Digital Signal Processing Reference
In-Depth Information
where m > 0 (m = 1, 2, 3,...). If (4.56) is compared with the identity
c n+s u −n−1 = c s
1
u
c n+s+1 u −n−1
u +
(4.58)
n=0
n=0
the following relation is obtained
c n+s+1 u −n−1
n=0
q (s)
r+1 e (s)
q (s)
1
u−q (s)
1
q (s 2 e (s)
r
1
= c s
−e (s 1
−e (s 2 −· · ·−
r+1 −· · ·
u−q (s)
2
u−q (s)
r+1
−e (s)
q (s)
1
u−γ (s)
1
(s 1 ] 2
u−γ (s)
2
(s r ] 2
u−γ (s)
= c s
(4.59)
−· · ·−
r+1 −· · ·
where
= q (s)
γ (s)
n
= q (s)
n
+ e (s)
n
(s)
n
] 2
n+1 e (s n .
(4.60)
On the other hand, by employing (4.21) and (4.27) we arrive at
+ e (s+1)
n−1
= α (s+1)
n−1
q (s)
n
+ e (s)
n
= q (s+1)
n
(4.61)
q (s)
n+1 e (s)
= q (s+1)
n
e (s+1)
n
= [β (s+1)
n
] 2
(4.62)
n
= α (s+1)
n−1
γ (s)
n
(s n ] 2 = [β (s+1)
] 2
(4.63)
n
so that
c n+s+1 u −n−1
n=0
q (s)
r+1 e (s)
q (s 2 e (s)
c s+1
u−q (s)
1
r
1
=
−e (s 1
−e (s 2 −· · ·−
r+1 −· · ·
u−q (s)
2
u−q (s)
r+1
−e (s)
(s+1 1 ] 2
u−α (s+1)
1
(s+1 r ] 2
u−α (s+1)
c s+1
u−α (s+1)
0
=
.
(4.64)
−· · ·−
−· · ·
r
Here, the second line of (4.64) is identical to the second line of (4.53), provided
that s + 1 is used instead of s, as it ought to be. As such, (4.64) is, in fact,
the proof that (4.56) is correct. Returning now to (4.57) for the odd part of
G CCF(s)
(u)
n
(u) = G CF(s)
G LCF(s)
o,n
(u) = G CCF(s)
o,n
2n+1 (u).
(4.65)
Hence, we see from (4.55) and (4.65) that the even and odd parts of the delayed
Lanczos approximants G LCF(s)
(u) and G LCF(s)
o,n (u) of order n (n = 1, 2, 3,...)
coincide with the delayed continued fractions G CF(s)
2n
e,n
(u) and G CF(s)
2n+1 (u) of or
ders 2n and 2n + 1, respectively.
Search WWH ::




Custom Search