112.5 Gbit/s Long-Reach Passive Optical Network with >31 dB Power Budget Enabled by a Semiconductor Optical Amplifier

112.5 Gbit/s Long-Reach Passive Optical Network with >31 dB Power Budget Enabled by a Semiconductor Optical Amplifier

---- Translated from the work by Ahmed Galib Reza et al.

Abstract

We experimentally demonstrate downstream transmission of 112.5 Gbit/s pulse amplitude modulation (PAM) signals in the O-band for future time-division multiplexed long-reach passive optical networks (LR-PONs). In this work, a commercial quantum-well semiconductor optical amplifier (QW-SOA) is deployed at the remote node (RN) of a 100G-class commercial PON for the first time to extend the transmission distance between the optical line terminal (OLT) and optical network units (ONUs). Furthermore, we investigate the performance of the QW-SOA when amplifying continuous-mode downstream PAM signals with uniform probabilistic shaping (PS), including PAM-4, PAM-8 and PS-PAM-8, and discuss its potential for downstream burst-mode operation. With the hard-decision low-density parity-check (HD-LDPC) bit error rate (BER) limit of 1×10⁻², over 50 km single-mode fiber, the SOA-amplified 112.5 Gbit/s transmission processed by a truncated T-spaced Volterra nonlinear equalizer achieves optical power budgets of 32.85 dB and 31.3 dB for PAM-4 and PS-PAM-8, respectively, satisfying the 31 dB Class N2 power budget requirement for PONs. To the best of our knowledge, this is the highest reported power budget for a 100G-class LR-PON adopting a commercial QW-SOA at a transmission distance of 50 km.

Keywords: Intensity modulation and direct detection; Passive optical network; Pulse amplitude modulation; Semiconductor optical amplifier; Probabilistic shaping

Passive optical networks (PONs) are a key enabling technology that delivers high-speed broadband access to end users efficiently at low cost. Driven by the rapid proliferation of cutting-edge disruptive technologies and services such as social networks, the Internet of Things (IoT), artificial intelligence (AI), and 4K/8K video streaming, existing PON technologies and architectures are facing new challenges to support unprecedented capacity growth while pursuing low latency and low cost.

25G PONs are currently under deployment. They adopt standard time-division multiplexing (TDM) technology to provide symmetric 25 Gbit/s bit rates in the O-band and meet the Class 1 power budget requirement (29 dB) via high launch power. 50G PON or higher-speed PONs (HSPs), expected to be commercially available by 2025, will still employ pulse amplitude modulation 2 (PAM-2), high launch power, and intensity modulation with direct detection (IM/DD) to reduce overall system cost and complexity. Meanwhile, standardization bodies including the International Telecommunication Union Telecommunication Standardization Sector (ITU-T) and the Institute of Electrical and Electronics Engineers (IEEE) are exploring cost-effective 100G PON or very-high-speed PON (VHSP) solutions that can coexist with legacy PONs and deliver higher bandwidth to support emerging next-generation technologies and services. Given the cost-sensitive nature of the PON market, higher-order modulation formats such as PAM-4 and PAM-8 serve as suitable candidates to boost the overall capacity of IM/DD PONs without requiring high-bandwidth components or migrating to coherent transmission. The 1310 nm transmission window is preferable for high-speed PONs owing to its nearly zero chromatic dispersion in the O-band. Although its attenuation coefficient is relatively higher than that of the C-band, this drawback can be mitigated by increasing the optical launch power into the fiber and/or introducing optical amplification using semiconductor optical amplifiers (SOAs). SOAs feature wide gain bandwidth, compact footprint, low cost, and great potential for monolithic or hybrid integration. It is well known that systems using high launch power and SOA amplification are generally limited by fiber-induced and SOA-induced nonlinear effects, including self-phase modulation, stimulated Brillouin scattering (SBS), stimulated Raman scattering (SRS), polarization dependence, inter-channel crosstalk, and pattern effects associated with gain saturation. In addition, multilevel signals are more vulnerable to linear and nonlinear signal distortions and demand a higher signal-to-noise ratio (SNR). Nevertheless, recent advances in digital signal processing (DSP) can alleviate such distortions and improve the capacity and coverage of IM/DD PONs. Various DSP schemes have been extensively studied and verified effective, such as feed-forward equalizers (FFEs), Volterra nonlinear equalizers (VNLEs), decision feedback equalizers (DFEs), and machine learning (ML). Another DSP technique called probabilistic shaping (PS) exhibits strong robustness against nonlinear effects and has recently been considered for systems adopting spectrally efficient modulation formats such as PAM-8 and PAM-16.

