1. Introduction

Multiwavelength pulse source is one of the essential technologies for providing cost advantages by generating multiple signals simultaneously and thereby eliminating some transmitters. In commercial transmission systems, which are already reaching an advanced stage of development, wavelength-division-multiplexing (WDM) technique is very important, not only to increase capacity of transmission by multiplexing, but also to enhance network flexibility and scalability [1]. In such systems, return-to-zero (RZ) and non-RZ (NRZ) signals are well-known practical candidates for optimum modulation formats. In general, the RZ signal has better performance than NRZ signal due to a better match of the RZ pulses to optical filters used at the WDM demultiplexer and receiver. On the other hand, the NRZ signal has better dispersion tolerance than the RZ signal. Since these features mainly depend on the pulsewidth of the signals, wide pulsewidth controllability without intersymbol interference between neighboring pulses is useful to optimize the signal waveform according to various transmission characteristics of existing systems [2, 3].

As one of the multiwavelength pulse sources, spectrum-slicing supercontinuum (SC) light source can generate multiple pulse trains from a single optical clock. So far, several transmission experiments over 100-Gb/s using a SC light source were already reported [4, 5]. However, in such pulse sources, it is difficult to tune the output pulsewidth, which is uniquely determined by a profile of the spectrum-slicing filter. On the other hand, optical pulse generators based on an optical modulator such as electroabsorption modulator (EAM) or LiNbO₃ modulator (LNM) are also useful to generate multiple pulse trains with desired carrier frequencies easily. However, it is difficult to tune the output pulsewidth and operating wavelength of the generated signals in wide tuning ranges.

All-optical wavelength converters are very attractive for such signal generation and conversion techniques without limits of electrical bandwidth. Semiconductor optical amplifier (SOA)-based wavelength converters have advantages of being compact, low-switching energy, and wide range of operating wavelengths [6]. Since the output signal in such converters is synchronized with a data signal and the output wavelength is originated from a continuous-wave (CW) probe light with an arbitrary wavelength, the converter enables us to generate a signal pulse train with a desired clock frequency and carrier wavelength easily. Although the generated signal has frequency chirping on both of leading and trailing edges, it is effective for transmission performance improvement due to interaction between the chirp and the dispersion in transmission line [6, 7, 8]. By injecting multiple probe beams into the converter, it is also possible to generate multiwavelength outputs [9], that is, converting a clock pulse train into several synchronized pulse trains with each different wavelength. Another attractive feature of SOA-based wavelength converters is the pulsewidth tunable operation, which is achieved with a delayed interferometric configuration such as terahertz optical asymmetric demultiplexer (TOAD) [10], symmetric Mach-Zehnder interferometer (SMZ) [11], and delayed interference signal wavelength converter (DISC) [12]. In such converters, the pulsewidth of the output signal can be controlled by adjusting the time delay between CW probe light’s paths in the interferometer. Moreover, the output pulse with a delay setting much longer than an input data pulsewidth forms a rectangular-like shape [13], which is useful for return-to-zero (RZ) to non-RZ (NRZ) signal format conversion [7, 14] and performance improvement of an optical RZ-receiver [15].

Previously, we proposed a multiwavelength synchronized pulse generator by a single SOA-based delayed interferometric switch, and successfully demonstrated a simultaneous 16-channel pulse trains generation with wide pulsewidth tunability at 10 Gb/s [16]. In this paper, the performances of the multiwavelength synchronized pulse generator are investigated. We evaluate the characteristics of the generated signals for a wide pulsewidth tuning range. The transmission performance and the availability of our proposed method are also investigated.

