Open Access
Add to BibSonomy
Abstract
We experimentally demonstrate the transmission of 100 Gbit/s and beyond vestigial sideband (VSB) n-level pulse amplitude modulation (PAM-n) signals, with a commercial 10 GHz-class directly modulated laser (DML) in the C-band, using optical filtering. To mitigate transmission impairments at the transmitter side, Nyquist pulse shaping and Kaiser Window filtering techniques are used to overcome the limited bandwidth of optoelectronic devices. At the receiver side, the joint nonlinear equalization based on cascaded multi-modulus algorithm (CMMA) and Volterra Filter (VF) is used to reduce the strong nonlinear impairments from chirp and chromatic dispersion (CD). 100 Gb/s PAM-4, 107.5 Gb/s PAM-4, and 101.25 Gb/s PAM-8 signals can be successfully transmitted over 45 km, 10 km and 10 km standard single-mode fiber (SSMF) under the bit-error-ratio (BER) of 3.8 × 10−3, respectively.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
The IEEE Standard 802.3bs-2017 has been developed by the IEEE P802.3bs 200 Gb/s and 400 Gb/s Ethernet Task Force and approved by the IEEE-SA Standards Board, and 4-lane × 100 Gbit/s/λ has been standardized to realize 400GBASE-DR4 [1]. Driven by bandwidth-hungry cloud services, high definition TV, and interactive games, there occurs an urgent requirement of transmission capacity for intra- or inter- data center interconnection (DCI). In typical data center applications, the intensity-modulation and direct-detection (IM/DD) transmission based on n-level pulse amplitude modulation (PAM-n) has been demonstrated with a net bit rate of more than 100 Gb/s, due to its lower cost, smaller footprint, and lower power consumption. Directly modulated laser (DML) or electro-absorption modulated laser (EML)-based transmitters are generally more preferred than externally modulated transmitters based on Mach-Zehnder modulators (MZM). For the DML-based IM/DD transmission system, there are two main transmission impairments: the nonlinear impairments from the interaction between chirp and chromatic dispersion (CD) [2] and the bandwidth constraint of the optoelectronic devices.
Recently, various advanced digital signal processing (DSP) algorithms have been proposed to handle these two limitations. By applying intensity directed equalizer based on feedforward equalizer and decision feedback equalizer (FFE/DFE), 56 Gb/s PAM-4 signal transmission over 43 km standard single-mode fiber (SSMF) was demonstrated using one 16.8 GHz DML in the C-band [2]. 64 Gb/s PAM-4 signal transmission over 70 km SSMF was demonstrated based on a O-band 18 GHz DML, using sparse Volterra filter (SVF) equalizer [3]. 100 Gb/s PAM-4 signal transmission over 15 km SSMF was reported using one 16.8 GHz DML in the C-band with pre-coding and maximum likelihood sequence estimation (MLSE) [4]. Single-wavelength 112 Gb/s Nyqiust PAM-4 signal transmission over 40 km SSMF was realized using one 22 GHz performance-enhanced O-band DML with FFE/DFE [5]. The FFE/DFE or modified methods with less taps have been widely researched in intra- or inter- DCI communication systems. However, the error propagation can degrade the FFE/DFE performance, especially when the signal has a strong nonlinear impairment. Moreover, the FFE/DFE requires a training sequence to obtain the equalizer coefficients, which will increase the system overhead and therefore is not suitable for the agile high-speed DCI networks. In [6], a novel constant-modulus-algorithm (CMA) was proposed for the compensation of the static CD and time-varying polarization mode dispersion (PMD), without using the training sequence. This blind equalizer can automatically find the optimal tap coefficients using the constant amplitude in a PAM-2 fiber transmission system, but it is not suitable for higher-order PAM-n optical communication systems. In this paper, we first use the cascaded multi-modulus algorithm (CMMA) blind equalizer as an effective solution for the high-speed PAM-n transmission system based on a DML, which can adapt varying radii of the PAM-n signal. Furthermore, most previous researches focus on 50 or 56 Gb/s PAM-4 signal transmission based on a 10 GHz-class DML in the C-band. Single-lane 100 Gb/s transmission is difficult to realize due to the strong bandwidth constraint of optoelectronic devices and the nonlinear impairments.
