Modulation format identification aided hitless flexible coherent transceiver

We propose a hitless flexible coherent transceiver enabled by a novel modulation format identification (MFI) scheme for dynamic agile optical networks. The modulation format transparent digital signal processing (DSP) is realized by a block-wise decisiondirected least-mean-square (DD-LMS) equalizer for channel tracking, and a pilot symbol aided superscalar phase locked loop (PLL) for carrier phase estimation (CPE). For the MFI, the modulation format information is encoded onto the pilot symbols initially used for CPE. Therefore, the proposed MFI method does not require extra overhead. Moreover, it can identify arbitrary modulation formats including multi-dimensional formats, and it enables tracking of the format change for short data blocks. The performance of the proposed hitless flexible coherent transceiver is successfully evaluated with five modulation formats including QPSK, 16QAM, 64QAM, Hybrid QPSK/8QAM and set-partitioning (SP)-512-QAM. We show that the proposed MFI method induces a negligible performance penalty. Moreover, we experimentally demonstrate that such a hitless transceiver can adapt to fast block-by-block modulation format switching. Finally, the performance improvement of the proposed MFI method is experimentally verified with respect to other commonly used MFI methods. ©2016 Optical Society of America OCIS codes: (060.2330) Fiber optics communications; (060.1660) Coherent communications; (060.4080) Modulation. References and links 1. Cisco Visual Networking Index, Forecast and Methodology, 2013–2018. [Online] Available at (http://www. cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11–481360.pdf). 2. B. Liu, X. Xin, L. Zhang, and J. Yu, “109.92-Gb/s WDM-OFDMA Uni-PON with dynamic resource allocation and variable rate access,” Opt. Express 20(10), 10552–10561 (2012). 3. I. Tomkos, S. Azodolmolky, J. Sole-Pareta, D. Careglio, and E. Palkopoulou, “A tutorial on the flexible optical networking paradigm: state of the art, trends, and research challenges,” Proc. IEEE 102(9), 1317–1337 (2014). 4. Z. Zhang, and C. Li, “Hitless multi-rate coherent transceiver,” in Proceedings of Signal Processing in Photonic Communications (2015), paper SpS3D.2. 5. D. A. A. Mello, V. N. Rozental, T. C. Lima, F. C. Pereira, A. N. Barreto, M. Camera, and G. Bruno, “Adaptive optical transceivers: concepts and challenges,” J. Commun. Information Sys. 29(1), 1–11 (2014). 6. K. Roberts and C. Laperle, “Flexible transceivers,” in Proceedings of European Conference and Exposition on Optical Communications (2012), paper We.3.A.3. 7. R. Boada, R. Borkowski, and I. T. Monroy, “Clustering algorithms for Stokes space modulation format recognition,” Opt. Express 23(12), 15521–15531 (2015). 8. R. Borkowski, D. Zibar, A. Caballero, V. Arlunno, and I. T. Monroy, “Stokes space-based optical modulation format recognition for digital coherent receivers,” IEEE Photonics Technol. Lett. 25(21), 2129–2132 (2013). 9. P. Isautier, A. Stark, and S. E. Ralph, “Autonomous software-defined coherent optical receivers performing modulation format recognition in Stokes-space,” in Proceedings of OFC (2014), paper OTh3B.4. 10. P. Isautier, J. Pan, and S. E. Ralph, “Stokes space-based modulation format recognition for autonomous optical receivers,” J. Lightwave Technol. 33(24), 5157–5163 (2015). 11. P. Isautier, J. Pan, and S. E. Ralph, “Autonomous receivers for complex format identification and demodulation,” in Proceedings of AVFOP (2014), paper TuB2. Vol. 24, No. 14 | 11 Jul 2016 | OPTICS EXPRESS 15642 #265423 http://dx.doi.org/10.1364/OE.24.015642 Journal © 2016 Received 17 May 2016; revised 23 Jun 2016; accepted 27 Jun 2016; published 30 Jun 2016 12. J. Liu, Z. Dong, K. P. Zhong, A. P. T. Lau, C. Lu, and Y. Lu, “Modulation format identification based on received signal power distributions for digital coherent receivers,” in Proceedings of OFC (2014), paper Th4D.3. 13. S. M. Bilal, G. Bosco, Z. Dong, A. P. Lau, and C. Lu, “Blind modulation format identification for digital coherent receivers,” Opt. Express 23(20), 26769–26778 (2015). 14. Q. Zhuge, M. Morsy-Osman, and D. V. Plant, “Spectral efficiency-adaptive optical transmission using time domain hybrid QAM for agile optical networks,” J. Lightwave Technol. 31(15), 2621–2628 (2013). 15. J. K. Fischer, C. Schmidt-Langhorst, S. Alreesh, R. Elschner, F. Frey, P. W. Berenguer, L. Molle, M. Nolle, and C. Schubert, “Generation, transmission and detection of 4D set-partitioning QAM signals,” J. Lightwave Technol. 33(7), 1445–1451 (2015). 16. Q. Zhuge, C. Chen, and D. V. Plant, “Low computation complexity two-stage feedforward carrier recovery algorithm for M-QAM,” in Proceedings of OFC (2011), paper OMJ5. 17. M. Xiang, S. Fu, L. Deng, M. Tang, P. Shum, and D. Liu, “Low-complexity feed-forward carrier phase estimation for M-ary QAM based on phase search acceleration by quadratic approximation,” Opt. Express 23(15), 19142–19153 (2015). 18. K. Piyawanno, M. Kuschnerov, and B. Lankl, “Low complexity carrier recovery for coherent QAM using superscalar parallelization,” in Proceedings of European Conference on Optical Communication (2010), paper We.7.A.3. 19. Q. Zhuge, M. Morsy-Osman, X. Xu, M. E. Mousa-Pasandi, M. Chagnon, Z. A. El-Sahn, and D. V. Plant, “Pilotaided carrier phase recovery for M-QAM using superscalar parallelization based PLL,” Opt. Express 20(17), 19599–19609 (2012). 20. M. Qiu, Q. Zhuge, W. Wang, M. Chagnon, F. Zhang, and D. V. Plant, “Optimized superscalar parallelization based carrier phase recovery for agile metro optical networks,” J. Lightwave Technol. 34(4), 1111–1119 (2016). 21. Y. Wang, E. Serpedin, and P. Ciblat, “An alternative blind feedforward symbol timing estimator using two samples per symbol,” IEEE Trans. Commun. 51(9), 1451–1455 (2003). 22. M. Selmi, P. Ciblat, Y. Jaouen, and C. Gosset, “Block versus adaptive MIMO equalization for coherent PolMux QAM transmission system,” in Proceedings of European Conference on Optical Communication (2010), paper Th.9.A.5. 23. S.-H. Fan, J. Yu, D. Qian, and G.-K. Chang, “A fast and efficient frequency offset correction technique for coherent optical orthogonal frequency division multiplexing,” J. Lightwave Technol. 29(13), 1997–2004 (2011). 24. X. Chen, A. Al Amin, and W. Shieh, “Characterization and monitoring of laser linewidths in coherent systems,” J. Lightwave Technol. 29(17), 2533–2537 (2012).


