The advent of spectrum crowdsensing systems is challenging the way the spectrum is monitored, thanks to the large deployment of Internet of Things (IoT) devices with Software-Defined Radio (SDR) capabilities. One critical problem of these networks is that they collect a massive amount of data not labeled. However, there is the need to classify the technologies used in this data for spectrum management policies such as spectrum allocation, spectrum sharing, dynamic spectrum access, and anomaly detection. Although raw In-Phase and Quadrature (IQ) data may be used to do so, they entail orders of magnitude larger backhaul bandwidth and storage capabilities, which can not be afforded in today’s deployments. On the other hand, the sensors’ equipment in crowdsensing systems is often limited (e.g., RTL-SDR dongle is limited to sampling the spectrum in only 2MHz of bandwidth).
In this work, we propose a framework to address this problem that relies solely on Power Spectrum Density (PSD) data collected by low-cost RTL-SDR sensors. First, we design a method for efficient transmissions detection that can run on the backend of the crowdsensing platform and can support near-real-time spectrum measurements. Second, we introduce a data-driven deep learning solution for technology classification, that leverages the above knowledge of transmission features and the frequency they use. The classifier uses a single PSD segment from which information is extracted and used within a network consisting of an Autoencoder and LSTM, to make inferences about the technology type. Our evaluation is conducted with real-world data collected by the crowdsensing system Electrosense. We show that the proposed model yields an average classification accuracy close to 95\% in a wide range of Signal-To-Noise ratio conditions, it can well discriminate between technologies that have the same wideband modulation schema, and it is robust to classification errors using just a portion of transmission’s bandwidth when the latter is greater than that of the sensor.
Alessio Scalingi is a Ph.D. student of the Pervasive Wireless Systems Group since January 2020. He received both his Bachelor’s and Master’s Degree in Computer Engineering from the University of Naples Federico II in 2015 and 2019, respectively. He held his research for his Master’s thesis in the Computer Science Lab at Saint Louis University (USA), where he focused on developing data-driven solutions for a system that proactively predicts Virtual Network Functions resources using artificial intelligence. His current main research interests include Collaborative Spectrum Sensing, Machine Learning, Spectrum Anomaly Detection, O-RAN, 5G Networks.
This event will be conducted in English