In this context, O-band 100G PONs may be built upon PAM-4 technology, adopting high optical launch power and optical amplifiers to satisfy the Class 1 loss budget. Recently, a new PON architecture named long-reach passive optical network (LR-PON) has attracted growing attention. It extends the typical 20 km transmission reach of conventional PONs up to 100 km. The major advantages of LR-PONs include converging metro and core access networks, extended coverage, and consolidation of central offices (COs), which reduces the total number of central offices in the network and lowers capital expenditures and operational expenditures (CAPEX/OPEX). High-speed LR-PONs with large splitting ratios can be realized with coherent technologies; however, this would push the cost, complexity and power consumption of standard PONs beyond acceptable levels. Using IM/DD technology, a reasonable target for next-generation LR-PONs is to support a net data rate of at least 100 Gbit/s under the Class 1 loss budget.

Multiple multiplexing technologies have been proposed to implement high-speed LR-PONs, including time-division multiplexing (TDM), orthogonal frequency-division multiplexing (OFDM), and wavelength-division multiplexing (WDM). For example, 100 Gbit/s DSP-enabled linear λ PAM-4 transmission over 50 km in the C-band has been reported using directly modulated lasers (DMLs) and SOAs under a 29.3 dB power budget. 100 Gbit/s PAM-4 signals have also been transmitted over more than 100 km in the C-band using an IQ Mach–Zehnder modulator (MZM) for dispersion pre-compensation combined with nonlinear DSP at a 29.3 dB power budget. 50 Gbit/s OFDM signal transmission has been demonstrated with an erbium-doped fiber amplifier (EDFA) under a 28 dB loss budget. In terms of wavelength management, TDM and OFDM are the simplest multiplexing schemes. However, they broadcast aggregated high-speed data on a single wavelength in the downstream direction, and each optical network unit (ONU) needs to process the entire data stream to transmit or receive its allocated portion of the aggregated data. Unlike TDM-PON and OFDM-PON, ONUs in WDM-PON can be equipped with low-bandwidth transceivers since they do not need to process all multiplexed data. Nevertheless, WDM-PON introduces substantial passive optical loss, making it challenging to implement LR-PONs with large splitting ratios using WDM. Moreover, ITU-T has selected TDM for 50G PON, which delivers a 50 Gbit/s data rate per channel. In this regard, next-generation 100G LR-PONs can achieve 100 Gbit/s for both upstream and downstream directions on a single wavelength over long distances while adopting TDM.

In this paper, we propose a novel LR-PON architecture design concept that utilizes the low-dispersion 1310 nm transmission window for long-distance data transmission and deploys an SOA at the remote node (RN) of the PON to overcome high transmission loss, which may also increase the number of subscribers supported by the network. Optically amplified remote nodes have previously been proposed for 10G PONs to compensate for transmission loss and passive splitting loss in the network. SOAs deployed at PON remote nodes can operate with burst-mode optical amplification or function as PON extenders, which has the potential to reduce PON cost and power consumption. Recently, SOA-amplified remote nodes have been demonstrated for 2×53 Gbit/s direct-detection dense wavelength-division multiplexing (DWDM) PAM-4 transmission. PONs impose stringent power budget requirements, which pose great challenges for intensity modulation/direct detection (IM/DD) systems operating at 100 Gbit/s or higher data rates. Furthermore, the performance of high-speed PONs carrying multilevel signals and amplified by SOAs is severely degraded by SOA-induced amplified spontaneous emission (ASE) noise and nonlinearities such as pattern effects. We experimentally demonstrate, for the first time, downstream single-channel 112.5 Gbit/s PAM transmission over 50 km of O-band single-mode fiber (SMF) in a next-generation 100G-class LR-TDM-PON using high launch power and an SOA-amplified remote node. Considering bandwidth limitations and SOA-induced nonlinearities, we investigate feasible implementations of higher-order modulation formats (PAM-4, PAM-8 and PS-PAM-8) and low-complexity DSP techniques (T-spaced feed-forward equalizers and truncated second-order Volterra nonlinear equalizers) in SOA-amplified LR-PONs to mitigate linear and nonlinear signal distortions. After Volterra nonlinear equalization, PAM-4 and PS-PAM-8 achieve power budgets of 32.85 dB and 31.3 dB at the bit error rate (BER) threshold of 1×10⁻², satisfying the 31 dB Class N2 power budget requirement defined for NG-PONs. To the best of our knowledge, this represents the highest power budget achieved over 50 km in a 100G-class LR-PON amplified by a commercial quantum-well (QW) SOA.

