1. INTRODUCTION

5G wireless communication service has been commercialized since its initial phase. Meanwhile, in the decade of 2021–2030, lots of research and development (R&D) for beyond-5G (B5G)/6G architectures, systems, and algorithms are being or will be carried out, providing technological candidates for standardization and deployment in order to better handle the traffic demand from futuristic applications of metaverse, holographic communications, extended reality (XR), digital twins, and so on [1]. Within these R&D activities, one important topic would be on the fronthaul (FH) segment in the cloud radio access network (C-RAN), which is responsible for efficient radio data delivery inside the functionally split beyond-5G (B5G)/6G base stations, i.e., between distributed units (DUs) and massive radio units (RUs). In the 5G initial phase, low-layer split (LLS) Option 7, 25G enhanced Common Public Radio Interface (eCPRI) interface, and possibly wavelength division multiplexing (WDM) optics may be mainly adopted for FH [2–4]. Nevertheless, considering the traffic growth due to higher frequency bands, increased antenna ports, and denser cell deployment, R&D of more advanced FH transport systems is still needed. Different from those for long-haul networks, the techniques for FH segment need to be high speed and low complexity simultaneously. Low latency is also crucial, the constraint of which depends on different 6G RAN protocols and services.

In particular, at large cell sites with multiple RUs in 6G, the aggregated FH traffic (CPRI-equivalent rate) might exceed 1 Tb/s or even 10 Tb/s [2]. Even after applying FH data compression techniques such as analog-to-digital-compression radio-over-fiber (ADX-RoF) [5–7], Tb/s may be inevitable. WDM with intensity-modulation direct-detection (IM-DD) is one potential choice for cost-effective high-capacity transport; C-band ${20} \times {25}\;{\rm Gb/s}$ over 10 km has been demonstrated by industry as a solution to 5G FH [8]. Looking forward to 6G FH, it is likely that the line-rate of each channel would be 100 Gb/s, 200 Gb/s, or even more [9,10]. In this high-bandwidth signal case, fiber chromatic dispersion (CD) in the C-band and even edge wavelengths of the O-band become a major issue as each wavelength channel could occupy 50 GHz or wider bandwidth.

In addition, while the reach of FH in the 5G initial phase may be around 10 km [3], the reach of 6G FH can potentially be 20 km and even 40 km, for example, in the scenarios of (i) routing (e.g., optics-based [11–13]) of FH traffic to/from a non-nearest DU for better coordinated processing [14] or due to content caching [15] and (ii) rural connectivity [16]. Therefore, reach extension and/or CD compensation techniques for the high-speed IM-DD FH are highly desirable. The technique and system could be applied to midhaul transmissions as well where the aggregated data rate is also demanding [17] and typical distance is longer than FH [3]. Illustrations of FH for the 5G initial phase and envisioned 6G FH are shown in Fig. 1.

 

Fig. 1. Illustration of (a), (b) fronthaul for the 5G initial phase and (c) envisioned 6G high-speed reach-extended fronthaul. FSO, free-space optical communication.

Download Full Size | PDF

Requirements for such reach extension include low complexity/cost and low latency overhead. Traditional dispersion compensation fiber (DCF) becomes less compelling considering latency. Optical approaches for reach extension have been discussed in the context of metro and access networks, equalizing the CD impact by either optical infinite impulse response (IIR) filters or optical finite impulse response (FIR) filters [18–23]. However, it is very challenging to have error-free performance for the high-speed system like 200 Gb/s using only optical equalizers.

