Ze Dong

(ZTE USA,Morristown,NJ 07960,USA)

AbstractNyquist wavelength-division multiplexing(N-WDM)allows high spectral efficiency(SE)in long-haul transmission systems.Compared to polarization-division multiplexing quadrature phase-shift keying(PDM-QPSK),multilevel modulation,such as PDM 16 quadrature-amplitude modulation(16-QAM),is much more sensitive to intrachannel noise and interchannel linear crosstalk caused by N-WDM.We experimentally generate and transmit a 6×128 Gbit/s N-WDM PDM 16-QAM signal over 1200 km single-mode fiber(SMF)-28 with amplification provided by an erbium-doped fiber amplifier(EDFA)only.The net SEis 7.47 bit/s/Hz,which to the best of our knowledge is the highest SEfor a signal with a bit rate beyond 100 Gbit/s using the PDM 16-QAM.Such SEwas achieved by DSPpre-equalization of transmitter-side impairments and DSPpost-equalization of channel and receiver-side impairments.Nyquist-band can be used in pre-equalization to enhance the tolerance of PDM 16-QAM to aggressive spectral shaping.The bit-error ratio(BER)for each of the 6 channels is smaller than the forward error correction(FEC)limit of 3.8×10-3 after 1200 km SMF-28 transmission.

Keyw ords 16-QAM;coherent detection;Nyquist wavelength-division multiplexing;Nyquist-band;pre-equalization;spectral efficiency;signal transmission

1 Introduction

W ith the commercialization of 100G Ethernet,spectralefficiency(SE)needs to be increased to meet the bandwidth requirements of next-generation optical transmission networks[1]-[4].Recent experiments have shown that a polarization-division multiplexing quadrature phase-shift keying(PDM-QPSK)transmission system can achieve a maximum SEof 4 bit/s/Hz[5],[6].To further increase SE,we can combine multilevel modulation formats such as 16 quadrature-amplitude-modulation(16-QAM),32-QAM,or 64-QAM with PDM.These modulation formats carry more than 4 bits per symbol.However,multilevel modulation not only requires larger optical signal-to-noise ratio(OSNR),it is more sensitive to nonlinear propagation impairments and laser phase noise.Thus,as a tradeoff,PDM 16-QAM that carries 8 bits per symbol is a promising candidate for improving SE.The potential of PDM 16-QAM has already been shown in both simulations[7]-[9]and experiments[10].

Two different schemes have been proposed for achieving very high SEwith PDM 16-QAM.The first scheme uses coherent optical orthogonal frequency-division multiplexing(CO-OFDM)[11].In experiments carried out in[12],a 485 Gbit/s CO-OFDM superchannel with PDM 16-QAM transmitted over 1600 km ultralarge-area fiber(ULAF)and standard single-mode fiber(SSMF)links.The second scheme is Nyquist wavelength-division multiplexing(N-WDM),which is based on optical pulses that have an almost-rectangular spectrum,and the bandwidth is ideally equal to the baud rate[13].Although these two schemes have the potential to perform the same,CO-OFDM requires channels to be synchronized,and the analog-to-digital converters(ADCs)need large receiver bandwidth.In practical implementation,N-WDM is much more robust to receiver constraints[14].Therefore,a combination of N-WDM and PDM 16-QAM is a promising solution for future large-capacity,high-SEopticaltransmission systems and networks.

In conventional direct-detection receivers,linear distortion caused by fiber chromatic dispersion(CD)in the optical domain is converted into a nonlinear distortion in the electrical domain.Using a linear baseband equalizer based on one received baseband signal only improves performance a little.In coherent-detection receivers,distortion caused by CD is converted linearly into the electrical domain.This explains why,in the case of CDonly,fractionally spaced equalizers with complex coefficients can potentially extend the system reach to distances that are only limited by the number of equalizer taps[15].However,when CDis ideally compensated,nonlinear propagation impairments and laser phase noise(which PDM 16-QAM is very sensitive to)eventually limits the maximum achievable transmission distance.Most of the complexity of building an equalizer in a coherent receiver can be avoided by adopting pre-equalization at the transmitter,where the data is stillin its uncorrupted form.

The large constellation size of PDM 16-QAM also makes the system sensitive to transmitter impairments such as nonlinear drive in the optical modulator and imbalance between the frequency responses of the in-phase(I)and quadrature(Q)channels.These cause signaldistortion and deteriorate system performance.Transmitter impairments in PDM 16-QAM can be precompensated with pre-equalization,which is not difficult to implement in the digital-to-analog converter(DAC)at the transmitter.Some studies have focused on electronic pre-equalization,which is now a well-known technique in optical communication[16]-[18].A further benefit of pre-equalization is that the transmitted spectrum can be optimized using Nyquist-band preshaping pulses,which allow narrower channel spacing and higher SE[18].

