Complex-Valued CNN Nonlinear Equalization Enabled 36-Tbit/s (45×800-Gbit/s) WDM Transmission Over 3150 Km Using Silicon-Based IC-TROSA

Abstract

The growing Internet traffic urgently needs large-capacity and cost-effective optical transmissions. To maintain system performance under low-cost conditions, the silicon-based integrated coherent transmit and receive optical sub-assembly (IC-TROSA) and the complex-valued convolutional neural network (CVCNN) algorithm provide an effective solution for high-capacity and long-distance WDM optical transmission. The proposed CVCNN can improve the system performance under nonlinear damage conditions, which fully considers the orthogonality of IQ signals in this paper. This algorithm exhibits different equalization performances for 64QAM signals under various encoding schemes considering 20%-overhead, achieving up to 2dB maximum decrease in the required optical signal-to-noise ratio at the optical back-to-back case. Regarding transmission distance, employing CVCNN extends the maximum reach from 3500 km to 3850 km. The paper also demonstrates the application of CVCNN in WDM systems, enhancing system performance across different WDM encoding schemes. Finally, the experiment verified that CVCNN requires fewer computational resources than real-valued convolutional neural networks (RVCNN).

Keywords

• Complex-value neural network
• 800 Gb/s/lane transmission
• long-haul WDM
• low-cost optical device