1. Introduction

Ultra-wideband light sources based on supercontinuum (SC) generation are useful in diverse applications such as optical communications [1–3], optical coherence tomography (OCT) [4,5], spectroscopy and nonlinear microscopy. Numerous methods have been reported in supercontinuum generation using different lasers and nonlinear-optic media. Mode-locked ultra-short-pulse lasers [3–7], Q-switched lasers [8,9] and even CW lasers [10,11] of various operation wavelengths have been employed as high-power pumping sources for SC generation, while dispersion-shifted fibers, photonic-crystal fibers [5,7,9] and tapered fibers [12] were proven to be successful candidates for the supercontinuum-generating media. However, development of compact, low-cost and reliable SC sources is still in demand to make them practical and easily accessible for uses in various applications. Especially, a compact pump laser is of prime interest because they take up most of the space, cost and the maintenance in the supercontinuum source.

Among various kinds of pumping lasers, laser diodes (LDs) are considered to be one of the most compact and reliable methods of generating short pulses although they can produce only low-power pulses. It is exclusively easy to obtain tens-of-picosecond short pulses with widely available telecom LDs by using a gain-switching technique. The pulse power can be boosted up with compact fiber-optic amplifiers that are widely used in optical communications. Erbium-doped fiber amplifiers (EDFAs) are suitable for the purpose because they are compact and capable of producing high-energy pulses of up to micro-joules [13]. A high-gain amplifier can be easily built up by concatenating the EDFA modules and is able to amplify the low-power diode-laser pulses to high-power ones such that they can induce nonlinear effects efficiently. Because this kind of the pulse laser or so-called, the amplified diode-laser pulse (ADP) source is composed of reliable telecom components and there is no troublesome additional cavity outside the diode laser, it is a reliable and compact method of producing high-power pumping pulses for SC generation.

There have been the previous attempts especially, in 1990s to make a compact supercontinuum generator based on the ADP source for the application of the WDM multi-channel transmissions [1,2]. They have tried to get high-repetition-rate pulses typically, 2.5 or 10 GHz in order to support high bit-rate transmissions. Because pulse energy is inversely proportional to the repetition rate of a pulse source for a given average output power, the peak power of the pulse that they have utilized is relatively low. Troublesome pulse compression has been necessary for SC generation with those high-repetition-rate low-energy pulses. The bandwidths of the generated supercontinuum have been narrow even with kilometer-long supercontinuum fibers. Furthermore, the pulses produced by gain-switched LDs are usually a little bit noisy and hence hardly satisfy the requirement of the communication applications. Because of these reasons, using an ADP source in SC generation has been almost forgotten although it provides a simple and reliable method in pumping for SC generation. However, in most of the applications other than communications, very high repetition and extreme stability are not considered to be essential. Instead, some of them such as OCT application require more bandwidth, spectral power and flatness. A broadband SC generation that satisfies those requirements can be easily obtained with a simple pulsed laser source based on an ADP by lowering the repetition rate and cascading the amplifier modules to form a high-gain amplifier.

In this paper, we report an effective method for ultra-wideband SC spanning more than an octave with a high-power pulsed laser source, in which the original seed pulse is generated by laser-diode gain-switching technique at a wavelength of 1550 nm and is amplified with high gain by cascaded EDFAs. To obtain high peak power that is necessary for efficient supercontinuum generation, the pulse repetition rate is decreased. The pulsed laser based on the ADP method works as a compact pumping source does not require any special techniques such as mode locking and pulse compression. The high-power output of this ADP source was launched into a telecom-grade dispersion-shifted fiber to generate ultra-wideband supercontinuum. The spectrum of the generated supercontinuum at its best spanns more than an octave from 0.8 μm to wavelengths longer than 1.8 μm, which is the upper limit of our spectrum measurement. To the best of our knowledge, it is the first octave-spanning supercontinuum source made of telecom-grade fiber-optic components without mode-locking technique. Various characteristics of our proposed SC generation are also investigated experimentally in this research to understand the basic features and to obtain flat supercontinuum spectrum.

