Sampling and Reconstruction of Signals

Lesson 12, Part I, INF44480

Andreas Austeng
Department of Informatics, University of Oslo
April 2011
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 1 / 82
Outline
Outline
1 Fourier transforms and the sampling theorem
The sampling theorem
Fourier transforms
Sampling and reconstruction of signals
2 Sampling of bandpass signals
Representation of bandpass signals
Sampling of bandpass signals
3 Multirate digital signal processing
Sampling rate conversion
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 2 / 82
Fourier transforms and the sampling theorem The sampling theorem
The sampling theorem
Alternative I:
If the highest frequency contained in an analog signal x
a
(t ) is
F
max
= B Hertz and the signal is sampled at a rate
F
s
> 2F
max
2B samples per second, then x
a
(t ) can be exactly
recovered from its sample values using the interpolation function
g(t ) =
sin 2Bt
2B
The sampling rate F
N
= 2B = 2F
max
is called the Nyquist rate.
Alternative II:
A bandlimited continuous-time signal, with highest frequency
(bandwidth) B Hertz, can be uniquely recovered from its samples
provided that the sampling rate F
s
> 2B samples per second.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 4 / 82
Fourier transforms and the sampling theorem The sampling theorem
Relations among frequency variables
We have that: = T, f = F/F
s
(or = /T, F = f F
s
)
Continuous-time signals Discrete-time signals
= 2F = 2f
radians
sec
, Hz
radians
sample
,
cycle
sample

1
2
f
1
2
< < /T /T
< F < F
s
/2 F
s
/2
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 5 / 82
Fourier transforms and the sampling theorem Fourier transforms
Fourier transforms
Discrete-Time Fourier Transform, DTFT;
X() = F{x[n]} =

n=
x[n]e
n
x[n] =
1
2
_
2
X()e
n
d
Continuous-Time Fourier Transform, CTFT:
X(F) = F{x(t )} =
_

x(t )e
2Ft
dt .
x(t ) = F
1
{X(F)} =
_

X(F)e
2Ft
dF.
Convolution integral:
y(t ) = x(t ) h(t ) y(t ) =
_

x()h(t )d.
H(t ): The Heaviside function (cont. time step-func.):
F{H(t )} =
_

H(t )e
2Ft
dt =
1
2
_
(t )
1
t

.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 6 / 82
Fourier transforms and the sampling theorem Fourier transforms
Fourier transforms ...
Properties valid for DTFT are usually valid for CTFT;
F{f g} = F{f }F{g}.
F{f g} =
_

__

f ()g(t )d
_
e
2Ft
dt
=
_

[e
2F
f ()d][e
2F(t )
g(t )dt ]
=
__

e
2F
f ()d
_ __

e
2F( )
g( )d
_
= F{f }F{g}, (t ).
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 7 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Sampling of analog signals
Periodic or uniform sampling; described by
x[n] = x
a
(nT), < n < ,
where x[n] is the discrete-time signal obtained by taking samples
of analog signal x
a
(t ) every T second.
T; sampling period/interval, 1/T = F
s
; sampling rate/frequency;
t = nT =
n
Fs
.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 8 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Sampling of analog signals (2)
The spectrum of x
a
(t ) (an aperiodic, nite energy signal), i.e. the CTFT;
X
a
(F) =
_

x
a
(t )e
2Ft
dt . (1)
x
a
(t ) can be recovered by the inverse CTFT
x
a
(t ) =
_

X
a
(F)e
2Ft
dF. (2)
The spectrum of x[n] (sampled version of x
a
(t )), i.e. the DTFT;
X() =

x[n]e
n
or X(f ) =

x[n]e
2fn
(3)
x[n] can be recovered by the inverse DTFT
x[n] =
1
2
_

X()e
n
d =
_ 1
2

1
2
X(f )e
2fn
df . (4)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 9 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Sampling of analog signals (3)
To determine a relationship between the spectra of analog and
digital signals we use that t = nT =
n
Fs
and f =
F
Fs
:
x[n] x
a
(nT)
(2)
=
_

X
a
(F)e
2nF/Fs
dF
(4)
=
_ 1
2

1
2
X(f )e
2fn
df =
1
F
s
_ Fs
2

Fs
2
X(F)e
2nF/Fs
dF.
Due to the periodicity of e
2nFs/F
, we can write
_

dF as

k=
_
(k+1/2)Fs
(k1/2)Fs
dF...
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 10 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Sampling of analog signals (4)
We then have that
1
F
s
_ Fs
2

