DigitalSignalProcessing ssing for OpticalAccess ccess Networks works

Jianjun Yu(Optics Labs，ZTE(TX)Inc.，NJ 07960，USA)

DigitalSignalProcessing ssing for OpticalAccess ccess Networks works

Jianjun Yu
(Optics Labs，ZTE(TX)Inc.，NJ 07960，USA)

In this paper，we investigate advanced digital signal process⁃ing(DSP)at the transmitter and receiver side for signal pre⁃equalization and post⁃equalization in order to improve spec⁃trum efficiency(SE)and transmission distance in an optical access network.A novel DSP scheme for this optical super⁃Nyquist filtering 9 Quadrature Amplitude Modulation(9⁃QAM)like signals based on multi⁃modulus equalization with⁃out post filtering is proposed.This scheme recovers the Ny⁃quist filtered Quadrature Phase⁃Shift Keying(QPSK)signal to a 9⁃QAM⁃like one.With this technique，SE can be increased to 4 b/s/Hz for QPSK signals.A novel digital super⁃Nyquist signal generation scheme is also proposed to further suppress the Nyquist signal bandwidth and reduce channel crosstalk without the need for optical pre⁃filtering.Only optical cou⁃plers are needed for super⁃Nyquistwavelength⁃division⁃multi⁃plexing(WDM)channelmultiplexing.We extend the DSP for short⁃haul optical transmission networks by using high⁃order QAMs.We propose a high⁃speed Carrierless Amplitude/Phase⁃64 QAM(CAP⁃64 QAM)system using directly modulated la⁃ser(DML)based on direct detection and digital equalization. Decision⁃directed least mean square is used to equalize the CAP⁃64QAM.Using this scheme，we generate and transmit up to 60 Gbit/s CAP⁃64QAMover 20 km standard single⁃mode fiber based on the DML and direct detection.Finally，several key problems are solved for real time orthogonal⁃fre⁃quency⁃division⁃multiplexing(OFDM)signal transmission and processing.With coherent detection，up to 100 Gbit/s 16 QAM⁃OFDMreal⁃time transmission is possible.

digital signal processing;high spectrum efficiency;super⁃Ny⁃quist;coherent optical transmission

1 Introduction

I n long⁃haul backbone networks and short⁃haul access networks，bandwidth demand has increased by 30%to 60%annually due to the rapid development of cloud computing，social media，and mobile data services. The trend of increasing service bandwidth requires a lower cost per bit;thus，high⁃speed optical transmission interfaces and high spectrum efficiency(SE)technologies have become more important.For example，a successful solution for the 100 Gbit/s long⁃haul system is to combine the single⁃carrier Polar⁃ization⁃Division Multiplexing Quadrature Phase Shift Keying (PDM⁃QPSK)modulation formatand the digital signal process⁃ing(DSP)based coherent detection.Transmission technologies providing high SE have also been widely investigated.These technologies fall into two main categories:those that reduce spectrum bandwidth and those that increase themodulation or⁃der.The former uses spectrum shaping technologies of optical or electrical domain filtering，which are also called Nyquistor Super⁃Nyquist technologies.The latteruseshigher⁃ordermodu⁃lation formats，such as 32 Quadrature Amplitude Modulation (32⁃QAM)，64⁃QAM，or even the QAMs thathave a higher or⁃der[1]-[26].However，the two technologies depend on the ad⁃vanced DSP of transmitters or receivers，and there are various restrictions such as high sensitivity to laser frequency offset，phase noise，and inter⁃symbol interference(ISI).There is also a variety of intra⁃channeland inter⁃channel impairments.

Higher⁃ordermodulation is the simplestsolution to high SE. This solution，however，causes impairment，requires high re⁃ceiver sensitivity and only covers a short distance.Compared with Quadrature Phase⁃Shift Keying(QPSK)signals，16⁃QAMsignals require 6 dB higher Optical Signal to Noise Ratio(OS⁃NR)，and the OSNR requirement increases exponentially with the increase of constellation points.As for optical fiber trans⁃mission，the nonlinearity of optical fibers restricts the launch power and significantly limits the OSNR.In addition，the Eu⁃clidean distance of constellation points in high SEmodulation is shorter，and there is even lower tolerance for the nonlineari⁃ty of optical fibers.In the latestexperiment[27]，the OSNR at BER=10⁃3for 16⁃QAMhas a penalty of 8 dB whereas for QPSK only a penalty of 1 dB.Therefore，it is very important to study the advanced DSPalgorithms for higher⁃orderQAM.Un⁃like QPSK，higher⁃order QAMneeds new solutions to the po⁃larization de⁃multiplexing，frequency offsetand phase recovery ofsignals.In field tests，dual⁃carrier 16⁃QAMsignalsare trans⁃mitted at512 Gbit/s for amaximum distance of 734 km in dis⁃persion⁃compensated fibers，while at the same time non⁃return⁃to⁃zero(NRZ)signals are co⁃propagated at 10 Gbit/swithin a 200GHzbandwidth[28].These resultsshow that it ischalleng⁃ing to use 16⁃QAMand 64⁃QAMto increase the SE.Due to smaller electrical bandwidth at the same bit rate，high SE transmission based on higher⁃order QAMcan bring a better system performance in a short⁃haul optical network that re⁃quiresahigherOSNR.

