VISTA-LoRa: Task-Oriented Semantic Visual Transmission for Ultra-Low-Bitrate LoRa
IEEE Wireless Communications Letters, vol.15, pp.3174-3178, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 15
- Publication Date: 2026
- Doi Number: 10.1109/lwc.2026.3691811
- Journal Name: IEEE Wireless Communications Letters
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC
- Page Numbers: pp.3174-3178
- Keywords: DCT, LoRa, Reed-Solomon, semantic reliability, task-oriented semantic communication, ultra-low-bitrate, visual IoT
- Middle East Technical University Affiliated: Yes
Abstract
Long Range (LoRa)’s ultra-low bitrate and strict payload constraints make conventional visual transmission infeasible for embedded Internet of Things (IoT) nodes. This letter presents VISTA-LoRa, a fully deterministic and training-free task-oriented semantic framework that transmits only clarity-selected and morphology-adapted visual regions required for a downstream detector. The VISTA-LoRa pipeline integrates clarity-aware spatio-temporal Region of Interest (ROI) selection, an airtime-normalised morphology-variant optimiser, and a Low-Frequency/High-Frequency (LF/HF)-weighted Discrete Cosine Transform (DCT) encoder with Reed-Solomon (RS) protection, followed by multi-scale semantic verification at the receiver. Experiments on real tracked frames show that VISTA-LoRa achieves the highest semantic reliability among the evaluated schemes, obtaining a mean receiver-side confidence of 0.78 and a 4.5–6.4× payload reduction while preserving perfect structural consistency. Such a payload-efficient, high-confidence semantic uplink is particularly valuable for delay-tolerant visual IoT applications such as remote inspection, environmental monitoring, and low-power autonomous systems that must operate within the severe energy, bandwidth, and duty-cycle restrictions of LoRa-class networks. Despite using stronger quantisation and spatial reduction, VISTA-LoRa maintains 91.7% of the original semantic confidence compared to 75.0% for baseline approaches under identical constraints demonstrating its suitability for visual reporting over ultra-low-bitrate LoRa links.