Han Li

Ultra-wideband optical transmission systems suffer from severe inter-channel performance disparities, necessitating effective launch power optimization. However, this objective is often hindered by the complexity or inaccuracy of traditional quality of transmission estimations. To address this limitation, we propose a physics-augmented residual neural network. By delegating primary physics to analytical models and focusing the data-driven task solely on quantifying residual discrepancies, this framework substantially narrows the learning space, enabling high-precision inference with sparse datasets. Using a tandem architecture, the pre-trained forward model is frozen to serve as a differentiable bridge, providing indirect supervision to guide an inverse model for launch power optimization. As a proof-of-concept, a real-time field trial of 120-channel transmission over a 67 km G.652.D fiber is demonstrated across an 18.7 THz S+C+L band. The proposed scheme achieves a quality of transmission estimation accuracy where 97.8% of errors fall within ±0.5dB. In view of the output power limits of S-band optical amplifiers, a flatness of the optical/generalized signal-to-noise ratio of ±0.27dB and ±0.36dB is realized within the individual S band and C+L band, respectively. Compared to purely black-box neural networks, our approach exhibits similar efficacy while reducing the required training samples by 67%.

Coherent passive optical networks (CPONs) are emerging as promising solutions for next-generation optical access networks, offering an enhanced power budget and improved system capacity. In this paper, as a proof-of-concept, 56 GBaud dual-polarized probabilistic shaped 4-ary pulse amplitude modulation (DP-PS-PAM4) with an entropy of 1.81 bits/symbol/polarization is experimentally investigated over a 20 km standard single-mode fiber (SSMF) upstream link using an intensity-modulation and coherent detection (IM-CD) structure, where the line data rate is about 200Gb/s/ λ . Besides, to meet the low-cost requirements of PONs, the many-to-one (MTO) mapping-based probabilistic shaping technique is adopted to implement DP-PS-PAM4 with low complexity. Experimental results show that the received optical power (ROP) sensitivity of DP-PS-PAM4 is about −27dBm in the optical back-to-back (OBTB) and 20 km transmission cases, outperforming dual-polarized uniformly distributed PAM4 (DP-UD-PAM4) by about 1 dB. In terms of power budget, a 36 dB power budget with a launch power of 9 dBm can be obtained by adopting DP-PS-PAM4, which can be further improved to about 40 dB (corresponding to an ROP of −31dBm) by using a maximum likelihood sequence detector (MLSD). To the best of our knowledge, this is a remarkable power budget metric for 200G PON. Moreover, since the symbol distribution of DP-PS-PAM4 is a dyadic distribution and deviates from the Maxwell–Boltzmann (MB) distribution, the performance gaps between the dyadic and MB distributions are also investigated. Experimental results show that the performance degradation induced by distribution deviation is negligible under the OBTB and 20 km conditions. The experimental results presented in this paper verify that MTO-based DP-PS-PAM4 can achieve superior performance, and it might be a promising solution for very-high-speed PON in the future.

Obtaining network delay data is crucial for various low-latency telecommunication applications. This work presents a method that employs periodic interaction messages based on precision time synchronization to measure the delay for each link segment in the network. The proposed method can automatically obtain full-network delay data with nanosecond-level precision in real time. The field trial results demonstrate the effectiveness and accuracy of this method. In addition, optimal path planning is also verified, which proves that the minimum path delay can be achieved for the services utilizing the measured delay data.

A novel, to our knowledge, PFC-assisted hitless OTN for separation of compute and data is proposed and demonstrated. Based on the new scheme, simulations are conducted and analyzed comprehensively. Simulations indicate that symmetric or asymmetric computing power distribution minimally affects efficiency. With consistent GPUs, larger models experience less impact from remote extension. When training with an appropriate number of GPUs compatible with the model, the degradation rate of training efficiency for models of different sizes remains comparable as the distance increases. Then, the effectiveness of the PFC-assisted hitless OTN is experimentally verified. By hitless protection, pipeline parallelism training efficiency is free from the switchover between primary and backup paths. With the PFC function on, the training efficiency could be improved from 52% to 99.51% over 120 km transmission, compared to the off case. Furthermore, different model sizes, transmission distances, and network convergence ratios are experimentally investigated by the field trial. When extending to 240 km, training efficiency is still up to 99.29%.

