Task-Oriented Prediction and Communication Co-Design for Haptic Communications

TL;DR

This paper introduces a task-oriented co-design framework for haptic communications that optimizes resource use by considering prediction errors and communication reliability, significantly reducing bandwidth needs.

Contribution

It presents a novel joint prediction and communication design tailored for task-specific requirements, incorporating real-world data and synthetic data generation for optimization.

Findings

Achieves up to 77.80% reduction in wireless resource consumption.

Demonstrates effectiveness using 5G New Radio as an example.

Validates the framework through numerical and experimental results.

Abstract

Prediction has recently been considered as a promising approach to meet low-latency and high-reliability requirements in long-distance haptic communications. However, most of the existing methods did not take features of tasks and the relationship between prediction and communication into account. In this paper, we propose a task-oriented prediction and communication co-design framework, where the reliability of the system depends on prediction errors and packet losses in communications. The goal is to minimize the required radio resources subject to the low-latency and high-reliability requirements of various tasks. Specifically, we consider the just noticeable difference (JND) as a performance metric for the haptic communication system. We collect experiment data from a real-world teleoperation testbed and use time-series generative adversarial networks (TimeGAN) to generate a large amount of synthetic data. This allows us to obtain the relationship between the JND threshold, prediction horizon, and the overall reliability including communication reliability and prediction reliability. We take 5G New Radio as an example to demonstrate the proposed framework and optimize bandwidth allocation and data rates of devices. Our numerical and experimental results show that the proposed framework can reduce wireless resource consumption up to 77.80% compared with a task-agnostic benchmark.