Abstract:
In extremely low signal-to-noise ratio (SNR) region, the useful features of the signal are weakened by higher-power noise, making it difficult for conventional direction-...Show MoreMetadata
Abstract:
In extremely low signal-to-noise ratio (SNR) region, the useful features of the signal are weakened by higher-power noise, making it difficult for conventional direction-of-arrival (DOA) estimation methods to adequately exploit and extract the low-SNR signal features. Thus, a generative adversarial network (GAN) is presented to learn the underlying features and complex distributions of high-SNR covariance matrices. The introduced GAN establishes a mapping between low-SNR and high-SNR covariance matrices, thereby generating first-rate high-SNR covariance matrices that closely resemble real high-SNR matrices. Also, it effectively captures signal features that are overwhelmed by excessive noise power. Additionally, to improve the performance of convolutional neural network (CNN)-based DOA estimation models in medium-to-high SNR ranges, a deep residual attention network (DRANet) is designed to significantly enhance DOA estimation accuracy in such SNR region. By integrating residual and attention modules, the network effectively filters key features. This enhances feature learning and adaptability, allowing it to capture DOA-related features more proficiently. The experimental results indicate that the developed GAN-DRANet approach can approach the CRLB in the extremely-low SNR range and improves the estimation resolution limits of the other two DL-based methods, DNN and CNN, in medium to high SNR conditions.
Published in: IEEE Internet of Things Journal ( Early Access )
Funding Agency:
Keywords assist with retrieval of results and provide a means to discovering other relevant content. Learn more.
- IEEE Keywords
- Index Terms
- Neural Network ,
- Deep Neural Network ,
- Low Signal-to-noise Ratio ,
- Convolutional Neural Network ,
- Deep Network ,
- Covariance Matrix ,
- Characteristic Signals ,
- Generative Adversarial Networks ,
- Attention Module ,
- Noise Power ,
- Residual Network ,
- High Signal-to-noise Ratio ,
- Can Approach ,
- Cramer-Rao Lower Bound ,
- Deep Residual Network ,
- Signal-to-noise Ratio Conditions ,
- Signal-to-noise Ratio Range ,
- Direction Of Arrival Estimation ,
- Deep Attention ,
- Signal-to-noise Ratio Region ,
- Model-based Methods ,
- Reconfigurable Intelligent Surface ,
- Convolutional Layers ,
- Antenna Array ,
- Multiple-input Multiple-output ,
- Binary Vector ,
- Sigmoid Activation Function ,
- Deep Learning ,
- Nonlinear Function ,
- Internet Of Things
- Author Keywords
Keywords assist with retrieval of results and provide a means to discovering other relevant content. Learn more.
- IEEE Keywords
- Index Terms
- Neural Network ,
- Deep Neural Network ,
- Low Signal-to-noise Ratio ,
- Convolutional Neural Network ,
- Deep Network ,
- Covariance Matrix ,
- Characteristic Signals ,
- Generative Adversarial Networks ,
- Attention Module ,
- Noise Power ,
- Residual Network ,
- High Signal-to-noise Ratio ,
- Can Approach ,
- Cramer-Rao Lower Bound ,
- Deep Residual Network ,
- Signal-to-noise Ratio Conditions ,
- Signal-to-noise Ratio Range ,
- Direction Of Arrival Estimation ,
- Deep Attention ,
- Signal-to-noise Ratio Region ,
- Model-based Methods ,
- Reconfigurable Intelligent Surface ,
- Convolutional Layers ,
- Antenna Array ,
- Multiple-input Multiple-output ,
- Binary Vector ,
- Sigmoid Activation Function ,
- Deep Learning ,
- Nonlinear Function ,
- Internet Of Things
- Author Keywords