Semantic Communications for Efficient Spectrum Maps Construction and Signal Source Location

Spectrum maps, which depict the distribution of radio signal strength, offer a potential solution to alleviate the issue of spectrum scarcity by optimizing network design. Nevertheless, the transmission of spectrum maps encounters difficulty due to the vast amount of data. To address this problem, the letter draws inspiration from semantic communication and proposes a framework of spectrum semantic transmission for the task of signal source localization. The quantization semantic extraction strategy is first proposed to acquire discrete spectrum semantics, which results in a significant reduction of data volume. Moreover, to achieve semantic restoration, the fully connected neural network is utilized to conduct nonlinear regression. Furthermore, we introduce an autoencoder-based completion scheme that effectively enhances the accuracy of signal source localization in the context of sparse data. Simulation results demonstrate the significant effectiveness of the framework, transmitting only 12.5% of data compared to the baseline while maintaining location robustness.