chapter 7 sampling - yuntech
TRANSCRIPT
2020/6/24
1
Signal and Systems
Chapter 7
Sampling
β’ Under some conditions, a continuous-time signal can be completely represented by and recoverablefrom knowledge of its values, at points equally spaced (not always) in time.
β’ Exploit sampling to convert a continuous-time signal to a discrete-time signal, process the discrete-time signal using a discrete-time system, and then convert back to continuous time.
2020/6/24
2
β’ Processing discrete-time signals is more flexible and is often preferable to processing continuous-time signals.
β’ This is due to in large part to the dramatic development of digital technology over the past few decades
β’ Resulting in the availability of inexpensive, lightweight, programmable, and easily reproducible discrete-time systems.
7.1 Representation of a continuous-time
signal by its samples: The sample theorem
β’ Figure 7.1, three different continuous-time
signals have identical values at integer
multiples of T.
2020/6/24
3
)()()(But
)()()(
321
321
txtxtx
kTxkTxkTx
7.1.1 Impulse-Train Sampling
β’ T: sampling period
β’ : fundamental frequency
β’ p(t): sampling function
β’ xp(t)=x(t)p(t)
Ts /2
2020/6/24
4
7.1.1 Impulse-Train Sampling
xp(t)=x(t)p(t)
n
nTttp )()(
k
sp
s
k
p
nn
p
kjXT
jX
kT
jP
djPjXjX
nTtnTxnTttx
tptxtx
))((1
)(
)(2
)( 299, p. 4.8, Example From
))(()(2
1)(
)()()()(
)()()(
2020/6/24
5
Impulse-Train Sampling
Figure 7.3(c):
There is no overlap between the shifted replicas
of X(j)
Figure 7.3(d): there is overlap between the shifted
replicas
Ms
MS
2
)(M
Ms
MsM
2
)(
2020/6/24
6
If s>2M, x(t) can be
recovered exactly from xp(t)
by means of a lowpass filter
with gain T and a cutoff
frequency greater than M
and less than s-M.
Sampling theorem:
β’ Let x(t) be a band-limited signal with π ππ = 0for π > ππ. Then x(t) is uniquely determined by
its samples π₯(ππ), π = 0,Β±1,Β±2,β¦ , if ππ > 2ππ,
where ππ =2π
π.
β’ Given these samples, we can reconstruct x(t) by
generating a periodic impulse train in which
successive impulses have amplitudes that are
successive sample values.
2020/6/24
7
β’ This impulse train is then processed through an ideal lowpass filter with gain T and cutoff frequency greater than ππ΄ and less than ππ βππ΄. The resulting output signal will exactly equal x(t).
β’ The frequency 2M, which under the sampling theorem, must be exceeded by the sampling frequency s, is commonly referred to as the Nyquist rate.
7.1.2 Sampling with a Zero-Order Hold
β’ In practice, narrow, large-amplitude pulses are relatively difficult to generate and transmit.
β’ It is often more convenient to generate the sampled signal in a form of referred to as a zero-order hold.
β’ Such a system samples x(t) at a given instant and holds that value until the next instant at which a sample is taken (Fig. 7.5)
2020/6/24
8
The output x0(t) of the zero-order hold can be
generated by impulse-train sampling followed by
an LTI system with a rectangular impulse response
(Fig. 7.6)
Sampling with
a Zero-Order
Hold
2020/6/24
9
To reconstruct x(t) from x0(t), we consider
processing x0(t) with an LTI system with
impulse response hr(t) and frequency
response Hr(j). (Fig. 7.7)
We wish to specify Hr(j) so that r(t)= x(t)
π»0 ππ = πβπππ/22 sin
ππ2
π, (7.7)
π»π ππ =ππππ2 π»(ππ)
2sin(ππ/2)
π
, (7.8)
π»0 ππ π»π ππ = π» ππ
2020/6/24
10
If the cutoff frequency of H(j) is s /2, the ideal
magnitude and phase for the reconstruction filter
following a zero-order hold is shown below:
Sampling with a Zero-Order Hold
In practice the frequency response in Eq. (7.8) cannot be
exactly realized, and thus an adequate approximation to it
must be designed.
