Semantic-Driven Adaptive Holographic Content Transmission Over a Wireless Network: A Gradient Projection Method-Assisted DRL Approach

This paper presents a semantic-driven adaptive framework for holographic content transmission over a wireless network comprising multiple base stations (BSs) and Internet of Things (IoT) devices. Within this framework, IoT devices capture holographic content and transmit it to BSs in a semantic-driven manner to accomplish diverse computational tasks, including classification, segmentation, and reconstruction. To optimize system performance, we formulate an optimization problem encompassing BS association, semantic compression ratio (SCR) selection, and power resource allocation, with the objective of maximizing task utility that quantifies task completion effectiveness. Given the non-convex nature and NP-hardness of the problem, we first derive the feasibility conditions for power allocation. Subsequently, we develop a gradient projection method (GPM)-assisted deep reinforcement learning (DRL) algorithm that learns optimal adaptive strategies for BS association and SCR selection, while employing GPM for efficient power allocation. Simulation results demonstrate that our proposed framework can improve task utility by up to 78% compared to the baselines.