Modulation recognition driven by signal enhancement
The existing modulation recognition algorithms based on deep learning theory require a large number of IQ signal samples during the training phase. It is difficult to obtain a large number of samples in complex electromagnetic environments, resulting in a decrease in the generalization performance o...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Beijing Xintong Media Co., Ltd
2024-04-01
|
| Series: | Dianxin kexue |
| Subjects: | |
| Online Access: | http://www.telecomsci.com/zh/article/doi/10.11959/j.issn.1000-0801.2024090/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The existing modulation recognition algorithms based on deep learning theory require a large number of IQ signal samples during the training phase. It is difficult to obtain a large number of samples in complex electromagnetic environments, resulting in a decrease in the generalization performance of modulation recognition algorithms based on deep learning. A signal enhancement based modulation recognition (SEBMR) algorithm was proposed to address the issue of poor network generalization ability. Firstly, a feature extraction and reconstruction module was designed to capture the global features of IQ signals. Secondly, an IQ signal enhancement network based on auxiliary classifier generative adversarial network (ACGAN) was proposed to achieve dual enhancement of sample quantity and quality. Finally, the support vector machine algorithm was employed to achieve modulation recognition and classification. To achieve recognition of debugging signals in complex channels, reconstructed signals representing global features were for training, and IQ baseband signals which experienced wireless fading were used for testing. The experimental results show that the proposed method can achieve better recognition accuracy performance in small sample training sets and fading channel environments compared to existing recognition methods based on long short-term memory (LSTM), convolutional neural network (CNN), attention mechanism, etc. |
|---|---|
| ISSN: | 1000-0801 |