1. Introduction

Fiber lasers have received a great deal of attention recently as low noise sources of femtosecond pulses at a variety of wavelengths. One of the first demonstrations of a passively mode-locked fiber laser was the figure eight fiber laser [1,2], which makes use of a nonlinear amplifying loop mirror (NALM) for modelocking [3].

One of the primary drawbacks of femtosecond fiber lasers is sensitivity to environmental fluctuations. Figure eight lasers for example can drop out of modelocking when subjected to mechanical perturbations or significant temperature changes. These instabilities can be largely eliminated however by the use of polarization maintaining (PM) fibers. PM fiber lasers modelocked with semiconductor, saturable absorber mirrors have been demonstrated to be robust sources of femtosecond pulses [4].

In spite of the environmental sensitivity, figure eight lasers, as well as polarization rotation modelocked ring lasers [5], remain popular lab tools because of the simple cavity layouts, inexpensive components, and easy construction. While the polarization rotation modelocked ring laser is not amenable to an all PM design, there is nothing in the cavity layout of the figure-eight laser that precludes the use of PM fibers.

In general, there has been very little work on PM figure-eight lasers. A self-starting, fully polarization maintaining figure-eight laser has been demonstrated [6], however the laser operated at a relatively low repetition frequency of 525 kHz and produced pulses of only 2.3 ps in width. The laser cavity was not dispersion managed which is important for achieving the shortest possible pulses in fiber lasers [7]. Most other work with polarization maintaining, passively -modelocked figure-eight laser have used PM fiber in only part of the cavity [8]. In this work we demonstrate, to the best of our knowledge for the first time, an all polarization maintaining passively modelocked, femtosecond figure-eight fiber laser with a dispersion managed cavity. The laser operated with repetition frequencies between 16 and 32 MHz and produced pulses with a spectral width of 10 nm that were de-chirped outside the laser cavity to a FWHM of 427 fs.

2. Polarization-maintaining, figure-eight fiber laser

A schematic of the laser cavity is shown in Fig. 1. In typical figure eight lasers mechanical polarization controllers are used to align the state of polarization to a point conducive to mode-locking. In this system, all fibers and components were PM, and no polarization controllers were used. Between 1 and 2 m of a specially designed, PM, high concentration Er-doped fiber was used. The PM Er fiber had a peak absorption at 1530 nm of 27.5 dB/m, a beat length at 1530 nm of 14.3 mm and dispersion parameter at 1530 nm of -34.4 ps/nm-km. The anomalous dispersion of the PM single mode fiber was used to balance the normal dispersion of the erbium-doped fiber.

The laser was pumped by a 980 nm laser diode delivering a maximum of 350 mW of power through a 980/1550 PM WDM. A PM amplitude modulator with a bandwidth of 2 GHz (JDS Uniphase) was used to initiate mode-locking. However, under normal femtosecond operation the RF drive was turned off, and passive mode-locking was maintained with the NALM. A single polarization, PM isolator performed the dual tasks of providing unidirectional operation in the non amplifying loop, as well as adding polarization dependant loss to the cavity, to provide operation on a single polarization state.

When the laser was first turned on, it operated in CW mode, even at the highest available pump power. The spectrum showed several narrow spikes, plotted in Fig. 2(a). When the amplitude modulator was driven with a purely sinusoidal drive at the cavity repetition frequency, the spectrum broadened slightly (red line, Fig. 2(a)) and a pulse train was observed using a fast photodiode and an oscilloscope. An expanded view of one of the peaks in Fig. 2(a) is shown in Fig. 2(b) on a linear scale. Assuming sech² pulses, the blue curve in Fig. 2(b) with the amplitude modulator on corresponds to bandwidth limited pulses with a FWHM of 18 ps. Femtosecond pulse operation was not observed with the amplitude modulator on and driven by a purely sinusoidal drive.

 

Fig. 1. Schematic of the polarization maintaining figure-eight erbium laser.

Download Full Size | PDF

 

Fig. 2. (a) Lasing spectrum when the RF modulator is off (red line) and when the amplitude modulator is driven by a pure sine wave at the cavity repetition frequency (blue line). (b) Close up of one peak in the spectrum from (a), plotted on a linear scale

Download Full Size | PDF

 

Fig. 3. Spectrum of the laser when passively modelocked. (a) Spectrum when 1.5 m of Er-doped fiber was used, for two different amounts of PM-SMF in the cavity. (b) Spectrum for f_rep=19.6 MHz from (a), plotted on a linear scale.

Download Full Size | PDF

When the RF drive to the amplitude modulator was turned off however, the laser was capable of transitioning into femtosecond operation. Once femtosecond pulsing was observed, the modulator was left off and pulsing was maintained by the passive modelocking operation of the NALM. Further details of the transition between the long, actively modelocked pulses, and the passively modelocked femtosecond pulses are given in Section 3. By driving the modulator at harmonics of the fundamental cavity repetition frequency, we could observe active harmonic modelocking, however, in these cases, when the modulator was turned off, we never observed the laser to transition to passive modelocking. The transition to passive modelocking was only observed when the modulator was driven at the fundamental cavity repetition frequency.