Introduction
Global IP traffic is predicted to increase by about threefold over the next 5 years due to the emergence of bandwidth-consuming services [1].Growing capacity demands motivate the network evolution from conventional fixed optical networks to future agile optical networks [2][3][4][5][6].A hitless flexible transceiver aims to adapt the transceiver configurations such as data rate according to the instantaneous link margin without interrupting network traffic.The link margin could vary over time because of aging, environmental change, adding or dropping wavelength-division multiplexing (WDM) channels, and so forth.Therefore, the overall capacity can be better exploited compared to conventional systems, where transceivers are provisioned based on end-of-life margin.Furthermore, as network traffic becomes more dynamic and unpredictable in future networks, hitless line rate changes implemented by switching modulation formats are expected to bring more benefits in improving capacity and saving power consumption [4].In this scenario, modulation format identification (MFI) is essential for reconfiguring digital signal processing (DSP) for signal recovery and demapping at the receiver-side (Rx).Recently, several MFI techniques based on the properties of specific standard QAM formats have been proposed [7][8][9][10][11][12][13]. For example, MFI can be achieved in the Stokes space by identifying either the number of clusters or the higher order statistics [7][8][9][10].It can also be implemented based on the power distributions of the received signals [11][12][13].However, those techniques cannot be easily extended to more complex modulation formats, such as hybrid QAM formats or multi-dimensional formats [14,15].Moreover, these techniques are unable to track a fast block-by-block change of modulation format because of the high computational complexity to obtain an accurate MFI.
On the other hand, carrier phase estimation (CPE) and adaptive equalization are compulsory for current optical coherent transmission [16,17].They should be formattransparent in a hitless flexible coherent transceiver in order to minimize power consumption.Recently, a superscalar phase-locked loop (PLL) was proposed for the parallel implementation of CPE for arbitrary modulation formats [18][19][20].However, its integration with a format-transparent parallel adaptive equalizer has not been investigated.
In this paper, we propose the design of a hitless coherent transceiver for agile optical networks.First, we adopt a block-wise decision-directed least-mean-square (DD-LMS) equalizer for channel tracking and a pilot symbol aided superscalar PLL for CPE.For the purpose of MFI, the modulation format information is encoded onto BPSK pilot symbols, which are initially used for the superscalar CPE.Therefore, the proposed MFI does not result in extra overhead.In both simulations and experiments with QPSK, 16QAM, 64QAM, Hybrid QPSK/8QAM, and set-partitioning (SP)-512-QAM, we demonstrate that the proposed MFI method induces a negligible performance penalty.Furthermore, our hitless transceiver can support fast modulation format switching on a block by block basis.This is demonstrated in a 6400 km standard single mode fiber (SSMF) transmission experiment where we show format switching between QPSK and Hybrid QPSK/8QAM, and in a 1600 km SSMF transmission experiment with format switching between 16QAM and SP-128-QAM.Finally, we compare the performance of the proposed MFI method with two commonly used MFI methods.The results show that the proposed MFI covers a wider range of modulation formats and is more reliable especially at low optical-to-signal noise ratio (OSNR).