With the development of high SE coherent detection and DSP，spectrum shaping technologies of NyquistWavelength⁃Division Multiplexing(N⁃WDM)and Super⁃NyquistWDM(SN⁃WDM)have become hot topics in the field of 100 Gbit/s long⁃haul transmission.Current research shows thatQPSKmodula⁃tion formatsprovide thebestbalancebetween SE and transmis⁃sion distance.Therefore，using spectrum shaping technologies to implementN⁃WDMor SN⁃WDMfor increasing the SE of the PDM⁃QPSK system is a promising and highly efficient solution for long⁃haul large⁃capacity optical transmission networks [11]-[21].However，filter shaping and DSPmay cause ISI，in⁃ter⁃channel crosstalk and noise amplification，allofwhich seri⁃ously affect system performance[11]-[16].When a linear equalization algorithm，such as the Constant Modulus Algo⁃rithm(CMA)，is used，high⁃frequency noise and inter⁃channel crosstalk in the signal spectrum is enhanced.To compensate for the impairment，extra processing is needed for noise sup⁃pression and multi⁃symbol detection decision.In[11]-[16]，a delay⁃and⁃add post⁃filter is used to suppress the enhanced noise.In addition，a 1⁃bitMaximum Likelihood Sequence Esti⁃mation(MLSE)is introduced to overcome the ISI impairment. However，some problems stillexist.

1)Although the post⁃filter combined with Constant Modulus Equalization(CMEQ)algorithm has been widely applied to 100 Gb/s and higher optical Nyquist and Super⁃Nyquist transmission[14]-[16]，some DSPmodules，including carri⁃er recovery，may stillbe affected by noise and crosstalk.

2)A Wavelength Selective Switch(WSS)，which is costly and difficult to integrate，cannotbe easily integrated with a tra⁃ditional optical transceiver，especially in amulti⁃channel system.

3)The instability of the filter centerwindowmay cause serious

deterioration of the system performance.

To solve these problems，we use a digital⁃to⁃analog convert⁃er(DAC)with a high sampling rate and high analog bandwidth. ThisDACshapes the spectrum through a digital filterwith hun⁃dredsof taps.This solution doesnotneed any extra device and parameters can be reset easily.It can be combined with other functions(such as tilt correction)of a transmitter，and WDMchannels can also bemultiplexed by optical couplers.

In thispaper，we study DSPalgorithms forhigh SE and long⁃haul transmission in opticalaccessnetworks.We describe post⁃processingat the receiverend and pre⁃processing at the trans⁃mitter end in an SN⁃WDMsystem.We propose a 9⁃QAM⁃like Multi⁃Modulus Equalization(MMEQ)DSP scheme based on optical super⁃Nyquist filtering.By using the Cascaded Multi⁃Modulus Equalization algorithm(CMMA)，this scheme directly restores QPSK signals from 9⁃QAM⁃like signals[29]-[30].In addition，for Quadrature Duobinary(QDB)signals，the system performanceofpost⁃filterCMEQ and MMEQ schemesare com⁃pared for different filter bandwidths，carrier spacing，and trans⁃mission distances.We transmit QPSK signals with an SE of 4 b/s/Hz.We also propose a novel digital Super⁃Nyquist signal generation scheme that reduces Nyquist signal bandwidth and channel crosstalk withoutusing optical pre⁃filtering.The spec⁃trum of Super⁃Nyquist 9⁃QAMsignals generated by using this scheme ismore effectively compressed than thatof regular Ny⁃quist QPSK signals.Only an optical coupler is needed to im⁃plementWDMchannelmultiplexing.After the 20%soft⁃deci⁃sion Forward Error Correction(FEC)overhead is removed，this scheme stillachievesa netSE of4 b/s/Hz.By using higher⁃or⁃der QAMtechnologies，we also apply DSP in short⁃hauloptical transmission networks.We propose a high⁃speed Carrierless Amplitude/Phase⁃64 QAM(CAP⁃64⁃QAM)system based on Directly Modulated Laser(DML)，direct detection，and digital equalization.CAP⁃64⁃QAMsignal equalization is implement⁃ed with the Decision Directed Least Mean Square(DD⁃LMS) algorithm.By using this scheme，we generate high⁃speed CAP⁃64⁃QAMsignals based on DML and direct⁃detection technolo⁃gies.We transmit the signals over a 20 km Standard Single⁃Mode Fiber(SSMF)ata record rate of60Gbit/s.Thispaperal⁃so describes the latest research progressof the DSP⁃based real⁃time coherentsystem.

The remainder of this paper is organized as follows.In sec⁃tion 2，we describe theMMEQ⁃based post⁃transmission DSPal⁃gorithms used in the Super⁃Nyquist system.In section 3，we discuss a novel digital Super⁃Nyquist signal generation scheme.This scheme reduces channel crosstalk without using optical pre⁃filtering.In section 4，we show the 64⁃QAMsignal transmission experiments based on Carrierless Amplitude/ Phase(CAP)algorithms.We also discuss the DSP⁃based real⁃time coherent system in section 5，and conclude the paper in section 6.

2 MMEQ-Based Post-Processing DSP Algorithm sUsed in Optical Super-Nyquist Channels

Optical super⁃Nyquistshaping can be implemented by 4⁃or⁃der super⁃Gaussian narrowband filtering，such asWSS[11]-[16].For PDM⁃QPSK signalswith a symbol rate of R s，we use a filter shaping devicewith 3 dB bandwidth less than or equal to R s in order to implementQDB spectrum shaping.Due to fil⁃teringeffect，4⁃pointQPSK signalscan be changed into 9⁃QAM⁃like signals from the viewpoint of constellation points.Com⁃pared with QPSK signals，QDB signals have a narrower spec⁃trum and side lobe suppression.In general，conventional Ny⁃quist signals are generated by a raised cosine function thathas a bandwidth equal to symbol rate.We propose a solution to reach the limit of the Super⁃Nyquist SE by adopting a filter with a 3 dB bandwidth less than signalbaud rate.Fig.1 shows the principle for the generation ofWDMchannelsof Super⁃Ny⁃quist filtering 9⁃QAM⁃like signals from QPSK signals，which is based on opticalGaussian filtering.

