Demonstration of Terabit/s LAN-WDM for the Evolution of B5G/6G Fronthaul Networks

Intensity modulation/direct detection (IM/DD) and coherent schemes are two kinds of promising solutions to achieve the evolution for B5G/6G fronthaul networks. IM/DD has more mature commercial deployment applications for short-reach applications with lower costs and simpler configurations compared with coherent solutions. In this article, we experimentally demonstrate a local area network-wavelength division multiplexing (LAN-WDM) IM/DD access system with a record bit rate of 2 Tb/s (250 Gbit/s × 8) probabilistic shaping eight-pulse amplitude modulation (PS-PAM-8) signal with joint equalization techniques to support the evolution of future B5G/6G mobile fronthaul networks. Considering the 20% Soft Decision Forward Error Correction (SD-FEC) threshold with the bit error rate (BER) of 2 × 10−2, we realize a record net bit rate of 1.6 Tb/s over 5 km fiber. This system can potentially support a 12-wavelengths transmission with the total line rate of 3 Tb/s (250 Gbit/s × 12). Moreover, we analyze the challenges of the evolution, usage scenarios, digital signal processing (DSP) implementation, transmission rate and distance, and the tolerance of dispersion with different wavelengths. The LAN-WDM can be considered as a feasible and potential candidate scheme to meet the evolution and deployment of B5G/6G Terabit/s fronthaul networks.


I. INTRODUCTION
L OOKING into the future, with the increasing wireless data and large-scale interconnections in beyond fifth-generation and sixth-generation (B5G/6G) fronthaul, a flexible, intelligent, and cost-effective fronthaul network is urgently needed to be further expanded to provide high-capacity service, higher traffic density and lower latency with hundreds of Gb/s or even Tb/s for a ubiquitous mobile society [1], [2]. Part of the access network can be used for fronthaul deployments, and in turn, fronthaul access can also prepare for the construction of the access network. Two major industrial organizations have been ongoing efforts on the evolution of access network standardization work: the ITU Telecommunication Standardization Sector (ITU-T) Question 2/Study Group 15 (Q2/SG15) and the IEEE 802.3 Ethernet Working Group. The access data rates have been upgraded by the Q2/SG15 group from approximately 2.5 Gb/s to 10 Gb/s, 50 Gb/s per wavelength, and upgraded by the IEEE 802.3 group from circa 1.25 Gb/s to 10 Gb/s, 25 Gb/s per wavelength in the past two decades [3], [4]. The peak data rate per device of fourth-generation (4G) up to 1 Gbps with that of 5G up to 10 Gbps, and the 6G system is expected to achieve a super-fast data rate in Tb/s [5]. The current 5G fronthaul which connects active antenna units (AAUs) and distributed units (DUs) is based on point-to-point (PTP) intensity modulation/direct detection (IM/DD) systems, with either dedicated fiber links or wavelength division multiplexing (WDM) links [6], [7]. Fig. 1 shows the evolution of future mobile fronthaul architecture. In the future B5G/6G network, with the shortening of the upgrading cycle, frequent replacement of remote hardware devices in different places will result in increasing operation and maintenance costs [1]. Nevertheless, in order to speed up the deployment of B5G/6G fronthaul network and adapt to the rapidly growing demand for higher capacity, simpler deployment and lower cost, the mature commercial components may be reused to upgrade as much as possible considering the trend of building base stations more and more densely. So as to meet the target of data communication up to Tb/s in B5G/6G fronthaul network, the single-channel line rate of 100 Gb/s or 200 Gb/s may be required using this architecture for fronthaul networks. Furthermore, the total rate of 12 wavelengths can clearly reach 1.2 Tb/s or 2.4 Tb/s, as depicted in Fig. 1. This architecture has favorable coexistence characteristics between 4G and 5G, and the possibility of B5G/6G co-existing with them in the future needs to be investigated further. To cope with this concern, several works have been done on evolution technology for fronthaul networks by many research groups in this 5G access network. At present, there are two technical routes that have been preferred to this issue: PTP IM/DD and point-to-multipoint (PTMP) coherent detection. Coherent optics has been widely deployed in long hauls and metro transmission systems and the coherent technology will likely be introduced into access networks owing to its superior receiver sensitivity and long-range operation in the near future. However, direct implementation of coherent optics in access networks is uneconomical on account of high costs, and which is also the main reason why coherence architecture is not commercially available at present. Moreover, they are of high complexity and power consumption in this fronthaul network, and have not yet been reduced to the same cost level as IM/DD. Therefore, IM/DD is still a more attractive approach due to its cost effectiveness and simpler configuration.
