Air-ODE Neural Network with Distributed RISs Aided Communication Systems

Analog machine-learning hardware platforms promise to be faster and more energy efficient than their digital counterparts. Specifically, over-the-air analog computation allows offloading computation to the wireless environment through carefully constructed transmitted signals. In addition, reconfigurable intelligent surface (RIS) is emerging as a promising solution for next-generation wireless communication networks, offering a variety of merits such as the ability to tailor the communication environment. Based on the merits brought by the RIS, we design and implement the residual-based block that uses over-the-air computation and demonstrate it for inference tasks in an ordinary differential equation (ODE) deep neural network. We engineer the ambient wireless propagation environment through distributed RISs to design such an architecture, which is termed as over-the-air ordinary differential equation (Air-ODE) neural network. In contrast to the conventional digital ODE-inspired neural network architecture, the Air-ODE block leverages the physics of wave reflection and the reconfigurable phase shifts of RISs to implement an ODE-based block in the analog domain. We then validate the entire Air-ODE neural network based on a complex-valued image reconstruction task. The simulation results illustrate that the analog Air-ODE can achieve similar performance to the digital ODE network and the deployment of the Air-ODE block can achieve 2.08 times and 1.29 times performance gain on peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), respectively.