Fiber Raman amplifiers (FRAs) have been extensively studied since the early years of nonlinear fiber optics [1] and currently find important applications in optical communications. The ability to obtain Raman gain at any optical wavelength, provided a suitable pump source is available, is particularly interesting because it makes the design of FRAs extremely flexible. However, the use of these amplifiers in applications outside telecommunications has been rather limited, mainly due to the low achievable signal peak powers. Due to the low Raman gain coefficient in optical fibers, FRAs pumped by continuous-wave (CW) sources tend to present lengths of at least a few kilometers, in which case nonlinearity-driven distortions limit the peak powers of sub-picosecond pulses to values of the order of a Watt. The long FRA lengths also mean that the typical net dispersions are high and that sub-picosecond pulses are considerably stretched during amplification. FRA length reduction can be obtained with pulsed pump sources. However, in this case pump-signal synchronization is required and pump peak powers must be as high as a hundred kilowatts so that substantial gain can be obtained in fiber lengths of the order of a meter [2].

The problem of pulse distortion in optical amplifiers is also present in conventional bulk gain media, although peak powers in the megawatt range are achievable in this case due to the much shorter interaction lengths. In such systems, peak power scaling to the terawatt level became possible with the development of the technique of chirped pulse amplification (CPA) [3]. CPA systems have also been demonstrated in all-fiber format using rare-earth-doped fiber amplifiers and chirped fiber Bragg gratings for pulse recompression [4, 5]. These systems benefit from the attractive all-fiber configuration characteristics (compactness, low electrical power consumption, alignment-free operation, to name a few) but peak powers achieved are still nonlinearity-limited to the kilowatt region because the compressed pulses travel along a length of optical fiber. The recent development of air-core photonic bandgap fibers (PBFs) with low losses [6], nonlinearity thresholds increased by a factor ~1000 [7], and high waveguide dispersion now allows for stretched pulse recompression in fiber with significantly reduced nonlinearity distortion. All-fiber CPA systems using air-core PBFs have been demonstrated [8, 9] and up to 0.86 MW peak power has been achieved in a CPA configuration that included fiber and bulk components [10].

Despite the benefits and widespread applicability of CPA, little work has been done to increase the peak powers achievable with FRAs. In one experiment [2], a 500 fs pulse was stretched to 23 ps in 100 m of fiber before being amplified in a 1-m FRA and recompressed in a bulk grating pair. To achieve a gain of ~42 dB, the FRA was synchronously-pumped with 50-ps pulses and peak powers as high as 150 kW. More recently, the generation of parabolic pulses, which present a linear chirp, was experimentally demonstrated using a CW-pumped FRA and the possibility of subsequent linear pulse compression was numerically investigated [11]. Nevertheless, to the best of our knowledge, no CPA systems have been demonstrated that utilized CW-pumped FRAs or that utilized FRAs in an all-fiber configuration.

In this Letter, we present an all-fiber integrated chirped pulse Raman amplification (CPRA) system that uses CW-pumped Raman amplifiers and an air-core PBF for pulse compression. ~50-ps input pulses were chirped through self-phase modulation (SPM) in two FRAs and subsequently compressed to sub-picosecond durations. Output peak powers were ~1 kW, significantly increasing the applicability of FRAs. As no components with operation restricted to a specific spectral band were employed, the results demonstrate the feasibility of all-fiber CPRA systems at virtually any optical wavelength. A numerical simulation was utilized to analyze the results.

The experimental configuration employed can be seen in Fig. 1. The input pulses were generated with a fiber-pigtailed distributed feedback (DFB) semiconductor laser operating at ~1542 nm that was biased with a near-threshold dc current and driven by 35-ps electrical pulses at a repetition rate of 50 MHz. The average optical power from the DFB laser under these conditions was as low as ~-24.6 dBm and, therefore, a Raman pre-amplifier (FRA1) was utilized. The pre-amplifier consisted of 9 km of a dispersion-shifted fiber (DSF) with measured losses at 1.55 and 1.45 µm of 0.25 and 0.32 dB/km, respectively. The fiber zerodispersion wavelength was found to be 1549.3 nm and its dispersion slope is estimated as 0.07 ps.nm^-2.km^-1. The DSF was pumped by a fiber Raman laser (Pump1) operating at 1455 nm and delivering 1.9 W of CW power into the gain fiber. The pump counter-propagated with the signal pulses and optical circulators (OCs) were used to insert and extract the pump and the signal to and from the DSF. The signal average power after FRA1 was 14.7 dBm, a large fraction of which, however, consisted of Raman amplified spontaneous emission (ASE). A tunable bandpass filter with a 3-dB bandwidth of 0.7 nm was used to spectrally clean the pulses and reduced the total average power to +0.6 dBm into the second FRA (FRA2).

