Intermodulation Interference Detection in 6G Networks: A Machine Learning Approach

TL;DR

This work tackles passive intermodulation interference in multi-band wireless systems by deploying a regression-based detector that operates at the network edge without test tones. It shows that IM-induced interference manifests as a slope in the received interference-power versus PRB pattern, enabling a simple linear-regression test with $R^2$ and $\beta_1$ as decision metrics. The approach achieves linear-time complexity in the number of PRBs and demonstrates strong detection performance on real operator data, highlighting practicality for real-time radio resource management in 6G architectures. The results indicate a robust, non-surgical method suitable for edge computing contexts like O-RAN, with potential extensions to handle spectral leakage and other non-IM interferences.

Abstract

This paper demonstrates the use of machine learning to detect the presence of intermodulation interference across several wireless carriers. We show a salient characteristic of intermodulation interference and propose a machine learning based algorithm that detects the presence of intermodulation interference through the use of supervised learning. This algorithm can use the radio access network intelligent controller or the sixth generation of wireless communication (6G) edge node as a means of computation. Our proposed algorithm runs in linear time in the number of resource blocks, making it a suitable radio resource management application in 6G.

Figures (5)

• Figure 1: Next-generation base station using machine learning in an edge node on radio measurements to detect the existence of intermodulation interference.
• Figure 2: OFDM tones (dashed) and their respective odd-order intermodulation products (solid).
• Figure 3: Windowing spectral plots: (a) gain vs frequency plot of a window function (black) showing the side lobes (dotted peaks) and a line (red) the slope of which is equal to the rolloff rate (b) spectral leakage impact on intermodulation interference (black) in a receive frequency band $B^\prime$ (shaded) with side lobe peaks (blue dots) and a fitted line (dashed red) of the windowed intermodulation products.
• Figure 4: Allocation of the control- and user-plane channels to the uplink PRBs.
• Figure 5: Simulation results: (a) RIP vs PRB for two cases: intermodulation interference is present and is not present (b) receiver operating characteristic plot (c) confusion matrix and (d) normalized run time vs number of PRBs.