2. Configuration and experimental setup

Figure 1(a) shows the configuration of multiwavelength optical pulse generator. The generator employs a TOAD configuration [10]. The interferometer in the generator consists of a short fiber loop with a polarization beam splitter (PBS) and a SOA that is placed at an arbitrary position in the loop with a variable delay line. CW probe beams are divided into clockwise (CLW) and counter-clockwise (CCW) traveling waves with equal amplitude at each orthogonally polarized state. When a linearly polarized clock pulse is injected and only traveling CLW into the loop, the CLW and CCW probe beams experience cross-gain modulation (XGM) and cross-phase modulation (XPM) in the SOA. The rise time of the probe phase change imparted by the clock pulse is primarily determined by a pulsewidth of the clock pulse, whereas the fall time is determined by a slow SOA’s carrier recovery time. Therefore, when the time slot of the pulse train is smaller than the carrier recovery time, the temporal phase changes of the CLW and CCW probe beams exhibit a sawtooth-like waveform. Between these probe lights, there exists a time delay determined by a delay setting Δt of the delay line into the loop as shown in solid and dashed curves of Fig. 1(b). After traveling the loop with 90° polarization rotation, the probe beams are recombined and directed to a 45° tilted polarizer through a circulator, half-wave plate (H), and quarter-wave plate (Q) at each orthogonally polarized state. Figure 1(c) shows the temporal differential phase response between the CLW and CCW probe beams. Since the differential phase response causes polarization rotation of the combined probe beams, the phase response governs whether the two beams at the polarizer’s output interfere constructively or destructively. The interferometric transmission T(t) of this configuration is described as

 

Fig. 1. (a) Configuration of multiwavelength optical pulse generator, (b) Temporal phase changes of the CLW (solid line) and the CCW (dashed line) probe beams, (c) Differential phase response between the two probe beams, and (d) Output intensity of the generated signal.

Download Full Size | PDF

where Φ(t) and Φ(t -Δt) are the temporal phase changes (dashed and solid curves in Fig. 1(b)), respectively. Φ ₀ is the constant phase offset between the two probe beams. The phase offset is adjusted by rotating the quarter-wave plate at the input of the polarizer, so that an optimal interferometric condition is achieved. As a result, the slow carrier recovery time is canceled out each other, and the output waveform with the pulsewidth determined by a delay setting Δt is obtained as shown in Fig. 1(d). It should be noted that the configuration to adjust delay setting is very simple compared with random data pattern switching such as wavelength conversion or optical time-division-multiplexing (OTDM) demultiplexing. In our case, the pulsewidth of the generated signal is solely determined by the relative position of the delay setting between arbitrary CLW and CCW signal pulses travelling into the loop. On the other hand, in the case of the random data switching, the precise offset position of the delay setting between signal pulses divided by the PBS is required [10]. As mentioned above, the synchronous timing and carrier-frequency of the output signal is uniquely determined by characteristics of the clock and probe signals, respectively. Thus, this technique enables us to generate synchronized clock signals with each desired wavelength even if the pulsewidth tuning is achieved by varying the delay setting Δt. In addition, it is possible to convert an incoming clock signal wavelength to an signal wavelength in a wide range of wavelengths, since a phase change based on the XPM in the SOA is weakly dependent on the operating wavelength [6].

 

Fig. 2. Experimental setup for multiwavelength synchronized signal generation and transmission of the generated signals.

Download Full Size | PDF

The experimental setup is depicted in Fig. 2. The 10 Gb/s optical clock pulse train with 18 ps pulsewidth was generated from an external-cavity laser diode (ECL) followed by a 10 GHz sinusoidally driven EAM with a swept frequency synthesizer (Willtron Ltd. Co., 6728B). The pulses were further compressed to 5.5 ps by using a spectrum-sliced supercontinuum light [17]. In this experiment, two kinds of pulse with each different pulsewidth were employed. The injected average power of the clock into the generator was 0 dBm. The CW probe lights (from 1 to 16 wavelengths) were generated from the ECLs according to number of the channels. To adjust input polarization states of these probe lights, polarization controllers at the output of each ECL and a polarizer at the input of the generator were employed. The injected total power of the probe lights was kept constant at 2 dBm, regardless of the numbers of the channel, in order to mitigate waveform change due to the carrier recovery time dependence on the input power of the probe lights. The details are described in Section 3. The SOA employed in the pulse generator was a polarization insensitive bulk type with 600-μm-long active layer. The injection current was 80 mA in the following experiments. At the output of the generator, the optical bandpass filter (BPF) was employed to demultiplex the generated signals, and remove the residual clock signal and SOA-induced amplified spontaneous emission (ASE) noise. In order to evaluate signal performances of the generated signals, the demultiplexed signal was amplified by an erbium-doped fiber amplifier (EDFA) with a BPF, and coded into a pseudorandom bit sequence (PRBS) with 2³¹-1 bit pattern length with a pulse pattern generator (PPG) and LiNbO₃ modulator (LNM). The phase shifter at the PPG input of the electrical clock was employed to tune the synchronous timing between optical clock and PRBS signals. The performances of the transmitted signals were estimated by using a 10 Gb/s error detector with a pre-amplifier.

