1. Introduction

Space-division multiplexing (SDM) using multi-core fiber (MCF) is one solution to meet the expected capacity demand for future data services being proposed for use in both long haul amplified transmission [1] and for short range links in access networks and data-centres [2]. Self-homodyne detection (SHD) [3, 4] has recently been proposed as a compatible technology for reducing costs in such space-division multiplexing (SDM) systems [5, 6]. In SHD, a transmitted pilot-tone (PT) originating from the transmitter laser, is space- or polarization-multiplexed with the data signal and used as local oscillator (LO) for coherent reception. Phase or near-phase coherency between the data signal and the PT yields phase noise cancellation (PNC) after coherent detection, which can be exploited to reduce the impact of laser phase noise and subsequently enable the use of spectrally efficient high-order quadrature-amplitude modulation (QAM) formats. Furthermore, as has also been demonstrated with shared carrier reception schemes in multi-core fiber (MCF) [7], where both signal cores and a PT are received with an intradyne detection (ID) receiver, PNC relaxes the requirement for high-speed carrier-phase tracking digital signal processing (DSP) in the receiver [8, 9]. SHD with a PT transmitted on an orthogonal polarization to the data has been previously demonstrated for a range of multi-level modulation formats [10–12], but suffers from a reduction of spectral efficiency (SE) of up to 50% compared to polarization division multiplexed (PDM) transmission. Although the SE may be improved by spectrally interleaving the PT with the signal [13], recent work on high channel-count SDM systems [1, 14–16] has opened the possibility of employing SHD with the PT transmitted through 1 SDM channel and the remainder used for data signals. Referred to as self-homodyne-MCF (SH-MCF) transmission, the SE reduction due to PT transmission compared to an equivalent ID scheme becomes inversely proportional to the number of SDM channels. An additional impairment in all SHD systems is the accumulation of noise along the PT transmission, which translates into an optical signal-to-noise ratio (OSNR) penalty with respect to intradyne detection (ID) systems. For typical system configurations, this penalty ranges from 1.5 to 4 dB and is largely determined by the bandwidth of the filter required to isolate the PT in a WDM configuration, the ratio of signal OSNR and PT OSNR, and the modulation format [18, 19].

Previously, the feasibility of SH-MCF was investigated experimentally with an external cavity tunable laser (ECTL) for linewidth sensitive 5 GBd quadrature-phase-shift-keyed (QPSK) signals [5, 6] and recently the prospect of employing SHD in multi-mode fiber systems was investigated numerically [20]. Here, we again focus on a 19-core MCF and expand on previous transmission results [21, 22]. On the same system, we investigate PNC performance in a single channel using both ECTL and distributed feedback (DFB) lasers for both QPSK and 16QAM modulation formats, showing for the first time that SH-MCF can outperform ID for linewidth limited, high-order QAM modulation. Finally, we describe high throughput SHD transmission using low-cost DFB lasers with MHz linewidth. This scheme is in contrast to previous high capacity demonstrations, which used DFBs as dummy channels with narrow linewidth transmitters used for measurement channel and receiver side LO. We transmit the PT on a single core of a 10.1 km MCF span with the remaining 18 cores used to transmit 25 GBd, 125 WDM QPSK and PDM-QPSK signals with 50 GHz channel spacing at 50 Gb/s and 100 Gb/s, respectively. Assuming a 7% overhead for forward-error correction (FEC), the total data throughput is 105.1 Tb/s for single polarization and 210 Tb/s for PDM signals with a SE of 16.7 b/s/Hz and 33.4 b/s/Hz, respectively.