In fact, in many situations, the output of the zero-order
hold is considered an adequate approximation to the
original signal by itself, without any additional lowpass
filtering, and present a possible, although admittedly very
coarse, interpolation between the sample values.
2020/6/24
11
7.2 Reconstruction of a signal from its
samples using interpolation
Interpolation: the fitting of a continuous
signal of a set of sample values
β Zero-order hold
β Linear interpolation (Fig. 7.9)
Back to Fig. 7.4,
π₯π π‘ = π₯π π‘ β β π‘ =π=ββ
β
π₯ ππ β π‘ β ππ
For ideal lowpass filter π» ππ in Fig. 7.4
β π‘ =ππππ ππ(πππ‘)
ππππ‘,
π₯π π‘ =π=ββ
β
π₯ πππππ
π
π ππ(ππ(π‘ β ππ))
ππ(π‘ β ππ).
2020/6/24
12
Reconstruction of a signal from its
samples using interpolation
Reconstruction of a signal from its
samples using interpolation
The zero-order hold is a very rough approximation,
although in some cases it is sufficient.
Higher order holds: a variety of smoother interpolation
strategies.
2020/6/24
13
Reconstruction of a signal
from its samples using
interpolation under
different sampling
frequencies
Reconstruction of a signal from its samples using
interpolation
Linear interpolation: sometimes referred to as a first-order hold with h(t) triangular, the associated transfer function is
π» ππ =1
π[sin
ππ2
π2
]2
2020/6/24
14
2020/6/24
15
7.3 The effect of undersampling:
Aliasing
When ππ < 2ππ, the spectrum of x(t), is no
longer replicated in ππ(ππ) and thus is no longer
recoverable by lowpass filtering. This effect, in
which the individual terms in Eq. (7.6) overlap,
is referred to as aliasing (ζ··η).
ππ ππ =1
π
π=ββ
β
π(π(π β πππ )) (7.6)
The effect of undersampling: Aliasing
)13.7(,......2 ,1 ,0),()( nnTxnTxr
)14.7(cos)( 0ttx
)()cos()(;6
5)(
)()cos()(;6
4)(
)(cos)(;6
2)(
)(cos)(;6
)(
00
00
00
00
txttxd
txttxc
txttxb
txttxa
srs
srs
rs
rs
2020/6/24
16
The effect of
undersampling:
Aliasing
2020/6/24
17
)15.7()cos()( 0 ttx
2020/6/24
18
β’ The sampling theorem explicitly requires that
the sampling frequency be greater than twice
the highest frequency in the signal, rather than
less than or equal to twice the highest
frequency.
β’ Sampling a sinusoidal signal at exactly twice
the highest frequency is not sufficient.
According to Prob. 7.39,
π₯π π‘ = πππ β cos(ππ
2π‘)
2020/6/24
19
Discussions:
(1) Perfect reconstruction of x(t) occurs only in the case in
which β = 0 or an integer multiple of 2π.
(2) Otherwise, the signal π₯π(π‘) does not equal π₯(π‘).
(3) An extreme case, when β = βπ/2, π₯π π‘ = sin(π
π
2π‘).
The values of the signal at integer multiples of the sample
period 2π/ππ are zero.
The effect of undersampling, whereby higher frequencies
are reflected into lower frequencies, is the principle on
which the stroboscopic effect is based.