Transceiver structure
The structure of the proposed hitless flexible coherent transceiver is depicted in Fig. 1.Before being converted into analog signals by digital-to-analog converters (DACs), the transmitted sequence is processed.In particular, a rate change controller is used to choose the transmitted modulation format according to the request.The transmitter-side (Tx) DSP consists of Nyquist pulse shaping and pre-compensation of channel impairments such as narrow filtering and fiber nonlinearities.The linear electrical-to-optical (E/O) conversion including linear radio frequency (RF) drivers and linear IQ modulators is essential to ensure the signal quality of various formats.At the Rx, after the digitization using analog-to-digital converters (ADCs), DSP is used to recover and decode the signal.The Rx DSP consists of two stages: 1) initialization, and 2) tracking.Chromatic dispersion (CD) compensation is first applied in the frequency-domain.QPSK training symbols are sent for initialization where two identical patterns are sent for the coarse synchronization using the autocorrelation metric [14].Then the butterfly filter pre-convergence is achieved using the constant modulus algorithm (CMA), based on the QPSK training symbols.Next, the equalization training symbols are used for initial frequency offset (FO) estimation based on the 4th power of the QPSK symbols.Then, fine synchronization is achieved using the cross-correlation between the received and the transmitted training symbols [14].At the tracking stage, clock jitter is tracked using the method described in Ref [21].Afterwards, channel tracking is implemented with the DD-LMS based block equalizer, which works well for parallel processing as detailed in Section 2.4 below.Then, the periodically inserted BPSK symbols are used for the FO tracking, CPE and MFI.The superscalar parallelization is used to minimize the feedback delay for CPE.

