EE 203: Signals and Systems

Notes on Nyquist Sampling Theorem and Discrete Fourier Transform

In this note, we state (and provide an idea of the proof for) the Nyquist sampling theorem
followed by an introduction to the concept discrete Fourier transform and its application to
computing sampled Fourier transform of a sampled signal.

1 Nyquist Sampling Theorem

Theorem 1 (Nyquist Sampling Theorem). Let f (t) be a bandlimited signal with bandwith
1
B Hz, that is, F () = 0, || > 2B. Then for every positive T (0, 2B ],

X h i
nT
F () = T f (nT )e , , , (Eq.1)
n=
T T
   
X t
f (t) = f (nT ) sinc n , t R. (Eq.2)
n=
T

Proof: By inverse Fourier transform,

1
Z
f (t) = F ()etd
2

1
Z
T
= F ()etd, (Eq.3)
2 T

where the second equality is a consequence of the fact that F () = 0 for all T
2B.
Now, applying Fourier series to F () (by the periodic extension technique) we get

X h i
nT
F () = Dn e , , ,
n=
T T

where Z
T T
Dn = F ()enT d = T f (nT ), n Z,
2
T
where second equality follows from (Eq.3) which further implies (Eq.1). Finally, it follows from
(Eq.1) and (Eq.3) that

1
Z
T
f (t) = F ()etd
2 T
Z
T T X
= f (nT )e(tnT ) d
2 T n=
Z
T X T
= f (nT ) e(tnT ) d
2 n= T

   
X t
= f (nT ) sinc n ,
n=
T

which implies (Eq.2). 

Remark 1. In Theorem , note that if the signal f (t) is not bandlimited but only essentially
bandlimited, that is, F () 1, || 2B then (Eq.1) and (Eq.2) will hold approximately.

2 Discrete-Time Fourier Series for Periodic Sequences

Definition 1. Let x[n] be a periodic sequence with period N, that is, x[n + N] = x[n], n Z.
The discrete-time Fourier series of x[n] is given by
2
X
x[n] = ak ekn N , n = hNi,
k=hN i

where
2
X
ak = x[n]ekn N , k = hNi.
n=hN i

3 Discrete-Time Fourier Transform

Definition 2. Let x[n] be a sequence. The discrete-time Fourier transform (DTFT) of x[n] is
given by
X

X(e ) = x[n]en , = [0 , 0 + 2].
n=

and the inverse discrete-time Fourier transform (IDTFT) of X(e ) is given by

1
Z
x[n] = X(e )en d, n Z.
2 2
4 Discrete Fourier Transform for Finite Length Sequences
Let x[n] be a finite length sequence of length N, that is, x[n] = 0, n < 0 or n > N 1. Let x[n]
be the periodic extension of x[n]. Then it follows from the discrete-time Fourier series that
2
X
x[n] = ak ekn N ,
k=hN i

where
N 1 N 1
1 X 2 1 X 2
ak = x[n]ekn N = x[n]ekn N .
N n=0 N n=0
The above discussion leads to the following definition.
Definition 3. Let x[n] be a finite length sequence of length N, that is, x[n] = 0, n < 0 or
n > N 1. Then the discrete Fourier transform of x[n] is given by another finite length
sequence of length N
N 1
2
=
X
X[k] x[n]ekn N , k = 0, . . . , N 1,
n=0

and the inverse discrete Fourier transform of X[k] is given by
N 1
kn 2
X
x[n] = X[k]e N , n = 0, . . . , N 1.
k=0

5 Discrete Fourier Transform for Sampled Signals

Theorem 2 (Discrete Fourier Transform). Let f (t) be a time and bandlimited signal with
1
bandwith B Hz and f (t) = 0, t 6 [0, T0 ]. Let T > 0 be such that T 2B and TT0 is some positive
integer N0 . Then
N 1
1 X
fm = F em0 , m = 0, . . . , N0 1, (Eq.4)
N0 =0
N
X 0 1

F = fm em0 , = 0, . . . , N0 1, (Eq.5)
m=0
2 2
where fm = T f (mT ) and F = F (0) and where 0 = T0
and 0 = N0
= 0 T .
Proof. Since f (t) is bandlimited it follows from (Eq.1) (see Theorem 1) that

X
F (0 ) = T f (mT )em0 T
m=
N
X 0 1

= T f (mT )em0 T , = 0, . . . , N0 1,
m=0
where the second equality is a consequence of timelimited signal assumption. This proves (Eq.5).
Now, it follows from (Eq.5) that
N
X 0 1 N
X 0 1 N
X 0 1

F ek0 = fm e(km)0
=0 =0 m=0
N
X 0 1 NX0 1

= fm e(km)0
m=0 =0
= N0 fk , k = 0, . . . , N0 1,

which proves (Eq.4). 

Remark 2. It is interesting to note that there does not exist a signal that is both time and band
limited. Hence, Theorem 2, though technically correct, is a vacuous result. However, it follows
from Remark 1 that the above result (specially (Eq.4) and (Eq.5)) will hold approximately for
timelimited and essentially bandlimited signals.

Remark 3. Note that (Eq.5) and (Eq.4) can be written as

F0 1 1 1 1 f0
F1 1 1 2 (N0 1)

f1

F2 1
= 2 4 2(N0 1) f2
.. .. .. .. .. .. ..


. . . . . . .
(N 1) 2(N 1) 2
FN0 1 1 0
0
0 1)
(N fN0 1

and
f0 1 1 1 1 F0

f1
1 2 N0 1
F1

f2 =
1 2 4 2(N0 1) F2 ,

.. .. .. .. .. .. ..


. . . . . . .
(N0 1)2
fN0 1 1 N0 1 2(N0 1) FN0 1
respectively, where = e0 . Hence, the computation of DFT is simply a multiplication of
a matrix with a vector. However, a typical length of these vectors can be in the order of 101 0
and hence an efficient algorithm is needed to compute the DFT (or IDFT). The fast Fourier
transform (FFT) is an algorithm that is quite an efficient method to compute DFT of a sequence,
which in turn can be used to get approximate (sampled) Fourier transform by using samples.