A Continuous-State Reinforcement Learning Strategy for Link Adaptation in OFDM Wireless Systems

Abstract

This paper presents a machine learning approach based on the concept of reinforcement learning andMarkov Decision Processes for link adaptation in orthogonal frequency-division multiplexing systemsthrough adaptive modulation and coding. Although machine learning techniques have attracted attentionfor link adaptation, the schemes proposed so far are based on off-line training algorithms, which makethem not well suited for real time and real-world operation. The proposed solution learns the bestmodulation and coding scheme for a given signal-to-noise ratio by interacting with the environment(the radio channel). Unlike the other solutions, it does not rely on an off-line and computationallyintensive training phase. Simulation results show that the proposed technique outperforms the wellknownsolution based on look-up tables for adaptive modulation and coding, and it can potentiallyadapt itself to distinct characteristics of the environment or the receiver radio frequency (RF) front end.