IM/DD has been widely used in 5G commercial systems. At any point in the moment, the industry will use the highest practical speed per channel and then use WDM to increase the capacity to meet any needs of the application. Several WDM schemes have been proposed and widely deployed [7], [8], [9], [10], [11], including coarse WDM (CWDM), local area network WDM (LAN-WDM), and dense WDM (DWDM). These solutions all have been deployed in existing commercial 5G applications in China. For example, the three major operators: China Mobile, China Telecom, China Unicom and they adopt Open-WDM/ (Medium WDM) MWDM, LAN-WDM and optical transport network (OTN)+DWDM, respectively. The MWDM is expanded from 6 wavelengths to 12 wavelengths by adding thermal electronic cooler (TEC) for temperature control based on the first 6 wavelengths of CWDM [7]. In 2022, one real-time field trial of a 300 Gb/s 12-channel MWDM system in a deployed 5G C-RAN fronthaul network was reported [8].
A new CWDM and circulator integrated semi-active system for 5G fronthaul was proposed and demonstrated [9]. Furthermore, the LAN-WDM can do smooth upgrade to save user deployment Fig. 2. Channel schedules of 12-standard-channel LAN-WDM and the chromatic dispersion coefficient versus wavelength ranges from 1270 nm to 1350 nm [13]. and spectrum resource, and reduce the system cost due to the existing mature deployment of industrial chain, which might be considered to be a potential way for front-haul network. The channel schedule of LAN-WDM is shown in Fig. 2. Twelve channels are used for fronthaul and consist of three parts, which are the first 4 wavelengths and the last 4 wavelengths of the LWDM standard and the 4 wavelengths of CWDM. Can LAN-WDM meet the evolution from 4G/5G deployment to B5G/6G Terabit/s fronthaul networks using deployed infrastructure?
The evolution of LAN-WDM based on for B5G/6G fronthaul network is facing several demanding challenges and bottlenecks, which can be summarized up as two main transmission impairments limiting the performance: the nonlinear impairments from the nonlinear region of the imperfect optoelectronic components, including the digital-to-analog converter, the photodiode, modulators, and complex channel environment, and linear impairments from the bandwidth constraint of the electro-optical devices and fiber chromatic dispersion (CD) [12]. Therefore, the achievable data rate and system capacity would severely deteriorate. Fig. 2 further shows the CD coefficient which is specified for G.652 type fibers for twelve wavelengths according to the ITU-T Recommendation G.652 standard [13], [14]. However, there is a lack of comprehensive research about the effect of CD on the evolution of LAN-WDM to future mobile fronthaul networks. Furthermore, to improve the loss budget in the O-band where transmission has low dispersion penalties, but also high fiber attenuation loss, semiconductor optical amplifier (SOA) is considered as an appropriate choice to serve as an amplifier in signal transmission. Because SOA has the features of low cost, high integrate ability, fast carrier dynamics and small physical footprint [15], [16], [17], [18]. Meantime, pulse amplitude modulation (PAM) format, a feasible modulation scheme for the high capacity fronthaul network, not only can improve spectral efficiency (SE) with low energy consumption and ensure a simple system architecture but also has stringent linearity requirements [19].
In this article, we extend our OFC contribution [18] to provide a more detailed demonstration and discussion whether the LAN-WDM can meet the evolution from 4G/5G deployment to B5G/6G Terabit/s fronthaul networks. Challenges and opportunities to achieve a super high data rate up to Tb/s as the target in B5G and 6G fronthaul network are analyzed. The transmission rate, transmission fiber length and the tolerance of dispersion with different wavelengths are also fully studied and discussed. In addition, we have experimentally demonstrated an 8-channel LAN-WDM IM/DD transmission system for B5G/6G fronthaul network and successfully achieve a new record bit rate of 2 Tb/s (250 Gbit/s × 8-channel) probabilistic shaping (PS)-PAM-8 signal with 1.6 Tb/s net rate over 5 km fiber transmission at soft decision forward error correction (SD-FEC) threshold with the bit error rate (BER) of 2 × 10 −2 with the aid of powerful equalization techniques, making it a promising technological application scheme with high capacity, low cost, and mature infrastructure.
