Lecture 4:Sampling and Reconstruction

• Sampling

• Data Reconstruction (Hold)

• Reading: Chapter 3 of the textbook

A Typical Digital Control System

r(t) e(t) Sampler e(kT ) m(kT ) m̄(t) Plant c(t)
Digital Filter Data hold
(A/D) D(z) (D/A) Gp (s)

H(s)

• In the above digital control system, sampling and data reconstruction

of signals are needed to interface the digital computer with the
physical world
Sampling

Continuous-time Discrete-time
Signals Signals

Data reconstruction

1
 Sampler
• A sampler obtains from a continuous-time signal a
discrete-time signal by sampling every T seconds
– Input is a continuous-time signal
– Output is a discrete-time signal e(kT), k=0,1,…
– Both signals have identical values at sampling moments
– Not a lossless procedure in general
– Not a time-invariant procedure

Reconstruction: Zero-Order Hold

• Data reconstruction devices are needed to convert the discrete-time
signal computed by the digital controller into a continuous-time
signal suitable for driving the actuator
– Ideally both signals should agree at the sampling moment: 0, T, 2T, …
• The simplest data reconstruction is zero-order hold
– Given a discrete time signal e(kT), k=0,1,…
– A continuous-time signal is produced
– The value of at any time t is equal to the value of e(kT) in the closest
sampling moment preceding t:

ē(t) = e(0)[u(t) − u(t − T )] + e(T )[u(t − T ) − u(t − 2T )] + · · ·

2
 Sampled-Data Control System
e(kT ) u(kT )
e(t) Digital u(t) y(t)
Sampler Data hold Plant
controller

• Digital controller is designed together with sampler and data hold

• Does there exist a transfer function from e(t) to u(t)?
– The answer is no
• How to characterize the input-output relation from e(t) to u(t) based
on the transfer function D(z) of the digital controller?
• Let us start from the simplest case: digital controller does nothing

e(t) e(kT ) ē(t)

Sampler Data hold

sampler and hold

Sampler and Zero-Order Hold

Sampler and zero-order hold

In general, original signal e(t) can not be fully reconstructed:

What is the effect of sampler and hold in frequency domain?

e(t) e(kT ) ē(t)

Sampler Data hold
E(s) E(z) Ē(s)

3
 Sampler and Zero-Order Hold
Laplace transform of
Z-transform of e(kT):
Laplace transform of
ē(t) = e(0)[u(t) − u(t − T )] + e(T )[u(t − T ) − u(t − 2T )] + · · ·
e−T s −T s −2T s
Ē(s) = e(0)[ 1s − s
] + e(T )[ e s − e s ] + ···

−T s

= [e(0) + e(T )e −T s
+ e(2T )e −2T s
+ · · ·] 1−es
∞
−T s

= [ k=0 e(kT )e −kT s
] 1−es

Independent of signal e(t)

Sampler and Zero-Order Hold (cont.)

In effect, the sampler and zero-order hold can be broken up in two steps:

E(s) T E ∗(s) Ē(s)

1−e−sT
Gh0(s) = s

Ideal sampler 0-th order data hold

In time domain:

e(t) T e∗(t) ē(t)

T
where
e∗ (t) = L−1 [E ∗ (s)] = e(0)δ(t) + e(T )δ(t − T ) + e(2T )δ(t − 2T ) + · · ·
is a string of impulses whose amplitudes are modulated by e(t)
∞
E ∗ (s) = k=0 e(kT )e−kT s

4
 Ideal Sampler
e(t) T e∗ (t)

e∗ (t) = e(0)δ(t) + e(T )δ(t − T ) + e(2T )δ(t − 2T ) + · · ·

= e(t) · [δ(t) + δ(t − T ) + δ(t − 2T ) + · · ·]

δT (t) δT (t)
δT (t) is a string of unit impulses T second apart:
T 2T 3T 4T 5T
δT (t)

e(t) Impulse e∗ (t)

