doubled mode-locked Nd:YVO₄ laser. The laser generated CW mode-locked output at an overall (green-yellow) efficiency of 25.6%. Compression of the 10 ps pump pulses down to 3.2 ps output pulses was observed when the cavity length was slightly longer than for perfect synchronization. We present a detailed characterization of the behavior of this laser.

2. Experiments

A 50-mm-long KGW crystal, AR coated at 532 nm, was used in the experiments as the SRS gain medium. The crystal was oriented such that the pump beam propagated along the N_p axis. A four mirror z-fold cavity was employed as depicted in Fig 1. The mirrors M1 and M2 were a pair of concave mirrors, each with a 20 cm radius of curvature, separated by approximately 23 cm. This mirror separation formed a resonator mode waist of radius of 33 μm at the centre of the KGW crystal, matching the one of the pump beam. The cavity was optimized to achieve the minimum lasing threshold. The angle of the z-fold cavity was set as small as possible at ~4 degrees to minimize the astigmatism of the cavity mode. M1 was a dichroic mirror with 90% transmission at 532 nm. While the laser was designed for the Stokes light to be output through mirror M4 with 5% transmission, there was also some leakage at 559 nm through the other mirrors. Accordingly, the reported output powers are the sum of the powers through the various mirrors. It’s possible to fabricate mirrors with close to ideal performance (for example using ion-beam-sputtering coating technology), and so this total output could be easily achieved in a single beam in an optimized arrangement.

 

Fig. 1. Laser cavity configuration. HWP, Half Wave-Pl ate @ 532 nm; PBS, polarizing beam splitter; M1, dichroic mirror 20 cm ROC; M2, High reflector 20 cm ROC; M3, flat high reflector; M4, output coupler 5% transmission; Δx, cavity detuning.

Download Full Size | PDF

The pump source was a frequency-doubled CW mode-locked Nd:YVO₄ laser (Spectra-Physics Vanguard 2000-HM532). The 2 W pump radiation was directly focused through M1 into the KGW crystal, matched to the cavity mode size by two lenses (f₁ = 20 cm, f₂ = 15 cm). The pump pulse duration was 10 ps, with an 80 MHz repetition rate. The pump was polarized along the N_m axis to match the 901 cm^-1 Raman shift, corresponding to a conversion of 532 nm to 559 nm.

 

Fig. 2. Average output power as a function of cavity length detuning.

Download Full Size | PDF

To characterize the laser, we measured the average output power and the temporal autocorrelation function as a function of cavity length and pump power. We define the cavity length detuning Δx as the difference in the cavity length from that corresponding to the minimum threshold for laser operation, so that Δx = 0 corresponds to perfect synchronization. For positive values of Δx, the Stokes pulse was slightly lagging the pump pulse on each round trip, whereas when Δx was negative the Stokes pulse preceded the pump pulse. The detuning was performed changing the position of M4 with a high precision translation stage.

The dependence of the output power on cavity length is shown in Fig. 2. The maximum output power was achieved at Δx = -60 μm. The power dropped off quickly for Δx > -60 μm, and decreased slowly for Δx < -60 μm. The measured beam quality (DataRay Inc. BeamScope P8) of the output yellow beam was M² < 1.05, slightly better than the pump which had M² = 1.2.

 

Fig. 3. Output pulse duration as a function of the cavity length detuning. Traces above the main curve represent measured autocorrelation functions for different lengths.

Download Full Size | PDF

Figure 3 shows the output pulse duration measured as a function of cavity detuning, using a commercial non-collinear second harmonic autocorrelator (Femtochrome Research Inc. FR-103XL). Under conditions of maximum output power, the Stokes pulse duration was approximately 8.5 ps, compared to the pump pulse duration of 10 ps. For larger Δx, however, we observed substantial shortening of the output pulses, with the minimum pulse duration of 3.2 ps observed when the cavity length was detuned to +8 μm, and the pump was set to maximum power of 1.6 W. Our retrieval of pulse duration Δτ from the autocorrelation traces assumed a Gaussian pulse shape for all Δx. However, we did observe changes in shape of the traces as the cavity was detuned, shown inset in Fig. 3. For cavities with Δx < -50 μm or Δx > 10 μm, the autocorrelations were close to Gaussian. For the shortest pulses the autocorrelation was peaked more strongly, consistent with a sech-squared or single-sided-exponential pulse shape. Using those fittings would retrieve pulse durations shorter than depicted in Fig. 3, dropping to a minimum of under 3 ps. Moving away from the position of maximum compression, the autocorrelation showed a growing pedestal, responsible for the discontinuity in the measured pulse duration at -45 μm.

Figure 4(a) shows the dependence of pulse duration on pump power. For each pump power, the cavity length had to be readjusted for optimum pulse compression. For lower pump powers, the best compression was achieved with a detuning closer to Δx = 0. The pulse duration decreased rapidly to below 3.5 ps as the power was increased, but showed little further shortening as the pump power was increased from 1.4 W to 1.6 W.

 

Fig. 4. (a) Average pulse duration when the cavity is optimized for shortest pulse duration for different pump powers. (b) Average output power as a function of pump power for two different regimes; cavity length optimized for maximum output power (filled squares) and shortest pulse duration (open circles).