The remainder of this article is organized as follows. In Section II, we give a detailed description of the experimental setup, including system parameters and transmission situations. In Section III, the digital signal processing (DSP) implemented at transmitter and receiver side are introduced and analyses of DSP parameters are shown. The results and performance comparisons, such as the optimization of the laser and SOA performance, transmission rate, transmission distance and dispersion tolerance and different channels after different transmission cases, are presented in Section IV. In Section V, we give extensively discussions. Finally, we conclude this article in Section VI.

II. EXPERIMENTAL SETUP
The experimental setup for LAN-WDM IM/DD transmission system with PS-PAM-8 is shown in Fig. 3. The transmitter consists of a high-speed arbitrary waveform generator (AWG), eight distributed feedback (DFB) lasers, two 65-GHz electrical amplifiers (EAs) and two intensity-modulators (IMs, EOSPACE AZ-DV5-65-PFA-PFA) with 6 dB bandwidth of 65 GHz. The eight O-band DFB lasers are multiplexed by a multiplexer (MUX) through odd and even channels in the link and the IM modulators modulate odd and even channels respectively. At the transmitter side (Tx), the 92-GBaud PAM-N drive signals are generated by the 92-Gsa/s AWG with the 3-dB analog bandwidth of 32-GHz with an offline Matlab program, and the AWG output separately drives the EAs and IM modulators of the odd and even channels, then the modulated signals are combined by an optical coupler (OC). The wavelengths have all the same modulation. Three different transmission cases are considered in this experimental system: Case 1: Back-to-back case without transmission fiber; Case 2: 3 km and 5 km standard single mode fiber (SSMF) transmission; Case 3: In order to support 10 km, 15 km and 20 km SSMF transmission, the modulated optical signals are amplified by a SOA boost amplifier before being transmitted into fiber. The SOA as a boost amplifier only be considered in case 3. The optimization of the SOA parameters will be discussed in more detail in Section IV. Then the 92-Gbaud modulated optical signal is transmitted over fiber and demultiplexed by a demultiplexer (DEMUX). One variable optical attenuator (VOA) is applied to adjust the input power into photodiode (PD) to emulate the receiver performance. What we measure is the input power into PD in the later part of our experiment. During measurement, one receiver is considered, and which is applied for the measurement of each wavelength channel. We take turns performing different wavelength channels and one channel's data is collected at a time.
At the receiver side (Rx), considering attenuation in O-band transmission, we use one SOA at the optical network unit (ONU) before VOA to amplify the received optical signal so that the 92-Gbaud PAM-N signal can be directly detected by a 100-GHz PD and amplified by another cascaded EA, and then captured by a real time digital storage oscilloscope (DSO) working at 128-GSa/s with 59-GHz bandwidth and processed by offline DSP. Fig. 4 shows the optical spectra of eight channels without modulation under back-to-back (BtB) case. In this article, 8-channel MUX and DEMUX were used in this demonstration due to the limitation of experimental conditions, but we also give a comprehensive analysis for 12-channel application in Section IV.

A. Tx DSP
The off-line Tx DSP block is shown in Fig. 5(a). In the Tx offline DSP, PAM-N signal with a length of 2 18 with Gray-mapping are generated [20]. To support 92-Gbaud signals transmission, the EAs and IMs at the transmitter-side may operate in nonlinear region. Therefore, we adopt PS technique scheme which can offer adaptive shaping gain to reduce the average power of the transmitted signal and improve receiver sensitivity with symmetric probabilistic distribution in order to offer adaptive shaping gain compared to regular modulation formats. By changing the rate parameter, the probability distribution of different constellation points can be changed. After oversampling to two samples per symbol (2-sps), a square root raised cosine filter (RRCF) with the optimal roll-off factor 0.01 and digital time-domain pre-equalization (Pre-EQ) using 19-tap constant-modulus-algorithm (CMA) equalizer FIR are used to adapt to further Nyquist pulse shape for the signal and pre-compensate the linear impairments during transmission, but effective power of signals at the output of AWG will decrease with the increase of the taps of FIR. Moreover, the symbol sequence is resampled to one sample per symbol (1-sps) to match the sampling rate of the AWG and loaded to the AWG at Baud-rate sampling. In resampling, Kaiser Window filter can be used to avoid frequency aliasing [21]. With the increase of the shape parameters that determines main-lobe width and side-lobe