In this Letter, we experimentally demonstrate a 16.7 Tb/s co-frequency co-time full-duplex 10 km mobile fronthaul on C band by harnessing the ultralow loss, nonlinearity, dispersion, and Rayleigh backscattering of anti-resonant hollow-core fiber. 112 GBaud four-level pulse amplitude modulation signals are adopted with direct detection. Bit error rates of all the channels are below 6.7% hard-decision forward error correction threshold of 3.8 × 10 ⁻³ . Few sensitivity penalties of inter-channel nonlinear interference and backward crosstalk are observed as 0.2 dB and 0.5 dB, respectively. This work shows that anti-resonant hollow-core fiber is a promising medium for implementing a high-capacity 6 G fronthaul architecture.

In recent years, anti-resonant hollow-core fibers (AR-HCFs) have made remarkable progress, surpassing the loss limit of conventional silica solid-core fibers, and achieving attenuation below 0.1 dB/km. Due to their ultra-low latency, negligible nonlinearity, and broad transmission bandwidth, AR-HCFs have attracted significant attention from both academia and industry, positioning them as a foundational technology for optical communications over the next half-century. However, large-scale industrial production of AR-HCFs remains underdeveloped, and prior research has been largely confined to laboratory-scale experiments using short fiber segments, failing to fully explore their performance potential or identify deployment-related challenges. To address this gap, we present the field deployments of four structurally distinct AR-HCF cables spanning 10∼42.7 km. Post-deployment measurements demonstrate an average link loss reduction from 0.599 dB/km to 0.113 dB/km, with the lowest recorded span loss reaching 0.065 dB/km, scoring the world’s first fiber cable deployment to break the 0.1 dB/km. Leveraging these deployed links, we conduct real-time coherent transmission experiments at 800 Gb/s to 1.2 Tb/s across S-, C-, and L-bands, validating the feasibility of co-time co-frequency full-duplex (CCFD) transmission in hollow-core fibers and characterizing the impact of gas absorption (e.g., CO_{2 id="240">} and H_{2 id="241">}O) on system performance. Despite these impairments, we achieve the first field demonstration of real-time terabit-scale coherent transmission over the S+C+L band, delivering a record aggregate capacity of 114.9 Tb/s over 137.36 km using 105 channels.

Low-noise microwave generation is crucial for advanced applications such as 6G millimeter-wave communications, satellite communications, and synthetic aperture radar systems. Traditional microwave sources encounter challenges in meeting stringent performance requirements while maintaining low size, weight, and power consumption (SWaP) at ever-increasing high frequencies. Photonic microwave generation is a promising solution to overcome these limitations, particularly when implemented with chip-scale integration using a simple yet efficient architecture. Here, we propose a chip-scale integrated photonic microwave generator (IPMG) scheme that features a low-noise hybrid InP / Si ₃ N ₄ comb laser based on the self-injection locking mechanism, in conjunction with the optic-electro-optic feedback to further enhance the RF generation performance. The proof-of-concept IPMG prototype has demonstrated superior performance, highlighting an ultra-narrow RF intrinsic linewidth of 0.8 Hz, low single-sideband phase noises of −92.1 dBc/Hz at 10 kHz offset and −128.3 dBc/Hz at 1 MHz offset, and excellent frequency stability with only 16 kHz fluctuations over 5 min. This work marks a substantial advancement in the development of fully integrated photonic microwave generators by unifying good performance, architectural simplicity, and low SWaP.

A real-time DWDM transmission system based on a unified C + L-band architecture is demonstrated. The system incorporates transceiver modules, wavelength-selective switches (WSS), and distributed Raman amplifiers (DRA), all compatible with the full C + L band. A major limitation in previous C + L-band systems is the requirement for separate optical amplifiers for each band, which increases system complexity and cost. This issue is addressed by employing an erbium-bismuth co-doped fiber amplifier (EBDFA), which achieves a 20 dB gain bandwidth of 100 nm. Enabled by these unified components, a total capacity of 62.1 Tb/s is transmitted over 80 km of G.652.D fiber using 70 WDM channels, with per-channel rates up to 1 Tb/s across a seamless 87 nm optical spectrum. The prototype highlights the feasibility of unified C + L-band optics for high-capacity, spectrally efficient, and simplified optical transport networks.