2. Experiment description

The experimental set-up used for high-capacity transmission of QPSK and PDM-QPSK signals is shown in Fig. 1.For these measurements, the transmitter consisted of 125 rack-mounted DFB lasers with frequencies ranging from 189.1 to 195.3 THz on the 50 GHz spacedITU-T grid with an average linewidth of 3.8 MHz. After independent polarization control (PC) and variable optical attenuator (VOA) units, the signals were multiplexed in temperature-controlled arrayed-waveguide gratings (AWGs) with modulation performed with 3 different transmitter configurations. Initially, the system was characterized using either a single DFB laser with 2.9 MHz linewidth or an external cavity-tunable laser (ECTL) with 100 kHz linewidth. To enable exploration of both 16QAM and QPSK transmission, these measurements used a dual-polarization (DP) dual-parallel Mach-Zehnder modulator, referred to subsequently and in Fig. 1 as an I-Q modulator, driven by the outputs of two digital-to-analog converters (DACs) operating at 14 GBd, to map decorrelated 2¹⁵-1 bit pseudo-random binary sequences (PRBS) onto single polarization or PDM QPSK and 16QAM signals, as shown in Fig. 1(a). Due to limitations of the DAC performance, for the high capacity transmission demonstration using the 125 DFB laser array, transmitter set-up later was replaced with two independent single polarization I-Q modulators with automatic bias control. For QPSK modulation, shown in Fig. 1(b), an interleaver was used to separate even and odd channels for decorrelated modulation and each modulator was driven by two 101-bit-decorrelated 2¹⁵-1 bit PRBS signals for I and Q from pulse-pattern-generators (PPGs) operating at 25 GBd. Odd and even channels were combined in a power coupler with path length aligned by an free-space optical delay (OD). For PDM-QPSK, a power coupler was used to split the signal in two components, each modulated independently and polarization-multiplexed in a polarization-beam combiner (PBC) before a 50/100 GHz interleaver and OD were used to decorrelate the odd and even channels as shown in Fig. 1(c).

 

Fig. 1 Experimental set-up for SDM-WDM-PDM self-homodyne transmission in a 19-core fiber.

Download Full Size | PDF

For all transmitter configurations, a 3 dB coupler was used before signal modulation to split carriers for data modulation from those to be transmitted as PTs. The signal path was then amplified in an erbium-doped-fiber-amplifier (EDFA) before modulation and ODs were used to ensure all optical paths were matched in length. The modulated signals were then further amplified in a tellurite EDFA (T-EDFA) before a 1x20 splitter was used to split the signal in to 18 different cores of the MCF via the SDM MUX. Variable length patch cables were used to decorrelate the signals in different fiber cores. After the T-EDFA, a VOA was used to control the total fiber launch power to be approximately 0 dBm and the remaining splitter outputs were used for power and spectrum monitoring. The PT path contained only an EDFA and VOA to control the fiber launch power to 2 dBm and was connected to the remaining input port of the SDM MUX [15]. The 10.1 km MCF span was a trench-assisted homogeneous 19-core fiber with average loss of 0.23 dB/km, chromatic dispersion of 19.5 ps/nm/km and highest inter-core crosstalk value of −23 dB below the signal for center core transmission [16].

The remaining coupler input was used for an ECTL signal with a low frequency (160 MHz) phase modulation, used to precisely align the signal and PT path lengths. This tone propagated through both the signal and PT paths and since it is cancelled by PNC when path lengths were matched, its detection and digital processing allows generation of a low speed tracking signal and compensate dynamic changes in the path length alignment. In practice, the phase modulation could be applied to signal channels directly and has been observed to have negligible performance impact, when paths are aligned [23]. For this set-up it was observed that precise, automated path control had minimal impact on bit-error-rate (BER) performance, although it is envisaged that such path adjustment is needed to compensate for path variations of installed MCF in longer spans that are subject to greater environmental fluctuations.

After transmission, the SDM-DEMUX split the output of each core into separate fibers at the input to a 1x20 optical switch. The switch selected the core under test for BER measurements whilst the PT path contained a variable OD for course alignment of the signal and PT path lengths for each core. Next, signal and PT were filtered by optical band-pass filters (OBPFs) with variable bandwidth after each of two T-EDFA amplification stages. Between the amplification and filter stages, both paths contained an additional 3 dB coupler, which was used for noise loading by adding amplified spontaneous emission (ASE) noise generated by the filtered output of EDFAs with patch cables used to ensure decorrelated noise on each receiver input. Next, VOAs and optical taps were used to maintain optical power levels of 7 dBm for the signal and 10 dBm for the PT at the signal and LO inputs of an optical modulation analyzer (OMA). The OMA consisted of a polarization diverse coherent receiver followed by a 50 GS/s digital sampling oscilloscope with 20 GHz analogue bandwidth. For comparison purposes, an ID receiver was also implemented using an ECTL internal to the OMA, with 100 kHz linewidth, and used to evaluate the transmission penalty of SHD with respect to ID. The PT polarization was manually optimized before each acquisition to maximize the received signal amplitude although we note that real SHD systems will likely require polarization alignment schemes to align the transmitted PT to the axis of the receiver polarization-beam splitter [24]. The DSP, described more fully in [8,9], was performed offline using MATLAB and included skew and chromatic dispersion compensation, normalization, multiple input-multiple output (MIMO) polarization alignment, residual carrier phase recovery and error counting. BER measurements were based on the average of 3x40 µs traces.

