Alternating minimax approximation with unequal restraints