1. Introduction

In comparison to either bulk or quantum-well active layer mode-locked semiconductor lasers, a quantum-dot Fabry-Perot laser exhibits a much narrower beat spectrum linewidth[1, 2]. For all-optical clock recovery, the timing jitter of the recovered clock signal is determined by both the phase noise transferred from the mode-locked laser and the timing jitter of the input data signal. A smaller beat spectrum linewidth accounts for a smaller amount of phase noise generated from the mode-locked laser and hence a smaller timing jitter of the recovered clock signal. The small linewidth of the QD-FP laser is attributed to four-wave-mixing in QD structures, which induces a strong phase correlation among the laser modes[1]. A significant reduction in the timing jitter of the recovered clock signal has been achieved at 40 Gb/s by using a mode-locked QD-FP laser, as compared to a bulk active layer self-pulsating distributed Bragg reflector laser[1]. This makes the QD-FP laser a very promising candidate for all-optical clock recovery in a 40 Gb/s 3R regenerator. However, since the emission spectrum exhibits multiple modes, the QD-FP laser cannot be used directly in a conventional 3R regenerator where the data signal is used as the pump signal for a nonlinear optical gate to modulate the recovered clock signal[3]. The multimode spectrum would preclude the use of wavelength division multiplexing and limit the transmission distance due to chromatic dispersion.

Previously a two-stage all-optical 3R (retiming, reshaping and reamplifying) regenerator has been demonstrated at 10 Gb/s using a self-pulsating DFB laser for clock recovery[4]. The regenerator uses cross-phase modulation (XPM) in a highly nonlinear fiber (HNLF) for retiming and self-phase modulation (SPM) in a HNLF for reshaping. In the retiming stage, the recovered clock signal serves as the pump signal to induce XPM on the input data signal (in contrast to a conventional nonlinear optical gate) and an offset filter is used to produce a retimed signal. The retimed signal is then reshaped by SPM in a HNLF and offset filtering[5]. In this type of all-optical 3R regenerator, the spectrum of the regenerated signal is determined by the amplitude response of the offset filter in the reshaping stage. Therefore, this approach can take advantage of the low timing jitter of the clock signal obtained from a 40 GHz QD-FP laser and accommodate the multimode spectrum of the laser. In the first demonstration of this type of regenerator, each stage was separately optimized and the input signal wavelength was not preserved due to different filter offsets[4].

In this paper, we demonstrate 40 Gb/s all-optical 3R regeneration that takes advantage of the low timing jitter achieved with a passively mode-locked QD-FP laser. To preserve the wavelength of the input signal, the filter offsets are chosen carefully so that both stages have good performance.The wavelength offsets for the filters in the retiming and reshaping stages, relative to the corresponding input signals, are -1 nm and 1 nm, respectively. The performance of the 40 Gb/s regenerator is characterized experimentally by measurements of the Q-factor, timing jitter and bit error ratio for the regenerated signal.

2. QD-FP laser and experimental setup

The single-section QD-FP laser used in the experiment has a quantum-dot active layer on an InP substrate, with a cavity length of about 1060 μm. The threshold current of the QD-FP is 14 mA. In the experiment, the QD-FP laser was biased at a DC current of 180 mA. The optical spectrum of the QD-FP laser is centered at 1528 nm[6]. More than thirty laser modes are contained within the 3-dB spectrum bandwidth. The passive mode-locking yields self-pulsating emission with a frequency of around 39.8 GHz. The measured 3-dB linewidth of the RF spectrum is 80 kHz[6]. Such a narrow RF spectrum (or beat spectrum) linewidth implies a strong phase correlation among the laser modes.

The experimental setup for 40 Gb/s all-optical 3R regeneration is illustrated in Fig. 1. A return-to-zero on-off-keying signal with a carrier wavelength of 1545 nm was generated using a cascaded Mach-Zehnder modulator based transmitter driven by a 39.8 Gb/s pulse stream with a 2³¹ – 1 pseudo random bit sequence. The data signal was applied to the QD-FP laser via an optical circulator (OC). The QD-FP laser has a narrow locking range (~5 MHz) due to the ultra-narrow beat spectrum linewidth. An optical bandpass filter (1.3 nm bandwidth) centered at 1529 nm was used to eliminate the data signal reflected from the QD-FP laser. Four laser modes were contained in the filtered spectrum of the recovered clock signal. The recovered clock signal and the data signal were coupled into a 2 km HNLF (nonlinearity of 10.6 W⁻¹km⁻¹, attenuation of 0.76 dB/km, and dispersion of -0.05 ps/nm-km at 1550 nm). A tunable optical delay (TOD) was used to adjust the relative time delay between the clock and data pulses thereby balancing the timing jitter of the rising and falling edges of the retimed signal. The clock signal was used as a pump signal to induce XPM on the data signal.

