Sasipim Srivallapanondh

The paradigm shift toward autonomous optical networking is transforming the transport layer into a widespread distributed sensing platform. However, traditional telemetry architectures, designed for low-frequency performance monitoring (typically sampled at 1 Hz or 15 min intervals), fail to meet the stringent requirements of physical sensing applications, which demand kHz-range sampling to capture seismic events by using indirect metrics such as state of polarization (SOP) rotations or the Jones rotation vector. This paper proposes a high-sampling-rate, open-source telemetry architecture designed to bridge this gap. Implemented as a containerized microservices stack orchestrated via Docker Compose, the framework leverages Apache Kafka to decouple high-velocity sensing streams from persistent storage, elevating raw sensor data to a “First-Class Citizen” for immediate consumption by AI/ML agents. This approach facilitates multi-modal sensor fusion by enabling the temporal correlation of heterogeneous metrics in real-time. We validate the architecture through an extensive experimental scalability analysis using a Monte Carlo approach. Results demonstrate that the system sustains a throughput of over 30,000 messages/s—approximately 15 times the requirement for high-fidelity sensing—while maintaining sub-second end-to-end latency ( <200ms ) in the optimal batching regime. These findings confirm the architecture’s suitability for critical early warning scenarios, enabling robust, low-latency intervention in disaggregated optical networks.

This article conducts a comparative study of the complexity reduction of neural network (NN) models for nonlinearity compensation used in digital subcarrier multiplexing (DSCM)-based optical communication systems. We employ the NN model based on bi-directional long short-term memory (biLSTM) and 1D-convolutional NN (1D-CNN) layers. To reduce the computation complexity of the proposed solution, weight clustering is applied to the NN. We specifically compare the performance of our proposed NN-based equalizer with traditional methods such as chromatic dispersion compensation (CDC), digital back-propagation (DBP) and alternative NN, based on triplet coefficients in perturbation analysis, proposed in previous literature, evaluating both Q-factor performance and computational complexity. Our findings show that the NN-based equalizer offers competitive Q-factor improvements while significantly reducing computational complexity, particularly when weight clustering is used. We show that the complexity of the NN can be reduced by up to 91.1% compared to the NN based on the perturbation analysis proposed in previous literature and by 31.5% compared to the DBP with 1 step per span. The results underscore the potential of NN-based approaches to deliver high-performance nonlinear compensation with lower computational demands, positioning them as a promising solution for future optical nonlinear equalizers.

We propose a low-complexity weight-clustered NN-based equalizer for digital subcarrier multiplexing systems. Our approach achieved a 91% complexity reduction compared to the perturbation-based model in previous literature and 93% relative to a non-clustered NN.

We address the development of efficient neural network (NN)-based post-equalizers in long-haul coherent-detection dense wavelength-division multiplexing (DWDM) optical transmission systems. To achieve a high level of generalization of the NN-based equalizers, we propose to employ multi-task learning (MTL). MTL refers to a single shared machine learning (NN) model that can perform multiple different (albeit related) tasks. We verify the good performance of the developed MTL equalizer model using experimental data as compared to the previously proposed approaches. Furthermore, we report how MTL can improve performance compared to single-task counterparts. We also demonstrate that reducing the complexity of the resulting MTL equalizer is possible without essential performance compromise.

Digital Twins (DT) are needed in elastic optical networks to enable further advance and pave the way towards autonomous networks. Dynamical models self-adjusting with their physical counterpart have to maintain low computational complexity and guarantee good accuracy, regardless of the network topology and the technology supporting the optical lightpath. In this work, we propose data-driven models based on cascaded learning for single carrier signal as well as digital subcarrier multiplexing signals. Firstly, we explore the potentials of such a DT by proposing an algorithm for lightpath provisioning. In addition, the performances of the proposed models are compared with the ones of analytical models. Secondly, we experimentally evaluate its accuracy to predict the impact of optical filtering penalties. The simulations show an overall good accuracy for different signal formats and demonstrate their viability for lightpath provisioning. Similarly, the experimental results demonstrate the feasibility of implementing such models to estimate the quality of transmission for digital subcarrier multiplexed signals.

