A novel and blind optical modulation format identification (MFI) scheme assisted by signal intensity fluctuation features is proposed for autonomous digital coherent receivers of next generation optical network. The proposed MFI scheme needn’t to pre-know OSNR value of incoming signal, even though it is well known that the intensity dependent features of the incoming signal changes as the change of OSNR performance. Here, the proposed scheme firstly utilizes two kinds of signal intensity fluctuation features, Godard’s criterion error and intensity noise variance, to construct a two-dimensional (2D) plane where three different regions which consist of QPSK region, 8QAM region, mixed 16/32/64QAM region can be found. Thus, we can firstly identify QPSK and 8QAM from the 2D plane, and then partition Godard’s criterion error method is introduced to further identify 16QAM, 32QAM and 64QAM in our proposed scheme. The performance of the proposed scheme is firstly verified by a series of numerical simulations in 28GBaud PDM-QPSK/-8QAM/-16QAM/-32QAM/-64QAM coherent optical communication systems. The results show that the lowest required OSNR values to achieve 100% recognition rate for all modulation format signals are even lower than or close to their corresponding theoretical 20% FEC limits (BER = 2.4 × 10−2). Finally, the feasibility is further demonstrated via a series of proof-of-concept experiments among 28GBaud PDM-QPSK/-8QAM/-16QAM, and 21.5GBaud PDM-32QAM systems under back-to-back and long-haul fiber transmission links (from 320 km to 2000km). Experiment results show that the proposed scheme is robust to both linear and nonlinear noise.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
The transmitter of next generation optical network is expected to be able to dynamically change signal transmission parameters (such as symbol rate, forward error correction (FEC) overhead, modulation format, and signal bandwidth) according to different link conditions and quality of service requirements [1–3], so that these parameters may be uncertain in the receiver of optical network nodes. Typically, the supervisory control layer of the optical networks would deliver these parameters to the receiver of optical network nodes. However, when the supervisory channel is disrupted or deemed too slow to allow fast and flexible provisioning, the control layer would not obtain these parameters information and ultimately the receiver can’t get these information. Meanwhile, it is practically not flexible to introduce or rely on additional cross-layer communication in next generation optical network. Thus, it is critical for the receiver of optical network nodes to autonomously and blindly monitor these transmission parameters without the supervisory control layer assistance. Over the past few years, a plethora of blind monitoring and identifying techniques have been proposed for obtaining these transmission parameters [4–7] in the receiver. In this paper, we only focus on the modulation format identification (MFI) technique.
Currently, coherent detection technique combined with digital signal processing (DSP) is considered as a prospective solution in next generation optical network. Not only can DSP technique be used to remove the need for dynamic polarization control, but also it may be applied to compensate and mitigate various transmission impairments (such as chromatic dispersion (CD), self-phase modulation (SPM) effect, cross-phase modulation (XPM) effect, polarization mode dispersion (PMD), polarization dependent loss (PDL), polarization mixing, frequency offset, laser phase noise, etc.). In these DSP techniques, some modulation format dependent algorithms (multi-modulus algorithm (MMA), frequency offset compensation, carrier phase recovery, decoder) require modulation format information (obtained through MFI technique) to choose their corresponding optimum algorithms for achieving optimized demodulation (different modulation formats contain different intensity and phase features, so their compensation or recovery algorithms may be different). On the other hand, MFI technique will be also indispensable for the application of existing optical performance monitoring (OPM) devices deployed at the intermediate network nodes . The actual modulation format information of the incoming signal (obtained through MFI technique) would enable the OPM devices deployed at the intermediate network nodes apply a monitoring technique  suitable for that specific modulation format. Therefore, it would be very important for the next generation intelligent receiver to identify the modulation format information autonomously and blindly.
In recent years, various MFI schemes have been proposed in optical communication systems. Among these schemes, aided identification schemes require some additional operations to encode modulation format information into the amplitude of the RF-pilot  or the FO distribution  at the transmitter. Other scheme, nonlinear power transformation , may require longer FFT to cope with reduction of PAPR caused by imperfect symbol partitioning for higher-order QAM signals. In addition, some signal amplitude/power feature schemes [13–18] may need to pre-know OSNR performance of the transmission system or require a large number of symbols to obtain optimized amplitude features. Stokes-space based schemes [19–23] are insensitive to polarization mixing, frequency offset, and phase noise, but there would be some difficulties in identifying higher-order modulation formats. Therefore, it will be very necessary to propose a novel and blind MFI algorithm which can be able to identify higher-order modulation formats without prior prediction of OSNR.
