Week 11 (Chapter 7.)
Sampling Theorem and Reconstruction form Samples
[Communications System I]
Prof. Young-Chai Ko
EE at KU
May 23, 2018
Outline
1 Introduction to ADC
2 Sampling and interpolation Sampled signal
Interpolation Sampling rate
Signal bandwidth and sampling rate
3 Sampling theorem Preliminary Proof
Frequency-domain representation of the sampled signal
Introduction to ADC
Analog-to-Digital Conversion (ADC)
We need ”sampling theorem” and ”quantization” for analog-to-digital conversion (ADC).
We will study the sampling theorem and the quantization.
Sampling and interpolation
Sampling
Sampling of the analog signal x(t)
”Sampling” can be explained by the following expression:
xδ(t) =
∞
X
n=−∞
x(nTs)δ(t − nTs)
where Ts is sampling interval.
Sampling and interpolation Sampled signal
Sampled signal: Discrete time signal
Sampling circuit
Sampled signal which is discrete-time signal
Sampling and interpolation Sampled signal
Two questions in sampling
1 Can we reconstruct x(t) from x(nTs)?
The sampling interval may matter.
2 If so, how?
Interpolation may matter.
Sampling and interpolation Interpolation
Reconstructing x(t) from x(nT
s) by interpolator
We can reconstruct x(t) (perfectly or maybe approximately) from x(nTs) by interpolation.
Example of linear interpolation.
- Is it perfectly reconstructed?
No.
- Can we do better if we have a certain good interpolator?
Yes. Use sinc function as an interpolator which will be explained later.
Sampling and interpolation Sampling rate
Does sampling interval T
smatter in reconstruction?
Consider Ts1 > Ts2, say Ts1 = Ts and Ts2 = 0.5Ts. Then the sampled (discrete-time) signal x(nTs1) and x(nTs2) may look like the following:
Sampling and interpolation Sampling rate
Remark
Sampling interval Ts is smaller which means the faster sampling rate, we will have better reconstruction.
However, note that the faster sampling rate means increasing the power consumption and difficulty in circuit design which increases the cost.
The question is what is the minimum sampling rate to reconstruct x(t) from x(nTs).
Sampling and interpolation Signal bandwidth and sampling rate
Relation between Signal bandwidth and sampling rate
Sampling and interpolation Signal bandwidth and sampling rate
Remark 1
x2(t) is changing faster than x1(t).
- It means the bandwidth of x2(t) is larger than the bandwidth of x1(t).
Remark 2
Which do you think the sampling rate should be faster between x1(t) and x2(t)?
- Answer is x2(t).
Remark 3
We can conjecture that the sampling rate must be related to the bandwidth of the signal.
Sampling theorem
Concept of Sampling theorem
Goal of sampling
We want to convert the analog signal to discrete time signal with a certain sampling rate.
Then we want to reconstruct perfectly x(t) from x(nTs).
Two things to consider
1 Sampling rate
How fast do we need to sample the analog signal?
Is there any minimum sampling rate without having any problem in reconstructing the signal perfectly?
2 Interpolation
If the sampling rate is fast enough by a certain rule, can we have the interpolator for prefect reconstruction.
Sampling theorem
Sampling theorem
Rate and interpolator in sampling theorem
Consider the analog signal x(t) with the signal bandwidth W . That means, |X(f )| = 0 for
|f | > W .
Sampling interval Ts must be faster than 2W1 for perfect reconstruction:
Ts ≤ 1 2W
The sinc function should be used for interpolator for perfect reconstruction.
b sinc(at)
where a and b are certain constants which will be explained in detail later.
Sampling theorem
Interpolating the discrete-time signal using sinc function
Sampling theorem Preliminary
Preliminary
1 Convolution property in Fourier transform
F [x1(t) × x2(t)] = X1(f ) ∗ X2(f ) and vice versa.
2 Fourier transform of periodic signal x(t) with period T0.
- Let {xn} denote the Fourier series coefficients corresponding to this signal. Then x(t) =
∞
X
n=−∞
xnej2π nT0t
By taking the Fourier transform of both sides and using the fact that F
ej2π nT0t
= δ
f − n T0
, we obtain
X(f ) =
∞
X
n=−∞
xnδ
f − n T0
.