On the other hand, a number of digital signal processing (DSP)-based CD compensation techniques have been discussed recently, such as the use of a nonlinear feedforward equalizer (FFE), a decision feedback equalizer (DFE), maximum-likelihood sequence estimation (MLSE), and judicious pulse shaping. 200 Gb/s IM-DD pulse amplitude modulation (PAM) transmissions over 20 km [with hard-decision forward error correction (HD-FEC) or 50G-PON FEC] and 40 km (with soft-decision FEC) in the low-CD O-band were reported [24–27]. In the C-band case, 120 Gb/s pulse-shaped PAM4 over 100 km was studied [28]; however, by far only a few works attempted to address a challenging condition of CD (e.g., in the C-band) in beyond-${200 \text{-Gb/s/}}\lambda$ IM-DD transmission with clearly disclosed DSP complexity [29,30]; dual RF chains or a costly sharp optical filter were additionally needed. Also, the achievable performance depends heavily on the DSP complexity, which may be calculated by the number of multiplications per symbol (NMPS) and number of taps/registers per symbol (NTPS). The NMPS and NTPS of a linear FFE, a nonlinear FFE, a DFE, and MLSE are summarized in Table 1.

Table 1. Complexity of Different FFEs, DFEs, and MLSEs^a$^{^,}$^b

View Table | View all tables in this article

For a nonlinear FFE, a second-order Volterra FFE (VFFE) and a low-complexity simplified polynomial FFE (SPFFE) utilizing absolute operation [31] are considered, the equalizer output of which, {${y_i}$}, is given by Eqs. (1) and (2), respectively ({$w_k^{(1)}$} and {$w_k^{(2)}$} denote equalizer weights, while {${x_i}$} denotes equalizer inputs):

For example, even for O-band 4 km transmission of 200G PAM4, the use of MLSE with NMPS of 128 and NTPS of 256 (i.e., $L = {2}$) was suggested [33]; thus, if 20 km or longer reach is targeted, the complexity of MLSE would be much increased. Currently, the NMPS and NTPS discussed for the intra-datacenter standardization may be slightly larger than 128 and 256, dominated by MLSE. In academia, the Rx-side NMPS of C-band 200G beyond-10-km IM-DD works were ${\rm NMPS} \gt {2000}$ [29] and ${\rm NMPS} \gt {5000}$ [30], implying the limited practicality. Furthermore, it is worth noting the difficulty of implementing DFE circuits with a large number of feedback-filter taps and high bandwidth; industry prefers that the number of feedback taps in a DFE is 2 or less [34,35]. Overall, it is challenging to handle high CD in 200G IM-DD systems with low-complexity digital equalizers.

Recently, we theoretically analyzed and performed proof-of-concept demonstration of an optoelectronic feedforward equalization (OE-FFE) approach for low-complexity adaptive CD compensation [36]. The low-cost optical part [a 1-tap optical delay line (ODL)] has ps-level latency, which drastically reduces the complexity of digital equalizer and also enables circuit-friendly all-FFE architecture. The 3.8 THz dense WDM (DWDM) grid in the C-band can be supported by a single 1-tap ODL and ${\sim}{30}$-tap FFE for 100G PAM4 over beyond 40 km [37]. However, the support of higher rates such as ${200}{\rm G}/\lambda$ as well as WDM operation are yet to be verified and thoroughly investigated, which are of potential interest for 6G advanced FH applications.

This work is extended from our 2023 Optical Fiber Communication Conference (OFC) paper [38]. In [38] (and [39]), we focused on different wired-wireless conversion/converging (WWC) techniques for FH (namely, DSP-enhanced RoF) and their performance, latency overhead, and power consumption. This work will extend to the complementary side, namely, the high-speed optical transmission. Considering 6G capacity and cost requirement, ${200}\;{\rm Gb/s/}\lambda$ based on IM-DD PAM4 and low-complexity DSP is targeted, which transparently supports digital RoF techniques like ADX-RoF [3rd Generation Partnership Project (3GPP) split Option 8] and Delta-sigma RoF [40–44], while also supporting LLS-based FH and high-layer split (HLS) data in the midhaul. Moreover, it has good compatibility with cost-effective ubiquitous Ethernet technology such as 200GE. We will show by extensive experiments how the OE-FFE-based system enables a minimal-complexity digital equalizer (NMPS down to 82) and HD-FEC codes even in the C-band and address the reach extension challenge in $200{\rm G/}\lambda$ fronthauling. Table 2 compares the achieved results with prior works. We also discuss the CPRI-equivalent rates and peak wireless data rates (PWDRs) supported by the systems, assuming two typical digital RoF technologies.

