Capacity and Flexibility Improvement of Traffic Aggregation for Fixed 5G:Key Enabling Technologies,Challenges and Trends

Jing Zhang ,Tianhai Chang ,Zhuqiang Zhong ,Xianqing Jin ,Shaohua Hu ,Taowei Jin ,Jiahao Zhou,Mingyue Zhu,Yi Yu,Jianming Tang,Liangchuan Li,Kun Qiu

1 School of Communication and Information Engineering,University of Electronic Science and Technology of China,Chengdu,Sichuan 611731,China

2 Optical Research Dept,Huawei Technologies,518129,Shenzhen,China

3 School of Electronic Engineering,Bangor University,Bangor,LL57 1UT,UK

Abstract:As the emergence of various highbandwidth services and the requirements to support 5G/Wi-Fi 6 wireless networks,the next generation fixed networks,i.e.F5G,are expected to be realized in the 5G era.F5G is endowed with new characteristics,including ultra-high bandwidth,all-optical connections and optimal service experience.With the prospect of optical-to-everywhere,optical technologies are used for mobile front-haul,mid-haul,and back-haul.Optical access networks would play an important role in F5G to support radio access network and fixed access network.Low-latency PON is a key for cost effective-haul traffic aggregation.In terms of signal transmission,intensity modulation directdetection (IM-DD) is a promising scheme due to its simple architecture.The fundamental challenge associated with direct-detection is the disappearance of the transmitted signal’s phase.In access network,the flexibility and low latency are the two key factors affecting service experience.In this article,we review the evolution of PONs and the challenges of current PONs in detail.We analyze key enabling digital signal processing (DSP) techniques,including detection linearization for direct-detection and simplified coherent detection,adaptive equalizers,digital filer enabled flexible access network and low-latency inter-ONU communications.Finally,we discuss the developing trends of future optical access networks.

Keywords: optical fiber communication;flexible access network;optical field recovery;inter-ONU communication

I.INTRODUCTION

The rapidly emerging network services including cloud computing,ultra-high definition video,virtual reality and real-time online games require unprecedently high bandwidth and ultra-reliable communication connections to be successfully delivered by access and aggregation networks [1].The European Telecommunications Standards Institute (ETSI) has started the definition and specification of F5G since 2020[2—4].The F5G prospect is to implement optical fiber communication in mobile front-haul,mid-haul,back-haul,and all wired communication applications[3,4].10-Gb/s passive optical network is an important stage in the evolution of optical access network[5].In fixed communication networks,optical transport networks act as the backbones of the networks,which are being rapidly advanced towards the provision of a wide diversity of new networking features such as on-demand bandwidth[6].From the network infrastructure simplification and integration point of view,it is greatly beneficial if direct handshakes can be conducted between optical transportation networks and access networks.In addition,the seamless convergence of fixed access networks and radio access networks (Wi-Fi and wireless) is also highly desirable in order to further reduce network latency,costeffectively enhance the network flexibility and coverage,and also hugely lower practical network implementation complexity and cost.Passive optical networks (PONs) are highly promising solutions for local area networks (LANs) [7].Compared with active switches embedded in core optical networks,optical passive power splitters in PONs just play a simple aggregation role,thus PONs offer the most attractive solution to wide-band optical access networks and flexible aggregation which play a vital role in 5th generation fixed networks.The Full Service Access Network(FSAN)Group,the ITU Telecommunication Standardization Sector (ITU-T) Question 2/Study Group 15(Q2/15),and the IEEE 802.3 Ethernet Working Group are the three major industry organizations working on the standardization of PONs,which has evolved from asynchronous transfer mode PON(ATM-PON),to broadband PON (BPON),to Ethernet PON (EPON),to gigabit-capable PON (GPON)and recently to XG-PON standards[8].In 2015,NGPON2 employing the time wavelength division multiplexing PON (TWDM-PON) technology has been standardized to offer an aggregate network capacity of 40 or 80 Gb/s using 4 or 8 wavelengths over a distance of 40 km.Being an extended variant of the aforementioned PONs,the first version of 50-Gb/s standardized PON is expected to be released soon and even 200 Gb/s PONs may be discussed in the near future.

