Semantic Extraction Model Selection for IoT Devices in Edge-Assisted Semantic Communications

Semantic communications offer the potential to alleviate communication loads by exchanging meaningful information. However, semantic extraction (SE) is computation-intensive, posing challenges for resource-constrained Internet of Things (IoT) devices. To address this, leveraging computing resources at the edge servers (ESs) is essential. ESs at the access points can support multiple SE models for uploaded SE tasks, making it crucial to select appropriate SE models based on diverse requirements and limited ES computing resources. In this letter, an SE model selection problem is studied in an edge-assisted semantic network. We aim to maximize the total semantic rate of all tasks under SE delay and accuracy requirements, and maximum ES computing capacity. The formulated NP-hard problem is transformed into a modified Knapsack problem equivalently. The proposed efficient approximation algorithm using dynamic programming can yield a guaranteed near-optimum solution. A key insight is revealed that the parameter $\varepsilon $ is an important indicator to balance the trade-off between the running time and obtained total semantic rate. Simulation results demonstrate the superior performance of proposed solution.

I. Introduction

With growing emerging applications and increasing data loads, the bottleneck of spectrum scarcity motivates a paradigm shift from conventional to semantic communications. As a novel paradigm focusing on the meaning of information rather than source data, semantic communications have revealed a significant potential to alleviate the communication loads over the network [1]. As a critical step, semantic extraction (SE) is to extract semantic information from the source data, which is computation-intensive [2], [3]. For example, the work in [4] used an universal model at devices, in which only simple background knowledge can be semantically extracted. Due to the limited capacity of internet of things (IoT) devices, it is challenging to implement the SE at device to satisfy diverse quality of service (QoS) requirements, especially for time-sensitive tasks.