1. Introduction

All-optical wavelength conversion is one of the essential functions in high-capacity wavelength division multiplexing (WDM) optical communication systems and the related all-optical networks [1]. The capacity of WDM networks is mainly limited by the number of channels used, wavelength congestion at network nodes, and system management requirements for data re-routing on link failure. The use of wavelength converters within the networks will thus increase the flexibility of data traffic management schemes. Furthermore, the recent progress of WDM system technology has resulted in the demonstration of systems with a single channel data rate faster than 160 Gbit/s [2], and the practical implementation of the corresponding high-speed wavelength converters is thus a significantly important issue of current research focus.

In order to realize all-optical wavelength converters operable at a data rate greater than 160 Gbit/s the most important component would be a nonlinear optical medium, which is mainly nonlinear optical fiber or nonlinear semiconductor optical devices (semiconductor optical amplifier (SOA) or electroabsorption modulator (EAM)). Compared to the semiconductor devices with a limited operating speed caused by the carrier-lifetime related bandwidth [3], optical-fiber-based devices have inherent benefits such as fast response time (a few fs), no need for electrical bias, and no heat dissipation.

A number of researches have been performed so far for developing high-performance all-fiber wavelength converters using the nonlinear effects such as four-wave mixing (FWM) [4] and cross-phase modulation (XPM) [5]. In the previous-demonstrations of fiber-based wavelength converters, generally speaking in any fiber-based nonlinear signal-processing devices, one of main drawbacks was the compactness and stability issue originating from the long fiber length. A number of researches have thus been performed to overcome the issue by developing new types of novel highly-nonlinear fiber; for example, highly-nonlinear dispersion-shifted fiber (HNL-DSF) [6] and holey fiber (HF) [7]. Although a variety of types of wavelength converters using those fibers have been successfully demonstrated, those fibers still require lengths of tens of meters for the implementation of wavelength converters, prohibiting their practical applications.

Recently, we demonstrated the fabrication of a novel optical fiber with an ultra-high nonlinearity using a Bi₂O₃ material [8]. Owing to both the high material nonlinear coefficient and the small mode field area, such a high nonlinearity of γ~1360 W^-1.km^-1 was successfully obtained. This means that only a meter or less in length would be long enough to generate nonlinear optical phase shift sufficient for obtaining various signal-processing functions [9].

In this paper we go on to perform a research on the application of the developed Bismuth nonlinear optical fiber (Bi-NLF) technology for the implementation of high-speed nonlinear signal-processing devices. More specifically, we provide what we believe to be the shortest all-fiber high-speed wavelength converter ever demonstrated that uses an only 1-m-long Bi-NLF with a normal group-velocity dispersion (GVD). The 1-m Bi-NLF is fusion-spliced to standard single mode fibers (SMF’s) and its nonlinearity is ~1100 W^-1km^-1. The length of the whole device is almost comparable to that of fiber-pigtailed semiconductor devices.

Our wavelength converter is based on the conventional Kerr shutter principle using nonlinear birefringence and its performance is evaluated at a data rate of 80 Gbit/s. One noticeable point in the use of such a short length of nonlinear fiber in the Kerr shutter is that we do not need any special scheme for suppression of signal polarization instability caused by temperature-dependent local birefringence fluctuation within a long length of the conventional optical fiber [10]. So far the optical fiber based Kerr shutter approach has not attracted so much attention for high-speed nonlinear signal-processing applications mainly due to the polarization instability problem despite a range of practical benefits such as simple implementation, no wavelength shift, and good temporal resolution [11]. The 1-m-long Bi-NLF solves this instability problem and enables error-free wavelength conversion of an 80-Gbit/s input signal.

2. Bi-NLF fabrication and optical property characterization

 

Fig. 1. (a) Measured optical spectrum for nonlinear four-wave mixing in the Bi-NLF fabricated. (b) Measured nonlinear phase shift versus input signal power (The dashed line is a linear fit).

