Page Menu
Home
c4science
Search
Configure Global Search
Log In
Files
F104878058
2_interpolation.tex
No One
Temporary
Actions
Download File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Award Token
Subscribers
None
File Metadata
Details
File Info
Storage
Attached
Created
Thu, Mar 13, 02:10
Size
22 KB
Mime Type
text/x-tex
Expires
Sat, Mar 15, 02:10 (2 d)
Engine
blob
Format
Raw Data
Handle
24871062
Attached To
R2653 epfl
2_interpolation.tex
View Options
\documentclass
[aspectratio=169]
{
beamer
}
\def\stylepath
{
../styles
}
\usepackage
{
\stylepath
/com303
}
\begin
{
document
}
\begin
{
frame
}
\frametitle
{
Overview:
}
\begin
{
itemize
}
\item
Polynomial interpolation
\item
Local interpolation
\item
Sinc interpolation
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Interpolation
}
\centering
$
x
[
n
]
\longrightarrow
x
(
t
)
$
\vspace
{
2em
}
``fill the gaps'' between samples
\end
{
frame
}
%% finite-length signal, 5 taps
\def\ta
{
1
}
\def\tb
{
2
}
\def\tc
{
0.7
}
\def\td
{
2.4
}
\def\te
{
-1.5
}
%% the tap plotting string
\def\taps
{
-2
\ta
-1
\tb
0
\tc
1
\td
2
\te
}
\def\plotTaps
{
\dspTaps
[linecolor=blue!70]
{
\taps
}}
%% Lagrange polynomials
\def\lpa
{
%
dup 1 add -1 div exch
%
dup -2 div exch
%
dup -1 add -3 div exch
%
-2 add -4 div
%
mul mul mul
}
\def\lpb
{
%
dup 2 add exch
%
dup -1 div exch
%
dup -1 add -2 div exch
%
-2 add -3 div
%
mul mul mul
}
\def\lpc
{
%
dup 2 add 2 div exch
%
dup 1 add 1 div exch
%
dup -1 add -1 div exch
%
-2 add -2 div
%
mul mul mul
}
\def\lpd
{
%
dup 2 add 3 div exch
%
dup 1 add 2 div exch
%
dup 1 div exch
%
-2 add -1 div
%
mul mul mul
}
\def\lpe
{
%
dup 2 add 4 div exch
%
dup 1 add 3 div exch
%
dup 2 div exch
%
-1 add 1 div
%
mul mul mul
}
\def\lagInterp
{
%
x
\lpa
\ta
mul
%
x
\lpb
\tb
mul
%
x
\lpc
\tc
mul
%
x
\lpd
\td
mul
%
x
\lpe
\te
mul
%
add add add add
}
%% cubic interpolator
\def\cubFunA
{
abs dup 2 exp -2.25 mul exch 3 exp 1.25 mul 1 add add
}
\def\cubFunB
{
abs dup dup 8 mul exch 2 exp -5 mul add exch 3 exp -4 add add -0.75 mul
}
\def\cubFun
#1
{
#1 sub
dup dup dup dup
%
-2 lt
{
pop pop pop pop 0
}
{
2 gt
{
pop pop pop 0
}
{
-1 lt
{
pop
\cubFunB
}
{
1 gt
{
\cubFunB
}
{
\cubFunA
}
ifelse
}
%
ifelse
}
%
ifelse
}
%
ifelse
}
\def\triInterp
{
%
x
\cubFun
{
-2
}
\ta
mul
%
x
\cubFun
{
-1
}
\tb
mul
%
x
\cubFun
{
0
}
\tc
mul
%
x
\cubFun
{
1
}
\td
mul
%
x
\cubFun
{
2
}
\te
mul
%
add add add add
}
\def\t
#1
{
\csname
t#1
\endcsname
}
\begin
{
frame
}
\frametitle
{
Example
}
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,yticks=1,xticks=none]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Example
}
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,yticks=1,xticks=none]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\pscurve
(-2,
\ta
)(0,0)(2,
\te
)
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Example
}
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,yticks=1,xticks=none]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\pscurve
(-2,
\ta
)(-1.75, 2)(-1.5, -1.5)(-1.25, 1.5)
%
(-1,
\tb
)(-0.75, 0.5)(-0.5, .6)(-0.25, .5)
%
(0,
\tc
)(0.25, -1)(0.5, 1)(0.75, -1)
%
(1,
\td
)(1.25, 1.5)(1.5, 2.5)(1.75, 1.8)(2,
\te
)
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Example
}
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,yticks=1,xticks=none]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\only
<2->
{
%
\dspFunc
[xmin=-2,xmax=2]
{
\triInterp
}
\dspFunc
[linestyle=dashed,xmax=-2]
{
\triInterp
}
\dspFunc
[linestyle=dashed,xmin=2]
{
\triInterp
}}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Interpolation requirements
}
\begin
{
itemize
}
\item
decide on
$
T_s
$
\item
make sure
$
x
(
nT_s
)
=
x
[
n
]
$
\item
make sure
$
x
(
t
)
$
is
{
\em
smooth
}
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Why smoothness?