Table 2. Recent Works of C-Band 200G IM-DD PAM Systems Using a Single MZM & PD^a

View Table | View all tables in this article

The rest of this paper is organized as follows. We provide in detail the experimental investigations single-channel 200G fronthauling up to 41.38 km in the C-band in Section 2. DWDM experiments are reported in Section 3. The supported CPRI-equivalent rate and PWDR are presented in Section 4. Finally, the work is concluded in Section 5.

2. LOW-COMPLEXITY SINGLE-$\lambda$ 200G FRONTHAUL UP TO 41 KM BASED ON OE-FFE

A. Single-Channel 200G over Beyond 40 km in the C-Band

We transmitted 224 Gb/s PAM4 over 41.38 km single-mode fiber (SMF)-based FH enabled by the OE-FFE technique. Figure 2(a) shows the experimental setup. At the transmitter side, 112 GBd electrical PAM4 signals were generated from an arbitrary waveform generator (AWG) (Keysight M8199A) operating at two samples per symbol. Neither digital pulse shaping nor pre-equalization was applied. This lowers the system cost and power consumption as a high-resolution wideband digital-to-analog converter (DAC) can be omitted. The peak-to-peak voltage of the signals at the output of the AWG was 800 mV. The signal was modulated onto a 100-kHz-linewidth optical carrier at 193.8 THz (${\sim}{1547}\;{\rm nm}$) via a single-drive Mach–Zehnder modulator (MZM) (3 dB bandwidth about 30 GHz, half-wave voltage ${\rm V}\pi = {1.9}\;{\rm V}$) biased around its quadrature point. The optical double-sideband (DSB) PAM4 signal was then transmitted over 41.38 km SMF. At the receiver side, an erbium-doped fiber amplifier (EDFA) was used to compensate for the fiber loss. The PAM4 signal was processed by a 1-tap ODL based on free-space components with delay (T) of about 5 ps (free spectral range $\approx {200}\;{\rm GHz}$). The 5 ps delay was chosen based on theoretically predicted system performance [36] and the compatibility with the standard WDM grid of 200 GHz. Subsequently, the signal was power-attenuated, detected by a 50 GHz PD, and captured by a 256 GS/s analog-to-digital converter (ADC) (Keysight real-time oscilloscope). The receiver-side offline processing includes resampling to two samples per symbol (SpS), a 7-tap low-pass filter (LPF) (the filter order was 6; the passband and stopband cutoff frequencies were 54 GHz and 65.88 GHz), downsampling to 1 SpS with synchronization, a symbol-spaced FFE (via recursive least squares adaptation with a 4600-symbol pilot), and PAM demodulation. The bit error rate (BER) was calculated by direct error counting (${\gt}{0.5}$ million bits counted for PAM4).

 

Fig. 2. (a) Experimental setup of single-channel 200G PAM4 over 41 km SMF. $\varphi$, phase shift; T, time delay. (b) Analytical and measured frequency responses. (c) BER versus the number of linear taps in the SPFFE and the second-order VFFE, when the PD input power was 6dBm. N_NL, number of nonlinear taps. (d) BER versus the PD input power. Inset (i) shows an amplitude histogram of equalized PAM4 @ ${+}{6}\;{\rm dBm}$ PD input; inset (ii) shows the corresponding evolution of normalized mean squared error (MSE).

