SmartMQ: Utility-Driven Semantic-Difference-Based Transmission Filtering for Message Queuing Telemetry Transport-Based Internet of Things Systems

Abstract

Preprint. This is the authors' own version and has not been peer reviewed. The proliferation of Internet of Things (IoT) devices generates large volumes of sensor data, much of it adding little information beyond prior transmissions. This paper presents SmartMQ, a semantic-aware transmission filtering framework operating at the application layer over the Message Queuing Telemetry Transport (MQTT) protocol without protocol modifications. SmartMQ couples a sensor-agnostic semantic difference module, built on a dimensionless embedding of the temporal character of the signal in which every feature is normalized online by a robust estimate of the stream's own noise scale, with a utility-driven decision mechanism. The decision jointly evaluates semantic relevance, energy state, communication delay, message cost, trend memory, information freshness, and reference deviation, a combination that, to our knowledge, no existing publisher-side MQTT filtering method provides: a trend-memory term graduates the incentive to publish with sustained suppression, and a graduated age-pressure term, backed by a hard freshness cap, guarantees bounded Age of Information. Because the embedding is dimensionless, publish decisions are invariant under positive affine transformations of the sensor output, confirmed by 100% decision agreement under unit conversions and rescalings of the temperature trace. Across four signal scenarios SmartMQ suppresses 81.2%-86.2% of samples and attains the highest matched-rate semantic selectivity among the compared methods (1.86x-2.16x, consistent across 20 independent runs per scenario), with reconstruction overhead below the injected noise level. Replaying real SHT30 temperature and humidity traces under a single identical configuration yields 85.5% and 81.6% suppression at 2.36x and 2.67x selectivity. The reproducibility package (implementation, SHT30 dataset, and scripts that regenerate every reported result) is archived at https://doi.org/10.5281/zenodo.21414411