Professional Documents
Culture Documents
Chapter 4: Discrete-Time Fourier Transform (DTFT) 4.1 DTFT and Its Inverse
Chapter 4: Discrete-Time Fourier Transform (DTFT) 4.1 DTFT and Its Inverse
Chapter 4: Discrete-Time Fourier Transform (DTFT) 4.1 DTFT and Its Inverse
Inverse DTFT: Let X (w) be the DTFT of x[n]. Then its inverse is inverse Fourier integral of X (w) in the
interval {−π , π ).
1 π
x[ n] = jwn
∫ X ( w)e dw (4.5)
2π −π
This is also called the synthesis equation.
π
Derivation: Utilizing a special integral: ∫ e jwn dw = 2πδ [ n ] we write:
−π
4.1
π π ∞ ∞ π ∞
− jwk
∫ X ( w)e
jwn
dw = ∫ { ∑ x[ k ]e }e jwn
dw = ∑ x[k ] ∫ e − jw[ n −k ] dw = 2π ∑ x[ k ]δ [ n − k ] = 2π .x[ n ]
−π − π k = −∞ k = −∞ −π k = −∞
Note that since x[n] can be recovered uniquely from its DTFT, they form Fourier Pair: x[ n] ⇔ X ( w).
∞
Convergence of DTFT: In order DTFT to exist, the series ∑ x[ n]e − jwn must converge. In other words:
n= −∞
M
X M ( w) = ∑ x[n ]e − jwn must converge to a limit X (w) as M → ∞. (4.6)
n =−M
Convergence of X m (w) for three difference signal types have to be studied:
∞
• Absolutely summable signals: x[n] is absolutely summable iff ∑ | x[ n] | < ∞ . In this case, X (w) always
n= −∞
exists because:
∞ ∞ ∞
| ∑ x[ n]e − jwn | ≤ ∑ | x[n ] | . | e − jwn |= ∑ | x[n ] | < ∞ (4.7)
n = −∞ n = −∞ n = −∞
∞
• Energy signals: Remember x[n] is an energy signal iff E x ≡ ∑ | x[n ] | 2 < ∞. We can show that X M (w)
n = −∞
converges in the mean-square sense to X (w) :
π
Lim ∫ | X ( w) − X M ( w) | 2 dw = 0 (4.8)
M →∞ −π
Note that mean-square sense convergence is weaker than the uniform (always) convergence of (4.7).
• Power signals: x[n] is a power signal iff
1 N
Px = Lim ∑ | x[ n] | < ∞
2
N →∞ 2 N + 1 n = − N
• In this case, x[n] with a finite power is expected to have infinite energy. But X M (w) may still converge
to X (w) and have DTFT.
• Examples with DTFT are: periodic signals and unit step-functions.
• X (w) typically contains continuous delta functions in the variable w.
4.2
4.2 DTFT Examples
Example 4.1 Find the DTFT of a unit-sample x[ n] = δ [ n].
∞ ∞
X ( w) = ∑ x[ n]e − jwn = ∑ δ [ n]e − jwn = e − j 0 = 1 (4.9)
n = −∞ n = −∞
Similarly, the DTFT of a generic unit-sample is given by:
∞
DTFT{δ [ n − n 0 ]} = ∑ δ [ n − n 0 ]e − jwn = e − jwn0 (4.10)
n = −∞
Example 4.2 Find the DTFT of an arbitrary finite duration discrete pulse signal in the interval: N 1 < N 2 :
N2
x[ n] = ∑ c k δ [n − k ]
k = − N1
Note: x[n] is absolutely summable and DTFT exists:
∞ N2 N2 ∞ N2
X ( w) = ∑ { ∑ c k δ [ n − k ]}e − jwn = ∑ c k { ∑ δ [n − k ]e − jwn } = ∑ c k e − jwk (4.11)
n= −∞ k = − N1 k = − N1 n = −∞ k = − N1
Example 4.3 Find the DTFT of an exponential sequence: x[ n] = a n u[ n ] where | a |< 1 . It is not difficult to see
that this signal is absolutely summable and the DTFT must exist.
