Adaptive Resource Allocation Optimization Using Large Language Models in Dynamic Wireless Environments

TL;DR

This work addresses constrained wireless resource allocation in dynamic environments where $NP$-hard problems and changing objectives hinder conventional DL methods. It introduces LLM-RAO, a prompt-based optimizer that uses a meta-prompt, $OPRO$ fine-tuning, and in-context learning to generate and refine RA solutions without retraining, while incorporating QoS and queue dynamics. The approach tackles OFDMA RB allocation, MU-MIMO mode selection, and user scheduling with two utilities: $\sum_k \hat{r}^{(k)}$ and $\sum_k \log(1+\hat{r}^{(k)})$. Empirical results on an IEEE 802.11ax uplink show up to 40% improvements over DRL and 80% over analytical methods, and up to 2.9× gains when objectives change, highlighting practical relevance for adaptive wireless networks.

Abstract

Deep learning (DL) has made notable progress in addressing complex radio access network control challenges that conventional analytic methods have struggled to solve. However, DL has shown limitations in solving constrained NP-hard problems often encountered in network optimization, such as those involving quality of service (QoS) or discrete variables like user indices. Current solutions rely on domain-specific architectures or heuristic techniques, and a general DL approach for constrained optimization remains undeveloped. Moreover, even minor changes in communication objectives demand time-consuming retraining, limiting their adaptability to dynamic environments where task objectives, constraints, environmental factors, and communication scenarios frequently change. To address these challenges, we propose a large language model for resource allocation optimizer (LLM-RAO), a novel approach that harnesses the capabilities of LLMs to address the complex resource allocation problem while adhering to QoS constraints. By employing a prompt-based tuning strategy to flexibly convey ever-changing task descriptions and requirements to the LLM, LLM-RAO demonstrates robust performance and seamless adaptability in dynamic environments without requiring extensive retraining. Simulation results reveal that LLM-RAO achieves up to a 40% performance enhancement compared to conventional DL methods and up to an $80$\% improvement over analytical approaches. Moreover, in scenarios with fluctuating communication objectives, LLM-RAO attains up to 2.9 times the performance of traditional DL-based networks.

Figures (5)

• Figure 1: OFDMA RBs in a 20 MHz channel.
• Figure 2: Overview of the proposed LLM-RAO. In the initial meta-prompt generation phase, the AP generates and transmits a meta-prompt including communication information of users and task description to the LLM server. During the solution inference process, the LLM server finds the optimal solution by iteratively generating solutions and receiving feedback on the score and constraint violations from the external toolkit. After completing the configured iterations, the LLM server delivers the final optimal solution to the AP.
• Figure 3: A detailed example of the inference process for solving the RA problem. The orange text represents the solution generated by the LLM server, and the green text describes the score for the generated solution as evaluated by external toolkits in the history of RA solutions and scores.
• Figure 4: Performance analysis of LLM-RAO and baseline methods in terms of data rate and proportional fairness under different scenarios. (a) scenario 1. (b) scenario 2. (c) scenario 3. (d) scenario 4.
• Figure 5: Evaluation of performance and adaptability of LLM-RAO and baseline methods under changing environment and scenario configuration. DRL (scenario 1, $K=50, R^{(k)} =0$) represents the DRL-based scheme using the model trained under scenario 1 with 50 users and no minimum rate constraint.