Bohan Sang

The rapid growth of data-intensive applications has intensified the demand for O-band intensity modulation and direct detection (IM/DD) systems designed for short- and medium-reach optical interconnects. However, the high symbol rates required for single-wavelength transmission impose stringent requirements on the channels. In this paper, we propose a novel channel estimation method that leverages the properties of sampling frequency offset (SFO) to directly estimate the channel impulse response (IR) in O-band wavelength-division multiplexing (WDM) IM/DD systems. By artificially introducing controlled SFO, we enable channel estimation with arbitrary sampling intervals, offering a new and intuitive approach to signal equalization. Theoretical derivations, numerical simulations, and experimental demonstrations validate the consistency and accuracy of the proposed method. We successfully demonstrate the transmission of 100-GBaud PAM-4 signals over 40-km fiber in an O-band WDM IM/DD system, achieving a total line rate of 800 Gbps (4×200 Gbps). Experimental results show that the estimated channel response can be used to construct zero-forcing filter, with the resulting equalization performance validating the accuracy of the estimated channel response and matching that of conventional adaptive feed-forward equalizers (FFEs). Unlike traditional works that treat SFO solely as a detrimental impairment, we demonstrate how to harness it to play its dual role as a powerful tool for channel estimation, paving the way for high-capacity, cost-effective optical interconnects in next-generation datacenter networks.

Photonics-assisted terahertz (THz) wireless communication systems have emerged as a promising solution for future THz-over-fiber (ToF) networks, offering both abundant bandwidth and seamless integration with optical infrastructure. However, existing THz systems typically operate within limited bandwidths (i.e., <0.1 THz), restricting their adaptability for diverse THz applications. Furthermore, laser phase noise and system nonlinearities significantly impair signal recovery at the receiver, while conventional approaches employing separate phase compensation and nonlinear equalization suffer from both high computational complexity and performance degradation. In this article, we present a novel low-complexity neural network architecture embedded with maximum likelihood phase retrieval (NN-eMLPR) for photonics-aided wireless transmission across the 0.33–0.52-THz band. Our solution integrates a maximum likelihood phase recovery layer directly within the neural network (NN) structure, enabling simultaneous equalization and phase noise compensation. Experimental results demonstrate successful 30-GBaud signal transmission throughout the 0.33–0.52-THz range. The NN-eMLPR achieves a 44.6% reduction in multiply-accumulate (MACC) operations compared to traditional phase recovery methods with NN while maintaining superior bit-error-rate (BER) performance, offering both computational efficiency and enhanced signal integrity across this ultrawide bandwidth.

In this study, we demonstrate a hardware-efficient post-equalization implementation for PAM8 signals on an FPGA platform by utilizing an additive power-of-two (APoT) quantization scheme. This approach reduces hardware resource consumption by over 70% compared to conventional methods while preserving critical performance metrics. In a 300-m W-band wireless communication experiment, the system reliably transmits 22.1184 Gbit/s PAM8 signals and satisfies the 2.4 × 10 ⁻² soft-decision forward error correction (SD-FEC) standard. Notably, under a host-assisted workflow, the APoT scheme achieves performance equivalent to 16-bit uniform quantization and exhibits superior stability than PoT quantization.

Simplified coherent detection presents a promising strategy for lowering receiver costs in photonics-aided terahertz (THz) wireless communication systems that operate at ultra-high data rates. However, the envelope detection is inherently limited by signal–signal beat interference (SSBI)-induced SNR degradation. While the Kramers-Kronig (KK) receiver configuration effectively mitigates SSBI, its practical deployment is hindered by the difficulty of implementing the Hilbert transform and nonlinear functions under low computational complexity. In this paper, A FPGA-based simplified parallel KK (SP-KK) receiver architecture is proposed and experimentally demonstrated. The key contribution is a design that approximates the nonlinear operation via a Taylor series expansion and incorporates a computationally efficient, multiplication-free FIR filter to execute the Hilbert transform. The SP-KK receiver’s performance was evaluated utilizing 16QAM single-sideband discrete multitone (DMT) signals across a converged 10-km fiber and 30-m wireless link operating at 322 GHz. The results demonstrate comparable performance to a conventional parallel KK receiver, with the primary advantage being a significant conservation of on-chip DSP resources. This work highlights the potential of integrating THz photonic components, electronic hardware, and sophisticated DSP algorithms to realize real-time simplified coherent high-speed wireless communication. This will provide us with a viable pathway toward future 6G mobile fronthaul and backhaul solutions.

This paper addresses a critical challenge in high-speed THz systems: sampling frequency offset (SFO) from both intrinsic device characteristics and critically, artificial introduction during system operation. We present a comprehensive study that uniquely examines operation-induced SFO alongside conventional device SFO through a novel sample-per-symbol (sps) framework. Our analysis reveals how artificial SFO, often undetected in standard system calibration, can significantly degrade performance beyond typical device limitations. The developed theoretical model precisely quantifies both SFO sources and their combined impact. Experimental validation in a photonics-aided THz-over-fiber polarization-division-multiplexing system demonstrates our method's effectiveness, successfully mitigating both device and artificially introduced SFO to achieve stable 60-GBaud transmission at 0.32-THz. This implementation establishes a record 576-Gbps/$\lambda$ per-channel capacity while providing crucial insights into preventing artificial SFO in practical deployments. The work offers researchers essential tools for identifying and correcting human-induced sampling errors in next-generation THz systems.

