Anti-resonant hollow-core fiber (AR-HCF) has exhibited the revolutionary potential in wide-spectrum optical transmission. We propose and demonstrate the co-transmission of O, S, C, L-band signals in a single AR-HCF. Enabled by 80Gbaud PAM8 modulated four independent tunable lasers in O, S, C, L-bands and linear equalizers, we achieve the 960Gb/s (240Gb/s/lambda & times;4 lambda) optical interconnect over 6.2km AR-HCF at 20% FEC limit of 2.4 & times; 10( -2) . By adjusting the laser wavelengths, the measured useful bandwidths within the 20% FEC limit are 35nm, 35nm, 30nm and 20nm in O, S, C, L-bands, respectively, which provides an upgrade solution to achieve high-capacity optical interconnects with 200Gb/s/lambda and beyond.

The escalation of symbol rates in next-generation coherent optical systems (e.g., 1.6 Tb/s and beyond) significantly exacerbates influence of transceiver frequency response impairments (FRIs). The system performance is degraded by FRIs of both transmitter (Tx) and receiver (Rx), being difficult to discriminate the individual contribution of each subsystem. Nevertheless, such discrimination is essential for the optimization of state-of-the-art transceiver designs, enabling effective mitigation of impairment propagation. To address this technical challenge, we propose a non-intrusive scheme for the simultaneous characterization and separation of Tx and Rx frequency responses (FRs). By leveraging tone-shift characteristics induced by frequency offset combined with multi-tone probe signals, the proposed method digitally separates the FRs of the Tx and Rx without requiring auxiliary components. Experimental results demonstrate the proposed scheme achieves high estimation accuracy, with deviations below 1 dB in amplitude and 0.25 rd in phase across a 20-dB bandwidth of 35 GHz. Furthermore, using the monitored responses for targeted pre- and post-compensation in a 69-GBaud DP-16QAM transmission system yields up to a 2-dB improvement in OSNR sensitivity at the 7% HD-FEC threshold. In addition, when combined with a reduced-tap MIMO equalizer, the proposed scheme still preserves a 1-dB OSNR sensitivity gain while reducing the equalization complexity by approximately 44%.. Therefore, the proposed scheme provides a practical solution for in-field transceiver characterization, paving the way for simplified DSP and enhanced performance in future ultra-high-speed coherent optical interfaces.

High-performance blue-violet lasers are essential for compact quantum sensing and metrology. However, light sources in this spectral range currently rely on bulky external cavities or complex nonlinear frequency conversion, limiting their robustness and field deployment. While monolithic InGaN-based distributed feedback laser diodes (DFB LDs) offer a path toward miniaturization, precisely targeting narrow atomic transitions remains challenging due to fixed grating periods and fabrication-induced wavelength deviations. In this work, we demonstrate a 420-nm DFB laser array with an ultrafine 0.2-nm channel spacing, enabled by a continuously phase-shifted grating design. This approach allows for precise wavelength control without compromising coupling efficiency, while significantly enhancing the array's robustness against fabrication variations. By integrating ten devices, the array spans a 2-nm spectral range around 420 nm, enabling coarse wavelength selection through individual device addressing. This strategy minimizes the required thermal or electrical tuning range, thereby keeping the LD within its optimal operating regime while reaching the target atomic transitions. The fabricated LDs exhibit a 40 mA threshold current, 1 W/A slope efficiency, and up to 40 mW single-mode output power, with tuning coefficients of 1.6 pm/mA and 20 pm/∘C for current and temperature, respectively. Optical heterodyne measurements reveal a Lorentzian linewidth of 9.8 MHz, highlighting a practical and manufacturable path toward robust blue-violet light sources for next-generation chip-scale atomic and quantum photonic systems.

We demonstrate a same-wavelength bidirectional transmission over a 10.9-km AR-HCF with 393×2 channels and 50-GHz spaced grid. The GMI-estimated net rate of 550.97 Tb/s indicates a record capacity for any single-core single-mode fiber.