Download Full Size | PDF

Figure 2(b) shows the channel frequency response estimated from the received 224 Gb/s PAM4 signal without and with the 1-tap ODL, respectively, after 41.38 km transmission. The deep spectral notches/zeros caused by the CD occurred at RF frequencies $f$ satisfying $\pi {\lambda ^2}{\textit{DL}}{f^2}/c =\def\LDeqbreak{}({2p - 1})\pi /2\;({p = 1,2, \ldots})$ ($\lambda$, wavelength; D, dispersion coefficient; L, SMF length; $c$, speed of light). It is confirmed that the 1-tap ODL successfully avoided the deep notches. The residual spectrum fluctuations or distortions are below 15 dB, which were handled by the following digital FFE. The frequency responses given by our theoretical model (${D} = {16}\;({\rm ps/nm} *{\rm km})$, $\varphi = {0.62}\pi$) are also plotted. The experimental frequency response matches well with theoretical prediction, except the slight decay in $f \gt {45}\;{\rm GHz}$ (which might be due to the bandwidth of the MZM).

Figure 2(d) depicts the BER versus the PD input power [or the received optical power (ROP) defined at the PD input] after 41.38 km transmission. We found that a nonlinear FFE was needed and applied an SPFFE. Its advantage is that the complexity of each nonlinear tap is almost the same as that of each linear tap (only an additional absolute operation is needed, which is simply a XOR operation on the sign bit in the digital circuit). Enabled by a 1-tap ODL and an SPFFE with 103 linear taps and 33 nonlinear taps (${\rm NMPS} = { 136}$; ${\rm NTPS} = {103}$), a 6.25% staircase HD-FEC-compliant [45] BER was achieved when the PD input power was ${+}{6}\;{\rm dBm}$ or higher. For reference, we also plot the threshold of a low-latency, low-power concatenated KP4-Hamming FEC [46], which was originally discussed for 200G/lane intra-datacenter scenarios. Figure 2(c) shows the impact of different numbers of linear and nonlinear taps. The performance of OE-FFE with a second-order VFFE was additionally depicted. When the second-order VFFE also had ${{\rm N}\_{{\rm NL}}} = {33}$ as in the SPFFE, lower BERs were achieved yet at the cost of excessive complexity (${\rm NMPS} \approx {1200}$); when the NMPS of the VFFE was comparable to that of the SPFFE (i.e., ${{\rm N}\_{{\rm NL}}} = {6}$), the BERs were higher. This indicates the merit of using an SPFFE in OE-FFE. The relatively high PD input power was because of the lack of a high-bandwidth transimpedance amplifier (TIA) after the PD. Potential solutions to increase loss budget of the FH system include adding a TIA, using optical amplifiers integrated in the receiver, adopting an avalanche PD, incorporating FEC with higher coding gain, and so forth [24,25]. The ODL with the same 5 ps delay also supports other fiber lengths, as shown next.

To our knowledge, this is among the first demonstrations of C-band IM-DD $200{\rm G/}\lambda$ over beyond 40 km with low-complexity optics (a 1-tap ODL) and low-complexity digital equalizer simultaneously. Remarkably, the NMPS achieved is as low as that of MLSE used in merely O-band 4 km transmission [33] and is 90% lower than those in [29,30], although the achieved reach is fourfold. The system can be useful for next-generation FH as well as midhaul applications.

B. Fiber-Amplifier-Free 200G over 20 km

Fiber-amplifier-free high-speed fronthauling may often be of interest in order to have small-footprint deployment and connections in 6G. The C-band has 40%–50% lower loss than the O-band, but CD represents a much more severe issue, which can be addressed by OE-FFE. We also tried this scenario. The experimental setup was similar to Fig. 3(a), except that the fiber length was 20 km and the EDFA was removed. The launched power was 9 dBm (emulated with an EDFA, since the output power from the MZM was about 6 dBm with 13 dBm available laser). In practice, the 9 dBm power can be obtained by inputting higher laser power such as 16 dBm, or by using a semiconductor optical amplifier [47–49]. In order to compensate for the lack of a high-bandwidth TIA and to improve receiver sensitivity, an electrical amplifier with 60 GHz bandwidth, 22 dB gain, and 6–7 dB noise figure was employed after the PD. The 1-tap ODL here has a delay ${T} \approx {5}\;{\rm ps}$ and about 5.5 dB loss.

