Real-time Ambient Temperature Estimation of Mobile Electronic Devices

The thermal response and operation reliability of a mobile device are greatly affected by the environmental temperature. While traditional mobile systems fail to sense the ambient temperature in real time without the help of external information. Here, we propose a learning-based framework, which encompasses off-line parameter extraction and online iterative calculation, to estimate the ambient temperature with merely internal thermal sensors measurements and the rough power information. We model the system with a low-order lumped-parameter thermal network. Our model has no special requirements for the placement of sensors, and the parameter identification does not rely on architectural details. We implement and validate our method on a real smartphone with actual benchmarks under changeable ambient temperature. Experimental results show that the online estimator has a mean absolute error of 0.7°C. Our work may benefit not only the dynamic thermal management but also the function expansion of various mobile platforms.