Deep Learning Solutions for Large Scale Wireless Spectrum Monitoring

The ability to represent complex thoughts into a structured symbolic set, a language, and communicate these thoughts over large distances is one of the greatest achievements of the human race. By the end of 19th century, we had enhanced the speed and range of communication by playing with one of the fundamental forces of nature, the electromagnetic force. Since then there were countless applications that made use of radio spectrum for communication and control. An increase in the demand for this limited radio spectrum resulted in administrative licensing of the spectrum from 1930s. Thus one of our precious natural resources, the radio spectrum got divided into various licensed and unlicensed frequency bands. The resulting diverse spectrum regulations across countries, the complex electromagnetic space and further chaotic wireless spectrum usages resulted in a minimal real time knowledge of the wireless spectrum usage across frequency, time and space. Wireless spectrum usage knowledge is critical for various applications ranging from spectrum enforcement to dynamic spectrum sharing. This dissertation discusses the need for wireless spectrum monitoring and relevant usecases. New frameworks for obtaining wireless spectrum data at large scale and further extracting information from this collected data are detailed in this thesis. First, a new crowdsourced collaborative spectrum monitoring initiative, Electrosense, is introduced, which improves the accessibility of spectrum data for the general public. Electrosense is a non-profit organization based in Switzerland, consisting of four research groups: IMDEA Networks, Spain, Sero systems, Germany, armasuisse, Switzerland and Networked Systems, KU Leuven, which aims at improving the way how the radio frequency spectrum is used. The monitoring network consists of low-cost spectrum sensing nodes designed to monitor wireless spectrum on a large scale. The designed framework scales easily for large scale deployments, data collection, data storage and further data processing. Further, an open spectrum data as a service model is introduced to allow users to build diverse applications using the collected spectrum data. Second, this thesis explores different techniques for extracting relevant information from the collected spectrum data. Detecting anomalous behaviour in the wireless spectrum is one of the important techniques for various usecases. It is a demanding task as wireless spectrum anomalies can take a wide range of forms from the presence of an unwanted signal in a licensed band to the absence of an expected signal. Spectrum Anomaly Detector with Interpretable FEatures (SAIFE), an Adversarial Autoencoder (AAE) based anomaly detector is proposed which can detect anomalies by formulating the anomaly detection problem as an unsupervised outlier detection problem using Power Spectral Density (PSD) data. The model achieves good anomaly detection and localization capabilities along with lossy PSD data compression up to 120X. Third, after initial anomaly detection from the PSD data or spectrograms, better understanding about the anomalous signal is achieved by classifying the signal modulation type. A new data-driven model for Automatic Modulation Classification (AMC) based on long short term memory (LSTM) is proposed which yields an average classification accuracy of close to 90% at varying SNR conditions ranging from 0dB to 20dB. To reduce the data communication overhead from distributed sensors, the feasibility of classification using averaged magnitude spectrum data and on-line classification on the low-cost spectrum sensors are studied. Furthermore, quantized realizations of the proposed models are analyzed for deployment on sensors with low processing power. Fourth, the main issues with unsupervised anomaly detection are discussed in detail and a new framework is proposed for semi-supervised anomaly detection on the wireless spectrum data for crowdsourced sensors. The importance of user feedback in the anomaly detection process is investigated, to improve the performance of unsupervised anomaly detection. Furthermore, schemes for generalizing user feedback across sensors are also developed to close the anomaly detection loop. Finally, the benefits of wireless spectrum awareness specific to a spectrum sharing scenario in cognitive radio is analyzed by organizing various open challenges. State-of-the-Art (SoA) solutions are evaluated in terms of achieving full spectrum awareness and network performance improvements with the provided situational awareness. The presented results emphasize the need for open spectrum sensing frameworks such as Electrosense that can provide online spectrum awareness for enhancing wireless network performance.status: publishe