In this paper, 100 Gb/s PAM-4 and PAM-8 signal transmissions based on a 10 GHz C-band DML and IM/DD have been experimentally demonstrated. The bandwidth constraint of the optoelectronic devices can be reduced by Nyquist pulse shaping and Kaiser Window filtering techniques at the transmitter side. The optical vestigial sideband (VSB) modulation by using an optical filter can mitigate the frequency-selective power fading. Furthermore, the joint nonlinear equalization based on CMMA and Volterra Filter (VF) at the receiver side is used to reduce the strong nonlinear impairments from the interaction between chirp and CD. Finally, 80 Gb/s PAM-4, 100 Gb/s PAM-4, and 101.25 Gb/s PAM-8 signals can be transmitted over 80 km, 45 km, and 10 km SSMF fiber with the bit-error-ratio (BER) under the 7% hard-decision forward-error-correction (HD-FEC) threshold of 3.8 × 10−3.
2. Principle of the transmitter and receiver offline DSP
2.1 Nyquist pulse shaping and Kaiser Window resampling
In the transmitter offline DSP, Gray-mapped PAM-n symbols with a length of 215 is generated, and then oversampled to two samples per symbol (2-sps). The filtering effect due to the limitation of the digital-to-analog converter (DAC) and DML bandwidth can result in inter-symbol interference (ISI) and restrict the baseband signal bandwidth. Root-raised-cosine (RRC) filter with the optimal roll-off factor is used to realize Nyquist pulse shaping and simultaneously reduce ISI. Figure 1(a) shows the frequency response of the RRC filter with different roll-off factors α for a given tap number (512). Decreasing α can reduce the baseband signal bandwidth. Next, the symbol sequence is resampled to 1-sps with Kaiser Window filter [7], which can avoid frequency aliasing, can be expressed as
where I0 is the zeroth-order modified Bessel function of the first kind, N is the length of the window, β is the shape parameter which determines main-lobe width and side-lobe level. The frequency response of the Kaiser Window with different lengths of FIR filter and shape parameters is shown in Fig. 1(b). Increasing β can decrease the amplitude of the side lobes and increase the energy concentration in the main lobe. Higher resampling accuracy will be attained by increasing the length of the window, but at the cost of a longer computation time. n = 10 and β = 5 are considered in our transmission experiment. Finally, the symbol sequence is quantized in 8-bits (−128~127) before it is loaded to the DAC. Figure 1(c) shows the PAM-4 symbols after quantization with the transmitter offline DSP. Insets (i) and (ii) are the electrical eye diagrams of the DAC output (10-ps/div, 50-mv/div) of the 50-Gbaud PAM-4 signal with and without Kaiser Window filter only, respectively. Clear eye diagrams can be observed at the cost of a slight power decrease. Insets (iii) and (iv) are the electrical eye diagrams with and without RRC pulse shaping after Kaiser Window filtering, respectively. With the Kaiser Window filter, we can find that electrical eye diagram can be legible.
Fig. 1 The offline DSP blocks. (a) and (b) are the frequency response of RRC filter and Kaiser Window filter, respectively. (c) PAM-4 symbols after quantization. Insets are the electrical eye diagrams of 50 Gbaud PAM-4: (i) with Kaiser Window filtering only, (ii) without Kaiser Window filtering only, (iii) with RRC filtering after Kaiser Window filtering, and (iv) without RRC filtering after Kaiser Window filtering.
