Deep Learning-Based In-Band Interference Detection and Classification

Abstract

Artificial intelligence has recently entered into the electromagnetic compatibility (EMC) area for design and analysis of EMC. Signal detection and classification are powerful tools for interference management to protect and ensure the availability of wireless communications. In this work, we study detection and classification of different types of interference signals, that interfere with a communication signal within the communication bandwidth. We propose two classification algorithms based on deep convolutional neural networks: a joint model based on one single neural network that distinguishes between all different types of interference, and a composite model based on multiple neural networks that each detects a distinct type of interference. The proposed algorithms are evaluated by Monte Carlo simulations. The composite model is shown to perform well in terms of high probability of correct classification and low probability of false classification. The joint model, however, tends to favor the pulsed interference signal and therefore yields too much false classifications.