1. Introduction

The combined requirements for increase in spectral efficiency and higher channel granularity, have created a considerable need for generation, copying and processing of high-rate data streams. In particular, recent efforts have focused on fidelity, transmission speeds and transmission distances of optical time division multiplexed (OTDM) data streams [1–2]. At the same time, transparent multicasting has been recognized as vital functionality for network configurations with significant asymmetry in the traffic direction flow. High quality multicasting is of particular interest in defense application, as well as commercial applications such as finances, high-bandwidth broadcasting or sensor networks [3].

A significant problem associated with conventional OTDM techniques is the impairment originating from coherent crosstalk created by the interference between adjacent pulse tails. Indeed, the coherent crosstalk can rapidly degrade the performance of the multiplexed signal [4]. Several approaches have been proposed to overcome this impairment, such as the use of pulses [5] with duty cycle below one third of the target bit slot, polarization multiplexing [2] or increase in extinction ratio (ER) between the pulse peak and the pulse tail [6]. The last technique offers the ability, at least in principle, to generate high-rate data channel without lowering the duty cycle of the generating pulse train [6]. Subsequent multicasting of OTDM signal poses an additional challenge that can be understood in terms of bandwidth and distortion requirements. When considering multicasting of ultrahigh-speed data, large bandwidths associated with excessively short pulses or pulse-to-pulse polarization diversity can add complexity and reduce the flexibility and robustness of the system. The preferred approach is therefore to avoid coherent crosstalk by using pulses with increased ER and only then attempt to perform all-optical multicasting to generate multiple, high-fidelity copies of the original signal.

Fiber optical parametric amplifiers (FOPA), besides their ability to project spectrally wide gain in any band of interest [7–8] also offer ultrafast response and high-efficiency conversion necessary for combined multicasting and regeneration functions. Multicasting in a single segment of highly non linear fiber (HNLF) can be done in one-pump configuration where the data is imposed on the pump and a number of CW signals are used as seeds. In this configuration, data is copied to the entire complement of idlers and signals [9]. Seemingly simple, this configuration could be, at least in principle, scaled to an arbitrary number of copies. In practice, however, this strategy is limited by the bandwidth and the gain of the one-pump FOPA. Indeed, the response of one-pump device with modulated pump is governed by the modulation instability and inherent distortions from the pump-signal nonlinear interaction responsible for the pulse steepening [10]. A two-pump FOPA architecture using continuous wave (CW) pumps can circumvent these limitations, particularly when operating in signal-pump linear regime [10]. In addition, a two-pump device directly generates three new data copies from a single data seed and provides diversity of conjugated and non-conjugated waves.

Following this strategy, in this paper we describe the generation of high quality 160 Gb/s data stream and its subsequent multicasting. In first stage of the experiment, one-pump FOPA idler wave was used to increase channel ER in excess of 20 dB and use it in OTDM not limited by coherent crosstalk. The 160 Gb/s data stream was subsequently multicast to 4 copies in a two-pump FOPA for the first time. Finally, we report impairment-free multicasting that resulted in 160 Gb/s data copies with a Q-factor in excess of 19 dB.

2. Experimental setup

The experimental setup is shown in Fig. 1 and consisted of four functional blocks. The first block was used for pulse regeneration, or, equivalently, increase of original pulse train ER. The seed pulse train was generated by a 40 GHz modelocked laser (MLL), and created a 40 GHz idler wave with significant extinction ratio improvement. The high ER pulse train was modulated at 40 Gb/s by the second block and subsequently sent to the third block used as the bit rate multiplier (BRM) not limited by the coherent crosstalk. The final processing block was represented by a two-pump multicasting device.

The MLL seed regeneration was performed in one-pump FOPA seeded by a continuous wave laser source. The pump was amplified modelocked laser at 1562.8 nm producing 1.9 ps full-width half maximum (FWHM) long pulses. After being amplified (EDFA₂), the pump was coupled into a 100 m long segment of highly non linear fiber (HNLF₁) through a WDM coupler. A continuous wave (CW) signal seed from an external cavity laser (ECL) at 1593.3 nm, was amplified (EDFA₁) and filtered by a 0.6 nm wide band pass filter (BPF₁) to eliminate excess amplifier spontaneous emission (ASE). A variable optical attenuator (VOA) was used to vary the power of the seed before it was coupled into the HNLF₁. The HNLF₁ was characterized by a 1561.8 nm zero dispersion wavelength (ZDW), a slope of 0.026 ps/nm²-km a β₄ of 2.4×10^-5 ps⁴/km, and a γ of 16 W^-1km^-1. Parametric interaction between the pulsed pump and the CW signal generated a pulsed idler wave centered at 1532.5 nm.