2.2 Joint nonlinear equalization base on CMMA and Volterra filter
In the receiver offline DSP, the captured offline data is first resampled to 1-sps with the matched Kaiser Window filter, and then a squaring time recovery is applied to remove the timing offset and jitter from the data. Afterwards, the joint nonlinear equalization algorithm with the optimized 19-tap CMMA [8] at 1-sps and 389-tap VF at 1-sps is used to reduce the nonlinear impairments. The tap coefficients of the CMMA blind equalizer can be given as
where μ is the filter convergence step, and the error function can be written asThe modified coefficient can be given as followsHere, A1 = (R1 + R2)/2, A2 = (R3 – R2)/2, …, Ak–1 = (Rk – Rk–2)/2, and Ak = (Rk – Rk–1)/2, where R1, R2, …, and Rk are the radii of the PAM-n signal symmetric levels. Figure 2(a) shows the schematic of the PAM-8 signal, and there are four radii of symmetric eight levels. Similarly, there are two radii of PAM-4 symmetric amplitudes. Compared with CMA, the CMMA algorithm calculates the output error by using multiple cascaded constant modes, to ensure that the filter error value of the high-order PAM-n signals is still zero under ideal conditions. We take the PAM-4 signal as an example, and the tap coefficients of CMMA after 50 Gbaud PAM-4 signal equalization with a 10 GHz DML are shown in Fig. 2(b). After around 5 × 105 symbols, the tap coefficients are convergent. The finite-impulse-response (FIR) of CMMA is shown in Fig. 2(c). Afterwards, the VF based on 2nd-order Volterra series is applied to further compensate for the nonlinear impairments. The VF can be expressed aswhere N1 and N2 are the tap numbers of the linear and nonlinear terms. wk1 and wk1k2 are the VF weight coefficients and can be updated according to the Least Mean Square (LMS) error function by using a reference training sequence, respectively. After CMMA and VF equalization, the decision-directed LMS (DD-LMS) equalizer with 189-tap at 1-sps can be applied before final decision, to further compensate for the channel response and to mitigate the implementation penalty of devices [9]. Finally, the BER can be calculated after PAM-n demodulation based on the recovered signal.Fig. 2 The offline CMMA blind equalizer. (a) The schematic for the radii of PAM-8 signal. (b) The tap coefficients of CMMA. (c) The FIR of CMMA.
3. Experimental setup
Figure 3 shows the experimental setup of the PAM-n IM/DD transmission system based on one 10 GHz-class DML in the C-band. At the transmitter side, the PAM-4 or −8 drive signals are generated by a DAC (Fujitsu) working at 80 Gsa/s with 3 dB analog bandwidth of 20 GHz. However, the clock leakage can always occur in the high-speed DAC and analog-to-digital converter (ADC) [10]. In [11], the Adaptive Notch Filter (ANF) algorithm was proposed to suppress the narrowband interference induced by the DAC clock leakage, and the results show that about 1.3 dB receiver sensitivity improvement can be obtained at the BER of 3.8 × 10−3 in a 112 Gb/s PAM-4 IM/DD system based on MZM. In this paper, the PAM-n signals from the DAC pass through one simple 40 GHz broadband balun (BB) for combining differential signals into a single-ended signal, to successfully suppress the 20 GHz DAC clock leakage and to avoid narrowband interference. Figure 3(a) shows the optical spectra with and without the clock leakage mitigation measured at the 0.02 nm resolution. One commercial DML (NLK1551SSC) is operated at 25 °C temperature. The single-ended signal is attenuated by a 3 dB attenuator (ATT) to reduce the impact of nonlinear impairments, and then amplified by a 25 dB electrical amplifier (EA) before driving the DML.
Fig. 3 Experimental setup for PAM-n IM/DD transmission system. (a) The optical spectra without and with DAC clock leakage mitigation. (b) The measured P-I-V curve of the 10 GHz DML. (c) The extinction ratio versus bias current with 50-Gbaud PAM-4 signal without OTF.
