1. Introduction

Variety of applications of pulsed mid-infrared sources in the 2 μm spectral region has inspired the community to develop novel concepts of Thulium-doped fiber lasers (TDFLs). Especially laser spectroscopy and environmental sensing benefits from the spectral properties of mode-locked TDFLs. Their emission spectrum (1.9 – 2.0 μm) overlaps with absorption lines of greenhouse gases: nitrous oxide (N₂O) and carbon dioxide (CO₂) [1]. The development of compact, broadband TDFLs (i.e. mode-locked all-fiber lasers) might strongly contribute to the development of robust, sensitive and transportable trace-gas detection systems.

Broad optical spectra might be generated from dispersion-managed fiber lasers. Fiber lasers with balanced or normal net dispersion typically require saturable absorbers with high modulation depths [2,3]. In case of graphene, the typical modulation depth of multilayer structures is limited to few percent [4]. Up till now, the broadest optical spectra and shortest pulses at 2 microns were generated using nonlinear polarization rotation (NPR) mechanism [5–7]. This effect acts as an artificial saturable absorber and indeed is considered to be the fastest SA for fiber lasers, enabling the generation of record-breaking pulse durations (e.g. 37.4 fs at 1560 nm [8]). Nevertheless, NPR lasers tend to be unstable, vulnerable to external disturbances and require careful polarization alignment. Thulium-doped lasers based on SESAMs were also reported, but the bandwidth was limited to 10 nm [9]. Graphene is an excellent material for the generation of broadband pulses from mode-locked lasers, since its saturable absorption effect occurs in extremely broad spectral range, from the visible to the mid-infrared, or even terahertz/microwave region [10]. The combination of almost wavelength-independent absorption of graphene with the extremely broad gain bandwidth of Tm-doped fibers might lead to the development of mode-locked lasers with broad emission spectra. At present the modulation depth of graphene-based saturable absorbers is most likely too low in order to support broadband dissipative soliton mode-locking in Tm-doped fiber lasers. This is the main reason why such laser was not demonstrated in the literature yet. Alternatively, broad emission bandwidth might be achieved in the so called noise-like pulse (NLP) regime. In NLP lasers, the output pulse consists of a bunch of short pulses with varying widths and intensities [11,12]. The autocorrelation trace of such bundle of pulses is characterized by a very narrow (usually femtosecond) spike, located on the top of a broad pedestal (typically ~ps-long) [13]. The emission spectrum might be very broad, comparable to or even broader than the gain bandwidth. However, to our knowledge, NLP lasers in the 1.9 – 2.0 μm region based on graphene saturable absorbers were not yet reported in the literature. In contrast to dissipative soliton lasers, NLP sources do not require saturable absorbers with high modulation depth. Since the intensity of the pulses is lower, the peak intensity at the SA surface is also limited. This allows to avoid optical damage of the SA. Up till now, several Erbium-doped NLP lasers were reported, with bandwidths exceeding 100 nm [14]. Some applications, like interferometry, might benefit from such laser behavior [15]. Also efficient supercontinuum generation using NLP lasers as pump sources was demonstrated [16]. Noise-like lasers operating in the 1.9 μm region were reported, however, the achieved bandwidths were at the level of tens of nanometers [17–21]. The pulse bunch energies achieved directly from the oscillator are usually also limited by nonlinear effects or the optical damage of the SA. The maximum energy so far was reported by S. Liu et al. (32.72 nJ) [19]. To overcome this limitation, one can use an external amplifier, which might be realized as a fully fiberized, monolithic construction, integrated with the oscillator. Such approach was applied previously to amplify soliton pulses at 1970 nm beyond the limit achievable directly from the laser [22].

In our previous report, we have demonstrated noise-like generation from a Tm-doped fiber laser based on NPE mechanism [7], with 1.3 nJ pulse bunch energy. Here, we demonstrate the generation and amplification of broad noise-like pulses in a compact, monolithic, and fully fiberized Tm-doped laser-amplifier system, with multilayer graphene saturable absorber responsible for the mode-locking. The setup generates 1.21 W of average power at 23.5 MHz pulse repetition rate. This is, to our knowledge, the most powerful noise-like laser operating at 2 microns reported so far. It is also the first demonstration of a graphene-based Tm-doped fiber laser operating in the NLP regime. Such compact, robust and cost-effective system might serve e.g. as pump source for mid-IR supercontinuum generation.