 

Fig. 1. Experimental setup for 40 Gb/s all-optical 3R regeneration. PC: polarization controller,

Download Full Size | PDF

The input powers to the HNLF for the clock and data signals were 16.3 dBm and 8.9 dBm, respectively. No stimulated Brillouin scattering was observed in the experiment since the QD-FP laser produces an optical clock signal with a multi-mode spectrum and a low spectral peak power. Thus, phase modulation of the clock signal was not necessary here while it was indispensable for the 10 Gb/s regenerator using a self-pulsating DFB laser[4]. A counter propagating Raman pump signal at 1425 nm with an input power of 26.8 dBm was employed to increase the XPM-induced spectral broadening and achieve low-noise distributed Raman amplification. It is worth noting that the Raman pump could be removed subject to the availability of a high-gain EDFA that provides sufficient XPM spectral broadening. While this would reduce the system complexity, it would also likely degrade the OSNR of the retimed signal. At the output of the HNLF in the retiming stage, an offset filter was used to select the XPM retimed components while eliminating the clock signal. For the reshaping stage, the retimed signal was boosted to a power of 21 dBm to induce SPM spectral broadening in a 4 km HNLF (nonlinearity of 10.6 W⁻¹km⁻¹, attenuation of 0.76 dB/km, and dispersion of -0.21 ps/nm-km at 1550 nm). For an offset filter centered at 1545 nm, the output signal had the same wavelength as the input signal. The offset filter of the retiming stage had a bandwidth of 1.0 nm. The bandwidth represents a trade-off between achieving a retimed signal with a small timing jitter (narrow filter to reject the original data signal) and a narrow pulse width to increases the SPM efficiency in the reshaping stage (wide filter). For the reshaping stage, the bandwidth of the offset filter was 0.55 nm. This yields a regenerated signal with a similar pulse shape and pulse width compared to the input signal.

3. Filter offset optimization

Figure 2 shows the dependence of the RMS timing jitter of the retimed signal and the Q-factor of the reshaped signal on the filter offset. All the measurements of the timing jitter and Q-factor were obtained using a sampling oscilloscope with precision timebase. The timing jitter is minimum at offsets of around ±1.0 nm. For the Q-factor measurement of the reshaped signal, the retiming stage was bypassed and the input signal to the regenerator was directly used to assess the reshaping stage. It is shown that a filter offset of larger than 1.0 nm in magnitude yields a Q-factor of more than 21 dB for the reshaped signal. As the spectra experienced different nonlinear optical processes (XPM for retiming and SPM for reshaping), the offset filtering imposes a different impact on the performance of the retiming and reshaping stages, as shown in Fig. 2. This means that the filter offsets of the retiming and reshaping stages should be selected carefully to obtain wavelength-preserving 3R regeneration with both low timing jitter retiming and high quality reshaping. It is clear that filter offsets of around ±1.0 nm are optimum in terms of the timing jitter and Q-factor for the regenerated signal. Therefore, filter offsets of -1.0 nm and +1.0 nm were used for the retiming and reshaping stages, respectively, to achieve wavelength-preserving regeneration.

 

Fig. 2. Timing jitter of the retimed signal (retiming stage only) and Q-factor of the reshaped signal (reshaping stage only) as a function of the filter offset. The OSNR of the input data signal is 22 dB (0.1 nm noise bandwidth).

Download Full Size | PDF

 

Fig. 3. Optical spectra measured with a resolution bandwith of 0.1 nm; (a) XPM broadening, (b) after retiming stage, (c) SPM broadening, and (d) after reshaping stage. Insets in (d) show the eye diagrams for the signal before (left) and after (right) the 3R regenerator with pulse widths of 10.1 ps and 9.2 ps, respectively (5 ps/div).