3. Experimental results

3.1 Minimum OSNR penalty characterization

Before high capacity transmission, receiver noise loading was used for BER measurements as a function of the OSNR, measured with 0.1 nm resolution, for a single wavelength with both DFB and ECTL transmitter for QPSK, 16QAM and their PDM variants at 14 GBd, using the transmitter shown in Fig. 1(a). As demonstrated analytically in [19], the filter bandwidth of the PT is crucial in determining the OSNR penalty with respect to an equivalent ID system. For this reason, and to enable 16QAM transmission, BER vs. OSNR measurements were first made using the narrowest achievable PT filter bandwidth of 0.12 nm with the signal filter bandwidth of 0.5 nm. Figure 2(a) shows BER vs. OSNR measurements for QPSK and 16QAM transmission using both SHD and ID detection with DFB and ECTL transmitters. Similarly, Fig. 2(b) shows the same for PDM equivalent modulation formats. Also shown in Fig. 2 are the theoretical predictions based on the model presented in [19].

 

Fig. 2 Measured BER as a function of OSNR (0.1nm resolution) for DFB and ECTL transmitters for (a) QPSK, 16QAM and (b) PDM-QPSK, PDM-16QAM.

Download Full Size | PDF

The results in Fig. 2 reveal that the observed imperfections in the transmitter caused large implementation penalties for all formats. Compared to the predicted curves, the measured penalty for ID with an ECTL at a BER of 10⁻³ was 2 dB and 4 dB for QPSK and 16QAM modulation formats, respectively. Additionally, further signal distortion at the output of one of the DACs induced larger penalties of 2 dB and 6.5 dB for PDM-QPSK and PDM-16QAM respectively. However, despite the additional OSNR requirement caused by the DACs, the relative performance of SHD compared to ID is similar to the predicted values in all cases and lower than previously observed for similar baud-rate signals [13, 22] due to narrow filtering used and the PT OSNRs exceeding that of the signal. The measured penalties were 2.3 dB and 3.8 dB for QPSK and 16QAM, respectively, with smaller penalties of 1.7 dB and 2 dB for their PDM variants as a result of a reduction of the signal-PT ASE beat noise beat term as the signal power per polarization is reduced with the same single polarization PT.

Figure 2 also shows that for QPSK and PDM-QPSK transmission, neither ID nor SHD are affected by the additional phase noise of the broader linewidth DFB transmitter. In contrast, we observed errors floor at BERs of 10⁻³ and 2x10⁻³ for 16QAM and PDM-16QAM, respectively, when using ID and a DFB transmitter, which is attributed to the higher sensitivity of these modulation formats to phase noise. This was not the case for SHD with 16QAM, where the additional phase noise is cancelled at the receiver leading to similar performance with both laser transmitters with no error floor visible for BERs of approximately 1x10⁻⁴ for the ECTL and 8x10⁻⁴ for the DFB laser with the maximum range of OSNR limited to 28dB by the low output power of the modulator. These results confirm that higher order modulation formats make increasing demands for narrow laser linewidth and that SH-MCF transmission could be useful to enable super-high capacity transmission.

