Experimental Analysis of Deep Learning-Based Modulation Classification in Indoor Environments

Abstract

In this paper, we detail a Deep Learning (DL) based RF signal classification in an indoor environment. Software Defined Radio (SDR) based transmitter and receiver units are used for generating and processing the dataset consists of BPSK, QPSK, GMSK, 16-QAM, 64-QAM, GFSK, and CPFSK waveforms at different positions inside a building layout. We analyze the classification accuracy of individual modulation types at different receiver positions with two different transmitter positions. It is seen that for a given transmitter position, some of the modulation types perform better than the others at different receiver positions. It is also seen that the number of locations where the classification accuracy is greater than a certain threshold value varies with the transmitter positions. The results suggest that DL-based models taking into account the transmitter and receiver positions can be used for efficiently designing cognitive radio-based communication systems for a given propagation environment in indoor scenarios.