Prepared by Dr Harish M S

Module 3
DFT Properties
DFT of a Time reversal sequence: If
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛) ↔ 𝑋(𝑘) Then
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥((−𝑛))𝑁 = 𝑥(𝑁 − 𝑛) ↔ 𝑋(𝑁 − 𝑘)
Proof: The N-point DFT of 𝑥(𝑛) isX(k)
N−1

X(k) = ∑ x(n)WNnk 0 ≤ 𝑘 ≤ 𝑁 − 1
n=0

Put 𝑚 = 𝑁 − 𝑛, 𝑛=𝑁−𝑚
−1 N−1
(N−m)k (N−m)k
X(k) = ∑ x(N − m)WN = ∑ x(N − m)WN
m=N m=0
N−1

X(k) = ∑ x(N − m)WNNk WN−mk

m=0

Put 𝑘 = 𝑁 − 𝑘
N−1
−(N−k)m
X(N − k) = ∑ x(N − m) WN
m=0
N−1

X(N − k) = ∑ x(N − m) WN−Nm WNmk

m=0

Since m is the dummy variable, it can be replaced by n, thus

N−1

X(N − k) = ∑ x(N − n) WNnk

n=0
N−1

X(N − k) = X((−k))N = ∑ x(N − n) WNnk 0 ≤ 𝑘 ≤ 𝑁 − 1

n=0

Therefore reversing the N-point sequence in time is equivalent to reversing the DFT values.
And we know that 𝑋 ∗ (𝑘) = X(N − k) = X((−k))N
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥((−𝑛))𝑁 = 𝑥(𝑁 − 𝑛) ↔ 𝑋 (𝑁 − 𝑘 )= 𝑋 ∗ (𝑘 )
Problem: Determine the 4-point DFT of the sequence 𝑥(𝑛) = (1,2,1) , and also compute the
DFT of 𝑥((−𝑛) )4

1
 Prepared by Dr Harish M S

Solution: For 4-point DFT𝑥(𝑛) = (1,2,1,0)

The 4-point DFT of 𝑥(𝑛) isX(k)
3

X(k) = ∑ x(n)W4nk 0 ≤ 𝑘 ≤ 3
n=0

X(k) = x(0)W40 + x(1)W4k + x(2)W42k + x(3)W43k

X(k) = 1 + 2W4k + W42k
X(k) = [4, −2j, 0,2j]
Let 𝑦(𝑛) = 𝑥((−𝑛) )4
𝑌 (𝑘) = 𝑋 (4 − 𝑘) = 𝑋 ∗ (𝑘) = X(4 − k) = X((−k))4
= {4,2𝑗, 0, −2𝑗}
DFT of circular time shifted sequence: If
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛) ↔ 𝑋(𝑘) Then
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥((𝑛 − 𝑙 ))𝑁 ↔ 𝑊𝑁𝑙𝑘 𝑋(𝑘)
Proof: The N-point IDFT of X(k) is 𝑥(𝑛)
N−1
1
x(n) = ∑ X(k)WN−nk 0 ≤ 𝑛 ≤ 𝑁 − 1
N
k=0

Put 𝑛 = 𝑛 − 𝑙
N−1
1 −(n−l)k
x(n − l) = 𝑥((𝑛 − 𝑙 ))𝑁 = ∑ X(k)WN
N
k=0
N−1
1
= ∑ X(k)WN−nk 𝑊𝑁𝑙𝑘
N
k=0
N−1
1
= ∑{X(k)WNlk }𝑊𝑁−𝑛𝑘 = X(k)𝑊𝑁𝑙𝑘
N
k=0
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
∴ 𝑥((𝑛 − 𝑙 ))𝑁 = ↔ 𝑊𝑁𝑙𝑘 𝑋(𝑘)
Problem: The 4-point DFT of x(n) is 𝑋(𝑘) = {1, 𝑗, −1, −𝑗}. Determine the DFTs of
i. 𝑥((𝑛 − 1))4
ii. 𝑥((𝑛 + 3))4
iii. 𝑥((𝑛 − 1))4 + 𝑥((𝑛 + 3))4