}
\begin
{
itemize
}
\item
jumps (1st order discontinuities) would require the signal to move ``faster than light''...
\item
2nd order discontinuities would require infinite acceleration
\item
...
\vspace
{
1em
}
\item
the interpolation should be infinitely differentiable
\vspace
{
1em
}
\item
``natural'' solution:
{
\color
{
red
}
polynomial interpolation
}
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Polynomial interpolation
}
\begin
{
itemize
}
\item
$
N
+
1
$
points
$
\rightarrow
$
polynomial of degree
$
N
$
\item
$
x_p
(
t
)
=
a_
0
+
a_
1
t
+
a_
2
t^
2
+
\ldots
+
a_N t^{N}
$
\item
straightforward approach:
\vspace
{
1em
}
\[
\left\{
\begin
{aligned}
x_p
(
0
)
&
=
x
[
0
]
\\
x_p
(
T_s
)
&
=
x
[
1
]
\\
x_p
(
2
T_s
)
&
=
x
[
2
]
\\
&
\ldots
\\
x_p
(
NT_s
)
&
=
x
[
N
]
\\
\end
{aligned}
\right
.
\]
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Polynomial interpolation
}
Without loss of generality:
\begin
{
itemize
}
\item
even polynomial degree
$
2
N
$
\item
fit polynomial over
$
2
N
+
1
$
data points symmetrically distributed around zero
\vspace
{
1em
}
\[
\left\{
\begin
{aligned}
x_p
(-
NT_s
)
&
=
x
[-
N
]
\\
x_p
((-
N
+
1
)
T_s
)
&
=
x
[-
N
+
1
]
\\
&
\ldots
\\
x_p
(
0
)
&
=
x
[
0
]
\\
&
\ldots
\\
x_p
((
N
-
1
)
T_s
)
&
=
x
[
N
-
1
]
\\
x_p
(
NT_s
)
&
=
x
[
N
]
\\
\end
{aligned}
\right
.
\]
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Polynomial interpolation
}
Without loss of generality:
\begin
{
itemize
}
\item
set
$
T_s
=
1
$
\item
find the
$
2
N
+
1
$
coefficients
$
a_
0
,
\ldots
, a_{
2
N}
$
by solving:
\[
\sum
_{k
=
0
}^{
2
N} a_k n^k
=
x
[
n
]
\qquad
n
=
-
N,
-
N
+
1
,
\ldots
,
0
,
\ldots
, N
-
1
, N
\]
% \vspace{1em}
% \[
% \left\{
% \begin{aligned}
% x_p(-N) &= x[-N] \\
% x_p(-N+1) &= x[-N+1] \\
% & \ldots \\
% x_p(0) &= x[0] \\
% & \ldots \\
% x_p(N-1) &= x[N-1] \\
% x_p(N) &= x[N] \\
% \end{aligned}
% \right.
% \]
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Lagrange interpolation
}
Let's use the power of vector spaces:
\begin
{
itemize
}
\item
$
P_N
$
: space of degree-
$
2
N
$
polynomials over
$
I_N
=
[-
N, N
]
$
\item
interpolation will be a linear combination of basis vectors for
$
P_N
$
\item
what is a good basis for
\textit
{
interpolation?
