Evolutionary Knowledge-Enhanced Semantic Communication: Transition From SISO to MIMO

Semantic communication (SemCom) conveys seman tic meaning rather than raw data to address growing data demands within limited spectrum resources. However, existing SemCom systems remain confined to single-input single-output (SISO) configurations, creating a gap with modern multiple input multiple-output (MIMO) infrastructure. While some recent studies have explored MIMO-enabled SemCom, they assume perfect channel state information (CSI), an impractical requirement given real-world interference and complex propagation. To bridge this gap, we propose the MIMO-augmented Knowledge Enhanced SemCom (MIMO-KESC) system, integrating MIMO specific modules including layer mapping, precoding, and channel equalization into conventional SemCom architecture. Its key innovation is a complex-valued neural equalizer that processes inherently complex signals to preserve phase information, delivering more accurate equalization than conventional minimum mean square error (MMSE) methods without perfect CSI. The system also incorporates a knowledge enhancement component for decoding highly compressed semantics. Simulations demon strate that MIMO-KESC delivers robustly superior transmission quality (e.g., higher Bilingual Evaluation Understudy (BLEU) scores) under diverse wireless conditions and antenna setups, paving the way for scalable, high-performance SemCom in future MIMO networks.