We successfully demonstrate a 106.25-Gbps PAM-4 bidirectional optical sub-assembly for optical access networks, including a driver amplifier and an electro-absorption modulated laser for a transmitter, a photodiode and transimpedance amplifier for a receiver, and an optical filter block. For its implementation, we propose design strategies providing an in-line arrangement of optical and electrical interfaces while ensuring optical alignment tolerance for easy assembly and reducing electrical crosstalk between the transmitter and receiver. Measured receiver sensitivity was <-11.4 dBm for the KP4 forward error correction limit during transmitter operation, and measured power penalty of 10-km single-mode fiber transmission was <0.9 dB.
Download full-text PDF |
Source |
|---|---|
| http://dx.doi.org/10.1364/OE.445814 | DOI Listing |
Publication Analysis
Top Keywords
Similar Publications
In this Letter, we propose a high-performance optimized detection scheme based on a neural network (NN) in a receiver digital signal processing (DSP) for bandwidth-limited intensity modulation and direct detection (IM/DD) transmission systems. The NN-based optimized detection scheme consists of two components, an NN-based lookup table (NN-LUT) and an NN-based log-maximum estimation with a fixed number of surviving state (NN-MAP) decoder. The NN-LUT provides more accurate and sufficient information (PI) to the decoder than the conventional filter-form PI without increasing computational complexity.
View Article and Find Full Text PDFO-band intensity modulation and direct detection (IM/DD) transmission offers a promising solution for high-speed data center interconnects (DCIs). Additionally, the introduction of bismuth-doped fiber amplifiers (BDFAs) results in less nonlinear impact and a higher link budget compared with semiconductor optical amplifiers (SOAs). However, with these key issues resolved, the system bandwidth limitation emerges as the next critical bottleneck for high-speed O-band DCI transmission that limits overall performance.
View Article and Find Full Text PDFHigh-capacity optical interconnects with short reach are hugely demanded driven by the exponential growth of data traffic. In this work, four-channel wavelength division multiplexing (WDM) uplink/downlink twin single-sideband (twin-SSB) signals are implemented by a wavelength selective switch (WSS) at once, which simplifies the structure of multi-channel SSB transmitters and reduces the cost of high-capacity optical interconnect. Compared to a double sideband scheme, it has been experimentally proven that the performance of SSB transmission over standard single-mode fiber (SSMF) at C-band with an ultra-high baud rate has been greatly improved, which has the ability to effectively overcome the power fading induced by chromatic dispersion in an intensity modulation and direct detection (IM/DD) system.
View Article and Find Full Text PDFAn interleaved odd and even samples electrical dispersion pre-compensation (pre-EDC) scheme with low complexity is proposed to alleviate the severe distortions caused by frequency-selective fading due to chromatic dispersion (CD) in the intensity-modulation and direct-detection (IM-DD) optical transmission systems. This scheme utilizes training-based interleaved dual finite impulse response (FIR) filters for odd and even samples pre-compensation, respectively, referred to as training-based ID FIRs-pre-EDC. Additionally, a Tx and Rx side joint scheme is proposed to enhance the performance of the training-based ID FIRs-pre-EDC by implementing the efficient reduced state maximum likelihood sequence estimation (RS-MLSE) at the receiver side.
View Article and Find Full Text PDFArticle Synopsis
- - The paper introduces a new multi-source domain transfer learning (MST) method designed to lower the training costs for deep neural network (DNN) equalizers used in intensity-modulation and direct-detection (IMDD) systems.
- - This MST approach differs from typical transfer learning by selecting and using data with various channel parameters to create a multi-source domain dataset, enhancing the model’s ability to generalize across different conditions.
- - The MST equalizer was tested in an 80Gb/s PAM-4 IMDD short-reach system and showed significant improvements, achieving a low bit error rate that meets industry standards while dramatically reducing both training time (by 87%) and the amount of training data needed (by 65
Want AI Summaries of new PubMed Abstracts delivered to your In-box?
Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!