}
% \item a basis for $P_N$ is the family of $2N+1$ Lagrange polynomials
% \[
% L^{(N)}_n(t) = \mathop{\prod_{k = -N}}_{k\neq n}^{N}\frac{t - k}{n - k} \qquad\qquad n = -N, \ldots, N
% \]
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Aside:
$
N
$
-degree polynomial bases on the interval
}
Given a set of
$
N
+
1
$
points, the degree-
$
N
$
polynomial interpolation is unique; however, when numerical issues come into play:
\vspace
{
1em
}
\begin
{
itemize
}
\item
naive basis:
$
1
, t, t^
2
,
\ldots
, t^{N}
$
; most general but very ill-conditioned
\item
Lagrange polynomials: equal degree, equally-spaced points but potential oscillation at edges
\item
Legendre polynomials: orthonormal, increasing degree, points denser at edges, better conditioning
\item
Chebyshev polynomials: orthonormal, increasing degree, optimal point placement (minimax criterion)
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Lagrange interpolation
}
\begin
{
itemize
}
\item
$
P_N
$
: space of degree-
$
2
N
$
polynomials over
$
I_N
=
[-
N, N
]
$
\item
a basis for
$
P_N
$
is the family of
$
2
N
+
1
$
Lagrange polynomials
\[
L^{
(
N
)
}_n
(
t
)
=
\mathop
{
\prod
_{k
=
-
N}}_{k
\neq
n}^{N}
\frac
{t
-
k}{n
-
k}
\qquad\qquad
n
=
-
N,
\ldots
, N
\]
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Lagrange polynomials for
$
I_
2
$
}
\begin
{
align*
}
L
^{
(2)
}_{
-2
}
(t)
&
=
\left
(
\frac
{
t + 1
}{
-2 + 1
}
\right
)
\,
\left
(
\frac
{
t
}{
-2
}
\right
)
\,
\left
(
\frac
{
t - 1
}{
-2 - 1
}
\right
)
\,
\left
(
\frac
{
t -2
}{
-2 - 2
}
\right
) =
\frac
{
(t+1)t(t-1)(t-2)
}{
24
}
\\
L
^{
(2)
}_{
-1
}
(t)
&
=
\left
(
\frac
{
t + 2
}{
-1 + 2
}
\right
)
\,
\left
(
\frac
{
t
}{
-1
}
\right
)
\,
\left
(
\frac
{
t - 1
}{
-1 - 1
}
\right
)
\,
\left
(
\frac
{
t -2
}{
-1 - 2
}
\right
) =
\frac
{
(t+2)t(t-1)(t-2)
}{
-6
}
\\
L
^{
(2)
}_{
0
}
(t)
&
=
\left
(
\frac
{
t + 2
}{
2
}
\right
)
\,
\left
(
\frac
{
t+1
}{
1
}
\right
)
\,
\left
(
\frac
{
t - 1
}{
- 1
}
\right
)
\,
\left
(
\frac
{
t -2
}{
-2
}
\right
) =
\frac
{
(t+2)(t+1)(t-1)(t-2)
}{
-4
}
\\
L
^{
(2)
}_{
1
}
(t)
&
= L
^{
(2)
}_{
-1
}
(-t)
\\
L
^{
(2)
}_{
2
}
(t)
&
= L
^{
(2)
}_{
-2
}
(-t)
\\
\end
{
align*
}
\end
{
frame
}
\def\LagPoly
#1#2#3#4
{
\only
<#1->
{
%
\dspFunc
[linecolor=#2]
{
x
\csname
lp#3
\endcsname
}
%
\only
<#1|handout:0>
{
\dspText
(0,1.3)
{
\color
{
#2
}
$
L_{#
4
}^{
(
2
)
}
(
t
)
$
}}}}
\begin
{
frame
}
\frametitle
{
Lagrange interpolation polynomials
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,yticks=0.5,height=5cm]
{
-2.5,2.5
}{
-0.8,1.6
}
\moocStyle
\begin
{
dspClip
}
%
\LagPoly
{
1
}{
green
}{
a
}{
-2
}
%
\LagPoly
{
2
}{
blue
}{
b
}{
-1
}
%
\LagPoly
{
3
}{
orange
}{
c
}{
0
}
%
\LagPoly
{
4
}{
black
}{
d
}{
1
}
%
\LagPoly
{
5
}{
cyan
}{
e
}{
2
}
%
\end
{
dspClip
}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Interpolation property of Lagrange polynomials
}
\[
m
\in
\mathbb
{N}
\quad\Rightarrow\quad
L^{
(
N
)
}_n
(
m
)
=
\left\{
\begin
{array}{ll}
1
&
\mbox
{if $n
=
m$}
\\
0
&
\mbox
{if $n
\neq
m$}
\end
{array}
\right
.
