On high-order error criteria for system identification

A new family of algorithms has been considered for the purpose of identifying FIR (finite impulse response) systems by means of minimizing the sum of the output error to the degree 2K, K = 1, 2, . . . . A heuristic argument is presented to the effect that utilization of high-order error criteria (K > 1) will yield, in certain cases, considerable improvement in the performance (lower estimation error can be expected for the same amount of data). A simple approximate expression is derived which allows the choice of the most suitable value of K for the given application.