Minimax stack filtering in a parameterized environment

An optimization algorithm based on the minimax error criterion is presented. The algorithm assumes a parameterized environment where the input to the stack filter can be modeled by some parameterized stochastic process. First, it is assumed that the signal parameters are fixed while the noise parameters vary over a predetermined range. The algorithm's task is to pick an optimal stack filter, among the set of all stack filters, which minimizes the worst effect of the noise in the minimax sense. Then, the optimization algorithm is generalized to allow parameter variations in the underlying input signal itself. This scheme can, for instance, deal with data-dependent noise. The optimization goal remains the same. Both algorithms can be solved using a 0-1 linear program. A linear program would suffice where soft decisions are acceptable.<>