Superscalar parallelization based carrier phase recovery
It is well known that the feedback delay of a PLL in conventional parallel processing is very large, resulting in a significantly reduced tracking speed and consequently a low laser linewidth tolerance.In order to reduce the feedback delay, a superscalar parallelization approach was proposed at the cost of additional buffers [18][19][20].In this scheme the received symbols are stored in a buffer with a size of S P × , where S is the block length for each parallel channel and P is the degree of parallelization.In order to reduce the feedback delay for the following PLL, the symbols are re-ordered to have the consecutive symbols in each parallel channel, as shown in Fig. 2. Pilot symbols are required to initialize the PLL for each block, because PLLs are independent across parallel channels and blocks.As illustrated in Fig. 2, the pilot symbols can be shared between two adjacent parallel channels.The red symbols are known pilot BPSK symbols used for CPE only, and the blue symbols are unknown BPSK symbols used for both CPE and MFI, which will be detailed in the next section.The overhead is 2 S .For the FO tracking, the average phase of each group of three known pilot symbols (red) is calculated and compared to the previous group of three pilot symbols to obtain one phase difference.We average over all of such phase differences within a buffer block to obtain an accurate FO.For CPE, in the first step we use three pilot symbols to calculate an initial phase by comparing the received symbols to the transmitted symbols.In the second step, after applying the initial phase to the pilot (red) and MFI (blue) symbols, we use the 2nd-power Viterbi and Viterbi (V&V) algorithm to extract a more accurate phase from both the pilot and the MFI BPSK symbols for the PLL initialization.Since cycle slip from the pilot symbols will lead to wrong symbol decisions for the entire parallel channel block, it is necessary to investigate the cycle slip effect on the proposed PLL initialization.Figure 3 Δ is the estimated phase error.The reference is the case where the MFI symbols are replaced by the known pilot symbols for the PLL initialization and the V&V algorithm is not applied, which is the same as the original scheme in [19].We can see that the probability of cycle slip decreases as the SNR increases.It decreases to about 10 −5 ~10 −6 for the proposed scheme (blue curve: 3 pilot symbols + 1 MFI symbol) at a SNR of 8 dB, which will have a very limited impact on the overall bit error ratio (BER).Furthermore, using V&V achieves a noticeable reduction in the cycle slip probability for SNR ≥ 8 dB.Even though the proposed scheme increases the cycle slip rate by roughly one order compared to the original scheme (reference: 4 pilot symbols) at low SNRs, it does not degrade the system performance for long-haul, regional and metro applications where the SNR is typically larger than 8 dB.
In order to further evaluate the performance of the proposed PLL initialization, we also plot the variance of the estimated phase error as a function of SNR in Fig. 3(b).Again, the second stage V&V phase estimation reduces the phase error variance, indicating an improved PLL initialization, especially for high SNRs.In particular, the performance of the proposed scheme (3 pilot symbols + 1 MFI symbol) approaches the original scheme (reference: 4 pilot symbols) at SNR = 8 dB, where the difference of the phase error variances is less than 0.4 dB.

Principle of the proposed MFI
Operation principle of the proposed MFI scheme is to encode the modulation format information onto the MFI BPSK symbols, which are periodically inserted as shown in blue part of Fig. 2. At the receiver before MFI, both phase noise and amplified spontaneous emission (ASE) noise on MFI symbols should be mitigated.For the phase noise mitigation, we use the known pilot symbols which are also used for the PLL initialization.Then to suppress the influence of ASE noise, especially at the low SNR regime, we propose to send repetitive MFI symbols and average them at the receiver.Assuming the phase noise is completely compensated, the received k -th symbol is represented by 2 2 ; , , where S P and N P are the signal and noise power, respectively.Then N MFI BPSK symbols are used for averaging as Therefore, the SNR after averaging is increased by a factor of N .Finally, the theoretical BER of the MFI symbols is obtained as  3) are also plotted.The performance penalty of the simulated BER with respect to the theory is due to the occurrence of cycle slip caused by phase noise at the low SNR regime.In order to achieve a BER lower than 1 × 10 −10 at SNR = 6.25 dB (the soft-decision threshold for QPSK at BER = 1e-2), we fix 8 N = in our following investigations.Note that since every 2 parallel channels contain 1 MFI BPSK symbol, and N identical MFI symbols are transmitted for averaging, the number of bits to encode the modulation format information is equals to (2 ) P N per superscalar block, where P is the total number of parallel channels.On the other hand, incorrect MFI at receiver side will inevitably cause misconfiguration of DSP for signal recovery, degrading system performance or even leading to system failure.In practical implementation, we can increase the averaging length to reduce the probability of incorrect MFI, in order to guarantee the correct MFI.

