A popular approach to solve this ill-posed problem is by the regularization technique [1], [2]. It restricts the admissible solution to be a smooth function. The problem can be formulated as minimizing an error function defined as,
where K is the total number of measurements available,
the first term is the data constraint, i.e., the residual error in
surface fitting to the measurements, the second term 
is the smoothness requirement placed on f, and 
is a regularization
constant which controls the tradeoff between the data constraint and
the smoothness constraint.