3. Characteristics of the pulse generator

As described in Section 2, the pulsewidth of the output waveform can be controlled by adjusting the time delay of the delay line. In order to obtain an optimum waveform of the output pulse for various time delay settings Δt, the phase offset Φ ₀ is also important parameter to determine the characteristics of the output waveform. The relation between the delay setting Δt and the phase offset Φ ₀ is described as [18]

 

Fig. 3. (a) Time delay setting dependence of the optimum phase offset of the generated signals. The dashed line and circles show the calculated and measured phase offsets, respectively. (b), (c) The calculated (dashed) and measured (solid) temporal waveforms of the generated signals at phase offsets Φ ₀ of 1.16 π (Optimized:(b)), and 1.04 π (Non-optimized:(c)).

Download Full Size | PDF

where t_R is the repetition period of the clock pulse, and ΔΦ is the phase shift of the CW probe beams as shown in Fig. 1(b). The parameters used for our case are t_R = 100 ps and ΔΦ = 0.40 π, respectively. We estimated the phase shifts of the probe beams by fitting the calculated dynamic phase change and the measured phase shift of the probe beams. The phase change and waveform by the generator were calculated from a simple simulation model based on a rate equation for the carrier density of the SOA [19]. In this study, the injected clock pulsewidth was 18 ps, while the carrier recovery time was 102 ps, which was measured at the operating condition as mentioned in Section 2. The output wavelengths of the clock and generated signals were 1560 nm and 1550 nm, respectively. We measured the carrier recovery time of the SOA and the phase shift of the CW probe beams by means of the XGM and XPM waveforms fitting method [20] utilizing our proposed generator shown in Fig. 1(a). These waveforms originated from the CLW and CCW probe beams at the output of the polarizer are obtained by adjusting the half-wave and quarter-wave plates in the generator. When the phase difference Φ(t) - Φ(t - Δt) was set to π in order to realize an ideal constructive interferometric condition of the generator, the optimum phase offset Φ ₀ calculated from Eq. (2) for varying the delay setting is depicted in the dashed line of Fig. 3(a). The circles show the measured optimum phase offsets by adjusting the quarter-wave plate in the generator. The measured phase offsets were determined by monitoring the output waveform, which was detected with a 30 GHz-bandwidth photo diode and subsequently displayed on a digital-sampling oscilloscope. The calculated phase offset agree well with the experimental data. The influence of the temporal waveforms with the delay setting Δt of 40 ps and different phase offsets is depicted in Figs. 3(b) and (c). When the phase offset was optimized, the waveform exhibited a high extinction ratio as shown in Fig. 3(b), while a non-optimized phase offset state in Fig. 3(c) resulted in a poor extinction ratio of the waveform. These results indicate that it is important to adjust the optimum phase offset setting precisely, in order to obtain a high quality of the generated signal.

 

Fig. 4. Dependences of the carrier recovery time of the injected probe power for single- and multiple-channel configurations.