∞ ∞ ∞ 1
X ( w ) = ∑ a n .u[ n ]e − jwn = ∑ a n .e − jwn = ∑ ( ae − jw ) n = (4.12)
n = −∞ n =0 n =0 1 − ae − jw
Observe the plot of the magnitude spectrum for DTFT and X M (w) for: a = 0.8 and M = {2,5,10,20 , ∞ = DTFT}
4.3
Example 4.4 Gibbs Phenomenon: Significance of the finite size of M in (4.6).
For small M , the approximation of a pulse by a finite harmonics have significant overshoots and
undershoots. But it gets smaller as the number of terms in the summation increases.
Example 4.5 Ideal Low-Pass Filter (LPF). Consider a frequency response defined by a DTFT with a form:
1 | w |< w C
X ( w) = (4.13)
0 w C < w < π
4.4
Here any signal with frequency components smaller than wC will be untouched, whereas all other frequencies
will be forced to zero. Hence, a discrete-time continuous frequency ideal LPF configuration.
4.3.4 Linearity:
a. Zero-in zero-out and
b. Superposition principle applies: A.x[n ] + B. y[n ] ⇔ A.X ( w) + B. X ( w) (4.18)
4.3.5 Time-Shift (Delay) Property:
x[ n − D ] ⇔ e − jwD . X ( w ) (4.19)
4.3.6 Frequency-Shift (Modulation) Property:
X [ w − wC ] ⇔ e − jwC n .x[ n] (4.20)
Example 4.6 Consider a first-order system:
y[ n] = K 0 .x[ n] + K 1 .x[ n − 1]
H ( jw ) = Y ( w) / X ( w) = K 0 + K 1 .e − jw
4.3.7 Convolution Property:
x[ n] * h[ n ] ⇔ X ( w).H ( w) (4.21)
4.3.8 Multiplication Property:
1 π
x[ n]. y[ n] ⇔ ∫ X (φ ).Y ( w − φ ) dφ (4.22)
2π −π
4.3.9 Differentiation in Frequency:
dX ( w)
j. ⇔ n.x[ n ] (4.23)
dw
4.6
4.3.10 Parseval’s and Plancherel’s Theorems:
∞ 1 π
∑ | x[ n ] | =
2 2
∫ | X ( w) | dw (4.24)
n= −∞ 2π −π
If x[n] and/or y[n] complex then
∞ 1 π
∑ x[ n ]. y * [n ] = *
∫ X ( w).Y ( w)dw (4.25)
n= −∞ 2π −π
Example 4.7 Find the DTFT of a generic discrete-time periodic sequence x[n].
Let us write the Fourier series expansion of a generic periodic signal:
N −1 2π
x[ n] = ∑ a k e jkw0n where w0 =
k =0 N
N −1 N −1 N −1
X ( w) = DTFT{x[n ]} = DTFT{ ∑ a k e jkw0 n ) = ∑ a k .DTFT{e jkw0n ) = ∑ a k .2πδ ( w − kw0 ) (4.26)
k =0 k =0 k =0
Therefore, DTFT of a periodic sequence is a set of delta functions placed at multiples of kw 0 with heights a k .
4.7
• If the LTI system is stable then h[n] must be absolutely summable and DTFT exists and is continuous.
• We can recover h[n] from the inverse DTFT:
1 π
h[ n] = IDTFT{H ( w)} = jwn
∫ H ( w).e dw (4.28)
2π −π
• We call | H ( w) | as the magnitude response and ∠H (w) the phase response
For M=6 we plot the magnitude and the phase response of this system:
Notes:
2πk sin( wM / 2 )
1. Magnitude response Zeros at w = where =0
M sin( w / w)
2. Level of first sidelobe ≈ −13 dB
3. Phase response with a negative slope of − ( M − 1) / 2
2πk sin( wM / 2)