II.CHALLENGES

Figure 1 shows an access network with optical-wire and optical-wireless convergence.The current PON is based on burst mode TDM access(TDMA),which is considered as packet switching in the optical domain.Although the burst mode based 50G PON might cover the majority of application scenarios at current stage,it is difficult to evolve to 200G and beyond due to burst mode operation principles.Considering the industrial developing trend and addressing the corresponding technical challenges,it is greatly beneficial that PONs can cost-effectively and flexibly deliver the following highly desired features: 1) DSP-enhanced adaptability to the physical layer system characteristics,2)DSP-enabled multiple channel aggregation and disaggregation with flexible channel count,bit rate variations and bandwidth elasticity,3) great potential of more effective and low-complex transceiver DSP,such as adaptive equalizations,forward error correction (FEC),and synchronization of optics-generated millimeter wave (mmW) signals,4) direct communications between optical network units (ONUs) without sending any traffic via the optical line terminal(OLT),5) DSP-enabled excellent compatibility with the existing mobile fronthaul networks,and 6) excellent transparency to flexible RAN functional splits.These efforts are capable of achieving better transmission performances or flexibility compromise cost and complexity.So,how to gain better cost-efficiency and lower complexity digital signal processing are still key issues that need to be solved.In this regard,we review the evolution of PONs,and present DSP enabled high speed access techniques,low-latency and flexible network architectures with photonic wireless convergence.

Figure 1.Optical access networks with optical-wire and-wireless convergence.

Figure 2.The optical field recovery with Gerchberg-Saxton[9].

Figure 3.112-Gb/s PAM-4 signal transmission over 50-km fiber with different equalizers.

III.DSP ENABLED PERFORMANCE IMPROVEMENT

3.1 Transmitter-side DSP

To cost-effectively support the evolution of PON toward 100 Gb/s per wavelength,the use of lowcost transceivers with on-off keying inevitably suffers strong bandwidth limitation effect,which results in inter-symbol interference(ISI).Simplified preequalization,duobinary,faster than Nyquist (FTN),and advanced modulation formats are effective methods to alleviate such an effect at the transmitter side.Pre-equalization and duobinary increase the peak-toaverage power ratio and/or reduce the system noise tolerance.FTN significantly improves the complexity at the receiver side.Advanced modulation formats,i.e.pulse amplitude modulation (PAM) and discrete multitone (DMT),with higher spectral efficiency are widely investigated to improve the spectral efficiency and combat the bandwidth limitation effect.However,advanced modulation formats require a higher signalto-noise ratio,which causes that a large link power budget is needed.To explore the optimum trade-off between the spectral efficiency and receiver sensitivity for a specific application case,constellation shaping has been proposed in IM-DD systems.The Flex-PAM with probabilistic shaping has been proposed to leverage limited bandwidth and SNR requirement[10,11].However,the granularity is limited because most of the shaping techniques are based on PAM-8[12].For a higher order PAM,it is very challenging to achieve enough OSNR for error free transmission [13].DMT is an effective modulation for shortreach applications,in which the well-known waterfilling improves its transmission capacity that suffers power fading in the IM-DD systems.Even with bitloading,fixed constellations of M-QAM just offer a coarse granularity in data rate improvement for a flat channel response.The above fact means that DMT cannot maximize its channel capacity approaching the Shannon limit.DMT with continuous adjustment of entropy,i.e.entropy loading,can relax the granularity and squeeze out the last few bits in band-limited and power fading IM-DD systems.From the information theory point of view,normalized generalized mutual information(NGMI)is straightforward to evaluate the achievable information rate(AIR)and predict the FEC overhead.Therefore,we can optimize the NGMI to maximize the transmission capacity.We can use the AIR as a target and the FEC overhead as a constraint to define the NGMI’s lower and upper boundaries for optimizing the NGMI performance [14].Moreover,machine learning can also be used to achieve arbitrary small information entropy resolutions to maximize the net rate[15—17].Although DSP at the transmitter side can improve the performance or transmission capacity,it is,however,a passive process as it just adapts the channel impulse response.To maximize the transmission capacity or optimize the transmission perforemance,we can jointly use the transmitter-and receiver-side DSP to optimize the NGMI and meet the achievable information rate (AIR) target [14].Moreover,after using the linearization of direct-detection schemes,the non-ideal channel response can be flattened partly so that the transmission capacity is maximized to approach the Shannon limit.