Fs
2
X(F)e
2nF/Fs
dF =
_

X
a
(F)e
2nF/Fs
dF
=

k=
_
(k+1/2)Fs
(k1/2)Fs
X
a
(F)e
2nF/Fs
dF
=

k=
_ Fs
2

Fs
2
X
a
(F kF
s
)e
2nF/Fs
dF
=
_ Fs
2

Fs
2
_

k=
X
a
(F kF
s
)
_
e
2nF/Fs
dF
which gives
X(F) = F
s

k=
X
a
(F kF
s
) or X(f ) = F
s

k=
X
a
((f k)F
s
) .
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 11 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Sampling of analog signals (5)
No aliasing if F
s
> 2B.
If so; x(t ) can be
reconstructed from
x[n].
If not; aliasing occur as
in (e).
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 12 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Reconstruction of analog signal
Given no aliasing, then
X
a
(F) =
_
1
Fs
X(F), |F| F
s
/2
0, |F| > F
s
/2
(5)
and analog signal can be reconstructed:
x
a
(t )
(2)
=
_

X
a
(F)e
2Ft
dF
(5)
=
1
F
s
_ Fs
2

Fs
2
X(F)e
2Ft
dF
(4)
=
1
F
s
_ Fs
2

Fs
2
_

n=
x[n]e
j 2Fn/Fs
_
e
j 2Ft
dF
=
1
F
s

n=
x[n]
_ Fs
2

Fs
2
e
j 2F(t n/Fs)
dF
=

n=
x
a
(nT)
sin[(/T)(t nT)]
(/T)(t nT)
.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 13 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Reconstruction of analog signal (2)
Ideal interpolation formula: x
a
(t ) =

n=
x
a
(nT)
sin[(/T)(t nT)]
(/T)(t nT)
.
Involves the function g(t ) =
sin(/T)t
(/T)t
appropriately shifted by nT,
n = 0, 1, 2, . . ., and multiplied or weighted by the
corresponding samples x
a
(nT) of the signal.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 14 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Folding ...
Time-freq. relationships ...
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 15 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Nonbandlimited signal (1)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 16 / 82
Fourier transforms and the sampling theorem Sampling and reconstruction of signals
Nonbandlimited signal (2)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 17 / 82
Sampling of bandpass signals
A bandlimited signal
%%
% Eksempel p? b?ndbegrenset si gnal :
%
% x ( t ) = a( t ) * si n ( 2 \ pi f _c t ) ;
%
% f _c : c ar r i erf r ekvens
% a( t ) : langsom var i er ende modul asj on
fs = 500; % Sampl i ngsf rekvens
Nsek = 2; % Ant al l sekunder
t = 0: 1/ fs : ( Nsek) ; % Ti d
f_c = 100; % Car r i er
an = 1 + 0. 5 * si n ( 2* pi *5*t) ; % Modul asj on
xn = an . * si n ( 2* pi *f_c*t) ; % Bandbegrenset si gnal
% Pl ot t av xn
f i gur e ( 1) ;
pl ot ( t, xn, ' o ' ) ; gr i d on
t i t l e ( ' Pl ot of bandl i mi t ed si gnal xn ' ) ; xl abel ( ' t i me [ sec ] ' ) ;
fname = ' BpSi gnTi d ' ;
cmd = [ ' p r i n t deps2c ' , fname, ' ; ' ] ;
eval ( cmd) ;
% cmd = [ ' epst opdf ' , fname , ' . eps ' ] ;
% uni x ( cmd) ;
% Pl ot t av spektrum t i l xn
f i gur e ( 2) ;
Hs = spectrum. periodogram( ' Hamming ' ) ;
psd( Hs, xn, ' Fs ' , fs, ' NFFT ' , 1024 , ' SpectrumType ' , ' t wosi ded ' , ' CenterDC ' ,
true) ;
axi s([fs/ 2 fs/ 2 100 0] )
fname = ' BpSi gnFrek ' ;
cmd = [ ' p r i n t deps2c ' , fname, ' ; ' ] ;
eval ( cmd) ;
%cmd = [ ' epst opdf ' , fname , ' . eps ' ] ;
%uni x ( cmd) ;
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
1.5
1
0.5
0
0.5
1
1.5
Plot of bandlimited signal xn
time [sec]
250 200 150 100 50 0 50 100 150 200 250
100
90
80
70
60
50
40
30
20
10
0
Frequency (Hz)
P
ow
er/frequency (dB
/H
z)
Periodogram Power Spectral Density Estimate
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 19 / 82
Sampling of bandpass signals
Example (Sampling of a complex bandlimited signal (1))
Given a complex bandpass analog
signal x
a
(t ) with nonzero Fourier
transform over the frequency range
[
1
,
2
].
What is the smallest sampling
frequency that can be used so that
x
a
(t ) can be recovered from its
samples x[n]?