In this paper, a blind and novel optical MFI scheme assisted by signal intensity fluctuation features extracted after constant modulus algorithm (CMA) is proposed for autonomous digital coherent receivers of next generation optical network. This paper expands our previous conference paper  to identify higher-order modulation format and achieve higher identification accuracy, and more detailed analysis about the influence of fiber nonlinearity effect are also presented in this paper. Here, the proposed scheme firstly utilizes two kinds of signal intensity fluctuation features, Godard’s criterion error and intensity noise variance, to construct a two-dimensional (2D) plane where three different intensity fluctuation regions which consist of QPSK region, 8QAM region, mixed 16/32/64QAM region can be found. Thus, we can firstly identify QPSK and 8QAM from the 2D plane, while the remaining high-order modulation formats would be very difficult to be separated in the 2D plane. It is well known that high-order modulation format signals have multiple intensity levels so that their features are very similar in a certain OSNR range. In order to mitigate this case, the proposed scheme introduces partition Godard’s criterion error operation to classify the remaining high-order modulation formats. The feasibility is firstly demonstrated via a series of numerical simulations in 28GBaud PDM-QPSK/-8QAM/-16QAM/-32QAM/-64 QAM coherent optical communication systems. The results show that the lowest required OSNR values to achieve 100% recognition rate for all modulation formats signals are even lower than or close to their corresponding theoretical 20% FEC limits (BER = 2.4 × 10−2). Subsequently, the proposed scheme is further demonstrated via a series of proof-of-concept experiments among 28GBaud PDM-QPSK/-8QAM/-16QAM, and 21.5GBaud PDM-32QAM systems under back-to-back and long-haul fiber transmission links (from 320km to 2000km).
2. Operation principle
At the receiver of next generation optical network, DSP techniques are introduced to compensate or mitigate link transmission impairments for achieving optimized demodulation of the transmission signal. The DSP architecture in coherent optical receiver is shown in the Fig. 1. Here, the received signals are firstly converted from analog signals to digital signals by the analog-to-digital converter (ADC). Then, we utilize CD compensation and timing recovery algorithm to compensate link CD and timing jitter, respectively. Subsequently, a 21-tap T/2 spaced butterfly equalizer based on CMA algorithm is applied to compensate the residual CD and polarization mode dispersion (PMD). In addition, it can fully equalize and demultiplex mPSK signals which include only one level. However, when the modulation format type is mQAM (m > 4), CMA algorithm can be used to perform both rough polarization demultiplexing and equalization, so it can be considered as modulation format independent algorithm. In our proposed scheme, the intensity fluctuation features extracted after CMA algorithm are applied to identify different modulation format types. Then, the modulation format information obtained from our proposed scheme can help the subsequent modulation format dependent algorithms (i.e. MMA, frequency offset estimation, carrier phase recovery) to select their corresponding optimum algorithms. For example, if the modulation format type of the incoming signal is QPSK, the polarization de-multiplexing algorithm, frequency offset compensation, and carrier phase recovery would choose CMA algorithm , 4-th frequency estimation algorithm , Viterbi-Viterbi phase estimation (VVPE) algorithm  to recovery the incoming signals. Or if the modulation format type is 16QAM, the “residual” polarization demultiplexing needs to select the corresponding MMA  to achieve finally demultiplexing operation, while the CMA algorithm just is used to carry pre-convergence or pre-equalization in here. Frequency offset compensation and carrier phase recovery would choose QAM frequency estimation algorithm  and feedforward carrier recovery , respectively.
2.1. Constructing intensity fluctuation 2D plane
The digital signals D(n) after CMA equalization are applied to obtain the signal intensity fluctuation features in our proposed scheme. As shown in Fig. 2, the intensity information |D(n)|2 of different modulation format signals represent unique features, because different modulation formats have different amplitude levels (i.e. QPSK with only one itensity level, 8QAM with two itensity levels, 16QAM with three itensity levels, 32QAM with five itensity levels, 64QAM with nine itensity levels). In addition, the intensity fluctuation features of different modulation formats would also be different, so we consider it as identification metric in our proposed scheme. Here, one of intensity fluctuation features known as the Godard’s criterion error [24,31–32] is defined as:
Since high-order modulation formats (i.e. 16/32/64QAM signals) have multiple amplitude levels, their intensity fluctuation features would be very close and be mixed in one region of 2D plane, so these high-order modulation formats cannot be directly distinguished from each other by using the 2D plane. Subsequently, we further introduce partition Godard’s criterion error to classify the remaining high-order modulation formats.
