Modulation Recognition IQ Samples Across Heterogeneous Sensors

In this paper, we design and present a testbed implementation and the resulting dataset for modulation recognition from real-world imperfect scans. We describe our efforts to build a testbed of heterogeneous spectrum sensors (low-cost RTL-SDR and mid-cost USRP) and a controlled transmitter in order to facilitate real-world data collection for modulation recognition from partial and biased scans. The testbed allows data collection, in the form of IQ samples, across five common modulation types (BPSK, QPSK, 8PSK, 16-QAM, 64-QAM) and with a controlled bandwidth overlap, 50% and 100%, between the transmitter and sensor. Using the data collected from the testbed, we then demonstrate modulation recognition from real-world, low-cost and imperfect scans. Our dataset is positioned to facilitate new research into signal recognition from partial and biased real-world spectrum data. Furthermore, the data can also facilitate research on the applicability of various sensor platforms to support real-world spectrum characterization.