Over-the-Air ODE-Inspired Neural Network for Dual Task-Oriented Semantic Communications

TL;DR

This paper introduces an innovative over-the-air ODE-inspired neural network leveraging reconfigurable intelligent surfaces for dual tasks in semantic communications, achieving high efficiency and performance comparable to digital methods.

Contribution

It proposes the Air-ODE network architecture that uses RIS and analog wave physics to implement ODE neural networks in the wireless environment, enabling dual tasks with improved spectrum efficiency.

Findings

Air-ODE achieves similar performance to digital ODE-inspired networks.

The Air-ODE block doubles reconstruction quality and tagging accuracy.

Analog Air-ODE network demonstrates effective dual-task performance in semantic communication.

Abstract

Analog machine-learning hardware platforms promise greater speed and energy efficiency than their digital counterparts. Specifically, over-the-air analog computation allows offloading computation to the wireless propagation through carefully constructed transmitted signals. In addition, reconfigurable intelligent surface (RIS) is emerging as a promising solution for next-generation wireless networks, offering the ability to tailor the communication environment. Leveraging the advantages of RIS, we design and implement the ordinary differential equation (ODE) neural network using over-the-air computation (AirComp) and demonstrate its effectiveness for dual tasks. We engineer the ambient wireless propagation environment through distributed RISs to create an architecture termed the over-the-air ordinary differential equation (Air-ODE) network. Unlike the conventional digital ODE-inspired neural network, the Air-ODE block utilizes the physics of wave reflection and the reconfigurable phase shifts of RISs to implement an ODE block in the analog domain, enhancing spectrum efficiency. Moreover, the advantages of Air-ODE are demonstrated in a deep learning-based semantic communication (DeepSC) system by extracting effective semantic information to reduce the data transmission load, while achieving the dual functions of image reconstruction and semantic tagging simultaneously at the receiver. Simulation results show that the analog Air-ODE network can achieve similar performance to the digital ODE-inspired network. Specifically, for the image reconstruction and semantic tagging task, compared with the analog network without the Air-ODE block, the Air-ODE block can achieve around 2 times gain in both reconstruction quality and tagging accuracy.