Fig.2 shows the DSPmodule process.Figs.2a and 2b showthe DSPmodule processes of two differentprocessing schemes. In[11]-[16]，the authorsuse the CMA and post⁃filterCMEQ al⁃gorithm.[29]and[30]introduce the MMEQ scheme which we have proposed recently.For the post⁃filterCMEQ，received sig⁃nals are first restored to QPSK signals and then converted into 9⁃QAM⁃like signalsby delay⁃and⁃add post⁃filter⁃ing to suppress noise.In the MMEQ scheme，however，we use the CMMA algorithm to restore QDB signals to 3⁃modulus 9⁃QAM⁃like signals and then obtain 9⁃QAMsignals by using an im⁃proved carrier phase recovery(CPR)algorithm. The detailed DSP algorithm is discussed in[29]，and Fig.2 shows the signal constellations after the processing of each DSP module.The main advantage of using the MMEQ algorithm to pro⁃cess QDB filtering signals is that the frequency response ofadaptive MMEQ tapshasa compres⁃sion effect on the high⁃frequency components，which，compared with CMEQ，avoids performance deteriora⁃tion caused by noise and crosstalk.Fig.2c shows the signal spectra of differentschemeswhen a 3 dB 22GHzQDB filter is used.Noise and crosstalk are improved，and the high⁃frequen⁃cy components near±R s/2 are restored by the CMEQ algo⁃rithm.However，noise and crosstalk are suppressed by the pro⁃cessing of the CMMA⁃based MMEQ algorithm.This sup⁃pressed noise ismainly the noise inside the channel near the high⁃frequency partand includes Amplified SpontaneousEmis⁃sion(ASE)noise and ISI.Therefore，with the CMEQ⁃based scheme，it is necessary to add a post⁃filter after the CPR pro⁃cess in order to suppressnoise and crosstalk[11]-[16].Howev⁃er，some DSPmodules are still affected by noise and crosstalk during CPR.Compared with the CMEQ algorithm，the MMEQ algorithm better suppresses noise and crosstalk during the ini⁃tialphaseofDSP，and improvessystem performance.

▲Figure1.Generation ofWDMchannelsof super⁃Nyquist filtering 9⁃QAM⁃likesignals from QPSK signals.

▲Figure2.DSPModule Process.

To compare the filter tolerance for noise and crosstalk of CMEQ and MMEQ algorithms，we design an 8⁃channel PDM⁃QPSK experimentwith 28⁃Gbaud QDB filtering.The transmis⁃sion rate of the system is 8×112Gbit/s，and the channel spac⁃ing is25GHz.Single⁃mode fiber⁃28(SMF⁃28)isused，and the circulating fiber loop isdivided into ten 88 km spans.Theaver⁃age lossofeach span is 18.5 dB and chromatic dispersion(CD) is 17 ps/km/nm.An EDFA is added before each 88 km fiber span to compensate for the fiber loss.In addition，a programma⁃ble WSS is introduced into the fiber loop to suppress ASE noise as an opticalbandpass filter(BPF).TheWSShasa 4⁃or⁃der Gaussian spectral featurewith a 3 dB bandwidth of2.2 nm. At the receiver end，a tunable BPFat 3 dB bandwidth of 0.34 nm is used to select the desired subchannel.Polarization⁃and phase⁃diversity homodyne coherent detection is also adopted. The External Cavity Laser(ECL)functioning as the Local Os⁃cillator(LO)in the transmitter or receiver has a linewidth of about100 kHz.Each Balanced Photodiode(BPD)iswith 3 dB bandwidth of 42 GHz.The average inputoptical power ofeach photodiode in BPD ranges from⁃20 dBm to 13 dBm.The pow⁃er of received signals is 3 dBm，and the power of the pre⁃ampli⁃fied LO before an opticalmixer is 20 dBm.A digital sampling oscilloscopewith a sampling rate of80GSa/sand bandwidth of 30 GHz is used for analog⁃to⁃digital conversion(ADC).Thecrosstalk ofadjacent channels issuppressed after the ADC and there isnoneed toadd an extra filter in offline processing.

The result shows that MMEQ provides a better BER than CMEQ with post⁃filtering because the formermore effectively suppresses noise and crosstalk.When the filter bandwidth is 20.1 GHz，the OSNR corresponding to BER=1 x 10⁃3in the MMEQ scheme isabout16.5 dB，which improvesby 1 dB com⁃pared with the post⁃filtering CMEQ scheme.In addition，the MMEQ scheme improves the filter tolerance for noise and crosstalk，and themaximum transmission distance of 25 GHz QDB signals can reach 2640 km.For post⁃filtering CMEQ，however，themaximum transmission distance at the BER be⁃low the FEC limit is about2000 km.Therefore，compared with the post⁃filtering CMEQ scheme，the MMEQ scheme provides better transmission and increases transmission distance by 32%when the BER is3.8×10⁃3.

3 Pre-Processing Algorithm for Generation and Processing of Digital Super-Nyquist Signal

Fig.3 shows the difference between the generation of DAC⁃based Super⁃Nyquist 9⁃QAMsignals and that of regular Nyquist QPSK signals.For the filtering of regular Ny⁃quist signals，only Square⁃Root⁃Raised⁃Co⁃sine(SRRC)filter is needed to generate Ny⁃quist pulses.However，when the channel spacing is less than the symbol rate，the bandwidth beyond the channel spacingmay cause serious crosstalk(Fig.3).To achieve Super⁃Nyquist transmission，we add a low pass filter(LPF)to generate super⁃Nyquist pulses.In this way，the signal spectrum is further compressed to reduce channel cross⁃talk.In our scheme，the LPF can be imple⁃mented by the QDB delay⁃and⁃add，and the z⁃transform of the related transfer function is

Thus，the LPF can be implemented by a 2⁃tap FIR filter，which converts QPSK signals into 9⁃QAMsignals[11]-[16].By cascading QDB and SRRC filters，the Super⁃Nyquist digital filter in the time domain can be ex⁃pressed as

where hsrrc(t)is the time domain pulse re⁃sponse of the SRRC filter[17]-[20]，and hQDB(t)is the impulse response of the QDB filter in(1).