2. Experimental setup

The setup of the experiment is depicted in Fig. 1(a). It consists of an oscillator and a fiber amplifier, which is directly spliced to the laser output. The seed comprises the following components: an integrated output coupler/isolator/wavelength division multiplexer (OC/ISO/WDM), a 17 cm piece of Tm-doped fiber (Nufern TSF-5/125, TDF), a polarization controller (PC), the graphene-based saturable absorber (GSA), and a segment of dispersion compensating fiber (DCF) with group velocity dispersion (GVD) of 0.0246 ps²/m at 1950 nm.

 

Fig. 1 Experimental setup of the noise-like oscillator and amplifier (a), measured nonlinear transmission curve of the 60-layer graphene/PMMA composite (b).

Download Full Size | PDF

The dispersion of the DCF was measured using white light interferometry, described in [23]. The oscillator is pumped by a home-build Er/Yb co-doped fiber laser, delivering 1.2 W of continuous-wave (CW) power at 1560 nm wavelength with 0.5 nm bandwidth (defined by a fiber Bragg grating, FBG). The length of the DCF used in the cavity was 6.71 m. The calculated net group delay dispersion (GDD) of the laser was normal, and equal to 0.011 ps². The noise-like pulses from the seed are directly introduced to an integrated fiber amplifier, which consists of an isolator (ISO), multimode pump combiner (MPC), a double-cladding Tm-doped fiber (CorActive DCF-TM-6/128, DC-TDF), a cladding-mode stripper (CMS), and a 99/1% coupler. The amplifier is pumped using a multimode laser diode, delivering up to 6 W of power at 791.7 nm wavelength.

The saturable absorber used in the oscillator is based on a 60-layer graphene/polymer composite. Sixty graphene monolayers were grown by chemical vapor deposition (CVD) and afterwards stacked layer by layer on a poly(methylmethacrylate) (PMMA) support, forming a stable, free-standing multilayer graphene foil. The details on the graphene/PMMA composite fabrication process were described in our previous report [4]. The power-dependent transmission of the fabricated SA was tested using a pulsed laser as an excitation source (Menlo T-Light, 1560 nm, 100 MHz repetition rate and 100 mW of average power). The maximum achievable fluence at the sample was at the level of 740 μJ/cm², while the pulse duration was 400 fs. The obtained power-dependent transmission curve together with theoretical fitting (formula valid for a fast saturable absorbers taken from [4]) is plotted in Fig. 1(b). The theoretical modulation depth of the SA should be at the level of 9.5% with 55.5% of non-saturable losses. However, the sample damages at flucences of 600 μJ/cm² (clearly visible roll-off in the transmittance curve). The maximum observed change in the SA transmittance was 6.9%. It is worth mentioning, that during experiments in the mode-locked Tm-doped laser no damage of the SA was observed, since the intra-cavity fluence does not reach the 600 μJ/cm² level.

3. Experimental results

All experimental data were gathered using the following measurement equipment: an optical spectrum analyzer (Yokogawa AQ6375), a 7 GHz radio frequency (RF) spectrum analyzer (Agilent EXA N9010A), a 6 GHz oscilloscope (Agilent Infiniium DSO91304A), a 16 GHz photodiode (Discovery Semiconductors DSC2-50S), an autocorrelator (Femtochrome FR-103XL), and an optical power meter (Gentec Maestro with XLP12-3S-VP detector).

Firstly, we have characterized the seed laser. After launching the pump with power levels above 1.09 W, the oscillator starts to operate in the NLP regime by itself. The best performance in terms of bandwidth was achieved with 1.25 W of pumping. Further optimization by slight adjustment of the PC allowed to observe 63 nm wide emission directly from the oscillator. The PC is not necessary for starting pulsed operation (it runs by itself after launching the pump), but allows to optimize the spectrum shape during operation (e.g. cancelling unwanted peaks originating from parasitic continuous-wave lasing). The optical spectrum is depicted in Fig. 2(a). The pulse repetition rate was 23.5 MHz, which corresponds to a ~8.86 m long cavity. The RF spectrum is depicted in Fig. 2(b). The high-resolution scan (with 33 Hz resolution bandwidth) reveals that the signal is characterized by a noise pedestal, typical for NLP lasers. However, the signal to noise ratio (SNR) is at the level of 57 dB, which is better than reported previously [19,21]. Despite the noisy output, the RF spectrum in the wide span (inset of Fig. 2(b)) constitutes a very broad comb of harmonics up to 7 GHz (limited by the analyzer bandwidth), indicating that a large number of longitudinal laser modes is phase-synchronized. The average output power measured at the 10% output coupler was 3.8 mW, which corresponds to 0.16 nJ of pulse bundle energy. Unfortunately, the sensitivity of our autocorrelator was too low to measure the pulse duration directly from the laser at this power level. Nevertheless, 63 nm wide emission is, to our knowledge, the broadest spectrum generated from a graphene-based TDFL. Figure 3 depicts the output pulse train recorded with different temporal ranges: 6 μs (a), 500 ns (b), and 2 ns (c). It can be seen, that the amplitudes of the pulses fluctuate randomly by about 25%, which indicates noisy operation of the oscillator. The measured width of a single pulse is approx. 100 ps, which results from the limited bandwidth of the used oscilloscope. The actual pulse bunch duration is most likely shorter and needs to be measured using an autocorrelator. It is also important to mention, that despite our efforts the laser did not operate in dissipative soliton mode-locking regime. Nevertheless, we strongly believe, that the laser can generate dissipative solitons after careful optimization of the SA technology, especially increasing the modulation depth and minimizing the damage threshold of the graphene SA (e.g. by making the graphene transfer process more clean).

