The shift toward Industry 4.0 has driven the adoption of smart factories, where modular and reconfigurable production systems require flexible and scalable communication solutions. To support the growing number of field-level devices—such as sensors, actuators, and controllers—wireless communication is emerging as a promising alternative to traditional wired systems. However, with limited radio resources and high device density, efficient spectrum reuse becomes critical.

This PhD investigates the use of 6G subnetworks—small, localized wireless networks embedded within machines and robots—to enable short-range, low-power communication among internal devices. While this approach enhances flexibility and spectrum reuse, it introduces a key challenge: inter-subnetwork interference, particularly in dense and dynamic factory environments.

The thesis proposes scalable methods for mitigating interference and optimizing radio resource usage. In the first phase, an unsupervised graph neural network-based solution is developed for transmit power and frequency allocation, designed to operate with minimal overhead in large-scale deployments. This ensures high spectral efficiency and reliable communication within each subnetwork. 

In the second phase, the research introduces a goal-oriented approach, where resource allocation is guided by the functional needs of the subnetwork (e.g., real-time control) rather than purely communication metrics. Novel decentralized methods are proposed for power and channel allocation, along with application-aware strategies like proximity control to further enhance efficiency in mobile and dynamic settings.

These methods significantly improve the scalability, and efficient reuse of limited radio resources for wireless communication in smart factories, offering key contributions to the design of future 6G-enabled industrial networks.