Block-wise equalization
Format-transparent DSP is essential to realize a low power consumption hitless flexible transceiver.In order to integrate an adaptive DD-LMS equalizer with the superscalar phase recovery algorithm, we propose to use a block-wise DD-LMS scheme.The equalization is performed by a butterfly L -tap 2 S T -spaced finite impulse response (FIR) filter.The received signal after the adaptive equalization for each polarization is obtained where xx x y y x y y w w w w are the corresponding adaptive filter taps for the current block l. ( ) where Θ represents the decision operation.n θ is the estimated phase noise in the PLL.M is the block size and it is equal to the buffer size in this work.l x μ and l y μ are the optimal stepsize and are given by (

Simulations
Simulations are conducted to investigate the performance of the proposed hitless flexible transceiver.In our simulations, the system symbol rate is fixed at 35 Gbaud.For the Tx Nyquist filter shaping, a 64-tap root raised-cosine (RRC) time-domain FIR filter with a rolloff factor of 0.1 is applied and the same filter is also used at the Rx for matched filtering.For the superscalar structure, each parallel channel contains 100 S = symbols, out of which 98 symbols carry data, resulting in an overhead of 2%.The degree of parallelization is 64 P = , leading to a buffer size of 6400 S P × = symbols per polarization.The buffer structure is the same as that shown in Fig. 2, where every two parallel channels include one MFI BPSK symbol and thus the total number of MFI symbols is 32 within one buffer block.Since we transmit 8 N = identical MFI symbols for averaging in order to mitigate ASE noise, the number of MFI bits per buffer block is (2 ) 4 P N = .Therefore, the modulation format information can be encoded in 4 bits, which leads to 16 different modulation formats.Table 1 gives an example of a 4-bit modulation format encoding table.As noted in the table we can encode arbitrary modulation formats, including hybrid QAM formats and multi-dimensional formats.In our configuration, the modulation format can be switched for each buffer block, which contains 6400 symbols.The combined linewidth and symbol duration product  First, the laser linewidth tolerance of the proposed hitless transceiver is investigated.DP-Hybrid QPSK/8QAM modulation format [14], 3) DP-16QAM, 4) a multi-dimensional modulation format of 512-SP-QAM [15], and 5) DP-64QAM, respectively.The "Reference" in the plots is the system using the original superscalar CPE with four known pilot BPSK symbols and no MFI symbols.Therefore, the overhead is the same for the "Reference" system and the proposed hitless system with MFI.Note that in the "Reference" case the V&V step is not applied.The processing delay of PLLs is set to four clock cycles.The parameters of the superscalar CPE are optimized for each S T υ Δ ⋅ value.As shown in Fig. 5(a), the performance penalty caused by integrating the MFI function with respect to the "Reference" is negligible.In particular, the induced penalty is less than 0.05 dB for various ) with respect to the number of total BPSK symbols for PLL initialization in two consecutive parallel channels is shown in Fig. 5(b).Here, we vary the number of BPSK symbols while keeping the length of parallel channel fixed at 100, leading to a varying pilot overhead.Note that in the "Reference" system the number of pilot BPSK symbols is the same as the combined number of MFI and pilot BPSK symbols in the system with MFI.As a result, the overheads for the two systems are always the same.The OSNR is chosen properly to separate curves.Only one MFI symbol is used in all the cases in Fig. 5(b) for our proposed system.Again, we note that the system with MFI shows very similar performance to the "Reference" system with different number of BPSK symbols.Next, we investigate the system tolerance to FO drifting and polarization rotation.Figure 6(a) shows the OSNR penalty at BER = 2 × 10 −2 as a function of the FO drifting speed.Obviously, the tolerance to the FO drifting decreases with higher level modulation formats.In the worst case of DP-64QAM, the FO drifting tolerance can still reach up to 1 MHz/μs without a performance loss, which is faster than the typical frequency variation rate of an external cavity laser (ECL) [23].Moreover, when the FO drifting speed is 2 MHz/μs, the performance penalty is only 0.07 dB.If needed, the tolerance to the FO drifting can be further increased by optimizing the length of the phase difference average block.On the other hand, Fig. 6(b) shows the OSNR penalty at BER = 2 × 10 −2 as a function of the polarization rotate speed.Here the channel is assumed to be ′ Γ as where Ω is the polarization rotate speed in rad/s.Again, the tolerance to Ω decreases with higher level modulation formats.Moreover, in the worst case of DP-64QAM, the tolerance is about 10 krad/s without any performance penalty.It is noteworthy that the tolerance to Ω can be improved by decreasing the buffer size.Therefore, for a specific transmission system, the superscalar buffer size should be designed while taking into account the tracking speed requirement.The experiments are conducted in a 34.94 GBaud single-channel coherent transmission system as outlined in Fig. 7. First, the offline generated real and imaginary components of the orthogonal polarizations are loaded to the transmitter module of a Ciena WaveLogic 3 transceiver, which incorporates a low-linewidth ECL, four high-speed DACs and a DP IQ modulator.The wavelength of the laser is set to 1554.54 nm.The output of the transmitter is boosted by an Erbium-doped fiber amplifier (EDFA).A variable optical attenuator (VOA) controls the launch power before the signals enter a re-circulating loop.The loop contains 320 km SSMF, and an EDFA is employed after every 80 km fiber to compensate the loss.After the loop, the signals are filtered, amplified and filtered again.An ECL-based integrable tunable laser assembly (ITLA) is employed as the LO for coherent detection.The total laser linewidth in the system is estimated to be <30 kHz based on a laser phase noise estimation method as described in [24].The optical-to-electrical conversion is achieved by 4 balanced photodiodes, and a four-channel real time oscilloscope with a sampling rate of 80 GSa/s per channel is used to digitize the waveforms.Finally, the captured waveforms are processed offline in MATLAB.The offline DSP at the Tx and Rx for the experiments is the same as that for the simulations.First, we investigate the back-to-back (B2B) performance for all five modulation formats, as shown in Fig. 8.The simulation results are also presented for comparison.We can see that the system with MFI and the "Reference" system without MFI show almost the same performance under various OSNRs in both simulations and experiments.Further, the transmission performance is evaluated.As shown in Figs.9(a)-(c), both systems achieve almost the same performance under different launch powers and the optimized launch power for DP-16QAM, 512-SP-QAM, DP-64QAM is 0, 1, and 1 dBm, respectively.With the optimized launch power, we evaluate the BER at different transmission distances as shown in Fig. 9(d).Again, both systems achieve identical performance indicating that the proposed implementation of MFI does not cause a system performance penalty., respectively.The measured SNR within each block is quite stable when we switch format.The fluctuation is within 0.3 dB, which is mainly caused by the limited number of symbols in the measurement.The BER values are similar for the same format.As expected, the achieved BERs for DP-QPSK and 128-SP-QAM are smaller than DP-Hybrid QPSK/8QAM and QP-16QAM, respectively.Thus, the proposed system achieves stable performance with a fast switching between modulation formats.Moreover, the hitless receiver can automatically identify the format for current block.It is worth mentioning that in addition to the DSP platform, which is the focus of this work, hitless flexible transceivers also require reconfigurable forward error correction (FEC) circuits and client interfaces to accommodate varying bit rates in a hitless manner.Similarly, the proposed MFI scheme can also be used to aid the hitless reconfiguration of these blocks.

