Efficient information-theoretic model input selection

Of fundamental importance to proper system identification and virtual sensing is the determination and assessment of an optimal set of input signals independent of the final model form. If the system is causal and deterministic, it is possible to efficiently compute an information-theoretic optimal input set for a desired uniform accuracy of the target estimate and maximal dimension of the candidate input set. A branch and bound combinatorial optimization algorithm based on an estimate of joint mutual information is presented as part of a total coherent methodology of input selection.