Figs.4a and 4d show the time domain pulse responses ofan SRRC⁃based regular Nyquist filter and a super⁃Nyquist filter based on cascaded QDB and SRRC filters. In the figure，the SRRC roll⁃off factor isset to 0.The Super⁃Ny⁃quist digital filter has a smaller resonance and faster conver⁃gence compared with the conventional Nyquist filter.Figs.4b and 4e are theeye diagramsofa conventionalNyquistQPSK 2⁃level baseband signal and a Super⁃Nyquist 9⁃QAM3⁃level baseband signal.Figs.4c and 4f show the electrical power spectra of Nyquist QPSK and Super⁃Nyquist 9⁃QAMsignals. The power spectrum of the Super⁃Nyquist signal ismore se⁃verely compressed;its spectrum side lobe is greatly sup⁃pressed;and its 3 dB bandwidth is less than one half of the baud rate.

4 CAP-64-QAMShort-Haul Transm ission Using Direct Detection and Advanced DigitalEqualization Technologies

▲Figure3.DAC⁃based Nyquistand Super⁃Nyquist9⁃QAMsignalgeneration.

▲Figure4.Comparison of Nyquistand super⁃Nyquist.

Because of the smaller electrical bandwidth at a given bit rate，high SE transmission based on higher⁃order QAMcan bring about better system performance in a short⁃haul opticalnetwork that requires a higher OSNR.On the other hand，with the rapid growth in the demand for short⁃haul communication bandwidth of optical links between access networks and data centers，to increase the transmission capacity isa hot topic[5]，[6].Considering costand complexity，intensitymodulation and direct detection(IM/DD)using higher⁃order modulation for⁃mats isauniversally feasible solution[5]，[6]，[31]-[41].IM/DD⁃basedmodulation technologies such asQAM⁃subcarriermodu⁃lation(SCM)[5]，[6]，pulse amplitudemodulation(PAM)[31]，Discrete Multi⁃Tone(DMT)or orthogonal frequency division multiplexing(OFDM)[32]，[33]，and CAP modulation[34]-[41]havebeen proposed.

Studieshave shown that the IM/DD⁃based CAP structure re⁃duces complexity and ensures good performance.It is still able to provide a high data transmission rate by using only a DML，a Vertical⁃Cavity Surface⁃Emitting Laser(VCSEL)，a photo⁃electric devicewith a limited bandwidth，orother cheap compo⁃nents[34]-[42].Compared with QAM⁃SCM[5]，[6]and OFDM[32]，[33]，CAP doesnot require complex⁃to⁃real conversion in the electrical domain，complexmixers，RF sources or optical in⁃phase/quadrature(I/Q)modulators.CAP also eliminates the discrete Fourier transform(DFT)used during OFDMsignal modulation and demodulation[40].Various CAP⁃based optical communication systems have been discussed in[35]-[42].In [38]，the authors describe how multiband CAP⁃QAMcan in⁃crease the bandwidth in short⁃haul communication.G.Step⁃niak et al.[42]propose a system based on CAP⁃16⁃QAMand CAP⁃64⁃QAM.However，the bit ratesof these systemsare only 2 Gbit/s and 2.1 Gbit/s，respectively.In[40]，a CMMA⁃based digital equalizer is used to equalize the ISIof CAP⁃16⁃QAMsignals，and this improves performance.However，higher⁃order modulation CAP systems，such as CAP⁃64⁃QAMwith a rate up to tens of gigabit per second，have not been demonstrated，and the corresponding digital equalization technologies also have notbeen deeply investigated.Thus，we propose and experimen⁃tally demonstrate a high⁃speed CAP⁃64⁃QAMsystem based on DML，directdetection，and digitalequalization technologies.

▲Figure5.CAPM⁃QAMtransm ittersand receiversusing DML,directdetection, and digitalequalization.

Fig.5 shows the principles of CAP M⁃QAMtransmitters and receivers thatuse DML，directdetection，and digitalequal⁃ization.CAPBell Labs firstproposed CAP，amulti⁃levelmulti⁃dimensionalmodulation formatsuitable for short⁃haul commu⁃nications[34]-[43].Thismodulation format is similar to QAMbutdoes not require an RF source.Two⁃dimensional CAP can be implemented by the two orthogonal filters fIand fQin Fig.5. The original bit sequence is firstmapped to the complex sym⁃bol of M⁃QAM，where Mis the level of the QAM.In order to match the sampling rate of the shaping filter，themapped com⁃plex symbol is then up⁃sampled.The sampling rate of the shap⁃ing filter is jointly determined by the data baud rate and DAC sampling rate.After the outputs of the two filters are com⁃bined，DAC processing is performed to form S(t)，which drives DML.The receiver uses direct detection.After the ADC pro⁃cessing，the signals are sent to twomatching filters to separate the in⁃phase and quadrature components.After down⁃sam⁃pling，linear equalization and M⁃QAMdemodulation，the origi⁃nalbitsequence isobtained.

fI(t)and fQ(t)represent a pair of quadraturematching filters，and MfI(t)and MfQ(t)are their corresponding shaping filters. These two pairs of filtersmake up a Hilbert pair between the transmitter and receiver.The two quadrature filters can be es⁃tablished by multiplying the SRRC of sine and cosine func⁃tions[43].Therefore，the relationships between the matching filters are MfIn(t)=fIn(⁃t)and MfQn(t)=fQn(⁃t).Because of the or⁃thogonality of the filters，in⁃phase and quadrature data can be obtained with the quadraturematching filters.In order to accu⁃rately recover the in⁃phase and quadrature data，synchroniza⁃tion during CAP demodulation is very important.Time errors ofmatching Finite Impulse Response(FIR)filters introduce se⁃rious ISI[40]-[42].Themost suitable sampling point is diffi⁃cult to determine;therefore，the deviation of the sampling time point causes subsequent signals to be severely affected by ISI and I/Q crosstalk，and this results in blurring and phase rota⁃tion of constellation points.

