Deep Learning Based Real-Time Spectrum Analysis for Wireless Networks

Wireless networks are indispensable in today's production and automation. In an industrial environment, they are used in particular for networking moving or inaccessible parts of a factory. Due to difficult signal propagation conditions as well as coexisting wireless networks, transmission errors that can lead to malfunctions in the application are often very difficult to identify. Wireless remote monitoring systems are essential tools for diagnosing such faults. This work is intended to contribute to the improvement of failure analysis tools for wireless networks. The fundamental approach relies on an extension of standard protocol analysis tools by sensitive measurements of a spectrum analyser. Using Machine Learning-based image processing algorithms, individual frames of different radio technologies are detected in real time and classified according to their communication standard. A subsequent statistical processing or anomaly detection provides an abstract view to the spectrogram and enables a cross connection to the protocol analysis. Using IEEE 802.11 as an example, it can be shown that frames and frame collisions can be detected with a high degree of accuracy. In conclusion, the performance of the developed method is evaluated and compared with a protocol-based monitoring system.