A Semantic-Aware Multipacket Parallel Transmission Scheme for Bandwidth-Constrained Networked Control Systems

This article investigates a semantic-aware multipacket parallel transmission (MPPT) scheme for networked control systems (NCSs) using an autoencoder-based encoding–decoding framework. First, semantic encoding and decoding schemes are proposed based on autoencoder technology, aiming to effectively enhance the information density of data packets and reduce bandwidth usage during MPPT in NCSs. Through abstract semantic representations, raw data are mapped into a lower dimensional space for semantic encoding. Subsequently, these semantic representations are transmitted and utilized for semantic decoding, achieving the reconstruction of raw data. Considering multiple factors, such as semantic decoding errors, delays, and disturbances, stability analysis and $H_\infty$ controller design methods of NCSs under the semantic-aware MPPT scheme are explored based on Lyapunov stability theory. Using gradient descent optimization, semantic encoding–decoding parameters are pretrained offline to enhance the collaborative efficacy of semantic encoder and decoder, concurrently meeting real-time requirements of NCSs. Guided by operational feedback, subsequent parameter refinement is achieved through an incremental search algorithm, aiming to improve the performance of data reconstruction. Finally, the effectiveness of the proposed method is demonstrated through the example of a six-area power system.