3.2 High Capacity Transmission Demonstration

A high capacity transmission demonstration was performed using SHD with 125 DFB laser transmitters by measuring the BER after transmission across the 19-core fiber. Due to the high inter-core of up to −23 dB below the signal the fiber [15,16] and the higher required OSNR observed due to the DAC performance, it was not possible to achieve sufficient OSNR at the receiver for 16QAM transmission. Hence, transmission measurements first used the transmitter set-up shown in Fig. 1(b) for QPSK transmission before being upgraded to PDM-QPSK, using the set-up shown in Fig. 1(c). For these measurements, it was necessary to use a broader PT filter bandwidth of 0.23 nm, similar to that of an AWG filter with 50GHz channel spacing, to allow some margin for laser wavelength drift and inaccuracy of manual filter tuning to reduce measurement time. For QPSK transmission, these filter bandwidths resulted in a penalty for SHD of 4.5 dB compared to ID in the single polarization case and 3.1 dB in the PDM case. Figure 3 shows the BER measured for each channel and core combination and average BER measurements for odd and even channels separately.

 

Fig. 3 Measured BERs for 125 wavelength channels in each of 18 signal cores and (inset) MCF core layout.

Download Full Size | PDF

The measured BERs varied from 2x10⁻⁴ to 2.7x10⁻⁶ with the average BER measured to be 3.6x10⁻⁵ and 4.5x10⁻⁵ for odd and even channels respectively. A systematic difference in performance between odd and even channels is observable across all channels in Fig. 3 and believed to originate from the unmatched characteristics of the 2 modulators. Some wavelength dependence is evident and attributed to the combination of EDFA and T-EDFA gain spectrums. For example, the combination of the 1530 nm gain peak in the C-band EDFAs and sharp roll off in the T-EDFAs led to large received power fluctuations in the shorter wavelength channels. All the measured BER is well below the 2x10⁻³ limit of commercially available FEC systems with 7% overhead. Hence, including the associated 7% overhead, the capacity of the whole system reduces from 112.5 Tb/s to 105.1 Tb/s.

Finally, the I-Q modulators were used to generate PDM-QPSK signals using the configuration in Fig. 1(c) and Fig. 4 shows the mean and individually measured BERs on all cores for 5 evenly spaced channels, including the center and edge wavelengths, and shows that BERs below 1x10⁻⁵ were achieved for all measurements. As observed in crosstalk measurements [15,16], the inner and center cores (1-7) suffer from higher inter-core crosstalk due to the increased number of neighboring cores and this is reflected in slightly higher average of the measured BERs. Furthermore, the cores adjacent to the PT have the highest measured BERs of all the outer cores, suggesting that the increased launch power of the PT caused some additional interference, the impact and power dependence of which could be an avenue of future study. Assuming that similar BER performance is achieved for all wavelength channels, this system represents doubling the capacity and spectral efficiency of the transmission system to 210 Tb/s and 33.4 b/s/Hz respectively, using low cost DFB laser transmitters, for the same 5.3% reduction in SE compared to an equivalent intradyne system using all cores for signal channels.

 

Fig. 4 Measured BER in each signal core for 5 wavelength channels for PDM-QPSK transmission in an MCF.

Download Full Size | PDF

4. Summary

We have demonstrated high capacity transmission using coherent self-homodyne detection (SHD) in a space-division-multiplexed link using a 19-core fiber. Characterization results showed that with reasonable optical filtering of the PT, the cost of linewidth tolerance and simplified receiver structure is an OSNR penalty ranging from of 1.7 dB for PDM-QPSK to 3.2 dB for single polarization 16QAM modulations. Furthermore, the characterization of 16QAM signals showed that phase noise cancellation in SHD enabled transmission with MHz linewidth lasers that resulted in error floors when using intradyne detection although it was not possible to achieve the required OSNR for high capacity 16QAM transmission in this fiber due to WDM and DSM crosstalk. Finally, we performed a high-capacity demonstration of and SDM-WDM-PMD transmission, where we transmitted the pilot-tone on a single core of a 19-core, 10.1 km span with the remaining 18 cores used to transmit QPSK and PDM-QPSK signals on 125 wavelengths with 50 GHz channel spacing at 25 GBd. Assuming a 7% overhead for forward-error correction, the total data throughput was 105.1 Tb/s for single polarization and 210 Tb/s for PDM signals with a SE of 16.7 b/s/Hz and 33.4 b/s/Hz, respectively. High-capacity transmission was achieved despite high inter-core crosstalk and laser linewidth and narrow channel spacing, showing that combining SHCD with MCF could enable high throughput, low-cost transmission in next-generation optical networks.