3.2 DSP Enabled Detection Linearization

A fundamental challenge associated with IM-DD systems is the disappearance of the transmitted signal’s phase information after direct detection in the receivers.Along with the nonlinear effects associated with IM-DD,the maximum achievable signal transmission capacity of such a transmission system is also limited by chromatic dispersion and direct detectioninduced signal-to-signal interference (SSBI).This means that the received signal not only contains the signal itself but also signal’s beating products.Optical field recovery is thus essential to realize the linearized detection for the IM-DD system.The present popular Kramers-Kronig receiver has been successfully implemented to extract a received complex-valued signal.However,the received signal has to meet the minimum phase requirement to fully reconstruct the complex envelope of the optical field impinging upon the receiver [18].Therefore,single sideband modulation with a relatively large optical carrier-to-signal power ratio (CSPR) is essential for the technique to operate.As the intensity modulators generate the inherently double sideband (DSB) signals,optical filters,Hilbert filters or complex modulators must be introduced in the transmission link.In addition,it is a waste of the digital-analog convertor (DAC) bandwidth when SSB modulation is employed.To recover the complex-valued DSB signals,the carrier-assisted differential detection (CADD) introduces the optical delay line (ODL) at the receiver side to construct the transfer function and then mitigate the SSBI effect by iterations with a relatively large carrier.A frequency gap around zero frequency should be retained to avoid the transfer function to have a null point near the zero frequency[19].The transmission performance is sensitive to ODL which is also related to the symbol rate.This implies that it would be very challenging for upgrading the systems.Except for these two schemes,Gerchberg-Saxton (G-S) developed by crystallographers Ralph Gerchberg and Owen Saxton is a very classic and simple phase retrieval algorithm that infers the electron beams phase distribution based on intensity-modulating two planes [20—22].It is an iterative algorithm and can be extended to retrieve the phase of a pair of light distributions.To implement the G-S algorithm,two or more independent photon diodes (PDs) are used for parallel detection at the receiver,where different dispersion mediums are introduced to different PDs so that two or more planes are constructed.Then,the received signal’s fields are recovered by iteration process.Besides the iterative algorithm,non-iterative method can also reconstruct the complex signal based on solution of the temporal transport-of-intensity equation (TIE) by using a dispersion medium [23,24].At the receiver side,two parallel PDs are used for detection.These iterative and non-iterative methods introduce more PDs and dispersion mediums,thus the network architecture is changed in physical layer.Alternatively,we regard the received optical current as the image plane,and the transmitted optical signal as the object plane as shown in Figure 2 [9].The main object is to minimize the errors and ensure perfect matches between the two intensity planes,i.e.the real optical field and the intensity of the estimated optical field after passing through the system.Pilot symbols and randomly allocated decision symbols are introduced into the transmitted frames in the G-S iteration algorithm to accelerate the convergence speed of the iteration process.The data-aided iterative algorithm can recover the optical field and transmitted data by one photodiode only even at the presence of severe CD-induced distortions.Numerical simulations show that the G-S approach is feasible to support 100-Gb/s PAM4 signal transmissions over 50-km SSMFs with overheads as low as 6.7%[9].

Figure 4.Schematic diagram of the concurrent Inter-ONU communications enabled IM-DD hybrid OFDM-DFMA PONs.US:upstream,DS:downstream,Tx: transmitter,Rx: receiver.De-mux: wavelength de-multiplexer[27].

