Reviewed by Lexie CornerNov 4 2024
A research group led by Professor Lilin Yi from the State Key Lab of Advanced Optical Communication Systems and Networks at Shanghai Jiao Tong University, China, has developed a novel method known as learnable DSP (LDSP). This method was recently published in the journal Light: Science & Applications.
To enhance the capacity of optical fiber communication networks, advancements in DSP technology are essential. While traditional DSP frameworks with a block-by-block design have reached a significant level of maturity, this architecture can lead to local performance minima.
Moreover, interference from linearities may adversely affect fiber nonlinear correction performance compared to standard DSP techniques. To fully evaluate the advantages of nonlinear DSPs and to pursue increased transmission capacity, it is crucial to optimize performance in linear DSPs.
This technique, which integrates deep learning optimization into the existing DSP framework, has yielded significant performance and efficiency improvements. They introduced their LDSP framework, which conceptualizes the DSP process as a deep learning structure while reusing conventional DSP modules.
LDSP enhances compensation for both linear and nonlinear performance by globally optimizing DSP parameters using backpropagation algorithms.
The study highlighted the efficiency of the LDSP architecture by demonstrating its ability to perform multiple tasks with a single module. The researchers also found that improved accuracy in linear compensation increases the precision of nonlinear estimates, resulting in remarkable nonlinear compensation (NLC) performance for LDSP-enhanced perturbation-based methods.
This underscores LDSP's potential as a novel and highly effective standard for linearity adjustment in optical fiber communications. Their experimental trials, which involved 1600 km of fiber transmission at 400 Gb/s, showed a significant improvement in signal performance.
The proposed optimization allows LDSP to achieve full module-level optimization, which increases the Q factor by approximately 0.77 dB for single-channel transmission and 0.56 dB for 21-channel transmission. Performance improvements were observed with NLC, yielding a maximum gain of 1.21 dB for single-channel and 0.9 dB for multi-channel transmissions. The researchers summarize the working premise of their LDSP framework as follows:
The researchers noted, “While the majority of the DSP modules remain the same, each DSP module is treated as a linear layer of a deep neural network (DNN), and its parameters are optimized using a learning algorithm through backpropagation, specifically the SGD method. This approach enables performance optimization from a global perspective, offering a more holistic and effective solution.”
“It is noteworthy that all DSP operations must be implemented differentially, and then the backpropagation algorithm can be performed normally. After passing through the entire LDSP module, the loss is computed based on the loss function, and gradients are calculated using the error backpropagation algorithm,” they added.
“Due to the DL framework, the LDSP is compatible with DL structures and can be extended to incorporate learnable perspectives for nonlinear compensation in the future. The LDSP could emerge as a new and highly efficient benchmark for linearity compensation, generating significant interest across various domains of nonlinear compensations and beyond,” the scientists concluded.
Journal Reference:
Niu, Z., et al. (2024) Learnable digital signal processing: a new benchmark of linearity compensation for optical fiber communications. Light Science & Applications. doi.org/10.1038/s41377-024-01556-5.