Digital Subcarrier Multiplexing (DSM) has emerged as a preferred solution over its Single-Carrier (SC) counterpart for high-speed long-haul Standard Single Mode Fiber (SSMF) transmission, owing to its flexible spectral allocation and adaptive transmission capabilities. However, fiber nonlinearity remains a fundamental limitation on the transmission bandwidth-distance product, while Digital Backpropagation (DBP) serves as a representative Nonlinearity Compensation (NLC) method. Traditional symmetric DBP assumes a uniform nonlinearity accumulation over the SSMF link, which deviates from the inherently asymmetric power profile in practice. Specifically, we propose an asymmetric DSM-DBP scheme based on a nonlinear accumulation asymmetry metric ΔE(ρ), which enables the efficient determination of the optimal dispersion splitting ratio ρ ∈ [0, 1] without exhaustive optimization. Consequently, we can efficiently determine the position of nonlinear compensation operators within each DBP step for high-speed long-haul DSM transmission systems. When the single-wavelength, four-subcarrier, 800 Gb/s Probabilistic Shaping Dual-Polarization 64-Quadrature-Amplitude-Modulation (PS-DP-64QAM) signals are transmitted, the asymmetric DSM-DBP increases the optimal launch power to 5dBm and achieves a Signal-To-Noise Ratio (SNR) improvement of 0.2 dB over symmetric DSM-DBP after the 1600 km SSMF transmission. Under a Normalized Generalized Mutual Information (NGMI) threshold of 0.89, the SSMF reach can be extended to 2400 km, representing a reach extension of approximately 160km compared to traditional symmetric DSM-DBP. Meanwhile, the robustness of the proposed asymmetric DSM-DBP is numerically verified for various modulation formats, variable baud rates, and different subcarrier configurations. These results highlight the scalability and practical relevance of the proposed asymmetric DSM-DBP for next-generation high-capacity optical networks.

We experimentally demonstrate a 400G 4λ WDM-PON system over C-band 20km~50km SSMF with below 12.5% SD-FEC limit of 1.0×10-2 enabled by optically filtered comb sideband modulation and linear equalizer only, achieving 44dB optical power budget.

The application of digital subcarrier multiplexing (DSM) in elastic optical networks necessitates advanced optical performance monitoring (OPM) for various transmission impairments owing to its flexible spectral allocation and adaptive transmission capability. In this work, we propose a two-interleaved pilot-tones (TIPTs)-assisted OPM approach that enables simultaneous estimation of multiple transmission impairments and system parameters, including frequency offset (FO), polarization rotation (PR), chromatic dispersion (CD), transmitter (Tx) and receiver (Rx) individual laser phase noise (LPN), and modulation format (MF). With optimal TIPT parameters, the simulation results reveal that the proposed OPM scheme can simultaneously achieve an FO estimation error below 35 MHz, a CD estimation error within 120 ps/nm after 2000 km standard single-mode fiber (SSMF) transmission, and robust PR tracking under 100 krad/s polarization variations. In particular, the explicit separation of Tx/Rx LPN effectively eliminates equalization-enhanced phase noise (EEPN), while the embedded amplitude-modulated TIPTs enable nearly 100% MF identification (MFI) even under optical signal-to-noise ratios (OSNRs) as low as 20 dB. The proposed OPM scheme is experimentally validated when the 80 GBaud DP-16QAM DSM signals are transmitted over a 1600 km SSMF. Those results confirm the accurate and robust operation of TIPTs-assisted OPM for future DSM-enabled elastic optical networks.

We demonstrate a record 12,113-km interstitial-tube-assisted double-nested anti-resonant nodeless fiber long-haul transmission using a 153.34-km recirculating loop, achieving GMI-estimated 24.8-Tb/s capacity with PCS-16QAM in C-band, the longest distance reported to date for hollow-core fiber.

For short-distance applications, Multi-mode fiber (MMF)-based optical communication systems have become an important solution due to their low cost and compatibility with existing infrastructure. However, the performance of MMF systems is significantly limited by mode-dependent loss (MDL) and mode dispersion (MD), especially in high-speed transmission scenarios. Diversity reception technology serves as an effective solution to significantly improve overall system performance by combining multiple signal paths. Our work demonstrates a neural network (NN)-enhanced diversity reception scheme that fundamentally addresses the limitations of conventional methods—selection combining (SC), equal gain combining (EGC), and maximal ratio combining (MRC) — in MMF systems. The experimental results for both 30-GBaud PAM-4 and 20-GBaud PAM-8 signal MMF transmissions demonstrate that the performance can be significantly improved by the proposed scheme and NN methods. Both 30-GBaud PAM-4 and 20-GBaud PAM-8 signal transmissions over 4.2-km MMF have been successfully achieved under 7% and 20% forward error correction (FEC) thresholds, respectively. It indicates that NN-based diversity reception can effectively mitigate MDL and MD in multi-mode transmission and improve system robustness and reliability to meet future communication requirements.

We experimentally demonstrate a flexible optical interconnect in O-band by using intensity modulation and direct detection (IM/DD) with probabilistically shaped eight-level pulse amplitude modulation (PS-PAM8) at 125 GBaud for 20 km-50 km reach application. With the aid of a semiconductor optical amplifier (SOA) and low-complexity linear equalization, a peak line rate of 360 Gb/s is successfully achieved. Considering 12.5% FEC limit of 1 & times; 10(-2) , the demonstrated system delivers a net throughput of 320 Gb/s. To the best of our knowledge, this is the first O-band demonstration of PS-PAM8 based flexible optical interconnects achieving beyond-300 Gb/s with only linear equalization over 20 km, highlighting their potential for costand power-efficient intra-data center applications.

