1. Introduction

In-line all-optical signal regenerators are likely to play an important role in future large-scale photonic networks, offering considerable increases in the transmission distances that can be achieved and additional flexibility in the network design. In their simplest form, signal regeneration devices cancel the effects of amplitude noise accumulation in transmitted signals (2R regeneration), and they typically employ a fast nonlinear element with a step-like power transfer characteristic [1]. Such elements can be implemented in optical fibers by making use of the ultrafast Kerr nonlinearity. Fiber-based nonlinear devices offer the advantages of being transparent to the bit rate (to rates that exceed several 100’s of Gb/s) and of not degrading the signal-to-noise ratio or the extinction ratio of the signal [2]. However, because of the relatively low nonlinearity of silica, fiber based regenerator devices usually have to be quite long-typically of order >100 m in length. The implementation of more compact nonlinear devices represents a challenge, and is currently a hot topic in fiber technology research. Recently, promising new highly nonlinear glasses have been developed, such as Bi-Oxide [3], lead silicate [4] or Chalcogenide glasses [5]. Fibers drawn from these compound glasses have shown to exhibit effective nonlinearities per unit length which are some orders of magnitude higher than standard single mode fibers, allowing for a drastic reduction in the length requirements of fiber-based nonlinear devices [6] and enabling the implementation of meterlong nonlinear switches with improved performance in terms of stability and input power requirements.

There have been several demonstrations of fiber-based 2R regenerators that make use of either self-phase modulation (SPM), cross-phase modulation (XPM), four-wave mixing (FWM), or even stimulated Raman scattering (SRS) [7–10]. Amongst these, SPM-based schemes have the advantage that no additional laser source is required, making the regenerator structure simpler [11]. In this paper, we demonstrate the performance of a 2-m long all-optical regenerator based on a bismuth-oxide-based nonlinear fiber (Bi-NLF) with an effective nonlinear coefficient γ of ~1100 W-1km-1. The 2R regeneration scheme we used is based on spectral broadening of the signal due to SPM in the Bi-NLF followed by narrowband offset filtering. The normal dispersion of the Bi-NLF at telecommunication wavelengths favored the generation of a smooth SPM spectrum [12], free from any noise arising from modulation instability, which would potentially be observed in anomalously dispersive fibers [13]. We present a complete characterization of the pulse spectral broadening, which is supported both by experimental data and numerical analysis, and which proves the suitability of this fiber for 2R regeneration. In our system experiments we demonstrate significant noise suppression and bit-error rate improvement of heavily degraded 10-Gb/s and 40-Gb/s returnto-zero (RZ) signals.

2. Experimental set-up and results

For the implementation of the SPM-based 2R regenerator, the amplified data signal propagates through the nonlinear fiber, and the spectrally broadened signal is then passed through a filter, which is centered at a wavelength slightly offset from the original signal [6]. At low powers (i.e. when the transmitted symbol is nominally a “zero”), the signal does not experience any spectral broadening during its propagation through the fiber, such that the signal is rejected by the filter (i.e. suppression of the “zero” bits). However, if the signal power is high enough (i.e. when the transmitted signal is nominally a “one”), then a portion of the SPM-broadened spectrum passes through the filter. Furthermore, since the spectral density of the broadened spectrum at the filter passband can be made to be relatively insensitive to the peak power of the input pulse, any amplitude fluctuations on the ‘one’ bits are reduced in the process (i.e. equalization of the “one” bits). Clearly, this behavior depends strongly on the launched power, filter offset and fiber characteristics. Note that in our investigation, we have aimed at achieving the optimum performance of the 2R regenerator without trying to minimize the input power-fiber length product. For this reason we consider relatively high power levels and wavelength detunings.

Our experimental set-up is shown in Fig. 1. The pulses at the transmitter have a FWHM of ~5 ps, and are derived from a 10 GHz actively mode-locked erbium fibre ring laser (EFRL) operating at 1555.7 nm. The pulses are then modulated with a 2³¹-1 pseudorandom bit sequence using a lithium niobate modulator. We artificially introduce amplitude noise to the pulses using a second modulator driven by a 15 MHz sinusoidal signal. The amount of induced amplitude jitter can be varied both by varying the amplitude of the frequency modulation as well as by degrading the extinction ratio between marks and spaces of the data, facilitated by choosing a non-optimum bias voltage for the modulator. This noise generation

scheme has allowed us accurate and flexible control over the noise introduced on the one and zero levels of our signal. The signal pulses are subsequently amplified by a high power Er/Yb amplifier and launched into the 2-m-long Bi-NLF. The nonlinear coefficient, dispersion, dispersion slope and loss of the Bi-NLF, as measured at 1550 nm, are 1100 W^-1km^-1, -260 ps/nm·km, 0.947 ps/nm²/km and 0.9 dB/m respectively. Both ends of the Bi-NLF are spliced to SMF connectors and the splicing losses at each end of the fiber are ~3 dB. Finally, a tunable 0.6 nm bandpass grating filter is used to filter the output of the regenerator. Note that the power levels quoted herein, correspond to the powers at the input of the Bi-NLF patchcord, and not to the input of the Bi-NLF itself.

