1. INTRODUCTION

The generation of optical pulses from a laser is an active research field due to its large application potential in a variety of domains, such as telecommunications, metrology, remote sensing, and material processing. In addition, extreme nonlinear optical effects can be explored because the pulses can be externally amplified and the high spatial coherence of the beam allows for tight focusing of the light to small transverse sections; hence extremely high peak intensities and fluences can be achieved [1].

Pulses in the picosecond range or shorter can be obtained by mode-locking (ML) of the laser [2]. The initial development of pulsed laser sources relied on solid-state lasers [1], but fiber lasers have emerged as a reliable alternative to bulky solid state lasers for optical pulse generation [3]. Besides compactness, they provide high beam quality with high efficiency, and the fibered cavity offers mechanical robustness; however, dispersion management is more problematic in fiber lasers due to fiber dispersion, and the optical powers that can be achieved in fiber lasers are usually lower than those obtained in solid-state lasers. Most fiber lasers rely on a fiber section doped with rare-earth ions, which provide gain upon optical pumping; yet, fiber lasers that use a semiconductor optical amplifier (SOA) as the gain element have also been developed, since they offer the simplicity of electrical pumping, high nonlinearity, rapid tuning, and a broad gain bandwidth [4–8].

Active ML of SOA-based fiber lasers in unidirectional ring cavities has allowed researchers to obtain optical pulses with a duration of 15 ps [9], which can be further amplified and compressed down to 50 fs [10]. Exploiting nonlinear polarization rotation of the field in the SOA, passive ML has been achieved in different situations. Yang et al. [11] have obtained 800 fs pulses at the fundamental ML frequency of 14 MHz in a unidirectional SOA-based fiber laser. In addition, dark and bright pulses can be obtained in a similar configuration [12]. In this case, the pulses arise from a square-wave intensity modulation that can be quite slow and that can be useful for generating optical clocks with high duty cycle, and the repetition rate of the pulses can be increased by operating the device in the harmonic ML regime. Recently, frequencies up to 12 GHz have been obtained by passive ML of an SOA-based, figure-of-eight laser (F8L) operating at 1.06 μm, with pulses of 29 ps duration; in this case, the repetition frequency can be selected by changing the bias current in the SOA [13].

In this work, we report on self-starting ML of an F8L based on an SOA working in the CL band that does not rely on nonlinear polarization rotation for achieving ML. The cavity is defined by standard off-the-shelf polarization maintaining (PM) components, and its simplicity makes it very attractive for applications because it does not require of any polarization controller or adjustment. Optical pulses with a duration of 247 fs have been obtained with a repetition rate that depends on the bias current into the SOA. Self-starting fundamental ML at a repetition rate of 18 MHz is obtained at the lasing threshold, and it evolves into harmonic ML up to 2.45 GHz upon increasing the bias current in the SOA.

2. EXPERIMENTAL SETUP

The experimental setup is schematically shown in Fig. 1. The F8L cavity is defined by two PM fiber loops coupled through a 50:50 PM fiber coupler. One of the loops (length $L_1\approx 6\mathrm{m}$) is purely passive, and it includes a PM optical circulator that imposes unidirectional operation of the loop. The other loop (length $L_2\approx 5\mathrm{m}$) includes a single-transverse-mode PM fiber-coupled SOA. The amplifier is strongly dichroic, with a difference of 18 dB in the polarized power levels emitted by amplified spontaneous emission, and it provides an optical gain bandwidth of 80 nm around a wavelength of 1550 nm for the dominant polarization. The SOA (Thorlabs BOA1004P) is thermally stabilized to better than 0.1°C, and it is driven by a stable current source. The laser output is split in two branches. One of them directs 10% of the output toward an optical spectrum analyzer (OSA, HP 86142A). The remaining 90% of the output is sent to a fiber-coupled amplified photodetector (New Focus 1544B) whose output is divided by a radio-frequency (RF) 3 dB coupler connected directly to a real-time digital oscilloscope (Agilent Infinium DSO 81204A) and to an RF spectrum analyzer (Anritsu MS2602A) after being amplified by an RF amplifier (Nuclétudes ALC2250). Autocorrelation measurements have been performed when possible by further splitting this branch in two; in one of these branches we insert another optical circulator followed by a PM erbium-doped fiber amplifier (Amonics, AEDFA-PM-CL-17-B-FA providing up to 17 dB gain over the C or L bands) that is used to amplify the laser output, which is finally sent to a background-free optical autocorrelator (Femtochrome Research Inc., FR-103HS).

 

Fig. 1. Schematics of the F8L setup. CIR, optical circulator. 50:50 and 95:5: fiber couplers that divert 50% and 5% of the light, respectively. All elements are PM.