 

Fig. 3. Experimental results of fiber-amplifier-free C-band 200G 20 km transmission by OE-FFE. (a) BER versus the number of linear and nonlinear taps in an SPFFE. (b) BER versus the PD input power.

Download Full Size | PDF

 

Fig. 4. Experimental setup of ${4} \times {200{\rm G}}$ C-band WDM fronthaul based on OE-FFE. PM, polarization-maintaining; SMF, single-mode fiber. Inset (i) shows the optical spectra of four wavelength channels launched into the fiber. Inset (ii) shows the optical spectra after 20 km of SMF and a 200 GHz 1-tap ODL. Inset (iii) shows the optical spectra of four channels after a 200 GHz demultiplexer (DeMux).

Download Full Size | PDF

Figure 3(a) shows the system BER versus the number of linear taps in an SPFFE when the PD input power was 0 dBm. Different numbers of nonlinear taps were tested. It was found that an SPFFE with 69 linear taps and 25 nonlinear taps (${\rm NMPS} = {94}$; ${\rm NTPS} = {69}$) was sufficient to have a BER performance well below the 11% HD-FEC limit [45], while more taps brought only slight performance gain. Thus, in the following measurements, we applied an SPFFE with 69 linear taps and 25 nonlinear taps. Figure 3(b) plots the measured BER versus the PD input power, showing a receiver sensitivity of 0 dBm. The performance might be improved if a TIA with less noise was available.

3. ${4} \times {200{\rm G}}$ WDM FRONTHAUL BASED ON OE-FFE WITH A SHARED 1-TAP ODL

To show the potential of FH capacity expansion together with reach extension, we experimentally demonstrated ${4} \times {200{\rm G}}$ DWDM transmission over 20 km based on OE-FFE. Figure 4 shows the experimental setup. Here we omit the discussion of detailed wavelength/frequency plan for the system and just take wavelengths around 1550 nm for instances. Moreover, we show ${4}\lambda$ demonstration as an example considering lab availability, which is expected to be scalable, e.g., to $8\lambda {-} 10\lambda$ (at 200 GHz spacing) at downlink/uplink each [3]. The four 112 GBd PAM4 signals were centered at 193.4 THz (${\sim}{1550.1}\;{\rm nm}$), 193.2 THz (${\sim}{1551.7}\;{\rm nm}$), 193.0 THz (${\sim}{1553.3}\;{\rm nm}$), and 192.8 THz (${\sim}{1554.9}\;{\rm nm}$) with 200 GHz spacing. At the transmitter side, 112 GBd PAM4 signals were generated from a 224 GS/s AWG outputting at two SpS. No digital pulse-shaping or pre-equalization was applied. The only DSP was to slightly pre-distort the four amplitudes from ${\rm \{} - {3},- {1},{1},{3\}}$ to $\{- {3},- {0.8},{1.1},{3\}}$ to improve system performance. The peak-to-peak voltage (Vpp) of the electrical driving signal was 850 mV.

To emulate the generation of independent four-wavelength signals, the “odd-and-even” method for de-correlation was applied. The 193.0 THz and 193.4 THz laser outputs are combined by a polarization-maintaining (PM) coupler and are fed into MZM $A$ with about 30 GHz 3 dB bandwidth and ${\rm V}\pi = {1.9}\;{\rm V}$. Then, they were modulated by one channel of AWG output. The 192.8 THz and 193.2 THz laser outputs are combined by another PM coupler and are fed into MZM $B$ with about 20 GHz 3 dB bandwidth and ${\rm V}\pi \lt {3.5}\;{\rm V}$. They were modulated by another channel of AWG output with a different data pattern. The MZM inputs were swapped when the performance of 192.8 THz and 193.2 THz channels were to be measured.

 

Fig. 5. Experimental results. (a) Estimated frequency response of four channels. (b) BER of the 192.8 THz channel versus the number of linear and nonlinear (N_NL) taps in an SPFFE. (c) BER versus the PD input power of four channels. (d) Amplitude histograms of four equalized channels.