\qquad\qquad
-
N
\leq
n,m
\leq
N
\]
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Lagrange interpolator for the data set
}
\[
x_p
(
t
)
=
\sum
_{n
=
-
N}^{N} x
[
n
]
L^{
(
N
)
}_n
(
t
)
\]
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Lagrange interpolation
}
The Lagrange interpolator
{
\em
is
}
the unique polynomial interpolation for the data set:
\begin
{
itemize
}
\item
a polynomial of degree
$
2
N
$
through
$
2
N
+
1
$
points is uniquely defined
\item
the Lagrangian interpolator satisfies
\[
x_p
(
n
)
=
x
[
n
]
\qquad
\mbox
{for $
-
N
\le
n
\le
N$}
\]
since
\[
L^{
(
N
)
}_n
(
m
)
=
\left\{
\begin
{array}{ll}
1
&
\mbox
{if $n
=
m$}
\\
0
&
\mbox
{if $n
\neq
m$}
\end
{array}
\right
.
\qquad\qquad
-
N
\leq
n,m
\leq
N
\]
\end
{
itemize
}
\end
{
frame
}
% slide showing current interpolator in red and past interpolators in blue
\def\interpolant
#1#2#3#4
{
%
\FPupn\n
{
#1 1 + 0 trunc
}
%
\only
<#1|handout:#1>
{
%
\dspTaps
{
#3
\csname
t#4
\endcsname
}
%
\dspFunc
[linewidth=0.5pt]
{
x
\csname
lp#4
\endcsname
\csname
t#4
\endcsname
mul
}}
%
\only
<
\n
-#2|handout:
\n
-#2>
{
\dspFunc
[linewidth=0.5pt,linecolor=blue!50]
{
x
\csname
lp#4
\endcsname
\csname
t#4
\endcsname
mul
}}}
%
\def\polyName
#1#2
{
\only
<#1|handout:#1>
{
$
x
[
#
2
]
L^{
(
2
)
}_{#
2
}
(
t
)
$
}}
\begin
{
frame
}
\frametitle
{
Lagrange interpolation
}
\begin
{
center
}
\begin
{
figure
}
$
\vphantom
{x
[
0
]
L^{
(
N
)
}_{
0
}
(
t
)
}
$
%
\polyName
{
2
}{
-2
}
\polyName
{
3
}{
-1
}
\polyName
{
4
}{
0
}
\polyName
{
5
}{
1
}
\polyName
{
6
}{
2
}
\\
\begin
{
dspPlot
}
[sidegap=0]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\begin
{
dspClip
}
\interpolant
{
2
}{
7
}{
-2
}{
a
}
%
\interpolant
{
3
}{
7
}{
-1
}{
b
}
%
\interpolant
{
4
}{
7
}{
0
}{
c
}
%
\interpolant
{
5
}{
7
}{
1
}{
d
}
%
\interpolant
{
6
}{
7
}{
2
}{
e
}
%
\only
<7-8|handout:7-8>
{
\dspFunc
[linewidth=2pt,xmin=-2,xmax=2]
{
\lagInterp
}}
%
\end
{
dspClip
}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
center
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Polynomial interpolation
}
key property:
\begin
{
itemize
}
\item
maximally smooth (infinitely many continuous derivatives)
\end
{
itemize
}
\vspace
{
2em
}
drawback:
\begin
{
itemize
}
\item
interpolation ``machine'' depend on
$
N
$
: we need to use a different set of polynomials if the length of the dataset changes
\end
{
itemize
}
\centering
\vspace
{
2em
}
can we find a ``universal'' interpolation machine?
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Relaxing the interpolation requirements
}
\setbeamercovered
{
transparent
}
\begin
{
itemize
}
[<+->]
\item
<1-2> decide on
$
T_s
$
\item
<1-2> make sure
$
x
(
nT_s
)
=
x
[
n
]
$
\item
<1|handout:0> make sure
$
x
(
t
)
$
is
{
\em
smooth
}
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Zero-order interpolation
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\only
<2->
{
%
\psline
(-2,
\ta
)(-1.5,
\ta
)(-1.5,
\tb
)(-.5,
\tb
)(-.5,
\tc
)(0.5,
\tc
)(0.5,
\td
)(1.5,
\td
)(1.5,
\te
)(2,
\te
)
}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Zero-order interpolation
}
\begin
{
itemize
}
\item
$
x_
0
(
t
)
=
x
[
\,
\lfloor
t
+
0
.