At the output of HNLF₁ segment, the pump and the seed waves were removed by a spectral block (WDM) and the idler wave was sent to the data modulation block. A variable optical attenuator (VOA) was used to control the input power to the 40 Gb/s amplitude modulator (AM) and the time delay between the pulse and the center of the bit slot was adjusted by the variable manual delay line (τ). The modulator was driven by a 2⁷-1 pseudo random bit sequence (PRBS) which was transferred onto the idler by pump-signal parametric process, generating high ER 40 Gb/s return-to-zero (RZ) data stream. The channel was subsequently amplified (EDFA₃) and sent to the bit rate multiplier block. A polarization maintaining (PM) BRM was used to generate 2⁷-1 PRBS at 160 Gb/s from the 2⁷-1 PRBS 40 Gb/s input, by cascading two rate-doubling stages. Each stage was constructed from a delay interferometer with differential delays of 1.5875 ns and 0.7938 ns in the first and the second stage, respectively.

 

Fig. 1. Experimental setup. MLL: modelocked laser. ECL: external cavity laser. VOA: variable optical attenuator. τ: optical delay line. AM: amplitude modulator. PM: phase modulator.

Download Full Size | PDF

The 160 Gb/s data channel was finally sent to a 2-pump parametric amplifier for multicasting. Lasers positioned at 1543 nm and 1581.8 nm were used as C-Band (Pump 1) and L-Band (Pump 2) pumps, respectively. The pumps were phase modulated with a single tone at 100 MHz to increase the stimulated Brillouin threshold before being amplified (EFDA₄ and EDFA₅), filtered by two 0.6 nm wide bandpass filters (BPF₂ and BPF₃) and combined with a C-L band WDM coupler. The two parametric pumps and the 160 Gb/s signal at 1532.5 nm were then coupled by a power (50:50) coupler into a 50 m segment of HNLF (HNLF₂) having similar characteristics as HNLF₁. The parametric interaction between the two pumps and the seed positioned within the 1- [9] band of the 2-pump FOPA generated three additional waves (idlers) in 1+ (1554nm), 2- (1571.3nm) and 2+ (1592.3nm) bands, all carrying the original 160 Gb/s data. A total of four copies were therefore achieved, with two non-conjugated 160 Gb/s channels positioned within 1- and 2- bands, and two conjugated channels generated within 1+ and 2+ bands. A WDM coupler was used at the output of the HNLF to filter the chosen wavelength, which was then amplified and monitored on an ultrafast optical sampling oscilloscope and an optical spectrum analyzer (OSA).

3. Experimental results

The output of the modelocked laser and its corresponding spectrum are shown in Fig. 2(a) and 2(b), respectively. The original pulse was characterized by a width of 1.9 ps, a signal to noise ratio (SNR) of 28 dB, and a 160 GHz pulse ER of 21.1 dB. We define the pulse ER as the ratio between the energy of the pulse within its designated bit slot (i.e. over 6.25 ps for 160 GHz rate) and the unwanted energy in the adjacent bit. The residual energy in the tail of the pulse poses a significant problem as it gives rise to coherent crosstalk between adjacent pulses after OTDM process. It has been shown that for a duty cycle of 40% and pulse tail ER of 27 dB will result in a 1 dB power penalty and that the duty cycle and ER can be traded off [3] to mitigate this penalty. Consequently, the requirements for pulse width and pulse ER for time division multiplexing at 160 Gb/s are demanding: one is not free to arbitrarily choose one of these parameters without compound effect on channel integrity and its transportability. The MLL-driven FOPA in this experiment was used to remove the coherent crosstalk impairment without reducing the duty cycle significantly, and thus rendering the overall robustness of the generated data stream. Intuitively, the ER improvement (regeneration) of the idler can be understood in terms of significantly lower conversion efficiency associated with the (pump) pulse tail when compared to the peak of the pump pulse [11].

 

Fig. 2. Optical sampling oscilloscope waveforms and optical spectra. (a) Output of the modelocked laser and (b) corresponding spectrum. (c) Idler generated in the 1-pump FOPA and (d) corresponding spectrum.

Download Full Size | PDF

The mixing of 175 mW average power pump and 40 mW CW seed in HNLF₁ generated a 125 mW peak power idler at 1532.5 nm. The spectrum at the output of HNLF₁ is shown in Fig. 3 indicating the absence of higher-order light generation [9]. The waveform and spectrum of the filtered idler are shown in Fig. 2(c) and 2(d), respectively. The pulse was slightly compressed to a width of 1.4 ps, and clearly exhibits the form associated with the regeneration function. While conversion efficiency of 5 dB was observed in the FOPA, it was not deemed critical for the experiment, as considerably higher pump power was available. Through the regeneration process, the pulse ER of the idler was increased to the minimum of 42 dB. The parametric stage provided ER improvement in excess of 20 dB, sufficient for high quality OTDM to 160 GHz. The SNR decreased slightly to 24.5 dB as noise was added during the parametric conversion and amplification process.

 

Fig. 3. Optical spectrum at the output of HNLF₁.