4. Jumps of π at w = where changes its sign.
M sin( w / w)
4.9
TABLE: 4.1 DISCRETE-TIME FOURIER TRANSFORM PAIRS
Signal DTFT
δ [n] 1
1 2π .δ ( w)
e jwC n 2π .δ ( w − wC )
N −1 N −1
jkw n
∑ a k .e C with NwC = 2π ∑ 2π .a k .δ ( w − kwC )
k =0 k =0
a n .u[n ]; | n |< 1 1
1 − a.e − jw
a | n| .; | n |< 1 1− a2
1 − 2 a. cos w + a 2
n.a n .u[ n ]; | n |< 1 a.e − jw
(1 − a.e − jw ) 2
n sin[ w( N + 1 / 2)]
rect [ ]
N sin[ w / 2]
sin wC n rect [ w / 2 w C ]
πn
4.10
TABLE 4.2 PROPERTIES OF DTFT
1. Linearity A.x1 [n ] + B.x 2 [ n] A. X 1 ( w) + B.X 2 ( w)
2. Time-Shift (Delay) x[n − N ] e − jwN . X ( w )
3. Frequency-Shift x[ n].e jwC n X (w − wC )
6. Periodic Signals x[ n] = x[ n + N ] ∞
∑ 2π .a k .δ ( w − kwC )
k = −∞
2π 1 − jkw n
wC = ak = ∑ x[n ].e C
N N <N >
Example 4.10 Response of an LTI system with H ( w) = DTFT{h[n ]} : Given x[ n] = e jwn ; a complex harmonic.
y[ n] = ∑ h[ k ].x[n − k ] = ∑ h[ k ].e jw ( n− k ) ={∑ h[ k ].e − jwk }.e jwn = H ( w). x[ n] (4.29)
k k k
Note that this is the ONLY time frequency-domain variable w and the time-domain variable n appear on the
same side of the equation. In all other cases, we have time domain variable in the time-domain and vice
versa. In calculus jargon, e jwn acts as the Eigenvector of the system.
4.11
4.5 FREQUENCY-SELECTIVE DISCRETE-TIME FILTERS
4.5.1 Ideal Low-Pass and High-Pass Filters
1 if | w |< wC 0 if | w |< w C
H LP ( w) = H HP ( w ) = (4.30a-d)
0 if w C <| w |< π 1 if wC <| w |< π
wC w n wC w n
h LP [ n ] = .Sinc( C ) h HP [ n ] = δ [n ] − h LP [ n] = δ [ n ] − .Sinc( C )
π π π π
4.5.2 Ideal Band-Pass and Band-Stop Filters
1 if | w − wC |< B / 2 0 if | w − wC |< B / 2
H BP ( w) = H BS ( w) = (4.31a-d)
0 elsewhere in ( −π , π ) 1 elsewhere in (−π , π )
h BP [ n ] = 2 . cos( w C n ). h LP [ n ] h BS [ n ] = δ [n ] − h BP [n ] = δ [ n] − 2 . cos(wC n).hLP [n ] w
wC = B / 2 C =B /2
4.12
• All of these ideal filters are non-causal and hence, non-realizable.
• They form benchmark for implementable real-life filters.
Consider an integer system, for which the input-output relationship is given by:
y[ n] = x[ n − k ], k ∈ Integer (4.32a)
The frequency response is computed:
Y ( w)
Y ( w) = e − jwk . X ( w) ⇒ H ( w) ≡ = e − jwk (4.32b)
X ( w)
The phase response of this system:
∠H ( w) = − wk (4.33)
is a linear function of the frequency variable w.
4.13
The corresponding system output is approximated by:
y[ n] ≈| H ( w 0 ) | .s[ n − τ G ( w)].e jw0 [ n −τ PH ( w0 )] (4.36)
It is easy to see that the phase delay τ PH contributes a phase shift to the carrier e jw 0 n , whereas the group
delay τ G causes a delay to the envelope s[n].
4.14
Pure delay, or All-Pass Filter:
When a system is a pure delay; i.e., its magnitude response is unity for all w and the phase is a linear
function of the delay τ .
| H ( w) |= 1 τ PH ( w) = τ G ( w) = 1 (4.37)
If the phase is linear but the magnitude may depend on w , then the system is labeled as a linear Phase
system:
H ( w) =| H ( w ) | .e j ∠H ( w) (4.38a)
where
∠H ( w) = − wτ (4.38b)
where the phase is a linear function of w with a slope − τ .
4.15