Download Full Size | PDF

3. EXPERIMENTAL RESULTS

Figure 2 displays the light-current characteristics of the F8L when the substrate temperature of the SOA is set at 20°C, as recorded with a power meter. Increasing the bias current $I$ into the SOA, the lasing threshold occurs at $I=I_{\mathrm{th}}\simeq 90\mathrm{mA}$, and the emitted average power starts growing, reaching 0.6 mW—which corresponds to an intracavity optical power around 12 mW—for a bias current $I=340\mathrm{mA}$. The LI curve presents a slight hysteresis close to threshold, and the F8L does not switch off until the bias current applied to the SOA is reduced below 85 mA; indeed, for currents below 120 mA, the power emitted is higher on the decreasing branch than on the increasing one.

 

Fig. 2. Output power of the F8L when the substrate temperature is $T=20°\mathrm{C}$ as the bias current of the SOA is increased (blue) and decreased (red).

Download Full Size | PDF

The optical spectra for several values of the bias current are shown in Fig. 3. At threshold ($I=95\mathrm{mA}$) the F8L has an optical spectrum centered at a wavelength $\lambda \simeq 1575\mathrm{nm}$. In this regime, the optical spectrum is rather broad ($\approx 25\mathrm{nm}$ full width at half-maximum, FWHM), and it exhibits some ripples that are more pronounced on the red wing of the spectrum, which we attribute to the residual reflectivities of the SOA facets and connectors in the setup—a similar ripple is observed in the optical spectrum of the amplified spontaneous emission of the SOA when the bias current is above 500 mA. However, the mode spacing of the F8L cavity is $\approx 18\mathrm{MHz}$, far too small to be resolved by the OSA. The optical spectrum remains essentially unaltered as the current is increased up to $I=110\mathrm{mA}$. For this current value, a peak appears on the red wing of the spectrum at wavelength $\lambda \approx 1584\mathrm{nm}$, the modulation of the spectrum becomes clearer, and the FWHM of the spectrum reduces to $\approx 3\mathrm{nm}$. Further increasing the current leads to a redshift of the dominant frequency up to $\lambda =1591\mathrm{nm}$ for $I=285\mathrm{mA}$, but the FWHM of the spectrum does not change noticeably. In all cases, the optical and RF spectra of the system were stable over tens of minutes.

 

Fig. 3. Optical spectra of the F8L output when the substrate temperature is $T=20°\mathrm{C}$ for bias currents (a) $I=95\mathrm{mA}$, (b) $I=110\mathrm{mA}$, and (c) $I=285\mathrm{mA}$.

Download Full Size | PDF

In the time domain, the former optical spectra correspond in most cases to the emission of regular trains of optical pulses, as shown in Figs. 4 and 5. Close to the lasing threshold, Fig. 4, the laser output exhibits a regular train of short pulses with a period of $\approx 56\mathrm{ns}$ that corresponds to the roundtrip time in the F8L cavity; on the oscilloscope screen, the pulses appear to have an FWHM of $\approx 70\mathrm{ps}$, which is the minimum pulse width that can be achieved with the detection bandwidth available in our system. This emission state corresponds to the fundamental passively mode-locked state of the system, and it occurs spontaneously. Its RF spectrum is displayed, with different degrees of frequency resolution, in Figs. 4(b)–4(d). The fundamental frequency of the RF spectrum occurs at 18.0623 MHz and exhibits 50 dB signal to noise ratio, typical of semiconductor lasers [14,15]. A flat comb of replicas of this peak is observed within 5 GHz. Indeed, the comb covers the full bandwidth of the RF spectrum analyzer, although its amplitude decreases above 5 GHz due to the bandwidth limit of the RF amplifier.

 

Fig. 4. (a) Time trace of the fundamental mode-locked emission of the F8L when the substrate temperature is $T=20°\mathrm{C}$ and the bias current $I=85.2\mathrm{mA}$. (b) RF spectrum of the pulse train obtained for the fundamental ML regime in the 0–5 GHz range. (c) Detail of the RF spectrum in the 0–200 MHz range. (d) Zoom around the fundamental peak of the RF spectrum at frequency $\sim 18\mathrm{MHz}$.

Download Full Size | PDF

 

Fig. 5. (a) Time traces of the F8L emission when the substrate temperature is $T=20°\mathrm{C}$ for different values of the bias current close to the lasing threshold. From top to bottom, $I=201.5$, 113.5, 102.8, and 93 mA. The traces have been offset by 100 mV for clarity. (b)–(e) RF spectra from 0 to 8.5 GHz of the pulse trains in (a).

Download Full Size | PDF

Further increasing the current leads to the appearance of new pulses within the round trip. Although the pulse heights change when the pulse number increases, their temporal profile on the oscilloscope screen remains unchanged. Close to the threshold, the coexistence of an increasingly large number of states with different numbers of pulses per round trip arranged in a multiplicity of forms is observed, but for still higher bias currents, different harmonic passively mode-locked states set in and remain stable for large current intervals [see Fig. 5(a)]. We have obtained stable repetition rates up to 2.45 GHz, which correspond to the 136th harmonic of the fundamental repetition rate. The RF spectrum of each of these states is shown in Figs. 5(b)–5(e). Again, a clear comb can be seen, with decreasing amplitude above 5 GHz due to the amplifier bandwidth limit, indicating that the pulse width is rather shorter than 100 ps.