2
 Prepared by Dr Harish M S

Solution:
i. 𝑥((𝑛 − 1))4
Let 𝑦1 (𝑛) = 𝑥((𝑛 − 1))4
𝑌1 (𝑘) = 𝑊4𝑘 𝑋(𝑘)
𝑌1 (0) = 𝑊40 𝑋 (0) = 1 × 1 = 1
𝑌1 (1) = 𝑊41 𝑋 (1) = −𝑗 × 𝑗 = 1
𝑌1 (2) = 𝑊42 𝑋 (2) = −1 × −1 = 1
𝑌1 (3) = 𝑊43 𝑋 (3) = 𝑗 × −𝑗 = 1
𝑌1 (𝑘) = [1,1,1,1]
ii. 𝑥((𝑛 + 3))4
Let 𝑦2 (𝑛) = 𝑥((𝑛 + 3))4
𝑌2 (𝑘) = 𝑊4−3𝑘 𝑋 (𝑘)
𝑌2 (0) = 𝑊40 𝑋(0) = 1 × 1 = 1
𝑌2 (1) = 𝑊4−3 𝑋(1) = −j × j = 1
𝑌2 (2) = 𝑊4−6 𝑋 (2) = −1 × −1 = 1
𝑌2 (3) = 𝑊4−9 𝑋(3) = j × −j = 1
𝑌3 (𝑘) = [1,1,1,1]
iii. 𝑥((𝑛 − 1))4 + 𝑥((𝑛 + 3))4
Let 𝑦3 (𝑛) = 𝑥((𝑛 − 1))4 + 𝑥((𝑛 + 3))4
𝑌3 (𝑘) = 𝑊4𝑘 𝑋 (𝑘) + 𝑊4−3𝑘 𝑋 (𝑘)
𝑌3 (0) = 𝑊40 𝑋(0) + 𝑊40 𝑋(0) = 1 + 1 = 2
𝑌3 (1) = 𝑊41 𝑋 (1) + 𝑊4−3 𝑋 (1) = 1 + 1 = 2
𝑌3 (2) = 𝑊42 𝑋(2) + 𝑊4−6 𝑋(2) = 1 + 1 = 2
𝑌3 (3) = 𝑊43 𝑋 (3) + 𝑊4−9 𝑋 (3) = 1 + 1 = 2
Problem: Consider the sequence 𝑥(𝑛) = (0,1,2, .3,4)
i. Sketch the sequence 𝑠(𝑛) with 6-point DFT 𝑆(𝑘) = 𝑋(𝑘)𝑊2𝑘 0≤𝑘≤5
ii. Determine the sequence 𝑦(𝑛) with 6-point DFT 𝑌(𝑘) = 𝑅𝑒(𝑋 (𝑘)) 0 ≤ 𝑘 ≤ 5
iii. Determine the sequence 𝑣(𝑛) with 6-point DFT 𝑉(𝑘) = 𝐼𝑚(𝑋(𝑘)) 0 ≤ 𝑘 ≤ 5
Solution: i. 𝑆(𝑘) = 𝑋(𝑘)𝑊2𝑘 0≤𝑘≤5
𝑆(𝑘) = 𝑋(𝑘)𝑊3𝑘6 0≤𝑘≤5

3
 Prepared by Dr Harish M S

𝑠(𝑛) = 𝑥((𝑛 − 𝑙))𝑁

𝑠(𝑛) = 𝑥((𝑛 − 3))6 = 𝑥(3 + 𝑛) 0 ≤ 𝑛 ≤ 5
∴ 𝑠(𝑛) = (3,4,0,0,1,2)
ii. 𝑌(𝑘) = 𝑅𝑒(𝑋(𝑘)) 0 ≤ 𝑘 ≤ 5
1 1
𝑌 (𝑘) = 𝑅𝑒(𝑋 (𝑘 )) = {𝑋 (𝑘) + 𝑋 ∗ (𝑘)} = {𝑋 (𝑘) + 𝑋(𝑁 − 𝑘)}
2 2
1 1 1 1
𝑦(𝑛) = {𝑥(𝑛) + 𝑥 ((−𝑛))6 } = { 𝑥 (𝑛) + 𝑥(6 − 𝑛)} = { 0, , 3,3,3 }
2 2 2 2
iii. 𝑉(𝑘) = 𝐼𝑚(𝑋(𝑘)) 0 ≤ 𝑘 ≤ 5
1 1
𝑉 (𝑘) = 𝐼𝑚(𝑋 (𝑘)) = {𝑋 (𝑘) − 𝑋 ∗ (𝑘)} = {𝑋(𝑘) + 𝑋(𝑁 − 𝑘)}
2𝑗 2𝑗
1 1 1 3 3 31
𝑣 (𝑛 ) = {𝑥 (𝑛) + 𝑥 ((−𝑛))6 } = { 𝑥(𝑛) + 𝑥(6 − 𝑛)} = { 0, , , , }
2𝑗 2𝑗 2𝑗 𝑗 𝑗 𝑗 2𝑗
1 1
{ 0, −𝑗 , −3𝑗, −3𝑗, −3𝑗, −𝑗 }
2 2
1 0≤n≤3
Problem: A sequence x(n) = { has an 8-point DFTX(k). Compute the DFT of
0 4≤n≤7
𝑥2 (n)and 𝑥3 (n) in terms of 𝑋(𝑘) where
1 n=0 0 0≤n≤1
𝑥 2 (n) = { 0 1≤n≤4 And 𝑥 3 ( n) = { 1 2≤n≤5
1 5≤n≤7 0 6≤n≤7
Solution:x(n) = (1,1,1,1,0,0,0,0), the 8-point DFT of 𝑥(𝑛) is 𝑋(𝑘) = {4, 1 − 2.4𝑗, 0, 1 −
0.414𝑗, 0, 1 + 0.414𝑗, 0,1 + 2.4𝑗}

By observing the above figure3.1, 𝑥2 (n) = 𝑥((𝑛 − 5))8

𝑋2 (k) = 𝑋(𝑘)𝑊85𝑘

4
 Prepared by Dr Harish M S

𝑋2 (k) = (4, 1 + 2.4𝑗, 0, 1 + 0.414𝑗, 1 − 0.414𝑗, 0,1 − 2.414𝑗)

