OPTIMIZATION OF INTELLIGENT TELECOMMUNICATION NETWORKS

Abstract

The paper presents the results of research on the use of machine learning in telecommunication networks and describes the basics of the theory of artificial intelligence. The impact of dynamic Bayesian network (DBN) and DNN on the development of many technologies, including user activity detection, channel estimation, and mobility tracking, is determined. The indicators of the effectiveness of communications based on the theory of information bottlenecks, which is at the junction of machine learning and forecasting, statistics and information theory, are considered. A neural network model that is pretrained for high-level tasks and divided into transmitter-side and receiver-side uses is investigated. The process of learning the model, which is performed after its adjustment, taking ino account the existing transmission channels, is considered. New ANN learning techniques capable of predicting or adapting to sudden changes in a wireless network, such as federated learning and multiagent reinforcement learning (MARL), are reviewed. The DBN model, which describes a system that dynamically changes or develops over time, is studied. The considered model provides constant monitoring of work and updating of the system and prediction of its behavior. Distributed forecasting of channel states and user locations as a key component in the development of reliable wireless communication systems is studied. The possibility of increasing the number of degrees of freedom of the generalized wireless channel G(E) in terms of: the physical propagation channel, the directional diagram of the antenna array and mutual influence, electromagnetic physical characteristics is substantiated. The impact of ultra-highresolution theory on the development of many technologies, including localization algorithms, compressed sampling, and wireless imaging algorithms, is also identified. Mathematical expressions for optimizing the functional characteristics of 5G/6G radio networks are presented using new, sufficiently formal and at the same time universal mathematical tools with an emphasis on deep learning technologies, which allow systematic, reliable and interpretable analysis of large random networks and a wide range of their network models and practical networks.