 

Fig. 1. Experimental set-up of the 2-m-long Bi-NLF-based 2R regenerator at 10 Gb/s and 40 Gb/s.

Download Full Size | PDF

 

Fig. 2. (a) Experimental SPM-induced spectral broadening for various input peak power levels for the 10 Gb/s case. The spectra are vertically offset for ease of reading. Resolution: 0.5 nm. Intensity and chirp profile of the pulses at (b) the input, and (c) the output of the regenerator measured using SHG-FROG.

Download Full Size | PDF

In order to assess the performance of the nonlinear thresholder we first studied the SPM-broadened spectra of the pulses for various input signal powers (see Fig. 2(a)). It can be seen that the combined effects of SPM and normal dispersion result in a smooth, almost flat-topped spectrum, which is desirable for 2R regeneration since it leads to a flatter power transfer function [6]. It is worth noting the asymmetric broadening behavior between the longer and shorter wavelengths is indicative that the input pulses are not symmetric transform-limited input pulses since such pulses would give symmetric SPM-induced spectral broadening (see red curve in Fig. 3(a)). Consequently, we decided to characterize the pulses entering the regenerator system using second harmonic generation frequency-resolved optical gating (SHG-FROG). As can be seen in Fig. 2(b), the chirp of these pulses exhibits higher order components, which are induced by a filter used in the transmitter block for the generation of the ~5 ps input pulses. By including this measured chirp within our numerical Split Step Fourier technique based nonlinear pulse propagation calculations we were able to accurately account for the observed experimental behavior (see blue curve in Fig. 3(a)). In order to make use of this flat spectral region during the implementation of the 2R regenerator, we tuned the filter towards longer wavelengths relative to the original signal wavelength, and achieved optimum regeneration performance for an average input power of ~31 dBm and a filter offset of 6.1 nm, (see dashed curve in Fig. 3(a)). We have also characterized the pulses at the output of the regenerator using the SHG-FROG technique, (see Fig. 2(c)), and obtained good quality, unchirped pulses with a FWHM of ~6.0 ps. Fig. 3(b) shows a measurement of the transfer characteristic of the regenerator. The relatively large offset wavelength, combined with the very flat SPM spectrum, ensures a clear two-level response between low and high powers.

 

Fig. 3. (a) Optical spectra at the input of the Bi-NLF, after propagation in the fiber (experimental and simulated traces) and at the output of the 0.6 nm filter, for Pin~31 dBm and operation at 10 Gb/s. The spectra are vertically offset for ease of reading. Resolution: 0.5 nm. (b) Transfer function of the regenerator. Note that the power levels correspond to the powers at the input of the Bi-NLF patch cord, and not to the input of the Bi-NLF itself.

Download Full Size | PDF

We then examined the noise-rejection properties of the system. Fig. 4(b) and (d) show the data signal at the input and output of the system when no amplitude noise is added to the pulses, and demonstrate that the reshaping system does not in itself introduce any additional noise to the signal. We next introduced noise to the input signal with a standard deviation in the amplitude of marks and spaces of 11% and 7% respectively, see Fig. 4(c). (Note that due to noise restrictions in our DCA-based measuring system, a 2% standard deviation in amplitude is measured for a noise-free signal). Fig. 4(e) shows the eye diagram of the received signal at the output of the bandpass filter, and indicates that the noise at the spaces has been suppressed, while amplitude equalization has been performed at the marks. The normal dispersion of the Bi-NLF at the wavelengths of operation ensures suppression of particular detrimental nonlinear effects, such as modulational instability, that can give rise to additional amplitude noise on the signal. The measured BERs as a function of the launched optical average power into the receiver are presented in Fig. 4(a). Penalty-free operation of the regenerator for a noiseless input signal is confirmed, while an improvement in the receiver sensitivity of more than ~5 dB is achieved for the case when amplitude noise is deliberately applied to the incident data pulses. This significant improvement is due to the flat-topped spectrum achieved in the Bi-NLF.

