Digital Signal Processing Reference
In-Depth Information
we arrive at
9
=
;
.
c
0
= d
1
+ d
2
+ d
3
++ d
K
c
1
= d
1
u
1
+ d
2
u
2
+ d
3
u
3
++ d
K
u
K
c
2
= d
1
u
1
+ d
2
u
2
+ d
3
u
3
++ d
K
u
2
K
.
c
n
= d
1
u
1
+ d
2
u
2
+ d
3
u
3
++ d
K
u
K
.
(2.156)
Hence, the obtained general term of the form
K
c
n
= d
1
u
1
+ d
2
u
2
+ d
3
u
3
++ d
K
u
K
=
d
k
u
k
(2.157)
k=1
coincides with the complete integral (2.138) of the K th order difference equa
tion (2.137). This sets the prescription for the usage of the first K results
c
0
,c
1
,c
2
,...,c
K−1
deduced from the generating fraction. As mentioned, these
results c
0
,c
1
,c
2
,...,c
K−1
contain the coe
cients a
0
,a
1
,a
2
,...,a
K−1
of the
numerator polynomial of the generating fraction. In order to find all the terms
c
0
,c
1
,c
2
,..., of the rhs of (2.154) to the recurring order K, where c
n
is the
general member of the Taylor series
∞
n=0
c
n
u
n
, it is su
cient to have only
the K starting values{c
r
}(0≤r≤K−1) that should be given in advance.
By means of these K initial c
n
's, one could continue the sequence forward or
backward using the difference equation (2.137). In this way, one can either
predict the new c
n
's beyond the originally given{c
r
}(0≤r≤K−1) or
retrieve all the preceding c
n
's by recurring with the descending value of the
su
ces. In other words, if the K terms are given c
s+1
,c
s+2
,...,c
s+K
, one
could prolong/extrapolate this latter sequence by means of the relationship
9
=
;
. (2.158)
c
s+K+1
=−
1
b
K
(b
K−1
c
s+K
+ b
K−2
c
s+K−1
++ b
0
c
s+1
c
s+K+2
=−
1
b
K
(b
K−1
c
s+K+1
+ b
K−2
c
s+K
++ b
0
c
s+2
.
Thus the new c
n
's are predicted using the previous ones, so in general
K−1
c
s+K+m
=−
1
b
K
(predicting the new c
n
′
s). (2.159)
b
r
c
r+s+m
r=0
Alternatively, the calculation could also be carried out by recurring towards
Search WWH ::
Custom Search