MIPT Develops Algorithm to Boost 5G Network Capacity
The software, called ALPACA, can handle demanding network tasks even in the most challenging mobile scenarios.
Neural networks are playing an increasingly prominent role in the telecom industry. One of the most promising applications is using them to improve the performance of network equipment. That is particularly important when building fifth-generation and next-generation communications infrastructure, where signal latency needs to be kept to a minimum.
System Predicts Radio Channel Behavior
Researchers at the Moscow Institute of Physics and Technology (MIPT) have taken that approach with an algorithm called ALPACA (Asymmetric Loss Prediction Algorithm for Channel Adaptation). Designed for 5G and next-generation networks, it can maintain good signal quality in challenging mobile scenarios, even when suboptimal transmission parameters were selected during network deployment. The system uses a convolutional recurrent neural network to predict potential changes in channel quality.
One of the key innovations is an asymmetric loss function. ALPACA is deliberately “penalized” for overly optimistic predictions that could cause transmission failures. The approach reduces channel resource consumption by up to 40% compared with existing solutions, increasing network capacity for ultra-reliable low-latency communications (URLLC) applications.
ALPACA Outperforms Conventional Tools
“The research is based on applying neural network models to a key challenge in URLLC networks – selecting the optimal data transmission mode under rapidly changing radio channel conditions. Conventional methods cannot always meet strict latency requirements measured in milliseconds and reliability requirements with packet loss probabilities below 10⁻⁵, particularly in scenarios involving highly mobile users. That led me to explore whether neural network algorithms could provide a better approach,” said Kirill Glinsky, first author of the study and a researcher at MIPT’s Laboratory of Intelligent Communication Systems and the Wireless Networks Laboratory at the Kharkevich Institute for Information Transmission Problems of the Russian Academy of Sciences.
The MIPT system significantly outperforms existing approaches designed for similar tasks. According to the researchers, current tools, including heuristic algorithms and conventional neural network models, either fail to predict channel changes accurately enough or do not account for the asymmetric consequences of prediction errors: An overly optimistic forecast risks packet loss and transmission failure, while an overly conservative one wastes network resources.
In Step With Global Telecom Trends
The development of ALPACA also reflects the latest global push to make communications systems more autonomous and increasingly manageable through software. As early as 2023, Russian researchers were describing potential ways to integrate artificial intelligence directly into 5G architecture.
In 2024, researchers at the Moscow’s HSE University Artificial Intelligence Research Center developed software for modeling radio channels in 5G and 6G wireless communications using ray tracing and machine learning. The software could determine how radio waves propagate between a transmitter and receiver, convert ray-tracing data into sequences of frames, and configure and train a neural network using those data.
The Technology Foundation Is Already in Place
Just as important, the technology base needed to deploy developments such as MIPT’s is taking shape. It includes new Russian-made base stations that support a software-based transition from LTE to 5G. Equipment from YADRO, for example, supports this capability, with production beginning in 2025. Under a national project, Russia plans to manufacture and deploy 75,000 domestically produced base stations by 2030.
MIPT’s work underscores the emergence of a software-driven, intelligent layer in Russia’s telecom infrastructure. Technologies of this kind could increase network capacity while reducing the need to install costly additional hardware. The development is particularly timely as Russia prepares to launch 5G in its cities with populations above one million and transition to domestically produced base stations.