Photonics-assisted millimeter-wave (mmWave) technologies, capable of supporting large signal bandwidths, hold great promise for future ultrahigh-speed wireless communication systems. Optical device-based intensity modulation, photodetector-enabled mmWave generation, and envelope detector (ED)-aided systems offer broadband signal transmission with reduced cost, system complexity, and carrier frequency drift. In this work, we propose a novel method to analyze the characteristics of sampling frequency offset (SFO) and leverage these properties to estimate the time-domain impulse response of a 145-GHz D-band mmWave intensity-modulation envelope-detection system. The proposed estimation method facilitates arbitrary sampling rate and timing estimation of the system’s time-domain impulse response. The validity and accuracy of the method are demonstrated through three key aspects: the frequency-domain shape of the estimated impulse response, the equalization performance, and the spectral characteristics of the constructed equalizer. The signal equalizer based on this method outperforms traditional adaptive blind equalization algorithms and training-sequence-based equalizers. In addition, the proposed estimation method is shown to be minimally influenced by the modulation format of the transmitted signal. Using this method, we successfully achieve reliable transmission of a 30-GBaud four-level pulse amplitude modulation (PAM-4) signal over a 50-m free-space photonics-assisted D-band 145-GHz mmWave system with intensity-modulation envelope detection. Finally, we illustrate how SFO, traditionally viewed as a source of signal degradation, can be harnessed as a resource to enhance system performance.

Recently, the continuous advancement of sixth-generation (6G) mobile communication technology has positioned the space-air-ground-sea integrated network (SAGSIN) as a highly promising architecture for 6G networks. However, emerging applications and intelligent access devices have significantly strained network capacity, making it increasingly difficult to meet the ever-growing demand. To address these challenges, high-frequency terahertz-waveband has been proposed as an attractive solution for next-generation communication networks. Terahertz (THz) technology, enables data rates from tens to hundreds of gigabits per second, thus making it suitable for ultrahigh-speed wireless communication. Our study explores integrating photonics-aided signal generation with THz transmission technologies in a long-haul wireless communication system with a partial over-the-sea transmission. However, nonlinear distortions often occur for the THz-band system, the traditional Volterra-series nonlinear equalizers (VNEs) do not perform well in our experiment since the Volterra series cannot give an accurate polynomial approximation of the nonlinear channel response for the complicated over-the-sea channel conditions. A complex-valued multiple-symbol output (CVMSO) neural network (NN) equalizer is proposed to solve this problem. In this article, we achieved the transmission of a 2 GBaud THz-band quadrature phase shift keying (QPSK) signal over a 20.1 km wireless link at 125 GHz using photonics-assisted technology. Meanwhile, we establish a 20.1 km THz wireless communication model over the sea, revealing that elevated temperature and humidity markedly enhance air attenuation loss, adversely affecting the THz communication system. With the CVMSO equalization method, we achieved a bit error rate (BER) below 1.56 $\times$ $10^{-2}$ and a 1.8 dB receiver gain compared with VNE. Furthermore, the CVMSO NN equalizer demonstrated a 5.7% reduction in computational complexity and a reduction of 2000 training lengths because it can utilize the network information trained by the current input symbols for equalizing following multiple symbols, thereby minimizing unnecessary computations. Moreover, we extend our research on the complex climate environment in the over-the-sea channel, instructive for long-haul THz over-the-sea communications.

In the field of passive optical networks (PONs), coherent sinking has emerged as a significant topic, particularly in the context of cost and complexity improvements. We experimentally demonstrate a cost-effective 100-/200-Gbits/s/λ coherent PON system employing dual-polarization quadrature phase shift keying (DP-QPSK) and 16-ary quadrature amplitude modulation (DP-16QAM). To achieve cost reductions in coherent PON implementations, we employ low-cost distributed feedback lasers (DFBs) as the sole active light sources, replacing the much more expensive external cavity lasers (ECLs) at both the optical line terminal (OLT) and the optical network units (ONUs). To effectively address the performance penalty associated with phase noise (PN) of DFBs characterized by MHz-class linewidths, a straightforward carrier phase recovery (CPR) is proposed via embedding a one-tap carrier phase estimator (CPE) with gradient-descent algorithm into a 4 × 4 multiple-input multiple-output (MIMO) equalizer following the least-mean square algorithm (LMS). This approach simplifies the structure and enables more precise cancellation of PN. We conduct a transmission performance comparison at 32 Gbaud between our proposed scheme and the blind phase searching algorithm (BPS) for single-carrier (SC) signals, as well as between our method and the frequency-domain pilot-tone carrier phase recovery (FPT-CPR) for two-sub-band digital subcarrier multiplexing (DScM) signals. Experimental results demonstrate the corresponding power budgets of 38.1, 30.8, 36.5, and 29.1 dB for SC-QPSK/16QAM and DScM-QPSK/16QAM at the 20% soft-decision forward error correction (SD-FEC) threshold of 2 × 10-2. Furthermore, complexity analysis of these phase noise estimation (PNE) methods reveals the superiority of our proposed scheme in balancing performance and computational complexity.