In this paper,we use PDM 16-QAM to generate a 16 Gbaud signalin an N-WDM channel on a 16 GHz grid.In section 2,we describe the principle of pre-equalization and compare bit-error ratio(BER)of a single channel with that of an N-WDM channel when both channels have Nyquist-band pre-equalization or no Nyquist-band equalization.Pre-equalization can compensate for transmitter impairments and reduce the effects of nonlinear propagation and laser phase noise.N-WDM with Nyquist-band pre-equalization is tolerance of narrowband filtering and crosstalk caused by adjacent channels.In section 3,we describe the setup of an experiment for generating and transmitting 6×128 Gbit/s N-WDM PDM 16-QAM signals over 1200 km SMF-28.Amplification is provided by an erbium-doped fiber amplifier(EDFA)only.In this experiment,spectral efficiency(SE)of 7.47 bit/s/Hz was achieved.To the best of our knowledge,this the highest SEever achieved for a signalwith a bit rate beyond 100 Gbit/s and using PDM 16-QAM.The BERfor all channels(with the average OSNRof 23.6 d B)is smaller than the forward-error-correction(FEC)limit of 3.8×10-3over 1200 km SMF-28 transmission.Section 4 concludes the paper.

▲Figure 1.The principle of pre-equalization.

2 Pre-Equalization

2.1 Principle

Instead of only building the equalizer at the receiver,the equalizer can also be built at the transmitter.Compared with a binary phase-shift keying(BPSK)signal,a high-level signal has much more uncontrolled nonlinearity because of imperfections in the DAC,electrical amplifier(EA),in-phase/quadrature modulator(I/Q MOD),optical filter,and ADC.Therefore,we first transmit a BPSK signal in the back-to-back(B2B)link in order to calculate the transfer function.This function is then used to pre-equalize the high-level signal and reverse the channeldistortion.

Fig.1 shows the principle for pre-equalization.One of the parallel Mach-Zehnder modulators(MZMs)in an I/Q MOD is driven by a 16 Gbaud binary signal so that BPSKis generated.This signal has 1.5×samples and a word length of 215-1,and it is generated by an arbitrary waveform generator(AWG).An electricallowpass filter(LPF)with 3 d B bandwidth of 7.5 GHz is used to suppress out-of-band noise from the AWG,which operates in the interleaver mode with a sample rate of 24 GSa/s.The continuous wavelength lightwave(CW1)generated by an external cavity laser(ECL)with a linewidth of less than 100 kHz and output power of 14.5 d Bm is used as both the signal source and the local oscillator(LO)source in self-homodyne coherent detection.Then,the optical BPSK signal passes through a wavelength selective switch(WSS)with a 12 GHz passband before coherent detection.A real-time scope with 3 d B bandwidth of 16 GHz captures the detected electrical signal,which is used to calculate the transfer function of the transmitter in the frequency domain.The transfer function is then used to pre-equalize the 4-level signal with a word length of 215-1.This signal is used to generate the optical 16-QAM via the I/Q MOD.A raised-cosine(R-C)filter with a roll-off factor of 0.99 is used to pulse shape the 4-level signal.Pre-equalization of the I and Q output of the AWG makes the performance of both outputs similar to each other;therefore,we simply choose the Ioutput of the AWG to implement pre-equalization in our experiment.

▲Figure 2.Optical spectra(0.02 nm resolution)of the pre-equalized 16 Gbaud PDM 16-QAMsignal.

The WSShas a 12 GHz passband and is used for pre-equalization.This WSSis different from the WSSwith 10 GHz passband used for the WDM channel in the next section.The reason for setting WSSat a different bandwidth for pre-equalization and for WDM channel shaping is to balance the pre-equalization effect and crosstalk from the neighboring channels.After pre-equalization,we show by experiment the generation and transmission of the 6×128 Gbit/s N-WDM PDM 16-QAM signal over 1200 km SMF-28 with EDFA-only amplification.The spectral efficiency is 7.47 bit/s/Hz if we assume 7%FEC overhead.

2.2 Experimental Results and Discussion

Fig.2 shows the optical spectra for 16 Gbaud PDM 16-QAM in the case of pre-equalization with and without Nyquist band.Nyquist band is the definition of the WSSwith a 12 GHz passband.Compared to pre-equalization without Nyquist band(Fig.3,dotted line),pre-equalization with Nyquist band results in a PDM 16-QAM optical spectrum that has the function of a Nyquist-like filtering profile.This can sufficiently compensate for narrowband filtering effects.After passing through the 10 GHz WSS,the PDM 16-QAM optical spectrum with Nyquist-band pre-equalization is much narrower.