Therefore，it isnecessary to usea linearequalizer after down⁃sampling to process complex signals，and the original signals can be obtained by QAMde⁃coding.In our system，both quadrature filters and matching filters are implementedby digi⁃tal FIR filters，and the tap lengths are T⁃OFL and R⁃MFL，respectively.The tap length of a FIR filter determines its time domain pulse shape and frequency response[40].Its im⁃pact on system performance is also consid⁃ered in theexperiment.

In the previous work，CAP signals are equalized by a two⁃stage equalization scheme that combines ISIequalization and phase re⁃covery algorithms.CMA is used first for pre⁃convergence，and then the CMMA algorithm is used for ISIequalization.For higher⁃orderCAP⁃QAMsignals，however，CMMA equalization is not ideal because the intervals in a QAMring are generally smaller than the minimum inter⁃symbol interval.Previous studies have shown that DD⁃LMS results in better SNR than CMMA for higher⁃order QAMsignals[22].On the other hand，the conver⁃gence of CMMA is based on the modulus value of a symbol，and this algorithm is independent of the phase.Therefore，phase recovery is additionally needed after CMMA in order to equalize crosstalk.

We propose a novel DSP algorithm for equalizing the ISI and crosstalk in CAP⁃QAMsignals.After CMA pre⁃conver⁃gence，a DD⁃LMSbased 1⁃levelequalizer is used to adjust the tap coefficientof the FIR filter.Figs.6a and 6b show the struc⁃ture and principle of the DD⁃LMS algorithm.The FIR filter used for CAP signalequalization isa butterfly⁃configured adap⁃tive digital filterwith a T/2 interval.Unlike DD⁃LMSused in a coherent optical system，the four time⁃domain tap coefficients of the FIR filter are all real numbers.ZI(n)and ZQ(n)respec⁃tively indicate the in⁃phase and quadrature signaloutputsafter the N th equalization of the filter.DI(n)and DQ(n)are the deci⁃sion resultsof in⁃phase and quadrature signals.Although the in⁃phase and quadrature signal inputs are independent，each out⁃put is related to both inputs.The error function of DD⁃LMS can beexpressed as

where eI(n)and eQ(n)are the error functions of in⁃phase and quadrature signals，respectively，and the four real⁃value FIR filters hii，hiq，hqi，and hqqare updated by the error function after decision:

This means that the ISI and crosstalk of in⁃phase and quadrature signals can be eliminated.

We compared the CMMA and DD⁃LMSschemes by simula⁃tion.Figs.6c and 6d show the CAP⁃64⁃QAMsignalequaliza⁃tion results.Fig.6c shows the relationship between the sam⁃pling deviation and phase rotation in the CMMA and DD⁃LMS equalization schemes，and the up⁃sampling rate is 8 Sa/sym⁃bol.The phase rotation introduced by the clock deviation can⁃not be compensated by CMMA.Extra phase recovery process⁃ing is necessary after CMMA.When the DD⁃LMSalgorithm is used，however，the phase of received signals can be correctly restored，because DD⁃LMS is very sensitive to phase informa⁃tion.Fig.6d shows the relationship between the Q value and SNR of received signals in differentequalization schemes.The results show that the Q value of CAP⁃64⁃QAMsignals is high⁃er in the DD⁃LMSalgorithm than in the CMMA algorithm，be⁃cause for CAP⁃64⁃QAMsignals，the error function of DD⁃LMS is based on the symbol interval whereas that of CMMA is based on the ring interval.For QAM，the ring interval is small⁃er than theminimum symbol interval in most cases;thus，the DD⁃LMSalgorithm performsbetter.

Theaforementioned resultsverify thatour proposed CAP⁃64⁃QAMsystem，which adopts DML，direct detection，and im⁃proved DD⁃LMSequalization，is very feasible.

5 DSP-Based Real-Time Coherent System

▲Figure6.Simulation comparison between CMMA and DD⁃LMSsolutions.

We have，for the first time，constructed a 100 Gbit/s single⁃band real⁃time coherent optical 16⁃QAM⁃OFDMtransmission system[44].The OFDMsignal has a high SE and its spectral resources can be dynamically allocated.It also can effectively resist chromatic dispersion(CD)in optical fiber transmission. Therefore，it is an advanced modulation format that has been continuously studied in the industry.Fig.7 shows the experi⁃mentsetup.The laser in the experimenthasan operatingwave⁃length of 1548.53 nm and a linewidth less than 100 kHz.After electrical amplification，16⁃QAM⁃OFDMsignals generated by DAC are used to drive the optical I/Q modulator.The DAC sampling rate is 62.895 GSa/s.In OFDMmodulation，the FFT size is1024，ofwhich 256 subcarriersare used to carry data，8 subcarriers are used to carry pilot signals，the first subcarrieriszero，and the other 759 subcarriersalso are all set to zero.In the experiment，the DFT⁃spread technology is used to evenly distribute SNR and reduce the peak⁃to⁃average power ratio (PAPR)in signal subcarriers.In addition，intra⁃symbol fre⁃quency⁃domain averaging(ISFA)isused to eliminate the influ⁃ence of noise in the optical channels in channel estimation. Thus，the BER of the system can be improved.After the in⁃verse fast Fourier transform(IFFT)，24 sampling points are used as cyclic prefix.At the receiver end，the ADC sampling rate is 41.93 GSa/s，and the bandwidth is 16 GHz.DAC and ADC resolutions are 8 bits and 6 bits，respectively.The FPGA chipmodels are Altera EP4S100G and Xilinx 6VSX475 FPGA [44].Fig.7 shows the photos of DAC and ADC，optical spec⁃trum and electrical spectrum of signals，and the flow chart of FPGA register transmission.In the absence of electronic dis⁃persion compensation(EDC)，the BER of 100 Gbit/s polariza⁃ tion divisionmultiplexing16⁃QAM⁃OFDMis less than 3.8×10⁃3after200 km transmission.

▲Figure7.100GHz16⁃QAM⁃OFDMreal⁃timesystem.