Next-Generation Long-Reach Passive Optical Network

Figure 1 illustrates the conceptual design of the proposed next-generation LR-PON. The presented architecture adopts optical signal amplification by placing an SOA at the remote node (RN) of the O-band next-generation LR-PON to compensate for the propagation loss of 50–100 km feeder fiber (FF). It enables the broadcast of TDM signals with sufficient optical power to reliably deliver 100G signals to all connected subscribers in the network via distribution fibers (DFs). A passive splitter is typically deployed in the network to broadcast optical signals in PONs. Introducing an SOA at the RN can increase the total loss budget of the PON, which is calculated as the sum of the feeder loss (optical loss from the optical line terminal (OLT) to the RN) and the access loss budget (optical power loss from the RN to optical network units (ONUs)). To the best of our knowledge, the performance of such SOA-amplified LR-PONs has not been investigated before. Several key factors may limit its performance, including fiber- and SOA-induced nonlinear distortion (subject to bandwidth constraints), amplified spontaneous emission noise, and photodetector (PD) sensitivity.

The experimental setup of the long-reach passive optical network (LR-PON) is shown in Figure 2. To realize the next-generation LR-PON, transmissions of approximately 112.5 Gbit/s PAM-4, PAM-8 and probabilistically shaped PAM-8 (PS-PAM-8) signals amplified by a commercial quantum-well semiconductor optical amplifier (QW-SOA) are demonstrated over 50 km single-mode fiber (SMF) in the O-band. In the off-line digital signal processing (DSP) module at the transmitter, gray-coded PAM-4 and PAM-8 symbols are generated in MATLAB with a sequence length of 2¹⁶ symbols and uniform distribution. For the generation of PS-PAM-8 symbols (Figure 2(a)), uniformly distributed binary data are split into two parts: the first two-thirds of the total binary data are mapped to unsigned PAM-8 levels following a Maxwell–Boltzmann (MB) distribution via a look-up table based on the cost-minimization distribution matcher (CMDM) algorithm according to a predefined entropy. After binary labeling using binary reflected Gray code (BRGC), the remaining one-third of the binary data are assigned to the sign bits of the binary labels. Finally, PS-PAM-8 is generated after bit-to-symbol (B2S) mapping. It is worth noting that part of the uniformly distributed binary signal data can be replaced by forward error correction (FEC) parity bits without modifying the probability distribution of the PS-PAM-8 symbols.

To maintain comparable data rates of approximately 112.5 Gbit/s, the symbol rates for PAM-4, PAM-8 and PS-PAM-8 modulation are set to 56.25 GBaud, 37.5 GBaud and 42 GBaud, respectively. The digital PAM-4, PAM-8 and PS-PAM-8 symbols are upsampled by a factor of 4, 6 and 5, respectively, and pulse-shaped using a root-raised-cosine (RRC) filter with a roll-off factor of 0.1. To mitigate the influence of modulator nonlinearity, arcsine pre-distortion is applied to the digital data to equalize the PAM symbol levels during modulation. The digital PAM signals are then loaded onto an arbitrary waveform generator (AWG), which operates at 225 GSa/s for PAM-4 and PAM-8 signals and 210 GSa/s for PS-PAM-8 signals. The AWG has an analog bandwidth of 70 GHz. In general, a larger peak-to-peak voltage (Vpp) swing is required as the modulation order increases; otherwise, elevated nonlinear signal distortion and bit error rate (BER) may occur. The Vpp output from the AWG is optimized to 260 mV for PAM-4 and PS-PAM-8 signals and 300 mV for PAM-8 signals. These signals are amplified by a radio-frequency (RF) amplifier with a bandwidth of 55 GHz. The RF amplifier provides a gain of 23 dB, and its 3 dB gain compression point (P3dB) occurs at 19 dBm. The amplified electrical signal drives a commercial lithium niobate (LiNbO₃)-based Mach–Zehnder modulator (MZM) biased at the quadrature point. The electro-optic (E-O) S21 response of the MZM shows a 3 dB modulation bandwidth of 30 GHz owing to its gradual roll-off, while its 6 dB bandwidth reaches 40 GHz due to bandwidth limitations. In the experiment, a high-power distributed feedback (DFB) laser serves as the optical carrier emitting at 1310 nm, delivering an optical output power of 16 dBm at a bias current of 325 mA. After modulation, the approximately 112.5 Gbit/s PAM signal with an optical power of 8.5 dBm is directly launched into single-mode fiber (SMF) for both back-to-back (BtB) system characterization and transmission experiments over 50 km.

