Russian operator T2 (formerly Tele2) has officially launched the large-scale implementation of an intelligent platform based on artificial intelligence. The new solution radically changes the approach to managing LTE networks, allowing mobile internet speed to increase by 10–15% solely through a software upgrade, without purchasing and installing additional hardware.

In conditions where access to foreign telecommunications equipment is limited and traffic demand is growing, the use of machine learning algorithms becomes not just a technological innovation, but a strategic necessity for maintaining connection quality.

From static to dynamic: how the algorithm works

Traditional base stations operate on the principle of static resource distribution. Available frequency bands (LTE-800, LTE-1800, LTE-2100, and LTE-2600) are divided according to pre-written templates. During peak hours, when many subscribers are within the range of a single cell, this approach leads to uneven load and a drop in speed.

The new T2 system is based on the concept of Self-Organizing Networks (SON). Instead of rigid templates, the neural network conducts continuous multi-factor analysis of the radio spectrum in real time. Algorithms assess critical parameters every second:

  • Received signal quality (RSRP and RSRQ indicators);
  • Current load of the base station sectors;
  • Technical characteristics of the subscriber's smartphone modem (LTE Cat category);
  • Traffic consumption profile (video streams, web surfing, or packet data transmission).

Based on this data, the system instantly selects the optimal combination of carrier frequencies for a specific device. This process is called dynamic Carrier Aggregation. Additionally, the mathematical model predicts the user's movement between coverage sectors, adjusting connection parameters in advance. This prevents session drops during the handover from one tower to another.

Economic effect and Shannon's theory

Implementing the AI module allows operators to expand network capacity without capital expenditure on purchasing new transceivers. This is particularly relevant in the current economic situation. The system also effectively combats inter-cell interference — the superposition of radio waves from neighboring transmitters, which is especially important in conditions of dense urban development.

From the perspective of fundamental communication theory, channel capacity is limited by bandwidth and the signal-to-noise ratio (Shannon-Hartley theorem). Applying machine learning algorithms allows the actual data transfer rate to approach this theoretical limit as closely as possible. This is achieved through targeted noise minimization and the dynamic combination of disparate frequency bands.

Comparative analysis shows that while standard cell operation provides a baseline speed, the AI-optimized mode provides a 10–15% increase. At the same time, congestion management changes from a simple packet queue to prioritization based on traffic profile analysis, which minimizes latency.

Scaling plans

At the current stage, the pilot project has been successfully completed in the Moscow region and St. Petersburg. The operator has confirmed that the technology is ready for widespread implementation. By the end of 2026, T2 plans to scale this technology to its entire network infrastructure in regions of presence, which should lead to an overall improvement in mobile internet quality for subscribers across the country.