Explainable Reinforcement Learning and Lightweight LLM-Assisted Resource Allocation for 6G Networks: A Compact Review

Abstract

The evolution of wireless communication toward sixth-generation (6G) networks has increased the complexity of resource allocation due to ultra-dense deployments, heterogeneous users, dynamic channel conditions, mobility, interference, and strict service-level requirements. Future 6G networks are expected to support ultra-reliable low-latency communication, massive machine-type communication, intelligent transportation, digital twins, industrial automation, and immersive applications. These services require real-time and adaptive allocation of radio and computing resources such as transmit power, bandwidth, spectrum blocks, scheduling priority, and edge processing capacity. Reinforcement Learning (RL) and Deep Reinforcement Learning (DRL) have emerged as promising solutions because they can learn adaptive allocation policies through interaction with the wireless environment. However, most RL-based resource allocation models operate as black-box systems and provide limited explanation for their decisions. Large Language Models (LLMs) can support reasoning and human-readable explanation, but their high computational cost makes direct edge deployment difficult. This paper presents a compact review of RL, DRL, LLM-assisted wireless management, lightweight language models, and Explainable Artificial Intelligence (XAI) for 6G resource allocation. The review identifies research gaps and highlights a future direction where RL performs real-time resource optimization, while XAI and lightweight LLM layers provide interpretable decision support.