By observing the above figure 3.2, 𝑥3 (n) = 𝑥((𝑛 − 2))8

𝑋3 (k) = 𝑋(𝑘)𝑊82𝑘
𝑋3 (k) = (4, −2.414 − 𝑗, 0, 0.414 + 𝑗, 0, 0.414 − 𝑗, 0 , −2.414 + 𝑗
Problem: Consider the sequence 𝑥 (𝑛) = 𝛿 (𝑛) − 2𝛿 (𝑛 − 5). Find
i. The 10-point DFT of 𝑥(𝑛)
2𝑘
ii. The sequence y(n) that has a DFT 𝑌(𝑘) = 𝑊10 𝑋(𝑘)
iii. The sequence y (n) that has a DFT 𝑌(𝑘) = 𝑊(𝑘)𝑋(𝑘), where 𝑊(𝑘) is the 10-point
DFT of sequence 𝑤(𝑛) = 𝑢(𝑛) − 𝑢(𝑛 − 6) and 𝑋(𝑘) is the 10-point DFT of x (n).
Solution:
i. The 10-point of 𝑥(𝑛)
𝑥(𝑛) = 𝑥(𝑛) = 𝛿 (𝑛) − 2𝛿(𝑛 − 5) = (1,0,0,0,0,1,0,0,0,0)
10-point DFT of x (n) is X(k) = ∑N−1 nk
n=0 x(n)WN 0 ≤ 𝑘 ≤ 9
5k
X(k) = 1 + W10 0≤𝑘≤9
X(k) = (2 0 2 0 2 0 2 0 2 0)
2𝑘
ii. The sequence y(n) that has a DFT 𝑌(𝑘) = 𝑊10 𝑋(𝑘)
𝑦(𝑛) = 𝑥((𝑛 − 2))10
𝑦(𝑛) = 𝑥 (8 + 𝑛) = {0,0,1,0,0,0,01,0,0}
iii. The sequence y (n) that has a DFT 𝑌(𝑘) = 𝑊(𝑘)𝑋(𝑘), where 𝑊(𝑘) is the 10-point
DFT of sequence 𝑤(𝑛) = 𝑢(𝑛) − 𝑢(𝑛 − 6) and 𝑋(𝑘) is the 10-point DFT of x (n).
5k
X(k) = 1 + W10 0≤𝑘≤9
𝑤(𝑛) = 𝑢(𝑛) − 𝑢(𝑛 − 6) = (1, 1, 1, 1, 1, 1, 0 , 0, 0, 0)

5
 Prepared by Dr Harish M S

k 2k 3k 4k 5k
W(k) = 1 + W10 + W10 + W10 + W10 + W10
5k k 2k 3k 4k 5k
𝑌(𝑘) = 𝑊 (𝑘)𝑋(𝑘) = (1 + W10 )(1 + W10 + W10 + W10 + W10 + W10 )
k 2k 3k 4k 5k 5k 6k 7k 8k 9k 10k
= 1 + W10 + W10 + W10 + W10 + W10 + W10 + W10 + W10 + W10 + W10 + W10
k 2k 3k 4k 5k 5k 6k 7k 8k 9k
= 1 + W10 + W10 + W10 + W10 + 2W10 + W10 + W10 + W10 + W10 + W10 +1
k 2k 3k 4k 5k 5k 6k 7k 8k 9k
= 2 + W10 + W10 + W10 + W10 + W10 + W10 + W10 + W10 + W10 + W10
𝑦(𝑛) = (2,1,1,1,1,1,1,1,1,1)
DFT of circular frequency shifted sequence: If
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛) ↔ 𝑋(𝑘) Then
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛)𝑊𝑁−𝑙𝑛 ↔ 𝑋((𝑘 − 𝑙 ))𝑁
Proof: The N-point DFT of 𝑥(𝑛) is X(k)
N−1

X(k) = ∑ x(n)WNnk 0≤ 𝑘 ≤𝑁−1

n=0

Put 𝑘 = 𝑘 − 𝑙
N−1
n(k−l)
X(k − l) = ∑ x(n)WN
n=0
N−1

X((k − l))N = ∑{x(n) WN−ln }WNnk

n=0
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛)𝑊𝑁−𝑙𝑛 ↔ 𝑋((𝑘 − 𝑙 ))𝑁
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛)𝑊𝑁𝑙𝑛 ↔ 𝑋((𝑘 + 𝑙 ))𝑁
Problem: Compute the 4-point DFT of x(n) = (1,0,1,0). Also compute 𝑦(𝑛) if 𝑌(𝑘) =
𝑋((𝑘 − 2))4
Solution: X(k) = 1 + W42k
X(k) = (2,0,2,0)
If 𝑌 (𝑘) = 𝑋((𝑘 − 2))4 then, 𝑦(𝑛) = 𝑥(𝑛)𝑊𝑁−2𝑛
𝑦(0) = 𝑥(0)𝑊𝑁0 = 1
𝑦(1) = 𝑥(1)𝑊𝑁−2 = 0 × (−1) = 0
𝑦(2) = 𝑥(2)𝑊𝑁−4 = 1 × 1 = 1
𝑦(3) = 𝑥(3)𝑊𝑁−6 = 0 × (−) = 0

6
 Prepared by Dr Harish M S

𝑦(𝑛) = {1,0,1,0}
DFT of complex-conjugate sequence: If
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛) ↔ 𝑋(𝑘) Then
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥 ∗ (𝑛) ↔ 𝑋 ∗ ((−𝑘))𝑁 = 𝑋 ∗ (𝑁 − 𝑘)
𝐷𝐹𝑇{𝑥 ∗ (𝑛)} = ∑𝑁−1 ∗ 𝑛𝑘
𝑛=0 𝑥 (𝑛)𝑊𝑁 } 0≤𝑘 ≤𝑁−1
𝑁−1

