Reducing Temperature Reading Errors in Embedded Systems

Modern embedded systems form the basis for applications that enable accurate and reliable acquisition of environmental data. In particular, the processing of analog signals such as temperature in microcontroller-based systems plays a critical role in real-time decision-making processes. Temperature measurement is not only a fundamental parameter in environmental monitoring, but also a key variable in a wide range of applications including industrial automation, smart buildings, energy management systems, medical devices, food storage and safety, and climate control systems. Accurate temperature sensing is essential for ensuring system efficiency, safety, and reliability; even small measurement errors can lead to significant performance degradation, energy losses, or unsafe operating conditions in sensitive applications. However, these analog signals produced by sensors are often affected by factors such as environmental noise, electromagnetic interference, and sensor accuracy, which directly impact the system's performance. Applying digital filtering techniques to such noisy measurements enhances signal stability, enabling more reliable and repeatable results. In this context, the Exponential Moving Average (EMA) filter, which has low computational costs, and the Kalman filter, which is more sophisticated and accuracy-focused, are two powerful approaches that are frequently compared. The EMA filter is a simple and fast technique that filters past data using exponential weights. Due to this feature, it is widely preferred in microcontroller systems with limited resources. On the other hand, the Kalman filter provides a statistical estimate by considering both the current measurement and the previous state of the system, making it superior in terms of accuracy. However, the Kalman filter is computationally more expensive due to its requirement for matrix operations and a system model. This study aims to experimentally compare the performance of EMA and Kalman filters on a microcontroller-based ambient temperature measurement system. In the experimental application conducted using an LM35 temperature sensor and an STM32F4 microcontroller platform, both filters were run on the same analog data set; the results were evaluated in terms of filtering accuracy, response time, and computational load. Furthermore, this study contributes to simplifying and improving temperature measurement in embedded systems by demonstrating how appropriate filtering techniques can effectively reduce noise and enhance measurement quality. The findings provide practical guidance for selecting the most suitable filtering method in different application scenarios, ultimately facilitating more accurate, stable, and efficient temperature monitoring solutions.