2.2. Partition Godard’s criterion error
As shown in Fig. 4, there are two kinds of partition operations for high-order modulation formats. In here, the first partition operation is to select the normalized intensity values range from 0.7 to 1.4 to calculate the normalized Godard’s criterion error by using Eq. (1). The particular intensity range is the middle intensity range of 16QAM signal which has three intensity ranges as shown in Fig. 4. In the particular range, 32QAM and 64QAM signals have different intensity distributions compared with 16QAM so that their normalized Godard’s criterion errors would also be different from that of 16QAM. Thus, we can use the case to separate 16QAM from the mixed high-order modulation formats region. The corresponding normalized Godard’s criterion errors for the partition operation 1 are calculated as depicted in Fig. 5(a). Here, the Godard’s criterion error of 16QAM is obviously far from that of 32QAM and 64QAM by using the partition operation 1. Meanwhile, a threshold value Th1 can be found to perfectly separate 16QAM from 32QAM and 64QAM even under relatively low OSNR condition. Subsequently, partition operation 2 is introduced to further distinguish 32QAM and 64QAM in this paper. The partition operation uses the greater than 0.85 normalized intensity interval to calculate the normalized Godard’s criterion errors as shown in in Fig. 4. Note that the intensity interval of the partition operation 2 is dependent on the intensity levels of 32QAM. In the specified intensity interval, the intensity distribution of 32QAM is different from that of 64QAM so that we can use the normalized Godard’s criterion errors of these two modulation formats to classify each other in the intensity interval. Meanwhile, an optimal threshold value Th2 can also be found to distinguish 32QAM and 64QAM under relatively low OSNR conditions as depicted in Fig. 5(b).
The flow chart of the proposed MFI scheme is depicted in Fig. 6. The identification process consists of three steps in here. The first step is to construct a 2D intensity fluctuation plane by Godard’s criterion error and intensity noise variance. QPSK and 8QAM signals can be separated in the 2D plane by using SVM technique. Then, the second step is to apply the partition operation 1 to calculate the Godard’s criterion errors for distinguishing 16QAM from mixed high-order modulation format region. Subsequently, the third step is to introduce the partition operation 2 to calculate the Godard’s criterion errors of different modulation formats for further identifying 32QAM and 64QAM. In this paper, the basic idea is to utilize the intensity fluctuation features of different modulation formats to identify each other.
As shown in Fig. 6, there are two-stage comparison or decision operations in the proposed scheme. In here, the comparison or decision operation just judges whether the Godard’s criterion error of the incoming signal is greater than or less than the corresponding threshold value (Th1 or Th2), so the time complexity of each comparison operation is O(1). Therefore, the time delay introduced by the two-stage comparison or decision operations can be almost ignored in the whole DSP module, and these comparison or decision operations would not limit the transmission rate of the link. For the SVM with RBF kernel, the time complexity of training process is about O(m3), and that of testing process is about O(m*d), where m is the number of support vectors which depend on the size of the training samples and d is the dimensionality of the feature vectors . Therefore, the total time complexity of SVM is about O(m3). The SVM can be well used to identify different features in 2D intensity fluctuation plane of our proposed scheme, but it may not be the best or only one technique to solve classification problem. Therefore, we would further explore the low time complexity classification technique in the future.
3. The numerical simulation results
The proposed scheme is firstly demonstrated via a series of numerical simulations in the optical communication commercial software VPI Tranmission Makers. In the transmitter of the coherent optical communication system, 28GBaud PDM-QPSK/-8QAM/-16QAM/-32 QAM/-64QAM signals are generated, and then passed through an additive Gaussian white noise channel under variable OSNR conditions. Subsequently, these incoming modulation format signals are detected by a coherent receiver, and compensated or monitored through off-line digital signal processing module. In all simulation systems, each OSNR (0.1nm ASE noise bandwidth) value is implemented with 100 independent simulations, and the OSNR interval is 1dB. Meanwhile, the OSNR ranges of QPSK, 8QAM, 16QAM, 32QAM, 64QAM are 7∼30dB, 9∼30dB, 14∼32dB, 15∼32dB, 16∼35dB, respectively. Our proposed MFI scheme is applied in off-line digital signal processing module to monitor modulation format type of the incoming signals.
