1. Introduction

To satisfy the ever-increasing capacity demand in optical fiber communications, both the spectral efficiency (SE) and the data rate carried by a wavelength channel have been increasing dramatically [1,2]. Channel data rates of 1 Tb/s and beyond have been demonstrated using the superchannel concept [3–6], achieving intrachannel SEs (ISE, defined as the net channel bit rate divided by the channel’s spectral width) between ~4 and ~7 b/s/Hz. Using high-level quadrature-amplitude modulation (QAM), ISEs beyond 7 b/s/Hz have been demonstrated at sub-Tb/s data rates [7–9]. Space-division multiplexing (SDM) is being considered as a promising candidate technology to dramatically increase per-fiber capacity [1,2,10]. Per-fiber capacities of over 100 Tb/s have been recently demonstrated by using SDM with multi-core fibers (MCFs) [11,12], surpassing with ease the highest capacity reported over single-mode fiber [13]. In this paper, we present in more depth our recent demonstration [14] of the generation and detection of a 1.12-Tb/s superchannel based on coherent optical orthogonal frequency-division multiplexing (CO-OFDM) with polarization-division-multiplexed (PDM) 32-QAM subcarriers, achieving an ISE of 8.6 b/s/Hz. We further leverage SDM to demonstrate a record aggregate ISE of 60 b/s/Hz per fiber over a 76.8-km MCF [14]. Finally, we discuss the impact of core-to-core crosstalk on the transmission performance of the high-ISE superchannel.

2. Superchannel generation at 1.12 Tb/s with an ISE of 8.6 b/s/Hz

Figure 1 shows the schematic of the experimental setup for the generation of the superchannel. An external cavity laser (ECL) at 1548.3 nm with a linewidth of ~100 kHz was used as the laser source. A 5-comb generator, based on a Mach-Zehnder modulator (MZM) driven by a 25.94-GHz sine-wave with ~3V_π amplitude, generated five frequency-locked carriers with a spacing of 25.94 GHz. A wavelength-selective switch (WSS) was configured to have a 3-dB bandwidth of 120 GHz to reject the unwanted harmonics generated by the 5-comb generator. A novel 4-comb generator, based on a nested MZM whose two branches were respectively driven by 3.24-GHz and 9.73-GHz sine-waves with ~1V_π amplitudes, quadrupled the number of frequency-locked carriers to 20 with a carrier spacing of 6.48 GHz. The phase between the two branches was set to π to suppress the unwanted DC carrier. The optical spectra of the generated carriers at different stages are shown as insets in Fig. 1. Remarkably, all the unwanted harmonics were rejected to be over ~35 dB down. The 20 carriers were then modulated by a PDM I/Q modulator to generate a PDM-32QAM-OFDM superchannel. Note that the 20 carriers had different optical phases, as previously shown for the MZM-based comb generator [15], so the 20 subchannels in the superchannel were effectively phase de-correlated, in addition to the intrinsic de-correlation of the subcarriers in each OFDM subchannel. It would be more preferred to also de-correlate the intensity profiles of these subchannels, but due to limited hardware resource, this experiment was conducted with the intensity profiles being correlated. The measured signal’s nonlinear tolerance is expected to be slightly worse than what it would be with both phase and amplitude de-correlation. The x- and y-polarization components of the PDM signal were independently modulated to better emulate a real transmitter. Four independent drive patterns were stored in two synchronized arbitrary waveform generators (AWGs), each having two 10-GS/s digital-to-analog converters (DACs). Pseudo-random bit sequences (PRBS) of length 2¹⁵-1 were used as the payload data. The IFFT size used for OFDM was 128, and the guard-interval (GI) was 2 samples, resulting in a small GI-overhead of 1.56%. Each polarization component of an OFDM symbol contained 78 32-QAM data subcarriers (SCs), 4 pilot SCs, one unfilled DC SC, and 45 unfilled edge SCs. The spectral bandwidth of each modulated subchannel was 6.48 GHz (=83/128×10GHz), and the 20 frequency-locked 6.48-GHz-spaced input carriers enabled seamless superchannel formation with a total bandwidth of 130 GHz, as shown in inset (f). Three correlated dual-polarization training symbols (TSs) [7] were used for every 697 payload OFDM symbols, resulting in a small TS-overhead of 0.43%. Excluding 7% overhead for forward-error correction [16], the net payload data rate of the superchannel was 1.12 Tb/s (=10GHz × 10b/s/Hz × 78/130 × 697/700 × 20/1.07), corresponding to a net ISE of 8.61 b/s/Hz (=1.12Tb/s/129.7GHz). With SDM in a seven-core fiber, the aggregate per-fiber ISE became 60 b/s/Hz. This superchannel could likely be put on a 150-GHz grid with <-40 dB crosstalk to neighbors, as indicated in inset (f), to achieve an aggregate SE of 52 b/s/Hz in a wavelength-division multiplexed (WDM) system.

