A constellation geometry shaping system and method for few-mode fiber space division multiplexing
By using end-to-end differentiable communication links and a dual-track training optimization strategy, the constellation geometry of the few-mode fiber system was optimized, which solved the performance bottleneck caused by mode-dependent loss and improved the robustness and transmission performance of the system.
Patent Information
- Application Number
- CN202610733286.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-26
- Publication Date
- 2026-07-17
AI Technical Summary
Existing geometric shaping methods struggle to balance worst-case mode performance in few-mode fiber systems, failing to effectively alleviate the performance bottleneck caused by mode-dependent loss (MDL). Furthermore, deep learning-based optimization schemes struggle to handle non-differentiable processing modules at the receiver, leading to performance discrepancies between the trained model and the actual system.
A constellation geometry shaping system for few-mode fiber spatial division multiplexing is provided, including an initialization configuration module, a trainable geometry shaping transmitter module, a few-mode fiber microchannel transmission module, a receiver-end differentiable digital signal processing module, and a dual-track training and verification module. The system optimizes the constellation point distribution to enhance robustness through an end-to-end fully differentiable communication link and a dual-track training optimization strategy.
It significantly improves the generalized mutual information of various spatial modes, reduces the system bit error rate, widens the optimal operating power range of the system, and enhances transmission robustness. It is suitable for high-speed optical fiber communication and spatial division multiplexing few-mode transmission networks.
Smart Images

Figure CN122419612A_ABST
