Sensor signal preprocessing techniques for analysis and prediction

This paper presents a signal processing technique that employs oversampling and identification of important samples to determine signal behavior and tendency. Sensor signal windows of random lengths are vectorized and classified to fit into only eight predefined types, and in conjunction with time indexes vectors, they can predict future values, steady state value and an estimation of the sensor signal function. The techniques developed allow the representation of any class of sensor signal for further analysis. The computational cost is quite low so they can be implemented in real time into smart sensors with low cost microcontrollers. Therefore, it is also an ideal technique to preprocess the sensor signal to mark regions of interest to more sophisticated processes.