level, the amplitude of the side lobes can be decreased and the energy concentration in the main lobe can be increased, and which can bring us higher resampling accuracy, but also extend the calculation time. Therefore, by varying the length of the window and shape parameters, we can find the optimal values so that obtains significant system improvement. Filtering effects can cause inter-symbol interference (ISI) limiting the baseband signal bandwidth. Moreover, time-domain digital Pre-EQ is also used to reduce the ISI because of the bandwidth limitation [15], and the compensating effect of the Pre-EQ will be discussed in more detail in Section IV. Pre-EQ and PS processing can result in limited peak power. Therefore, the clipping method is adopted as an effective method to reduce the peak-to-average power ratio (PAPR) of PS signals [22]. Before the transmission experiments, we analyzed the PAPR of higher-order PAM signals based on PS techniques. As shown in Fig. 6, the PAPR of PS-PAM-8 signals under different time-domain Pre-EQ and clipping conditions is given. The method of time domain Pre-EQ does not need to add additional DSP module at the Rx, nor does it need strict symbol synchronization, so which is simpler to realize than the method of frequency domain Pre-EQ. Firstly, the PS-PAM-8 signals without Pre-EQ and clipping have lower PAPR. After PS-PAM-8 signals are only Pre-EQ in time domain, PAPR will increase a lot. Time domain digital Pre-EQ method is suitable for signal with limited bandwidth. However, we can find that the disadvantage of Pre-EQ is that the signal PAPR is very high, especially for PS-PAM-8 signals, which may exceed the linear operating range of EAs and optical modulators and other photoelectric devices, resulting in nonlinear signal distortion. Therefore, there is no doubt that the system performance will deteriorate. In addition, it can be seen that the curve of the PAPR of PS-PAM-8 signal after Pre-EQ and clipping is cut-off at PAPR of 4.5 dB. After resampling, the signals shaping introduced by Pre-EQ still remains. Fig. 7(a) and (b) respectively depict the channel response with or without digital Pre-EQ in time domain. It can be observed that high frequency part of the signal without digital Pre-EQ is restricted. After the Pre-EQ process, the relatively flat channel response is obtained. Twelve channels are used for fronthaul and consist of three parts according to the Fig. 2. In our experiment, we firstly named 8 standard wavelengths of LAN-WDM as Ch1 to Ch8. The remaining four wavelengths are named Ch9 to Ch12 in order of wavelength from smallest to largest. Fig. 8(a) and (b) show the optical spectra of 92-Gbaud PS-PAM-8 signals without and with Pre-EQ or clipping under BtB case in the channel 1 and 8, respectively. It can be observed that the central frequency is relative stable. And it can be seen that the signal carrier to signal power ratio (CSPR) after Pre-EQ will greatly increase, and the  signal effective power will decrease as the relative direct current (DC) component is larger. Meantime, when optimizing the number of taps for Pre-EQ FIR, the compromise between BER performance of the system and the effective power degradation is obvious [23]. Therefore, it is necessary to optimize the balance between the FIR used for Pre-EQ and the effective power of the output signal. Before the fiber transmission measurement, the Pre-EQ FIR is estimated based on the transfer function of constant modulus algorithm (CMA) equalizers under 92-Gbaud on-off keying (OOK) signal after BtB case. After steady convergence, the tap coefficients of FIR for Pre-EQ in the time domain can be calculated from the CMA filter after normalization and symmetry processing.

B. Rx DSP
The off-line Rx DSP block is shown in Fig. 5(b). The Rx offline DSP includes resampling to two samples per symbol(2sps), squaring time recovery applied to remove the timing offset and jitter from the data, T/2-spaced CMA equalizer used to compensate for the linear impairments, T-spaced nonlinear equalization (NLE) base on Volterra Filter used to reduce the linear and nonlinear impairments, and T-spaced decision-directed least mean square (DD-LMS) equalizer used to further compensate for the channel response and to mitigate the implementation penalty of devices before final decision. In our previous works [15], we have extensively investigated typical TX and RX DSP, including the optimizing numbers of equalizer taps. Before the transmission experiment, the number of equalizer taps with different DSP operation conditions is optimized, and the taps numbers of CMA, NLE, DD-LMS are fixed at 19, 189 and 189 respectively [15]. The joint equalization based on CMA and NLE with DD-LMS has the best system performance and then is used for the experiment. Finally, the BER can be calculated after PAM-N demodulation based on the recovered signal.