6 Conclusion

In this paper，we studied the high SE and long⁃haul trans⁃mission DSP algorithms in optical access networks.In order to improve the SE of QPSK signals，we proposed two Super⁃Ny⁃quistWDMalgorithmsbased on pre⁃processing at the transmit⁃ter end and post⁃processing at the receiver end.We also pro⁃posed and verified an MMEQ DSP scheme for optical super⁃Nyquist filtering 9⁃QAM⁃like signals.By using the CMMA al⁃gorithm，this scheme can directly restore QPSK signals from 9⁃QAM⁃like signals.For QDB signals，we compared the system performance of post⁃filter CMEQ and MMEQ schemes for dif⁃ferent filter bandwidths，carrier spacing，and transmission dis⁃tances.In addition，we proposed a novel digital super⁃Nyquist signal generation scheme that further compresses Nyquist sig⁃nal bandwidth and reduces channel crosstalk without the need for optical pre⁃filtering.The spectrum of super⁃Nyquist 9⁃QAMsignals ismore compressed than that of regular Nyquist QPSK signals，and only an optical coupler is needed to imple⁃mentsuper⁃NyquistWDMchannelmultiplexing，with a net SE up to 4 b/s/Hz(after removing the 20%soft⁃decision FEC over⁃head).We also expanded the higher⁃orderQAMDSPalgorithm to short⁃haul optical transmission networks and proposed and experimentally verified high⁃speed CAP⁃64⁃QAMsystems based on DML，direct detection，and digital equalization tech⁃nologies.DD⁃LMS is used for CAP⁃64⁃QAMsignal equaliza⁃tion.By using this scheme，we achieved 60 Gbit/s CAP⁃64⁃QAMtransmission over 20 km SSMF based on DML and direct detection.We firstachieved a 100 Gbit/s single⁃band polariza⁃tion division multiplexing 16⁃QAM⁃OFDMreal⁃time coherent optical transmission system.For the DSP algorithms of a real⁃time system，we proposed a novel solution，which simplifies the complex floating pointmultiplication into simple XOR op⁃erations by the comparison of only the symbol bits of signals. Thus，the algorithm complexity of time⁃domain synchronization and frequency offset estimation can be greatly reduced.In the experiment，distortion⁃less DFT⁃spread technology is used to reduce the PAPR of OFDMsignals.In addition，ISFA is used to eliminate the influence of noise in the optical channels in channel estimation.In thisway，the BER of the system can be improved.Our research provides a reliable alternative for real⁃time transmission in a local area network(LAN)with a channel rateup to100Gbit/s.

[1]B.Swanson and G.Gilder，“Estimating the exaflood—the impact of video and richmedia on the internet—a zettabyteby 2015，”Discovery Institute，2008.

[2]R.W.Tkach，“Scaling optical communications for the nextdecade and beyond，”Bell LabsTechnol.，vol.14，no.4，pp.3-9，2010.

[3]R.Essiambre and R.W.Tkach，“Capacity trends and limits ofoptical communi⁃cation networks，”Proc.IEEE，vol.100，no.5，pp.1035-1055，May 2012.doi: 10.1109/JPROC.2012.2182970.

[4]C.V.N.Inde，“Forecastand Methodology 2010-2015，”Cisco Systems，2011.

[5]K.Szczerba，B.E.Olsson，P.Westbergh，etal.，“37Gbit/s transmission over200 m ofMMF using single cycle subcarriermodulation and a VCSELwith 20 GHz modulation bandwidth，”36th Eur.Conf.Exhibition Opt.Commun.，Torino，Italy，2010，pp.1-3.doi:10.1109/ECOC.2010.5621209.

[6]A.S.Karar and J.C.Cartledge，“Generation and detection of a 56 Gbit/s signal using a DML and half⁃cycle 16⁃QAMNyquist⁃SCM，”IEEE Photon.Technol.，vol.25，no.8，pp.757-760，2013.

[7]K.Roberts，M.O’sullivan，K.T.Wu，et al.，“Performance of dual⁃polarization QPSK for optical transport systems，”Lightwave Technol.，vol.27，no.16，pp. 3546-3559，2009.

[8]J.Renaudier，O.Bertran⁃Pardo，H.Mardoyan，et al.，“Performance comparison of40G and 100G coherent PDM⁃QPSK for upgrading dispersionmanaged legacy systems，”OFC 2009，San Diego，USA.

[9]C.Xie，G.Raybon，and P.J.Winzer，“Transmission ofmixed 224⁃Gbit/sand 112⁃Gbit/s PDM⁃QPSK at 50⁃GHz channel spacing over 1200⁃km dispersion⁃man⁃aged LEAF®spans and three ROADMs，”Lightwave Technol.，vol.30，no.4，pp. 547-552，2012.

[10]J.X.Cai，C.R.Davidson，A.Lucero，etal.，“20 Tbit/s transmission over 6860 km with sub⁃Nyquist channel spacing，”Lightwave Technol.，vol.30，no.4，pp. 651-657，2012.

[11]J.Li，Z.Tao，H.Zhang，etal.，“Spectrally efficientquadrature duobinary coher⁃ent systems with symbol⁃rate digital signal processing，”Lightwave Technol.，vol.29，no.8，pp.1098-1104，2011.

[12]J.Li，E.Tipsuwannakul，T.Eriksson，M.Karlsson，and P.A.Andrekson，“Ap⁃proaching Nyquist limit inWDMsystems by low⁃complexity receiver⁃side duo⁃binary shaping，”Lightwave Technol.，vol.30，no.24，pp.1664-1676，2012.

[13]Z.Jia，J.Yu，H.Chien，Z.Dong，and D.Huo，“Field transmission of 100G and beyond:multiple baud ratesandmixed line ratesusing Nyquist⁃WDMtechnolo⁃gy，”Lightwave Technol.，vol.30，no.24，pp.3793-3804，2012.