2. Amplified diode-laser pulse source

 

Fig. 1. Schematic view of the ADP (amplified diode-laser pulse) source used in SC generation

Download Full Size | PDF

Figure 1 shows the schematic of the ADP (amplified diode-laser pulse) source used in SC generation. A 1550-nm distributed feed-back laser-diode (DFB-LD) was operated in gain-switching mode by injecting electric current pulses (FWHM ~200 ps) with an appropriate DC bias. The repetition rate of the output pulse is determined by an external clock of adjustable frequency between 1 kHz and 100 MHz. And the produced optical pulses are transmitted through three cascaded optical amplifiers. The amplifiers are low-cost EDFA gain-block modules with uncooled 980-nm pump LDs inside. To minimize the automatic gain saturation caused by ASE re-amplification in cascaded EDFAs, optical band-pass filters (BPFs) of 0.8-nm-wide pass band were placed between the EDFAs. The small-signal gain of each individual EDFA is more than 20 dB. However, the total gain of three cascaded EDFAs in series is slightly more than 40 dB because the gain compression is inevitable even with no input.

The ADP source has an advantage of producing pulses with arbitrary repetition rate. For a given output average power, the ADP source can produce pulses of high peak power at a low repetition rate e.g. 1 MHz. Even though a mode-locked laser can produce pulses with shorter pulse-widths, it can rarely be operated at such low repetition rates. Requiring long cavity length more than 200 meters, a mode-locked laser with <1 MHz repetition is usually unstable and sensitive to environmental changes and also suffer from intra-cavity nonlinear processes. On the contrary, the repetition rate in the ADP source is determined by the seeding laser-diode and freely adjustable electrically. The fundamental lower limit of the repetition rate without energy dissipation comes from the life time of the erbium ion state and is typically less than 1 kHz in the ideal situation. However, self-saturation involved with ASE re-amplification determines the effective lower limit of the repetition rate for the high-gain cascaded amplifiers. For less than a certain repetition rate, the pulse energy does not increase significantly as the repetition rate decreases as Galvanauskas et al. reported previously [13].

The average output power of our composed ADP source differed with the repetition rate and was around 50 mW. Assuming that the contribution of ASE power is less than a half of the total output power, the pulse energy was estimated to be >30 nJ for 1-MHz repetition rate. The average power of the laser diode before amplification was measured to be 1.2 μW at 1-MHz repetition and was amplified up to tens of mili-watts with >40 dB gain. The pulse-width after EDFA 1 was measured with an SHG-based autocorrelator and was found to be 30 ps under Gaussian pulse shape approximation. Because the high-power pulses experience significant nonlinear processes in EDFA 2 and EDFA 3, the pulse-widths after the amplifiers increase slightly and vary with the operation conditions but were measured not to exceed 50 ps in all the cases. Therefore, the pulse peak power was estimated to be about 1 kW.

Although the ADP source contains many components inside, all of them except for driving electronics are available in the form of inexpensive and robust sub-blocks owing to mature telecom industries. Besides, the operation principle is not complex but straightforward so that it is easy to build and operate and requires minimal maintenance. The pulse source has all the features required to make the supercontinuum generation practical and feasible for various applications.

3. Supercontinuum generation

 

Fig. 2. Spectra of the supercontinua generated with DSFs (λ₀ = 1504 nm) of different lengths (0, 3, 5, 15, 50 and 200 m) by pumping 1-MHz 48-mW_avg pulses at 1550 nm. They are displayed with 20 dB offset for better clarity and visibility.

Download Full Size | PDF

The high-power optical pulses at 1550 nm generated from the ADP source was coupled to a nonlinear-optic fiber of supercontinuum generation, or so-called, a supercontinuum fiber. A dispersion-shifted fiber whose zero-GVD wavelength or namely, zero-dispersion wavelength (ZDW, λ₀) is 1504 nm was used for the purpose. The fiber is not a specialty fiber of highly nonlinear design that are typically obtained by reducing the core size excessively but is a communication-grade fiber. This kind of DSFs is easy to fabricate and serves little insertion loss after fusion splicing.

Very wide supercontinua were obtained with the DSF. Figure 2 shows the spectra of the supercontinua generated with DSFs (λ₀ = 1504 nm) of different lengths (0, 3, 5, 15, 50 and 200 m). They are displayed with 20 dB offset for better clarity and visibility. The average power of the pump was 48 mW with pulse repetition rate of 1 MHz. The amount of useless ASE inside the output power was estimated to be less than 9 mW in that the power remained to be residual at 1550 nm even after 200-m-long interaction. So the net pulse average power was estimated to be approximately 39 mW and the corresponding pulse energy was 39 nJ. The SC spectra were measured with an optical-spectrum analyzer (OSA) with a 10-nm spectral resolution. We could not measure the spectra beyond 1770 nm, which is the upper limit of the OSA measurement range.