The first step of the proposed scheme is to apply SVM technique to classify QPSK and 8QAM signals from all incoming signals in 2D intensity fluctuation plane. The SVM technique uses a kernel function to map the data to a higher dimensional space where the separation of the data is easier. It is well known that kernel function could be considered as a measure of similarity between input data points. Thus, the choice of the kernel function would have an influence on the classification performance of SVM technique. As shown in Fig. 7(a), we compare the classification performance of the linear and radial bias function (RBF) SVM kernel functions under different training sample percent of total samples. The total 10300 samples which are composed of the samples of all modulation formats under different OSNR conditions are applied in the SVM techniques for training and testing. As shown in Fig. 7(a), with the number of training samples increasing (note that the percent of total samples in Fig. 7(a) represents the number of training samples), the performance of both two kernel functions is greatly improved. Compared with the linear kernel function, we consider that RBF-SVM kernel function is more suitable for the proposed MFI scheme because the identification performance can be close to 100% correct identification rate in small percent of total samples (the training samples are relatively less). Under weighing identification performance (the correct identification rates) and time complexity (the number of training samples), the proposed scheme chooses 50% and 50% of the total samples to construct training and testing sample subsets for all simulation systems. Subsequently, the other important factor, the number of symbols, is considered in here. It is well known that the signal intensity fluctuation features may become clearer when a large numbers of symbols are applied to calculate these features. Here, the total correct identification rates with different numbers of symbols for each modulation format sample are calculated in Fig. 7(b). It can be seen that 100% correct identification rate can be achieved when the number of symbols for each modulation format sample is 6000. Therefore, the proposed scheme would select 6000 symbols to calculate intensity fluctuation features for each sample in subsequent identification process.
In order to verify the feasibility, the proposed scheme with the optimal number of training samples and optimal symbol number is performed in back-to-back transmission link as shown in Fig. 8. The lowest required OSNR values to achieve 100% recognition rate for 28GBaud PDM-QPSK/-8QAM/-16QAM/-32QAM/-64QAM signals are 7dB, 9dB, 16dB, 17dB, 16dB, respectively. As shown in Fig. 8, various colors dot lines are applied to mark their theoretical OSNR values of 20% FEC limits. In here, the corresponding lowest required OSNR values are much lower than their theoretical 20% FEC limits for all modulation formats. Thus, the effective MFI scheme with good identification performance would have the potential to be used in the next generation digital coherent receiver.
4. Experiment setup and results
A series of proof-of-concept experiments with flexible modulation format transmitters and the autonomous coherent receivers enabled the proposed blind MFI scheme are performed as shown in Fig. 9. In flexible format transmitter, pseudo-random bit sequence (PRBS) with a word length of 215−1 is generated and mapped into mQAM modulation format. A square root raised cosine (SRRC) filter with a roll-off factor of 1.0 is applied to shape the mQAM signal. Meanwhile, pre-distortion operation is performed to compensate the frequency roll-off of digital-to-analog converters (DACs). The light with ∼1549.3nm wavelength and ∼100 kHz linewidth from an external cavity laser is modulated by an integrated IQ modulator driven by two ports DACs operating at 64GSa/s with 25GHz analog bandwidth to generate 28GBaud QPSK, 8QAM, 16QAM, and 21.5GBaud 32QAM optical signals. Then, polarization multiplexing operation is emulated by employing a coupler, two polarization controllers (PCs), an optical delay-line (ODL), two variable optical attenuators (VOAs), and a polarization beam splitter (PBS) for obtaining PDM-QPSK/-8QAM/-16QAM/-32QAM optical signals. The signals are transmitted in the back-to-back link and various long-haul fiber transmission links. The fiber transmission link is composed of multi-spans single mode fiber (SMF) whose dispersion parameter, attenuation coefficient, and nonlinear coefficient are D = 16.9ps/nm/km, α = 0.2dB/km, and γ = 1.27km−1•W−1, respectively. Fiber loss of each span is compensated using an erbium doped fiber amplifier (EDFA) with noise figure of ∼5dB. An optical spectrum analyzer (OSA) is applied to measure the OSNR performance of system. In the receiver, the received signals are detected by an integrated coherent receiver, and then sampled by a real-time digital oscilloscope with 80GSamples/s and 33GHz electrical bandwidth. Finally, the proposed MFI method is achieved in off-line digital signal processing module. In all proof-of-concept experiments, each OSNR value is implemented with 100 independent implementations, and the OSNR interval is 1dB. Meanwhile, the RBF-SVM kernel function is used in the identification process for all proof-of-concept experiments. The percent of training samples and symbol number are also the same with the simulation system.
