No abstract

We live in an interconnected world. The IoT, Internet of Things, is a reality. There are more things connected to internet than people on the planet. The communication channels has been established and standardized, things can be easily identified. We know how to interchange data but we do not know exactly what to interchange. There is a need to reduce the traffic over the networks when the number of nodes increases. Things could be sensors or devices with sensors. Things generate signals and since that the world is inferred through these signals, there is a need to interchange knowledge at low bit rates in a standardized way. The technology explosion that invades our lives has an influence not only on what is plausible to be standardized but opens possibilities not even thought. In the sensor world, intelligence has been concentrated in the point of acquisition, inside the transducer. Considering CPU capabilities, cost, low consumption and reduction of size, the sensor signal will be processed entirely into the smart sensor, even for cheap and simple sensors. The network will be used to interchange more meaningful information or knowledge. This paper focuses on algorithms suitable for any sensor signal that start from oversampled data and end with knowledge about the sensor signal behavior in a normalized scheme. The proposed standard is an important step towards an interconnected and manageable world.

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.

The need for the use of sensors networks in ever more efficient manner drives research methods for better information management. It would be useful to decrease the amount of managed data. Often we are interested in few noteworthy information of a signal (for example period, amplitude, time constant, steady state value, etc…) not in the whole waveform. The idea is to take less data, but acquire the same information. In a highly oversampled signal, each single sample does not carry a lot of information. From this point, two different algorithms are compared, in which only few samples are stored or transferred. This paper describes these two algorithms: the first one is the segmentation and labeling algorithm, also proposed for the definition of the new standard of the IEEE 1451, and the second one is based on compressive sensing theory. These two algorithms are compared, the simulations results are shown, and it is discussed which case could be more suitable for

No abstract

The

No abstract

No abstract