In addition to the aforementioned optical field recovery methods,adaptive equalization is a reception linearization method in terms of mitigating both the band-limited and power fading effects.The linear feed forward equalizer (FFE) can shorten the channel impulse response and shape the signal and noise,leading to effective compensations for small ISI.The nonlinear Volterra filter(VF)is another equalizer for reducing nonlinear distortions.However,these two equalizers are finite impulse response (FIR) filters with no poles in the Z-plane,this means that spectral zeros caused by fiber dispersion still exist.The infinite impulse response (IIR) filter has its poles with a decision device,so the decision feedback equalizer(DFE)can compensate for distortions caused by spectral nulls.However,DFE has relatively large complexity and long iteration because of its feedback structureinduced increase in equalization tap.Meanwhile,error propagation also occurs due to DFE’s decision errors.Hybrid equalizers that combine different equalizers,such as FFE+DFE or VF+DFE (VDFE) have the ability of flattening channel response so as to mitigate the ISI and power fading effects [25].In Figure 3,we have experimentally demonstrated a 112-Gb/s PAM-4 signal transmissions over 50-km SSMFs with BERs below the 20%FEC.The power fading effect is gradually alleviated from linear equalization to hybrid equalizer[26].SSBI is mitigated and the signal field can be recovered.In general,the equalizationbased linearized direct-detection has huge complexity since the equalizations requires large numbers of taps for linear and nonlinear distortion compensation,which increases with transmission capacity.As a direct result,the exact DSP design is link configuration dependent,indicating the current DSP design cannot meet the 200G PON’s low-complexity requirement.

3.3 Simplified Coherent Detection

Despite all existing PONs utilizing direct-detection,coherent detection with advanced modulation formats has attracted increasingly attentions as the PONs are envisaged to target at 200 Gb/s or even higher bit rates per wavelength.The EU FP7 cost-effective coherent ultra-dense-WDM-PON for lambda-to-the-user access(COCONUT) project explores the “wavelength-tothe-user” concept and investigates optical coherent techniques to realize a new fully scalable optical access network with significantly extended network operation dimensions[28].The significant advantageous of a coherent system is the relatively large power of a local oscillator (LO),which enables the linear detection of a signal in the presence of low received powers.This may be very favorable for access networks with limited power budgets.However,since its high complexity,cost and power consumption of digital coherent demodulation algorithms are used in core networks,e.g.polarization-and phase-diverse intradyne (PPDI) digital coherent receivers prevent their wide applications in the PON scenes.The most important challenge is to simplify the optical coherent systems to reduce the cost and complexity.Since the main power consumption or complexity in coherent systems is the DSP at the receiver,such as carrier recovery and polarization demultiplexing,the transmitter uses conventional polarization multiplexed I/Q modulator in general.Advanced modulation formats and constellation shaping can be used to improve the spectral efficiency and realize the rate adaption.The frequency stability and phase noise due to the laser sources are the fundamental problems for digital coherent demodulation.The coherent-lite PON that has been proposed to use the same light source or locked sources as the optical carrier and LO,which can significantly relax the carrier recovery requirement [29—31].Thus,larger linewidth lasers,e.g.in an order of megahertz,can been used as the sources in the coherent optical systems.However,its coherence may be destroyed after transmitting over different fibers and injection locking is not easy to implement for multipoint to point laser control.Polarization-diversity also comes at the expense of significant complexity,so polarization-independent coherent reception is also a key issue.Various techniques have been proposed to achieve the polarization-independent performance[32],such as the symmetric 3times3 coupler[33,34],polarization-time block coding [35,36],and significantly simplified coherent detection.However,all these techniques sacrifice one polarization state or require polarization-diversity signal generation in the transmitter[28].Therefore,many technical challenges must be solved in coherent receiver designs in order to be practically implemented in future access networks.