= ∑ 𝑥 ∗ (𝑛)𝑊𝑁𝑛𝑘 𝑊𝑁−𝑛𝑁
𝑛=0
𝑁−1

= ∑{𝑥(𝑛)𝑊𝑁−𝑛𝑘 𝑊𝑁𝑛𝑁 }∗
𝑛=0
𝑁−1
(𝑁−𝑘)𝑛 ∗
∑{𝑥(𝑛)𝑊𝑁 }
𝑛=0

= 𝑋 ∗ (𝑁 − 𝑘)
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
∴ 𝑥 ∗ (𝑛) ↔ 𝑋 ∗ ((−𝑘))𝑁 = 𝑋 ∗ (𝑁 − 𝑘)
DFT of Multiplication of two sequences
1
Prove that DFT{x1 (n)x2 (n) = N X1 (k) ⊛ X2 (k)

Proof: The N-point DFT of x1 (n)x2 (n) is

N−1

DFT{x1 (n)x2 (n)} = ∑{x1 (n)x2 (n)}WNnk 0≤ 𝑘 ≤𝑁−1

n=0
1
Let x2 (n) = N ∑N−1 −nl
l=0 X 2 (l)WN 0 ≤ 𝑛 ≤ 𝑁 − 1

N−1 N−1
1
DFT{x1 (n)x2 (n)} = ∑{x1 (n)[ ∑ X2 (l)WN−nl ] WNnk
N
n=0 l=0

Interchange the order of summation we get

N−1 N−1 N−1
1 (k−l)n 1 1
= ∑ X 2 (l) ∑{x1 (n)WN = ∑ X2 (l) X1 ((k − l))N = X1 (l) ⊛ X2 (l)
N N N
l=0 n=0 l=0

Since l is the dummy variable, it can be replaced by k

1
∴ DFT{x1 (n)x2 (n) = X (k) ⊛ X2 (k)
N 1

7
 Prepared by Dr Harish M S

2πn
Problem: Compute the quantity∑N−1 ∗
n=0 x1 (n)x2 (n), where x1 (n) = x2 (n) = cos , 0≤𝑛≤
N

𝑁−1
2πn −j2πn 2πn j2πn
j −j
2πn 2πn ∗ e N +e N e N +e N
Solution:∑N−1 ∗
n=0 x1 (n)x2 (n) = ∑N−1
n=0 cos (cos ) = ∑N−1
n=0 ( )( )
N N 2 2
N−1 N−1 N−1 N−1
1 2π2n −j2π2n 1 1 2π2n 1 −j2π2n
= ∑ (1 + ej N + 1 + e N ) = ∑ 2 + ∑ ej N + ∑ e N
4 4 4 4
n=0 n=0 n=0 n=0
N−1 N−1 N−1
1 1 1 1 1
= ∑ 1 + ∑ WN−2n + ∑ WN2n = N + 0 + 0 = N
2 4 4 2 2
n=0 n=0 n=0
2πn
Problem: Compute the quantity∑N−1 ∗
n=0 x1 (n)x2 (n), where x1 (n) = cos and x2 (n) =
N
2πn
sin , 0≤ 𝑛 ≤ 𝑁−1
N
2πn −j2πn 2πn −j2πn
j j
2πn 2πn e N +e N e N −e N
Solution:∑N−1 ∗ N−1
n=0 x1 (n)x2 (n) = ∑n=0 cos ( sin N )∗ = ∑N−1
n=0 ( )( )
N 2 2j
N−1 N−1 N−1
1 2π2n −j2π2n 1 2π2n 1 −j2π2n
= ∑ (ej N − e N ) = ∑ ej N − ∑ e N
4j 4j 4j
n=0 n=0 n=0
N−1 N−1
1 1
∑ WN−2n − ∑ WN2n = 0 + 0 = 0
4j 4j
n=0 n=0

Parseval’s theorem (Inner product theorem or energy relation theorem)

For complex valued sequences 𝑥(𝑛) and 𝑦(𝑛), then
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑥(𝑛) ↔ 𝑋(𝑘)
𝑁−𝑃𝑜𝑖𝑛𝑡 𝐷𝐹𝑇
𝑦(𝑛) ↔ 𝑌(𝑘) Then
𝑁−1 𝑁−1
1
∑ 𝑥(𝑛)𝑦(𝑛 )∗ = ∑ 𝑋(𝑘)𝑌 ∗ (𝑘)
𝑁
𝑛=0 𝑘=0

Proof: Consider the summation

𝑁−1

∑ 𝑥(𝑛)𝑦(𝑛)∗
𝑛=0
1
We know that N-point IDFT of y (n) is 𝑦(𝑛) = 𝑁 ∑𝑁−1 −𝑛𝑘
𝑘=0 𝑌(𝑘)𝑊𝑁 , then
𝑁−1 𝑁−1 𝑁−1 ∗
1
∑ 𝑥 (𝑛)𝑦(𝑛)∗ = ∑ 𝑥(𝑛) ( ∑ 𝑌(𝑘)𝑊𝑁−𝑛𝑘 )
𝑁
𝑛=0 𝑛=0 𝑘=0

