Xiang Zhou
(AT&TLabs-Research,Middletown,NJ 07748,USA)
Abstract This paper gives an overview of the generation and transmission of 450 Gb/s wavelength-division multiplexed(WDM)channels over the standard 50 GHz ITU-Tgrid at a net spectral efficiency(SE)of 8.4 b/s/Hz.The use of nearly ideal Nyquist pulse shaping,spectrally-efficient high-order modulation format,distributed Raman amplification,distributed compensation of ROADM filtering effects,coherent equalization,and high-coding gain forward error correction(FEC)code may enable future 400G systems to operate over the standard 50 GHz grid optical network.
Keyw ords spectral efficiency;optical;modulation format;coherent
O ptical transport costs have traditionally been lowered by increasing per-channel data rates and spectral efficiency(SE).With 100 Gb/s wavelength-division multiplexed(WDM)technology being made commercially available in 2010,research is now being conducted on per-channel bit rates beyond 100 Gb/s that have spectral efficiencies greater than 2 b/s/Hz.It is likely that 400 Gb/s willbe the next-generation transport standard[1].From an historical point of view,increasing the transport interface rate in proportion to SEhas minimized the cost per bit transmitted.In line with this trend,SEof 8 b/s/Hz might be needed for future 400 Gb/s systems(Fig.1).Such SEenables future 400 Gb/s systems to operate over existing optical networks with 50 GHz WDM channel spacing;therefore,it is very attractive from a carrier’s perspective.
Because of severallimitations,transmitting 400 Gb/s per-channel signals on the 50 GHz WDM grid is very challenging.First,according to Shannon theory,11.76 d B signal-to-noise ratio(SNR)is required for an 8 b/s/Hz 400 Gb/s system,and this is higher than in current 2 b/s/Hz 100 Gb/s systems(without taking fiber nonlinearity into consideration).Second,fiber nonlinearity limits the allowable launch power and,consequently,the achievable signal SNR.Furthermore,a higher-SEmodulation format is less tolerant of fiber nonlinearity because of the reduced Euclidean distance.Third,non-ideal passband shapes from optical network components,such as the widely used reconfigurable add/drop multiplexer(ROADM),cause significant channel narrowing.Finally,a high-SEmodulation format is less tolerant of laser phase noise,which may introduce extra penalty.
Spectrally efficient,high-order coherent modulation formats,coherent detection,transmitter-and receiver-side digital signal processing(DSP),distributed Raman amplification,high-coding-gain forward error correction(FEC)code,and new low-loss,low-nonlinearity fibers have allbeen considered as enabling technologies for next generation high-speed transport systems.These technologies are currently being explored by the research community.Single-channel bit rates beyond 100 Gb/s have been demonstrated using single-carrier high-order coherent modulation formats(up to 448 Gb/s)[2]and multicarrier-based high-order coherent modulation formats(up to 448 Gb/s[3],[4]and 1.2 Tb/s[5]).For WDM transmission,the following have been demonstrated using polarization-division-multiplexed(PDM)-16 QAM,digital coherent detection,distributed Raman amplification,and new ultralarge-area fiber(ULAF):50 GHz-spaced,10×224 Gb/s over 12×100 km with SEof 4 b/s/Hz[6];70 GHz-spaced,10×456 Gb/s over 8×100 km with SEof 6.1 b/s/Hz[7];and 100 GHz-spaced,3×485 Gb/s over 48×100 km with SEof 4 b/s/Hz[8].
Recently,400G transmission over the standard 50 GHz WDM grid has been demonstrated using PDM-32 QAM[9]-[11].In[10],8×450 Gb/s WDM signals transmitted over 400 km of ULAFfiber and passing through one 50 GHz grid wavelength-selective switch(WSS)-based ROADM was demonstrated with net SEof 8 b/s/Hz.This was the first demonstration of a 400G WDM system over the standard 50 GHz grid opticalnetwork.In[11],the transmission reach was extended to 800 km by introducing a broadband optical spectral-shaping technique to compensate for ROADM filtering effects.This is the longest transmission distance beyond 4 b/s/Hz that has been demonstrated for WDM SE.Key enabling technologies and experimental results are reviewed in the following sections.
In section 2,the 450 Gb/s PDM-Nyquist-32 QAM transmitter is described.In section 3,the coherent receiver and DSPalgorithms are presented.In section 4,two WDM transmission experiments and back-to-back are presented.In section 5,a summary is given.