Performance comparison between different MFI methods
The performances of different MFI techniques are compared in this section.It should be pointed out that existing MFI methods are mostly designed for standard M-QAM modulation formats.An exception is Ref [11], which proposed a MFI method for Hybrid QAM modulation formats.However, this method requires an additional procedure examining the statistical distribution of received signal's radius.Moreover, it cannot be easily extended to higher level Hybrid QAM such as Hybrid 8QAM/16QAM.On the other hand, to the best of our knowledge, no MFI methods have been proposed for multi-dimensional modulation formats.For example, polarization switched QPSK (PS-QPSK) format cannot be distinguished from DP-BPSK since they exhibit the same power distributions and Stokes distributions, as shown in Fig. 11.In contrast, with our proposed MFI method, both multidimensional and Hybrid QAM formats can be easily identified by decoding the corresponding BPSK symbols for MFI and referring to an encoding table.In order to compare the performance of the proposed MFI method with other methods, the probability of successful MFI with respect to OSNR for DP-16QAM under B2B transmission is shown in Fig. 12(a), where we conduct 1000 times independent MFI for each OSNR for the probability calculation [7].The parameters of the superscalar parallelization structure in this experiment are the same as those described in Section 3.1, and the symbol rate is 34.94 Gbaud, which is the same as Section 3.2.We can see that for our proposed MFI no error is observed over the OSNR range down to 14 dB.The k-means clustering based Stokes MFI method [7] and the feature-based MFI method [12] are evaluated for comparison.Their successful probabilities become less than 100% when the OSNR is reduced to be less than 19~20 dB.Note that the required OSNR to achieve BER = 2 × 10 −2 is 18.4 dB in our experiments, as shown in Fig. 8. Next, the relationship between the probability of correct MFI and transmission distance is shown in Fig.
12(b).Again, no error is observed using the proposed MFI method with a transmission distance of 5440 km.However, the distance to assure 100% correct MFI is shortened to 1600 km and 2240 km for the Stokes MFI and feature-based MFI methods, respectively.Therefore, we conclude that our proposed MFI method not only suits for all kinds of modulation formats but also shows more accurate MFI compared to other MFI methods especially under the low OSNR and long-haul transmission conditions.

