Optimal design of multi-objective multi-sensor systems

In many measurement applications when several parameters need to be determined, multi-sensor systems are used. A maximum accuracy for each parameter is usually desired. However, the measurements may be nonlinear functions of the parameters and in the inverse problem for one parameter all measurements are used. Thus, a design that is good for the determination of one parameter may be bad for another. A compromise is required to satisfy the accuracy demands for all of them. Common methods to tackle these kind of problems are known as optimal design of experiments. This paper presents an extension of these methods with a special focus on multi-objective multi-sensor systems with suboptimal estimators. For error propagation, the approach makes use of the so-called unscented transformation, which is an efficient way to determine mean and variance of random distributions which undergo a non-linear transformation.