Distributed interference management in dense subnetworks
Author
Kristensen, Jeppe Bro
Term
4. term
Education
Publication year
2022
Submitted on
2022-05-24
Pages
70
Abstract
Interferens er en grundlæggende udfordring i tætte, multiagent radiokommunikationsmiljøer. In‑X‑delnet forventes at håndtere ekstreme krav i 6G. Dette projekt undersøger, om forstærkningslæring, specifikt Q‑learning, er egnet til distribueret interferenshåndtering via kanalallokering i frekvensdomænet i tætte In‑X‑delnet. Metoden er en Monte Carlo‑simulation af et muligt brugsscenarie, der omfatter tæt delnetsudrulning, radiopropagationsmodeller og en Markov‑beslutningsproces med tilstands‑, aktions‑ og belønningsdesign. Ydelsen evalueres i en simuleringsramme, og resultaterne antyder, at Q‑learning kan anvendes til ressourceallokering. Samtidig begrænses præstationen af valgte tilstandskvantiserings‑tærskler og ikke‑uniforme aktionsfordelinger.
Interference is a fundamental challenge in dense, multi‑agent radio systems. In‑X subnetworks are envisioned to meet the extreme requirements of 6G. This thesis investigates whether reinforcement learning, specifically Q‑learning, is a suitable approach for distributed interference management via frequency‑domain channel allocation in dense In‑X subnetworks. The method is a Monte Carlo simulation of a plausible use case that includes a dense subnetwork deployment, radio propagation models, and a Markov decision process with defined state, action, and reward. Performance is evaluated within this simulation framework, and the results indicate that Q‑learning can be used for resource allocation, with performance constrained by the chosen state quantization thresholds and non‑uniform action distributions.
[This summary has been generated with the help of AI directly from the project (PDF)]
Documents