Wlan interference identification using a convolutional neural network for factory environments

Abstract

—Factory communication systems require highly reliable links with predictable performance and quality of service in order to avoid outages that can damage the production-line process. Communication anomalies can be caused by narrowband interference which is difficult to identify and track from the time-domain information only. This paper describes a methodology for classifying increasing severity and types of interference in order to improve throughput prediction. Received signal strength (RSS) data is collected from both a ray-tracing simulation and a Wireless Local Area Network (WLAN) measurement campaign with a transmitter mounted on an actual automated guided vehicle (AGV). Scalogram time-frequency images are computed from the RSS signal and a convolutional neural network (CNN) is then trained to recognize the spectral features and enable the interference classification. The block random interference could be correctly classified on over 65% of the occasions in the ray-traced channel at 30 dB SNR.