
292
B. Yang and F. Gersemsky
It is easy to verify that the total computational complexity of this algorithm is still at the
order
O(nr).
9 Perform the update
{v~(t - 1), Ai(t- 1)1i=
1,...,r(t-
1)},AN(t- 1), x_(t)
{v~(t),Ai(t)[i= 1,...,r(t-
1)}, AN(t)
via the PASTd algorithm and Eq. (14)
9 Apply AIC or MDL to estimate the number of signals r~(t) from (15) and
set
r(t)= r,(t) + 1
9 IF
r,(t) < r,(t-
1)
Remove {v~(t),
Ai(t)li
= r(t) + i,..., r(t- I)} which are no longer needed
ELSEIF
r,(t) > ro(t- 1)
Generate a new auxiliary eigencomponent by computing
v_.,(t)(t)=z__,(,_~)+x(t)/ll~(,_x)+~(t)ll
and
A,(,)(t)= ...