2687-0517 19 2 2026 1-122

RAR RUS 7-15 Skrebenkov Danil skrebenkov_di@spbstu.ru Budanov Dmitriy dmitriy.budanov@gmail.com Spiking neural network analog two-variable neuron implementation Currently, the problem of using neural networks in embedded systems, which have constraints on power consumption and area occupied, is relevant. At the same time, there is growing interest in spiking neural networks – more biologically realistic neural networks with potentially higher energy efficiency. This paper presents a review of existing hardware implementations of spiking neural networks, implemented both in the digital and analog domains. A hardware implementation of analog spiking neuron and synapse fabricated in Mikron JSC 180 nm CMOS technology with a reduced supply voltage of 0.6 V is presented. According to the simulation results, the energy consumption of a single neuron was 1.46 pJ/spike. The paper also provides a formula for comparing energy consumption between neurons fabricated in different technologies and with different supply voltages. According to this formula, the energy consumption of the implemented neuron is comparable to that of the referenced sources. 10.18721/JCSTCS.19201 004.032.26 spiking neural network power consumption spiking neuron analog neuron analog synapse https://infocom.spbstu.ru/article/2026.89.1/

RAR RUS 16-30 0000-0002-1497-2893 Peter the Great St. Petersburg Polytechnic University Guo Chenxi chenxiguo.academic@gmail.com

St. Petersburg, Russian Federation

Potekhin Vyacheslav slava.potekhin@mail.ru TFCA-GRU: A temporal and feature cooperative attention GRU for remaining useful life prediction of bearings across conditions Accurate prediction of the remaining useful life (RUL) of rolling bearings under different operating conditions is essential to enable proactive maintenance of industrial systems. To address the limitations of traditional recursive models in capturing complex time dependencies and distinguishing sensor contributions, we propose a temporal and feature cooperative attention-based gated recursive unit (TFCA-GRU). By integrating temporal and feature channel attention mechanisms, the model adaptively focuses on degradation-related time steps and key sensor features to build more robust RUL models from multidimensional time-frequency fusion data. Therefore, the TFCA-GRU enables sensitive extraction of critical features from long-term dependencies and local feature information. The validity of the proposed TFCA-GRU is verified using the bearing dataset from the XJTU-SY and the IEEE PHM 2012 Prognostic Challenge. This paper also provides the model’s intrinsic interpretability through structured non-uniform attentional learning across time and feature dimensions. This reveals TFCA-GRU’s selective attention to information time intervals and feature dimensions, which improves the transparency of the model. Experimental results show that the proposed TFCA-GRU performs better than some existing prediction methods in bearing RUL prediction across operating conditions. 10.18721/JCSTCS.19202 004.8 remaining useful life deep learning attention mechanism gated recurrent unit time-frequency feature fusion model interpretability https://infocom.spbstu.ru/article/2026.89.2/

RAR RUS 31-43 F-6480-2013 7004013271 0000-0002-5637-1420 Peter the Great St.Petersburg Polytechnic University Lev V. lev.utkin@mail.ru

Polytechnicheskaya, 29, St.Petersburg, Russia, 195251

Peter the Great St. Petersburg Polytechnic University Konstantinov Andrei andrue.konst@gmail.com

St. Petersburg, Russian Federation

0000-0003-2275-1473 Peter the Great St. Petersburg Polytechnic University Kirpichenko Stanislav kirpichenko.sr@gmail.com

St. Petersburg, Russian Federation

0000-0001-8749-9470 Peter the Great St. Petersburg Polytechnic University Verbova Natalia

St. Petersburg, Russian Federation

Survival analysis using Mahalanobis distance in Kernels of the Beran estimator Survival analysis is critical for modeling time-to-event data across medicine, engineering, and economics. The Beran estimator serves as an efficient nonparametric tool for estimating conditional survival functions under censoring. However, the conventional Beran estimator relies on the Euclidean distance for kernel weighting, a metric that fails to account for heterogeneous scales and correlations among covariates. This limitation can result in suboptimal weighting when features exhibit multicollinearity or varying measurement units. To address this, we propose the M-Beran estimator, a generalized formulation that incorporates the Mahalanobis distance into the kernel smoothing scheme. This modification ensures scale invariance and leverages the underlying correlation structure of the feature space, transforming the kernel into an elliptically symmetric form. We evaluate the proposed method through numerical experiments on real-world survival datasets. Results demonstrate that the M-Beran estimator outperforms the conventional Euclidean-distance-based Beran estimator, offering improved predictive accuracy. 10.18721/JCSTCS.19203 004.85 machine learning survival analysis Beran estimator Mahalanobis distance Harrell’s C-index survival function https://infocom.spbstu.ru/article/2026.89.3/

