Abstract: With the development of internet of things (IoT) technology, the number of communication protocols is increasing. However, a single IoT device supports a limited range of protocols, making it difficult to meet the needs of future applications. Current solutions have limitations: multi-mode communication chips, while supporting multiple protocols, are limited in quantity and lead to excessive complexity of integrated circuits; Cross-Technology Communication technology, although enabling interconnection between heterogeneous protocols, is constrained by the physical layer, resulting in limited precision of signals; Software-Defined Radio technology allows flexible adjustment of modulation through software, but it fails to meet real-time communication requirements and involves a long development cycle for platform adaptation. To address these issues, this article proposes a neural network-defined ubiquitous communication technology. By analyzing the mathematical theory of signal modulation, this technology utilizes neural network modules to equivalently implement signal modulations of multiple protocols. It can be deployed on various hardware platforms that support neural networks. Leveraging the parallel computing acceleration of AI chips, it significantly improves modulation efficiency. Meanwhile, utilizing the autonomous learning ability of neural networks, it can optimize modulation results, achieving self-evolution of signal modulation.