IV.FLEXIBLE PON ARCHITECTURES

4.1 Requirements of Flexible 5G and Beyond Fronthaul

The high bandwidth,low latency and ultra-dense connection in 5G and beyond networks have to deliver impose unprecedented pressures on current mobile fronthaul networks.The existing fronthaul’s common public radio interface (CPRI) defined with fixed line bit rates cannot cost-effectively satisfy 5G and beyond networks’ requirements mainly because of its low spectral efficiency of digitizing orthogonal frequency division multiplexing(OFDM)IQ signals into binary bit streams.More importantly,for the conventional CPRI-based fronthaul,the required transport bandwidth scales with antenna port count [37].This considerably hinders the current RANs from utilizing 5G massive MIMO and beaming forming techniques.In addition,due to the emerging heterogeneous broadband services and applications,the inelasticity,inflexibility and low spectral efficiency associated with the current CPRI-based fronthaul make the current RANs incapable of cost-effectively addressing the explosive growth of mobile traffic with a large diversity of traffic characteristics and highly dynamic mobile traffic patterns.For effectively enabling the implementation of 5G and beyond networks,flexible functional splitting will be adopted depending upon application scenarios and 8 functional options are introduced by the 3rd Generation Partnership Project (3GPP).The current CPRI-based fronthaul corresponds to the functional split option 8.

To support flexible functional splitting,evolved CPRI(eCPRI)has been developed,which exploits existing IP/Ethernet transport techniques to achieve statistical multiplexing and simplifies the transmission of operations,administration and maintenance (OAM),and synchronization information,as well as ensuring network security (based on IPsec and MACsec).However,similar to its predecessor CPRI,eCPRI is also based on binary bit streams over the fronthaul,which unavoidably results in low spectral efficiency.To achieve bandwidth-efficient fronthaul,transmitting aggregated multiple mobile channels over analogue optical fronthaul links is highly desirable.For achieving this,use can be made of either extra radio frequency components such as up-/down-converters,I/Q-mixers [38] or DSPs such as Fast Fourier Transform (FFT)/Inverse FFT -based multichannel aggregation [39],and Walsh code sequence-based codedivision multiplexing[40].In comparison with the solutions using extra radio frequency components,DSPenabled multiple channel aggregation and disaggregation are attractive due to its cost efficiency and inherent digital-domain operation configurability and flexibility.However,to significantly reduce channel interferences,guard subcarriers,frequency domain windowing,or channel interference cancellation are required for the abovementioned DSP-enabled multiple channel aggregation and disaggregation techniques.Furthermore,it is also expected to use DSP to achieve improved channel bandwidth elasticity,flexible channel count variation,excellent adaptability to transmission system impairments and inherent transparency to flexible RAN function splitting.In addition,IMDD PONs are promising for cost-effectively providing broadband connectivity for fronthaul.

4.2 Low Transmission Latency Requirements in 5G and Beyond Fronthaul

The maximum transport network latency in fronthaul is primarily decided by the RAN internal processing time constraints that depend on hybrid automatic repeat request (HARQ) loop,scheduling,and etc.The requirement of low end-to-end latency for 5G applications determines the maximum allowable latency of fronthaul,which is within the range of 250μs.Considering a typical 20-km transmission scene and the transmission delay is about 5μs/km,so the roundtrip time for the fiber propagation is 200μs.It can be concluded that the time delay from fiber dominate the overall end-to-end latency.In addition,the data exchange among different end-users becomes popular in a single converged network infrastructure.As such,end-user data traffic has to be transmitted to the OLT via upstream links to provide such services in conventional PON-based fronthaul.Then,the information is re-modulated onto the downstream carriers to finally route to the destined ONU.This inevitably causes additional transmission latency,bandwidth occupation and extra signal processing.Thus,it is straightforward to provide end-users with direct communications without passing the data traffic to the OLT so that the end-to-end latency can be reduced.However,previously reported schemes for achieving direct communications between ONUs need to introduce extra optical components [41] or to add multiple additional transceivers[42],which results in huge modifications to the conventional PON architectures.Moreover,these schemes have limited transparency and flexibility for future converged networks,since there are a wide diversity of key network operation features that are envisaged to be accommodated simultaneously.More importantly,the upstream bandwidth needs to be relaxed to meet the upstream transmission requirement.

