Efficient Data Dissemination via Semantic Filtering at the Tactical Edge

In modern tactical operations, efficient and effective communication combined with robust data processing can significantly enhance the decision-making processes. However, these environments often face challenges such as limited bandwidth, high latency, and network interruptions, which can hinder rapid decision-making. Contemporary solutions include data compression, efficient protocols, and modern data formats. Yet, more powerful solutions, that go beyond merely reducing data and packet size are needed. One promising approach focuses on the Value of Information (VoI), specifically extracting semantics from data. In this paper, we present a Semantic Filtering Platform (SemFil-P), capable of effectively minimizing the quantity of transmitted data by reducing the number of semantic redundancies (e.g. two nearly identical images taken within a short time window). The platform was developed entirely with open-source software components to ensure full reproducibility. The platform’s core component, the Embedding Component, leverages state-of-the-art Large and Small Language and Vision Models to transform any complex data into a semantic vector representation known as an embedding. Finally, we demonstrate the effectiveness of the platform in a simulated scenario using the MiniNet-WiFi emulator, showcasing substantial savings in bandwidth while preserving the semantic value of the data.