Download Full Size | PDF

Figure 4(b) shows average output powers for two different regimes; cavity detuned for maximum output power (filled squares), and detuned for shortest pulse duration (open circles). When operating in the first regime, the maximum CW output power was 410 mW for an incident power of 1.6 W, reaching a maximum green to yellow optical conversion efficiency of 25.6%. The slope efficiency for this case was 42%. For operation at minimum pulse duration, the maximum measured output power was 290 mW, which is an optical conversion efficiency of 18%. However, the slope efficiency showed a significant drop when the pump power was > 0.9 W. We attribute this change in slope to the effects of the pulse compression in the oscillator, as discussed below. The lowest lasing threshold measured was for Δx = 0 (by definition), where the pump power was 360 mW. We did not observe any cascading of the Raman conversion to second or higher Stokes order, owing to the high 98% round-trip cavity losses at the second-Stokes wavelength.

3. Discussion and conclusion

The key feature of the results presented is the very sensitive conditions required for pulse shortening, with cavity detunings over a range of just ~80 μm. Compare this to the spatial extent of the 10 ps pump pulse of ~3 mm. This sensitivity is very different to the operation of the systems reported by different groups that use crystalline and gaseous picosecond Raman oscillators synchronously pumped by Q-switch mode-locked lasers [6, 9–13], which generate trains of typically 20 – 40 separate ps pulses. In those experiments the Stokes field was built up from noise in just a few tens of round-trips. This required a gain of hundreds of percent per round trip with strong reshaping of the Stokes pulse on each pass, resulting in much more relaxed bounds on the tolerated cavity detuning. As those systems relied on the high gain produced in the Raman medium, much higher pump peak powers were required for effective compression, and so picosecond pulses with energies of up to 1 mJ were used.

In our case of a continuous train of mode-locked pulses the round trip gain was of the same order as the output coupling. This regime is much closer to that in previous work on synchronously pumped OPOs. Comparing the curves depicted in Fig. 1 in reference [14] and the Fig. 3 of this paper, we observe many similarities: the pulse compression occurred within a very small region at slightly positive detunings; for longer and shorter cavities the compression was much smaller, with a longer plateau on the size corresponding to negative detuning. The compression of pulses in synchronously pumped OPO’s is produced by the group velocity mismatch between the pump and the generated pulses [14, 15], yielding compression factors greater than 20. In those experiments, the authors explained that the idler overtook the pump pulse because of a larger group velocity. The leading edge of the idler is amplified as it overtakes and depletes the pump pulse; the trailing edge of the idler pulse sees lower gain since it interacts with already-depleted sections of the pump pulse. This preferential amplification of the leading edge of the idler pulse leads to pulse compression [14].

We suggest that the basic physics of compression is likely to be the same, with group velocity mismatch driving the compression. We calculated, using the Sellmeier equations in [16], a group delay mismatch of 83 fs/mm in KGW, which over the 25 mm confocal length of the cavity waist results in the Stokes pulse overtaking the pump pulse by 2.1 ps on each pass. Compare this to the mismatch of 1.6 ps per pass between the pump and idler in the OPO in [14]. This is a big enough fraction of the 10 ps pump pulse to allow compression, although at a loss of efficiency since the 3.2 ps compressed pulse does not interact with the entire 10 ps pump pulse. We note that the high group delay dispersion in KGW (458 fs²/mm at 559 nm) results in a similar group delay mismatch as that calculated in [14], despite the relatively smaller difference between the pump and Stokes wavelengths.

The main difference between our work and the OPOs is that the instantaneous χ⁽²⁾ interaction is replaced with a non-instantaneous χ⁽³⁾ Raman interaction. In the SRS interaction, there is a build-up of the coherent oscillation of the vibrational mode over the dephasing time T₂, equal to 1.96 ps for the 901 cm^-1 mode in KGW. For Stokes pulses with duration comparable to T₂ or shorter (so-called transient SRS), this build up leads to enhanced scattering of the pump at the trailing edge of a Stokes pulse. This will cause enhanced amplification of the tail of the pulse compared to the case in OPOs. We suggest that this is the factor responsible for limiting the effectiveness of the compression that we observed once the Stokes pulse duration approached T₂. The effects of transient-SRS will also be responsible for the fact that the system can tolerate far larger cavity length detunings in the negative direction: in this case the Stokes pulse leads the pump pulse, and this better aligns the maximum scattering strength at the tail of the Stokes pulse with the peak of the pump pulse.

In conclusion, we have demonstrated a simple and efficient method for generating yellow intense short-pulse radiation by the operation of a CW synchronously pumped mode-locked Raman oscillator. The output pulses were compressed down to 3 ps (from a 10 ps pump), generating 0.29 W at 559 nm, with green – yellow conversion efficiencies up to 18% when best compression occurred. This technique allows easy wavelength-conversion of industry-standard mode-locked lasers using robust crystalline technology, and is ideal where a simple, reliable source of short-pulse yellow-orange radiation is needed.

Further investigation using numerical modeling is required to fully determine the roles of both group delay dispersion and transient SRS, and to identify the best conditions to achieve both efficiency and compression. A promising route for achieving shorter pulses is to use LiNbO3 thanks to its shorter dephasing time of 400 fs [17]. However, the gain of this crystal is generally lower than for KGW. We also highlight the possibility for generating a range of different wavelengths using cascading of the SRS process to higher Stokes orders [8], potentially allowing additional shortening.