Packet Header Recognition Utilizing an All-Optical Reservoir Based on Reinforcement-Learning- Optimized Double-Ring Resonators

Zheng Li, Xiaoyan Zhou, Zongze Li, Guanju Peng, Yuhao Guo, Lin Zhang

Abstract

Optical packet header recognition is a critical signal processing task in optical communication networks. This work proposes an all-optical reservoir utilizing integrated double-ring resonators (DRRs) as nodes for fast and accurate header recognition. A deep reinforcement learning (DRL) algorithm is employed to maximize the delay-bandwidth product (DBP) of these nodes, offering full parameter space optimization and rapid convergence. Notably, optimized DBPs for cascaded, parallel, and embedded DRR configurations reached a consistent maximum value of 1395 ps·GHz, suggesting a global theoretical limit. The system achieved word-error rates (WER) as low as $5 \times 10^{-4}$ for 3-bit and $9 \times 10^{-4}$ for 6-bit header recognition tasks, representing a ten-fold improvement over previous benchmarks.

Introduction

In the quest for high-bandwidth, low-power optical networks, optical packet switching remains a primary objective. A major hurdle is the latency and power consumption associated with optical-to-electrical-to-optical (OEO) conversion required for packet header recognition. All-optical reservoir computing (RC) provides a pathway to process these signals entirely in the optical domain. While on-chip microring resonators are promising for compact RC implementations, they often face a trade-off between signal delay and bandwidth. This study utilizes multi-ring structures and advanced optimization to overcome these limitations.

System Architecture and Methodology

The proposed architecture integrates hardware optimization with a robust reservoir computing framework.

Reservoir Computing Framework

The system operates by first masking the 10 Gbps input signal to enrich its features using optical delays and attenuators. These signals then pass through a reservoir of optimized DRR nodes and a Semiconductor Optical Amplifier (SOA) for nonlinear activation. Finally, a photodetector captures the output, which is processed via ridge regression to determine the final header recognition weights.

Schematic of the all-optical reservoir computing system architecture.

DRL-Based Hardware Optimization

To maximize the performance of the DRR nodes, a Regularization-Assisted Asynchronous Advantage Actor-Critic (RA-A3C) algorithm was developed. This DRL approach was used to optimize the coupling coefficients and waveguide lengths across various DRR topologies (cascaded, parallel, embedded, and 3×3 coupler-based), targeting a high DBP and a "flat-top" delay spectrum to maintain signal integrity at high data rates.

Key Insights and Performance Results

Optimization Efficiency

The RA-A3C algorithm demonstrated superior efficiency compared to traditional methods like Genetic Algorithms (GA) or Particle Swarm Optimization (PSO). As shown in the performance comparison, RA-A3C converged to the global optimum in under 5 minutes, whereas traditional methods were significantly slower and often failed to reach the same peak performance. This suggests that the DRL agent effectively learned the underlying physics of the parameter space.

Comparison of DBP optimization performance between RA-A3C and traditional algorithms.

Recognition Accuracy and Compactness

The optimized reservoir achieved a Word-Error Rate (WER) of $5 \times 10^{-4}$ for 3-bit headers and $9 \times 10^{-4}$ for 6-bit headers. These results are an order of magnitude better than previous state-of-the-art implementations. Furthermore, the high DBP of the optimized rings allows for significant signal delay within a compact chip footprint, avoiding the need for large waveguide spirals or external fiber loops.

Conclusion

This research highlights the potential of DRL-optimized all-optical reservoirs for high-speed signal processing in optical networks. By maximizing the delay-bandwidth product of on-chip resonators, the system achieves unprecedented accuracy in packet header recognition. Future work will focus on implementing fully optical training weights and scaling the architecture for more complex, higher-bit-count header protocols.

Summary updated

Abstract

Optical packet header recognition is an important signal processing task of optical communication networks. In this work, we propose an all-optical reservoir, consisting of integrated double-ring resonators (DRRs) as nodes, for fast and accurate optical packet header recognition. As the delay-bandwidth product (DBP) of the node is a key figure-of-merit in the reservoir, we adopt a deep reinforcement learning algorithm to maximize the DBPs for various types of DRRs, which has the advantage of full parameter space optimization and fast convergence speed. Intriguingly, the optimized DBPs of the DRRs in cascaded, parallel, and embedded configurations reach the same maximum value, which is believed to be the global maximum. Finally, 3-bit and 6-bit packet header recognition tasks are performed with the all-optical reservoir consisting of the optimized cascaded rings, which have greatly reduced chip size and the desired “flat-top” delay spectra. Using this optical computing scheme, word-error rates as low as 5×10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-4</sup> and 9×10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-4</sup> are achieved for 3-bit and 6-bit packet header recognition tasks, respectively, which are one order of magnitude better than the previously reported values.
BibTeX
@article{li2023packet,
  author = {Zheng Li and Xiaoyan Zhou and Zongze Li and Guanju Peng and Yuhao Guo and Lin Zhang},
  title = {{Packet Header Recognition Utilizing an All-Optical Reservoir Based on Reinforcement-Learning- Optimized Double-Ring Resonators}},
  journal = {IEEE Journal of Selected Topics in Quantum Electronics},
  year = {2023},
  doi = {10.1109/jstqe.2023.3307420},
}
Figure