The stair-like structure in the correlation integral of measured data is in
vestigated. It is shown that the stairs are caused by the limited floating-
point accuracy of the data in the measurement, and are completely different
from the noise in the measured phenomenon. It is found that a modified sin
gular value decomposition filter with very large projection dimension can i
mprove the precision of the data and smooth the stairs in the correlation i
ntegral.