Abstract Automatic modulation recognition (AMR) is a key link in communication signal sorting, electromagnetic situational awareness, and electronic reconnaissance. Deep learning has an extremely strong feature extraction and classification capability and has stronger performance than traditional modulation recognition methods. However, most of the current modulation recognition techniques use a single feature source, which makes the neural network unable to obtain a full range of communication signal features, resulting in non-ideal recognition results. For this reason, this paper proposes the use of multi-transform domain feature inputs to improve the modulation recognition performance and solve the problem of recognition rate under low signal-to-noise ratio and small data sets.
Wang et al. (Fri,) studied this question.