Compressed radio transmission of spatial field measurements by virtual sensors

The remote exploration or monitoring of an environment often includes sensor measurements at multiple probe points and reconstruction of the spatial distribution of the observed physical quantity by a regression model. Especially for long distances between the observer and the environment, required data volume for transmitting a parametric description of the spatial distribution becomes critical. Simple physical models or assumption of parametric base functions do not provide sufficient prediction accuracy. Statistically based methods for field reconstruction such as Kriging or Gaussian Process Regression provide good accuracy, even if the measurements are overlaid with noise, provided all sensor data is transmitted. The new method presented in this paper calculates a small set of quasi optimal virtual sensor positions located in-between the real sensors. By transmitting only the predicted values of these virtual sensors, the spatial field can be reconstructed with less transmitted data and without significantly increasing the prediction error. The new approach was verified in a simulation scenario for a temperature field caused by diffusion and advection phenomena, which yielded a data compression by a factor of up to four. For large variations of the number of sensors and of the magnitude of measurement noise, the prediction error was always lower compared with the parametric base function models.