After 0~80 km SSMF transmission, two experimental cases are considered and compared, that is, case-1 with Erbium-doped fiber amplifier (EDFA) only and case-2 with an optical tunable filter (OTF) between two-stage EDFAs. The OTF with a passband of 0.3 nm is applied to generate optical VSB signal to mitigate the frequency-selective power fading so as to increase the CD tolerance [12,13]. Then, the filtered optical signal passes through a variable optical attenuator (VOA), which is applied to adjust the received optical power for sensitivity measurement. At the receiver side, the signal is detected by a 50 GHz photo detector (PD) and amplified by a 60 GHz EA, and then captured by a real-time 160 GSa/s oscilloscope with 62 GHz bandwidth and processed by offline DSP. Figure 3(b) shows the measured P-I-V curve of the 10 GHz DML. The threshold current is about 20 mA. We also measured the extinction ratio (ER) versus the drive current for the 50 Gbaud PAM-4 optical signal without OTF as shown in Fig. 3(c). Higher output power can help suppress the transient chirp and enhance the device bandwidth, but it will also decrease the ER [14]. For the 50 Gbaud PAM-4 signal, the optimum drive current, optical power and ER are 100 mA, 6.6 dBm, and 1.15, respectively. The ER needs to be optimized for different PAM-n transmission rates and fiber distances.
4. Experiment results and discussions
4.1 The optimization of experimental parameters
Before the transmission experiment, we firstly take the 50 Gbaud PAM-4 signal at back-to-back (BTB) case as an example to optimize the roll-off factor of RRC and the number of equalizer taps. The BER under different roll-off factors for Nyquist pulse shaping at 4 dBm received optical power (ROP) is shown in Fig. 4(a), and the optimized Nyquist pulse shaping roll-off factor is 0.4. The relationship between the BER and the tap number of FFE/DFE only and CMMA only is shown in Fig. 4(b) at 2 dBm ROP with the same tap number. The main purpose of FFE/DFE and CMMA is to pre-converge data for Volterra Filter processing. We can see that the FFE/DFE is invalid, because the limitation of transmission impairments and DML bandwidth can result in error propagation. Insets (i) ~(iv) show the symbols and eye diagrams after 19-tap FFE/DFE only and 19-tap CMMA only, respectively. The taps number of FFE/DFE and CMMA are very small, and the BER performance is manly promoted by Volterra Filter. Figure 4(c) shows the BER versus different equalizer tap umbers for different cases at 2 dBm ROP. We can observe that the joint nonlinear equalization based on CMMA and VF with 189-tap DD-LMS has the best performance. The optimized tap numbers of CMMA and VF are 19 and 389, respectively.
Fig. 4 Optimization of experimental parameters. (a) The BER versus roll-off factor at 4-dBm ROP. (b) The BER versus the number of equalizer taps with FFE/DFE only and CMMA only at 2-dBm ROP. (c) The BER versus the number of VF equalizer taps. Insets (i) ~(iv) show the symbols and eye diagrams after 19-tap FFE/DFE only and 19-tap CMMA only, respectively.
4.2 PAM-4 transmission experiment results
We have initially tested the performance of the PAM-4 transmission without OTF. The BER versus bit rate for BTB, 10 km, 20 km, and 30 km without optical filter are shown in Fig. 5(a). For the fixed ROP at 4 dBm, it is shown that 80 Gb/s (40 Gbaud) were achieved for all distances, and 100 Gb/s was achieved for 10 km and 20 km at the BER of 3.8 × 10−3 (7% HD-FEC threshold), respectively. Especially, 107.5 Gb/s (53.75 Gbaud) was also achieved for BTB and 10 km. This is the highest data rate for a single-lane IM-DD system working in the C-band based on a 10G-class DML. Figure 5(b) shows the BER performance against the ROP with the 100 Gb/s (50 Gbaud) PAM-4 signal for BTB, 10 km, 20 km, and 30 km without optical filtering, respectively. We can observe that 100 Gb/s was transmitted for 10 km and 20 km without any CD compensation methods at the BER of 3.8 × 10−3. But, 30 km transmission cannot be achieved under the HD-FEC threshold. Insets (i) and (ii) show the recovered symbols and eye diagram for 50 Gbaud PAM-4 after 20 km SSMF at 2 dBm ROP, respectively.
Fig. 5 The BER for different fiber distances without OTF versus (a) different bit rates; (b) received optical power. Insets (i) and (ii) are recovered symbols and eye diagram, respectively.