Conclusion
In this paper, we propose a hitless coherent transceiver enabled by a novel MFI scheme for agile optical networks.At the receiver, we utilize a block-wise decision-directed least-meansquare (DD-LMS) equalizer for channel tracking and a pilot symbol aided superscalar phase locked loop (PLL) for phase tracking, in order to realize a low-complexity hitless flexible receiver with parallel processing.In addition, we propose to insert BPSK symbols periodically for the purpose of MFI.This MFI method is capable of supporting arbitrary modulation formats and tracking a fast switching of modulation formats.These MFI BPSK symbols are also used to initialize the superscalar PLLs.We then show that the proposed MFI induces a negligible OSNR penalty relative to a system without MFI at the same overhead in both simulations and experiments.In particular, the transmissions with various formats including QPSK, 16QAM, 64QAM, Hybrid QPSK/8QAM and SP-512-QAM are conducted to evaluate the performance of the proposed hitless transceiver.The hitless transmission with a block-by-block modulation format switching is successfully demonstrated in the experiment.Finally, we demonstrate that the proposed MFI method achieves a superior performance when compared with other existing MFI methods.
(a) shows the probability of cycle slip ( initialization as a function of the signal to noise ratio (SNR) per symbol, where θ

Fig. 3 .
Fig. 3. (a) Probability of cycle slip versus SNR per symbol.(b) Phase error variance versus SNR per symbol.

Figure 4
Figure 4 plots the simulated BER as a function of SNR under different linewidth and symbol duration product ( S T υ Δ ⋅ ).The theoretical curves derived from Eq. (3) are also plotted.The

Fig. 4 .
Fig. 4. BER versus SNR per symbol with different numbers of MFI BPSK symbols for averaging under the condition of various S T υ Δ ⋅ .

Δ
represent the L elements of the input sequences corresponding to the output block l .In order to update the filters for the next block ( 1 l + ), a cost function for block l is calculated, based on the phase recovered symbols after the following superscalar phase recovery with parallel data processing[22].The adaptive filter taps for block ( 1 l + ) are then obtained by are the cost functions in block l and are given by 10 −5 and the frequency offset is 1 GHz.Differential coding is not required due to the use of pilot symbols.The channel is modeled by a unitary 2 × 2 matrix Γ , which rotates the horizontal and vertical states of polarization at the transmitter to a pair of arbitrary but orthogonal states as φ are the azimuth and elevation rotation angles, respectively,

Figure 5 (
a) shows the OSNR penalty to achieve a BER = 2 × 10 −2 as a function of for the following formats: 1) dual-polarization (DP) QPSK, 2) a

Fig. 9 .
Fig. 9. BER versus launch power for (a) DP-16QAM after 1920 km SSMF transmission, (b) 512-SP-QAM after 960 km SSMF transmission, (c) DP-64QAM after 320 km SSMF transmission.(d) BER versus transmission distance under the optimized launch power.Next, the concept of hitless rate change is demonstrated by switching formats block by block.Each block contains 6400 symbols as identical to the buffer size.Within each bock, we measure the SNR (obtained by measuring the noise variance on the received symbols) and the BER for the interleaved DP-QPSK and DP-Hybrid QPSK/8QAM signals after 6400 km SSMF transmission and for the interleaved DP-16QAM and 128-SP-QAM signals after 2240