Download Full Size | PDF

The RZ 40 Gb/s signal generated by high-speed modulator was amplified by EDFA3 and is shown in Fig. 4(a). No filtering was performed after EDFA₃ due to the lack of suitable bandpass filters at these wavelengths, and the amplified spontaneous emission (ASE) noise resulted in a decrease of SNR to 22.1 dB. Consequently, we expect that OSNR penalty taken within the modulation stage can be reduced by insertion of the bandwidth-tailored optical filter. The pulse width of the modulated pulse was increased to 1.8 ps due to the residual dispersion in the EDFA. The spectrum of the 40 Gb/s signal is shown in Fig. 4(b). This signal was used to seed the BRM block whose delays were adjusted in order to obtain a 2⁷-1 PRBS at 160 Gb/s from the original 40 Gb/s input. The modulated 160 Gb/s data stream resulting from two-stage OTDM is shown in Fig. 4(c), indicating exceptionally clean eye characterized by a wide opening and low jitter signatures.

 

Fig. 4. Optical sampling oscilloscope waveforms and optical spectra. (a) Output of amplitude modulator driven by a 2⁷-1 PRBS at 40 GHz and (b) corresponding spectrum. (c) 160 Gb/s waveform at the output of the BRM and (d) corresponding spectrum.

Download Full Size | PDF

The pulse width was further increased from the original MLL width to 2.3 ps due to the residual dispersion experienced between the FOPA output and the sampling scope input. The final SNR was maintained at 22.1 dB and the final Q-factor was measured to be 20.1 dB. The spectrum (Fig. 4(d)) shows the rejection of 3 frequency modes (out of the original 4) separated by 160 GHz.

Finally, the last block performed multicasting of the original 160 Gb/s data to 3 additional wavelengths to demonstrate 1-to-4 multicasting process in a single fiber. All-optical replication was obtained by a 2-pump parametric amplifier where the 160 Gb/s data was used as a signal seed within the 1- band, i.e. wavelength band below pump 1 at 1543 nm. The average powers inside the HNLF were measured to be 0.5 mW for the data seed and 600 mW for each pump, thus generating true linear operation of the two-pump parametric device. Idler 1, positioned in the 1+ band, had the worst conversion efficiency of -10 dB. The other 2 idlers (Idler 2 and Idler 3 in the 2- and 2+ band, respectively) experienced conversion efficiencies of -7 dB and -6 dB, respectively. The sensitivity of the preamplified sampling oscilloscope was sufficient to perform the measurement of the lowest efficiency idler (1+), thus obviating the need for higher pump power or using longer HNLF section to increase overall efficiency. The optical spectrum at the output of the multicasting stage is shown in Fig. 5.

 

Fig. 5. Optical spectrum at the output of the multicast stage (after HNLF₂).

Download Full Size | PDF

At the output of the multicast stage, the worst performing copy (Idler 1) at 1553 nm was filtered by a WDM coupler and sent to the ultrafast optical sampling scope. The waveform (Fig. 6(a)) shows the 160 Gb/s copy of the original data (Fig. 4(c)). The pulses experienced finite compression within the FOPA and the pulse width was measured to be 1.85 ps, reducing the final data duty cycle from 0.5 to 0.4. In order to observe the waveform on the sampling scope, the idler was amplified (EDFA₆) to 3 dBm average power. Similar to the modulation stage measurements, the bandpass filters at these wavelengths were not available during the experiment, thus forcing us to accept the combination of the data channel, EDFA ASE and noise from the FOPA into the instrument (Fig. 6(b)); a simple path for the improvement would be the appropriate filtering of the observed pulses. The SNR of the copied idler was measured at 18 dB, indicating the excessive penalty taken in absence of final idler filtering. A measured eye diagram indicated high-fidelity multicasting, even for the lowest efficiency idler wave. The performance of the signal copy was characterized by Q- factor measured at 19.1 dB. We note that out of four generated copies, Idler 1 had the lowest performance due to the 4 dB difference in conversion efficiency with respect to 1571.3 and 1593.3 nm idlers. The observed minimum Q-factor was more than 10 dB above the expected FEC limit of 8.3 dB.

 

Fig. 6. (a) Optical waveform of the multicast 160 Gb/s Idler1 at 1553 nm and (b) corresponding spectrum.

Download Full Size | PDF

4. Conclusions

We have reported the experimental demonstration of a 160 Gb/s OTDM data stream generation and subsequent all-optical multicasting. The report represents the first use of two-pump parametric device for high-rate data multicasting in linear parametric regime. The performance of the OTDM generation was increased by the regeneration of the seed 40 GHz MLL pulse train. To achieve this, MLL was used as a pump in a one-pump FOPA leading to ER increase in excess of 20dB with respect to the seed source. The technique, used for the first time in this experiment, allowed the creation of temporally compressed pulse train with an ER of 42 dB. High ER pulse train was used to subsequently generate 160 Gb/s PRBS bit stream with a Q of 20.1 dB. A two-pump FOPA was used to copy the 160 Gb/s data channel into four spectrally distinct copies. Measurements of the worst performing data copy indicated a minimal SNR of 18 dB and Q-factor of 19.1 dB, guaranteeing the performance well above the conventional FEC limit.