After transmission over the 50 km feeder fiber, a semiconductor optical amplifier (SOA) placed at the remote node amplifies the received signal to compensate for transmission loss and improve the power budget. For the back-to-back case, a variable optical attenuator (VOA) is placed before the SOA to adjust the input power to the SOA. To emulate the splitting loss of a typical time-division multiplexed passive optical network (TDM-PON) system and tune the received optical power (ROP), the optically amplified signal is attenuated by another VOA before direct detection via a 70 GHz bandwidth photodetector (PD). As shown in Figure 2, the received optical power is always measured at the input of the photodetector. After detection, the signal is amplified by a 55 GHz RF amplifier with 23 dB gain and captured by a real-time oscilloscope (RTO) operating at 256 GSa/s for signal recovery via off-line digital signal processing. In the receiver-side DSP, the captured signal is matched-filtered by an RRC filter, resampled to one sample per symbol, time-aligned, and equalized using a 32-tap feed-forward equalizer (FFE) and a truncated Volterra nonlinear equalizer (VNLE) for comparison. The VNLE adopts 32 linear taps and ten nonlinear taps to compensate residual nonlinearities involving up to three neighboring samples. After equalization, the signal is demodulated, and the bit error rate (BER) performance is evaluated by error counting.

Experimental Results

Figure 3(a) plots the gain of a commercial InP/InGaAsP multiple-quantum-well (MQW) SOA against the input power to the SOA (Pin) under bias currents of 125 mA, 175 mA and 225 mA. Regardless of the bias current, the SOA gain continuously decreases as the input power rises. It is observed that the SOA input power at the 3 dB compression point (where the SOA gain drops by 3 dB from the unsaturated value) ranges from −12 dBm to −10 dBm, depending on the SOA bias current. Operating beyond the P3dB point (Pin > P3dB) drives the SOA into saturation. In the gain-saturated region, the SOA exhibits pattern effects due to the carrier lifetime in its active region. At low input power (Pin = −19 dBm ≪ P3dB), the SOA operates in the unsaturated regime and achieves higher gains of approximately 20.3 dB, 23.5 dB and 25.3 dB for bias currents of 125 mA, 175 mA and 225 mA, respectively. In this regime, the SOA features a higher noise figure (NF) because spontaneous-spontaneous beat noise dominates during amplification. Accordingly, the amplified spontaneous emission noise introduced by the SOA predominantly determines the signal quality.

Figure 3(b) shows the SOA gain as a function of bias current, with the SOA input power fixed at −10 dBm. It can be seen that the SOA gain can be increased simply by applying a higher bias current. Nevertheless, the SOA gain curves indicate that gain saturation occurs at higher bias currents, which may introduce gain-saturation-induced nonlinear signal distortion during amplification. Given the trade-off between SOA gain and nonlinear signal distortion, selecting a proper SOA bias current for amplifying high-speed multilevel signals is essential.

The optical spectra of the 56.25 GBaud PAM-4 signal before and after amplification by the semiconductor optical amplifier (SOA) are measured at a resolution bandwidth of 0.01 nm, as shown in Figure 4. The input signal power to the SOA is −10 dBm, and the bias current is fixed at 125 mA. A variable optical attenuator (VOA) is used to maintain identical power levels before and after the SOA (the SOA provides a gain of approximately 17.3 dB in this case), enabling a direct comparison of the optical signal-to-noise ratio (OSNR) before and after SOA amplification. It can be observed that the OSNR degrades by approximately 10 dB after SOA amplification. Nevertheless, the measured OSNR of the amplified signal remains as high as 42 dB, which supports high-speed transmission of higher-order modulation formats such as PAM-4 and PAM-8. In addition, the SOA offers a 3-dB bandwidth of roughly 70 nm centered at 1310 nm.

