Dataset for modulation classification and signal type classification for multi-task and single task learning

Wireless signal characterization is a growing area of research and an essential tool to enable spectrum monitoring, tactical signal recognition, spectrum management, signal authentication for secure communication, and so on. Recent years have witnessed several deep neural network models to perform single task signal characterization such as radio fingerprinting for emitter identification, automatic modulation classification, spectrum sharing, etc. However, with the emergence of 5G and the prospects of beyond 5G communication, there has been an increased deployment of edge devices that requires lightweight neural network models to perform signal characterization. To this end, a multi-task learning model that can perform multiple signal characterization tasks with a single neural network model has been proposed. However, due to the novel nature of multi-task learning as applied to signal characterization, there is a lack of a corresponding dataset with multiple labels for each waveform. In this paper, we openly share a synthetic wireless waveforms dataset suited for modulation recognition and wireless signal (protocol) classification tasks separately as well as jointly. The waveforms comprise radar and communication waveforms generated with GNU Radio to represent a heterogeneous wireless environment.