As is known to all, feedforward equalizer (FFE)/ decision feedback equalizer (DFE) or modified related methods requires a training sequence to obtain the equalizer coefficients, and which will increase the system overhead. The performance can be degraded by the error propagation which can be resulted in by the limitation of transmission impairments, especially for the signal suffering from severe nonlinearity. The CMA without using the training sequence has been widely used in optical communication systems for the compensation of the static CD and time-varying polarization mode dispersion (PMD). This blind equalizer can automatically find the optimal tap coefficients using the constant amplitude. In this article, the use of CMA can be summarized for two purposes: one is to pre-converge data for next stage NLE processing and small tap numbers are used, but with relatively high BER, the other is to get FIR in order to be used for signal Pre-EQ at the Tx, as described in Section A. The Volterra NLE based on the 2nd-order Volterra theory is used to compensate device or CD induced linear and nonlinear impairments [24]. The tap coefficients are initially updated by the training symbols according to the LMS optimization.
As a proof-of-concept stage, the evolution of LAN-WDM to B5G/6G fronthaul network is difficult to realize due to the linear impairments from the strong bandwidth constraint of the optoelectronic devices and fiber CD and the nonlinear impairments from the nonlinear region of the electro-optical components. In order to reduce the impairments of the limiting the system performance, we take appropriate action to ensure that the system has the significant BER performance, such as PAM-8 modulation format, Pre-EQ, PS, clipping, high-bandwidth devices with better linearity, and SOA pre-amplifying.

A. Laser and SOA Performance Optimization
By varying shaping factor of Maxwell-Boltzmann (MB) distribution, PS-PAM-8 symbols with different entropies (3 bit/symbol, 2.93 bit/symbol, 2.71 bit/symbol, 2.5 bit/symbol, and 2.17 bit/symbol) are generated. We keep the same total data rates (the entropy × baud rate) throughout the whole experiment, for 92-Gbaud PS-PAM-8, the corresponding total data rates are 276 Gb/s, 270 Gb/s, 250 Gb/s, 230 Gb/s, and 200 Gb/s, respectively. Bandwidth limitation of cascade devices, such as AWG,  electrical driver and modulator, is one of the most important problems for LAN-WDM to satisfy B5G/6G Terabit/s transmission. Pre-EQ is an effective method to solve this problem. We use one VOA between the SOA and the PD to adjust the input power into PD to emulate the receiver performance, and keep the injection power into the PD successively reduced by about 2 dB by adjusting the VOA. The BER performance versus input power into PD for BtB case w/o and w/ Pre-EQ or clipping process in the channel 8 is measured and shown in Fig. 9. The 92-Gbaud PS-PAM-8 signals after processing without Pre-EQ or clipping, and with both Pre-EQ and clipping under BtB case are all tested. First, without Pre-EQ and clipping, the system performance suffers severe degradation caused by ISI under bandwidth constraint as analyzed in Section III. Then, with Pre-EQ and clipping, the system performance is significantly improved. The input power into PD for 92-Gbaud PS-PAM-8 under BtB case is approximately −6.8 dBm in the channel 8 at SD-FEC threshold (2 × 10 −2 ).
Moreover, in order to optimize the parameters of AWG and SOA, we investigate the BER performance with different AWG and SOA operation conditions. Fig. 10 illustrates the BER versus AWG output voltage in the channel 1 and 8, respectively. Based on the BER results, when the optimal drive voltage values both are 270 mV for the channel 1 and channel 8, we can gain the optimal system performance. It shows that the optical AWG output voltage by channel 1 for transmission is the same as by Fig. 11. Measured BER versus SOA current in the channel 1 and 8 after BtB case, respectively. channel 8. The laser has the same effect on different wavelengths. To further improve the system power budget, a SOA at O-band is used to pre-amplify the optical power before PIN-PD detecting for better performance, as described in Section II. For longer distance transmission, SOA usually works with high operating current to obtain high optical power gain. However, the higher saturation current induces higher amplified spontaneous emission (ASE) noise, and which could cause SOA work in nonlinear region so that the system performance will be even worse and may lead to signal quality degradation [22]. In general, the ASE noise can be suppressed by an optical filter, but it will increase the system overhead due to its extra insert loss and larger size [25], [26], [27], and so no optical filter is used in this experiment. Therefore, it is necessary to optimize the parameters for SOA in such high-speed transmission system. We measure the BER performance with 200 Gb/s PS-PAM-8(2.17 bit/symbol) in the channel 1 and 8 at an input power into PD of −6.8 dBm after BtB case with different working current, which are presented in Fig. 11. The BER performance is more susceptible to noise for small input power into PD. There is a trade-off between noise and nonlinear impairments. Based on the BER results, the optimal operating current values are 110 mA and 60 mA for the channel 1 and channel 8, respectively. Then, they are used for the tests. Fig. 11 shows that the optimal working current of the SOA by channel 1 for transmission is higher than by channel 8. The gain of SOA is not flat under different wavelengths, so we optimize the parameter of SOA for different wavelengths. It has been demonstrated that the gain provided by the SOA are also different for different wavelengths. In addition, for different link lengths, the choice of SOA uniform parameters with slight performance penalty is sufficient for 5 km transmission range for most scenarios for dense mobile base stations deployment.

