We demonstrate a 10 Gsymbol/s, 1024 quadrature amplitude modulation (QAM) 160 km coherent transmission with an injection locking technique. Our newly developed, pilot-assisted adaptive equalizer has greatly improved the precision of waveform distortion compensation, and this has enabled us to increase the symbol rate to 10 Gsymbol/s in a 1024 QAM transmission. Thus, we could realize a 200 Gbit/s, 1024 QAM transmission over 160 km with a potential spectral efficiency of 12.6 bit/s/Hz.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Recently, an injection locking technique has attracted much attention as a precise optical carrier-phase synchronization scheme and has been applied to multi-level digital coherent optical transmissions [1, 2]. This technique enables us to realize low noise carrier-phase locking between the transmitted data signal and a local oscillator (LO) with a simple configuration . Furthermore, a great advantage of this scheme is that there is no need to use a narrow linewidth laser as an LO [3, 4].
We have already developed an injection-locked homodyne detection system to which we introduced a continuous wave (CW) pilot tone (PT) signal close to the quadrature amplitude modulation (QAM) spectrum and we used this as a seed signal for injection locking into an LO . With this system, we have transmitted 552 Gbit/s, 46 Gsymbol/s-64 QAM , 320 Gbit/s, 20 Gsymbol/s-256 QAM , and 216 Gbit/s, 12 Gsymbol/s-512 QAM  signals over 160 km. However, we have found it difficult to increase the QAM multiplicity to 1024 at a symbol rate exceeding 10 Gsymbol/s, where we used an adaptive finite impulse response (FIR) filter based on a decision directed least mean square (DD-LMS) algorithm  to compensate for time-varying waveform distortion. When the optical signal-to-noise ratio (OSNR) is low, the waveform equalization performance in such an equalizer is degraded for an ultra-multi-level QAM transmission, where filter tap coefficients are updated with a decision threshold.
In this study, we greatly improve the waveform distortion compensation performance by developing a pilot-assisted adaptive equalizer with an FIR digital filter  for our injection-locked transmission scheme. With this technique, we were able to increase the symbol rate from 3 Gsymbol/s  to 10 Gsymbol/s in a 1024 QAM transmission and realize a single-channel 200 Gbit/s transmission over 160 km with the highest symbol rate at 1024 QAM. The potential spectral efficiency (SE) reached as high as 12.6 bit/s/Hz. Here, we clarify the importance of this equalization scheme in an ultra-high speed and ultra-multilevel QAM coherent transmission.
2. Experimental setup for pol-mux, 10 Gsymbol/s-1024 QAM injection-locked coherent transmission over 160 km
Figure 1 shows our experimental setup for a 10 Gsymbol/s, 1024 QAM injection-locked coherent transmission. The configuration is similar to that described in our previous papers [4,7] except for the data signal including pilot symbols generated from an arbitrary waveform generator (AWG) and the time-domain waveform equalization scheme at the digital signal processor (DSP).
At the transmitter, the output beam of an external cavity laser diode (ECLD)  was modulated by an IQ modulator with a 10 Gsymbol/s, 1024 QAM baseband signal and a PT signal from the AWG. The AWG was operated at 60 Gsample/s with an 8-bit resolution and directly generated a 1 Vpp (peak-to-peak) 1024 QAM signal . Here, the pattern length was 4096 symbols including 19 header symbols (0.5%) for synchronization and 32 pilot symbols (0.8%) for waveform equalization. In our system, the pattern length was limited by the memory size of the demodulation software. The bandwidth of the baseband data signal was reduced to 6 GHz using a root raised cosine filter with a roll-off factor of 0.2. The frequency detuning between the data carrier and the PT was set at 6.56 GHz. The power ratio between the PT and the data signal, Ppilot/Pdata was optimized to −10 dB to obtain the best demodulation performance . Here, polarization multiplexing was carried out by using a polarization-multiplexing emulator with a symbol delay of 1 ns. These signals were transmitted over 160 km ultra-large-area (ULA) fiber with a launch power of 1 dBm (−2 dBm/pol data, −12 dBm/pol PT). The average fiber loss of each span was 16.4 dB, which was compensated for with a Raman amplifier and an EDFA. The Raman amplifiers and EDFAs provided gains of 9.5 and 6.9 dB, respectively.
