RATE SPLITTING FOR INTERFERENCE CHANNELS WITH DEEP REINFORCEMENT LEARNING

Date

2024-1-24

Author

Irkıçatal , Osman Nuri

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

65
views

53
downloads

Cite This

In recent advancements within communication systems, the rate-splitting multiple access (RSMA) technique has emerged as a crucial strategy to address interference, a persistent challenge in modern communication systems. This study examines the detailed application of precoding methodologies within RSMA, focusing on the complex environment of multiple-antenna interference channels and leveraging the capabilities of deep reinforcement learning. The primary objective is to optimize precoders and allocate transmit power for both common and private data streams, requiring a nuanced approach involving multiple decision-makers within a continuous action space. To address this challenge, the study proposes the utilization of a multi-agent deep deterministic policy gradient (MADDPG) framework. Within this framework, decentralized agents operate with partial observability but collectively learn from a centralized critic, navigating a multi-dimensional continuous policy space to optimize actions. Simulation outcomes highlight the effectiveness of the proposed rate-splitting method, achieving the information-theoretical upper bound for the sum rate in the single-antenna scenario. Even in multiple-antenna settings, its performance closely approaches this theoretical limit, outperforming benchmarks set by other techniques such as MADDPG without rate-splitting, maximal ratio transmission, zero-forcing, and leakage-based precoding methods. These compelling results emphasize the promising potential of this deep reinforcement learning-driven RSMA approach in communication systems by substantially mitigating interference and optimizing transmission rates and overall system performance.

Subject Keywords

Multi agent deep deterministic policy gradient (MADDPG), Rate-splitting multiple access (RSMA), Deep reinforcement learning, Interference channels

URI

https://hdl.handle.net/11511/108334

Collections

Graduate School of Natural and Applied Sciences, Thesis