Download Full Size | PDF

In order to obtain a Bi₂O₃-based optical fiber with an ultra-high nonlinearity for this experiment, an optical fiber with a step-index structure was designed and fabricated. Note that GVD property of the fiber is also a very important parameter to take into account for its successful application to nonlinear optical signal-processing devices. In designing an optical fiber for all-optical nonlinear signal processing, a normal GVD is essentially required at the signal wavelength since the use of an anomalous dispersion fiber inevitably results in signal coherence degradation induced by modulation instability [12]. Due to very high normal dispersion characteristic of the Bi₂O₃ materials at a wavelength band of ~1550 nm, the GVD requirement was automatically satisfied without using any special waveguide design. The refractive indices (n) of the core and the cladding glasses were 2.22 and 2.13, respectively. We could easily ensure the sufficient thermal stability during the fiber drawing process since the core and the cladding glasses possess the same softening temperature and thermal expansion coefficient. The measured numerical aperture (NA) and the cut-off wavelength of the fabricated fiber were 0.64 and 1380 nm, respectively. The mode field diameter of this fiber was measured to be 1.98 µm and the fiber propagation loss at 1550 nm was 0.8 dB/m. The GVD at a wavelength of 1550 nm was measured to be -260 ps/nm-km. Note that only one meter piece of the fiber was used in this experiment and the overall GVD was just -0.26 ps/nm. No birefringence was observed in this particular 1-m-long fiber. Further details of the fabrication process are fully described in Ref. [8].

Fusion splicing of newly-developed optical fibers to conventional standard single mode fibers (SMF) is another important procedure from a perspective of practical implementation of fiber-based devices. Special care was taken of controlling arc strength and time due to the material discrepancy between Bi₂O₃ and Silica, and a high NA silica fiber (NA~0.35) was employed between them to achieve good intermediate mode matching. We define the splicing loss as the overall loss from the Bi-NLF to the SMF including the loss between the high NA fiber and the SMF. The input and output splicing losses were measured to be 2.6 dB and 3.2 dB, respectively, which are acceptable for this experiment although more efforts are required for their further reduction.

We next performed the nonlinear property characterization of the fabricated fiber by measuring the nonlinear phase shift of continuous wave (CW) light incurred by a beat signal propagating in the fiber [13]. Figure1 (a) shows measured optical spectrum for four-wave mixing between the two spectral components of the beat signal in the Bi-NLF and the corresponding nonlinear phase shift versus input signal power is shown in Fig. 1(b). From the nonlinear phase shift curve, the nonlinear parameter γ was estimated to be ~1100 W^-1km^-1 at 1550 nm which is ~1000 times larger than that of the conventional SMF.

3. Experimental results

 

Fig. 2. Experimental setup for the 80-Gbit/s wavelength converter using a 1-m-long Bi-NLF.

Download Full Size | PDF

Figure 2 shows the experimental setup for our proposed 80-Gbit/s wavelength converter using a 1-m-long Bi-NLF. 3-ps soliton pulses at a wavelength of 1550 nm were first generated from a 10-GHz harmonically, regeneratively mode-locked erbium-fiber ring laser (EFRL) and subsequently modulated to provide a 2³¹-1 pseudorandom bit sequence (PRBS) at 10 Gbit/s by use of a high-speed LiNbO₃ modulator. This 10-Gbit/s pseudorandom data stream was then amplified to a 15-dBm average power and multiplexed to an aggregate bit rate of 80 Gbit/s using a commercially-available three-stage passive multiplexer (MUX). The multiplexed data pulse stream was then amplified to an average power of ~25 dBm (after the 3-nm bandpass filter) and served as a control beam for wavelength conversion. A probe beam was generated from a CW external cavity laser tunable in a range from 1530 nm to 1580 nm and amplified to an 18-dBm optical power. The 80-Gbit/s control data stream and the probe beam were combined by a 50:50 coupler and launched into the 1-m-long Bi-NLF. Polarization controllers (PC’s) were included on both the control and the probe launching paths into the Bi-NLF so that they were polarized at the fiber output end at a 45° angle of their polarization directions. In order to ensure the 45° angle, a fiberized polarizer attached with a degree-scaled half-wave plate at the front end was employed. The procedure for ensuring the 45° is as follows. We first adjusted the polarization of the probe beam to exhibit the maximum power at the polarizer output and then rotate the half-wave plate attached to the polarizer by 45°. We then launched the pump beam and adjusted its polarization to exhibit the maximum power at the polarizer output. In this adjustment the relative polarization direction of the two beams was assumed to maintain along such a short length of the Bi-NLF since the polarizer was located at the output end of the Bi-NLF. The polarizer was operating as a probe beam blocker without control pulses. A 3-nm bandpass filter was employed at the output of the polarizer to filter out residual control beam components.

Our wavelength converter is based on the conventional Kerr shutter using nonlinear birefringence. XPM between the control and the CW probe beams results in nonlinear birefringence for the CW beam where these beams overlap temporally within the fiber. This nonlinearly-polarization-rotated signal can then be converted to a wavelength-converted signal by passing it through a polarizer, which serves to eliminate the residual CW beam with no polarization rotation [14]. In order to implement a stable and high-performance Kerr shutter using a conventional optical fiber an additional scheme for suppression of signal polarization instability induced by temperature-dependent local birefringence fluctuation owing to its long fiber length is essentially required [10]. However, the use of a very short, ultra-high nonlinearity fiber could significantly reduce the polarization instability problem [15].

