Jaroslaw E. Prilepsky

Since a significant power burden in modern digital AI processors lies in linear operations, optics’ inherent linearity and parallelism offer an attractive route to analogue processors for multiply-and-accumulate (MAC) operations. We present a free-space dot-product multiplier that uses coherent detection to enable signed vector representation. The key innovation lies in a strictly common-path architecture where a single spatial light modulator (SLM) performs both phase modulation and optical fan-in, while the reference beam is generated within the active aperture itself. Such fan-in embedding allows for a leaner and phase-drift-tolerant architecture, as well as active misalignment compensation and enhanced parallelism. With commercially available device formats, our system is capable of performing dot products of 8-bit vectors of size N = 2025 or, alternatively, 4-bit vectors of size N = 32400 with RMS error as low as 0.06% (relative to full scale), with an equivalent SNR of 42 dB. Fan-in embedding allows us to seamlessly partition the SLM aperture to perform independent dot products of reduced-size vectors simultaneously with little precision penalty. Although simulation-based, our work evaluates key practical device limitations and identifies mitigation strategies for issues likely to arise during experimental deployment. Beyond the results themselves, we believe this to be a novel design that represents a step towards robust coherent optical processors, with the potential to contribute to future platforms that reduce the energy burden of linear computation.

Machine learning-based failure management in optical networks has gained significant attention in recent years, but severe class imbalance, where normal instances far outnumber failure cases, remains a considerable challenge. While pre- and in-processing techniques have been widely studied, post-processing methods are largely unexplored. We present a direct comparison of pre-, in-, and post-processing approaches for class imbalance mitigation in failure detection and identification using an experimental dataset. For failure detection, post-processing, particularly threshold adjustment, yields the highest F1 score improvement of up to 15.3%, while random under-sampling offers the fastest inference. In failure identification, generative AI methods deliver the most significant performance gains up to 24.2%, whereas post-processing has a limited impact in multi-class settings. When class overlap exists and latency is critical, over-sampling methods like synthetic minority over-sampling technique (SMOTE) are most effective; without latency constraints, meta-learning excels. In low-overlap scenarios, generative AI approaches provide the best performance with minimal inference time.

Power efficiency is a major challenge in modern optical fiber communication, driving efforts to reduce the computational complexity of chromatic dispersion compensation (CDC) algorithms. While various reduction strategies exist, many lack hardware implementations to validate their practical benefits regarding energy efficiency and hardware resource usage. This paper analyzes the tap overlapping effect in CDC filters for coherent receivers, introduces in detail the Time-Domain Clustered Equalizer (TDCE) to leverage this effect, and presents the validation through FPGA implementation for fiber lengths up to 640 km. Furthermore, we introduce an innovative parallelization method, and evaluate the memory impact on energy efficiency, chip area, and resource utilization. A fair comparison is conducted against the state-of-the-art frequency-domain equalizer (FDE). Furthermore, we demonstrate that TDCE is compatible with machine learning optimization, enabling up to 33.3% complexity reduction. Our findings highlight that implementation strategies—such as parallelization and memory organization—are as crucial as computational complexity in determining hardware efficiency. Despite TDCE's higher theoretical computational complexity, its optimized hardware implementation achieves up to 70.7% energy savings and a 71.4% reduction in multiplier usage compared to FDE. These results emphasize that computational complexity alone is not a reliable predictor of energy efficiency and multiplier usage in CDC filters.

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 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.

This paper reviews recent advancements in neural network (NN)-based equalizers for coherent optical communication systems, with a focus on their architecture and complexity. Various approaches are surveyed across transmission schemes, including single-channel, wavelength-division multiplexing, and digital subcarrier multiplexing systems. The study highlights the effectiveness of NNs in mitigating nonlinear impairments, improving signal quality, and transmission capacity. An overview of reduced-complexity NN-based equalizers in different transmission schemes is also provided, demonstrating their potential for real-time applications in future high-speed optical networks.

This paper reviews recent advancements in neural network (NN)-based equalizers for coherent optical communication systems, with a focus on their architecture and complexity. Various approaches are surveyed across transmission schemes, including single-channel, wavelength-division multiplexing, and digital subcarrier multiplexing systems. The study highlights the effectiveness of NNs in mitigating nonlinear impairments, improving signal quality, and transmission capacity. An overview of reduced-complexity NN-based equalizers in different transmission schemes is also provided, demonstrating their potential for real-time applications in future high-speed optical networks.