Semantic Wireless Networks With Minimal Energy Consumption

In this letter, we develop a novel resource allocation scheme for energy-efficient semantic-aware wireless networks. The wireless users (WUs) extract semantic information from their raw source data during the common feature extraction (FE) phase, and then transmit their semantic features to their paired semantic decoders at the BS over the uplink channels. In random fading channels, the semantic similarity between reconstructed and source data may occasionally fall below some desired threshold resulting in a communications outage, which leads to the notion of average semantic rate. The proposed schemes minimize total energy consumption of WUs by deriving closed-form global optimal solutions for the transmit powers, processor frequencies and transmission duration of each WU, and duration of common FE phase. Compared to conventional bit-based communications, energy gain of proposed schemes increases rapidly with increase of semantic rate and average SNR.