At the receiver, a distributed feedback laser diode (DFB LD)  used as an LO was injection-locked by a PT signal extracted with an etalon filter. Here, the PT was coupled into the LO with an injection power of −10 dBm . The output power and the locking range of the LO were 10 dBm and 5.8 GHz, respectively . In our injection locking circuit, the LO was phase-locked to the data signal through the PT with a phase noise of 0.24 degrees. This value is sufficiently small for the demodulation of 1024 QAM data. The phase noise characteristic of the PT signal remained unchanged after transmission. The transmitted 1024 QAM data signals were homodyne-detected with the frequency-shifted, injection-locked DFB LD output. The frequency shift of 6.56 GHz given in a PT at the transmitter was compensated for by the LiNbO3 (LN) intensity modulator driven at 6.56 GHz. This enabled the homodyne-detection. The optical path length difference between the QAM data path and the LO path including the injection locking circuit was adjusted to within a difference of 10 cm. Since the AWG and the synthesizer used to drive the LN intensity modulator were not synchronized, there was a small static phase difference and a slow phase fluctuation between them. We could compensate for these phenomena by employing a DSP at the receiver. However, there was no frequency offset between the frequency-shifted LO and the carrier frequency of transmitted data signal. Figure 2 shows the electrical spectrum of the homodyne-detected 10 Gsymbol/s, 1024 QAM signal and PT in a back-to-back condition. The homodyne-detected 1024 QAM signals were then analog-to-digital (A/D) converted by using a digital oscilloscope operated at 40 Gsample/s with an 8-bit resolution and demodulated with a DSP in an offline condition [4, 7].
At the DSP, we included a newly-developed pilot-assisted adaptive FIR filter to compensate for the time-varying waveform distortions as well as conventional polarization demultiplexing , frequency domain equalization (FDE) , and digital back propagation . The tap number of the pilot-assisted adaptive FIR filter was 99.
Figure 3(a) shows the configuration of a conventional FIR filter, and Figs. 3(b) and 3(c), respectively, compare the differences in the FIR tap coefficient updating processes with a DD-LMS method and a pilot-assisted method. Figure 3(b) represents the data frame configuration used in the DD-LMS method without pilot symbols. Here, we show a 64 QAM constellation when data symbols with the maximum amplitude are received. Although the two symbols, x(t1) and x(t2) originally have the same amplitude, the symbol datum x(t2) exceeds the symbol decision threshold due to large waveform distortions that occur during the transmission. With this DD-LMS algorithm, an error signal is calculated after the decision threshold. Therefore, in an ultra-multi-level QAM transmission, which accompanies many symbol decisions, the FIR tap coefficients are sometimes updated with an incorrect error signal. This results in the degradation of the demodulation performance. To avoid incorrect FIR tap coefficient updating, training sequences are generally employed with lower-order QAM signals or known symbols before the operation of a DD-LMS algorithm with a higher-order QAM signal . However, in our system, static waveform distortions caused by hardware imperfections were removed by employing high-resolution FDE in a training mode. Therefore, we did not use any training sequences with lower-order QAM signals or known symbols in the DD-LMS algorithm-based equalizer. In contrast, Fig. 3(c) shows the data frame configuration adopted in the pilot-assisted method, and depicts part of the 64 QAM constellation map with received pilot symbols x(t1) and x(t2) including waveform distortions. In the pilot-assisted equalizer, the FIR tap coefficients are simply updated so that an error signal between an actual received pilot symbol (blue) and a known ideal pilot symbol (red) becomes zero. Here, we used a fixed pattern QPSK signal with an amplitude of 0.52 V as a pilot symbol. The disadvantage of this pilot-assisted equalizer is the reduction in the transmission speed. However, in the present case, we used only 0.8% of the original data as the pilot symbols, and so the SE was not greatly reduced.
