1. Introduction

In recent years, coherently detected PDM-QPSK emerged as a favored solution for 100G Ethernet transport over transoceanic distances [1-4]. For WDM transmission on a 50 GHz channel grid, a reach of 9000 km has been demonstrated using pure-silica core fiber and amplification by erbium-doped fiber amplifiers (EDFA) only [3]. By using narrow optical pre-filtering of the 112 Gb/s PDM-QPSK signals and a maximum a posteriori (MAP) detection algorithm, transmission over 10610 km was achieved on a 33-GHz channel grid [4]. Both [3] and [4] used a span length of ≈50 km and EDFA-only amplification. In [2], a 33 GHz channel grid was achieved using optical Nyquist filtering. While [3] and [4] assumed 7% overhead for a hard-decision forward error correction (FEC), 20% was assumed in [2], resulting in a BER of 10⁻² at the FEC limit. This enabled a transmission reach of 10000 km.

Two years ago, Karlsson [5] and Agrell [6] as well as Bülow [7] explored the optimal placement of constellation points in the four-dimensional (4-D) signal space spanned by the electric field components in optical single-mode fibers. A numerical optimization using sphere packing in 4-D space identified PS-QPSK as the most power efficient 4-D modulation format [5]. Compared to PDM-QPSK at a BER of 10⁻³, it offers approximately 1 dB sensitivity improvement for the same bit rate [6].

The transmission performance of PS-QPSK signals was investigated by numerical simulations for symbol rates of 28 GBd and 37 GBd over 1800 km standard single-mode fiber (SSMF) as well as non-zero dispersion-shifted fiber links in [8] and for a symbol rate of 37 GBd over a 2000 km SSMF link in [9]. These results indicate that PS-QPSK signals are not only more power efficient but also show better resilience against nonlinear impairments in WDM systems, making PS-QPSK an interesting option for ultra-long-haul transmission.

The WDM transmission performance of 40 Gb/s PS-QPSK signals operating on a 50 GHz grid was also experimentally investigated in [10] and [11]. Both contributions use a PS-QPSK transmitter consisting of two serial stages. First, a QPSK signal is generated by using an I/Q-modulator. A second stage consisting of a 3-dB coupler, two synchronized Mach-Zehnder modulators and a polarization-beam combiner switches the polarization state. Ultra long-haul transmission over an uncompensated SSMF link with EDFA-only amplification was reported in [10]. Changing the modulation format from PDM-QPSK to PS-QPSK increased the maximum reach by about 30% to 13640 km. In [11], the authors experimentally compare WDM transmission of PS-QPSK signals and PDM-QPSK signals over a 1000-km SSMF link. In both experiments, the PS-QPSK signals showed a better resilience against nonlinear impairments.

In [12], we report on the generation of 28 GBd (84 Gb/s) PS-QPSK signals using an integrated dual-polarization (DP) I/Q-modulator and assessed the single-channel transmission performance over transoceanic distances up to 12500 km. In this contribution, we increase the symbol rate to 37.3 GBd to generate 112 Gb/s PS-QPSK signals. The performance of these signals is evaluated in back-to-back and WDM transmission using a fiber loop with ultra-large effective area fiber (ULAF) spans with up to 15000 km transmission distance. Furthermore, the performance is compared to that of PDM-QPSK signals at the same bit rate of 112 Gb/s.

2. Experimental setup

The schematic of our experimental setup is shown in Fig. 1 . We used eight tunable lasers (line width ~100 kHz) as light sources to generate 50 GHz spaced WDM channels using twomodulator rails for the odd and even channels. The upper rail consists of a single integrated DP-I/Q-modulator (Fujitsu FTM7977HQA), which can create both PDM-QPSK and PS-QPSK signals by simply changing its driving signal sequences d₁…d₄. To create 112 Gb/s PDM-QPSK, four 28 Gb/s pseudo-random binary sequences (PRBS) of length 2¹⁵-1 are generated from a four-channel bit pattern generator (BPG), with d₂, d₃ and d₄ mutually delayed relative to d₁ by 16383, 8191 and 24574 bits, respectively. For the generation of 112 Gb/s PS-QPSK, d₄ needs to be coded as XOR combination of d₁, d₂ and d₃ [5]. Furthermore, the bit rate of the driving signals needs to be increased to 37.3 Gb/s. Since this bit rate is above the maximum specified bit rate of 32 Gb/s of the employed BPG (SHF 12103A), it cannot generate freely programmable data sequences to accomplish the XOR operation. However, it can still generate four independent 37.3 Gb/s PRBS (after custom calibration by the manufacturer). We utilized a special property of PRBS to mimic the XOR coding required for PS-QPSK. It was pointed out in [13], that a PRBS preserves its structure when XOR-ed with a circularly delayed (by i bits) replica of itself, i.e. if d(n) is a PRBS of length L, then d(n) ⊕ d(n-i) = d(n-k) for i ≠ 0, m⋅L and k ≠ 0, i with i, k, m and n being integers and ⊕ indicating XOR-operation. The resulting shift k can be found numerically by evaluating the cross correlation of d(n) and d(n-k). Using the delays for d₁…d₃ as before and applying this rule twice to d₄ = d₁⊕d₂⊕d₃ results in a required shift of 20739 bits for d₄ relative to d₁. The resulting eye diagrams generated with the DP-I/Q modulator are shown in Fig. 2 (b) for PS-QPSK and Fig. 2 (e) for PDM-QPSK.