◀Figure 1.Projected demand for SEfor thenext-generationtransport standard.
To overcome limited digital-to-analog converter(DAC)bandwidth,a frequency-locked five-subcarrier generation method is used to create the 450 Gb/s per-channel signal[10],[11].Fig.2 shows the demonstrated 450 Gb/s PDM-Nyquist-32 QAM transmitter.The output from a continuous-wave(CW)laser with line width of approximately 100 kHz is split by a 3 d B optical coupler(OC).One output is sent to a Mach-Zehnder modulator(MZM-1)driven by a 9.2 GHz clock in order to generate two 18.4 GHz-spaced subcarriers per channel(the two first-order signal components)that are offset from the original wavelengths by±9.2 GHz.After an erbium-doped fiber amplifier(EDFA)and a 12.5/25 GHz interleaver filter(ILF),the original wavelengths and second-order harmonics are suppressed by more than 40 d Brelative to the first-order components(Fig.2a).The signal is then equally split between two outputs of a polarization beam splitter(PBS)prepared by a polarization controller(PC).The two subcarriers on one PBSoutput are sent to an IQ modulator(IQ MOD1),driven by a pre-equalized 9 Gbaud Nyquist 32-QAM signalwith 215-1 pseudorandom pattern length.The Nyquist pulse shaping has roll-off factor of 0.01,and the digital Nyquist filter has a tap length of 64.Fig.3 shows the Nyquist filter impulse response used in this experiment and the resulting eye diagram of the generated 32-QAM baseband signal in one quadrature.Frequency-domain based pre-equalization[12]is used to compensate for the band-limiting effects of the DACs,which have 3 d B bandwidths less than 5 GHz at 10 bit resolution and a 24 GSa/s sample rate.Fig.4 shows the relative amplitude spectra of the generated 9 Gbaud Nyquist 32-QAM baseband electrical drive signals(after DACs)with and without pre-equalization.The filtering effects caused by the DACs are compensated for using frequency-domain-based digital pre-equalization.
Asecond Mach-Zehnder modulator(MZM-2)driven by a 9.2 GHz clock is placed at the second PBSoutput to generate first-order signal components at 0 GHz and 18.4 GHz offsets from the originalwavelength.After MZM-2,the signals pass through two 25/50 GHz interleavers to suppress the 0 GHz signal components and the unwanted harmonics(Fig.2b).The second ILFre-inserts the original CWsignal(from the second 3 d B OC output),resulting in three 18.4 GHz-spaced subcarriers from the original wavelength.These three subcarriers pass through an IQ modulator(IQ MOD2)that is driven by a second pre-equalized 9 Gbaud Nyquist 32-QAM signal with 215-1 pseudorandom pattern length and originating from a second DAC.Then,the sets of two and three 45 Gb/s subcarriers are passively combined and polarization multiplexed with 20 ns relative delay.This results in a 450 Gb/s signal that occupies a spectralwidth of 45.8 GHz,sufficiently confined to be placed on the 50 GHz ITU grid(Fig.5).practical system,frequency-domain-based equalization is more efficient than time-domain-based equalization.Next,polarization recovery and residual CD compensation are simultaneously performed with four complex-valued,51-tap,T/2-spaced adaptive FIRfilters,optimized using a two-stage equalization strategy.The classic constant-modulus algorithm is used in the first stage for pre-convergence,and then a decision-directed least-mean-square algorithm is used for steady-state optimization.
▲Figure 2.450 Gb/s PDM-Nyquist 32-QAMtransmitter.
▲Figure 3.(a)Interpolated impulse response of the Nyquist filter and(b)the resulting eye diagram of the generated baseband 32-QAMsignal.
▲Figure 4.Measured relative amplitude spectra of 9 Gbaud Nyquist 32 QAMbaseband electrical drive signals(after DAC)(a)without pre-equalization and(b)with pre-equalization.