8
 Prepared by Dr Harish M S

𝑁−1 𝑁−1 𝑁−1

1
∑ 𝑥(𝑛)𝑦(𝑛)∗ = ∑ 𝑥(𝑛) ∑ 𝑌 ∗ (𝑘)𝑊𝑁𝑛𝑘
𝑁
𝑛=0 𝑛=0 𝑘=0

Interchange the order of summation, we get

𝑁−1 𝑁−1 𝑁−1 𝑁−1
1 1
∑ 𝑥 (𝑛)𝑦(𝑛)∗ = ∑ 𝑌 ∗ (𝑘) ∑ 𝑥(𝑛)𝑊𝑁𝑛𝑘 = ∑ 𝑌 ∗ (𝑘)𝑋(𝑘)
𝑁 𝑁
𝑛=0 𝑘=0 𝑛=0 𝑘=0
𝑁−1 𝑁−1
1
∴ ∑ 𝑥(𝑛)𝑦(𝑛)∗ = ∑ 𝑌 ∗ (𝑘)𝑋(𝑘)
𝑁
𝑛=0 𝑘=0

This is general expression for Parseval’s theorem. In the special case where𝑦(𝑛) = 𝑥(𝑛), then
𝑁−1 𝑁−1
1
∑ 𝑥(𝑛)𝑥(𝑛)∗ = ∑ 𝑋 ∗ (𝑘)𝑋(𝑘)
𝑁
𝑛=0 𝑘=0
𝑁−1 𝑁−1
1
∑ |𝑥(𝑛) |2 = ∑ |(𝑋(𝑘))|2
𝑁
𝑛=0 𝑘=0

Which express the energy in 𝑥(𝑛) in terms of the frequency component 𝑋(𝑘).
Problem: Let 𝑥 (𝑛) = (1,2,0,3, −2,4,7,5) with 8-point DFT 𝑋(𝑘) Evaluate the following
i. X(0)
ii. X(4)
iii. ∑7𝑘=0 𝑋(𝑘)
iv. ∑7𝑘=0|𝑋(𝑘)|2
Solution:
i. X(0)
7

𝑋(𝑘) = ∑ 𝑥(𝑛)𝑊8𝑛𝑘
𝑛=0
7

∴ 𝑋 (0) = ∑ 𝑥(𝑛) = 𝑥(0) + 𝑥(1) + 𝑥 (2) + 𝑥(3) + 𝑥(4) + 𝑥(5) + 𝑥(6) + 𝑥(7)
𝑛=0

𝑋(0) = 1 + 2 + 0 + 3 − 2 + 4 + 7 + 5 = 20
ii. X(4)
7

𝑋 (𝑘) = ∑ 𝑥(𝑛)𝑊8𝑛𝑘
𝑛=0

9
 Prepared by Dr Harish M S

7 7

𝑋 (4) = ∑ 𝑥 (𝑛)𝑊8𝑛4 = ∑ 𝑥(𝑛)(−1)𝑛

𝑛=0 𝑛=0

= 𝑥(0) − 𝑥(1) + 𝑥 (2) − 𝑥(3) + 𝑥(4) − 𝑥(5) + 𝑥(6) − 𝑥(7)}

𝑋 (4) = 1 − 2 + 0 − 3 − 2 − 4 + 7 − 5 = −8
iii. ∑7𝑘=0 𝑋(𝑘)
1
We know that 𝑥 (𝑛) = 8 ∑7𝑘=0 𝑋(𝑘)𝑊9−𝑛𝑘

If 𝑛 = 0, we get
7
1
𝑥(0) = ∑ 𝑋(𝑘)
8
𝑘=0
7

∑ 𝑋(𝑘) = 8𝑥(0) = 8 × 1 = 8
𝑘=0

iv. ∑7𝑘=0|𝑋(𝑘)|2
7 7
1
∑|𝑥(𝑛)|2 = ∑ |(𝑋(𝑘))|2
8
𝑛=0 𝑘=0
7

∴ ∑ |(𝑋(𝑘))|2
𝑛=0
7

= 8 ∑ |(𝑥(𝑛))|2
𝑛=0

= 8(𝑥(0)2 + 𝑥(1)2 + 𝑥(2)2 + 𝑥(3)2 + 𝑥(4)2 + 𝑥(5)2 + 𝑥(6)2 + 𝑥(7)2 )

8{(1)2 + (2)2 + (0)2 + (3)2 + (−2)2 + (4)2 + (7)2 + (5)2 } = 864
Problem: let 𝑋(𝑘) be the 14-point DFT of length 14 real sequences 𝑥(𝑛). The First 8-samples of
𝑋(𝑘) are given by
𝑋(0) = 12
𝑋(1) = −1 + 𝑗3
𝑋(2) = 3 + 𝑗4
𝑋(3) = 1 − 𝑗5
𝑋(4) = −2 + 𝑗2
𝑋(5) = 6 + 𝑗3
𝑋(6) = −2 − 3𝑗

