Synchronization Mechanism: Preliminary Attempt to Enable Wireless Semantic Communication With Heterogeneous Knowledge Bases

Despite the potential of semantic communication for 6G, the assumption of homogeneous knowledge bases (KBs) between communication parties cannot be guaranteed. The heterogeneity of KBs (KB-heterogeneity) degrades performance. Therefore, this letter proposes a communication-efficient synchronization mechanism to mitigate the performance degradation level (PDL) caused by KB-heterogeneity, especially the varied data distribution. The proposed mechanism re-frames the problem as an estimation problem of the source model, involving transmitting the semantic information of shared data as a ‘synchronization (sync) signal’ before data transmission. Furthermore, the letter introduces Sync-signal-driven model estimation (SSME) and data-model iterative optimization (DAMIO) synchronization algorithms to utilize the sync signal for source model estimation. Compared with SSME, DAMIO improves source model estimation accuracy, when shared data and source data have significantly different distributions. Experimental results demonstrate that the proposed mechanism and algorithms effectively reduce PDL in both intra-group and inter-group KB-heterogeneity scenarios, while still meeting wireless resource constraints.