The carrier frequency and phase are recovered after the initial equalization.The frequency offset between the LO and signal is estimated by using a constellation-assisted two-stage blind frequency search method[12].The frequency offset is scanned at a step size of 10 MHz and 1 MHz,and the optimal frequency offset is the one that gives the minimum mean-square error.For each trial frequency,the carrier phase is recovered on a best-effort basis using a newly proposed hybrid blind-phase search(BPS)and maximum likelihood(ML)phase-estimation method[13].Decisions made after phase estimation are then used as reference signals for mean-square error calculation.This frequency recovery method is applicable for any modulation format and only uses tens of symbols(96 are used in the experiments in section 3)to reliably recover the carrier frequency.The carrier phase is estimated using a two-stage method;that is,the carrier phase recovered from the previous symbol is used as an initial test-phase angle.The signal decided on after the initial test-phase angle is then used as a reference for a more accurate ML-based phase recovery using a feed-forward configuration[13].For the first block of data,the initial phase angle is obtained using BPS[14].To reduce the probability of cycle-slipping(no differentialcoding/encoding is used in the experiments in section 3),sliding-window-based symbol-by-symbol phase estimation is used.To calculate the bit-error ratio(BER),errors are counted for more than
Figure 5.▶Measured opticalspectrum of thegenerated 450 Gb/s PDM-Nyquist 32-QAM signal consisting of five subcarriers.
A DSP-enabled coherent receiver is used to detect and demodulate the received PDM-Nyquist-32 QAM signal.The polarization-and phase-diverse coherent receiver front-end consists of a polarization-diverse 90-degree hybrid,a tunable external cavity laser(ECL)with approximately 100 kHz line width that serves as the local oscillator(LO),and four balanced photodetectors.An optical tunable filter(OTF)with approximately 50 GHz-3 d B bandwidth is used to select the desired channel for detection.The subcarriers are selected by tuning the LO to within 200 MHz of their center frequencies.A four-channel real-time sampling scope with 50 GSa/s sample rate and 16 GHz analog bandwidth performs sampling and digitization(ADC),followed by post-transmission DSPof the captured data on a desktop computer.
Fig.6 shows the flow chart for the offline receiver DSP.After digitally compensating for sampling skews and hybrid phase errors in the front-end,and after anti-aliasing filtering,the 50 GSa/s signal is down-sampled to a rate double the symbol rate.Then,the bulk chromatic dispersion(CD)is compensated for using a fixed T/2-spaced finite impulse response(FIR)filter with 72 complex-value taps.In a 1.2×106 bits of information.
▲Figure 6.Post-transmission offline DSPflow chart.
▲Figure 7.Experimentsetup for 8×450 Gb/s over 400 km transmission.
Two 450 Gb/s per-channel WDM transmission experiments using PDM-Nyquist-32 QAM were performed[10],[11].In the first experiment,no optical pulse shaping was used to compensate for the filtering effects caused by the 50 GHz grid ROADM.In the second experiment,a liquid-crystal-on-silicon(LCoS)-based flexible-bandwidth WSSwas used as a broadband optical pulse shaper to mitigate the ROADM filtering effects.
4.1 8×450 Gb/s over 400 km Without Optical Shaping
Fig.7 shows the experiment setup for WDM transmission of 8×450 Gb/s PDM-Nyquist 32 QAM signals over 400 km.The eight 450 Gb/s C-band channels are based on odd(192.30-192.60 THz)and even(192.35-192.65 THz)sets of multiplexed,100 GHz-spaced ECLs.These ECLs are combined using a 3 d BOCand are modulated in the 450 Gb/s PDM-Nyquist-32 QAM transmitter(Fig.1).The measured optical spectrum of a single 450 Gb/s 32-QAM channelis shown in Fig.5,and the eight-channel WDM spectrum prior to transmission is shown in Fig.8.Fig.9 shows the back-to-back BERfor a single subcarrier operating at 9 Gbauds,a single 450 Gb/s channel comprising five subcarriers,and one of the center channels of the 8×450 Gb/s WDM 50 GHz-spaced channels.The optical signal noise ratio(OSNR)for the single subcarrier in Fig.9 is a scaled result obtained by multiplying the actual OSNRof the single subcarrier signal by five.No ROADM filtering was used in these back-to-back measurements.Fig.9 also shows the recovered Nyquist-32 QAM constellation diagram at an OSNRof 38.9 d B for a single 450 Gb/s channel.For comparison,a theoretical curve is included in Fig.9.There is an approximately 6 d B implementation penalty at 2×10-3BER.Because digital Nyquist pulse shaping is used,the OSNRpenalty at 2×10-3BERfrom interchannel WDM crosstalk is very small,even without narrow optical filtering.This is because the 450 Gb/s signal is well confined within a 45.8 GHz bandwidth.The OSNR penalty from intersubcarrier crosstalk is less than 1 d B.Aportion of the intersubcarrier crosstalk originates from the out-of-band aliased spectral components from the electrical drive signals.