5
\rfloor
\,
]
,
\qquad
-
N
\leq
t
\leq
N
$
\vspace
{
1em
}
\item
$
\displaystyle
x_
0
(
t
)
=
\sum
_{n
=
-
N}^{N}x
[
n
]
\rect
(
t
-
n
)
$
\item
interpolation kernel:
$
i_
0
(
t
)
=
\rect
(
t
)
$
\item
$
i_
0
(
t
)
$
: ``zero-order hold''
\item
interpolator's support is 1
\item
interpolation is not even continuous
\end
{
itemize
}
\end
{
frame
}
\def\interpolant
#1#2#3#4
{
%
\FPupn\n
{
#1 1 + 0 trunc
}
%
\only
<#1>
{
%
\dspTaps
{
#3
\csname
t#4
\endcsname
}
%
\dspFunc
[linewidth=0.5pt]
{
x
\dspRect
{
#3
}{
1
}
\csname
t#4
\endcsname
mul
}}
%
\only
<
\n
-#2>
{
\dspFunc
[linewidth=0.5pt,linecolor=blue!50]
{
x
\dspRect
{
#3
}{
1
}
\csname
t#4
\endcsname
mul
}}}
%
\begin
{
frame
}
\frametitle
{
Zero-order interpolation
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,height=5cm,width=10cm]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\interpolant
{
2
}{
7
}{
-2
}{
a
}
%
\interpolant
{
3
}{
7
}{
-1
}{
b
}
%
\interpolant
{
4
}{
7
}{
0
}{
c
}
%
\interpolant
{
5
}{
7
}{
1
}{
d
}
%
\interpolant
{
6
}{
7
}{
2
}{
e
}
%
\only
<7->
{
%
\psline
(-2,
\ta
)(-1.5,
\ta
)(-1.5,
\tb
)(-.5,
\tb
)(-.5,
\tc
)(0.5,
\tc
)(0.5,
\td
)(1.5,
\td
)(1.5,
\te
)(2,
\te
)
}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
First-order interpolation
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,height=5cm,width=10cm]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\psline
(-2,
\ta
)(-1,
\tb
)(0,
\tc
)(1,
\td
)(2,
\te
)
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
First-order interpolation
}
\begin
{
itemize
}
\item
``connect the dots'' strategy
\vspace
{
1em
}
\item
$
\displaystyle
x_
1
(
t
)
=
\sum
_{n
=
-
N}^{N}x
[
n
]
\,
i_
1
(
t
-
n
)
$
\item
interpolation kernel:
\[
i_
1
(
t
)
=
\begin
{cases}
1
-
|t| & |t|
\le
1
\\
0
&
\mbox
{otherwise}
\end
{cases}
\]
\item
interpolator's support is 2
\item
interpolation is continuous but derivative is not
\end
{
itemize
}
\end
{
frame
}
\def\interpolant
#1#2#3#4
{
%
\FPupn\n
{
#1 1 + 0 trunc
}
%
\only
<#1|handout:#1>
{
%
\dspTaps
{
#3
\csname
t#4
\endcsname
}
%
\dspFunc
[linewidth=0.5pt]
{
x
\dspTri
{
#3
}{
1
}
\csname
t#4
\endcsname
mul
}}
%
\only
<
\n
-#2|handout:
\n
-#2>
{
\dspFunc
[linewidth=0.5pt,linecolor=blue!50]
{
x
\dspTri
{
#3
}{
1
}
\csname
t#4
\endcsname
mul
}}}
%
\begin
{
frame
}
\frametitle
{
First-order interpolation
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,height=5cm,width=10cm]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\interpolant
{
2
}{
7
}{
-2
}{
a
}
%
\interpolant
{
3
}{
7
}{
-1
}{
b
}
%
\interpolant
{
4
}{
7
}{
0
}{
c
}
%
\interpolant
{
5
}{
7
}{
1
}{
d
}
%
\interpolant
{
6
}{
7
}{
2
}{
e
}
%
\only
<7|handout:7>
{
\psline
(-2,
\ta
)(-1,
\tb
)(0,
\tc
)(1,
\td
)(2,
\te
)
}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Third-order interpolation
}
\begin
{
itemize
}
\item
$
\displaystyle
x_
3
(
t
)
=
\sum
_{n
=
-
N}^{N}x
[
n
]
\,
i_
3
(
t
-
n
)
$
\item
interpolation kernel obtained by splicing two cubic polynomials
\item
interpolator's support is 4
\item
interpolation is continuous up to second derivative
\end
{
itemize
}
\end
{
frame
}
\def\interpolant
#1#2#3#4
{
%
\FPupn\n
{