Anti-resonant hollow core fiber (AR-HCF) is a promising alternative for next-generation optical systems, given their theoretical potential of achieving low loss and ultra-low Rayleigh backscattering over ultra-wide bandwidth, ideally overcoming the bottleneck of SSMF-based systems. This paper studies the Rayleigh backscattering characteristics of AR-HCF and interconnection technologies for HCF-SMF, which are of great significance to integrate AR-HCF with the existing SSMF-based network and thus enhance the system capacity by utilizing direction as a new dimension. We also report on the experimental demonstration of high-capacity same-wavelength bidirectional coherent wavelength division multiplexing (WDM) transmission over 11.04-km continuous long-length AR-HCF of the nested anti-resonant nodeless type. By leveraging low Rayleigh backscattering of AR-HCF, we achieved bidirectional transmission across the S+C+L bands (2 x 156-nm bandwidth) utilizing probabilistic constellation shaping 256QAM, with a net GMI throughput of 251.4 Tbit/s in one direction and 251.2 Tbit/s in the opposite direction. The results show an enormous potential for ultra-high capacity transmission using AR-HCF.

A neural network equalizer based on multi-task learning is proposed, which achieves resource efficiency through shared network parameters while enhancing system performance by leveraging inter-wavelength channel correlations. The scheme includes a signal preprocessing unit and a main multi-task learning (MTL) network to achieve architectural decoupling between linear compensation and nonlinear feature learning. A C-band 960 Gb/s (240 Gb/s/lambda & times; 4 lambda) PAM4 WDM IM/DD transmission system was experimentally constructed using four independent lasers over 1.4 km anti-resonant hollow-core fiber (AR-HCF). Experimental results demonstrate the proposed MTL equalizer consistently outperforms single-task equalizers across the tested channels while reducing parameter count by approximately 57%, significantly improving both equalization performance and computational efficiency.

Emerging disaggregated and distributed learning infrastructures require data center networks that offer ultra-low latency and high efficiency. However, conventional networks relying on TCP sockets or RoCE incur significant overhead due to recurrent conversions between the PCIe bus and network protocols. PCIe-based networking emerges as a promising solution, particularly for accessing disaggregated memory, storage, and GPU resources. Motivated by its potential, this paper introduces a novel architecture for PCIe-over-optical networks, designed for scalable inter-machine memory communication. By using optical fiber as the medium while preserving the native PCIe protocol stack, our system facilitates reliable, direct memory access across host domains. An experimental prototype was implemented to validate the proposed design. The results demonstrate robust PCIe transmission over normal QSFP optical modules with near-lossless performance compared to native PCIe. Comparative analysis shows clear advantages of our approach over existing RoCE-based AI scale out solutions.

Anti-resonant hollow-core fiber (AR-HCF) has emerged as a promising transmission medium for long-reach high-capacity optical fiber communications owing to its unique properties of low transmission loss, low nonlinearity, and low latency. In this Letter, we experimentally demonstrate a fully loaded C-band transmission with a generalized mutual information-estimated capacity of 50.14 Tb/s over a total fiber distance of 557.6 km based on a 21.9-km low-loss self-developed AR-HCF. The transmission is realized using a 32.8-km recirculating loop loaded with 120 wavelength channels on 50-GHz grids. To extend the loop length and improve system performance for convenient lab demonstration, we propose a recirculating loop architecture that leverages the intrinsically low Rayleigh backscattering of AR-HCF. To the best of our knowledge, this represents the highest reported C-band transmission capacity for HCF links exceeding 100 km.

Free-space optical communication (FSOC) is a key technology for sixth-generation (6G) networks. To achieve high coupling efficiency for the single mode fiber (SMF) receiver, conventional beam alignment methods typically require splitting the received signal beam to measure the angular deviation relative to the fiber center, leading to additional optical power loss and system complexity. Although laser nutation algorithms allow precise alignment without beam splitting, their narrow field of view (FOV) limits their applicability. We propose a wide-field nutation system with multi-core fiber (MCF), achieving a common optical path FSOC architecture that removes the need for a separate detection optical branch. An enhanced laser nutation algorithm is developed to extend its FOV by utilizing the structure characteristics of MCF. A framework for MCF-based beam alignment is established to improve the convergence ability. Experimental validation is conducted to optimize the operational parameters of the laser nutation process, achieving a fine alignment FOV of +/- 2.4 mrad. Finally, a 3.71 km lake-crossing experiment was conducted, achieving automatic alignment with a link loss of 31.25 dB and stable coherent communication at 8 and 16 Gbps, error-free without forward error correction.