2020/6/24
20
7.4 Discrete-Time Processing of
Continuous-Time Signals
Through the process of periodic sampling with
the sampling frequency consistent with the
conditions of the sampling theorem, the
continuous-time signal is exactly
represented by a sequence of instantaneous
values
)(][ nTxnx cd (7.16)
)(txc
)(nΞ€xc
7.4 Discrete-Time Processing of Continuous-Time Signals
Figure 7.19 will be referred to as continuous-to-discrete-
time conversion and will be abbreviated C/D. The reverse
operation (D/C) corresponding to the third system in Figure
7.19
2020/6/24
21
)(][ nTyny cd
: continuous-time frequency variable
Ξ©: discrete-time frequency variable
The continuous-time Fourier transforms of π₯π(π‘)and π¦π(π‘) are ππ(ππ) and ππ(ππ), respectively,
while the discrete-time Fourier transform of
π₯π[π] and π¦π[π] are ππ(ππΞ©) and ππ(π
πΞ©), respectively.
2020/6/24
22
2020/6/24
23
π₯π π‘ =
π=ββ
β
π₯π(ππ)πΏ(π‘ β ππ)
And since the transform of πΏ(π‘ β ππ) is πβππππ, it follows that
ππ ππ = Οπ=βββ π₯π(ππ) π
βππππ
ππ ππΞ© =
π=ββ
β
π₯π π πβπΞ©n
=
π=ββ
β
π₯π(ππ)πβπΞ©n
ππ ππΞ© and ππ ππ are related through
ππ ππΞ© = ππ πΞ©/π
ππ ππ =1
πΟπ=βββ ππ π π β πππ (7.6)
ππ ππΞ© =1
πΟπ=βββ ππ π Ξ© β 2ππ /π (7.23)
2020/6/24
24
ππ ππΞ© is a frequency-scaled version of ππ ππ and, in
particular, is periodic in Ξ© with period 2π.
β’ Recovery of the continuous-time signal π¦π(π‘)from this impulse train is then accomplished
by means of lowpass filtering.
2020/6/24
25
2020/6/24
26
The overall system of Fig. 7.24 is equivalent to a C-T LTI
system with frequency response Hc(j) which is related to the
D-T frequency response through
)()()( Tj
dcc eHjXjY (7.24)
2/ ,)(
2/ ,0)( s
Tjd
s
eH
c jH
(7.25)
Figure 7.26
2020/6/24
27
jjH c )(
c
c
j
c jH
,
,0)(
,)(
TjeH j
d
(7.26)
(7.27)
(7.28)
7.4.1 Digital differentiator
Figure 7.28
2020/6/24
28
By considering the output of the digital differentiator for a continuous-time sinc
input, we may conveniently determine the impulse response hd[n] of the discrete-
time filter in the implementation of the digital differentiator. With reference to
Figure 7.24, let
[Example 7.2]
where T is the sampling period. Then
P.593
which is sufficiently band limited to ensure that sampling xc(t) at frequency Οs =
2Ο/T does not give rise to any aliasing. It follows that the output of the digital
differentiator is
For xc(t) as given by eq. (7.29), the corresponding signal xd[n] in Figure 7.24
may be expressed as
That is, for n β 0, xc(nT ) = 0, while
P.594
2020/6/24
29
which can be verified by lβHΓ΄pitalβs rule. We can similarly evaluate yd[n] in Figure
7.24 corresponding to yc(t) in eq. (7.30). Specifically,
which can be verified for n β 0 by direct substitution into eq. (7.30) and for n = 0 by
application of lβHΓ΄pitalβs rule.
Thus when the input to the discrete-time filter given by eq. (7.28) is the scaled unit
impulse in eq. (7.31), the resulting output is given by eq. (7.32). We then conclude
that the impulse response of this filter is given by
Note that the impulse
response hd [n] derived
here is equivalent to
that in eq. (7.28a).P.594
In Example 7.2, we apply a trick to find the discrete-time filter that has the effect
equivalent to that of a continuous-time filter in the bandlimited case. That is:
(1) First, find the output yc(t) of the continuous-time system if the input xc(t) is
the sinc function as in equation (7.29).
(2) Then, the impulse response of the equivalent discrete-time system is hd [n] =
T yc(nT ).