Fig.3 shows the measured electricalspectrum for the generated 16 Gbaud 4-level signal from the AWG in the case of pre-equalization with and without Nyquist band.Certain high-frequency components that are lost because of aggressive spectral filtering are pre-recovered.The 5 d B dip at 3 GHz in both traces is the reflection of the pre-distortion caused by the 7.5 GHz LPFshown in Fig.1.

3 Generation and Transmission of a 6×128 Gbit/s N-WDM PDM 16-QAM Signal

3.1 Experimental Setup

Fig.4 shows the experimental setup for generating and transmitting a 6×128 Gbit/s N-WDM PDM 16-QAM signal.The two 16 Gbaud electrical 16-QAM signals are generated from AWG1and AWG2.CW1has a frequency spacing of 1550.10 nm at 48 GHz,and CW2has a frequency spacing of 0.384 nm at 48 GHz.Both are generated by two ECLs,each with a linewidth of less than 100 kHz and output power of 14.5 d Bm.Two I/Q MODs are used to modulate CW1and CW2with the Iand Q components of the 64 Gbit/s,16 Gbaud electrical 16-QAM signals.This occurs after power amplification using four broadband electricalamplifiers(EA).To generate an optical 16-QAM signal,the two parallel MZMs in an I/Q MOD are biased at the null point and driven at full swing to achieve zero-chirp,0-phase andπ-phase modulation.The phase difference between the upper and lower branches of the I/Q MOD is controlled atπ/2.After boosting the power with polarization-maintaining EDFAs(PM-EDFAs),PDM for each path is realized by the polarization multiplexer.This comprises a polarization beam splitter(PBS),which halves the signal,an optical delay line(DL),which provides a delay of 150 symbols,and a polarization beam combiner(PBC),which recombines the signal.In the upper path,the optical PDM 16-QAM signal is then halved again by a polarization-maintaining optical coupler(PM-OC1).Here,the signal passing through the upper branch is handled by an intensity modulator(IM1)that is driven by a 16 GHz sinusoidal radio frequency(RF)signal and that is DC-biased at the null point.The signal passing through the lower branch is handled by DL3.So do the operation for the lower path.IM1 and IM2 are used for optical carrier suppression(OCS)modulation[19];in other words,the function of IM1 and IM2 is to produce two copies of the original signal that are+16 GHz and-16 GHz relative to the original signal center.The four branches,with the uppermost and lowermost branches each including two subcarriers,are spectrally filtered and combined using a programmable 4-channel WSSwith a 10 GHz passband on a 16 GHz grid.

▲Figure 3.Electricalspectrum of the pre-equalized 16 Gbaud 4-level(in-phase of 16-QAM)signal considering and not considering the Nyquist-band.

The insertion loss of the WSSis 7 d B.

The WDM signals are launched into the recirculating loop of 5×80 km SMF-28 with three circles.Each span has an average loss of 18 d Band chromatic dispersion of 17 ps/km/nm at 1550 nm when there is no opticaldispersion compensation.EDFA is used to compensate for the loss of each span.The totallaunched power(after EDFA)into each span is 10 d Bm,which corresponds to approximately 1 d Bm per channel at 128 Gbit/s.A tunable optical bandpass filter(BOF)with a bandwidth of 1.27 nm is used in the loop to remove the ASEnoise from each circle of the recirculating loop.At the receiver,a tunable optical filter(TOF)with 3 d B bandwidth of 0.35 nm is used to choose the desired channel.An ECLwith a linewidth of less than 100 kHz is used as the LO.A polarization-diverse 90 degree hybrid is used for polarization-and phase-diverse coherent detection of the LO and received opticalsignal prior to balanced detection.The digital scope performs ADC at 50 GSa/s with an electrical bandwidth of 9 GHz.

In the DSP,electrical polarization recovery is performed using a three-stage blind equalization scheme.First,the clock is extracted using the square-and-filter method,and the digitalsignal is resampled at twice the baud rate of the recovered clock.Second,a T/2-spaced time-domain finite impulse response(FIR)filter is used to compensate for CD.The filter coefficients are calculated using the known fiber CD transfer function and the frequency-domain truncation method.Third,two complex 13 tap,T/2-spaced adaptive FIRfilters are used to retrieve the modulus of the 16-QAM signal.