Download Full Size | PDF

In our proposed generator, the output signal wavelengths were originated from multiple CW probe beams, and output waveforms were determined by a pulsewidth of the clock signal, a time delay of the delay line, and a carrier recovery time of the SOA. Moreover, in the case of SOA-based switching operations such as our proposed scheme, CW probe lights also act as a holding beam, which enhance the carrier recovery time of the SOA [21]. This means that the recovery time becomes shorter as the injected probe power is increased. Therefore, when the total injected probe power becomes larger by increasing the numbers of the channel, the acceleration of the recovery time may influence the output waveform. Figure 4 shows the dependences of the carrier recovery time of the probe power injected to the converter. The pulsewidth and the repetition rate of the optical clock were 5.5 ps and 622 MHz, respectively. The input pulse energy was approximately 100 fJ, which corresponds to the average power of 0 dBm at 10 GHz repetition rate. In Fig. 4, the carrier recovery time became shorter as the input power of the probe light was increased for single-channel operation. The open circles in Fig. 4 show the dependence of the recovery time on the injected probe power per channel when the total injected power was kept 2 dBm for various channel numbers of the probe waves. The recovery times were almost constant for all channel configurations, although the recovery time of multiple-channel operation is usually shorter than that of single channel operation for the same input probe power level per channel. Therefore, in our proposed scheme, the injected CW probe power was kept constant, regardless of the number of the channels. In addition, small wavelength dependence of the recovery time was observed, even though the wavelengths of the multiple-channel were aligned randomly.

The generated signals are synchronized with the timing of the injected clock signal into the generator. To estimate synchronization performance in our proposed scheme, we investigated the timing jitter noise induced by the process of the signal generation. Since the RMS (root-mean-square) value of the timing jitter can be calculated from the single-sideband (SSB) phase noise spectrum of the generated signal [22], we compared the RMS timing jitter of the clock signal to that of the generated signals, and evaluated the synchronization characteristics of the signals. The noise spectrum was detected with a high-speed photodiode and measured electrically with an RF spectrum analyzer. Figure 5(a) shows the wavelength dependences of the generated signal and clock signal (open circle) versus the timing jitter for varying the clock signal wavelengths. For all operating wavelengths, the timing jitters of the generated signal were slightly larger than those of the clock signals, and the average induced jitter noises were approximately 10 fs to 20 fs. These indicate that the influences of the jitter noise induced by the generator were very small. Moreover, there were no dependences of the jitter noise for the clock and signal wavelengths, and the differences between clock and signal wavelengths. In the case of spectrum-slicing SC light sources, the induced timing jitter noise depends on the SC spectrum waveform, and becomes larger as the output signal wavelength is shifted from the clock signal wavelength [23]. On the other hand, in our proposed method, the low jitter deviations were observed in a wide range of wavelengths. The main factor of the noise in our method is due to ASE noise of the SOA. The clock signal fluctuation induced by the ASE noise influences the phase change imparted by the clock signal in the SOA, while the fluctuation of the probe beam influences its phase change and the generated signal. Therefore, the noise figure of the SOA is closely related to the synchronization characteristics of the generated signal. The noise figure of the employed SOA became higher at a shorter wavelength, and the measured noise figures were 7.0 dB at 1530 nm and 6.0 dB at 1565 nm, respectively. However, in this experiment, no influences of the wavelength dependence of the noise figure were observed. We believe that such little difference of the noise figure does not influence the synchronization characteristics of the generated signal. Figure 5(b) shows the dependence of the timing jitter on the time delay setting Δt. The wavelengths of the clock and probe signals were 1550 nm and 1540 nm, respectively. The jitter deviations for all delay settings were approximately ±10 fs. These indicate that the timing jitter dependences of the pulsewidth of the generated signal were very small. In other words, good synchronization performances in the wide pulsewidth tuning range can be achieved by our proposed generator.

 

Fig. 5. (a) Output signal wavelength and (b) time delay setting dependences of the RMS timing jitter.