Supplement
P.594
2020/6/24
30
7.4.2 Half-Sample Delay
)()( txty cc
)()( jXejY c
j
c
cje
c jH
,
otherwise,0)(
(7.33)
(7.34)
Figure 7.29
2020/6/24
31
,)( /Tjj
d eeH
][][Ξ€
nxny dd
(7.35)
(7.36)
7.4.2 Half-Sample Delay
Figure 7.30
yc(t) is equal to samples of
a shifted version of the band-limited
interpolation of the sequence xd[n].
2020/6/24
32
Example 7.3 (1/2)
Example 7.3 (2/2)
2020/6/24
33
7.5 Sampling of Discrete-Time Signals
β’ 7.5.1 Impulse-train sampling
β’ 7.5.2 Discrete-time decimation and
interpolation
7.5 SAMPLING OF DISCRETE-TIME SIGNALS
7.5.1 Impulse-Train Sampling (1/3)
β’ The new sequence resulting from the sampling
processed in equal to the original sequence at
integer multiples of the sampling period N and is zero
at the intermediate samples; that is,
Nnnx
p nx ofmultipleintegeran if,][
otherwise,0][
][nxp
][nx
2020/6/24
34
7.5.1 Impulse-Train Sampling (2/3)
β’ The effect in the frequency domain of discrete-time
sampling is seen by using the multiplication property
developed in Section 5.5. Thus, with
k
p kNnkNxnpnxnx ][][][][][
Figure 7.31 Discrete-time sampling
2
)( )()(P2
1)( deXeeX jjj
p
2020/6/24
35
7.5.1 Impulse-Train Sampling (3/3)
β’ AS in Example 5.6, the Fourier transform of the
sampling sequence is
where ,the sampling frequency , equals .
Combining Eqs. (7.40) and (7.41), we have
K
sKN
P )(2
)(e j
1
0
)()(
1)(
N
k
kjj
pseX
NeX
][np
s N/2
Sampling of a discrete-time signal: the new sequence π₯π[π]resulting from the sampling process is equal to the original sequence π₯[π] at integer multiples of the sampling period Nand is zero at the intermediate samples;
π₯π π = απ₯ π , if π = an integer multiples of π0, otherwise
π₯π π = π₯ π π π =
π=ββ
β
π₯ ππ πΏ[π β ππ]
ππ πππ =1
2πΰΆ±2ππ πππ X ππ πβπ dΞΈ
π πππ =2π
π
π=ββ
β
πΏ(π β πππ )
ππ πππ =1
π
π=0
πβ1
π(ππ πβπ )
2020/6/24
36
Figure 7.33 (1/2)
2020/6/24
37
Figure
7.33
(2/2)
Figure 7.34: Relationship between xp[n] corresponding to sampling and
xb[n] corresponding to decimation
2020/6/24
38
In between the sampling instants, π₯π[π] is known to be zero. Therefore,
it is inefficient to represent, transmit, or store the sampled sequence.
The sampled sequence π₯π[π] is typically replaced by a new sequence
π₯π[π]= π₯π ππ = π₯ ππ .
Decimation: the operation of extracting every Nth sample.
ππ πππ =
π=ββ
β
π₯π[π]πβπππ
=1
π
π=0
πβ1
π₯π[ππ] πβπππ
Let π = ππ, ππ πππ =
π=ππ
π₯π[π]πβπππ/π
ππ πππ =
π=ββ
β
π₯π[π]πβπππ/π = ππ π
πππ
Since π₯π[π]=0 when n is not an integer multiple of N.
β’ The spectra for the sampled sequence and the
decimated sequence differ only in a frequency
scaling or normalization.
β’ The effect of decimation is to spread the
spectrum of the original sequence over a larger
portion of the frequency band.
2020/6/24
39
Figure 7.35 Frequency-domain illustration of the
relationship between sampling and decimation
Figure 7.36 (1/2)
2020/6/24
40
Figure 7.37 (1/2)
2020/6/24
41
Example 7.5
2020/6/24
42
Homework
β’ 7.3, 7.6, 7.8, 7.10