 

Fig. 2 Optical (a) and RF (b) spectra of the seed laser (prior to amplification).

Download Full Size | PDF

 

Fig. 3 Fast oscilloscope traces of the NLP laser output (prior to amplification) recorded with different time scales: 500 ns/div (a), 50 ns/div (b), and 500 ps/div (c).

Download Full Size | PDF

Next, the pulses from the seed laser were launched directly to the amplifier, without any dispersive lines for pulse broadening, like in chirped pulse amplification (CPA) schemes. Figure 4 shows the evolution of the optical spectrum at different amplifier output power levels (a), and the corresponding autocorrelation traces (b). In general, the obtained optical spectra are narrower than the initial spectrum, which is caused most likely by the gain narrowing effect in the amplifying medium [24]. Typically for NLP lasers, the autocorrelation traces contain a narrow peak on a broader pedestal. In our experiments, at highest pumping power the width of the spike was 255 fs, which is shorter than in previous reports [13,19]. However, the spike FWHM is not related to the duration of the pulse envelope. It is a coherent artifact that arises from the substructure of noise-like pulses [25]. Interestingly, the FWHM of the pedestal broadens with increasing pump power, which is most likely due to the dispersion of the amplifier: at higher pumping levels the spectral width broadens, which implies a larger temporal walk-off of the pulses in the bundle.

 

Fig. 4 Optical spectra (a) and pulse autocorrelations (b) after amplification, obtained at different output power levels.

Download Full Size | PDF

Figure 5(a) shows the measured RF spectrum of the amplified signal. It can be seen, that the noise level is kept still at similar level in comparison to the oscillator (53 dB SNR). The maximum average output power was 1.21 W, obtained at 6 W of pumping. The output power vs. pump power characteristic (see Fig. 5(b)) is free of any roll-over, which indicates, that the output power is only pump power limited. Since the noise-like emission is a bunch of pulses with varying amplitudes and width, the exact pulse energy in this regime cannot be determined. However, the energy of one NLP bunch can be roughly estimated by dividing the average output power by the repetition frequency (analogously to other authors [17–21]). Based on such calculations, the estimated energy at highest pump power is at the level of 51.5 nJ, which is significantly higher than in previous reports on Tm-doped NLP lasers at 2 μm: 32.72 nJ by S. Liu et al. [19], 17.3 nJ by X. He et al. [18], and 1.27 nJ by Q. Wang et al. [21].

 

Fig. 5 Measured RF spectrum after amplification (a), amplifier average output power vs. launched pump power relation (b).

Download Full Size | PDF

4. Summary and outlook

Summarizing, we have demonstrated generation of broadband noise-like emission from a Tm-doped fiber laser based on multilayer graphene SA at 23.5 MHz pulse repetition rate. The pulses were also amplified in a compact, monolithic, and fully fiberized Tm-doped amplifier (without the necessity of temporal pulse stretching or compression). The amplifier delivered 1.21 W of average power at 6 W of pumping. This result outperforms the previous reports on NLP Tm-doped fiber lasers in terms of average output power and pulse energy. In our future work the source will be examined as pump for mid-infrared supercontinuum generation in highly nonlinear optical fibers with zero-dispersion wavelength located around 1.95 μm. Based on previous literature report, the obtained power levels should be sufficient to observe significant spectral broadening in e.g. photonic crystal fibers. Broadband sources of mid-IR radiation exceeding 2 μm can be afterwards used as e.g. signal sources for chromatic dispersion measurements in optical fibers via white light interferometry.