Fiber Kerr nonlinearity is a fundamental limitation to the achievable capacity of long-distance optical fiber communication. Digital back-propagation (DBP) is a primary methodology to mitigate both linear and nonlinear impairments by solving the inverse-propagating nonlinear Schrödinger equation (NLSE), which requires detailed link information. Recently, the paradigms based on neural network (NN) were proposed to mitigate nonlinear transmission impairments in optical communication systems. However, almost all neural network-based equalization schemes yield high computation complexity, which prevents the practical implementation in commercial transmission systems. In this paper, we propose a center-oriented long short-term memory network (Co-LSTM) incorporating a simplified mode with a recycling mechanism in the equalization operation, which can mitigate fiber nonlinearity in coherent optical communication systems with ultralow complexity. To validate the proposed methodology, we carry out an experiment of ten-channel wavelength division multiplexing (WDM) transmission over 1600 km standard single-mode fiber (SSMF) with 64 Gbaud polarization-division-multiplexed 16-ary quadrature amplitude modulation (16-QAM) signals. A 0.51 dB Q ² factor gain is observed with the Co-LSTM equalization, which is comparable to that of DBP. The complexity of the Co-LSTM equalization is only 5.2% of that of the conventional bi-directional LSTM, and 28.4% of that of the DBP method with a single step per span. In principle, the complexity of the Co-LSTM with a simplified mode is almost independent of the transmission distance, which shows an essential benefit over the DBP method that determined by the optical signal evolution along the fiber link. The proposed Co-LSTM methodology presents an attractive approach for low complexity nonlinearity mitigation with neural networks.

Nonlinear frequency division multiplexing (NFDM) transmission scheme has attracted great interest in optical fiber communication systems due to its potential for surpassing the Kerr nonlinearity limit. There have been very active studies on single-channel NFDM transmission systems recently. However, up to now, few experiments have validated the feasibility of applying NFDM into wavelength division multiplexing (WDM) optical communication systems toward high-capacity transmission for wideband coverage. Moreover, the overall performance of the NFDM system is restricted due to many challenges in theoretical design and practical implementation. Novel neural network-based equalization methods contribute to eliminating multi-dimensional crosstalk, hence prompting the NFDM system performance. In this article, we experimentally demonstrate a high-capacity NFDM-WDM transmission system enabled by the newly developed digital signal processing (DSP) algorithms. With data modulation on the continuous spectrum in nonlinear frequency domain, a 25-channel WDM system with polarization division multiplexing (PDM) 16-ary quadrature amplitude modulation (16-QAM) transmitting over 800 km standard single-mode fiber (SSMF) is successfully achieved. To mitigate the undesirable correlations between the nonlinear continuous spectral components, a nonlinear spectrum equalization scheme based on artificial neural network (ANN) is utilized in the receiver side DSP. Through this ANN-based spectrum equalization, a 0.63 dB Q ² -factor gain is achieved compared to the conventional equalization scheme. A record transmission capacity of 10.83 Tb/s is achieved in a 25-channel WDM system over 800 km SSMF, with a remarkable single-channel line rate of 433.3 Gb/s. Our work reports the highest capacity NFDM-WDM transmission up to date and helps to further explore the inherent ability of NFDM to resist fiber channel nonlinearity.

The exponential growth in data-driven applications like artificial intelligence and metaverse poses an increasing demand for integrated optical interconnects with efficient and cost-effective high-capacity solutions. For short-reach optical interconnects dominated by intensity-modulated direct detection (IM-DD), high-speed and large-capacity transmission enabled by high-bandwidth electro-optic devices is always highly desired. In this article, based on a thin film lithium niobate (TFLN) modulator with a 3-dB bandwidth larger than 110 GHz fabricated by ourselves, we experimentally demonstrate ultrahigh-speed optical interconnects with probabilistic shaping pulse amplitude modulation (PS-PAM) signals. With linear pre-equalization and Volterra nonlinear equalizer (VNLE), a 120 Gbaud PS-PAM16 signal with a bit rate up to 408 Gb/s is generated. Compared to the regular-PAM8 signal with the same bit rate of 360Gb/s, the PS-PAM signal demonstrates a receiver sensitivity gain of 2.2 dB. With the help of VNLE, 384.5 Gb/s PS-PAM16 signal can be transmitted over 500 m standard single-mode fiber (SSMF) at the C band. The ultra-broadband TFLN enabled transmission provides a promising solution for future 400G per wavelength data-center interconnects.

Single-sideband (SSB) signaling is a promising technology for short and medium reach optical transmission. We propose a novel architecture of linear network-assisted long short-term memory network (LN-LSTM) to mitigate optical transmission impairments in SSB systems. With a combination of a linear network and an input-pruned LSTM network, LN-LSTM achieves better performance and lower computation complexity compared with conventional LSTM equalization. With LN-LSTM, we achieve 6.4 Tb/s ( $32\times 200$ Gb/s) wavelength division multiplexing (WDM) SSB transmission over 320km standard single-mode fiber (SSMF) with a bit error rate (BER) below the 20% soft-decision forward error-correction (SD-FEC) threshold of $2.4\times 10^{-2}$ .

Coherent passive optical networks (PONs) have emerged as a promising candidate for next-generation access networks due to their superior receiver sensitivity, high data rates, extended reach, and enhanced flexibility. However, their large-scale deployment is hindered by the complexity and cost of coherent receivers, particularly at the optical network unit (ONU). In this work, we propose and experimentally demonstrate a flexible 200 Gb/s coherent time-and-frequency-division multiple access PON for the downstream, leveraging a simplified ONU architecture and a low-cost MHz-linewidth distributed feedback (DFB) laser. A tailored subcarrier allocation scheme is introduced to enable heterodyne detection with polarization diversity, supporting flexible-rate reception of single or multiple subcarriers via local oscillator (LO) wavelength tuning. Efficient carrier recovery is achieved through pilot-tone-aided digital signal processing, eliminating the need for time-domain pilot symbols and significantly reducing frame overhead. We experimentally demonstrate flexible data rate reception from 50 to 200 Gb/s. A power budget exceeding 32.5 dB is achieved after 25 km standard single-mode fiber transmission using a 1 MHz linewidth DFB LO and a 512-symbol preamble, showcasing high sensitivity and robust performance. This work paves the way for the practical deployment of cost-effective coherent PON systems.