Xa()
1 2
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 20 / 82
Sampling of bandpass signals
Example (Sampling of a complex bandlimited signal (2))
Obvious choice, the Nyquist rate: 2
2
.
But if x
a
(t ) is modulated, i.e.
y
a
(t ) = x
a
(t )e
(
2
+
1
)t /2
, then y
a
(t )
becomes a (complex) low-pass signal
with spectrum as seen below.
y
a
(t ) may be recovered from samples
taken at the sampling rate
2
0
=
2

1
.
Question to ask: Can x
a
(t ) be uniquely
recovered from its samples x
a
(nT
s
)
provided that T
s

1
?

Ya()
0 0
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 21 / 82
Sampling of bandpass signals
Example (Sampling of a complex bandlimited signal (3))
If x
a
(t ) is sampled with a sampling frequency
s
, the spectrum of
the sampled signal is
X
s
() =
1
T
s

k=
X
a
( k
s
).
In order to avoid interference between shifted spectra, we must
have that

s

1
or
s

2

1
.

X()

1
-3s
2
-3s
1
-2s
2
-2s
1
-s
2
-s
1

2
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 22 / 82
Sampling of bandpass signals
Example (Sampling of a complex bandlimited signal (4))
If
s

2

1
, then x
a
(t ) may be
uniquely reconstructed from x
s
(t )
using a bandpass lter with a
frequency response as shown to the
right.
If
s
=
2

1
, the reconstruction
lter is a complex bandpass lter with
impulse response
h
a
(t ) = T
s
sin(
s
t /2)
t
e
(
2
+
1
)t /2

Hr ()
Ts
1 2
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 23 / 82
Sampling of bandpass signals Representation of bandpass signals
Representation of bandpass signals (1)
Any real-valued bandpass signal x
a
(t ) can be represented by an
equivalent lowpass signal x
LP
(t ).
x
a
(t ): Real-valued bandpass signal with frequency concentrated
around F
c
.
If
+
is the signal containing the positive frequencies in x
a
(t ), then
x
a
(t ) can be represented by a low-frequency signal:
x
LP
(t ) = e
2Fct

+
(t ) (complex envelope of x
a
(t ))
= x
I
(t ) + x
Q
(t ) (in-phase & quadrature components)
= A(t )e
(t )
e
piFct
(envelope & phase)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 24 / 82
Sampling of bandpass signals Representation of bandpass signals
Representation of bandpass signals (2)
To develop a mathematical representation of x
a
(t ); start with the
signal
+
(t ) containing the positive frequencies in x
a
(t ).
Then:
+
(F) = 2V
a
(F)X
a
(F),
where V
a
(F) is the Heaviside function and X
a
(F) = F[x
a
(t )].
The analytic signal or the pre-envelope of x
a
(t ) given as:

+
(t ) =
_

+
(F)e
j 2Ft
dF = F
1
[2V
a
(F)] F
1
[X
a
(F)]
=
_
(t ) +
j
t
_
x
a
(t ) = x
a
(t ) +j
1
t
x
a
(t )
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 25 / 82
Sampling of bandpass signals Representation of bandpass signals
Representation of bandpass signals (3)

+
(t ) = x
a
(t ) +j x
a
(t ),
x
a
(t ) =
1
t
x
a
(t ) =
1

xa()
t
d.
x
a
(t ) viewed as output of the lter with impulse response
h
Q
(t ) =
1
t
when excited by x
a
(t ).
Such lter called Hilbert transformer and
F[h
Q
(t )] = H
Q
(F) =
_
_
_
j F > 0
0 F == 0
j F < 0.
|H
Q
(F)| = 1 and (F) = /2, F > 0 and (F) = /2, F < 0,
i.e. 90

phase shift.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 26 / 82
Sampling of bandpass signals Representation of bandpass signals
Representation of bandpass signals (4)

+
(t ) a bandpass signal;

lowpass repr. X
LP
(F) =
+
(F +F
c
)

x
LP
(t ) =
+
(t )e
j 2Fc t
= [x
a
(t ) +j x
a
(t )]e
j 2Fc t
= x
I
(t ) +jx
Q
(t ).
This gives (from x
a
(t ) +j x
a
(t ) = [x
I
(t ) +jx
Q
(t )]e
j 2Fc t
)
x
a
(t ) = x
I
(t ) cos 2F
c
t x
Q
(t ) sin 2F
c
t and
x(t ) = x
I
(t ) sin 2F
c
t +x
Q
(t ) cos 2F
c
t .