B. Transmission Rate
Based on the optimized parameters of the devices, we measure the BER performance versus different bit rates for 92-Gbaud PS-PAM-8 with Pre-EQ and clipping at an input power into PD of −6.8 dBm after BtB case, 5 km fiber and 10 km fiber in the channel 1 and 8, respectively, as shown in Fig. 12. We can observe that 200 Gb/s, 230 Gb/s, 250 Gb/s and 270 Gb/s PS-PAM-8 can be achieved after 10 km fiber under 2 × 10 −2  The maximum transmission bit rate varies in different channels. Bit rate is limited by achievable component bandwidth (BW) and the dispersion of different wavelengths. It is possible to reduce component BW requirement with Pre-EQ. The effect of the Pre-EQ has been discussed in more detail in Section A. With the development of DSP technology and continual advances in optical and electrical technology, the use of current commercial devices combined with high-efficiency DSPs to achieve better transmission rates can tremendously reduce device costs, power and integration level and increase in density in the future [12], [28], [29].

C. Transmission Distance and Dispersion Tolerance
In addition, in order to investigate the dispersion tolerance for different wavelengths other than the near-zero-dispersion one used in this research study, we further measure the BER versus different G.652 type fibers length for 250 Gb/s PS-PAM-8 (2.71-bit/symbol) at an input power into PD of −6 dBm in the channel 1 and 8, as shown in Fig. 13. Considering the 20% SD-FEC limit (2 × 10 −2 ), the transmission length of 250 Gb/s PS-PAM-8 in the channel 1 is 5 km. It can be transmitted for 20-km through the channel 8 under SD-FEC threshold (2 × 10 −2 ). The channel 8 indicates better performance than the  channel 1 in terms of transmission distance. It has been demonstrated that the suffering effects of the chromatic dispersion are different for various channels. The maximum absolute value of the chromatic dispersion of the channel 1 is 99.2 ps/nm, and the dispersion of channel 8 over 20-km fiber is nearly 24.6 ps/nm. The CD after the channel 8 transmission has less impact on the system performance due to near zero-dispersion at O-band than the channel 1. In all, the closer the wavelength is to 1310 nm, the less chromatic dispersion (positive or negative) affects according to the ITU-T Recommendation G.652 standard [13], [14].
Even if the experiment adopted 8-channel MUX and DE-MUX, the transmission feasibility of the remaining four channels could also be evaluated by the dispersion index. Fig. 14 exhibits the maximum and minimum chromatic dispersion versus fiber length for the channel 9 (1269.23 nm) and 12 (1332.41 nm) according to the ITU-T Recommendation G.652 standard [13], [14]. Among them, the wavelength values of the channel 9 and 12 correspond to the minimum and maximum wavelengths of the remaining four CWDM channels, respectively. The standard provides chromatic dispersion coefficient at intervals of 10 nm. The wavelengths of channel 1 and channel 9 are both closer to 1270 nm. Therefore, in our experiment, we use the chromatic dispersion coefficient at 1270 nm for both channel 1 and channel 9. Based on the results in the Fig. 15, the chromatic dispersion values of the two channels are within the range (±99.2 ps/nm). Therefore, it is completely feasible to reach the total rate of 3Tb/s (250 Gbit/s × 12) over 5 km fiber transmission with 12 wavelengths, and which can meet the channel transmission requirements and support B5G/6G compatible coexistence in the future. Meantime, considering 3 radio remote unit (RRU) antennas for B5G and 3 AAU antennas for 6G for a base station, a total of 6 pairs of optical modules (OM) for 12 wavelengths are needed, as shown in Fig. 1. . We can find that there is no problem with 8 channels transmitting 5 km simultaneously. The ability of the last four channels (channel 5 to 8) to resist dispersion is stronger after 15 km and 20 km fiber under 2 × 10 −2 threshold and the channel 8 shows better performance than other channels with the increase of transmission distance. For future B5G/6G Terabit/s fronthaul networks, 5 km range is sufficient for most scenarios for dense mobile base stations deployment. Fig. 17 shows the different investigated graphical illustrations of probabilities and the eye-diagrams of the received signals for 92-Gbaud PS-PAM8 signals with different entropies (2.5 bit/symbol and 2.71 bit/symbol) over 5-km fiber for the channel 8.