Modulator

Remarks
• Ideal sampler is introduced for theoretical convenience and does not
correspond to physical entity
• Ideal sampler is linear but not time-invariant; can map different e(t)
to the same e*(t)
• Thus ideal sampler cannot be described by a transfer function

• If e(t) is discontinuous at a sampling time kT, then in the ideal

sampler, e(kT) is taken to be the right limit of e(t) at time kT:
e∗(t) = e(0+ )δ(t) + e(T + )δ(t − T ) + e(2T + )δ(t − 2T ) + · · ·

E ∗(s) = ∞ + −kT s
k=0 e(kT )e

• If e(t) contains impulses itself at some sampling times, then e*(t)

does not exist (not likely in practice)

5
 Summary: First Perspectives of Sampler and Hold
In the time domain:

e(t) T e(kT ) ē(t)

Zero-Order Hold
Sampler

In the frequency domain (a mixture of z- and Laplace transforms):

E(s) T E(z) Ē(s)

Zero-Order Hold
Sampler
∞ −k
E(z) = k=0 e(kT )z is called the z-transform of E(s)

Summary: Second Perspectives of Sampler and Hold

In the time domain:

e(t) T e∗ (t) ē(t)

T
Ideal sampler Data hold

In the frequency domain:

E(s) T E ∗(s) Ē(s)

1−e−T s
Gh0 (s) = s

∞
E ∗(s) = k=0 e(kT )e
−kT s
is called the star-transform of E(s)

6
 Relating the Two Perspectives

e(kT )
E(z)

e(t)
E(s) ē(t)
Ē(s)
s
−T
e
1− s
=
e∗(t) (s)
G h0
E ∗(s)

Finding Star-Transform
∞
• Given E(s), find E ∗ (s) = k=0 e(kT )e
−kT s


• Approach I: E ∗ (s) = [residues of E(λ) 1−e−T1 (s−λ) ]
where the summation is over all poles of E(λ)
(Note: E(λ) needs to be a rational function and ….)
∞ e(0)
• Approach II: E ∗(s) = T1 k=−∞ E(s + jkωs ) + 2
where ωs = 2π
T is the radian sampling frequency
(Derivations can be found in the Appendix III)

• Approach III: find E(z) either from the table on pp. 676,
 1
or by E(z) = [residues of E(λ) 1−z−1 eT λ , then
]
E ∗ (s) = E(z)|z=eT s

7
 Examples
z 1 1
E(s) = 1
s
E(z) = z−1 = 1−z −1 E ∗ (s) = 1−e−T s

z 1 1
E(s) = 1
s+1
E(z) = z−e−T = 1−e−T z−1 E ∗ (s) = 1−e−(1+s)T

Example
1 T ze−aT
E(s) = (s+a)2 E(z) = (z−e−aT )2 E ∗ (s) =

8
 More Examples
1
E(s) = (s+1)(s+2) E ∗ (s) =

ω z sin(ωT )
E(s) = s2 +ω 2 E(z) = z 2 −2z cos(ωT )+1

E ∗ (s) = 1 1 1 e−T s sin ωT

2j [ 1−e−T s ejωT − 1−e−T s e−jωT ] = 1−2e−T s cos ωT +e−2T s

Effect of Time Delay on Star-Transform

e1 (t) T e∗1 (t) e(t) = e1 (t − t0 ) T e∗ (t)

E1 (s) E1∗ (s) E(s) = E1 (s)e−t0 s E ∗ (s) =?

• In the special case when t0=nT is an integer multiple of the sampling

period T, we have the simple relation
E ∗ (s) = e−nT s E1∗ (s)

• For general t0, the above is NOT true

9
 Property I of E*(s)
E*(s) is periodic in s with period jωs = j 2π
T

E ∗ (s + jmωs ) = E ∗ (s)

Example:
1 1
E(s) = s
E ∗ (s) = 1−e−T s
1
E(s) = 1
s+1
E ∗ (s) = 1−e−(1+s)T

Proof: From property

1
∞ e(0)
E ∗ (s) = T k =−∞ E(s + jkωs ) + 2

Or from

Property II of E*(s)
If E(s) has a pole at p, then E*(s) has poles at p ± jmωs
E(s) E ∗(s)

