The implementation of representative new models in the AMR field on four different data sets (RML2016.10a, RML2016.10b, RML2018.01a, HisarMod2019.1) provides a unified reference for interested researchers.
Digital signal processing paper link: https://www.sciencedirect.com/science/article/pii/S1051200422002676
Arxiv link: https://arxiv.org/abs/2207.09647
1. Abstract
Automatic Modulation Recognition (AMR) detects the modulation scheme of a received signal in order to further process the signal without prior information and provides basic functionality when such information is missing. Recent breakthroughs in deep learning (DL) have laid the foundation for the development of high-performance DL-AMR methods for communication systems. Compared with traditional modulation detection methods, the DL-AMR method has achieved excellent performance, including high recognition accuracy and low error rate, due to the powerful feature extraction and classification capabilities of deep neural networks. Despite its great potential, DL-AMR methods also bring problems in complexity and interpretability, which affect practical deployment in wireless communication systems. This article aims to provide a review of current DL-AMR research, focusing on appropriate DL models and benchmark datasets. We further provide comprehensive experiments to compare state-of-the-art models for single-input single-output (SISO) systems from accuracy and complexity perspectives, and propose applications of DL-AMR in new precoding-inclusive multiple-input multiple-output (MIMO) scenarios. . Finally, existing challenges and possible future research directions are discussed.
2. Data set
3. Accuracy Curve
Figure 1 (a), (b), (c), (d) are the accuracy rates on RML2016.10a, RML2016.10b, RML2018.01a, HisarMod2019.1 respectively.
4. Training cost comparison
Table 1 Comparison of model size and training overhead on four data sets
5. Confusion matrix comparison
Figure 2 Confusion matrix comparison of 14 models on four different data sets (A, B, C, and D represent the four data sets respectively)
6. Remarks
If the above content is helpful to your research, please cite our article. If the methods involved in the article are helpful to your research, please also cite their article. (The models reproduced in the article are based on our understanding and the results of reproducing the models given in the article may have some differences from the original article.)
Relevant code and instructions can be found on github: https://github.com/Richardzhangxx/AMR-Benchmark