Wireless Channel-adaptive Sensor-integrated Reconfigurable Intelligent Surface based on Deep Neural Network

This paper presents functional reconfigurable intelligent surface (RIS) with sensing capability that is applicable to various wireless communication environments at microwave, millimeter-wave (mmWave), and potentially sub- Terahertz spectrum. The devised deep-learning-based model enables the sensing system to be consistently applied to any propagation environment. The sensing information is used to redirect the RIS in accordance with the wireless channel condition. Thus, this enables the featured RIS to be cognitive without the need of any external feedback as suggested in various literature in the past. The presented RIS unit cell entails two varactor diodes on the top patch unit cells to enable phase shift of the reflected signal. For ease of implementation, via walls are used to minimize phase distortions due to mutual coupling while reflecting the incoming electromagnetic waves and signals to intended targets while simultaneously sensing the target's location. The proposed methodology is exemplified at 28 GHz using a 10 x 10 unit cells configuration. Measured and simulated results confirm that the deep neural networks have 99 % accuracy in inferring the test data and the devised 1-bit quantization unit cells redirect the signal from 15 ° to 60°. This novel concept ascertains low-cost, practical employment of a truly intelligent RIS for future wireless applications.