 

Fig. 1 Schematic of the experimental setup of the 1.12-Tb/s superchannel transmitter. Insets (a-f) show the schematic outputs (in the frequency domain) and the measured optical spectra at different stages in the transmitter. EDFA: Erbium-doped fiber amplifier; PBC: polarization-beam combiner.

Download Full Size | PDF

3. SDM in a 7-core fiber for an aggregate ISE per fiber of 60 b/s/Hz

For SDM-based transmission, and as shown in Fig. 2 , the superchannel was split into 8 copies by a 1×8 splitter, whose seven outputs were delay de-correlated and amplified by seven erbium-doped fiber amplifiers (EDFAs) before launching into a 76.8-km seven-core-fiber [12] through a tapered multi-core connector (TMC). After transmission, a second TMC was used to couple out the signals, which were then amplified to compensate for the fiber loss. An optical switch (SW) was used to direct the received signal from each of the seven cores to a digital coherent receiver with offline digital signal processing (DSP). An optical local oscillator (OLO) was another ECL whose frequency was tuned to the center of each of the 20 subchannels. Electronic low-pass filters (LPFs) with 6-GHz bandwidth were used before the analog-to-digital converters (ADCs) to select a subchannel for measurement. Digitized waveforms of 1-million samples each were processed offline in a computer to perform electronic dispersion compensation, nonlinear compensation (NLC), polarization de-multiplexing, frequency/phase recovery, and bit error ratio (BER) measurement using previously reported PDM-OFDM algorithms [7].

 

Fig. 2 Schematic of the experimental setup for superchannel transmission and detection. Insets: (a) a typical recovered signal constellation of the 32-QAM subcarriers; (b) illustration of launching into the seven cores of the MCF with multiple multi-level signals.

Download Full Size | PDF

4. Experimental results

Figure 3 shows the measured back-to-back BER performance of the 1.12-Tb/s PDM-32QAM-OFDM superchannel (averaged over all 20 subchannels). The required optical signal-to-noise ratio (OSNR), defined with a 0.1-nm noise bandwidth, is 30.5 dB at BER=4.6×10⁻³, the threshold of a 7%-overhead hard-decision FEC [16]. Compared to the theoretical performance, the implementation penalties were 2.3 dB and 3.7 dB at BER=4.6×10⁻³ and 2×10⁻³, respectively, much improved over previous 32-QAM results [7,8]. Figure 4 shows the received signal Q² factor of the center subchannel (averaged over the center core and an outer core) versus signal launch power per core (P_in). Equal power was launched into each core and the optimum power is found to be 2 dBm. With the use of a single-step NLC [17], the optimum Q² factor was increased by 0.7 dB. Figure 5 shows the measured BER performance of the superchannel after passing through the seven cores of the 76.8-km MCF simultaneously, with 2-dBm launch power into each core. The mean and the worst BER values are, respectively, 3.8×10⁻³ and 4.3×10⁻³, which are both lower than the assumed FEC threshold. The received OSNR values are also shown in Fig. 5. The mean OSNR is ~34.5 dB, indicating a moderate transmission penalty of ~1.1 dB in Q² factor.

 

Fig. 3 Measured back-to-back BER performance of the 1.12-Tb/s PDM-32QAM-OFDM superchannel and theoretical limit. Inset: typical recovered subcarrier constellation at OSNR = 40 dB.

Download Full Size | PDF

 

Fig. 4 Q²-factor (derived from BER) of the center subchannel as a function of signal launch power. Inset: recovered subcarrier constellation after transmission.

Download Full Size | PDF

 

Fig. 5 Measured BER performance of the superchannel after passing through the seven cores of the 76.8-km MCF simultaneously, with 2-dBm launch power into each core. Core 1 here is the center core.