The spectra of the supercontinua as shown in Fig. 2 spreads form 0.9 to 1.8+ μm, nearly one-octave in the measured range. We believe that the spectrum might be extending up to wavelengths far above 2 μm. The output was delivered by an SMF-28^TM-compatible single-mode fiber. In order to test the single-mode property of the output, we applied a 5-turn bend of the output fiber with diameter of 2.5 cm. This bend gives severe loss to the second-order mode at wavelengths from 0.9 μm to 1.2 μm that are below the intrinsic cutoff wavelength of the fiber. Only negligible loss of less than 0.3 dB (5 %) was observed at that band after applying the bend, which proves that the output of the short wavelengths from 0.9 μm to 1.2 μm was provided in the form of the fundamental mode. And the spectral components of the longer wavelengths above 1.2 μm are naturally of the fundamental mode owing to the single-mode property of the fiber. The stability of the spectrum was also tested. Spectra of the supercontinuum were measured repeatedly during 30 minutes. Only small deviation of less than ±3 % (±0.1 dB) was observed throughout the entire SC band.

Although wider spectrum has been observed with a longer fiber, Fig. 2 shows that the spectral components of wavelengths from 1.4 to 1.5 μm were found to experience power loss after the first 15 meters in the DSF. A negative peak was observed around 1490 nm (201 THz) as indicated with a black arrow in Fig. 2. The frequency difference between the pump and the negative peak is ~ 10 THz in the spectrum, which can be a sign of stimulated Raman interaction. In stimulated Raman scattering (SRS), the positive gain peak can be found at about -13 THz and the negative gain peak, at about +13 THz apart from the pump frequency in silica-based fibers. The origin of the loss at that band will be discussed later in this paper.

 

Fig. 3. Maximum spectral broadening in the short-wavelength side, Δf_max as a function of the fiber length.

Download Full Size | PDF

The spectrum broadening as a function of the propagation distance or the fiber length is calculated from the data shown in Fig. 2. Due to the limitation of the spectrum measurement, spectral broadening only in the short-wavelength side is considered. The blue-shift edge frequency is defined as the highest optical frequency whose power spectral density (PSD) is 50-dB lower than that of the pump at 1550 nm (193 THz). The amount of the maximum spectral broadening in the short-wavelength side that is denoted by Δf_max, is the difference between the blue-shifted edge frequency and the pump frequency. Figure 3 shows the increase in the maximum spectral broadening in the short-wavelength side, Δf_max as a function of the fiber length. Note that the fiber length in the x-axis is scaled in log. The blue-shift exhibits a logarithmical growth by the fiber length up to 10 meters and the growth rate decreases after the point. The efficiency of the spectral broadening decreases dramatically after the first 10 m of the supercontinuum fiber.

4. Temporal walk-off in SC generation

 

Fig. 4. Relative time delays of the spectral components of the supercontinuum generated with the DSFs of 0, 3 and 15 meters in the same condition as that of Fig. 2.

Download Full Size | PDF

In order to understand the saturation of spectral broadening after a certain propagation distance of the SC fiber, we have also measured the relative temporal delays of the spectral components of the supercontinua. Each pulse component was optically filtered by a monochromator and was measured in time domain by an oscilloscope with a high-speed photo-detector (15-GHz bandwidth) to determine the relative time delay. The relative time delay that the monochromator makes at each wavelength was measured and calibrated using another monochromator. Figure 4 shows the relative time delays of the spectral components of the supercontinuum generated with the DSFs of 0, 3 and 15 meters. At each wavelength, 50-%-maximum and 25-%-maximum points are denoted as rectangular and triangular spots, respectively. All the measured pulses had singular amplitude maxima in the middle of the pulses between the rising edges and the falling edges.

All the traces show discontinuity of time delays around the pump wavelength, 1550 nm. It exists even at the output of the ADP source and does not change as a function of the propagation distance in the DSF. We think that the discontinuity comes from the nonlinear interactions inside the optical amplifiers. As shown in Fig. 2, the pump pulse after the amplifiers has exhibited a broader spectrum than that of the LD due to some nonlinear processes already in the amplification stages. Because the last EDFA module is operated in deep saturation, we believe that resonant nonlinearity can give high wavelength-dependent index modulation involved with narrow-bandwidth hole-burning effects.