In our case, the rather low optical power emitted by semiconductor lasers does not allow performing a direct autocorrelation measurement, and amplification of the pulse train is required, as is often the case when dealing with semiconductor lasers [14–16]. However, the optical spectra of the emission from the F8L lies precisely at the transition from the C-band to the L-band; hence amplification through our fiber amplifier will distort the pulse shapes rather strongly, and autocorrelation measurements will not be meaningful. In order to measure the pulse width, we have set the substrate temperature of the SOA to 40°C; hence the operation wavelength of the laser is redshifted to 1580 nm or higher, that is, well within the L-band. Figure 6 displays the main characteristics of the pulse train when the bias current in the SOA is set to 204 mA: the optical spectrum [blue trace in Fig. 6(a)] presents a rather triangular form with maximum power at 1596 nm and a spectral FWHM of $\sim 7\mathrm{nm}$. The pulse train has a period of $\sim 700\mathrm{ps}$ that corresponds to a dominant frequency in the RF spectrum of $\sim 1.43\mathrm{GHz}$, with clear harmonics; note however that in this case the pulse train presents some degree of amplitude modulation that manifests as an increased noise floor around the peaks in the RF. This modulation of the pulse amplitude is induced by the proximity (in terms of bias current) of the mode-locked state with frequency $\sim 1.4\mathrm{GHz}$ that is portrayed in Figs. 6(a)–6(c). Further increasing the current reduces the modulation amplitude, but the SOA has not been set at very high current in order to avoid damage in the device as the substrate temperature is rather high. The optical spectrum of the pulses after passage through the L-band fiber amplifier is also shown (brown curve in a), and it can be seen that—in spite of the appearance of a shoulder arising from amplified spontaneous emission in the L-band amplifier—more than 95% of the input pulse energy is linearly amplified. In these conditions, the autocorrelation trace of the amplified pulses displays a clear peak with a FWHM of 382 fs. After fitting to hyperbolic secant pulses (see brown line), the FWHM pulse width then corresponds to 247 fs, and the resulting time–bandwidth product is 0.31, quite close to the expected theoretical limit. At still higher currents [see Figs. 6(d)–6(f)], the pulse width remains the same, although the repetition rate of the pulse train has increased up to 2.1 GHz and there is some degree of amplitude modulation. The variations in pulse amplitude are on the order of 10% on the oscilloscope screen, but they cannot be assessed from the autocorrelation traces because the pulse period ($\simeq 480\mathrm{ps}$) is almost twice the maximum delay available in our autocorrelator.

 

Fig. 6. (a) Optical spectra of the pulses before (blue) and after (brown) amplification by the L-band fiber amplifier. (b) RF spectrum and (c) intensity autocorrelation trace (blue) and ${\mathrm{sech}}^2$ fit (brown) of the amplified pulses for an SOA bias current $I=204\mathrm{mA}$. (d) Through (f) same magnitudes for an SOA bias current $I=300.2\mathrm{mA}$. In all these measurements, the SOA temperature is $T=40°\mathrm{C}$.

Download Full Size | PDF

In order to clarify the physical origin of these ML dynamics, a PM optical isolator was inserted between the 50:50 coupler and the SOA. It was observed that ML appeared only when the isolator was set such that light traveled from the coupler toward the SOA. In this configuration, the light entering the SOA through the rear facet is strongly amplified and partially fed back onto the opposing facet of the SOA after reaching the fiber-loop mirror. If the direction of the optical isolator is reversed, the amplified light cannot be fed back onto the SOA, which experiences only feedback due to the weak spontaneous emission in the counterpropagating direction. In our opinion, the physical mechanism underlying pulse formation in this system is the interaction between the counterpropagating pulses that is at work in colliding-pulse mode-locked lasers (see e.g., [17]), which is based on the ultrafast dynamics of the gain in active semiconductor systems as described in [18]. This same mechanism is, in our opinion, the one that leads to active ML by counterpropagating injection of external pulses, as, for example, in [9,10,19,20].

4. CONCLUSIONS

We have experimentally studied the ML dynamics of a F8L which has a strongly dichroic SOA as gain medium in a simple cavity composed of off-the-shelf PM components. The simplicity of the device, whose only active element is the SOA and which does not require any polarization controller, makes it very attractive for several applications. Increasing the bias current in the SOA leads to self-starting ML at the fundamental repetition rate of 18 MHz just after crossing the lasing threshold, and further increasing the bias current into the SOA leads to the progressive emission of additional optical pulses in each round trip and eventually to mode-locked emission at increasingly high harmonics of the fundamental repetition rate, up to 2.45 GHz. The intensity autocorrelation of the amplified mode-locked pulses has an FWHM duration of 382 fs, which corresponds to a pulse duration of 247 fs. The physical mechanism leading to ML seems to be the interaction of the counterpropagating waves in the SOA, and work is in progress to better substantiate this hypothesis.