Real-World Considerations for Deep Learning in Wireless Signal Identification Based on Spectral Correlation Function

This paper proposes a convolutional neural network (CNN) model which utilizes the spectral correlation function (SCF) for wireless radio access technology identification of GSM, UMTS, and LTE signals without any priori information about bandwidth and/or the center frequency. The sensing and classification methods are applied on the baseband equivalent signals. The proposed SCF based CNN method is implemented in two different settings entitled CASE1 and CASE2. In CASE1, signals are jointly sensed and classified. In CASE2, sensing and classification are conducted in a sequential manner. The proposed CNN method eliminates threshold estimation processes of the classical estimators. It also eliminates the need to know the distinct features of signals beforehand. Over-the-air real-world measurements are used to show the robustness and the validity of the proposed method compared to the existing deep learning networks. The real-world measurement-driven spectrum sensing and classification results denote that the proposed CNN architecture is more robust, memory-efficient, and fast trainable compared to the existing deep learning networks and conventional detectors. The performance of SCF based CNN method is discussed in detail for both settings. Furthermore, the novel dataset is shared along with this study to make it publicly available.