×
× Primary strip
×

Example:
1
E (s) = 1
s
E ∗ (s) = 1−e−T s

1
E (s) = 1
s+1
E ∗ (s) = 1−e−(1+s)T

10
 Data Reconstruction

Second Perspectives of Sampler and Hold

In the time domain:

e(t) T e∗ (t) ē(t)

T
Ideal sampler Data hold

In the frequency domain:

E(s) T E ∗ (s) Ē(s)

1−e−T s
Gh0 (s) = s

11
 Shannon’s Sampling Theorem
When is the ideal sampler an invertible procedure: from the
sampled signal e*(t) we can uniquely reconstruct e(t)?
∞
E ∗ (s) = T1 k =−∞ E (s + jkωs ) + e(0)
2

Thus, the Fourier transform of e*(t) can be obtained from the Fourier transform
of e(t) as:
1
∞ e(0)
E ∗ (jω) = T k=−∞ E(jω + jkωs ) + 2

• Shannon’s Sampling Theorem: the signal e(t) can be uniquely determined

from its sampling e*(t) with sampling period T if its Fourier transform E(jω)
contains no frequency component greater than π/T

• Or equivalently, one has to sample the signal twice as fast as the fastest
changing part in e(t) in order to be able to reconstruct it uniquely

Data Reconstruction in Frequency Domain

e(t) e∗(t) ē(t)

E(jω) ∗ Ē(jω)
E (jω)

Ideal sampling Data reconstruction

∞ e(0)
E ∗ (jω) = 1
T k=−∞ E(jω + jkωs ) + 2 Ē(jω) = Gh0 (jω)E ∗ (jω)
(periodic extension in freq. domain) (data hold transfer function)

• Data reconstruction transfer function is normally a low pass filter that

extracts the low frequency part of E ∗(jω)
• By the Shannon’s Sampling Theorem, a lossless data reconstruction is
possible only if E(jω) does not contain frequency higher than ω2s = Tπ
• In this case, the perfect data reconstruction is achieved by adopting the
ideal low pass filter as the data hold transfer function (physically infeasible)

12
 Zero-Order Hold
• A physically feasible (causal) data hold transfer function is given by
−T s
Gh0 (s) = 1−es
with unit impulse response 1
T
• Frequency response
Gh0 (jω) = 1−e−jωT
jω
|Gh0 (jω)| = T | sin(πω/ω
|πω/ωs |
s )|

Example
Time Doman Frequency Doman
E(jω)
e(t) = 2 cos(ω1 t)
−ω1 ω1

Ideal sampling

E ∗ (jω)

−ωs −ω1 ω1 ωs

Zero-order hold

Ē(jω)

−ωs −ω1 ω1 ωs

13
 First-Order Hold
Idea: Given a sequence of sampled data e(kT ), k = 0, 1, . . .
reconstruct the signal between sampling times approximately by first order Euler
expansion
eo (t) ≃ e(kT ) + e′ (kT )(t − kT ), kT ≤ t < (k + 1)T
e(kT )−e((k−1)T )
≃ e(kT ) + T (t − kT )

Example of applying first-order hold

T 2T

Transfer Function of First-Order Hold

Unit impulse response Why?

1
2T
0
T
−1
1 2
eo (t) = u(t) + T tu(t) − 2u(t − T ) − T (t − T )u(t − T )
1
+u(t − 2T ) + T
(t − 2T )u(t − 2T )

2
1+T s 1−e−T s
Gh1 (s) = T s

14
 Fractional-Order Holds
Idea: Given a sequence of sampled data reconstruct the signal between
sampling times approximately as follows
eo (t) ≃ e(kT ) + ηe′ (kT )(t − kT ), kT ≤ t < (k + 1)T

≃ e(kT ) + η e(kT )−e((k−1)T

T
)
(t − kT )

Unit impulse response

1
2T
0
T
−1
−T s
η
Ghη (s) = (1 − ηe−T s ) 1−es + T s2 (1 − e−T s )2

15