Download Full Size | PDF

Figure 3 depicts the optical spectra of the data signal at different stages of the 3R regenerator. Figures 3(a) and (b) are the XPM broadened spectra of the data signal before and after the offset filter, respectively. The SPM broadened spectrum is illustrated in Fig. 3(c). Figure 3(d) shows the SPM broadened spectrum after the offset filter, where the center wavelength corresponds to the input data signal. The insets in Fig. 3(d) shows eye diagrams for an undegraded input signal with a Q-factor of 24.6 dB and the corresponding output signal. The regenerated signal is similar to the input signal in terms of pulse shape and pulse width.The measured pulsewidths for the signals before and after regeneration were 10.1 ps and 9.2 ps, respectively.

 

Fig. 4. Q-factor improvement (in dB) of the regenerated signal with respect to the input signal. Insets show the eye diagrams for an input signal at a Q-factor of 16 dB and the corresponding regenerated signal (5 ps/div).

Download Full Size | PDF

4. Regenerative performance

 

Fig. 5. Timing jitter of the regenerated signal and the recovered clock signal as a function of the input signal timing jitter. Inset shows the optical trace for the recovered clock signal for an input jitter of 900 fs(10 ps/div).

Download Full Size | PDF

By decreasing the optical signal-to-noise ratio (OSNR), the Q-factor and timing jitter of the input data signal were degraded. The improvement in the Q-factor of the regenerated signal is illustrated in Fig. 4. For input signals with Q-factors below 22 dB (which corresponds to an OSNR of less than 26 dB with a noise bandwidth of 0.1 nm), a significant improvement is obtained with the 3R regenerator. For an input signal with a Q-factor of 16 dB, regeneration improves the Q-factor by 4 dB. As shown in the eye diagrams inset in Fig. 4, the regenerator suppresses noise for both the “one” and “zero” bits.

Figure 5 shows the dependence of the root-mean-square (rms) timing jitter of the regenerated signal and recovered clock signal on the rms timing jitter of the input signal. The regenerator effectively reduces the timing jitter; for an input signal timing jitter of 900 fs, the output timing jitter is less than 550 fs. The saturation of the output timing jitter for large values of the input timing jitter is due to the high tolerance of the QD-FP laser to the input signal jitter. As shown in Fig. 5, the timing jitter of the recovered clock signal is 350 fs (200 fs as determined by integrating the single sideband RF spectrum) and does not change appreciably as the input signal timing jitter increases from 400 fs to 900 fs. Some amount of amplitude jitter is transferred from the input data signal to the regenerated signal. This amplitude jitter affects the oscilloscope measurement of the timing jitter and contributes to the larger timing jitter for the regenerated signal compared to the recovered clock signal.

Figure 6 shows the measured bit error ratio (BER) as a function of the OSNR of the input data signal. In the experiment, an optical preamplifier was used for the receiver. The OSNR was adjusted by using a broad-band noise source to add optical noise to the input signal. The input power of the signal to the preamplifier was adjusted by a variable optical attenuator. For an input power of -15 dBm, the impact of the receiver noise on the BER is negligible and the BER performance is dominated by the properties of the regenerator. Compared to the result without regeneration, the regenerator does not cause a BER penalty while suppressing amplitude noise and timing jitter. To illustrate the improvement offered by the regenerator when receiver noise is subsequently added to the regenerated signal, the BER was also measured for a low input power of -30 dBm. This emulates the subsequent degradation that a regenerated signal would incur in a practical system. A 2.5 dB improvement in the required OSNR at a BER of 10⁻⁹ was obtained with the 3R regenerator.

Since the QD-FP laser is sensitive to the state-of-polarization of the input signal, an input data signal with a fixed state-of-polarization was used in the 3R regeneration experiment. It has been shown that polarization-insensitive clock recovery based on a QD-FP laser can be obtained by using a pre-processing stage [7], and that the two-stage 3R regenerator can be polarization-insensitive with careful design [4]. Consequently, a polarization-insensitive 3R regenerator based on a QD-FP laser for clock recovery is deemed possible.

 

Fig. 6. Measured BER as a function of the input signal OSNR.

Download Full Size | PDF

5. Conclusion

We have experimentally demonstrated the application of a mode-locked QD-FP laser in a wavelength preserving all-optical 3R regenerator at 40 Gb/s. The benefit obtained from the low timing jitter of the QD-FP laser and careful design of the retiming and reshaping stages has been characterized in terms of the Q-factor, timing jitter, and BER performance.