The spectrum of the laser while passively modelocked is shown in Fig. 3. For this particular case, 1.5 m of PM Er was used. Figure 3(a) shows the spectra obtained for two different amounts of SMF in the cavity. The effect of managing the net dispersion in the cavity by balancing anomalous and normal dispersion fibers can be seen in the spectral width and the spacing of the Kelly sidebands [7]. Figure 3(b) shows the spectrum for a repetition frequency of 19.6 MHz plotted on a linear scale. The average output power for this spectrum was -13 dBm.

 

Fig. 4. (a) Passively modelocked pulse train measured with a fast photodiode and oscilloscope. (b) Measured auto-correlation of the passively modelocked pulse train. The FWHM of the correlation was 659 fs, corresponding to a FWHM of 427 fs for a sech² pulse.

Download Full Size | PDF

 

Fig. 5. Broadest measured modelocked spectra for different lengths of erbium-doped fiber in the cavity

Download Full Size | PDF

Measurements of the temporal characteristics of the laser when passively modelocked are shown in Fig. 4. A typical pulse train, with the laser operating at 27.5 MHz is shown in Fig. 4(a). Figure 4(b) shows a measured autocorrelation from the pulse train in Fig. 4(a). The pulses exiting the laser had a small amount of positive chirp, requiring a short length (~1 m) of normal dispersion fiber spliced to the output pigtail. The normal dispersion fiber was cut back until the shortest correlation was measured. The FWHM of the correlation was 659 fs, which corresponds to a pulse FWHM of 427 fs assuming a sech² pulse. The FWHM of the spectrum corresponding to this correlation was 9.3 nm which gives a time-bandwidth product of 0.50, somewhat higher than the value of 0.31 expected for sech² pulses. This discrepancy is most likely due to the negative dispersion fiber used to de-chirp the pulse outside the cavity. This fiber had a positive dispersion slope at 1535 nm, compared to the negative dispersion slope of the PM erbium fiber used intra-cavity.

The PM erbium-doped fiber was cut back, and at the same time, the single mode fiber in the cavity was trimmed in order to produce the broadest modelocked spectrum possible. The results of this measurement are shown in Fig. 5. In this way, the measured pulse repetition rate was varied from as low as 16 MHz to as high as 32 MHz.

Once the laser was modelocked the pulsed operation was extremely robust. The fibers could be strongly mechanically perturbed without interrupting the pulsed operation or changing the lasing spectrum. The robust nature of the modelocking was attributable to the all-PM configuration. The measured polarization extinction ratio at the output port of the laser was 13 dB.

 

Fig. 6. Measurement of the temporal evolution of the laser output when the RF drive power to the amplitude modulator was switched off. (a) Instance when laser reverted to CW oscillation. (b) Instance when the laser transitioned to passive modelocking.

Download Full Size | PDF

3. Initiation of passive modelocking

Although the laser was very stable once passive modelocking was achieved, self-starting operation of the passive modelocking was not observed. The amplitude modulator was used to provide picosecond pulses that seeded the femtosecond pulse operation. Once the laser was passively modelocked and operating with femtosecond pulses, the amplitude modulator was kept off. Details of the transition between active modelocking with picosecond pulses, and passive modelocking with femtosecond pulses are provided in this section.

Initially the amplitude modulator was driven with a purely sinusoidal drive with a frequency equal to the fundamental cavity frequency. Doing so produced an actively modelocked pulse train with a bandwidth limit of roughly 18 ps. When the drive power to the amplitude modulator was turned off, in the time domain, relaxation oscillations were observed that quickly subsided, and the laser returned to CW operation. This behavior is shown in Fig. 6(a).

Occasionally however, the laser was observed to transition to passive modelocking when the RF power to the amplitude modulator was turned off. The temporal behavior in such a case is plotted in Fig. 6(b). In this instance, the picosecond pulses left over from the active modelocking after the modulator was turned off provided a sufficient seed to initiate passively modelocked femtosecond pulses.

The laser most easily achieved pulsed operation at the highest pump powers. In turn this meant that when the laser transitioned to passive modelocking it typically oscillated with multiple pulses in the cavity, and the spectrum showed a corresponding interference pattern as plotted in Fig. 7. However, once passive modelocking was initiated the pump power could be reliably turned down and the extra pulses would drop out, leaving only a single pulse in the cavity, also plotted in Fig. 7. The spectra in Fig. 7 have been offset vertically for clarity.

In general, the transition from active modelocking with the RF power on to passive modelocking with the RF power off would occur roughly once every 20 to once every 100 times the RF power was turned off, or even less frequently. We also observed that achieving passive modelocking became much more difficult the more additional fiber was added to (or removed from) the cavity and the pulse spectrum got narrower.