Abstract Random lasers rely on multiple scattering in disordered media to generate emission with complex spectral behavior. While their high-entropy output is valuable for random number generation, the inherent unpredictability has historically limited their utility in structured information processing. Here, we demonstrate the coexistence of spectral randomness and determinism within these systems. Using deep neural networks for multi-dimensional spectral analysis, we identify inter-modal correlations that permit partial recovery of spectral intensity components, despite temporal fluctuations. We then develop a key generation and distribution scheme utilizing the dual nature of random lasers: the disorder ensures cryptographic randomness, while the underlying order facilitates accurate key transmission. This work presents a framework for physical-layer security, suggesting potential applications in photonic-based cryptography and secure communications.

In passive optical networks (PONs), the optical power budget serves as a key determinant of system scalability. A higher power budget allows for larger split ratios over a given transmission medium, thereby enabling more end users to share the same infrastructure and reducing the cost per subscriber. In this paper, we introduce a carrier-assisted complex-valued double-sideband direct detection PON architecture that eliminates the need for a local oscillator laser at the optical network unit of coherent PON while maintaining a high optical power budget. The system leverages a deep-learning-enabled optimal direct-detection receiver to achieve high launch power and receiver sensitivity without the complexity of coherent detection. Using the proposed architecture, we experimentally demonstrate an optical power budget of 42 dB for 100 Gb/s downstream transmission over 20 km standard single-mode fiber (SSMF), meeting the 15% soft-decision forward error correction (SD-FEC) threshold of 2×10 −2 . To the best of our knowledge, this represents the highest optical power budget for a 100 Gb/s direct-detection PON using SSMF. Furthermore, by employing our self-developed 20 km anti-resonant hollow-core fiber, we demonstrate an optical power budget of 49 dB under the 15% SD-FEC threshold, marking a record-high optical power budget achieved to date for 100 Gb/s PON.

Coherent optical communication offers superior spectral efficiency and substantial transmission capacity, making it indispensable for accommodating the exponential growth of global data traffic. Next-generation data-centers demand higher baud rates and improved spectral efficiency, which in turn require optoelectronic devices for both ultra-broadband operation and high modulation efficiency. However, conventional modulator design commonly faces a fundamental trade-off between electro-optic bandwidth and modulation depth. To address this limitation, we present a segmented modulation strategy and fabricate segmented I/Q modulator on the thin-film lithium niobate (TFLN) platform. It features a 3-dB bandwidth exceeding 110 GHz while maintaining a high signal-to-noise ratio for high-order signal generation. For high baud rate signal generation beyond the limitation of digital-to-analog convertor sampling rate, we achieve 320-Gbaud single-polarization quadrature phase shift keying (QPSK) and 310-Gbaud dual-polarization (DP) QPSK signal based on polybinary coding scheme. For high spectral efficiency transmission, we demonstrate 2.02-Tb/s line rate probabilistic shaped (PS) 144-ary quadrature amplitude modulation (144-QAM) signal over 125-km single-mode fiber, facilitated by the residual carrier-based phase tracking. The results highlight the potential of segmented IQ modulator, in conjunction with advanced digital signal processing, can pave the way for next-generation high-speed coherent optical communications.

We demonstrate a simplified bidirectional PON over 22km AR-HCF, achieving 200-Gb/s downstream and 50-Gb/s upstream while maintaining >40dB power budgets, enabled by shared single laser source for downstream coherent detection and upstream intensity modulation.

Leveraging the extremely low Rayleigh backscattering and ultralow loss characteristics of the hollow core fiber, we experimentally demonstrate a real-time same wavelength bidirectional transmission system over a 10.9-km nested anti-resonant nodeless fiber (NANF) link, which achieves a bidirectional net capacity of 235.2 Tbit/s using commercial real-time coherent optical transceivers. The system used a fully loaded S+C+L band WDM configuration over 19.65-THz spectrum, with 131 channels on a 150 GHz grid (51 in S-band, 40 in C-band, and 40 in L-band), resulting in 262 simultaneously transmitted channels bidirectionally. We deployed commercial optical modules across the operating bands, using 800 G modules in the S and L bands and a hybrid of 1.2 T and 800 G modules for the conventional and extended C band, respectively. The successful real-time operation of all channels within the BER requirement validates hollow-core fiber as a viable platform for wideband, same-wavelength bidirectional transmission, underscoring its potential for next-generation ultra-high-capacity optical interconnects.