 

Fig. 1. Experimental setup of the chirped pulse Raman amplification system.

Download Full Size | PDF

FRA2 was virtually identical to FRA1. Again, 9 km of the same model of DSF was employed as the gain fiber. The characteristics of this DSF are believed to be very similar to those of the gain fiber in FRA1, but the zero-dispersion wavelength was measured to be 1553.5 nm. The pump was another fiber Raman laser, which operated at 1440 nm yielding a CW power into FRA2 of 1.3 W. An average signal power of 22 dBm was obtained after FRA2 and was launched into 10 m of air-core PBF (Crystal Fibre model AIR-10-1550). The PBF was fusion spliced to a conventional fiber in its input end and had a total loss of 2.2 dB. The manufacturer-quoted 1542-nm PBF dispersion and dispersion slope values were ~840 ps.nm^-1.km^-1 and ~22 ps.nm^-2.km^-1, respectively, but a relatively high deviation from these values can be expected, as discussed in [8]. To analyze the pulses at different points along the configuration, an optical spectrum analyzer, an autocorrelator and a streak camera were utilized.

Figure 2 shows a comparison of the pulse spectra just before FRA2 and at the configuration output. As expected, the spectrum at the FRA2 output (not shown) was virtually identical to that at the PBF output, indicating that no optical nonlinearity was detectable in the PBF. The bandpass filter efficiently removed the ASE and the pulse spectrum at the FRA2 input shows an extinction ratio of ~45 dB. The figure also clearly shows that the occurrence of SPM is concentrated in FRA2, at the output of which a square spectral profile with a FWHM bandwidth of ~8.8 nm was observed.

Figure 3 depicts the amplitude-normalized temporal pulse profile obtained at several positions along the configuration. The output pulses were measured with the autocorrelator while the streak camera was employed in the other cases. The pulses at the DFB laser output and after FRA1 were practically identical and, thus, only the latter is shown in the figure. The ~50-ps duration of these pulses is greater than that of the electrical pulses possibly due to the laser chip capacitance. The use of the filter does not affect the pulse duration in a measurable way. From the average power and the pulse duration at the FRA2 input it is calculated that the peak power at this point is ~0.40 W. The spectral broadening observed in FRA2 then helps temporally stretching the pulses and their duration at the FRA2 output is ~68 ps. The pulse peak power at this point can be calculated as ~45 W.

 

Fig. 2. Pulse spectra at the FRA2 input (blue) and at the PBF output (red).

Download Full Size | PDF

 

Fig. 3. Streak camera traces of the pulses at the input (blue) and output (green) of FRA2 and autocorrelation of the pulses at the PBF output (red).

Download Full Size | PDF

The autocorrelation of the pulses at the PBF output is quite clean and has a FWHM duration of 1.29 ps, corresponding to a pulse duration of ~830 fs if a sech² pulse profile is assumed. The average power at the PBF output was 100 mW, of which ~6% corresponds to ASE. Therefore, still assuming a sech² profile, an output peak power of ~2 kW can be estimated. However, the autocorrelation trace shows the presence of a low amplitude pedestal extending ~20 ps beyond the pulse, that somewhat decreases the peak power. This pedestal is a consequence of nonlinear chirp in the compressed pulses arising from SPM in FRA2 and from the PBF higher-order dispersion. Incomplete linearization of the SPM-induced chirp [1, 12] in FRA2 can be anticipated from the relatively small pulse duration increase accompanying the large spectral width increase obtained in FRA2.

To assist the data analysis and to better estimate the output peak power, a pulse propagation numerical simulation was utilized that was based on the split-step Fourier method [1]. It took into account attenuation, dispersion, dispersion slope, SPM, and Raman gain from an external pump. Note that, in principle, the equation describing the pulse propagation is coupled with one describing the pump evolution along the FRAs due to the Raman gain dependence on the locally available pump power. As a simplification, the FRA pump power distributions were separately determined with the use of another numerical simulation that modeled a Raman amplifier with counter-propagating, CW, signal and pump. The input CW signal power used in this simulation was equal to the experimental average input power. Such an approximation is valid because in the experiment the pump interacts with over two thousand signal pulses and, effectively, it is the average signal power that is of relevance for its depletion.