[14]D.Dong，J.Yu，Z.Jia，et al.，“7×224 Gbit/s/ch Nyquist⁃WDMtransmission over 1600⁃km SMF⁃28 using PDM⁃CSRZ⁃QPSK modulation，”IEEE Photon. Technol.Lett.，vol.24，no.13，pp.125-129，2012.

[15]J.Yu，Z.Dong，H.Chien etal.，“Transmission of200 GPDM⁃CSRZ⁃QPSK and PDM⁃16 QAMwith a SE of 4 bit/s/Hz，”Lightwave Technol.，vol.31，no.5，pp. 515-522，2013.

[16]H.Chien，J.Yu，Z.Jia，Z.Dong，and X.Xiao，“Performance assessmentofnoise⁃suppressed Nyquist⁃WDMfor terabit superchannel transmission，”Lightwave Technol.，vol.30，no.24，pp.3965-3971，2012.

[17]J.Wang，C.Xie，and Z.Pan，“Generation of spectrally efficient Nyquist⁃WDMQPSK signals using DSP techniques at transmitter，”OFC 2012，Los Angeles，USA.

[18]B.Pardo，J.Renaudier，P.Tran，etal.，“Submarine transmissionswith spectral efficiency higher than 3 bit/s/Hz using Nyquist pulse⁃shaped channels，”OFC 2013，Anaheim，USA.

[19]Q.Juan，B.Mao，N.Gonzalez，N.Binh，and N.Stojanovic，“Generation of 28GBaud and 32GBaud PDM⁃Nyquist⁃QPSK by a DAC with 11.3GHz analog bandwidth，”OFC 2013，Anaheim，USA.

[20]J.Wang，C.Xie，and Z.Pan，“Generation of spectrally efficientNyquist⁃WDMQPSK signalsusing digital FIR or FDE filtersat transmitters，”Lightwave Tech⁃nol.，vol.30，no.24，pp.3679-3686，2012.

[21]J.X.Cai，“100G transmission over transoceanic distancewith high spectraleffi⁃ciency and large capacity，”Lightwave Technol.，vol.30，no.24，pp.3845-3856，2012.

[22]X.Zhou，L.E.Nelson，P.Magill P，etal.，“High spectral efficiency 400 Gbit/s transmission using PDMtime⁃domain hybrid 32-64QAMand training⁃assisted carrier recovery，”Lightwave Technol.，vol.31，no.21，pp.999-1005，2013.

[23]X.Zhou，J.Yu，M.F.Huang，and Y.Shao，“64⁃Tbit/s，8 bit/s/Hz，PDM⁃36QAMtransmission over 320 km using both pre⁃and post⁃transmission digi⁃tal signalprocessing，”Lightwave Technol.，vol.29，no.11，pp.571-577，2011.

[24]Y.Koizumi，K.Toyoda，M.Yoshida，and M.Nakazawa，“1024 QAM(60Gbit/s) single⁃carrier coherent optical transmission over 150 km，”Opt.Express，vol. 20，no.21，pp.12508-12514，2012.

[25]K.Toyoda，Y.Koizumi，T.Omiya，etal.，“Marked performance improvementof 256QAMtransmission usinga digitalback⁃propagationmethod，”Opt.Express，vol.20，no.21，pp.19815-19821，2012.

[26]Y.Koizumi，K.Toyoda，T.Omiya T，et al.，“512 QAMtransmission over 240 km using frequency⁃domain equalization in a digital coherent receiver，”Opt. Express，vol.20，no.21，pp.23383-23389，2012.

[27]P.J.Winzer，A.H.Gnauck，S.Chandrasekhar，et al.，“Generation and 1，200⁃km transmission of448⁃Gbit/s ETDM56⁃Gbaud PDM16⁃QAMusinga single I/ Qmodulator，”OFC 2010，San Diego，USA.

[28]F.Buchali，K.Schuh，D.Rosener，et al.，“512⁃Gbit/s DP⁃16⁃QAMfield trial over 734 km installed SSMFwith co⁃propagating 10 Gbit/s NRZ neighbors in⁃corporating soft⁃FEC decoding，”OFC 2010，San Diego，USA.

[29]J.Zhang，J.Yu，N.Chi，etal.，“Multi⁃Modulus Blind Equalizations for Coher⁃entQuadrature Duobinary Spectrum Shaped PM⁃QPSK Digital Signal Process⁃ing，”Lightwave Technol.，vol.31，no.21，pp.1073-1078，2013.

[30]J.Zhang，B.Huang，and X.Li，“Improved quadrature duobinary system perfor⁃mance using multi⁃modulus equalization，”IEEE Photon.Technol.Lett.，vol. 25，no.16，pp.1630-1633，2013.

[31]R.Rodes，M.Müeller，B.Li B，etal.，“High⁃speed 1550 nm VCSEL data trans⁃mission link employing 25 GBd 4⁃PAMmodulation and hard decision forward error correction，”Lightwave Technol.，vol.31，no.21，pp.689-695，2013.

[32]T.Tanaka，M.Nishihara，T.Takahara，et al.，“50 Gbit/s class transmission in singlemode fiberusing discretemulti⁃tonemodulation with 10G directlymodu⁃lated laser，”OFC 2012，Los Angeles，USA.

[33]R.P.Giddings，X.Q.Jin，E.Hugues⁃Salas，et al.，“Experimental demonstra⁃tion of a record high 11.25 Gbit/s real⁃time optical OFDMtransceiver support⁃ing 25 km SMF end⁃to⁃end transmission in simple IMDD systems，”Opt.Ex⁃press，vol.18，no.6，pp.5541-5555，2010.

[34]J.L.Wei，J.D.Ingham，D.G.Cunningham，R.V.Penty，and I.H.White，“Per⁃formance and power dissipation comparisons between 28 Gbit/s NRZ，PAM，CAP and optical OFDMsystems for data communication applications，”Light⁃wave Technol.，vol.30，no.26，pp.3273-3280，2012.