Download Full Size | PDF

4. Improvement of transmission performance

In this section, transmission performance improvement by the use of the generated signals in realistic lines is experimentally studied. As mentioned in Section 1, the pulsewidth-tunability of the pulse generator enables us to optimize the transmission performance by adjusting the pulsewidth of the signal. A work by Yu et al. has shown that a RZ pulse generator based on four-wave mixing in fiber could improve the transmission performance by converting a pulsewidth of the pulse train into a longer pulsewidth [24]. However, no transmission performances for multiwavelength signal generation and its detailed comparison of the other signal formats were reported. Here, we demonstrate the transmission experiments for wide pulsewidth tunable and multiple-channel operations. The experimental setup is shown in Fig. 2. To compare the transmission performance of the waveform-converted signal to that of the clock (non-converted) and conventional signals, RZ and NRZ signals were employed. The RZ signal was generated by the EAM-based pulse generator, which obtain a pulse train with 18 ps pulsewidth as described in Section 2. This RZ signal is also used as the clock signal of the proposed pulse generator. Although the EAM imposes a large positive chirp, which limits the signal transmission distance, the RZ signal by EAM is widely used for conventional RZ signal because of its small size and low-drive voltage. On the other hand, the NRZ signal was generated by an ECL and subsequent a LNM, which enables us to exhibit chirp-free modulation. The generated signals were demultiplexed by a BPF with 0.6-nm bandwidth. The output power of the LNM at the input of the transmission line was kept –3 dBm to suppress nonlinear effects in fiber. The transmission qualities for transmission lines were estimated by using an error detector with a pre-amplified receiver.

 

Fig. 6. Time delay setting dependence of the error-free sensitivity for various cumulative dispersion values.

Download Full Size | PDF

For conventional transmission lines over 10 Gb/s, the transmitted signals were degraded by the dispersion and nonlinearity in fiber. Especially, the chromatic dispersion influences strongly the transmission performances in such systems. Here, we investigate the dispersion dependence on the performance of the signals by using various transmission lines with different cumulative dispersion values. These dispersions were made by the combinations of various optical fibers with different cumulative dispersions and fiber lengths, which were from –101.83 ps/nm to +460 ps/nm and from 0.5 km to 45 km, respectively. In these measurements, no dependence of the error-free (BER=10^-9) sensitivities on the fiber length were observed for various cumulative dispersions. Figure 6 shows the time delay dependence of the error-free sensitivities for various cumulative dispersion values in the case of single-channel operation. As the cumulative dispersion was increased, the optimum delay settings, which obtained a minimum sensitivity, were shifted to longer delay setting. This indicates that there exists an optimum delay setting corresponding to a cumulative dispersion, and the pulsewidth tuning enables us to optimize the transmission performance with an arbitrary cumulative dispersion, although the sensitivity at the optimum delay setting gradually degraded by increasing the dispersion. These characteristics are due to the combination of the pulsewidth of the generated signal and the dispersion in fiber. In general, the RZ signal with a short pulsewidth has a better sensitivity than that of the signal with a long pulsewidth, since it is due to a better match to the BPF used at the demultiplexer and receiver [25]. On the other hand, the RZ signal with a shorter pulsewidth is more sensitive to the dispersion. Thus, the minimum sensitivity, which is determined by the combination of the signal pulsewidth and the dispersion, was shifted to longer pulsewidth setting, and degraded by increasing the dispersion.

 

Fig. 7. Cumulative dispersion dependence of the error-free sensitivity for the conventional RZ and NRZ signals, and the waveform-converted signals with different operating wavelengths.