Under back-to-back link, the performance of the proposed scheme is verified with different OSNR conditions as shown in Fig. 10. Meanwhile, the dot lines are applied to mark their corresponding theoretical OSNR values of 20% FEC limits for different modulation formats. The experimental results show that the lowest required OSNR values to achieve 100% recognition rate for PDM-QPSK/-8QAM/-16QAM/-32QAM is lower than or close to the their theoretical 20% FEC limits. Compared with numerical simulation results as shown in Fig. 8, PDM-QPSK and PDM-8QAM can still be identified with 100% correct identification rates in the whole OSNR ranges. Note that the lowest required OSNR value of PDM-16QAM signal is slightly larger about 1dB than the theoretical 20% FEC limits. The case may be caused by the undesirable signal generation for high-order modulation format in proof-of-concept experiments. We believe that the case can be mitigated through optimizing high-order modulation format signal generation.
Subsequently, for demonstrating the performance of the proposed scheme under nonlinear impairments, a series of long-haul transmission experiments are carried out under different launch powers. The launch powers of PDM-QPSK with 2000km, PDM-8QAM with 2000km, PDM-16QAM with 1040 km, and PDM-32QAM with 320km are varied in the ranges of −4 ∼10dBm, −3∼6dBm, −3∼8dBm, and −6∼11dBm, respectively. Here, the error vector magnitude (EVM) is introduced to evaluate the performance of different modulation formats. As depicted in Fig. 11, when the launch power is greater than 2 dBm, all modulation formats transmission systems are found to be in the nonlinear region where the system performance is dominated by fiber nonlinearity. When the launch power is less than 2 dBm, the penalty is dominated by noise. Next, the correct identification rates of various modulation formats under different launch powers are calculated as shown in Fig. 11. The all modulation formats can have almost 100% identification rates even at high launch powers (strong fiber nonlinearity impairments). Here, the identification rates of PDM-QPSK/-8QAM/-16QAM/-32QAM remain 100% even at 10dBm, 5dBm, 6dBm and 11dBm launch powers, respectively. Note that the decline of identification rate for PDM-8QAM and PDM-16QAM are because high-order modulation formats have smaller Euclidean distance than QPSK so that these modulation formats may be more susceptible to link impairments after long-haul fiber transmission. Meanwhile, the 100% identification rate of PDM-32QAM is obtained even at 11 dBm launch power due to the relatively short fiber transmission link. Thus, the proposed MFI scheme has a high tolerance for the fiber nonlinearity impairments, and I believe that it has the potential to be applied in long-haul fiber transmission link.
5. Discussion and conclusion
In this paper, we have proposed a blind and robust modulation format identification method assisted by signal intensity fluctuation features for autonomous digital coherent receivers of next-generation optical network. The proposed MFI scheme needn’t to pre-know the OSNR value of the incoming signal despite the intensity dependent features would change as the change of OSNR. Meanwhile, since CMA equalization applied in the proposed scheme is able to compensate the residual CD and PMD effects, the influence of these effects are not considered in this paper. The feasibility is verified via a series of numerical simulations and proof-of-concept experiments among PDM-QPSK/-8QAM/-16QAM/-32QAM/-64QAM systems under back-to-back and long-haul fiber transmission links (from 320km to 2000km). The results show that the lowest required OSNR values of the proposed scheme for all modulation format signals can achieve 100% MFI accuracies under less than or close to their corresponding theoretical FEC limit. Meanwhile, the results show that the proposed scheme is robust to both linear and nonlinear noise. I believe that the proposed scheme has the potential to be applied in the next-generation optical communication system.
National Key Research and Development Program of China (2019YFB1803500); National Natural Science Foundation of China (NSFC) (61860206006); Key Project of Sichuan Province of China (2019ZDZX0007).