 

Fig. 1 (a) Experimental setup of 112 Gb/s PS-QPSK and DP-QPSK eight-channel WDM transmitter, circulating transmission loop and coherent receiver. (b) detailed schematic of the transmitter setup, (c) PolSK modulator and Pol-MUX emulator used for the neighboring channels. ECL: external cavity laser, DFF: D-flip-flop, Pol-Ctrl: polarization controller, MZM: Mach-Zehnder modulator, BPF: optical band-pass filter, VOA: variable optical attenuator.

Download Full Size | PDF

 

Fig. 2 (a) Spectrum of 8 × 112 Gb/s PS-QPSK after 50 GHz ILV (red) and after 10000 km transmission (blue), (b) eye diagram of PS-QPSK before (top) and after (bottom) 50 GHz ILV, (c) received back-to-back constellation for 112 Gb/s PS-QPSK, 40 dB OSNR (d) –(f) spectra, eye diagram and constellation for 112 Gb/s PDM-QPSK.

Download Full Size | PDF

The lower modulator rail consists of a single-polarization I/Q modulator (Fujitsu FTM7961EX) driven by ${\overline{d}}_1$ and ${\overline{d}}_2$ in order to generate a single-polarization QPSK signal. For the generation of 112 Gb/s PS-QPSK, the I/Q modulator is followed by a polarization-shift keying (PolSK) stage as proposed in [14]. In the PolSK stage, the 37.3 GBd QPSK signal is split by a polarization beam splitter (PBS) and amplitude-modulated by two inversely driven (d₄ and${\overline{d}}_4$) Mach-Zehnder modulators (Oclaro SD40). The signals are subsequently recombined in a polarization beam combiner (PBC). For generation of 112 Gb/s PDM-QPSK, the BPG was clocked at 28 GBd and the PolSK stage was replaced by a polarization-multiplex (Pol-MUX) emulator, consisting of a PBS at its input, a passive delay stage with a relative delay of 33 ns (924 symbols) for decorrelation and a PBC to recombine the polarization tributaries.

Because of its better PS-QPSK performance, we used the upper DP-I/Q modulator rail throughout the measurements to modulate the even or odd channel group under test, whereas the lower rail was used for the complementary odd or even neighbor channels, respectively.

After boosting by EDFAs, the two modulated optical combs are combined with a programmable optical interleaver filter (ILV) onto the 50 GHz ITU grid and sent to the 250 km recirculating fiber loop. The loop timing is determined by two acousto-optical switches (AOS) driven by complementary control pulses. The EDFA-amplified transmission loop consists of two 80 km spans and one 90 km span of ULAF (kindly provided by OFS), with chromatic dispersion (CD) of 20 ps/(km∙nm), dispersion slope of 0.07 ps/(km∙nm²), attenuation of 0.184 dB/km, and an effective area of 126 µm² (parameters averaged over fiber segments). The fiber launch powers and unity loop gain are set by variable optical attenuators included in the EDFAs. A 5 nm optical band pass filter and a programmable gain equalization filter (GEQ) were used to suppress noise accumulation outside the signal band and to correct wavelength dependent EDFA gain. We also inserted a second AOS inside the loop (not shown in setup) to compensate for the Doppler frequency shift induced by the other loop-internal AOS. Optical WDM spectra after the transmitter and after transmission over 10000 km are shown in Fig. 2 (a) and (d).

At the receiver, the signal can be artificially noise-loaded for back-to-back performance characterization. Subsequently, the channel under test is selected by a 0.4 nm optical band pass filter before being fed into a polarization-diversity optical 90°-hybrid. The local oscillator (LO) laser is an ECL with ~100 kHz line width. After detection by four balanced photo-detectors (BD) the signals are digitized by a real-time oscilloscope (Tektronix DPO72004) with 20 GHz bandwidth and 50 GSa/s sampling rate. Digital signal processing is performed offline and includes corrections of optical frontend imbalances, CD and residual frequency offset. An adaptive 2 × 2 MIMO equalizer separates the signal polarizations and compensates for residual CD, PMD and laser phase noise. The filter coefficients are updated by a modified constant modulus algorithm (CMA) as proposed in [15]. After pre-convergence is obtained, the equalizer is switched to decision-directed mode. The equalizer feedback criterion as well as the symbol decision is based on minimum Euclidean distance in 4-D space. Finally, after symbol de-mapping and synchronization, bit errors are counted. The received constellation diagrams for maximum OSNR are shown in Fig. 2 (c) and (f) for PS-QPSK and PDM-QPSK respectively.