Download Full Size | PDF

Before FH transmission, we emulated a total launched power of 8.5 dBm. This corresponds to a per-channel power of 6 dBm (assuming 3.5 dB insertion loss of the DWDM multiplexer), which can be readily met by MZM $A$ with 13 dBm laser in the current setup. After fiber transmission, the four channels were processed by a single 1-tap ODL, amplified, and wavelength-demultiplexed. We used a 200 GHz tunable optical bandpass filter (OBPF) for demultiplexing. A 50 GHz PD and a 256 GS/s ADC were used to receive and digitize each wavelength channel. The receiver-side offline DSP is similar to that described in the previous section, except that a 9-tap LPF at 3 SpS (filter order was 8; the passband and stopband cutoff frequencies were 56 GHz and 68.32 GHz) was applied.

Insets (i)–(iii) of Fig. 4, respectively, show optical spectra of four wavelength channels launched into the fiber, optical spectra after the 200 GHz ODL, and optical spectra of four channels after demultiplexing. Note that, although the optical spectra at the output of the 1-tap ODL may be reminiscent of single-sideband (SSB), the design objective is different: there’s no requirement or effort needed to maximize the sideband rejection (as can be seen, the vestigial part is clearly visible), which would require complicated and costly optical filters; instead, the objective is to remove the deep notches in the baseband response by the 1-tap ODL and leave the residual fluctuations to the adaptive digital FFE.

Figure 5(a) shows the frequency response estimated from the received signals at four wavelengths with the 1-tap ODL, respectively, after 20 km transmission. The response of the signal at 193.0 THz without the 1-tap ODL is also plotted. Nine deep spectral notches can be seen, while a single 1-tap ODL successfully avoided all of them at all wavelengths. Also, the experimental frequency responses match with theoretical prediction. Figure 5(b) shows the BER of the 192.8 THz channel (the worst-performance channel) versus the number of linear/nonlinear taps in an SPFFE; 59 linear and 23 nonlinear taps (${\rm NMPS} = { 82}$; ${\rm NTPS} = {59}$) were applied in the subsequent measurements according to the result. Figure 5(c) shows the BER of four channels versus the PD input power. All four channels reached 11.1% HD-FEC threshold at a PD input power of ${\lt}{4}\;{\rm dBm}$, achieving ${200} + {\rm Gb/s}$ per-channel data rate excluding FEC overhead. Besides, the channel at 193.4 THz reached KP4-Hamming FEC limit. Figure 5(d) depicts PAM4 amplitude histograms of equalized four channels. The shapes of equalized PAM4 histograms were slightly deviated from the ones assuming Gaussian noise, which is due to (1) the square-law detection after optical amplifier which changed the noise distribution from Gaussian distribution to Chi-square distribution [50,51] and (2) further interaction with the 1-tap ODL and the nonlinear digital FFE.

To our knowledge, this is among the first demonstrations of IM-DD DWDM ${200{\rm G}}/\lambda$ over 20 km in the challenging C-band with cost-shared low-complexity optics (a 1-tap ODL) and a low-complexity digital equalizer. The system can potentially scale to more wavelengths.

4. DISCUSSION ON THE SUPPORTED CPRI-EQUIVALENT RATE AND PEAK WIRELESS DATA RATE

Considering Option-8-based FH, here we analyze the supported CPRI-equivalent rate and PWDR by the demonstrated systems. Two examples of digital RoF technology are considered, namely, space-time ADX-RoF [5,38,39] and Delta-sigma RoF [41]. It is assumed the HD-FEC ensures an error-free FH. Figure 6 shows experimentally measured radio signal quality (characterized by error vector magnitude or EVM) versus bandwidth conversion efficiency (BCE) of space-time ADX-RoF [39] and Delta-sigma RoF techniques [41]. BCE is defined as transported radio signal bandwidth divided by optical transceiver bandwidth [38,39] (for optical PAM, its baudrate is assumed as bandwidth). Note that the BCEs in Fig. 6 assume optical on-off keying modulation; BCE is doubled when optical PAM4 is used.

