Performance Evaluation of DL Models of Low-Size Datasets for AMC in MIMO System
Islam Helmy, Wooyeol Choi · 2024
Automatic modulation classification (AMC) is one of the foremost challenges in multi-input multi-output (MIMO) systems. In literature, various approaches have been presented where the deep learning (DL) models show remarkable performance. However, the existing methods mainly focus on training using large datasets, which is insignificant in high-mobility environments where rapid changes occur. In this study, we investigate the performance of existing DL models for small datasets and reduce the complexity by eliminating the equalizer. We apply two DL models to estimate the channel modulation among four modulation techniques. We also vary the number of transmitters for different datasets where the number of receiver antennas is 4 and 8, respectively. Furthermore, we verify the signal-to-noise ratio (SNR) of 0 to 40 dB. The results show that the convolutional neural network (CNN) performs better than the multi-layer perceptron (MLP) applied to the different datasets.