In supercontinuum output after 3-m-long DSF, all the pulses at different wavelengths arrives almost at the same time within the pulse duration time of 30 ps. Whereas the pulses at short wavelengths in the supercontinuum output after 15-m DSF shows significant temporal walk-off to the pump pulse and the other spectral components. When the walk-off becomes larger than the pulse-width, the generation of new spectral components at the blue-shift edge must have become even less efficient. It gives limitation on the effective interaction length of the beneficial blue-shift processes. This problem of walk-off explains the saturation of spectral broadening after propagating 15 m of DSF in our experiment. On the contrary, there must be long-lasting influences of the pump on the spectral components between 1.4 and 1.5 μm because of small group velocity difference around that region. The excessive interaction might have caused the loss of 1.4–1.5-μm band observed in the output spectra of 50-m and 200-m DSFs as shown in Fig. 2. The role of the walk-off will be discussed further in section 6.

5. Time-gating scheme to suppress ASE

We can expand the output spectrum further to the shorter wavelengths with pump pulses of higher peak power by suppressing ASE more and further reducing the repetition rate. Even though ASE re-amplification has been minimized with the band-pass filters in our ADP source as shown in Fig. 1, ASE noise within the pass band of the filters still exists and contributes to the saturation of the last EDFA. That’s why we could not get more peak power of the pulse by reducing the repetition rate further below 1 MHz. By using a temporal gating scheme we can reduce the ASE noise more. A time-gating element can reject much of the ASE power effectively between the EDFAs since the pulse power of interest is concentrated only in a narrow time slot (<100 ps). An acousto-optic modulator (AOM) with high extinction ratio of more than 40 dB was placed inside the ADP source between the 2nd EDFA (EDFA 2) and the second band-pass filter (BPF 2) and was operated synchronously with the pulse generation. It passes light only within 1.0-μs time slots containing the pulses and suppresses the other ASE. And we have reduced the pulse repetition rate down to 50 kHz to increase the peak power. Figure 5 shows the spectrum of the supercontinuum generated with 50-kHz repetition rate with the time-gating ASE suppression scheme using a 20-m-long DSF. The average power of the input to the DSF was 12 mW. The peak power of the pump was estimated to be about 5 kW, which is five times large than that of the case shown in Fig. 2. The output supercontinuum spectrum reaches 0.8 μm at the blue-shifted edge and gives very flat spectrum. Only the blue-shifted part from 0.8 to 1.55 μm achieves nearly an octave. Moreover, the flat-spectrum band within which the PSD variation is less than 3 dB ranges over a little more than 600 nm from 0.9 to 1.5 μm. The success of the time-gating method in obtaining wider spectrum proves the importance of ASE suppression in the ADP source for wide-band SC generation.

 

Fig. 6. Spectrum of the supercontinuum generated by 50-kHz 12mW_avg ADP pumping with the time-gating ASE suppression scheme, using a 20-m-long DSF.

Download Full Size | PDF

6. Discussion

To understand the mechanisms involved in the supercontinuum generation using fiber waveguides, lots of theoretical and experimental studies have been done and reported previously [15–18]. The role and the significance of each nonlinear-optic effect have been proved to be different case by case, depending on the pulse-widths and the wavelengths of the pump. When pumped by the femtosecond pulses, it is known that the evolution and the fission of higher-order solitons are more important [15,18]. In the case of the picosecond or nanosecond pulses, four-wave mixing (FWM), modulation instability (MI) and stimulated Raman scattering (SRS) are believed to be dominant processes for SC generation while the effect of the self-phase modulation (SPM) is almost negligible [16,17].

The supercontinuum generation with the ADP source reported in this paper is also based on the picosecond pump pulse of which peak power is in the order of a few kilowatts. Although the SPM-induced effects cannot be neglected, it is thought to play a minor role in generating the extraordinarily wide-band spectra in our experiments. On the other hand, parametric FWM is thought to be important to understand the wide-band supercontinuum generation. It is known that when pumping at the wavelength of slightly anomalous dispersion, or a little bit longer than the ZDW of the supercontinuum fiber, the phase-matching condition for efficient FWM can be accomplished in the existence of SPM. This parametric process is also interpreted as MI gain of which bandwidth is proportional to the square root of the pump power [14]. This MI process and SPM-induced effect must have been responsible for the initial broadening in our SC generation. And in the next stage, the phase matching can also be accomplished near the ZDW especially at the wavelengths of slightly normal dispersion. Various kinds of parametric processes of degenerate and non-degenerate cases can be possible and may be cascaded to broaden the spectrum very widely. This kind of two-step growth of supercontinuum has also been analyzed and reported in the previous studies [16]. SRS is also considered to play an important role in the SC generation. Although SRS by itself gives red-shifting effect, it is known that it may stimulate the blue-shifting process when it is coupled with FWM [17]. The spectral asymmetry found in the supercontinua reported in this paper shows the presence of the SRS effect. The loss of 1.4-to-1.5-μm band after the first 15 meters of the DSF that was pointed out in section 3 can be explained by the SRS effect. Because the phase matching of FWM was established by help of SPM, the gain bandwidth of the FWM would be collapsed as the pump intensity contributing to the SPM was decreased after a certain propagation distance. After pump power dissipating, only red-shifting process might have dominated with the presence of higher spectral densities around wavelength of 1.55 μm since the blue-shifting energy transfer from the pump to the band had been exhausted.