We then constructed an 80-to-10-Gbit/s data-demultiplexing switch to assess the wavelength conversion performance in terms of bit error rate (BER). The 80-to-10-Gbit/s demultiplexing switch is also based on a 2-m-long Bi-NLF. Our demultiplexing switch basically uses the principle of the Kerr shutter and its switching performance is further enhanced by the additional use of the wavelength blueshift of data pulses, which is induced by cross-phase modulation from the trailing edge of the control pulse. Further details of our demultiplexing switch are fully described in Ref. [9]. The demultiplexed signal was then fed into a receiver module with no optical preamplifier to measure BER performance.

 

Fig. 3. Measured output optical spectrum after the polarizer together with that after the filter.

Download Full Size | PDF

First, we characterized our wavelength converter in the spectral domain. The measured output optical spectrum after the polarizer is shown in Fig. 3. The CW probe beam is readily converted into an 80-Gbit/s signal due to nonlinear birefringence induced by the control signal. It is clearly evident that high spectral quality beam is successfully obtained at the output with such a short length of Bi-NLF

 

Fig. 4. (a) Measured autocorrelation traces of the 1555-nm original data pulses and the 1545-nm wavelength-converted pulses. (b) Measured temporal width of the wavelength-converted pulses as a function of probe beam wavelength.

Download Full Size | PDF

We then performed the autocorrelation measurement of the wavelength-converted pulses. Figure 4(a) shows autocorrelation traces of the 1555-nm original data pulses and the 1545-nm wavelength-converted pulses. A Gaussian wavelength-converted pulse of 3.3-ps temporal width was obtained and the calculated time-bandwidth product was ~0.37. We then measured the temporal width of the wavelength-converted pulses as a function of probe beam wavelength for the purpose of confirming the wavelength tunability and the results are plotted in Fig. 4(b). The temporal width of the converted pulses was observed to linearly increase from ~3 to ~4 ps as the probe wavelength decreases from 1550 to 1530 nm. The walk-off between the control and the probe beams is believed to lead to the uneven output pulse widths over the 20-nm bandwidth.

Finally, we performed the BER measurement on one wavelength-converted signal at 1545 nm to assess the system impact of using our proposed wavelength converter. Figure 5(a) shows measured eye diagrams for the input control and the wavelength converted 80-Gbit/s pulses together with those of the 10-Gbit/s back-to-back and the 10-Gbit/s demultiplexed signals. High quality of wavelength conversion and subsequent demultiplexing is clearly evident from the graph. The conversion performance was quantified with BER, and error-free operation was readily achieved with a power penalty of ~2 dB relative to the 10-Gbit/s back-to- back, as shown in Fig. 5(b). The power penalty is believed to be associated with multi-path interference of the probe beam due to reflection at the splicing points between the Bi-NLF and the high NA silica fibers. Owing to the material refractive index discrepancy at the splicing points, some degree of light reflection was observed. Note that the 2-dB penalty includes the penalty induced by the 80-to-10-Gbit/s demultiplexer, and thus the penalty associated with wavelength conversion should be less than 2 dB.

 

Fig. 5. (a) Measured eye diagrams for the 80-Gbit/s input control and the 80-Gbit/s wavelength-converted signals together with those of the 10-Gbit/s back-to-back and the 10-Gbit/s demultiplexed signals (input pulses: 1555 nm, wavelength-converted pulses: 1545nm). (b) Measured BER’s for the 80-to-10-Gbit/s demultiplexed wavelength-converted signal and the 10-Gbit/s back-to-back signal.

Download Full Size | PDF

4. Conclusion

We have experimentally demonstrated the use of our fabricated 1-m-long Bi₂O₃-based stepindex type optical fiber with an ultra-high nonlinearity of ~1100 W^-1km^-1 at 1550 nm for wavelength conversion of high-speed OTDM signals. The Kerr-shutter-based wavelength converter showed stable operation without signal polarization instability. Error-free wavelength conversion of an 80-Gbit/s data signal was readily achieved and wavelength tuning over a 20-nm range was also obtained. The application of the proposed wavelength converter for over 160-Gbit/s systems should be possible with both further increase in the control pulse power and the use of stable 160-Gbit/s multiplexed data pulses. More efforts are also required to reduce both splicing loss and GVD of Bi-NLF’s. The Bi-NLF has been investigated as a strong candidate for practical implementation of a range of nonlinear optical signal processing devices, and further applications are certain to be found in the near future.