The two adaptive FIRfilters are based on the classic constant modulus algorithm(CMA)and are followed by a three-stage CMAfor multimodulus recovery and polarization demultiplexing.Carrier recovery is performed in the subsequent step.The 4th power is used to estimate the frequency offset between the LO and the received optical signal.Phase recovery is achieved by using feed-forward and least-mean-square(LMS)algorithms for offset compensation.Finally,differential decoding is used to calculate the BERafter the decision.

3.2 B2BExperimental Results and Discussion

Fig.5 shows B2B BERversus OSNRfor 16 Gbaud PDM 16-QAM at the N-WDM channel of 1550.10 nm.The scope has a bandwidth of 9 GHz and a sample rate of 50 GSa/s.There are four different cases:B2B single channel with and without Nyquist-band pre-equalization,and N-WDM PDM 16-QAM with and without Nyquist-band pre-equalization.

In the cases involving WDM,the six-channel 16 Gbaud PDM 16-QAM is on a 16 GHz grid,and all the channels of the previously mentioned four cases pass through the WSSwith a 10 GHz passband.In the case of Nyquist-band pre-equalization,the required OSNRat the BERof 3.8×10-3is 20.6 d B for a single channel.For the corresponding WDM case,the required OSNRpenalty can be neglected because pre-equalization can precompensate for transmitter impairments as and reduce the effects of nonlinear impairments and laser phase noise.Our experiments indicate that 10 GHz filtering bandwidth with 12 GHz Nyquist-band pre-equalization is narrow enough to avoid crosstalk from adjacent channels.The required OSNRpenalty at BERof 1×10-3increases to 1 d B when the ADC bandwidth is 16 GHz.The required bandwidth of the ADC plays an important role in detecting the N-WDM PDM 16-QAM signal.An ADC with bandwidth of half the baud rate frequency is beneficial for suppressing the noise and neighboring channel signal.For a single channel that passes through the 10 GHz WSSand has no Nyquist-band pre-equalization,the required OSNRat BERof 3.8×10-3is 24 d B,and the required OSNRpenalty is approximately 4 to 5 d B.PDM 16-QAM is quite sensitive to narrow filtering effects and noise.For N-WDM without Nyquist-band pre-equalization,there is an extra 1.5 d BOSNRpenalty caused by crosstalk from adjacent channels.

▲Figure 5.B2BBERversus OSNRfor a 16 Gbaud PDM 16-QAM signal atthe N-WDMchannel of 1550.10 nm.

▲Figure 6.Opticalspectrum(0.1 nm resolution)before and after 1200 km SMF-28 transmission with EDFA-only amplification,and(a)the obtained signal constellations before transmission and(b)the obtained signalconstellations after transmission.

3.3 Experimental Results for Transmission over 1200 km SMF-28

Fig.6 shows the opticalspectra before and after 1200 km SMF-28 transmission with EDFA-only amplification.The corresponding constellations are shown in Fig.6(a)and(b).The OSNRof the signal is 23.6 d B after transmission over 1200 km SF-28.Fig.7 shows BERversus power launched into each fiber span for an N-WDM channel at 1550.10 nm after 1200 km SMF-28 transmission.Launch power of-1 d Bm gives the best BERperformance.Fig.8 shows BER versus transmission distance for an N-WDM channel at 1550.10 nm after transmission.The launched power of-1 d Bm per channel is optimal.After transmission over 1200-km SMF-28,the BERfor each of the six channels(with average OSNRof 23.6 d B)is smaller than the FEC limit of 3.8×10-3(Fig.9)[20].

Figure 7.?BERversus launch power for an N-WDMchannel at 1550.10 nm after transmission over 1200 km SMF-28.

Figure 8.?BERversus transmission distancefor an N-WDM channelat 1550.1 nm.

Figure 9.?BERfor each of the6 channels aftertransmission over 1200 km SMF-28 atthe optimal launch power.

4 Conclusion

We have experimentally demonstrated the generation and transmission of a 6×128 Gbit/s N-WDM PDM 16-QAM signal on a 16 GHz grid over 1200 km SMF-28 with EDFA-only amplification.The net SEis 7.47 bit/s/Hz,which to the best of our knowledge is the highest SEfor a signal with a bit rate beyond 100 Gbit/s using PDM 16-QAM.Such SEwas achieved by DSPpre-equalization of transmitter-side impairments and DSPpost-equalization of channel and receiver-side impairments.The benefit of Nyquist-band pre-equalization scheme has been shown experimentally.BERfor all channels is smaller than the FEC limit of 3.8×10-3over a 1200 km SMF-28 transmission link.