Professional Documents
Culture Documents
1
2.4.2
1 T
2
x(t) = sinc(2t/T ) (2.80)
t ←→ f
−1 1
T T
Combining sinc(t) ↔ rect(f ) with the time scaling T
property, x(at) ↔ |a|−1X(f /a), yields
( Figure 2.16: The Fourier transform pair sinc(2t/T ) ↔ T
rect(f T /2).
, |f | < T1
T 2
T
sinc(2t/T ) ↔ rect(f T /2) = 2 (2.81)
2 0, |f | > T1
2.4.4
2.4.3
2.4.5 2.4.6
In practice, we usually have to restrict our signals A softer restriction is that the signal x(t) has a
to not involve frequencies outside a certain nite fraction η of its energy inside the frequency band
band, [−W, W ] say. [−W, W ]; that is,
RW
However, mathematically such a strict restriction |X(f )|2df
on the frequency content of a signal cannot be
R−W
∞ =η (2.82)
−∞|X(f )|
2df
combined with the assumption of nite time-
duration!
2.4.7 2.4.8
Sampling Theorem
It certainly seems reasonable to say that the signal
rect(t/T ) has a Fourier bandwidth of W = 1/T Hz Consider the function
as our denition implies. √ sin 2πW t √
φ(t) = 2W = 2W sinc(2W t) (2.84)
To be more precise, we could say that the 90%- 2πW t
energy Fourier bandwidth of the signal rect(t/T ) is with Fourier transform
W = 1/T Hz, but we will usually not have need for ( 1
1 f √ , |f | < W
such precision. Φ(f ) = √ rect = 2W (2.85)
2W 2W 0, |f | > W
2.4.11 2.4.12
φ(t) is conned to the frequency band [−W, W ]. Orthogonality is an important notion in signal
k analysis.
Versions of φ(t) delayed by 2W form the set of
orthonormal functions It means that
Z
k √ k ∞
ek , l = k
φk (t) = φ t − = 2W sinc 2W t − φk (t)φ∗l (t)dt = (2.87)
2W 2W −∞ 0, l = 6 k
(2.86)
where ek is the energy of φk (t).
where k is an integer.
2.4.13
2.4.14
2.4.16
2.4.15
2.4.19 2.4.20
2.4.22
2.4.21
Because the functions sinc 2W t − 2W
k
are
Let x(t) be any signal whose Nyquist rate is 2W . 1
essentially 0 outside an interval of length W
By sampling x(t) at its Nyquist rate, we obtain n= seconds centered at their peaks, it follows from
2W T samples in the interval of T seconds starting (2.90) that we can write
1
at time − 12 2W = 1
− 4W . n−1
X k k
We now assume that the length of this time
x(t) ≈ x sinc 2W t − ,
T 2W 2W
k=0
interval is very large so that n >> 1.
1 1
− <t<T − (2.92)
4W 4W
2.4.23 2.4.24
2.4.26
Let TN denote the smallest τ such that ψ(t) is The Shannon bandwidth B of the basis signal ψ(t)
orthogonal to every time-shift of itself ψ(t − kτ ) is B = 1/2TN , or equivalently,
by a nonzero multiple of τ.
2B = 1/TN basis functions per second (2.94)
We call TN the Nyquist-shift of the basis signal
ψ(t).
The Nyquist-shift
of the signal φk (t) =
1
sinc 2W t − k
2W is TN = 2W .
2.4.28
2.4.27
Theorem 2.3
The full space of signals whose Fourier transform (Fundamental theorem of bandwidth)
is essentially zero for |f | ≥ W and that are
The Shannon bandwidth B of a basis signal is at
essentially time-limited to an interval of length T =
most equal to its Fourier bandwidth W; equality
mTN seconds has, according to the dimensionality
holds when the signal is a sinc function.
theorem, dimension n = 2W T while the dimension
of the subspace of these signals that can be
The signicance of B<W is that the considered
represented withψ(t) and its shifts is 2BT . Thus
signal and its shifts can then represent only
we must have 2BT ≤ 2W T , or equivalently
a proper subspace of the space of bandlimited
B≤W signals.
2.4.29 2.4.30
2.4.32
2.4.31
T T
= √ (rect(f T /3))(1 + 2 cos(ωT /3)) (2.96)
3 3
Figure 2.18: The triple-sinc signal and its Fourier transform.
2.4.34
2.4.33
Example 2.8
The Fourier bandwidth of the signal ψ(t) is seen to
3 Consider the basis signal
be W = 2T .
The Nyquist-shift of this signal is TN = T and 7
X
1
8 iT
hence its Shannon bandwidth is B = 2T .
ψ(t) = ai rect t− (2.98)
T 8
We see that i=0
2.4.35 2.4.36
2.4.37