Improved technique for statistically optimum inversion of optical data in presence of measurements and model noise

The problems of optimum inversion in the presence of random noise are analyzed. Two main kinds of noise are considered: the random errors of measurements and random errors of physical model. It is studied the optimization of the numerical inverse problem solution concerning both noises. Using the statistical estimation ideas is discussed for this consideration. Specific features of every noise to influence on the limitation of information content of the optic experiment and on implementation of inversion are distinguished. The quantitative criteria to evaluate information content of input data and procedure of their interpretation are proposed. The latter is aimed to optimize the solution in presence of random errors of the model as well as errors of measurements and, moreover, to correct used model by the measurements being interpreted. An arbitrary accompanied and a priori information can be used. For example, a priori estimations of the sought and model parameters, correlations between them, non-negativity of values etc., can be included. The peculiarity of the inversion method is an essentially large number of variables and increased stability should be provided. The original iterative process of linear inversion characteristic to statistical optimizations are being proposed for this in the algorithm elaborating.