A Hybrid Machine Learning Framework For Intelligent Interference Classification And Adaptive Mitigation In 5G/6G Wireless Networks

Abstract

The advancements in wireless networks towards 5G and beyond (6G) have led to dense spectrum use, heterogeneous radio-access technologies (RATs), ultra-massive connectivity, and network dynamism, making interference between RATs and user equipment (UEs) inevitable. Efficient management of interference is essential for ensuring reliable wireless communications. Traditional interference mitigation approaches often rely on heuristic-based threshold settings, pre-configured resource allocation plans, or signal-processing techniques which may not be adaptive enough for time-varying interference scenarios. While some recent works have shown promise in leveraging machine learning (ML) algorithms to detect and mitigate wireless interference, existing interference classification approaches mostly treat interference classification and mitigation problems separately or are proposed for a single interference scenario. In this paper, we propose a Hybrid Machine Learning Framework for Intelligent Interference Classification and Adaptive Mitigation in 5G/6G Wireless Networks. Our framework leverages multi-domain wireless features consisting of signal strength, SINR, spectrum attributes, time-domain attributes, and interference power together with a hybrid ensemble learning model to classify wireless interference. We propose an Interference Severity Estimation module that quantifies the detrimental impact of detected interference, followed by an Adaptive Mitigation Decision module that determines the most suitable mitigation technique conditioned on interference type and severity. Extensive experimentation results demonstrate that our proposed framework achieves an intelligent and adaptive solution to become interference-aware in wireless networks. We evaluate our framework based on classification accuracy, precision, recall, F1-score, interference reduction, SINR gain, computational complexity, and inference latency.