Download Full Size | PDF

To investigate the dispersion tolerance for various signal formats, we measured the error-free sensitivity dependence of the cumulative dispersion. Figure 7 shows the error-free sensitivities for the RZ and NRZ signals at 1550 nm, and the waveform-converted signals with different three wavelengths. All the delay settings of the converted signal were optimized to obtain a minimum sensitivity at each cumulative dispersion. Although the RZ signal has a better sensitivity of about 2 dB than NRZ signal at zero dispersion, the power penalty became drastically larger by increasing the dispersion. On the other hand, the power penalties of the converted signals could be kept lower by optimizing the delay setting. Especially at the cumulative dispersion of 460 ps/nm, the error-free sensitivities of the converted signals with delay setting of 70 ps were improved by about 3 dB in comparison with the NRZ signal. This is due to not only optimization of the pulsewidth of the waveform-converted signal, but also frequency chirp of the converted signal. The pulses of the converted signal have a red chirp at both of the leading and trailing edges, since the rapid phase changes by the XPM as shown in Fig. 1(b) cause the chirp on both edges of the pulses. When the pulses interact with the anomalous dispersion of the fiber, the leading and trailing edges with the red chirp are delayed with respect to the unchirped center of the pulses, and cause pulse steepening at the leading edges, and broadening at the trailing edges. This pulse formation induces the pulse peaking, and enables to improve the transmission performance [7, 8]. Therefore, the transmission performances of the signals by using the proposed generator can be drastically improved in comparison with the conventional RZ and NRZ signals. We believe that further improvements are also possible for a larger negative cumulative dispersion, since the waveform-converted pulses with the chirp on both edges interact with the normal dispersion of the fiber, and also induce the pulse peaking. The improved performances were also obtained in the wavelength range of 1530 nm to 1562 nm, because the pulse generator also acts as a wavelength converter. Since the RZ signal has a large positive chirp and a limited range of the operating wavelengths due to EAM characteristics, the signal conversion with our proposed method is effective to improve not only the sensitivity, but also the operating wavelength range of the conventional RZ signal.

We also demonstrate 70 km transmission experiments of the waveform-converted signal and NRZ signal for simultaneously generated 8 and 16 channels. These channel configurations were equally spaced with 200 GHz. The 70 km transmission line consisted of a 25 km standard single-mode fiber (SMF), 25 km dispersion-shifted fiber (DSF) with an anomalous dispersion, and 25 km DSF with a normal dispersion. These dispersion values at 1550 nm were 17 ps/nm/km, 4.32 ps/nm/km, and – 1.8 ps/nm/km, respectively. The total cumulative dispersion value was 403 ps/nm (as shown in dashed line of Fig. 7). Figure 8 shows the error-free sensitivities after 70 km transmission at 8 and 16 channels. The delay setting of 70 ps, which referred to the result of Fig. 7, and the phase offset at an optimized state were also fixed for all channels. The sensitivities of 8 and 16 channels were about 3 dB and 1.5 dB higher than that of the conventional NRZ signal, respectively. Thus, the improved transmission performances were obtained by adjusting the waveform according to the cumulative dispersion. The inset shows the eye pattern of the waveform-converted signal at a given channel of 16 channels configuration after 70 km transmission. These results indicate that our proposed method, which can optimize the transmission characteristics, is useful for WDM transmission systems.

 

Fig. 8. Error-free sensitivities of the waveform-converted signal and the NRZ signal after 70 km transmission at each channel for simultaneously generated 8 and 16 channels. The inset shows the eye pattern of the waveform-converted signal at a channel of 16 wavelengths after transmission.

Download Full Size | PDF

5. Conclusion

We have investigated performances of a widely pulsewidth-tunable multiwavelength pulse generation by a SOA-based delayed-interferometric switch. The optimized waveform of the output signal with an arbitrary pulsewidth can be obtained by adjusting simultaneously the time delay and phase offset in the generator. For multiwavelength signal generation with our proposed pulse generator, it is important that the injected probe power is kept constant, regardless of the number of channels. We have confirmed that the waveform-converted signals are fully synchronized with an optical clock signal, and obtain high signal qualities for the wide pulsewidth tuning range and multiple-channel outputs up to 16 wavelengths. We have also experimentally verified that the pulsewidth tuning by our proposed method enables us to optimize the transmission characteristics with an arbitrary cumulative dispersion. The waveform-converted signals for single- and multiple-channel operations show the improved transmission performances compared to the conventional RZ and NRZ signals. We conclude that our proposed method is attractive for optimizing the transmission performance in conventional WDM transmission systems.