Deep Learning Semantic Compression: IoT Support over LORA Use Case

Long Range (LORA) networks are serious candidates to support Internet of Things (IoT). Despite the scalability and range of LORA, yet many IoT devices need to send much more data than what is possible in this band. In this paper, a deep learning compression method to squeeze data and transfer its semantic is presented. Data from IoT equipment is considered as a time series and trains a neural network. Resulting neural network weights are periodically sent instead of sending all the IoT raw data. Anomalies are locally detected by a similar neural network and sent separately. The resulting architecture makes it feasible to use LORA for IoT devices that generate very large amounts of data.