Year: 2024
Type: article
Abstract: Aiming at the deep learning-based radio signal modulation mode recognition algorithm, a Transformer network model of radio signal modulation mode recognition algorithm is proposed. Firstly, the Transformer network model is built, which contains 12 layers of Transformer encoder and 3 layers of fully connected layers; secondly, the original data is partitioned into sequences with a fixed window size, and the models with different window sizes are compared, and the model with the optimal window size is comprehensively selected for experimental comparisons; lastly, the optimal model is compared with the other benchmark models, and the modulation signal is used to recognize the modulation mode of radio signals using the RaidoML2016.10a international standard dataset. The sample is the RaidoML2016.10a international standard dataset to train the model. The experimental results show that the Transformer model algorithm can achieve recognition accuracy of 94.63% under the condition of signal-to-noise ratio of 16 dB, as well as an average recognition accuracy of 63.18%, and the maximum recognition accuracy and average recognition accuracy are higher than those of the current benchmark models by 2%~11% and 1%~7%, respectively. The model has superior recognition and fast convergence speed.
DOI: 10.54097/a54bpx86
Institution: Southwest Minzu University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4405809001