Finally, we evaluated the bit error rate (BER) from 205 kbit data. There was a slow phase fluctuation between the injection-locked LO and the transmitted data signal due to their optical path difference. We compensated for such a slow phase fluctuation by incorporating an adaptive FIR filter in the receiver DSP.
3. Transmission results
First, we evaluated the BER characteristics of 1024 QAM before and after 160 km transmissions at symbol rates of 3, 5, 6, 7, 8 and 10 Gsymbol/s. Figure 4(a) shows their BER characteristics. In Fig. 4(b), we also show the theoretical BER curves of 1024 QAM as a function of OSNR at a 0.1 nm bandwidth for different symbol rates. In Fig. 4(a), the transmission powers at each symbol rate are individually optimized. Here, the optimum launch powers at symbol rates of 3, 5, 6, 7, 8, and 10 Gsymbol/s were −2, −1, −1, 0, 1, and 1 dBm, respectively. We obtained all the BER characteristics after a 160 km transmission with homodyne-detected data signals that had an OSNR of 38 dB at a 0.1 nm bandwidth, while we measured all the back-to-back BERs with data signals that had a 44 dB OSNR. When the OSNR is 44 dB, in theory a BER below 1.0 x 10−5 can be achieved even at 10 Gsymbol/s. However, in this experiment, the back-to-back BERs were 4.2 x 10−3 at 10 Gsymbol/s and around 1.0 x 10−4 at 3 Gsymbol/s. These OSNR penalties compared with the theory may be caused by the insufficient SNR of the electrical baseband signal generated by the AWG and nonlinear waveform distortions caused by the non-ideal frequency response characteristics of the hardware such as the IQ modulator. When the symbol rate increases, the BERs degrade. This includes the OSNR penalty. All the BERs after transmission were below the forward error correction (FEC) threshold of 3.8 x 10−2 for a 25.5% overhead .
Next, we detail the transmission results at 10 Gsymbol/s. Figure 5 shows the BER of a demodulated 10 Gsymbol/s, 1024 QAM signal after a 160 km transmission obtained for various powers launched into each fiber span. On the basis of these results, the launch power was optimized to 1 dBm. The BER degraded at launch powers below 1 dBm due to the OSNR degradation, while it degraded due to fiber nonlinearity at launch powers above 1 dBm. The optical spectra of the 10 Gsymbol/s, 1024 QAM data signal before and after transmission are shown in Fig. 6, where the spectral resolution was 0.1 nm. Here, the OSNR degradation was 6 dB during the 160 km transmission.
We evaluated the relationship between the demodulation performance of a 1024 QAM signal and the number of pilot symbols inserted into 4096 symbols of data. Figure 7 shows the BER characteristics of a 10 Gsymbol/s-1024 QAM signal at the maximum OSNR after a 160 km transmission as a function of the pilot symbol number. When this number was below 32 symbols, the BER characteristics were degraded over the FEC threshold with a 25.5% overhead. This is because the ambiguity of the waveform distortion compensation increases with a decrease in the pilot symbol number. On the other hand, the BERs obtained with more than 32 pilot symbols converged to below the FEC threshold level with a 25.5% overhead. Here, the BER saturation may be attributed to some from of fiber nonlinearity such as cross-phase modulation. A lower number of pilot symbols give a better SE. From these results, we set the pilot symbol number at 32.