Figure 5 presents the measured eye diagrams of 56.25 GBaud PAM-4, 37.5 GBaud PAM-8 and 42 GBaud probabilistically shaped PAM-8 (PS-PAM-8) signals in the back-to-back configuration at a received optical power (ROP) of 0 dBm for three different input powers into the semiconductor optical amplifier (SOA). Throughout the experiment, the entropy of the PS-PAM-8 signal is kept at 2.675 bit/symbol to maintain comparable achievable net data rates across all modulation formats. To mitigate system bandwidth limitations, a 32-tap feed-forward equalizer (FFE) operating at the sampling rate is adopted at the receiver. In Figure 5(a), pronounced gain saturation is observed at an SOA input power of −1 dBm, especially for the PAM-8 signal, whose sixth to eighth amplitude levels are heavily compressed and barely distinguishable. The PS-PAM-8 eye diagram reveals that probabilistic shaping offers strong resilience against SOA-induced pattern effects and nonlinear compression at high-energy PAM levels. Reducing the number of signal levels, e.g., adopting PAM-4, can alleviate gain saturation, yet at the cost of spectral efficiency, which imposes stricter requirements on wideband optical/electrical components and high-speed digital-to-analog/analog-to-digital converters (DACs/ADCs). At an SOA input power of −19 dBm (Figure 5(b)), the SOA operates in the unsaturated regime, and amplified spontaneous emission (ASE) noise from the SOA dominates the system performance for all modulation formats. The eye diagrams verify that higher-order modulation formats demand a higher signal-to-noise ratio (SNR) to achieve identical performance. PS-PAM-8 exhibits slightly better eye opening than PAM-8, indicating that probabilistic shaping provides limited improvement against SOA ASE noise. Among the three tested SOA input power conditions, the best eye openings for all modulation formats considered in this work are obtained at −10 dBm, which demonstrates that optimized input power and bias current are critical to achieving optimal system performance over extended optical transmission distances.

Next, we experimentally investigate the influence of the input power dynamic range of the semiconductor optical amplifier (SOA) on system performance. In Figure 6(a), we plot the bit error rate (BER) versus SOA input power for approximately 112.5 Gbit/s PAM-4, PAM-8 and PS-PAM-8 signals after feed-forward equalization (FFE) under back-to-back transmission. The SOA is biased at a constant current of 125 mA, and the received optical power (ROP) is fixed at 0 dBm for all cases. In this figure, the equalized BER performances of PAM-4, PAM-8 and PS-PAM-8 are marked by squares, circles and stars, respectively. It can be seen that the PAM-8 signal exhibits the worst performance among the three modulation formats at any SOA input power. It can barely reach the hard-decision low-density parity-check (HD-LDPC) BER threshold of 1×10⁻² only within a narrow SOA input power range from −7 dBm to −13 dBm. This supports our observation from Figure 5 that PAM-8 is more sensitive to linear and nonlinear distortions and imposes stricter signal-to-noise ratio (SNR) requirements. Unlike PAM-8, PS-PAM-8 is robust against such impairments introduced by the SOA; it can readily achieve HD-LDPC-level BER over a wide SOA input power window from −1 dBm to −15 dBm, which is essential for practical implementation in passive optical network (PON) scenarios. The BER performance of PAM-4 indicates that SOA-induced distortions only cause severe degradation when the four-level signal is amplified under heavy saturation or deep unsaturation. All multilevel modulation formats suffer from nonlinearities and level-dependent noise. The equalized BER trends reveal performance degradation originating from gain saturation and amplified spontaneous emission (ASE) noise in the saturated and unsaturated SOA operating regions. Nevertheless, optimal BER performance for all modulation formats is achieved only when the SOA input power falls within the optimal range of −7 dBm to −10 dBm, where the best trade-off between gain saturation and ASE noise is attained. Therefore, it is straightforward to conclude that the optimal strategy to minimize BER degradation is to optimize the power launched into the SOA and/or adopt more advanced signal processing techniques. Although this work focuses on downstream transmission, the above results show that the SOA dynamic range for PAM-4 is well above 18 dB at the HD-LDPC BER limit, making it promising for burst-mode operation. By contrast, the SOA dynamic range of PS-PAM-8 is around 3.5 dB narrower (approximately 14.5 dB) compared with PAM-4. A wider SOA dynamic range is required for burst upstream transmission, since the received optical power varies from burst to burst due to different path lengths between the optical line terminal (OLT) and optical network units (ONUs). In the proposed LR-PON architecture, amplified upstream transmission can be realized using either a shared or dedicated SOA at the remote node. In the shared-SOA configuration, a single SOA located at the remote node amplifies both upstream and downstream signals simultaneously. This introduces nonlinear signal distortion arising from cross-gain modulation (XGM) between downstream and upstream signals, which can be aggravated as burst-mode upstream packets turn on and off, though the feasibility of this scheme has been reported in Reference 34. Alternatively, the dedicated-SOA configuration employs two separate SOAs together with two three-port optical circulators at the remote node, with one SOA allocated for each transmission direction. In this configuration, the dominant impairments are excess ASE noise and pattern effects induced by gain saturation, as described in this paper.