SCO RUS 44-51 Pilipko M.M. m_m_pilipko@rambler.ru Morozov Dmitriy dvmorozov@inbox.ru A frequency-selective circuit in 180 nm CMOS A frequency-selective device has been developed in the form of a rejector active filter based on transconductance amplifiers and capacitors in the 180 nm CMOS technology of Mikron JSC. The layout dimensions are 62 x 26 µm. Post-layout simulation results are as follows: rejection frequency is 312.5 MHz; passband gain is 1.44 dB; rejection is 23.1 dB; power consumption is 4.8 mW; total harmonic distortion is 0.07%. A comparison with the results from other studies is performed. The device can be applied in optical communication systems. 10.18721/JCSTCS.19204 621.3.049.774.2 CMOS frequency-selective device rejector active filter transconductance amplifier element imitation https://infocom.spbstu.ru/article/2026.89.4/

RAR RUS 52-60 Special Technological Center Ltd. Klimenko Denis d.klimenk0@yandex.ru

St. Petersburg, Russian Federation

Peter the Great St.Petersburg Polytechnic University Nikitin Aleksandr nikitin@mail.spbstu.ru

Polytechnicheskaya, 29, St.Petersburg, 195251, Russia

Peter the Great St. Petersburg Polytechnic University Stroganov Alexander lemyr103@gmail.com

St. Petersburg, Russian Federation

Design of power limiter for antenna-feeder system for direction finding protection on GaAs pHEMT This paper presents the development results of a power limiter for the 30–6500 MHz frequency range, implemented as a microwave monolithic integrated circuit (MMIC) based on a domestic GaAs pHEMT technology. Implementing the power limiter as a monolithic integrated circuit reduces the weight and size of the final device compared to a hybrid implementation. The power limiter consists of three cascaded stages connected in parallel, comprising Schottky diodes interconnected by microstrip transmission lines that compensate for the parasitic capacitances of the diodes. The developed power limiter exhibits an insertion loss of less than 0.7 dB, an input and output VSWR of less than 1.2, an input 1 dB compression point of at least 22.8 dBm, and a maximum input power of at least 40.7 W. The MMIC die dimensions are 1400 × 1300 µm. The developed power limiter can be used in the input stage of receiving devices, such as antenna feeder systems for direction finding. 10.18721/JCSTCS.19205 621.372.552 microwave monolithic integrated circuit power limiter antenna-feeder system GaAs pHEMT Schottky diode https://infocom.spbstu.ru/article/2026.89.5/

SCO RUS 61-69 Pilipko M.M. m_m_pilipko@rambler.ru Morozov Dmitriy dvmorozov@inbox.ru A spiking neural network in 180 nm CMOS A hardware implementation of a spiking neural network for classifying font images of letters and numbers has been developed in the 180 nm CMOS technology of Mikron JSC. The layout dimensions are 500 × 550 µm. The supply voltage is 1.0 V. The average power consumption is approximately 0.06 mW. The clock frequency is up to 10 MHz. The neural network contains four layers of 15, 8, 8, and 10 nodes, respectively. The input data and synapse weights have a bit width of 3 bits. A comparison with the results from other studies is performed. 10.18721/JCSTCS.19206 621.3.049.774.2 CMOS analog neural network leaky integrating neuron https://infocom.spbstu.ru/article/2026.89.6/

RAR RUS 70-80 0009-0003-6281-6726 Chernyi Vitaliy vitaly.g.cherny@gmail.com 0000-0001-6650-6491 Peter the Great St.Petersburg Polytechnic University Bolsunovskaya Marina bolsun_hht@mail.ru

Polytechnicheskaya, 29, St.Petersburg, 195251, Russia

Analysis of the influence of weighting coefficients in a significance metric for 3D scene objects in preliminary rendering optimization setup This paper presents an experimental analysis of the behavior of a significance metric for 3D scene objects used for preliminary rendering optimization setup in the Unity environment. The influence of three object characteristics is considered, from which the final significance metric is formed: geometric complexity (polygon count), spatial size, and the object’s potential as an occluder, calculated based on visibility analysis performed using ray casting. Combining these features with adjustable weights makes it possible to obtain a unified estimate of the potential importance of an object and its impact on rendering during the configuration of optimization approaches. The results show that geometric complexity provides strong separation of objects into groups, assigning 22.6% of the objects in the experimental scene to the high-significance category. The balanced configuration and the dominance of other coefficients individually demonstrate lower differentiation, yet accurately identify the most complex objects intentionally placed in the scene, while classifying over 70% of the remaining models as low-significance. Regardless of the chosen configuration, the method correctly identifies candidates clearly suitable for further optimization. 10.18721/JCSTCS.19207 004.925 computer graphics 3D scene analysis visibility analysis rendering optimization Unity engine https://infocom.spbstu.ru/article/2026.89.7/