10
 Prepared by Dr Harish M S

𝑋(7) = 10
Find the remaining 6-samples of 𝑋(𝑘). And also find
a. 𝑥(0)
b. 𝑥(7)
c. ∑13
𝑛=0 𝑥(𝑛)

d. ∑13
𝑛=0|𝑥(𝑛)|
2

Solution:
Since 𝑥(𝑛) is real
𝑋(𝑘) = 𝑋 ∗ (𝑁 − 𝑘)
𝑋 (𝑘) = 𝑋 ∗ (14 − 𝑘)0 ≤ 𝑘 ≤ 13
𝑋 (8) = 𝑋 ∗ (6) = −2 + 3𝑗
𝑋(9) = 𝑋 ∗ (5) = 6 − 𝑗3
𝑋(10) = 𝑋 ∗ (4) = −2 − 2𝑗
𝑋 (11) = 𝑋 ∗ (3) = 1 + 5𝑗
𝑋 (12) = 𝑋 ∗ (2) = 3 − 4𝑗
𝑋(13) = 𝑋 ∗ (1) = −1 − 3𝑗
1
a. 𝑥(𝑛) = 𝑁 ∑𝑁−1 −𝑛𝑘
𝑘=0 𝑋(𝑘)𝑊𝑁

13
1
𝑥 (0) = ∑ 𝑋 (𝑘 )
14
𝑘=0

1
= {𝑋(0) + 𝑋 (1) + 𝑋 (2) + 𝑋(3) + 𝑋(4) + 𝑋 (5) + 𝑋 (6) + 𝑋(7) + 𝑋(8)
14
+ 𝑋 (9) + 𝑋 (10) + 𝑋 (11) + 𝑋 (12) + 𝑋 (13)}
1
= {12 + 1 + 3𝑗 + 3 + 𝑗4 + 1 − 𝑗5 + −2 + 𝑗2 + 6 + 𝑗3 − 2 − 3𝑗 + 10 − 2 + 3𝑗 + 6 − 𝑗3
14
− 2 − 2𝑗 + 1 + 5𝑗 + 3 − 4𝑗 + −1 − 3𝑗} = 2.2853
1
b. 𝑥(𝑛) = 𝑁 ∑𝑁−1 −𝑛𝑘
𝑘=0 𝑋(𝑘)𝑊𝑁

13 13
1 −7𝑘
1
𝑥 (7) = ∑ 𝑋(𝑘)𝑊14 = ∑ 𝑋(𝑘)(−1)𝑘
14 14
𝑘=0 𝑘=0

11
 Prepared by Dr Harish M S

13
1
= ∑ 𝑋(𝑘)(−1)𝑘
14
𝑘=0

1
= {𝑋(0) − 𝑋 (1) + 𝑋 (2) − 𝑋(3) + 𝑋(4) − 𝑋 (5) + 𝑋 (6) − 𝑋(7) + 𝑋(8)
14
− 𝑋 (9) + 𝑋 (10) − 𝑋 (11) + 𝑋 (12) − 𝑋 (13)}
1
= {12 − 1 − 3𝑗 + 3 + 𝑗4 − 1 + 𝑗5 − 2 + 𝑗2 − 6 − 𝑗3 − 2 − 3𝑗 − 10 − 2 + 3𝑗 − 6 + 𝑗3 − 2
14
− 2𝑗 − 1 − 5𝑗 + 3 − 4𝑗 + 1 + 3𝑗} = −0.8571
c. ∑13
𝑛=0 𝑥(𝑛)
13

𝑋(𝑘) = ∑ 𝑥(𝑛)𝑊𝑁𝑛𝑘
𝑛=0
13

𝑋(0) = ∑ 𝑥(𝑛)
𝑛=0
13

∑ 𝑥(𝑛) = 12
𝑛=0

d. ∑13
𝑛=0|𝑥(𝑛)|
2

13 13 13
1 1
∑ |𝑥(𝑛) |2 = ∑|(𝑋(𝑘))|2 = ∑ 𝑋(𝑘)𝑋 ∗ (𝑘) =
14 14
𝑛=0 𝑘=0 𝑘=0
13 13 13
1 1
∑ |𝑥(𝑛) |2 = ∑|(𝑋(𝑘))|2 = ∑ 𝑋(𝑘)𝑋 ∗ (𝑘) =
14 14
𝑛=0 𝑘=0 𝑘=0
13
1 1
= ∑ 𝑋 (𝑘)𝑋 ∗ (𝑘) = {𝑋 (0)𝑋 ∗ (0) + 𝑋 (1)𝑋 ∗ (1) + 𝑋(2)𝑋 ∗ (2) + 𝑋(3)𝑋 ∗ (3) + 𝑋 (4)𝑋 ∗ (4)
14 14
𝑘=0

+ 𝑋 (5)𝑋 ∗ (5) + 𝑋 (6)𝑋 ∗ (6) + 𝑋 (7)𝑋 ∗ (7) + 𝑋 (8)𝑋 ∗ (8) + 𝑋 (9)𝑋 ∗ (9)
+ 𝑋(10)𝑋 ∗ (10) + 𝑋 (11)𝑋 ∗ (11) + 𝑋(12)𝑋 ∗ (12) + 𝑋 (13)𝑋 ∗ (13)}
1
{12 × 12 + (−12 + 32 ) + (32 + 42 ) + (12 + 52 ) + (−22 + 22 ) + (62 + 32 )
14
+ (−22 + −33 ) + 10 × 10 + (−22 + 32 ) + (62 + −32 ) + (−22 + −22 )
+ (12 + −52 ) + (32 + −42 ) + (−12 + −32 )} = 35.51
Problem: If DFT and IDFT operators are cascaded, what is the overall output?
Solution:

