研究目的
Investigating the performance of a dynamic neural network based equalization scheme for 50 Gb/s PAM-4 signal transmissions using 10G-class optical devices.
研究成果
The dynamic neural network based equalization scheme significantly improves the performance of 50 Gb/s PAM-4 signal transmissions using 10G-class optical devices, achieving more than 2.5 dB power penalty improvement compared to MLSE. The scheme shows good tolerance to transmission distance.
研究不足
The study is limited by the bandwidth of the device, fiber dispersion, and fiber nonlinearity, which introduce inter-symbol interference (ISI) problems. The complexity of the neural network structure may also pose challenges.
1：Experimental Design and Method Selection:
The study employs a dynamic neural network based equalization scheme to improve the performance of 50 Gb/s PAM-4 signal transmissions. The methodology includes the use of delay blocks to construct parallel multidimensional data as input for the neural network.
2：Sample Selection and Data Sources:
The experiment uses a 10G-class electro-absorption modulated laser (EML) and a 12 GHz positive intrinsic-negative (PIN) for signal transmission over 10 km and 15 km standard single mode fiber (SSMF).
3：List of Experimental Equipment and Materials:
Equipment includes an Arbitrary waveform generator (AWG), a 10 GHz EML based TOSA, a 12 GHz PIN-TIA, and a DSO with a sampling rate of 80 GSa/s.
4：Experimental Procedures and Operational Workflow:
The signal from the AWG drives the EML, which transmits the signal over SSMF. The received optical signal is converted to an electrical signal by the PIN-TIA and sampled by the DSO. Digital signal processing is completed offline.
5：Data Analysis Methods:
The performance of the dynamic neural network based equalizer is compared to direct-decision and MLSE equalizers through BER performance analysis.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