In order to make the transition from active to passive modelocking more reliable, we added an additional sinusoidal amplitude modulation to the RF power that then drove the optical amplitude modulator. The RF power, which was initially a sine wave with a frequency in the MHz range, equal to the cavity repetition frequency, was additionally amplitude modulated with a sine wave in the kHz frequency range. In general, with the amplitude modulated RF power, the laser produced Q-switched pulses roughly 1 μs in width underneath which lay ps pulses. Such a Q-switched, actively modelocked pulse train is shown in Fig. 8(a). In this case, the RF power was amplitude modulated at 38 kHz.

 

Fig. 7. Spectrum of the laser at the highest pump power immediately after transitioning to passive modelocking, showing multiple pulses in the cavity, compared to the spectrum after the pump power has been turned down so that only a single pulse was in the cavity. Spectra have been offset vertically for clarity.

Download Full Size | PDF

 

Fig. 8. Q-switched, modelocked pulse trains observed when the RF power was amplitude modulated. (a) Modulation frequency = 38 kHz. (b) Modulation frequency = 49 kHz.

Download Full Size | PDF

At certain resonance frequencies, when the amplitude modulation on the RF drive power was equal to the relaxation oscillation frequency seen in Fig. 6(a), or a subharmonic of the relaxation frequency, then bursts of passively modelocked pulses could be observed. Figure 8(b) shows the time trace when the amplitude modulation frequency was 49 kHz, which is approximately half the oscillation relaxation frequency observed in Fig. 6. One Q-switched spike can be seen to seed a burst of femtosecond pulses, that lasts approximately 50 μs.

Under such conditions the bursts of modelocked pulses were observable in the optical spectrum as well. Figure 9 shows the optical spectrum when the RF power was amplitude modulated. Two spectra are plotted for a modulation frequency of 38 kHz, compared to a modulation frequency of 49 kHz. The spectra have been offset vertically for clarity. When the modulation frequency approached the resonance of the oscillation relaxation frequency, the pulse bursts became visible in the optical spectrum as a broad noise plateau.

With the amplitude modulated RF power driving the optical amplitude modulator, the transition to passively modelocked pulses could be obtained much more readily. For a given length of erbium fiber, the SMF length was required to be within 1 m of the zero net dispersion point of the cavity. By fine tuning the MHz frequency of RF power, passive modelocking could be observed even with the RF power on, and femtosecond pulses could be maintained when the RF power was turned off nearly 100% of the time.

 

Fig. 9. Optical spectrum observed when the RF drive power was amplitude modulated at 38 kHz compared to 49 kHz. The spectra have been offset vertically for clarity.

Download Full Size | PDF

4. Discussion and Conclusions

By generating an actively Q-switched, modelocked pulse train, the figure-eight laser was readily able to enter the femtosecond pulsed regime. This situation can be understood by estimating the peak power of the pulses under the various operating conditions.

When the laser was passively modelocked with a single pulse in the cavity the minimum average power at the output port required to maintain modelocking was -13.8 dBm. For pulses with a repetition rate of 27 MHz, the minimum pulse width was approximately 400 fs. Thus the peak pulse power was approximately 3.8 W. These pulses corresponded to the minimum peak power fs pulses needed for switching in the NALM.

When the laser was actively modelocked at the highest available pump powers, the output power was -12 dBm, and the minimum pulse width was approximately 18 ps, for a pulse peak power of approximately 0.13 W. However when the laser was actively Q-switched and mode-locked at the maximum pump power, the output power increased to -10.5 dBm. Furthermore the output of the laser was a 49 kHz Q-switched envelope with a width of 0.93 μs, and a 27 MHz, 18 ps pulse train underneath the Q-switched envelope. This pulse train corresponds to peak powers of 4 W. So, by generating the Q-switched, actively modelocked pulse train, the laser can reach the peak powers required for switching in the NALM, and thereby readily achieve passive modelocking.

Ideally one would like to achieve self-starting operation in order to be able to eliminate the amplitude modulator altogether. In the prior demonstration of the PM figure-eight laser, it was observed that self starting was only obtained for certain coupling ratio’s for the center splitter [6]. Therefore, going forward we plan to investigate alternate splitting ratios in the dispersion managed figure-eight laser.

In conclusion, we have demonstrated a polarization maintaining, passively modelocked, figure-eight fiber laser with a dispersion managed cavity, producing pulses with a 10 nm spectral width at repetition frequencies between 16 and 32 MHz. The average power was -13 dBM, and the polarization extinction ratio was 13 dB. When passively modelocked, the laser was environmentally stable and extremely robust; mechanical perturbations did not interrupt mode-locking or change the spectral shape of the pulses. The pulses were de-chirped outside the laser cavity to a minimum width of 427 fs.