To evaluate the impact of SOA bias current on transmission performance, Figure 6(b) presents the 32-tap FFE-equalized BER results of all modulation formats under several SOA bias currents after 50 km transmission, with the SOA input power set to −10 dBm and ROP fixed at 0 dBm. It is observed that the BER of all modulation formats degrades as the bias current increases. The PAM-4 eye diagrams at bias currents of 125 mA and 250 mA shown in the inset of Figure 6(b) demonstrate higher ASE noise and nonlinear signal distortion at larger bias currents. The uniform PAM-8 signal fails to achieve a BER below the HD-LDPC threshold at any tested bias current. On the other hand, probabilistically shaped PAM-8 (PS-PAM-8) can meet the target BER of 1×10⁻² when the SOA bias current is set to 150 mA or lower. At a bias current of 125 mA, the SOA provides a sufficient gain of 17.3 dB for amplifying high-order PAM signals at the remote node or local switching in optical access networks, and this operating condition will be selected for the subsequent experiments in this work.

In Figure 7(a), we plot the bit error rate (BER) as a function of received optical power (ROP) for the approximately 112.5 Gbit/s pulse amplitude modulation (PAM) signals under the back-to-back link condition. During the experiment, the bias current and input signal power of the semiconductor optical amplifier (SOA) are fixed at 125 mA and −10 dBm, respectively. For comparison, both the feed-forward equalizer (FFE) and the truncated Volterra nonlinear equalizer (VNLE) are adopted for symbol equalization in this case, represented by solid lines and dashed lines, respectively. It is worth noting that the adoption of the VNLE substantially increases the complexity of digital signal processing (DSP). For instance, an N-tap FFE requires N multiplications and N−1 additions to equalize one PAM symbol. By contrast, a truncated VNLE with N linear taps, M nonlinear taps and consideration of up to k neighboring samples requires

multiplications and

additions. This figure illustrates the BER performances of 56.25 GBaud PAM-4, 37.5 GBaud PAM-8 and 42 GBaud probabilistically shaped PAM-8 (PS-PAM-8) signals, marked by squares, circles and stars, respectively. For PAM-4 and PS-PAM-8 signals, the benefit brought by the VNLE is marginal, and no substantial gain can be observed compared with the FFE. This indicates that when the SOA operates in the linear regime, system bandwidth limitation and amplified spontaneous emission (ASE) noise dominate the system impairments. In contrast, the performance of PAM-8 remains limited by residual SOA nonlinearity; the VNLE achieves approximately 1.2 dB gain over the FFE at the hard-decision low-density parity-check (HD-LDPC) threshold.