For WDM transmission,the eight 450 Gb/s signals pass through a 1×850 GHz-spaced WSSbased on liquid-crystal technology in order to emulate the filtering by a ROADM.Odd and even channels are sent to separate WSSoutput ports for maximum filtering,and a relative delay of 175 symbols decorrelates the odd and even channels before they are recombined using a 3 d B OC.Filtering from the WSS passband is significant because the-3 d Bbandwidth is 42.2 GHz,and the-6 d B bandwidth is 46.6 GHz(Fig.10).The transmission line after the ROADM consists of four 100 km spans of ULAFwith,at 1550 nm,average Aeff of 135μm2,average attenuation of 0.179 d B/km,and average dispersion of 20.2 ps/nm/km.The span inputs are spliced to standard single-mode fiber jumpers,and a 1450/1550 nm WDM coupler is included for the counter-propagating Raman pumps at the span outputs.The span losses are 19.2,19.6,19.2,and 18.9 d B.Hybrid Raman-EDFAs are used,with an on/off Raman gain of 11 d B per span from approximately 1450 nm pumps.Because of the gain-flattened EDFAs and the narrow total bandwidth(approximately 3 nm)of the eight 450 Gb/s channels,after 400 km the spectrum is flat to within 1 d B for a large range of span input powers.
Figure 8.▶Measured opticalspectrum of the generated 8×450 Gb/s WDMsignals.
▲Figure 9.Measured back-to-back performance under three different conditions.
◀Figure 10.Power transmission of the 50 GHz-grid WSS used to emulate theROADM.
Fig.11 and Fig.12 show the results of the 8×450 Gb/s transmission experiment.The BERs of the five subcarriers of the center channelat 192.50 THz are measured because the total launch power to the spans is varied.From the average BERof the five subcarriers,the optimum launch power per span is 11 d Bm,an average of 2 d Bm per 450 Gb/s channel and-5 d Bm per subcarrier(Fig.11).At this launch power,the OSNRis 34 d B per 0.1 nm after the 400 km transmission.It is assumed that digitalsignals from allfive subcarriers coexist on one silicon chip,and the FEC is encoded on a per-channel basis,not on a per-subcarrier basis.Thus,the net BERof the 450 Gb/s channel is the average BERof the subcarriers[15].Fig.12 shows the performance of each of the eight 450 Gb/s DWDM channels at the optimum 11 d Bm total launch power.The average BERof the five subcarriers of all eight channels is better than 3.8×10-3,which is lower than the 4.5×10-3BERthreshold for a 7%continuously interleaved BCH code.The inset in Fig.12 shows the optical spectrum after 400 km,and spectral filtering by the WSSis evident.
4.2 5×450 Gb/s over 800 km with Optical Shaping
Fig.13 shows the experiment setup for WDM transmission of 5×450 Gb/s PDM-Nyquist-32 QAM over 800 km.An
LCoS-based dynamic,broadband optical spectralshaper with 1 GHz resolution is followed by a booster EDFA and is inserted before the 50 GHz grid ROADM.The optical spectral shaper pre-compensates for the ROADM filtering.Pre-compensation results in enhanced interchannel WDM crosstalk because of the limited channelisolation of the 50 GHz WSS.Therefore,a 50/100G interleaver is used inside the ROADM emulator to combine the odd and even channels and further suppress interchannel WDM crosstalk.The recirculating loop contains the same four 100 km spans of ULA fiber with 11 d B on/off Raman gain,as previously described for the 400 km experiment.After two circulations(800 km),the spectrum of the five 450 Gb/s channels was flat to within 1 d B for span input powers ranging from 6 to 12 d Bm.The optimaltotal launch power at the span inputs was 9 d Bm.
The optical spectra of a single 450 Gb/s 32-QAM channel before and after the 50 GHz ROADM are shown in Fig.14(a)and(b),respectively.The thin lines show the spectra without optical spectral shaping.When the signal without optical spectral shaping passes through the 50 GHz ROADM,a significant amount of filtering occurs.The power loss due to filtering can be largely precompensated for by using broadband optical spectral shaping,shown by the thick lines in Fig.14(a)and(b).The optical spectra of the five 450 Gb/s DWDM signals that have been spectrally shaped prior to and after the ROADM are shown in Fig.14(c).The filtering effects of all five channels have been largely precompensated for.
