Resource allocation based on radio intelligence controller for Open RAN toward 6G

In recent years, the open and standardized interfaces for radio access networks (Open RAN), promoted by the standard organization O-RAN alliance, demonstrate the potential to apply artificial intelligence in 6G networks. Among O-RAN, the newly introduced radio intelligence controller (RIC), includin...

Full description

Saved in:
Bibliographic Details
Main Authors: Wang, Qingtian, Liu, Yang, Wang, Yanchao, Xiong, Xiong, Zong, Jiaying, Wang, Jianxiu, Chen, Peng
Other Authors: School of Computer Science and Engineering
Format: Article
Language:English
Published: 2023
Subjects:
Online Access:https://hdl.handle.net/10356/171576
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:In recent years, the open and standardized interfaces for radio access networks (Open RAN), promoted by the standard organization O-RAN alliance, demonstrate the potential to apply artificial intelligence in 6G networks. Among O-RAN, the newly introduced radio intelligence controller (RIC), including near-real-time RIC and non-real-time RIC, provides intelligent control of the radio network. However, existing research on RIC only focuses on the implementation of interfaces and progress, while ignoring the resource allocation between near-RT RIC and non-RT RIC which is essential for ultra-low latency in 6G networks. In this paper, we propose a reinforcement learning-based resource allocation scheme that minimizes service latency by optimizing requests allocated and processed between near-RT RIC and non-RT RIC. Specifically, we aim at improving the request acceptance and minimum the average service latency, in our policy, we apply the Double DQN to decide whether the requests are processed at near-RT RIC or non-RT RIC and then allocate the near-RT RIC resource to finish the requests. Firstly, we define and formulate the resource allocation problem in RIC by the Markov decision process framework. Then we propose an allocation scheme based on the Double Deep Q network technique (Double DQN), with two variations (Double DQN with cache and Double DQN without cache) for handling different request types. Extensive simulations demonstrate the effectiveness of the proposed method in offering the maximum reward. Additionally, we conduct experiments to analyze the updating of cached AI models and the results show that the performance of the proposed method is always optimal compared to other algorithms in terms of latency and accepted number of requests.