Automating Labeling of ML Clusters using KPM Data for Interference Detection

Abstract

SenseORAN is a framework designed to enhance spectrum sensing and radar detection capabilities within the Citizens Broadband Radio Service (CBRS) band by leveraging the Open Radio Access Network (O-RAN) infrastructure. It aims to improve the detection of radar signals that may interfere with 5G signals by integrating AI and machine learning models deployed as xApps on Near-Real-Time RAN Intelligent Controllers (Near-RT RIC). 

Wfocus on automating the process of labeling Key Performance Metrics (KPMs) data collected from a softwareized LTE network. To address this challenge, we employ k-means clustering algorithms to process the KPM data and partitions data into distinct clusters based on similarities. In our implementation, the algorithm analyzes the KPM data and automatically detects which clusters represent interference and which do not. 

Our methodology involves several key steps: first, preprocessing the KPM data to ensure it is suitable for clustering, including dropping features with little predictive value and handling missing values. Next, we apply the k-means algorithm to identify clusters within the data. We then validate the clusters to ensure they accurately represent interference patterns. Finally, we integrate the labeled data into our machine learning models, which are deployed as xApps on the Near-RT RIC. This automated labeling process significantly reduces the manual effort required and provides a scalable and efficient solution for real-time interference detection in wireless communications.