4.3 Digital Filtering-Enabled Network Flexibility and Latency Reduction

Recently,a centralized,software-defined networking (SDN) controller-managed,digital filter multiple access (DFMA) PON was proposed,in which dynamic software-reconfigurable orthogonal digital shaping and matching filters are used to combine or separate multiple gapless channels of arbitrary bandwidth granularity in different frequency regions of the transceivers without employing extra analog hardware.It potentially provides multichannel fronthaul solutions for 5G and beyond.The key advantages of DFMA PONs are summarized as follows: 1)DSP-enabled dynamic network flexibility,elasticity capability and reconfigurability[43,44],2) Capability of SDN-based abstraction and virtualization for all layers including the physical layer,3) offering a useful framework for developing universal optical transceivers for cost-sensitive ONUs,4) transparency to signal bit rate,signal modulation format,network topologies and existing PON access techniques,5)excellent backward compatibility and great potential for supporting future C-RANs with flexible functional splits.DFMA-PON based fronthaul has been numerically explored and experimentally demonstrated in[45].More recently,an IM-DD hybrid DSB/SSB OFDMDFMA PON is subsequently proposed[46],in which for upstream transmissions,each ONU uses specific digital shaping filters (SFs) to locate the OFDM signals at different sub-wavelength spectral regions,which is similar to the conventional DFMA PONs.While in the OLT,a single FFT operation and the following data recovery DSP processes are implemented in a pipelined way to simultaneously demultiplex and recover the signals from various ONUs without employing the corresponding matching filters(MFs).This can bring a 100-fold reduction in the receiver DSP complexity when the total number of ONUs are larger than 32.Besides,it has higher tolerance to digital filter characteristic variations,transmission impairments and transceiver sample timing offset.

Furthermore,concurrent inter-ONU communication can be achievable in hybrid OFDM-DFMA PONs by operating these simple alterations to the remote node,whose principle is illustrated in Figure 4 [27].For inter-ONU communications and upstream,the inter-ONU communication and digital filtered OFDM signal for upstream are combined in the digital domain.After passing through a DAC,the generated electrical signals in each ONU are sent to intensity modulators to perform electrical-to-optical conversion with different wavelengths,and then transmitted to the remote node through the distribution fiber.The remote node incudes a 3×N optical coupler and an optical isolator.One portion of the combined optical signals from ONUs are transmitted to the OLT through the coupler and the feeder fiber.Whilst the other portion of the optical signals is redirected to each ONU via the coupler to achieve inter-ONU communication.At the OLT,the received aggregated optical signal is fed to a PIN detector to realize square-law photo detection,then digitized by an ADC and demodulated after sideband identification procedures,where the desired upstream signal bands are extracted.At the ONU side,the received optical signal first passes a de-multiplexer to separate the downstream and inter-ONU signal.After that,the downstream and inter-ONU signals can be demodulated respectively with DSP procedures similar to those used in upstream signal recovery.

Figure 5.Measured BER transmission performances of the IM-DD hybrid SSB OFDM-DFMA PON and corresponding radio frequency spectra of upstream and inter-ONU sides received signals[27].

Figure 5 illustrates the measured BER performances of the IM-DD hybrid SSB OFDM-DFMA PON,where two ONUs are adopted for concurrent 2-km inter-ONU communication and 25km upstream transmission[27].Benefited from the adaptive bit-power loading,all the channels have similar BER performances at the 20% SD-FEC limitation threshold.The insets of Figure 5 show the constellations of corresponding channels when the received optical power is -4dBm.As seen in the spectra,the two channels locate at the lower frequency regions are least impacted by power fading effects thus they are assigned for upstream traffic,while the other two channels,which are affected by strong power fading effects,are assigned for inter-ONU communications.The benefit is that the upstream transmission is not affected because the inter-ONU communications uses the band above 12GHz.Therefore,such band allocation can achieve a higher overall spectral efficiency.

Compared with the previously reported schemes that only achieving direct ONU communications,the proposed PON has the following advantages: 1)concurrent upstream and inter-ONU communications have no requirement of considerable modifications on installed PONs,2) significantly improved overall spectral efficiency by adaptively allocating the available band for inter-ONU communications and upstream transmissions,3) DSP-based reconfigurations without requiring any reconfigurations in transceiver hardware and also the remote node.Generally speaking,the high radio frequency spectral region suffers the severe frequency selective power fading effect,which can be used for inter-ONU communications.The experimentally results show that such a RF spectral region allocation can improve the aggregated transmission capacity by a factor of 1.3.Apart from the transmission capacity improvements,the effect of the Brillouin backscattering effects on the inter-ONU communications is negligible by suitably allocating the upstream and inter-ONU communication spectra.