Download Full Size | PDF

Since spatial crosstalk will be a major design factor for future SDM systems, it is of interest to investigate the impact of core-to-core crosstalk on the transmission performance of this 32-QAM superchannel. Figure 6 shows the BER performance of the superchannel after passing through the center core (core index 1) of the 76.8-km MCF with 2-dBm launch power but with different signal power loadings on the 6 outer cores. The mean BER for the case with no signals in the outer cores is essentially the same (within experimental error) as that with equal signal power in all cores, indicating negligible core-to-core crosstalk penalty. Considering the low total crosstalk from the 6 outer cores to the center core of about −35 dB at the signal wavelength [12], we can calculate the expected crosstalk penalty using the results from a recent study on the impact of coherent crosstalk on n-QAM signals [18]. The crosstalk penalty on the center-core signal when all the six outer cores are populated with signals each having the same power as the center-core signal is expected to be ~0.08 dB. This confirms that the low-crosstalk seven-core fiber can indeed support single-span transmission of high-level modulation formats such as 32-QAM with negligible penalty.

 

Fig. 6 Measured BER performance of the superchannel after passing through the center core of the 76.8-km MCF with P_in^center = 2 dBm but different signal power loadings on the 6 outer cores.

Download Full Size | PDF

Regarding longer transmission distances, it has been shown recently that the low-crosstalk MCF can be used for long-haul multi-span transmission with negligible crosstalk penalty for PDM-QPSK signals [19]. To further investigate the impact of crosstalk accumulated over long transmission distances, we next study the crosstalk penalty for high-level formats such as 32-QAM in long-haul transmission. To this end, we emulate higher levels of crosstalk as follows: We start with a measured 32-QAM waveform (raw data, measured after transmission at 2-dBm launch power with 50-GS/s sampling) as our signal in the center core, and add six randomly delayed and polarization-scrambled copies of this same signal to emulate interferers from the outer cores. Figure 7 shows the resulting crosstalk penalty on the center-core 32-QAM-OFDM superchannel when the crosstalk level is increased 10 times compared to a single fiber span. Assuming linear crosstalk accumulation with transmission length, this should correspond to a 10-span transmission system. Note that without sufficient de-correlation between signal and crosstalk, the crosstalk penalty can be “artificially” small due to the high tolerance of OFDM to multipath interference through the use of GI. In the experiment, the GI was 200 ps, corresponding to ten 50-GS/s samples. When the random delays were limited to 10 samples, the crosstalk-induced penalty is only ~0.15 dB, but when the random delays were set to be >200 samples (or >4 ns) for sufficient de-correlation, the crosstalk-induced Q²-factor penalty is ~0.8 dB. This indicates that it is desirable to further reduce the core-to-core crosstalk of the MCF in order to support long-haul transmission of signals with high-level modulation formats.

 

Fig. 7 BER performance of the superchannel after passing through the center core of the 76.8-km MCF with an emulated 10-fold increase in the crosstalk level.

Download Full Size | PDF

5. Discussion

To reduce the cost per bit in MCF-based SDM transmission systems compared to conventional systems using parallel strands of single-core fibers, there are still several aspects that need to be addressed. As illustrated in Fig. 8 , the desired enabling components include MCFs with further reduced loss, crosstalk, and nonlinear coefficient, multi-core Erbium-doped fiber amplifiers with direct coupling to the MCF, reconfigurable optical add/drop multiplexers (ROADMs) able to work with MCF in an integrated fashion, as well as photonic integrated circuits (PICs) that take advantage of the high mode density of MCF to reduce transponder size and cost.

 

Fig. 8 Illustration of key enabling components desired for future MCF-based SDM transmission.MC-EDFA: multi-core Erbium-doped fiber amplifier; TX/RX: transmitter/receiver.

Download Full Size | PDF

6. Summary

We have experimentally demonstrated the generation and detection of a 1.12-Tb/s PDM-32-QAM-OFDM superchannel with a small implementation penalty and with a record ISE of 8.6 b/s/Hz for Tb/s-class superchannels. Key enablers include the generation of 20 high-quality frequency-locked carriers, simultaneous modulation of both polarizations of the PDM signal, and a low overhead used for OFDM signal processing. We have further demonstrated a record aggregate per-fiber ISE of 60 b/s/Hz by transmitting the 1.12-Tb/s superchannel over a 76.8-km seven-core fiber, with negligible core-to-core crosstalk penalty. We have discussed implications of core-to-core crosstalk in long-haul transmission systems through crosstalk emulation. Our demonstration shows the potential of the combination of high-SE signal formats and MCF-based SDM, together with the development of other enabling multi-core-specific components, to dramatically increase the achievable spectral efficiency as well as the capacity of a single fiber for sustaining the capacity growth of future optical transport systems.