We compared the convergence characteristics of the FIR filter based on the DD-LMS algorithm with those of our pilot-assisted FIR filter. Figure 8 shows the BER characteristics of a 10 Gsymbol/s-1024 QAM signal after a 160 km transmission as a function of data frame length used to calculate the updated FIR tap coefficient. The closed squares show results for the DD-LMS-based equalizer and the red circles are those of our pilot-assisted equalizer with 32 pilot symbols. With the conventional DD-LMS FIR, the BER characteristics did not converge, because there were some decision errors in the updating process of the FIR tap coefficient. On the other hand, the BERs obtained with the pilot-assisted equalizer converged to below a 25.5% FEC threshold with a data frame length of over 30. These results indicate that there is a great advantage in the present equalization scheme regarding the fast and lower BER convergence characteristics of an FIR filter in ultra-multi-level QAM transmission.
Figure 9(a) and 9(b), respectively, show the constellations of 10 Gsymbol/s, 1024 QAM signals after transmission at an OSNR of 38 dB obtained by employing a conventional equalizer without pilot symbols and the present pilot-assisted equalizer. When we used the conventional equalizer, we could not demodulate the 10 Gsymbol/s, 1024 QAM signal due to insufficient waveform distortion compensation. On the other hand, the pilot-assisted equalizer enabled us to demodulate a 1024 QAM signal with a BER of 3.3 x 10−2.
The BER characteristics of a 200 Gbit/s, 10 Gsymbol/s 1024 QAM transmission as a function of OSNR are shown in Fig. 10. The blue line shows the results obtained under a back-to-back condition. The red line shows the transmission results. An insufficient SNR for the homodyne detection caused an error floor under the back-to-back condition even with a high OSNR. We attributed this to the insufficient SNR of the electrical baseband signal generated from the AWG. After the transmission, we obtained a BER of 3.8 x 10−2 at an OSNR of 36 dB. The OSNR penalty at that BER, which represents a 25.5% FEC threshold for error-free transmission, was 5.5 dB for both polarizations. The BERs for both sets of polarization data were below the FEC threshold at an OSNR of 38 dB.
Figure 11(a) and 11(b) show constellations of X-pol data obtained under a back-to-back condition and after a 160 km transmission, respectively. The error vector magnitudes (EVMs) were 1.32% and 1.76%, respectively. Here, 200 Gbit/s data were transmitted within an optical bandwidth of 12.56 GHz including a PT, which corresponds to a potential SE of 12.6 bit/s/Hz by taking account the 25.5% FEC overhead and the 19 header symbols and 32 pilot symbols.
We successfully demonstrated a single-channel 200 Gbit/s, 1024 QAM injection-locked coherent transmission. By developing a new pilot-assisted adaptive equalization scheme, the compensation performance for waveform distortions was greatly improved compared with that of a conventional equalizer. Using this scheme, we succeeded in transmitting pol-mux, 1024 QAM data at a symbol rate as high as 10 Gsymbol/s over 160 km with a potential SE of 12.6 bit/s/Hz. These results provide useful knowledge for realizing ultrahigh-speed optical transmission with ultra-high spectral efficiency.
Research and Development Project toward 5G Mobile Communication Systems of the Ministry of Internal Affairs and Communications, Japan.
References and links
1. S. Adhikari, S. Sygletos, A. D. Ellis, B. Inan, S. L. Jansen, and W. Rosenkranz, “Enhanced self-coherent OFDM by the use of injection locked laser,” in Proceedings of Optical Fiber Communication Conference (Optical Society of America, 2012), paper JW2A.
2. Z. Liu, J. Y. Kim, D. S. Wu, D. J. Richardson, and R. Slavík, “Homodyne OFDM with optical injection locking for carrier recovery,” J. Lightwave Technol. 33(1), 34–41 (2015). [CrossRef]
3. A. C. Bordonalli, C. Walton, and A. J. Seeds, “High-performance phase locking of wide linewidth semiconductor lasers by combined use of optical injection locking and optical phase-lock loop,” J. Lightwave Technol. 17(2), 328–342 (1999). [CrossRef]
4. Y. Wang, K. Kasai, M. Yoshida, and M. Nakazawa, “Single-carrier 216 Gbit/s, 12 Gsymbol/s 512 QAM coherent transmission over 160 km with injection-locked homodyne detection,” in Proceedings of Optical Fiber Communication Conference (Optical Society of America, 2017), paper Tu2E.1.