4.4 Seamless Optical-Wireless Convergence

Owing to the shortage of RF spectrum,the convergence of optical-wireless networks is beneficial,as new spectra of mmW,terahertz(THz) and even petahertz (PHz) might be exploited by collaboratively integrating optics/wireless-based technologies[47].For indoor scenarios,optical fibers can be deployed close to mobile users in a room,as the global mobile suppliers association predicts that>85%of mobile traffic will occur indoor in the near future.For outdoor scenarios,PHz/THz wireless links can be applied for mobile backhauls in rural areas where optical fibers cannot be easily deployed,as shown in Figure 1.

Compared with PHz/optical signals generated with lasers,mmW or even THz signals usually suffer severe phase noise/distortions.In the conventional way of generating a mmW/THz signal with an electronic oscillator,many stages of frequency doubling enlarge the unwanted phase noise effect.To address the phase noise challenge,the mmW/THz signals can be generated optically by beating between two optical waves at different wavelengths in a photodetector.This normally requires very narrow linewidth lasers operating at precise and stable wavelengths.Two methods can be utilised to optically generate tunable mmW/THz signals: opto-electronic oscillators (OEOs) (<50 GHz)and heterodyning of two continuous wave(CW)lasers (>50 GHz) [48].The resulting phase noise associated with these two methods does not depend on carrier frequency.However,the complexity and cost of these two methods are too high to be implemented in cost-sensitive application scenarios,as a large number of costly optical/electrical components are required to achieve high-quality mmW/THz RF signals.Compared to the aforementioned approaches,DSP can be used to estimate and subsequently eliminate the phase noise and frequency offset induced by two low-cost standalone lasers in a practical transmission system.For typical OFDM-based wireless transmissions,the distorted phase of the mmW/THz signal can be easily extracted with the FFT.The phase difference between two consecutive OFDM symbols with pilot subcarriers is used to estimate the laser phase noise,and the frequency offset between two lasers and the target mmW/THz carrier frequency,which are then digitally compensated.The integration of optical fibers into the wireless networks brings an opportunity of using DSP to further enhance the bandwidth capacity and flexibility of existing WiFi and 5G systems.

V.TRENDS AND CONCLUSION

To practically implement DSPs in cost-sensitive F5G scenarios,considerable advances still need to be made in the following aspects: The low complexity and power dissipation of any DSP solutions are vital for their future deployments in PONs.The combination of optical field recovery with simplified equalization may be capable of reducing the DSP power dissipation.For current TDMA,all transceivers in burst mode networks require burst mode amplifiers,analog/digital converters,and full bandwidth access to the aggregation rate.Apart from being expensive,this design also suffers from poor bandwidth utilization efficiency and technical difficulties in cost-effectively manufacturing burst mode analog devices and circuits.Moreover,coherent detection technologies can support line rates of beyond 200G,and enable transceiver flexibility with the aid of DSP and advanced modulation format.DSP can bring frequency division multiplexing access to PONs,in which the transceiver bandwidth in a terminal node can be much smaller than that of the aggregation node.For coherent detection-based PONs,the fundamental problem is the light source that may prevent the carrier recovery and polarization division de-multiplexing from being conducted properly.The rapid progress in flexible architectures for 5G and beyond networks are currently being made and attracting great attention from both industry and academia.For the techniques transmitting analogue signals over fronthaul suffer from noise,nonlinearity and other penalties severely,which may cause them unreliable due to the lack of error correction schemes in comparison with the CPRI and eCPRI-based schemes.For practical implementation of inter-ONU communications,the overall fronthaul architecture should be designed to have small equipment footprint,low deployment cost,reduced overall power consumption,and high bandwidth efficiency,as well as excellent flexibility,upgradability and scalability.Digital filter and inter-ONU communications may have contributions to flexibility and low latency access.

ACKNOWLEDGEMENT

This work was supported by National Science Foundation of China(NSFC)(61871082 and 62111530150),and Fundamental Research Funds for the Central Universities(ZYGX2020ZB043 and ZYGX2019J008).