 

Fig. 6. Experimental radio signal quality versus BCE [38,39] of space-time ADX-RoF and Delta-sigma RoF [41].

Download Full Size | PDF

Table 3. Supported CPRI-Equivalent Rate and Peak Wireless Data Rate^a

View Table | View all tables in this article

CPRI-equivalent rate here is calculated as 40 times of aggregated radio signal bandwidth, assuming that each I/Q sample is represented by 15 bits plus 1-bit control word, with coding overhead of 10/8. Note that no radio signal oversampling is assumed, corresponding to the lower bound of rate estimation.

As for the PWDR, since the calculation method for 6G is not yet defined, we estimate the PWDR based on the formula in Section 4.A.2, TS 38.306, of the latest 3GPP Release 17 [52]:

In the case of FH with space-time ADX-RoF, by leveraging the ${4} \times {200{\rm G}}$ WDM system, a maximum of a 10.4 Tb/s CPRI-equivalent rate can be achieved (with 16-QAM radio modulation). Note that the asymmetric MIMO feature (i.e., the number of MIMO layers is assumed to be one-fourth of the number of antenna ports) was fully leveraged during ADX processing. If we consider some margin for practical scenarios, e.g., BCE is compromised by 50%, each 200G channel still supports a ${\gt}{1}\;{\rm Tb/s}$ CPRI-equivalent rate with 64-QAM/16-QAM. Meanwhile, the PWDR of more than 300 Gb/s is achievable with 1024-QAM; this potentially meets the 6G target peak rate of ${\gt}{100}\;{\rm Gb/s}$ [53] and even Tb/s-level [54] with WDM channel expansion.

In the case of FH with Delta-sigma RoF, considering its feature of high oversampling ratio, the BCE is usually lower than ADX-RoF. Nevertheless, with optical PAM4, a 1.28 Tb/s CPRI-equivalent rate and ${\gt}{190}\;{\rm Gb/s}$ PWDR can be achieved assuming 256-QAM radio modulation.

5. CONCLUSION

With the continuous demand from wireless connectivity as well as the emergence of futuristic applications, FH networks are expected to be increasingly important in 6G, which face challenges in capacity and complexity/cost. As extended investigations to our OFC2023 paper, this paper focuses on the high-speed, transparent, and low-complexity optical transmission techniques for 6G FH. Advanced scenarios of reach-extended (20 km and even 40 km) ${200}\;{\rm Gb/s/}\lambda$ IM-DD FH have been discussed with experimental results. Notably, based on the OE-FFE technique, we have achieved up to 41 km transmissions of 200G single-$\lambda$ and WDM signals in the C-band with negligible delay of optics (several picoseconds) and a digital equalizer complexity at most comparable to that of O-band 4 km transmission in standardization discussion [33]. Even with 3GPP split Option 8 and extended reach, the FH systems support CPRI-equivalent rates up to 10 Tb/s and a ${\gt}{300}\;{\rm Gb/s}$ PWDR, respectively, potentially meeting the 6G objective. The presented systems can be promising technological candidates for high-speed, reach-extended 6G fronthauling requiring low complexity. Moreover, OE-FFE-based IM-DD could be potentially useful in other short-reach scenarios targeting high capacity and low complexity, such as midhaul transport, campus/inter-datacenter networks [24,30,55], and optical access networks [10], provided that application-specific requirements (e.g., on the transceiver density, loss budget, and/or temperature tolerance [56]) are met.

IM-DD and coherent schemes are two main evolution schemes for mobile x-haul. Coherent technology has superior receiver sensitivity and can fully compensate for the CD. When applying it for RAN scenarios, the system complexity, cost, and power consumption should be effectively reduced [17,57–59]. Our scheme keeps the simple architecture of IM-DD while extending the reach by incorporating OE-FFE; this technological alternative may co-exist with (or complement) simplified coherent solutions in the next-generation x-haul depending on the scenario-specific requirements.