Powering the IoT through embedded machine learning and LoRa

The Internet of Things (IoT) technology is rapidly changing the way we live and the number of connected devices are increasing at an exponential pace. However, two key challenges are the battery life for off-grid IoT applications and the ability of edge devices to communicate over long range. Raw data transmission poses as a very power hungry activity for any device. The conventional cellular wide area networks are power-hungry and incompatible for battery-operated IoT devices. There is a need for low-power edge computing devices that reduce the transmission payload and integrate Low-Power Wide-Area Network (LPWAN) technologies, which offer a wide range connectivity while still providing a long battery life. One of the most promising LPWAN technologies today is LoRa. In this paper, we present a solution that employs machine learning on the edge device and performs low power transmission through LoRa. We demonstrate the use case of our solution through a field trial conducted in China for sow activity classification. By implementing embedded machine learning with LoRa, we could compress the transmitted data by 512 times and extend the battery life by three times. A very low energy expenditure of 5.1 mJ per classification result is achieved.