12
 Prepared by Dr Harish M S

𝑁−1

𝑋(𝑘) = ∑ 𝑥(𝑚)𝑊𝑁𝑚𝑘
𝑚=0
𝑁−1
1
𝑥 ′ (𝑛) = ∑ 𝑋(𝑘)𝑊𝑁−𝑛𝑘
𝑁
𝑘=0
𝑁−1 𝑁−1
1
𝑥 ′ (𝑛) = ∑{ ∑ 𝑥(𝑚)𝑊𝑁𝑚𝑘 } 𝑊𝑁−𝑛𝑘
𝑁
𝑘=0 𝑚=0

Interchange the order of summation we get

𝑁−1 𝑁−1
1
𝑥 ′ (𝑛) = ∑ 𝑥(𝑚) ∑ 𝑊𝑁𝑚𝑘 𝑊𝑁−𝑛𝑘
𝑁
𝑚=0 𝑘=0
𝑁−1 𝑁−1
′(
1 (𝑚−𝑛)𝑘
𝑥 𝑛) = ∑ 𝑥(𝑚) ∑ 𝑊𝑁
𝑁
𝑚=0 𝑘=0
𝑁−1
(𝑚−𝑛)𝑘 𝑁 𝑖𝑓 𝑚 = 𝑛
𝑥(𝑚) ∑ 𝑊𝑁 = {
0 𝑚≠𝑛
𝑘=0
𝑁−1
′(
1
∴ 𝑥 𝑛) = ∑ 𝑥 (𝑚)𝑁𝛿(𝑚 − 𝑛) = 𝑥(𝑛)
𝑁
𝑚=0

𝑥 ′ (𝑛) = 𝑥(𝑛)
Problem: If the two DFT operators are cascaded, what is the overall output?
Solution:

𝑁−1

𝑋(𝑘) = ∑ 𝑥(𝑚)𝑊𝑁𝑚𝑘
𝑚=0

13
 Prepared by Dr Harish M S

𝑁−1

𝑥 ′ (𝑛) = ∑ 𝑋 (𝑘)𝑊𝑁𝑛𝑘
𝑘=0
𝑁−1 𝑁−1
′(
𝑥 𝑛) = ∑{ ∑ 𝑥(𝑚)𝑊𝑁𝑚𝑘 } 𝑊𝑁𝑛𝑘
𝑘=0 𝑚=0

Interchange the order of summation we get

𝑁−1 𝑁−1
′(
𝑥 𝑛) = ∑ 𝑥(𝑚) ∑ 𝑊𝑁𝑚𝑘 𝑊𝑁𝑛𝑘
𝑚=0 𝑘=0
𝑁−1 𝑁−1
′( (𝑚∓𝑛)𝑘
𝑥 𝑛) = ∑ 𝑥(𝑚) ∑ 𝑊𝑁
𝑚=0 𝑘=0
𝑁−1
(𝑚+𝑛)𝑘 𝑁 𝑖𝑓 𝑚 = −𝑛
𝑥(𝑚) ∑ 𝑊𝑁 = {
0 𝑚 ≠ −𝑛
𝑘=0
𝑁−1
′(
∴ 𝑥 𝑛) = ∑ 𝑥(𝑚)𝑁𝛿(𝑚 + 𝑛) = 𝑁𝑥((−𝑛))𝑁
𝑚=0

𝑥 ′ (𝑛) = 𝑁𝑥((−𝑛))𝑁
Problem: Let 𝑥(𝑛) be a length N with N-point DFT 𝑋(𝑘). The DFT operation is represented

as 𝑋(𝑘 = 𝐹{𝑥(𝑛)}, determine 𝑦(𝑛) = 𝐹{𝐹 {𝐹{𝐹{𝑥(𝑛)}}} }

Solution:𝑦(𝑛) = 𝐹 {𝐹 {𝐹{𝐹{𝑥(𝑛)}}}} = 𝐷𝐹𝑇 {𝐷𝐹𝑇 {𝐷𝐹𝑇{𝐷𝐹𝑇{𝑥(𝑛)}}}} = 𝑁 2 𝑥(𝑛)

Linear Filtering Methods based on the DFT

Use of DFT in linear filtering (Use of DFT in linear convolution)
Suppose that we have a finite duration sequence 𝑥(𝑛) of length L which excites an FIR filter
having an impulse response ℎ(𝑛) of length M. The output of 𝑦(𝑛) of the filter in the time domain
can be expressed as a convolution of 𝑥(𝑛) and ℎ(𝑛), that is

𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛) In fact the 𝑦(𝑛), consists of 𝑀 + 𝐿 − 1 samples. In the frequency domain,
the same linear convolution 𝑦(𝑛) can be obtained by performing N-point 𝐷𝐹𝑇 for which 𝑁 ≥
𝑀 + 𝐿 − 1, that is
𝑌 (𝑘) = 𝑋 (𝑘)𝐻(𝑘)

14
 Prepared by Dr Harish M S

𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛) = 𝐼𝐷𝐹𝑇{𝑌(𝑘)} = 𝐼𝐷𝐹𝑇{𝑋 (𝑘)𝐻(𝑘)}

