Towards Efficient Event-Based Sampling for Goal-Oriented Semantic Communication

English (OpenAlex)

Massive Internet of Things (IoT) networks require communication-efficient strategies to manage large volumes of time-sensitive and task-relevant data. Traditional time-based sampling often leads to redundant transmissions and inefficient bandwidth utilization. This paper proposes a goal-oriented event-based sampling (EBS) framework that prioritizes semantically relevant and timely updates while reducing communication overhead. We consider two baseline techniques, namely Send-on-Delta (SOD) and Send-on-Area (SOA), and investigate predictive extensions based on linear prediction (LP) and quadratic regression (QR), including SODwLP, SODwQR, SOAwLP, and SOAwQR. In addition, an Adaptive Send-on-Delta (ASOD) strategy is introduced to improve responsiveness under non-stationary signal conditions. To better evaluate semantic fidelity, the proposed framework adopts a multidimensional assessment based on normalized mean square error (NMSE), principal component analysis (PCA) distance, spectral similarity, and Age of Incorrect Information (AoII). The methods are validated using an Electric Arc Furnace (EAF)-oriented power quality monitoring scenario, where harmonic-rich and time-varying signals are of practical interest. The revised evaluation further includes robustness analysis over AWGN and Rayleigh channels, a comparison of reconstruction methods using zero-order hold (ZOH) and first-order hold (FOH), and a communication-fidelity trade-off analysis. The results reveal a clear trade-off between communication reduction and semantic fidelity. On average over the considered threshold range, SOA achieves the highest data reduction, whereas ASOD provides the strongest overall performance in terms of reconstruction accuracy, structural similarity, harmonic preservation, and timeliness-aware behavior. The results also show that QR-based variants may offer stronger compression in some operating regions, but at the cost of substantially higher computational burden. Overall, the proposed framework offers a scalable and communication-efficient solution for real-time monitoring in industrial IoT and cyber-physical systems.