Thesis

Automated management of radio spectrum and transmission power in heterogeneous shared spectrum networks

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2025
Thesis identifier
  • T17195
Person Identifier (Local)
  • 201981876
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • The huge benefits of internet connectivity and the impact of communication on human existence make connectivity extremely important in today’s world. Digitization has impacted education, health, commerce, and governance and recently ignited the fourth industrial revolution. Increased demand for wireless communication services has driven the need for cheaper wireless telecommunication infrastructures and affordable connectivity, which are some of the benefits of dynamic spectrum access (DSA) technologies. DSA technologies’ spectral maximization uses the spectrum-sharing paradigm that allows a timed or space shared use of spectrum. This permits a central spectrum coordination of vertical (unequal priority) access fixed nodes and a device-based (distributed) coexistence management of equal priority (horizontal) sharers. A detailed study of distributed coexistence management techniques/protocols revealed that flexible spectrum access is achieved when devices use similar techniques/protocols (homogeneous networks) and when this is not the case (heterogeneous networks) there is a huge contention for limited spectrum. Furthermore, homogeneous and heterogeneous networks suffer contention when the number of available resources is fewer than the number of requesting radios. This thesis investigates the coexistence management of unequal priority of typical DSA systems in two countries and highlights spectral availability in the two nations. It quantifies the impact of government policy on spectral availability. It also bolstered the huge information overload necessary for existing central coordination systems and the challenge of coexistence management of dynamically located radios. This work further addresses the high contention among dynamically located radios (nodes/base stations/access points), operating as equal priority users, for limited available spectrum and the huge information overload in central coordination by framing a central artificial intelligence (AI) algorithm for optimal resource assignment to dynamically located nodes. Its Artificial Intelligence models (trained Reinforcement learning algorithms) are designed to optimize the reuse of spectrum at different transmitter power levels among such nodes while simultaneously limiting interference among them. These minimize inter-device interference and maximize their signal-to-noise plus interference ratio (SINR), enabling spectral overlay, underlay, and reducing information overhead. Thus permitting more equal access devices to share resources without harmful interference. Two AI models, a two-stage optimization RL algorithm (TSA) and a joint optimization RL algorithm (JOA), are designed to solve the optimization problem and learn to assign spectrum and power resources to devices. The TSA used two reward functions, while the JOA used a single reward function to arrive at optimal solutions. These were compared with DSA’s random and recursive resource assignment. Two indices assessed the number of nodes with good SINR experience (assignment performance) when two to four available channels were assigned to 3 to 8 radios or nodes. The TSA and random assignment were inconsistent in providing nodes with good quality of service (fair assignment) and a reasonable request performance (assigning resources to requesting nodes). The JOA model resulted in a close to exclusive (ideal) resource assignment in its assignment performance and was at par with device requests with the two staged and random assignments in most scenarios examined. JOA, therefore, resulted in an average of 20% increase in request assignments as against exclusive assignments and an above 20% in assignment performance compared with other techniques in all network scenarios examined. These performance outcomes are helpful in shared spectrum technologies that adopt a random or recursive approach in resource assignment. An AI algorithm can improve the quality of service and number of nodes using limited available resources at the request instance. It is also valuable for regulators, as intelligent resource sharing can increase the number of nodes that share resources. In these scenarios, the node’s properties provided individualistic resource assignment, while the predictive algorithm provided an instantaneous search for optimal resource sharing. Thus taking advantage of nodes’ ability to accommodate a level of interference. Future works include improving the state space RL optimization formulation and training episodes. Also, advancements in deep Q-learning may solve increased state space dimensions, reducing the effect of state space approximation.
Advisor / supervisor
  • Stewart, Robert W. (Electrical engineer)
  • Crawford, David (David H.)
Resource Type
DOI
Funder

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of thesis T17195 2025-01-24 Public Download