1. A. Napoli, M. Bohn, D. Rafique, A. Stavdas, N. Sambo, L. Potì, M. Nölle, J. Fischer, E. Riccardi, A. Pagano, A. Giglio, M. Moreolo, J. Fabrega, E. H. Salas, G. Zervas, D. Simeonidou, P. Layec, A. Errico, T. Rahman, and J. Giménez, “Next generation elastic optical networks: the vision of the European research project IDEALIST,” IEEE Commun. Mag. 53(2), 152–162 (2015). [CrossRef]
2. B. Chatterjee, N. Sarma, and E. Oki, “Routing and spectrum allocation in elastic optical networks: A tutorial,” IEEE Commun. Surveys Tuts. 17(3), 1776–1800 (2015). [CrossRef]
3. Y. Yin, L. Liu, R. Proietti, and S. Yoo, “Software defined elastic optical networks for cloud computing,” IEEE Netw. 31(1), 4–10 (2017). [CrossRef]
4. Z. Dong, F. N. Khan, Q. Sui, K. Zhong, C. Lu, and A. Lau, “Optical performance monitoring: A review of current and future technologies,” IEEE/OSA J. Lightwave Technol. 34(2), 525–543 (2016). [CrossRef]
5. M. Ionescu, M. Sato, and B. Thomsen, “Cyclostationarity-based joint monitoring of symbol-rate, frequency offset, CD and OSNR for Nyquist WDM superchannels,” Opt. Express 23(20), 25762–25772 (2015). [CrossRef]
6. H. Xu, Y. Feng, J. Yuan, X. Zhang, and C. Bai, “Intelligent bandwidth-estimation technique for orthogonal frequency division multiplexing-based elastic optical networking,” J. Opt. Commun. Netw. 8(12), 938–946 (2016). [CrossRef]
7. L. Guesmi, A. M. Ragheb, H. Fathallah, and M. Menif, “Experimental demonstration of simultaneous modulation format/symbol rate identification and optical performance monitoring for coherent optical systems,” IEEE/OSA J. Lightwave Technol. 36(11), 2230–2239 (2018). [CrossRef]
8. F. Khan, Y. Zhou, Q. Sui, and A. Lau, “Non-data-aided joint bit-rate and modulation format identification for next generation heterogeneous optical networks,” Opt. Fiber Technol. 20(2), 68–74 (2014). [CrossRef]
9. D. J. Ives, B. Thomsen, R. Maher, and S. Savory, “Estimating OSNR of equalised QPSK signals,” Opt. Express 19(26), B661–B666 (2011). [CrossRef]
10. M. Xiang, Q. Zhuge, M. Qiu, X. Zhou, M. Tang, D. Liu, S. Fu, and D. Plant, “RF-pilot aided modulation format identification for hitless coherent transceiver,” Opt. Express 25(1), 463–471 (2017). [CrossRef]
11. S. Fu, Z. Xu, J. Lu, H. Jiang, Q. Wu, Z. Hu, M. Tang, D. Liu, and C. Chan, “Modulation format identification enabled by the digital frequency-offset loading technique for hitless coherent transceiver,” Opt. Express 26(6), 7288–7296 (2018). [CrossRef]
12. G. Liu, R. Proietti, K. Zhang, H. Lu, and S. Yoo, “Blind modulation format identification using nonlinear power transformation,” Opt. Express 25(25), 30895–30904 (2017). [CrossRef]
13. F. Khan, K. Zhong, W. A. Arashi, C. Yu, C. Lu, and A. Lau, “Modulation format identification in coherent receivers using deep machine learning,” IEEE Photonics Technol. Lett. 28(17), 1886–1889 (2016). [CrossRef]
14. S. Bilal, G. Bosco, Z. Dong, A. Lau, and C. Lu, “Blind modulation format identification for digital coherent receivers,” Opt. Express 23(20), 26769–26778 (2015). [CrossRef]
15. X. Lin, Y. A. Eldemerdash, O. A. Dobre, S. Zhang, and C. Li, “Modulation classification using received signal’s amplitude distribution for coherent receivers,” IEEE Photonics Technol. Lett. 29(21), 1872–1875 (2017). [CrossRef]
16. J. Liu, Z. Dong, K. Zhong, C. Guo, A. Lau, C. Lu, and Y. Lu, “Signal power distribution based modulation format identification for coherent optical receivers,” Opt. Fiber Technol. 36, 75–81 (2017). [CrossRef]