We have also assessed the Bi-NLF 2R regenerator at 40 Gb/s. The experimental setup is slightly modified in this case, as outlined in Fig. 1, in order to maintain the same duty cycle and hence achieve similar peak power levels at the Bi-NLF. The pulses at the transmitter are therefore compressed down to ~1.4 ps using a dispersion decreasing fiber (DDF). The SHGFROG characterization of these pulses is shown in Fig. 5(a), indicating a pedestal at a level of ~12 dB below the pulse peak, as well as a fairly linear chirp due to pulse evolution within the DDF. Due to the difference in the characteristics of the starting pulses relative to the 10 Gb/s case, we performed a further assessment of the performance of the nonlinear thresholder. For our set-up we have found that we can achieve optimum regeneration performance for an average power of 31 dBm, (see black solid curve in Fig. 5(c)), and an offset filtering of ~3.9 nm, (see dashed curve in Fig. 5(c)). The obtained spectrum is symmetric about the central wavelength, so similar performance is expected for both positive and negative filter offsets compared to the input signal wavelength. A comparison of the experimentally obtained spectrum with our simulations, (see red curve Fig. 5(c)), has shown that the ~2 dB peak at the central wavelength is caused by the finite extinction ratio of the data modulator, and the existence of ghost pulses in the zero-slots. To simulate this, we have also investigated numerically the SPM broadening of the data signal, when a modulator with a finite extinction is considered, (see Fig.5(c) blue curve), which is in good agreement with our experiment.

 

Fig. 4. Eye diagrams of the input and output of the system for no added noise (b and d) and some induced noise ((c) and (e)) (10ps/div), and corresponding BER measurements (a).

Download Full Size | PDF

 

Fig. 5. Experimental results for the 40Gb/s experiment: Intensity and chirp profile of the pulses at (a) the input and (b) the output of the regenerator measured using SHG-FROG. c) Experimental, and simulated SPM-induced spectral broadening when a modulator with finite, or infinite extinction is considered. Figure also shows the optical spectra at the input of the Bi-NLF, dotted curve, and after the 0.6nm filter, dashed curve. The spectra are vertically offset for ease of reading. Resolution: 0.5nm. d) Nonlinear transfer characteristic of the regenerator.

Download Full Size | PDF

In our experiment we have chosen to place the offset filter at wavelengths shorter than the original data wavelength in order to reduce the ASE noise levels introduced by the EDFA, which peak towards longer wavelengths. The measured transfer characteristic of the 2R regenerator is presented in Fig. 5(d). Note that the flat top of the curve is reached at a lower peak power than in the 10 Gb/s case, due to the smaller filter offset chosen in this case. We have characterized the pulses at the output of the regenerator using the SHG-FROG technique, (see Fig. 5(b)), and obtained a very similar behavior to the previous case shown in Fig. 3(c). This implies that the features of the output pulses were mainly determined by the optical filter used. Note that no pedestal is observed on the output pulses. In order to evaluate the noise-rejection of our system we have repeated similar measurements to those made for the 10 Gb/s case. The system is first assessed for the case when no amplitude noise is added onto the data, demonstrating that the technique does not in itself introduce any additional noise to the signal, (see Fig. 6(a) and (c)). We next introduce amplitude noise to the input signal with a ~11% and 4% standard deviation in the amplitude of both marks and spaces respectively, (see Fig. 6(b)), and assess the eye diagrams of the received signal at the output of the bandpass filter, (see Fig. 6(d)). The eye opening is clearly improved during the regenerative process. The signal was next demultiplexed into four 10 Gb/s constituents using an electro-absorption modulator (EAM), and BER measurements were performed on the four demultiplexed signals, (see Fig. 6(e)). For power levels <-16 dBm, the noise-degraded signals do not allow error-free operation of the receiver. However, error-and penalty-free operation is achieved after the 2R regenerator, with a receiver sensitivity improvement which exceeds 2 dB in all cases.

 

Fig. 6. Eye diagrams of the input and output of the system at 40Gb/s when no noise ((a) and (c)) or some amplitude noise ((b) and (d)) is induced (10ps/div). Corresponding BER measurements are also presented (e).

Download Full Size | PDF

6. Conclusions

The use of a 2-m-long highly nonlinear bismuth-oxide-based fiber in a simple 2R regeneration scheme based on spectral broadening due to SPM and offset filtering is reported at repetition rates of 10 Gb/s and 40 Gb/s. The regenerator benefits from the strong normal dispersion of the Bi-NLF in addition to its high nonlinearity. The performance of the regenerator was tested with heavily degraded input signals using SHG-FROG characterization, and both eye diagram and BER measurements. Even in such a short piece of fiber, and at power levels which are readily achieved using commercial amplifiers, an improvement in the receiver sensitivity of ~5 dB at 10 Gb/s and >2 dB at 40 Gb/s was achieved. We believe that this experiment highlights the potential of compound glass fiber technology for future compact and robust alloptical signal processing applications.