Neural Reconstruction of Nonlinear Sensor Input Signal

The paper presents a new approach to analysis metrological properties of neural networks used for reconstruction of input signal of a nonlinear sensor. The general idea of the reconstruction realization consists in its decomposition to static and dynamic parts properties of which are investigated independently. The analysis of the process of signal conversion and reconstruction is made by using the error model containing both propagation of error from input to the output and composition of the propagated errors with the errors introduced by elements realizing the conversion and reconstruction. Theoretical considerations have been illustrated by results obtained from measurement and simulation experiments.