Carrier components: cos 2F

c
t and sin 2F
c
t .

x
I
(t ) and x
Q
(t ): Low freq. amp. modulators.

x
I
(t ) and x
Q
(t ): quadrature components.
Alternatively:
x
a
(t ) = Re{[x
I
(t ) +jx
Q
(t )]e
j 2Fct
} = Re[x
LP
(t )e
j 2Fct
].

x
LP
(t ): Complex envelope of x
a
(t ) and it is the equivalent lowpass
signal.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 27 / 82
Sampling of bandpass signals Representation of bandpass signals
Representation of bandpass signals (5)
x
I
(t ); Known as in-phase component
x
Q
(t ); Known as quadrature component
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 28 / 82
Sampling of bandpass signals Representation of bandpass signals
Representation of bandpass signals (6)
Letting x
LP
(t ) = A(t )e
j (t )
,
A(t ) =
_
x
2
I
(t ) +x
2
Q
(t ) and (t ) = tan
1
(x
Q
(t )/x
I
(t )),
then a third possible representation is
x
a
(t ) = Re[x
LP
(t )e
j 2Fct
] = A(t ) cos[2F
c
t + (t )].

A(t ): Envelope of x
a
(t ).

(t ): Phase of x
a
(t )
Fourier transform of x
a
(t ):
X
a
(F) =
_

x
a
(t )e
j 2Ft
dt
=
_

{Re[x
LP
(t )e
j 2Fct
]}e
j 2Ft
dt
=
1
2
_

[x
LP
(t )e
j 2Fct
+x

LP
(t )e
j 2Fct
]e
j 2Ft
dt
=
1
2
[X
LP
(F F
c
) +X

LP
(F F
c
)].
Any bandpass signal x
a
(t ) can be represented by an equivalent
lowpass signal x
LP
(t ).
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 29 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (1)
Know: A signal with highest frequency B can uniquely be
represented by samples taken at the minimum (Nyquist) rate of
2B samples per second.
What if the signal is band limited; B
1
F B
2
:

Either sample at sample frequency 2B

2

or sample at 2 (B
2
B
1
)/2.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 30 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (2)
Integer band positioning (1)
Sampling of a bandpass signal at rate F
s
= 1/T produces a
sequence x[n] = x
a
(nT) with spectrum
X
a
(F) =
1
T

k=
X
a
(F kF
s
)

Positioning of shifted spectra controlled by F

s
.

Remember; real signals have two spectral bands. Care must be

taken!
Integer band positioning:

Restrict the higher frequency of the band to be an integer multiple

of the bandwidth, i.e. F
H
= mB.

m = F
H
/B; band position.

No aliasing with F
s
= 2B.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 31 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (3)
Integer band positioning (2)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 32 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (4)
Integer band positioning (3)
Signal can be reconstructed as x
a
(t ) =

n=
x
a
(nT)g
a
(t nT)
where g
a
(t ) =
sin Bt
Bt
cos 2F
c
t
g
a
(t ) is equal to the ideal interpolation function for lowpass
signals, modulated by the carrier frequency F
c
.
By properly choosing the center frequency F
c
of G
a
(F), we can
reconstruct a continuous-time signal with spectral band centered
at F
c
= (kB +B/2), k = 0, 1, . . ..

With k = 0 we obtain the baseband signal; down-conversion.

When the band position is an even integer, the baseband spectral
structure is inverted.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 33 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (5)
Arbitrary band positioning (1)
To avoid aliasing for spectral bands with arbitrary positioning, the
following must be true (see gure)
1
2F
H
kF
s
2
(1 k)F
s
2F
LP
We have to choose both k and F
s
.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 34 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (6)
Arbitrary band positioning (2)
Include guard bands when choosing k and F
s
to ensure sampling
frequency tolerances.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 35 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (7)
Sampling using bandpass signal representation (1)
May sample
+
(t ); the analytic signal (the pre-envelope of x
a
(t ) at
rate F
s
= 2B independent of band location).

This complex signal has one band (no symmetry)!

Needs B complex samples or 2B real samples.

