.png){kind=icon}
.png){kind=icon}
Return
A multichannel framework for multitask learning fusion in modulation recognition tasks
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2025.112601.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Automatic modulation recognition is an essential process linking signal detection to signal demodulation, and it is a promising technology for improving spectrum usage efficiency in cognitive radio. With the advent of fifth generation mobile communication technology, wireless communication systems have had massive data throughput. As a result, integrating artificial intelligence techniques with modulation recognition has emerged as a key focus area in the communication sector. To enhance the precision of signal modulation recognition, this study proposes a multitask learning fusion multichannel network modulation recognition framework—Multitask Residual Convolution Long Short Memory-Transformer Deep Neural Network. The primary task network is a two stream network composed of Convolutional Neural Network module, Long Short-Term Memory module and Transformer-Encoder module, allowing for the simultaneous extraction of both time and frequency features from the signal. The auxiliary task network is composed of residual convolution module paired with Transformer-Encoder module, designed for extracting the power spectral density characteristics of straightforward signals. Ultimately, the features derived from both the primary task network and the auxiliary task network are combined. The auxiliary network enhances the primary task network’s ability to characterize features, thereby boosting the neural network’s overall versatility and precision. The experimental findings indicate that the proposed model achieved peak recognition accuracies of 99.73%, 93.9%, and 94.1% on the three datasets, respectively. Moreover, the recognition accuracy of the proposed model is better than the baseline model in the low Signal-to-Noise Ratio environment (-18 decibel ∼0 decibel).
Journal
IF:
8
Papers:
5.4K
Citations:
3.5W
Organization
No organization information available