The contribution of this paper is twofold. First, it is shown that while ro
bust in terms of the average output error, the least squares estimate is se
nsitive to outliers with respect to the maximum output error. In fact the w
orst case output error of the least squares can go unbounded. Then, a const
rained logarithmic least squares for system identification is proposed. Ana
lytic center algorithms are presented to solve this constrained logarithmic
least squares problem.