#1 1 + 0 trunc
}
%
\only
<#1|handout:#1>
{
%
\dspTaps
{
#3
\csname
t#4
\endcsname
}
%
\dspFunc
[linewidth=0.5pt]
{
x
\cubFun
{
#3
}
\csname
t#4
\endcsname
mul
}}
%
\only
<
\n
-#2|handout:
\n
-#2>
{
\dspFunc
[linewidth=0.5pt,linecolor=blue!50]
{
x
\cubFun
{
#3
}
\csname
t#4
\endcsname
mul
}}}
%
\begin
{
frame
}
\frametitle
{
Third-order interpolation
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,height=5cm,width=10cm]
{
-2.5,2.5
}{
-2,3
}
\moocStyle
\plotTaps
\interpolant
{
2
}{
7
}{
-2
}{
a
}
%
\interpolant
{
3
}{
7
}{
-1
}{
b
}
%
\interpolant
{
4
}{
7
}{
0
}{
c
}
%
\interpolant
{
5
}{
7
}{
1
}{
d
}
%
\interpolant
{
6
}{
7
}{
2
}{
e
}
%
\only
<7|handout:7>
{
\dspFunc
[xmin=-2,xmax=2]
{
\triInterp
}}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Local interpolation schemes
}
\[
\displaystyle
x
(
t
)
=
\sum
_{n
=
-
N}^{N}x
[
n
]
\,
i_c
(
t
-
n
)
\]
\vspace
{
1em
}
Kernel must satisfy the interpolation property:
\begin
{
itemize
}
\item
$
i_c
(
0
)
=
1
$
\item
$
i_c
(
m
)
=
0
$
for
$
m
$
a nonzero integer.
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Local interpolators
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0]
{
-2.8,2.8
}{
-.5,1.3
}
\moocStyle
\only
<1|handout:1>
{
\dspFunc
{
x
\dspRect
{
0
}{
1
}}
\dspText
(0,1.5)
{
$
i_
0
(
t
)
$
}}
\only
<2-|handout:2->
{
\dspFunc
[linecolor=lightgray]
{
x
\dspRect
{
0
}{
1
}}
\dspText
(0,1.5)
{
$
i_
0
(
t
)
$
}}
\only
<2|handout:2>
{
\dspFunc
{
x
\dspTri
{
0
}{
1
}}
\dspText
(0,1.5)
{
$
i_
1
(
t
)
$
}}
\only
<3-|handout:3->
{
\dspFunc
[linecolor=lightgray]
{
x
\dspTri
{
0
}{
1
}}
\dspText
(0,1.5)
{
$
i_
1
(
t
)
$
}}
\only
<3|handout:3>
{
\dspFunc
{
x
\cubFun
{
0
}}
\dspText
(0,1.5)
{
$
i_
3
(
t
)
$
}}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
%\begin{frame}
% \frametitle{Comparison}
% \center
% \begin{figure}
% \begin{dspPlot}[sidegap=0]{-2.5,2.5}{-2,3}
% \moocStyle
% \plotTaps
% \dspFunc[xmin=-2,xmax=2]{\lagInterp}
% \only<2->{\psline[linecolor=cyan](-2, 1)(-1.5, 1)(-1.5, 2)(-.5,2)(-.5,0.7)%
% (0.5,0.7)(0.5,2.4)(1.5,2.4)(1.5,-1.5)(2,-1.5)}
% \only<3->{\psline[linecolor=orange](-2, 1)(-1, 2)(0,0.7)(1,2.4)(2,-1.5)}
% \only<4->{\dspFunc[linecolor=green,xmin=-2,xmax=2]{\triInterp}}
% \end{dspPlot}
% \end{figure}
%\end{frame}
\begin
{
frame
}
\frametitle
{
Local interpolation
}
key property:
\begin
{
itemize
}
\item
same interpolating function independently of
$
N
$
\end
{
itemize
}
\vspace
{
2em
}
drawback:
\begin
{
itemize
}
\item
lack of smoothness
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Polynomial interpolation
}
key property:
\begin
{
itemize
}
\item
maximally smooth (infinitely many continuous derivatives)
\end
{
itemize
}
\vspace
{
2em
}
drawback:
\begin
{
itemize
}
\item
interpolation kernels depend on
$
N
$
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
A remarkable result:
}
{
\Large
\[
\lim
_{N
\rightarrow\infty
} L^{
(
N
)
}_n
(
t
)
=
\sinc\left
(
t
-
n
\right
)
\]
}
\vspace
{
2em
}
\begin
{
itemize
}
\item
in other words:
$
\lim
_{N
\rightarrow\infty
} L^{
(
N
)
}_n
(
t
)
=
L^{
(
\infty
)
}_
0
(
t
-
n
)
$
\item
in the limit, local and global interpolation are the same!