Sampling theorem Preliminary
1 Convolution property of FT
2 FT of periodic signal x(t)
X(f ) =
∞
X
n=−∞
xnδ
f − n T0
.
3 FT of impulse train x(t) = P∞
n=−∞δ(t − nT0)
X(f ) = F
∞
X
n=−∞
δ(t − nT0)
=
∞
X
n=−∞
xnδ
f − n T0
where xn is the Fourier coefficients when x(t) is expressed in Fourier series form.
Sampling theorem Preliminary
Now let us find xn:
xn = 1
T0 Z
T0
δ(t − nT0)e−j2π
n T0t
dt
= 1
T0
Z
T0
δ(t)e−j2π
n
T0(t+nT0)
dt
= 1
T0 Z
T0
δ(t)e−j2π
n T0t
ej2πn2
| {z }
=1
dt
= 1
T0 Z
T0
δ(t)e−j2π
n T0t
dt
= 1
T0
Sampling theorem Preliminary
Sampling theorem Preliminary
1 Convolution property of FT
F [x1(t) × x2(t)] = X1(f ) ∗ X2(f )
2 FT of periodic signal x(t) with period T0
X(f ) =
∞
X
n=−∞
xnδ
f − n T0
.
where xn is the Fourier coefficients in Fourier series form of x(t).
3 FT of impulse train with period T0
X(f ) = F
∞
X
n=−∞
δ(t − nT0)
= 1
T0
∞
X
n=−∞
δ
f − n T0
Sampling theorem Proof
Proof of sampling theorem
Let us denote the sampled signal as
xδ(t) =
∞
X
n=−∞
x(nTs)δ(t − nTs)
which can be rewritten as
xδ(t) = x(t) ×
∞
X
n=−∞
δ(t − nTs)
where we make use of the property:
x(t)δ(t − t0) = x(t0)δ(t − t0)
Sampling theorem Proof
Fourier transform of xδ(t):
Xδ(f ) = F [xδ(t)]
= F
x(t) ×
∞
X
n=−∞
δ(t − nTs)
By convolution property of FT, we have
Xδ(f ) = F [x(t)] ∗ F
∞
X
n=−∞
δ(t − nTs)
= X(f ) ∗ F
∞
X
n=−∞
δ(t − nTs)
= X(f ) ∗ 1 Ts
∞
X
n=−∞
δ
f − n Ts
= 1
Ts
∞
X
n=−∞
X
f − n Ts
Sampling theorem Proof
The sampled signal xδ(t) is given as
xδ(t) =
∞
X
n=−∞
x(nTs)δ(t − nTs)
Then, its Fourier transform is given as
Xδ(f ) = X(f ) ∗ 1 Ts
∞
X
n=−∞
δ
f − n Ts
= 1
Ts
∞
X
n=−∞
X
f − n Ts
Sampling theorem Frequency-domain representation of the sampled signal
Frequency-domain representation of the sampled signal
Sampling theorem Frequency-domain representation of the sampled signal
What happened if
T1s
≥ 2W ?
Sampling theorem Frequency-domain representation of the sampled signal
We can reconstruct X(f ) from X
δ(f ) if
T1s
≥ 2W ?
We can reconstruct X(f ) from Xδ(f ) by passing through ideal lowpass filter.
Sampling theorem Frequency-domain representation of the sampled signal
Reconstruction of X(f ) from X
δ(f )
We can reconstruct X(f ) from Xδ(f ) such as:
X(f ) = Xδ(f )H(f ) where H(f ) is given as
H(f ) = Ts
Y
f 2W
Then
X(f ) = Xδ(f )TsY f 2W
Sampling theorem Frequency-domain representation of the sampled signal
Taking the inverse Fourier transform of both sides, we obtain x(t) = xδ(t) ∗ 2W Tssinc(2W t)
=
∞
X
n=−∞
x(nTs)δ(t − nTs)
∗ 2W Tssinc(2W t)
=
∞
X
n=−∞
2W Tsx(nTs)sinc(2W (t − nTs))
−
∞
X
n=−∞
x(nTs)sinc
t Ts
− n
Sampling theorem Frequency-domain representation of the sampled signal
What happened if
T1s
< 2W ?
There is no way to reconstruct X(f ) from Xδ(f ).