Unlike SC generation with nanosecond pump pulses, the evolution of the supercontinuum in our case cannot be fully explained with the concept of CW interactions. Based on the evolution of spectral components shown in Fig. 2 and the temporal walk-off of each frequency component within the spectrum shown in Fig. 4, we believe that the temporal walk-off problem between spectral components is the major obstacle for obtaining broader spectrum in our SC generation. While the interaction length of the wide-band processes is limited by the walk-off or namely, group velocity difference, very long interaction length is expected for the SRS process in the vicinity of the ZDW. It explains the efficient red-shifting effect found in the 1.4–1.5-μm wavelength band in a long fiber. Thus there must be an optimal length of the supercontinuum fiber to get a flat spectrum at the region of the wavelengths shorter than the pump. It is not thought to be beneficial to make excessive interactions by using a fiber longer than a certain length after which severe walk-off occurs between the spectral components.

In this context, tens-of-picosecond pulse that we used for SC generation sounds optimal for the pump of supercontinuum generation. The effective length of the supercontinuum fiber is less than a hundred meters with the picosecond pulse pumping. Longer duration of the pump pulse would permit longer interaction length but more than a hundred meters is not practical, especially for PCFs having high attenuation. Moreover, the energy per pulse needs to be increased to maintain the peak power in a long-duration pump pulse. Too energetic pump and supercontinuum pulse are not easy to treat for the source itself and for the applications. It also means that we can obtain only a low repetition rate less than 10 kHz for a given practical output average power. On the contrary, the pump pulse of even shorter duration i.e. sub-picosecond pulse is not easy to generate. The ultra-short femtosecond pulse is fragile and easily dispersed due to its intrinsic broadband nature. If the peak power of the ultra-short pulse were not far higher than kilowatts, the efficiency of the parametric FWM would be limited by the fast walk-off. Thus the supercontinuum generation by pumping the ultra-short pulse should also rely on the other mechanism, which must be the soliton-related effects. Therefore, the supercontinuum generation using the ADP source looks advantageous in many aspects. The tens-of-picosecond duration of the pump pulse enables fully wide broadening with only tens-of-meter supercontinuum fibers. The output pulse is not too energetic and can be given at relatively high repetition rates of megahertz’s. The generation of the SC pulse can be triggered electrically. The output spectra are relatively flat without fine structures that are usually found in the SPM-induced broadening effects.

7. Conclusion

In this paper, we reported an efficient method to generate ultra-wideband supercontinua by using a compact pumping source that is composed of only telecom-grade fiber-optic components. Compared with the other high-power short-pulse generation methods such as mode locking, the amplified diode-laser pulse (ADP) source gives a simple and robust scheme for SC generation. We have demonstrated that single-mode octave-spanning supercontinua can be obtained easily with the pulse source and the conventional telecom-grade DSF. We expect that wider spectrum could be generated by using highly nonlinear DSFs or by employing more powerful EDFAs. In this study, various characteristics of the supercontinuum generation were also investigated experimentally: spectral evolutions and time delays of the spectral components as a function of the fiber length. We have shown that the walk-off between the pump and the generated spectral components limits the effective interaction length of the SC generation and may determine the optimal length of the supercontinuum fiber. We have also observed that excessive SRS interaction may break the flatness at the band of wavelengths slightly shorter than the pump. We have also obtained a very flat and very wideband supercontinuum by using a time-gating ASE suppression method with help of the knowledge on the optimal length of the fiber. The bandwidth of the flat-spectrum band was more than 600 nm, within which the power spectral density exhibits only small wavelength-dependent variation of <3 dB. This success also proves the importance of ASE suppression in the ADP source for wide-band SC generation. Because our proposed supercontinuum source is composed of only telecom-grade fiber-optic components, we expect that it will be a practical pulsed broadband source for many applications.