This paper presents a comparative study of neural network (NN) models for nonlinear compensation in digital subcarrier multiplexing (DSCM) systems. We employ the NN model based on bi-directional long short-term memory (biLSTM) and 1D-convolutional NN (1D-CNN) layers. To provide a lower complexity solution, weight clustering is applied to the NN. We specifically compare the performance of our proposed NN-based equalizer with traditional methods such as Chromatic Dispersion Compensation (CDC), Digital Back-Propagation (DBP) and alternative NN, based on triplet coefficients in perturbation analysis, proposed by Bakhshali, et al., evaluating both Q-factor performance and computational complexity. Our findings demonstrate that the NN-based equalizer offers competitive Q-factor improvements while significantly reducing the computational complexity, particularly when employing weight clustering. We show that the complexity of the NN can be reduced by up to 91.8% when compared to the NN based on perturbation analysis proposed by Bakhshali, et al. and by 31% when compared to the DBP with 1 step per span. The results underscore the potential of NN-based approaches to deliver high-performance nonlinear compensation with lower computational demands, making them a promising solution for future optical nonlinear equalizers.

Experimental results based on offline processing reported at optical conferences increasingly rely on neural network-based equalizers for accurate data recovery. However, achieving low-complexity implementations that are efficient for real-time digital signal processing remains a challenge. This paper addresses this critical need by proposing a systematic approach to designing and evaluating low-complexity neural network equalizers. Our approach focuses on three key phases: training, inference, and hardware synthesis. We provide a comprehensive review of existing methods for reducing complexity in each phase, enabling informed choices during design. For the training and inference phases, we introduce a novel methodology for quantifying complexity. This includes new metrics that bridge software-to-hardware considerations, revealing the relationship between complexity and specific neural network architectures and hyperparameters. We guide the calculation of these metrics for both feed-forward and recurrent layers, highlighting the appropriate choice depending on the application's focus (software or hardware). Finally, to demonstrate the practical benefits of our approach, we showcase how the computational complexity of neural network equalizers can be significantly reduced and measured for both teacher (biLSTM+CNN) and student (1D-CNN) architectures in different scenarios. This work aims to standardize the estimation and optimization of computational complexity for neural networks applied to real-time digital signal processing, paving the way for more efficient and deployable optical communication systems.

The recurrent neural network (RNN)-based equalizers, especially the bidirectional long-short-term memory (biLSTM) structure, have already been proven to outperform the feed-forward NNs in nonlinear mitigation in coherent optical systems. However, the recurrent connections still prevent the computation from being fully parallelizable. To circumvent the non-parallelizability of recurrent-based equalizers, we propose, for the first time, knowledge distillation (KD) to recast the biLSTM into a parallelizable feed-forward 1D-convolutional NN structure. In this work, we applied KD to the cross-architecture regression problem, which is still in its infancy. We highlight how the KD helps the student's learning from the teacher in the regression problem. Additionally, we provide a comparative study of the performance of the NN-based equalizers for both the teacher and the students with different NN architectures. The performance comparison was carried out in terms of the Q-factor, inference speed, and computational complexity. The equalization performance was evaluated using both simulated and experimental data. The 1D-CNN outperformed other NN types as a student model with respect to the Q-factor. The proposed 1D-CNN showed a significant reduction in the inference time compared to the biLSTM while maintaining comparable performance in the experimental data and experiencing only a slight degradation in the Q-factor in the simulated data.

For the first time, multi-task learning is proposed to improve the flexibility of NN-based equalizers in coherent systems. A "single" NN-based equalizer improves Q-factor by up to 4 dB compared to CDC, without re-training, even with variations in launch power, symbol rate, or transmission distance.

To reduce the complexity of the hardware implementation of neural network-based optical channel equalizers, we demonstrate that the performance of the biLSTM equalizer with approximated activation functions is close to that of the original model.