▲Figure 11.BERfor the five subcarriers of the center DWDMchannel at192.50 THz after 4×100 km transmission for a range of totallaunch powers into the fiber spans.
▲Figure 12.Average BERof the five subcarriers of alleightchannels after 400 km transmission with optimum launch power.The insetshows the opticalspectrum after 400 km.
Fig.15 shows the measured OSNRsensitivity for the 450 Gb/s PDM-Nyquist-32 QAM signalfor a single channel without ROADM optical filtering(the back-to-back case with no optical shaping),with ROADM filtering and optical spectral shaping,and with ROADM filtering but without optical shaping.For comparison,a theoretical OSNRsensitivity curve is also shown in Fig.15.Compared with the previous experiment,back-to-back sensitivity is improved by about 2 d B at 2×10-3BER.This improvement is achieved mainly by improving bias optimization for the two IQ modulators.With ROADM filtering,optical spectral shaping improves OSNRsensitivity by more than 2 d Bat 2×10-3BER.
Fig.16 shows the optical spectrum after 800 km transmission.The measured OSNRis 31 d B per 0.1 nm.Fig.17 shows the measured BERof all five WDM channels at the optimal total launch power of 9 d Bm(2 d Bm per channel).The inset in Fig.17 shows the measured BERfor the center channel located at 192.5 THz as the total launch power into the spans is varied.The BERof all five channels is better than 3.8×10-3,and the worst subcarrier BER is 4.3×10-3,both of which are lower than the 4.5×10-3BERthreshold for a 7%continuously interleaved BCHcode[16].
This paper describes two high-SE 8.4 b/s/Hz 400G transmission experiments.These experiments show for the first time that 400 Gb/s per channel WDM signals can be transmitted up to 800 Km(eight 100 km spans)over the conventional 50 GHz ITU-Tgrid and passing through one 50 GHz grid ROADM.These results are achieved by using a spectrally efficient high-order modulation format,that is,Nyquist-shaped PDM-32 QAM,as well as pre-and post-transmission digital equalization.Low-nonlinearity fiber and low-noise Raman amplification can also be used to address the reduced nonlinear and noise tolerance of the high-order modulation format.
To overcome the bandwidth limitation of the available DACs,a frequency-locked five-subcarrier generation method with high sideband suppression is used to create the 450 Gb/s PDM-Nyquist-32 QAM signals.By using five frequency-locked subcarriers and near ideal Nyquist pulse shaping,the 450 Gb/s spectrum has a signal spectral width of about 45.8 GHz,well within the 50 GHz channel spacing.Using multiple subcarriers(with Nyquist pulse shaping)within each channelallows all-optical subwavelength grooming,which may be useful for future 400 Gb/s and ultrahigh-speed systems.
Another key enabling technology is LcoS-based broadband optical spectral shaping.This spectral-shaping technique can be used to mitigate the narrow ROADM filtering effects.An opticalspectral shaper could be designed within each ROADM to compensate for ROADM filtering effects in a distributed manner.Distributed ROADM filtering compensation has some advantages over transmitter-side pre-equalization or receiver-side post-equalization because,unlike transmitter-side pre-equalization,distributed compensation does not require more launch power into the fiber.Therefore,it does not increase fiber nonlinearity.Distributed compensation also does not increase noise components,unlike receiver-side post-equalization.A drawback of distributed compensation is the need for extra optical amplification to compensate for the loss caused by the optical spectralshaper.
▲Figure 13.Setup for 8×450 Gb/s transmission over 800 km.
▲Figure 14.(a)Optical spectra for a single channel with and withoutoptical spectral shaping;(b)before and after the ROADM,and(c)the spectra of the five DWDMsignals with opticalspectralshaping before and after the ROADM(launch to the fiber).
▲Figure 15.OSNRsensitivity for the generated 450 Gb/s signal.
▲Figure 17.Measured BERof the five 450 Gb/s DWDMchannels after 800 km transmission.The inset shows measured BERfor the center DWDMchannelversus total launch power for all five 450 Gb/s channels.
▲Figure 16.Measured opticalspectrum after 800 km transmission.