Deep Learning-based Automatic Modulation Recognition Algorithm in Internet of Things

Automatic modulation recognition (AMR) is one of the most promising topics in internet of things (IoT), which endows the capability of adaptive modulation to adapt various complicate environment. This paper proposes a deep learning-based method to distinguish frequency shift keying (FSK), phase shift keying (PSK) and quadrature amplitude modulation (QAM) with high accuracy. We train convolution neural network (CNN) on amplitude and phase (AP) samples, which are two-dimension matrices consisting of amplitudes and phases extracted from complex-valued baseband signals. The performance of our proposed method is confirmed in both light-of-sight (LOS) and non-light-of-sight (NLOS) channel. In addition, considering that IoT devices may not have powerful computing and sufficient power, it is proposed that the application of AP samples with appropriate dimension for CNN training can reduce consumed power of IoT devices with maintaining algorithm performances.