Deep learning powers reconfigurable intelligent surfaces in terahertz communications

The proliferation of Internet of Things (IoT) devices and broadband multimedia applications is increasing the demand for wireless data traffic and reinforcing the search for innovative solutions in wireless communications.

Advances in the application of reconfigurable intelligent surfaces for terahertz communications are reported. In a research paper published in intelligent computingA team of researchers led by Zhen Gao from Beijing Institute of Technology has introduced a new physical signal processing technique that leverages deep learning to enhance the functionality of reconfigurable intelligent surfaces in terahertz communication systems.

Reconfigurable intelligent surfaces are an innovative technology that passively reflects electromagnetic signals in a desired direction by adjusting the phase and amplitude of the elements. This ability to dynamically manipulate signals provides significant advantages over traditional communication systems, especially in indoor environments where the complexity of signal propagation can limit performance.

This technology can be integrated into current terahertz large-scale multiple-input multiple-output communication systems and passively reflects electromagnetic signals in a desired direction by manipulating their phase and amplitude, resulting in significant beamforming gains. to address the unique challenges of free-space loss and atmospheric attenuation. In the terahertz band.

Obtaining accurate channel state information is important for communication systems that use reconfigurable intelligent surfaces. Solutions based on compressed sensing and deep learning have been considered, but challenges remain in terms of computational complexity and storage requirements. Furthermore, existing studies often assume perfect channel state information and overlook the practical considerations of incomplete channel state information conditions.

A new transmission architecture based on deep learning is designed for large-scale multi-input, multi-output terahertz communication systems using reconfigurable intelligent surfaces. Their channel extrapolation method performs better reconstruction of channel state information than traditional alternatives while significantly reducing pilot overhead. Moreover, their beamforming method is more robust to incomplete channel state information.

This study introduces two methods:

• The SFDCExtra method is a spatial frequency domain channel extrapolation network method that uses deep learning to extrapolate a complete spatial frequency channel from a limited received pilot signal in communication systems using reconfigurable intelligent surfaces.
• The HBFRPD method uses deep learning to design the refractive phase of a hybrid beamformer and reconfigurable intelligent surface, especially in scattering-rich indoor scenarios brought about by incomplete channel state information and complex channel characteristics. Deal with challenges.

The effectiveness of the method was evaluated by numerical simulation. The SFDCExtra method aims to improve the efficiency and accuracy of channel estimation in wireless communication systems. This method provides promising advances in channel estimation performance while minimizing pilot overhead by exploiting spatial frequency correlation.

The researchers conducted a comprehensive evaluation, comparing it against various benchmark algorithms and assessing its robustness under different channel conditions and pilot configurations. Through detailed analysis and performance comparisons, the method demonstrates its effectiveness and versatility in revolutionizing channel estimation methods for next-generation communication architectures.

Researchers compared the performance of HBFRPD with other methods in multi-user communication systems. We tested the sum rate achieved by different methods assuming perfect channel state information and found that this method outperforms the others, especially at high transmit power, and is superior to non-repetitive It was observed that the properties allow for fast calculations. Additionally, incomplete channel state information causes interference between users to negatively impact the sum rate.

The results show that HBFRPD is robust to channel state information errors and outperforms other algorithms in such scenarios. The cumulative distribution function further supports the good performance and shows a higher probability of achieving the desired sum rate compared to traditional methods under conditions of incomplete channel state information.