Figure 6(a) shows the optical spectra of the 50 Gbaud PAM-4 without and with OTF measured at the 0.02 nm resolution. Figure 6(b) shows the electrical spectra of the received 50 Gbaud PAM4 signal for different cases. We can find that the electrical bandwidth of the received signal is about 24 GHz in BTB case, and the severe power fading induced by the interaction between chirp and CD is observed after 45 km SSMF transmission without OTF. After optical filtering, the notches due to the power fading can be mitigated. Figure 6(c) shows the BER performance versus ROP for 50 Gbaud PAM-4 without and with OTF after 45 km fiber. Under the 7% HD-FEC threshold, the use of OTF can increase the transmission distance from less than 20 km to 45 km. Insets (i) and (ii) are the recovered eye diagram and histogram at 0-dBm ROP after 45 km fiber with OTF, respectively. After 80 km fiber with OTF, 80 Gb/s PAM-4 can be under the BER of 3.8 × 10−3 as shown in Fig. 6(d).
Fig. 6 (a) The optical spectra of 50 Gbaud PAM-4 with and without OTF; (b) the electrical spectra of received 50 Gbaud PAM-4 signal for different cases; (c) the BER versus ROP for 50 Gbaud PAM-4 with and without OTF after 45-km fiber; (d) the BER versus ROP for 40-, 45-, and 50-Gbaud with OTF after 80-km fiber, respectively. Insets (i) and (ii) are recovered eye diagram and histogram at 0 dBm ROP, respectively.
4.3 PAM-8 transmission experiment results
Finally, we tested the performance of PAM-8 transmission with two cases, and the optimized DML drive current is 103 mA. As an illustration, Fig. 7(a) shows the optical spectra of 33.75 Gbaud PAM-8 without and with OTF measured at the 0.02 nm resolution. At the fixed 4 dBm ROP, the BER versus bit rate for BTB, 1 km, and 10 km without optical filter are shown in Fig. 7(b). 75 Gb/s (25 Gbaud) and 101.25 Gb/s (33.75-Gbaud) PAM-8 signals are under the 7% HD-FEC threshold for BTB and 1 km cases, and cannot be transmitted over 10 km fiber without optical filtering. The BER performance versus ROP for 33.75 Gbaud PAM-8 without and with OTF after 10 km fiber transmission is shown in Fig. 7(c). We can observe that 10 km transmission can be achieved under the HD-FEC threshold. Insets (i) and (ii) show the recovered symbols and eye diagram for 33.75 Gbaud PAM-8 after 10-km SSMF at 4-dBm ROP, respectively. Thanks to the joint nonlinear equalization algorithm based on CMMA and VF, the recovered PAM-8 signal histogram is shown in Fig. 7(d). The CMMA blind equalization is very effective for high-order PAM-n signals.
Fig. 7 (a) The optical spectra of 33.75 Gbaud PAM-8 with and without OTF; (b) the BER versus bit rates for different fiber distances without OTF; (c) the BER versus ROP for 33.75 Gbaud PAM-8 with and without OTF after 10 km fiber; (d) the amplitude distribution histogram of recovered PAM-8 signal at 4 dBm ROP. Insets (i) and (ii) are recovered symbols and eye diagram at 4 dBm ROP after 10 km fiber with OTF, respectively.
5. Conclusion
We experimentally demonstrated Nyquist PAM-4 and −8 signals IM/DD transmission system based on a 10 GHz-class DML with optical filtering and effective DSP, respectively. Thanks to RRC pulse shaping, Kaiser Window filtering, and the combination of CMMA and VF nonlinear equalization, 80 Gb/s PAM-4, 100 Gb/s PAM-4, and 101.25 Gb/s PAM-8 signals can be transmitted over 80 km, 45 km and 10 km SSMF with a BER of 3.8 × 10−3, respectively. 107.5 Gb/s PAM-4 transmission over 10 km below the BER of 3.8 × 10−3 without optical filtering can also be achieved. This is the first time to demonstrate >100 Gb/s per lane IM/DD transmission using a 10G-class DML in the C-band for four-lane 400-GE inter-DCI applications.
Funding
National Natural Science Foundation of China (NSFC) (61527801, 61675048, 61720106015, 61835002, 61805043).