17. Y. Zhao, C. Shi, D. Wang, X. Chen, L. Wang, T. Yang, and J. Du, “Low-complexity and nonlinearity-tolerant modulation format identification using Random Forest,” IEEE Photonics Technol. Lett. 31(11), 853–856 (2019). [CrossRef]
18. Z. Zhao, A. Yang, and P. Guo, “A modulation format identification method based on information entropy analysis of received optical communication signal,” IEEE Access 7, 41492–41497 (2019). [CrossRef]
19. R. Boada, R. Borkowski, and I. Monroy, “Clustering algorithms for Stokes space modulation format recognition,” Opt. Express 23(12), 15521–15531 (2015). [CrossRef]
20. L. Jiang, L. Yan, A. Yi, Y. Pan, T. Bo, M. Hao, W. Pan, and B. Luo, “Blind density-peak-based modulation format identification for elastic optical networks,” IEEE/OSA J. Lightwave Technol. 36(14), 2850–2858 (2018). [CrossRef]
21. W. Zhang, D. Zhu, Z. He, N. Zhang, X. Zhang, H. Zhang, and Y. Li, “Identifying modulation formats through 2D Stokes planes with deep neural networks,” Opt. Express 26(18), 23507–23517 (2018). [CrossRef]
22. A. Yi, L. Yan, H. Liu, L. Jiang, Y. Pan, B. Luo, and W. Pan, “Modulation format identification and OSNR monitoring using density distributions in Stokes axes for digital coherent receivers,” Opt. Express 27(4), 4471–4479 (2019). [CrossRef]
23. T. Bo, J. Tang, and C. Chan, “Modulation format recognition for optical signals using connected component analysis,” IEEE Photonics Technol. Lett. 29(1), 11–14 (2017). [CrossRef]
24. L. Jiang, L. Yan, A. Yi, Y. Pan, T. Bo, M. Hao, W. Pan, and B. Luo, “Robust and blind modulation format identification for elastic optical networks,” inProc. ECOC’2018, paper We2.20 (2018).
25. K. Kikuchi, “Polarization-demultiplexing algorithm in the digital coherent receiver,” in2008 Technical Digest of IEEE/LEOS Summer Topical Meeting (IEEE/LEOS, 2008), paper MC2.2 (2008).
26. A. Leven, N. Kaneda, U. Koc, and Y. Chen, “Frequency Estimation in Intradyne Reception,” IEEE Photonics Technol. Lett. 19(6), 366–368 (2007). [CrossRef]
27. A. J. Viterbi and A. M. Viterbi, “Nonlinear estimation of PSK-modulated carrier phase with application to burst digital transmission,” IEEE Trans. Inf. Theory 29(4), 543–551 (1983). [CrossRef]
28. 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 Proc. OFC’2009, Paper OWG3 (2009).
29. M. Selmi, Y. Jaouën, and P. Cibalt, “Accurate digital frequency offset estimator for coherent polmux QAM transmission systems,” inProc. ECOC’2009, paper P3.08 (2009).
30. T. Pfau, S. Hoffmann, and R. Noe, “Hardware-efficient coherent digital receiver concept with feed forward carrier recovery for M-QAM constellations,” IEEE/OSA J. Lightwave Technol. 27(8), 989–999 (2009). [CrossRef]
31. D. Godard, “Self-recovery equalization and carrier tracking in two-dimensional data communication systems,” IRE Trans. Commun. Syst. 28(11), 1867–1875 (1980). [CrossRef]
32. L. Jiang, L. Yan, A. Yi, Y. Pan, M. Hao, W. Pan, B. Luo, and Y. Jaouën, “Chromatic dispersion nonlinear parameter and modulation format monitoring based on Godard's error for coherent optical transmission systems,” IEEE Photonics J. 10(1), 1–12 (2018). [CrossRef]
33. Y. Bazi and F. Melgani, “Toward an optimal SVM classification system for hyperspectral remote sensing images,” IEEE T. Geosci. Remote 44(11), 3374–3385 (2006). [CrossRef]
34. Y. Li, N. Hua, Y. Yu, Q. Luo, and X. Zheng, “Light source and trail recognition via optical spectrum feature analysis for optical network security,” IEEE Commun. Lett. 22(5), 982–985 (2018). [CrossRef]