\end
{
itemize
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Sinc interpolation formula
}
{
\Large
\[
x
(
t
)
=
\sum
_{n
=
-
\infty
}^{
\infty
}x
[
n
]
\,\sinc\left
(
t
-
n
\right
)
\]
}
\end
{
frame
}
\def\smooth
{
10 div 360 mul dup sin exch dup 2 mul sin exch dup mul 360 div sin add add 2.2 div 0.2 add
}
\def\interpolant
#1
{
\dspSinc
{
#1
}{
1
}
#1
\smooth
mul
}
\def\interpolant
#1#2#3
{
%
\FPupn\n
{
#1 1 + 0 trunc
}
%
\only
<#1>
{
%
\listplot
[plotstyle=LineToXAxis, linestyle=solid,
%
showpoints=true, dotstyle=*, linewidth=
\dspStemWidth
,
%
dotsize=
\dspDotSize
]
{
#3 #3
\smooth
#3 #3
\smooth
}
\dspFunc
[linewidth=0.5pt]
{
x
\dspSinc
{
#3
}{
1
}
#3
\smooth
mul
}}
%
\only
<
\n
-#2>
{
\dspFunc
[linewidth=0.5pt,linecolor=blue!50]
{
x
\dspSinc
{
#3
}{
1
}
#3
\smooth
mul
}}}
%
\begin
{
frame
}
\frametitle
{
Sinc interpolation
}
\begin
{
center
}
\begin
{
figure
}
\begin
{
dspPlot
}
[yticks=1,xticks=100,sidegap=0]
{
-4.5,7.5
}{
-0.8,1.2
}
\moocStyle
\SpecialCoor
\dspSignal
[linecolor=blue!70]
{
x
\smooth
}
\interpolant
{
2
}{
6
}{
0
}
\interpolant
{
3
}{
6
}{
1
}
\interpolant
{
4
}{
6
}{
2
}
\interpolant
{
5
}{
6
}{
3
}
\interpolant
{
6
}{
6
}{
4
}
\only
<7-8>
{
\multido
{
\n
=-4+1
}{
12
}{
%
\dspFunc
[linewidth=0.5pt,linecolor=blue!50]
{
x
\dspSinc
{
\n
}{
1
}
\n
\smooth
mul
}}}
\only
<8-9>
{
\dspFunc
{
x
\smooth
}}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
center
}
\end
{
frame
}
% L_0^N(t) = \prod_{k=1}^{N} [t^2/k^2 - 1]
\def\sincApp
#1
{
%
1
1 1 #1
{
%
dup mul
x dup mul
exch div
1 exch sub
mul
}
for
}
\def\plotApp
#1#2
{
%
\only
<#1|handout:#1>
{
\dspFunc
{
\sincApp
{
#2
}}
\dspLegend
(-14,1.2)
{
green
{
$
\sinc
(
t
)
$
}
darkred
{
$
L_
0
^{
(
#
2
)
}
(
t
)
$
}}}}
\begin
{
frame
}
\frametitle
{
Convergence: graphical ``proof''
}
\begin
{
align*
}
L
^{
(N)
}_
n(t)
&
=
\mathop
{
\prod
_{
k = -N
}}_{
k
\neq
n
}^{
N
}
\frac
{
t - k
}{
n - k
}
\\
\\
L
_
0
^{
N
}
(t)
&
=
\mathop
{
\prod
_{
k = -N
}}_{
k
\neq
0
}^{
N
}
\frac
{
t - k
}{
-k
}
=
\mathop
{
\prod
_{
k = -N
}}^{
-1
}
\frac
{
t - k
}{
-k
}
\mathop
{
\prod
_{
k = 1
}}^{
N
}
\frac
{
t - k
}{
-k
}
\\
&
=
\mathop
{
\prod
_{
k = 1
}}^{
N
}
\frac
{
t + k
}{
k
}
\mathop
{
\prod
_{
k = 1
}}^{
N
}
\frac
{
t - k
}{
-k
}
\\
&
=
\mathop
{
\prod
_{
k = 1
}}^{
N
}
\left
(1 -
\frac
{
t
^
2
}{
k
^
2
}
\right
)
\end
{
align*
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Convergence: graphical ``proof''
}
\center
\begin
{
figure
}
\begin
{
dspPlot
}
[sidegap=0,xout=true]
{
-15,15
}{
-.5,1.3