Finally, in Figure 7(b), we plot the equalized BER results of approximately 112.5 Gbit/s PAM-4, PAM-8 and PS-PAM-8 signals after transmission over 50 km single-mode fiber (SMF) for long-reach passive optical network (PON) applications. Unlike the back-to-back case, BER improvements can be observed for PAM-8 and PS-PAM-8 signals with the truncated Volterra nonlinear equalizer (VNLE). This is attributed to enhanced nonlinear effects induced by the high launch power into the fiber, as well as the interaction between SOA chirp and fiber dispersion, which imposes more severe impairments on PAM-8 and PS-PAM-8 signals. After 50 km transmission, the PAM-8 signal can barely achieve HD-LDPC-level BER performance at a received optical power (ROP) of 0 dBm. By contrast, with VNLE, PS-PAM-8 reaches the HD-LDPC BER threshold of 1×10⁻² at an ROP of −5.5 dBm, yielding around 1.2 dB receiver sensitivity improvement compared with the feed-forward equalizer (FFE). In addition, the PAM-4 signal achieves identical performance at an ROP of approximately −7.05 dBm using either FFE or VNLE. Although PAM-4 delivers the best BER performance, its shorter symbol period demands higher system bandwidth, lower jitter and low-latency data processing, which may potentially increase implementation cost and complexity. Given the SOA gain of approximately 17.3 dB at an input power of −10 dBm and a bias current of 125 mA, together with the HD-LDPC threshold of 1×10⁻², optical power budgets of 32.85 dB and 31.3 dB are achieved for PAM-4 and PS-PAM-8, respectively, satisfying the Class C2 power budget requirement of next-generation passive optical network 2 (NG-PON2). Equalized with FFE, PS-PAM-8 still maintains a Class C1-compatible power budget of 29 dB. As stated above, the total power budget of the long-reach PON (LR-PON) after 50 km transmission is obtained by summing the feeder and access budgets. Figures 7(c)–7(e) show the histograms of PAM-4, PAM-8 and PS-PAM-8 symbols recovered by VNLE at an ROP of 0 dBm, respectively. It can be seen from the histograms of PAM-4 and PS-PAM-8 that the probability density functions (PDFs) of each symbol level differ from those of conventional PAM-8, and the level overlap is significantly reduced. This demonstrates that probabilistic shaping (PS) offers robustness against gain saturation by lowering the probability of high-amplitude symbols, which results in better-separated equalized histograms and substantially reduced inter-symbol interference (ISI) compared with conventional PAM-8 signals.

Conclusion

In summary, we experimentally demonstrate downstream transmission of single-channel 112.5 Gbit/s PAM-4, PAM-8 and PS-PAM-8 signals over 50 km for time-division multiplexed long-reach passive optical network (TDM LR-PON) applications. By adopting high launch power and SOA amplification at the remote node (RN), optical loss budgets of 32.85 dB and 31.3 dB are achieved for PAM-4 and PS-PAM-8 after equalization with the T-spaced truncated Volterra nonlinear equalizer (VNLE), satisfying the Class C2 power budget requirement of NG-PON2. When equalized with the T-spaced feed-forward equalizer (FFE), the PS-PAM-8 LR-PON still meets the Class C1 power budget requirement. Although the truncated VNLE provides certain resilience against SOA nonlinear distortion, more sophisticated and advanced digital signal processing (DSP) techniques may further boost the system performance.

A typical TDM-PON employs a shared feeder fiber for upstream and downstream transmission. All optical network units (ONUs) are connected to the remote node (RN) via distribution fibers of different lengths to transmit and receive data to/from the optical line terminal (OLT) simultaneously. This configuration results in bursty upstream traffic, while downstream traffic in PONs is continuous, as described in this work. Nevertheless, the BER results show that the proposed SOA-based LR-PON can operate over a wide dynamic range, making it suitable for burst-mode upstream operation with dedicated SOAs (one for upstream and one for downstream) at the remote node in the presence of nonlinear pattern effects. As mentioned earlier, upstream transmission can also be realized using a shared SOA at the remote node depending on the PON design; in this case, the dominant SOA-induced nonlinear impairment shifts from pattern effects to cross-gain modulation (XGM) imposed by upstream (downstream) signals on downstream (upstream) signals. It is important to quantify the performance of SOA-amplified bidirectional LR-PONs under various SOA configurations with concurrent burst upstream and continuous downstream data, which will constitute an important topic for our future research. Nevertheless, the key results on high-capacity long-reach downstream transmission presented in this paper are meaningful for the development of next-generation 100G LR-PONs.