References
1. IEEE 802.3bs-2017 [Online]. https://standards.ieee.org/standard/802_3bs-2017.html.
2. K. Zhang, Q. Zhuge, H. Xin, M. Morsy-Osman, E. El-Fiky, L. Yi, W. Hu, and D. V. Plant, “Intensity directed equalizer for the mitigation of DML chirp induced distortion in dispersion-unmanaged C-band PAM transmission,” Opt. Express 25(23), 28123–28135 (2017). [CrossRef]
3. F. Gao, S. Zhou, X. Li, S. Fu, L. Deng, M. Tang, D. Liu, and Q. Yang, “2 × 64 Gb/s PAM-4 transmission over 70 km SSMF using O-band 18G-class directly modulated lasers (DMLs),” Opt. Express 25(7), 7230–7237 (2017). [CrossRef] [PubMed]
4. H. Xin, K. Zhang, Q. Zhuge, L. Yi, H. He, W. Hu, and D. Plant, “Transmission of 100Gb/s PAM4 Signals Over 15km Dispersion-unmanaged SSMF Using a Directly Modulated Laser in C-band,” in Proceedings of European Conference on Optical Communication (IEEE, 2018). [CrossRef]
5. W. Wang, P. Zhao, Z. Zhang, H. Li, D. Zang, N. Zhu, and Y. Lu, “First demonstration of 112 Gb/s PAM-4 amplifier-free transmission over a record reach of 40 km using 1.3 μm directly modulated laser.” in optical Fiber Communication Conference (2018), paper Th4B. 8.
6. F. F. Dai, “Electronic equalizations for optical fiber dispersion compensation,” Opt. Eng. 46(3), 035006 (2007). [CrossRef]
7. J. Shi, J. Zhang, Y. Zhou, Y. Wang, N. Chi, and J. Yu, “Transmission performance comparison for 100Gb/s PAM-4, CAP-16 and DFT-S OFDM with direct detection,” J. Lightwave Technol. 35(23), 5127–5133 (2017). [CrossRef]
8. X. Zhou, J. Yu, and P. D. Magill, “Cascaded two-modulus algorithm for blind polarization de-multiplexing of 114-Gb/s PDM-8-QAM optical signals,” in optical Fiber Communication Conference (2009), paper OWG3.
9. F. Li, J. Yu, Z. Cao, J. Zhang, M. Chen, and X. Li, “Experimental demonstration of four-channel WDM 560 Gbit/s 128QAM-DMT using IM/DD for 2-km optical interconnect,” J. Lightwave Technol. 35(4), 941–948 (2017). [CrossRef]
10. Y. Zhu, W. Peng, Y. Cui, C. Kan, F. Zhu, and Y. Bai, “Comparative digital mitigations of DAC clock tone leakage in a single-carrier 400G system,” in Optical Fiber Communication Conference (2015), paper Th2A.17. [CrossRef]
11. F. Li, D. Zou, L. Ding, Y. Sun, J. Li, Q. Sui, L. Li, X. Yi, and Z. Li, “100 Gbit/s PAM4 signal transmission and reception for 2-km interconnect with adaptive notch filter for narrowband interference,” Opt. Express 26(18), 24066–24074 (2018). [CrossRef] [PubMed]
12. S. Zhou, X. Li, L. Yi, Q. Yang, and S. Fu, “Transmission of 2 × 56 Gb/s PAM-4 signal over 100 km SSMF using 18 GHz DMLs,” Opt. Lett. 41(8), 1805–1808 (2016). [CrossRef] [PubMed]
13. J. Lee, N. Kaneda, and Y. K. Chen, “112-Gbit/s Intensity-Modulated Direct-Detect Vestigial-Sideband PAM4 Transmission over an 80-km SSMF Link,” in Proceedings of European Conference on Optical Communication, (2016), paper M.2.D.3.
14. D. Mahgerefteh, Y. Matsui, X. Zheng, and K. McCallion, “Chirp managed laser and applications,” IEEE J. Sel. Top. Quantum Electron. 16(5), 1126–1139 (2010). [CrossRef]