The two simulations used the dispersion, dispersion slope and attenuation values mentioned above for the DSFs composing both FRAs. The nonlinear coefficients of these fibers were adjusted to within reasonable values for DSFs [13] and finally set to 1.87 W^-1km^-1. The Raman efficiencies for FRA1 and FRA2, respectively, were experimentally measured and set to 6.96×10^-1 and 7.87×10^-1 W^-1km^-1 in the simulations. The PBF Raman efficiency and nonlinear coefficient were set to zero while its 2.2 dB loss was taken into account. The results were found to be extremely sensitive to the dispersion of this fiber. Good agreement with the experiment was obtained with a dispersion value of 692 ps.nm^-1.km^-1 and a dispersion slope of 20.1 ps.nm^-2.km^-1. The pulses from the DFB laser were assumed to have a transform-limited sech² profile and losses associated with the filter and circulators were taken into account.

The simulations confirm that both the spectral and temporal profile changes in FRA1 are minimal. The spectral width agreement between experiment and simulation at the output of FRA2 is very good, although the patterns within the overall envelopes somewhat differ. This discrepancy can be accounted for by the fact that in reality the DFB laser pulses are expected to exhibit a complicated chirp profile rather than be transform limited. Figure 4 shows the temporal profile comparison of experimental and simulated pulses at the output of FRA2. The curve agreement is fairly good, although the simulated pulses are ~12% longer. The figure also shows the simulated chirp profile, from which it is seen that significant deviation from linearity takes place at points where the pulse intensity is still more than 50% of the peak value.

 

Fig. 4. Temporal profiles of the experimental (blue) and simulated (red) pulses after FRA2; simulated chirp profile (green) also shown.

Download Full Size | PDF

The experimental and simulated autocorrelations of the PBF output pulses are compared in the inset of Fig. 5. The overall trace agreement, including the pulse width and extension of the pedestal, is extremely good, although ripples are visible in the simulated trace. As discussed below, these ripples are mainly due to a nonlinear SPM-induced chirp. Close inspection reveals that a similar structure is present in the experimental trace, although with a much reduced amplitude. The difference between the experimental and simulated traces indicates some discrepancy in the two chirp profiles, which may be traced back to the assumption of unchirped pulses at the FRA1 input. The main part of Fig. 5 shows the simulated PBF output pulse chirp and intensity profiles, which are complicated due to imperfect pulse compression. The intensity profile yields a pulse duration and peak power of 940 fs and 890 W, respectively. Thus, the pulse net compression factor is ~53 while the net peak power increase factor is as high as ~730. Note that the output peak power is about 3 orders of magnitude higher than those typically obtained with FRAs without the use of a CPA scheme. The output pulses can be readily used in applications such as nonlinear microscopy and time-gated imaging [14]. It is, therefore, clear that the scheme described here may significantly contribute to a wider utilization of all-fiber Raman-based pulse sources.

The simulations also reveal that optimizing the PBF length and setting its dispersion slope to zero would yield a peak power of 1.08 kW. However, even in this case both the intensity and chirp profiles exhibit complicated oscillations due to the SPM-induced nonlinear chirp. Linearization of this chirp profile is difficult because FRA2, assembled in a counterpropagating signal-pump geometry, has its gain concentrated near its output end. Thus, the threshold for SPM is reached rather late in the fiber and significant temporal stretching does not have time to take place. Chirp linearization would require the inclusion of a fiber at the FRA2 output for additional stretching or a co-propagating signal-pump FRA scheme. The latter approach is attractive because it can generate linearly-chirped parabolic pulses [11] that may, after compression in the PBF, lead to a significant further peak power increase. However, in this signal-pump configuration care may have to be taken with the build up of ASE between signal pulses.

Note that in the amplification/compression scheme described here pulses become chirped along the amplifier, unlike in most CPA systems where chirping takes place prior to the amplification stage so that a linear chirp profile is achieved. However, as FRAs are typically several kilometers long, the interplay between SPM and dispersion in the gain fiber reduces the degree of nonlinearity in the chirp profile. As a consequence, configuration effectively acts as a CPA system.

 

Fig. 5. Simulated intensity (blue) and chirp (red) profiles of the PBF output pulses; inset: experimental (blue) and simulated (red) PBF output pulse autocorrelations.

Download Full Size | PDF

In conclusion, an all-fiber scheme for obtaining ultrashort optical pulses with increased peak powers was proposed and demonstrated. It is based on a chirped pulse amplification configuration in which CW-pumped stimulated Raman scattering is the only source of gain and temporal pulse compression is obtained in an air-core photonic bandgap fiber. The scheme is simple to operate and may be constructed at any wavelength within the silica lowloss transmission window. With ~50-ps initial pulses a net temporal compression factor of ~53 was achieved. The peak power of the sub-picosecond output pulses approached a kilowatt, representing an increase by about 3 orders of magnitude relative to the value typically obtainable with FRAs in the absence of the chirped pulse amplification scheme. While higher peak powers may still be achievable through system optimization, the current source can already find application in a number of applications outside the optical telecommunications field.