Estimation and improvement of data corruption in serial-optical communications with machine learning in embedded systems

In outdoor environments, sunlight causes corruptions in serial optical communication, rendering communication with electricity meters impossible. Dynamically adjusting the infrared light intensity of the Bluetooth optical port reader used for meter reading provides an alternative solution to this issue. A model capable of determining the appropriate emission intensity for environmental conditions has been developed using TensorFlow-Lite. This model is implemented on a microcontroller, which adjusts the necessary emission intensity for communication based on ambient conditions read through a light sensor connected to the microcontroller, and also identifies situations where communication is not feasible. As a result of this study, ambient light levels were read with a light sensor attached to the microcontroller and classified using the TensorFlow-Lite model. Consequently, it was observed that the deep learning network adjusted the emission intensity according to varying outdoor conditions, where sunlight intensity could be high or low, thereby ensuring the feasibility of optical communication.