RAR RUS 81-94 Shaidullin Renat renat.a.s@yandex.ru Method for implementing scheduling systems in discrete-continuous production: the case of oil refineries This paper addresses the complex problem of production scheduling for continuous production facilities, such as oil refineries and petrochemical plants. The typical challenges of existing models are analyzed in detail, including the difficulty of accounting for all technological and logistic constraints, operational challenges, and the limited time available for schedule preparation and verification. As a solution, a five-step implementation methodology is proposed: 1) model aggregation to ensure solvability and accuracy, 2) selection of data updating methods (manual, semi-automated, or fully automated), 3) user adaptation through a structured planning process and automatic checks, 4) ensuring additivity between monthly plans based on linear programming and shift-daily schedules, and 5) automatic monitoring of plan and actual asynchrony to detect deviations. The methodology is embedded into the system implementation process, linking methodological principles with each project stage. It constitutes a replicable approach applicable to petrochemical and other continuous production facilities, ensuring the creation of a robust and adequate system capable of operating under the high dynamics and complexity of modern production. 10.18721/JCSTCS.19208 658.5:004.9 production scheduling oil refinery planning models implementation methodology APS https://infocom.spbstu.ru/article/2026.89.8/

RAR RUS 95-108 Deeb Ali dib_a@spbstu.ru Khokhlovskiy Vladimir 78v.kh77@gmail.com Structure-aware decision support system for MOO-driven control Cyber-physical control systems must satisfy strict constrains on the magnitude of control actions and the rate of signal change, while simultaneously balancing multiple, often conflicting, control objectives. In multi-objective optimization (MOO) problems, the optimizer naturally produces a set of competing control candidates. However, in closed-loop operation, a single implementable control action must be selected at each sampling instant. In practice, this selection is typically performed through fixed scalarization rules, which can lead to performance degradation under changing operating conditions in the case of non-convex Pareto fronts or poorly scalable objectives. This work aims to improve the quality of control based on MOO through a decision support system that accounts for the structure: the proposed architecture decouples online Pareto-front construction from action selection. A heuristic multi-objective optimization algorithm based on the NSGA-II genetic algorithm generates feasible control sequences under magnitude and rate constraints, while a learned selector, denoted as Learned Pareto Filter (LPF), evaluates these candidates using structural features that account for both individual characteristics of the solutions and their interrelations within the Pareto set. The method is evaluated on two benchmark problems. First, a system with an analytically known Pareto front structure is used. Second, the approach is validated on a nonlinear boiler-turbine load-changing scenario, widely used in constrained nonlinear control studies. The results demonstrate that LPF-based selection produces smoother control actions, improved tracking performance, and greater robustness to regime changes compared to objective-based scalarization methods. The findings confirm that, in Pareto-based control, closed-loop quality depends critically on the selection mechanism rather than solely on the optimizer generating the Pareto set. 10.18721/JCSTCS.19209 004.891.3:519.853 multi-objective optimization model predictive control decision support systems Pareto selection graph neural networks constrained nonlinear control https://infocom.spbstu.ru/article/2026.89.9/

RAR RUS 109-122 Peter the Great St. Petersburg Polytechnic University Fershtadt Mikhail tral1930@mail.ru

St. Petersburg, Russian Federation

Shashikhin Vladimir Adaptive weight adjustment in multicriteria control of nonlinear systems based on a Lyapunov function The paper develops an approach to the control of nonlinear dynamic systems within a multicriteria optimization framework in which criterion weights are adjusted during the control process. The method relies on a Lyapunov function that is included both in the performance functional and in trajectory constraints. A numerical algorithm is proposed for dynamic tuning of the criterion weights. The criteria are stability and control energy consumption, while their relative importance changes depending on the current system state and trajectory. Adaptation is performed based on the current value of the Lyapunov function and the system state, which enables automatic prioritization of objectives without operator intervention. A numerical analysis of the system behavior under different weight-adaptation schemes is carried out. For the considered models, the simulation results show that adaptive weight adjustment can provide faster decay of the Lyapunov function and lower control energy consumption compared with fixed scalarization. The method can be applied to control problems for nonlinear and multi-agent systems, including cyber-physical complexes, autonomous agents, and energy nodes. 10.18721/JCSTCS.19210 519.8 multicriteria optimization Lyapunov function adaptive weights nonlinear systems stability https://infocom.spbstu.ru/article/2026.89.10/