}
\moocStyle
\dspFunc
[linecolor=green]
{
x
\dspSinc
{
0
}{
1
}}
\plotApp
{
2
}{
100
}
\plotApp
{
3
}{
200
}
\plotApp
{
4
}{
300
}
\end
{
dspPlot
}
\end
{
figure
}
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Convergence: mathematical intuition
}
\begin
{
itemize
}
[<+->]
\item
$
\sinc
(
t
-
n
)
$
and
$
L^{
(
\infty
)
}_n
(
t
)
$
share an infinite number of zeros:
\vspace
{
1em
}
\begin
{
align*
}
\sinc
(m-n)
&
=
\delta
[m-n]
\qquad
m,n
\in
\mathbb
{
Z
}
\\
L
^{
(N)
}_
n(m)
&
=
\delta
[m-n]
\qquad
m,n
\in
\mathbb
{
Z
}
,
\quad
-N
\le
n,m
\le
N
\end
{
align*
}
\end
{
itemize
}
\end
{
frame
}
%\begin{frame}
% \frametitle{Convergence: Euler's ``proof'' (1748)}
% consider the $N$ roots of unity for $N$ odd:
% \begin{itemize}
% \item $z = 1$
% \item $z = e^{\pm j\frac{2\pi}{N} k}$ for $k = 1, \ldots, (N-1)/2$
% \item (set $\omega_N = 2\pi/N$)
% \end{itemize}
%\end{frame}
%
%\begin{frame}
% \frametitle{Convergence: Euler's ``proof'' (1748)}
% group the complex conjugate roots pairwise in the factors the polynomial $z^N-1$:
% \[
% z^N-1 = (z-1)\prod_{k=1}^{(N-1)/2} (z^2 - 2z\cos(\omega_N k) + 1)
% \]
% \vspace{2em}
% in general:
% \[
% z^N-a^N = (z-a)\prod_{k=1}^{(N-1)/2} (z^2 - 2az\cos(\omega_N k) + a^2).
% \]
%\end{frame}
%
%\begin{frame}
% \frametitle{Convergence: Euler's ``proof'' (1748)}
%first trick: replace $z$ and $a$ by $z = (1+x/N)$ and $a = (1-x/N)$:
%\begin{eqnarray*}
% \lefteqn{\left(1+\frac{x}{N}\right)^N - \left(1-\frac{x}{N}\right)^N =} \nonumber \\
%&=& \frac{4x}{N}\prod_{k=1}^{(N-1)/2} \left(1-\cos(\omega_N k) + \frac{x^2}{N^2}(1+\cos(\omega_N k))\right) \\
%& = & \frac{4x}{N}\prod_{k=1}^{(N-1)/2} (1-\cos(\omega_N k)) \left(1 + \frac{x^2}{N^2}\cdot\frac{1+\cos(\omega_N k)}{1-\cos(\omega_N k)}\right) \\
%& = & Ax\,\prod_{k=1}^{(N-1)/2}\left(1 + \frac{x^2\,(1+\cos(\omega_N k))}{N^2\,(1-\cos(\omega_N k))}\right)
%\end{eqnarray*}
%\end{frame}
\begin
{
frame
}
\frametitle
{
Convergence: Euler's ``proof'' (1748)
}
\centering
very cute (if non-rigorous) proof -- see handout or book for details
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Convergence: rigorous proof
}
\centering
uses the properties of Fourier series expansions -- see handout or book for details
\end
{
frame
}
\begin
{
frame
}
\frametitle
{
Sinc interpolation formula for any
$
T_s
$
}
{
\Large
\[
x
(
t
)
=
\sum
_{n
=
-
\infty
}^{
\infty
}x
[
n
]
\,\mbox
{sinc}
\left
(
\frac
{t
-
nT_s}{T_s}
\right
)
\]
}
\end
{
frame
}
\end
{
document
}
Event Timeline
Log In to Comment