[35]J.D.Ingham，R.Penty，I.White，and D.Cunningham，“40 Gbit/s carrierless amplitude and phase modulation for low⁃cost optical data communication links，”OFC 2012，Los Angeles，USA.

[36]R.Rodes，M.Wieckowski，T.T.Pham，et al.，“Carrierless amplitude phase modulation of VCSEL with 4 bit/s/Hz spectral efficiency for use in WDM⁃PON，”Opt.Express，vol.19，no.27，pp.26551-26556，2011.

[37]M.B.Othman，X.Zhang，L.Deng L，etal.，“Experimental investigations of 3D/ 4D⁃CAPmodulation with DM⁃VCSELs，”IEEE Photon.Technol.Lett.，vol.24，no.22，pp.2009-2012，2012.

[38]M.I.Olmedo，T.Zuo，J.B.Jensen，etal.，“Towards 400GBASE 4⁃lane solution using directdetection ofmultiCAPsignal in 14GHzbandwidth per lane，”OFC 2013，Anaheim，USA.

[39]J.Wei，L.Geng，D.G.Cunningham，R.V.Penty，and I.White，“100 Gigabit Ethernet transmission enabled by carrierless amplitude and phasemodulation usingQAMreceivers，”OFC 2013，Anaheim，USA，pp.2065-2067.

[40]L.Tao，Y.Wang，Y.Gao，etal.，“Experimental demonstration of10Gbit/smulti⁃level carrier⁃lessamplitudeand phasemodulation for short range optical com⁃munication systems，”Opt.Express，vol.21，no.5，pp.6459-6465，2013.

[41]J.Zhang，J.Yu，F.Li，etal.，“11×5×9.3Gbit/sWDM⁃CAP⁃PON based on op⁃tical single⁃side bandmulti⁃levelmulti⁃band carrier⁃less amplitude and phase modulation with direct detection，”Opt.Express，vol.21，no.5，pp.18842-18848，2013.

[42]G.Stepniak and J.Siuzdak，“Transmission beyond 2 Gbit/s in a 100m SIPOF with multilevel CAPmodulation and digital equalization，”OFC 2013，Ana⁃heim，USA.

[43]G.H.Im，D.D.Harman，G.Huang，et al.，“51.84 Mbit/s 16⁃CAPATMLAN standard，”IEEE J.Sel.AreasComm.，vol.13，no.4，pp.620-632，1995.

[44]X.Xiao，F.Li，J.Yu，etal.，“100⁃Gbit/s single⁃band real⁃time coherentoptical DP⁃16⁃QAM⁃OFDMtransmission and reception，”OFC 2014，San Francisco，USA.

Biography raphy

Jianjun Yu received his PhD degree in electricalengineering from Beijing Universi⁃ty of Posts and Telecommunications in 1999.He works for ZTE Corporation as the chief scientist on high⁃speed optical transmission and director of optics labs in North America.He is also a chair professorat Fudan University and adjunctprofes⁃sor and PhD supervisor at the Georgia Institute of Technology，Beijing University of Posts and Telecommunications，and Hunan University.He has authored more than 100 papers for prestigious journals and conferences.Dr.Yu holds 8 U.S.patents with 30 others pending.He is a fellow of the OpticalSociety ofAmerica.He is editor⁃in⁃chiefof Recent Patentson Engineering and an associate editor for the Journal of Lightwave Technology and Journal ofOptical Communicationsand Networking.Dr. Yuwasa technicalcommitteememberat IEEELEOS from 2005 to2007 and a tech⁃nical committeemember ofOFC from 2009 to 2011.

Roundup

Domestic4GMobilePhoneShipmentsReached 31.6Million

December 12，2014—China Academy of Telecommuni⁃cation Research of MIIT has announced its analysis report ofdomesticmobile phonemarketofNovember 2014.

In November 2014，the overall shipments of mobile phones reached 44.543 million units in China.Among which 2Gmobile phone shipments was of 5.742 million，3Gmobile phonewasof7.2million and 31.601million for 4G ones.4G mobile phone shipments continued to in⁃crease rapidly，which wasmore than 4 times of 3Gmobile phone shipments.

During January to November in 2014，mobile phone shipments accumulated to 407 million in China.Among which 2Gmobile phone shipments was of 53.935 million，3G and 4Gmobile phoneswere of213million and 140mil⁃lion，respectively.

In November 2014，122 new mobile phonemodels was available in themarket.In which 2Gmobile phones are of 29models，3Gmobile phones are 19 and 4G are 74.Dur⁃ing January to November in 2014，there are 1952 newmod⁃els in total.Among which the new 2G phone models are 364，3G new phone models are 870 and 4G are 718.

(source:c114)

MIITPlans to Open Broadband AccessMarket

November 28，2014—Recently，MIIT issued the“Opin⁃ions On Open Access To Broadband Market(Draft)”，which regulates that private enterprises should be encour⁃aged to participate in the construction and operation of broadband access network infrastructure;encourage the private enterprises to participate in the relevant invest⁃mentand carry out cooperation with infrastructure compa⁃nies and provide broadband resale services etc.Themain three telecom operators shall not sign exclusive agree⁃ments with private enterprises and dynamic adjustment mechanism ofpricesshould beestablished.

The first batch of pilot cities include:Taiyuan，Shenyang，Harbin，Shanghai，Nanjing，Hangzhou，Ningbo，Xiamen，Qingdao，Zhengzhou，Wuhan，Changsha，Guang⁃zhou，Shenzhen，Chongqing and Chengdu.The pilot time is3 years.

(source:c114)

t

2014⁃10⁃10

10.3969/j.issn.1673-5188.2014.04.006

http://www.cnki.net/kcm s/detail/34.1294.TN.20141224.0955.001.htm l,pub lished online Decem ber 24,2014

This wo rk is suppo rted by the High Techno logy Research and Developm en t Prog ram of China(“863”Prog ram)under Gran tNo. 2012AA011303 and 2013AA010501 and National Nature Science Foundation o f China under Gran t No.61325002.