Robust-adaptive processing of sensor signals in conditions of noise, outliers, and measurement omissions

DOI: 10.31673/2412-9070.2026.318112

Abstract

The paper substantiates the problem of ensuring robust filtering in IoT systems in the presence of missing measurements, outliers, and noise distortions. It also demonstrates the limitations of traditional methods, which usually assume regular sampling, complete data availability, and relatively stable measurement conditions. It is noted that in real Internet of Things systems, sensor data are often affected by unstable communication channels, limited computational resources, external disturbances, and the physical characteristics of sensing devices. These factors complicate the direct application of classical state estimation algorithms without additional adaptation to incomplete and unreliable data streams. The paper considers the specific features of applying state estimation and recursive filtering methods to real-time sensor signal processing tasks. It is shown that the use of imputation methods is not always appropriate for such systems, since it may lead to additional delays, distortion of the informational content of the signal, and incorrect uncertainty estimation. Moreover, artificial reconstruction of missing values may create a misleading impression of measurement reliability, which is undesirable for monitoring systems where the estimation result is used for further analysis or decision-making. A robust adaptive approach to state estimation is proposed, which explicitly accounts for missing measurements in a recursive filtering system. Unlike approaches that preliminarily fill in data gaps, the proposed algorithm treats the absence of a measurement as a separate situation within the estimation process. During data gaps, the algorithm uses the process model to predict the current state while increasing the associated uncertainty according to the reduced amount of available information. When new measurements are received, the method adaptively regulates their influence on the estimate update, reduces the effect of anomalous values, and prevents abrupt jumps in the estimated state. Computer simulation and experimental validation of the proposed approach are performed using a dataset obtained from a real sensor, which contains noise, outliers, and frequent missing measurements. The obtained results show that the proposed algorithm provides stable and smoothed state estimates, remains operational even during prolonged periods of data absence, and maintains a consistent evolution of uncertainty.
It is also shown that the algorithm limits the influence of isolated anomalous measurements without losing the ability to respond to actual changes in the system state. The practical value of the approach is emphasized for IoT systems with limited computational resources, particularly in the context of Edge Computing, where local data processing, timely decision-making, and reduced dependence on cloud infrastructure are important. The proposed approach can be used in monitoring, diagnostics, and real-time sensor signal analysis tasks under conditions of incomplete data availability and imperfect sensing

Keywords: Internet of Things, real-time signal processing, adaptive filtering, sensor data, outliers, measurement gaps.