3. Results and discussion

The measured back-to-back BER as a function of OSNR (for 0.1 nm noise bandwidth) is shown in Fig. 3 (a) . Solid lines indicate the theoretical noise-limited BER derived for PS-QPSK in [6] and for PDM-QPSK in [16]. For a BER of 10⁻³, the measured required OSNR for PS-QPSK is ≈13.5 dB, while it is ≈13.9 dB for PDM-QPSK, resulting in 0.4 dB sensitivity advantage of PS-QPSK. This is in contrast to the theoretically predicted difference of about 1 dB. The reduced sensitivity advantage for PS-QPSK is due to the higher implementation penalty (≈1.2 dB at a BER of 10⁻³) compared to PDM-QPSK (≈0.6 dB). Comparing single-channel and WDM results in Fig. 3, no penalty due to crosstalk is observed. Therefore, we attribute the higher implementation penalty to the increased symbol rate (by a factor of 4/3) for PS-QPSK, where bandwidth limitations of the electrical and electro-optical components are already more pronounced. The sensitivity advantage of PS-QPSK decreases to 0.3 dB near the FEC limit of 3.8∙10⁻³. Figure 3 (b) and (c) show plots of the phase plane of X- and Y-polarizations of the received PS-QPSK and PDM-QPSK signals at a BER of 10⁻³. Due to the higher symbol distances, PS-QPSK can tolerate more noise, resulting in enlarged constellation points, while achieving the same BER.

 

Fig. 3 (a) back-to back performance for PS-QPSK and PDM-QPSK in a single-channel and WDM scenario: solid lines indicate theoretical limits and curves with symbols are measured values, (b) and (c) Plots of the received phase planes for PS-QPSK and PDM-QPSK, respectively, both for a BER of 10⁻³.

Download Full Size | PDF

In Fig. 4 (a) , the BER versus fiber launch power per channel for one channel at a wavelength of 1550.92 nm after WDM transmission is shown over various numbers of loop roundtrips. Blue symbols indicate PS-QPSK measurements, while red symbols denote PDM-QPSK. For launch powers below −1 dBm per channel, i.e. in the linear regime, the performance of both modulation formats is similar, which was expected considering the similar back-to-back performance. At higher launch powers, nonlinear effects start to limit the transmission performance. In Fig. 4 (a) it is seen that the optimum launch power is between 1 dB and 2 dB higher for 112-Gb/s PS-QPSK, depending on the transmission length, which indicates a higher nonlinear tolerance of this format compared to 112-Gb/s PDM-QPSK. A similar behavior was observed for example in numerical simulations [8,9] and experimental investigations [11] of a SSMF link. However, this may depend on the actual system configuration, since measurements at 42.9 Gb/s over 80 km SSMF spans [10] did not indicate such a nonlinear advantage for PS-QPSK.

 

Fig. 4 (a) BER vs. fiber launch power per channel for various transmission lengths, blue curves show PS-QPSK, red curves show PDM-QPSK, (b) BER vs. transmission distance (optimized launch power for every distance).

Download Full Size | PDF

In Fig. 4 (b), the BER as a function of distance is shown for optimum launch powers at each distance. Assuming a FEC limit of 3.8∙10⁻³ as upper BER limit, the achievable transmission distances are ≈12000 km and ≈11000 km for PS-QPSK and PDM-QPSK, respectively. To ensure that all eight WDM channels are well below the FEC threshold, the BER of the individual channels were measured at the optimum channel launch power ( + 1 dBm and −1 dBm) after 11000 km and 10000 km transmission for PS-QPSK and PDM-QPSK, respectively. The measured BER as well as the received OSNR for all WDM channels are shown in Fig. 5 (a) and (b) .

 

Fig. 5 BER and OSNR vs. measured channel index (shortest wavelength: 1549.715 nm, longest wavelength: 1552.524 nm), blue circles indicate BER and red triangles denote OSNR. (a) PS-QPSK for + 1 dBm channel launch power after 11000 km, (b) PDM-QPSK for −1 dBm channel launch power after 10000 km transmission.

Download Full Size | PDF

4. Conclusion

In this paper, we present the experimental generation and ultra-long-haul transmission of 8 × 112-Gb/s PS-QPSK on a 50 GHz channel grid over up to 11000 km ULAF with EDFA amplification. We also compare PS-QPSK to PDM-QPSK at the same bit rate. Regarding the noise-limited performance, the theoretically predicted sensitivity improvement for PS-QPSK at a BER of 3.8∙10⁻³ is 0.8 dB whereas we measured only 0.3 dB improvement. We attribute this to the required higher symbol rate of PS-QPSK and the therefore associated higher implementation penalty due to bandwidth constraints of our available components.

Ultra-long-haul transmission results ≥10000 km reveal a higher optimum channel power for PS-QPSK modulation, denoting its higher nonlinear tolerance and, resulting from that, an increased transmission reach by ≈10%.