5. Y. Wang, K. Kasai, M. Yoshida, and M. Nakazawa, “120 Gbit/s injection-locked homodyne coherent transmission of polarization-multiplexed 64 QAM signals over 150 km,” Opt. Express 22(25), 31310–31316 (2014). [CrossRef] [PubMed]
6. K. Kasai, M. Yoshida, and M. Nakazawa, “552 Gbit/s, 46 Gbaud, 64 QAM coherent transmission over 160 km with simple LD-based injection-locked homodyne detection,” in Proceedings of European Conference on Optical Communication (Optical Society of America, 2016), paper W.4.P1.SC5.51.
7. Y. Wang, K. Kasai, M. Yoshida, and M. Nakazawa, “320 Gbit/s, 20 Gsymbol/s 256 QAM coherent transmission over 160 km by using injection-locked local oscillator,” Opt. Express 24(19), 22088–22096 (2016). [CrossRef] [PubMed]
8. S. U. H. Qureshi, “Adaptive equalization,” Proc. IEEE 73(9), 1349–1387 (1985). [CrossRef]
9. M. P. Yankov, E. P. da Silva, F. Da Ros, and D. Zibar, “Experimental analysis of pilot-based equalization for probabilistically shaped WDM systems with 256QAM/1024QAM,” in Proceedings of Optical Fiber Communication Conference (Optical Society of America, 2017), paper W2A.48.
11. K. Kasai, M. Nakazawa, Y. Tomomatsu, and T. Endo, “1.5μm, mode-hop-free full C-band wavelength tunable laser diode with a linewidth of 8 kHz and a RIN of -130 dB/Hz and its extension to the L-band,” Opt. Express 25(18), 22113–22124 (2017). [CrossRef] [PubMed]
12. H. Ishii, K. Kasaya, and H. Oohashi, “Spectral linewidth reduction in widely wavelength tunable DFB laser array,” IEEE J. Sel. Top. Quantum Electron. 15(3), 514–520 (2009). [CrossRef]
13. T. Kan, K. Kasai, M. Yoshida, and M. Nakazawa, “42.3 Tbit/s, 18 Gbaud 64 QAM WDM coherent transmission over 160 km in the C-band using an injection-locked homodyne receiver with a spectral efficiency of 9 bit/s/Hz,” Opt. Express 25(19), 22726–22737 (2017). [CrossRef] [PubMed]
15. K. Ishihara, T. Kobayashi, R. Kudo, Y. Takatori, A. Sano, E. Yamada, H. Masuda, and Y. Miyamoto, “Coherent optical transmission with frequency-domain equalization,” in Proceedings of the European Conference on Optical Communications (Optical Society of America, 2008), Paper We2E3. [CrossRef]
16. C. Paré, A. Villeneuve, P. A. Bélanger, and N. J. Doran, “Compensating for dispersion and the nonlinear Kerr effect without phase conjugation,” Opt. Lett. 21(7), 459–461 (1996). [CrossRef] [PubMed]
17. K. Kikuchi, “Digital coherent optical communication systems: fundamentals and future prospects,” IEICE Electron. Express 8(20), 1642–1662 (2011). [CrossRef]
18. K. Sugihara, Y. Miyata, T. Sugihara, K. Kubo, H. Yoshida, W. Matsumoto, and T. Mizuochi, “A spatially coupled type LDPC code with an NCG of 12dB for optical transmission beyond 100 Gb/s,” in Proceedings of Optical Fiber Communication Conference (Optical Society of America, 2013), paper OM2B.4.