+
(t ) = x
a
(t ) +j x(t ), x(t ) =
1
t
x
a
(t ) =
1

x()
t
d
Must sample x
a
(t ) and its Hilbert transform x(t ) and use complex
bandpass interpolation (as example given before).
Solution: Shift the signal by F
c
= (B
1
+B
2
)/2 for instant by
multiplying with the two quadrature carriers + lowpass lter and
sample each stream at 2 (B
2
B
1
)/2 i.e. represent the signal
with 2 (B
2
B
1
) samples per second.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 36 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (8)
Sampling using bandpass signal representation (2)
(u
c
(1) == x
I
(t ) and u
s
(t ) == x
Q
(t ))
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 37 / 82
Sampling of bandpass signals Sampling of bandpass signals
Sampling of bandpass signals (9)
Sampling using bandpass signal representation (3)
If F
c
+B/2 = kB, k pos. int., B = (B
2
B
1
) and 2B = 1/T;
x(nT) = x
I
(nT) cos(2F
c
nT) x
Q
(nT) sin(2F
c
nT)
= x
I
(nT) cos
n(2k 1)
2
x
Q
(nT) sin
n(2k 1)
2

For n = 2m i.e. even and T

1
= 2T = 1/B,
x(2mT) x(mT
1
) = x
I
(mT
1
) cos(m(2k 1)) = (1)
m
x
I
(mT
1
).

For n = 2m1 i.e. odd and T

1
= 2T = 1/B,
x(2mT T) x(mT
1
T
1
/2) = (1)
m+k+1
x
Q
(mT
1
T
1
/2).
Thus;

Even number samples produces the lowpass signal component

x
I
(t )

Odd number samples produces the lowpass signal component

x
Q
(t )
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 38 / 82
Multirate digital signal processing Sampling rate conversion
Sampling rate conversion
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 40 / 82
Multirate digital signal processing Sampling rate conversion
Decimation by a factor D (1)
Assume signal x(n) with spectrum X(w) which is nonzero in the
frequency interval 0 |w| or |F| F
x
/2.
Choosing every Dth sample reduces the sampling rate by a factor
D, but aliasing occurs!
The bandwidth of x(n) must, therefore, be reduced to
F
max
= F
x
/2D.
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 41 / 82
Multirate digital signal processing Sampling rate conversion
Decimation by a factor D (2)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 42 / 82
Multirate digital signal processing Sampling rate conversion
Interpolation by a factor I (1)
To increase the sampling rate with a factor of I, interpolate I 1
new samples between successive values of the signal.
One way; add I 1 zeros between every successive values of
x(n). Call the new sequence v(m):
v(m) =
_
x(m/I), m = 0, I, 2I, . . .
0, otherwise
v(m) has z-transform V(z) =

m=
v(m)z
m
= X(z
I
).
V(w
y
) is an I-fold periodic repetition of X(w
x
).
v(m) must be passed through a lowpass lter H
I
(w
y
) to create the
up-sampled sequence
H
I
(w
y
) =
_
I, 0 |w
y
| /I
0, otherwise
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 43 / 82
Multirate digital signal processing Sampling rate conversion
Interpolation by a factor I (2)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 44 / 82
Multirate digital signal processing Sampling rate conversion
Example (Up- and down-sampling)
A signal x[n] has Fourier transformation X() as shown in the gure below.

X()
0 /3/2

1
The signal is used as input to the systems I and II, which are dened in the
gure below:
I: x[n] 3 w
1
[n] H
0
(z) z
1
[n] 2 y
1
[n]
II: x[n] 2 w
2
[n] 3 z
2
[n] H
0
(z) y
2
[n]
M means downsampling by a factor M (keep every Mth sample) and
N means zero-interpolation by a factor N (insert N 1 zeros
between each sample). H
0
(z) is an ideal low-pass lter with a cut-off
frequency w
c
= /3 and gain equal to 2.
For both systems, sketch the Fourier transformation of the signals
w
1
[n], w
2
[n], z
1
[n], z
2
[n], y
1
[n] og y
2
[n].
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 45 / 82
Multirate digital signal processing Sampling rate conversion
Example (Up- and down-sampling)

W1()
/6 0 /3 /2

11

Z1()
/6 0 /3 /2

2

Y1()
/6 0 /3 /2

1

W2()
/6 0 /3 /2

1
1
2

Z2()
/6 0 /3 /2

1
2

Y2()
/6 0 /3 /2

1
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 46 / 82
Multirate digital signal processing Sampling rate conversion
Conversion by a rational factor I/D (1)
Cascade an interpolator with a decimator.
Remember: Up-sample rst, downsample second.

Preserve the desired spectral characteristics

The two lters can be combined

Possible with multistage implementations.

I =

L
i =1
I
i

D =

J
i =1
D
i

Design lters for each stage to avoid aliasing.

AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 47 / 82
Multirate digital signal processing Sampling rate conversion
Conversion by a rational factor I/D (2)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 48 / 82
Multirate digital signal processing Sampling rate conversion
Conversion by a rational factor I/D (3)
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 49 / 82
Multirate digital signal processing Sampling rate conversion
Subband Coding of Speech Signals
mpeg, jpeg ...
AA, INF4480 (I/UiO) Sampling and reconstruction April 2011 50 / 82