V. DISCUSSIONS
Evolution schemes: IM/DD and coherent schemes are the two main kinds of promising mobile x-haul evolution solutions. As the demand for high-speed metro and access optical transmission growing, the application potential of IM/DD systems based optical communication technology is more obvious owing to their cost effectiveness, simpler configuration and low-power consumption. Therefore, IM/DD has more mature commercial deployment applications for short-reach applications. However, some trends indicate that the IM/DD data rate capacity is capped at 100 Gbps and can reach more than 400 Gbps for complex IM/DD systems, but which can result in displaying inappropriate sensitivity values. In addition, coherent scheme can also be introduced due to its superior receiver sensitivity and frequency selectivity. Cost, complexity, and high-power consumption are among the challenges that significantly limit the commercial application [30], [31], [32], [33], [29]. Therefore, it is essential to solve the problems of introducing coherent optics into fronthaul network. Moreover, IM/DD would might be needed for considering 200 Gbps/λ optical network in the face of operating in the O-band [3], which could be because dispersive effects become just too severe for IM/DD to handle. In this work, 250 Gbps/λ optical network has been proven possibly in laboratories also with IM/DD using advanced DSPs.
DSP complexity, power consumption and cost: With the development of DSP technology, the use of current commercial devices combined with high-efficiency DSPs to achieve better transmission rates can reduce device costs and integration level. However, at high symbol rates, bandwidth-limited IMDD systems suffer from the strong intersymbol interference (ISI) that requires powerful equalization at the receiver. Therefore, several modified efficient equalizations with lower operational complexity are proposed and demonstrated [34]. The DSP application-specific integrated circuits (ASIC) power is nominally about 50% of the optics module. Fortunately, Moore's law promises the permanent reduction of ASIC power consumption [35]. 800G 7-nm PAM4 DSP chip has been developed and demonstrated in 800 G QSFP-DD Transceiver [36]. The power consumption of the QSFP-DD transceiver is 14.43 Watts with 3.47 V input voltage at 70 degrees' Celsius case temperature. Meanwhile, broadband electrical and optoelectronic components are being designed and developed to facilitate high symbol rates transmission [37]. By implementing DSP and coding techniques with lower complexity, the transmission performance and spectral efficiency are optimized. In addition, the ASIC power consumption depends on the target application and operation environment. The choice of DSP parameters is sufficient for 5 km transmission range for most scenarios for dense mobile base stations deployment. It's trade-off in the design of lower complexity, lower power DSP ASIC's for direct detect and coherent, pluggable optical modules for 5/6G mobile X-haul applications.

VI. CONCLUSION
In this work, a LAN-WDM IM/DD access system for the evolution of future mobile fronthaul networks with low cost using bandwidth-limited optics has been experimentally demonstrated based on joint equalization techniques, and successfully achieved a record bit rate of 2 Tb/s (250 Gbit/s × 8) PS-PAM-8 signal with 1.6 Tb/s net rate over 5 km fiber transmission under 20% SD-FEC threshold (2 × 10 −2 ) with the aid of SOA, PS technique, Pre-EQ, clipping and advanced DSPs. It is the first time to demonstrate the LAN-WDM can meet the evolution of B5G/6G Terabit/s fronthaul networks, which can effectively utilize the limited resources and improve the performance of the fronthaul network, making it a promising scheme with high capacity, low cost, and mature infrastructure. It is foreseeable that further innovations and experimental investigation will be made in this area to make the technology and industry chain more mature.