In turn this implies that the N-point circulation of 𝑥(𝑛) with ℎ(𝑛) must be equivalent to the
linear convolution of 𝑥(𝑛) with ℎ(𝑛).
Problem: By means of DFT and IDFT, determine the response of the FIR filter with impulse
1, 2, 3 1, 2, 2 ,3
response ℎ(𝑛) = { } to the input sequence 𝑥 (𝑛) = { ].
↑ ↑
Solution: 𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛), 𝑁 ≥ 𝑀 + 𝐿 − 1, 𝑁 ≥ 4 + 3 − 1 = 6
1, 2, 3, 0,0,0 1, 2, 2 ,1,0 ,0
ℎ (𝑛 ) = { } , 𝑥 (𝑛 ) = { }
↑ ↑
𝐻 (𝑘) = 1 + 2𝑊6𝑘 + 3𝑊62𝑘 And 𝑋(𝑘) = 1 + 2𝑊6𝑘 + 2𝑊62𝑘 + 𝑊63𝑘
𝑌(𝑘) = 𝑋(𝑘)𝐻 (𝑘) = (1 + 2𝑊6𝑘 + 3𝑊62𝑘 )(1 + 2𝑊6𝑘 + 2𝑊62𝑘 + 𝑊63𝑘 )
𝑌 (𝑘) = 1 + 2𝑊6𝑘 + 2𝑊62𝑘 + 𝑊63𝑘 + 2𝑊6𝑘 + 4𝑊42𝑘 + 4𝑊63𝑘 + 2𝑊64𝑘 + 3𝑊62𝐾 + 6𝑊63𝑘
+ 6𝑊64𝑘 + 3𝑊65𝑘
𝑌(𝑘) = 1 + 4𝑊6𝑘 + 9𝑊62𝑘 + 11𝑊63𝑘 + 8𝑊64𝑘 + 3𝑊65𝑘
y(n) = {1,4,9,11, 8,3}
Filtering of long data sequences
In practical applications a linear filtering operation is performed on the long input
sequence 𝑥(𝑛). This linear filtering operation is performed using the DFT and IDFT by
segmenting a long data sequences into fixed size blocks prior to the filtering operation. Since the
filtering is linear, successive blocks can be processed one at a time via DFT, and the output
blocks are fitted together to form the overall output signal. Two methods called the overlap-save
method and the overlap-add method. For both methods we assume that the FIR filter impulse
response ℎ(𝑛) has duration M. the input data sequence is segmented into blocks of L points,
where by assumption, 𝐿 >> 𝑀.
Overlap-save method

Let the length of FIR filter impulse response ℎ(𝑛) be M. The input long data sequence 𝑥(𝑛) is
segmented into number of blocks 𝑥1 (𝑛), 𝑥2 (𝑛), 𝑥3 (𝑛), … … each blocks consists of 𝑁 = 𝐿 +
𝑀 − 1 and 𝐿 >> 𝑀. The first (𝑀 − 1) samples in 𝑥1 (𝑛) are zeros and L samples are from 𝑥(𝑛) .

15
 Prepared by Dr Harish M S

For remaining data blocks such as𝑥2 (𝑛), 𝑥3 (𝑛), … … , the first (𝑀 − 1) sequences are from
previous blocks followed by L new data points from 𝑥(𝑛) as shown in below figure.

And

Now we have to perform N-point circular convolution 𝑥1 (𝑛) with ℎ1 (𝑛), 𝑥2 (𝑛) with
ℎ1 (𝑛), 𝑥3 (𝑛) with ℎ1 (𝑛) and so on using DFT and IDFT method. We obtain the
output 𝑦1 (𝑛) = 𝑥1 (𝑛) ⊛ ℎ1 (𝑛), 𝑦2 (𝑛) = 𝑥2 (𝑛) ⊛ ℎ1 (𝑛), 𝑦3 (𝑛) = 𝑥3 (𝑛) ⊛ ℎ1 (𝑛) and so
on. To obtain the final output 𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛), we have to combine 𝑦1 (𝑛), 𝑦2 (𝑛), 𝑦3 (𝑛)
and so on as follows

16
 Prepared by Dr Harish M S

Problem: Find the output y(n) of the filter whose impulse response ℎ(𝑛) = {1,1,1} and the input
signal 𝑥(𝑛) = (3, −1, 0, 1, 3, 2, 0, 1, 2, 1) using overlap save method.
Solution: Output 𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛), 𝑀=3
𝑁 = 𝐿 + 𝑀 − 1, 𝐿 ≥ 𝑀, 𝐿=3
∴ 𝑁 = 3 + 3 − 1 = 5, thus we have to perform 𝑁 = 5 point circular convolution. The signal
𝑥(𝑛) is subdivide into 𝑥1 (𝑛) = (0,0, 3, −1,0 ), 𝑥2 (𝑛) = (−1,0, 1, 3, 2), 𝑥3 (𝑛) = (3, 2, 0, 1,2)
and 𝑥4 (𝑛) = (1,2,1,0,0). The ℎ(𝑛) is modified to ℎ1 (𝑛) = (1, 1, 1, 0, 0).
𝑦1 (𝑛) = 𝑥1 (𝑛) ⊛ ℎ1 (𝑛)
𝑌1 (𝑘 ) = 𝑋1 (𝑘)𝐻1 (𝑘) = (3𝑊52𝑘 − 𝑊53𝑘 )(1 + 𝑊5𝑘 + 𝑊52𝑘 ) = 3𝑊52𝑘 + 2𝑊53𝑘 + 2𝑊54𝑘 − 𝑊55𝑘
𝑌1 (𝑘) = −1 + 3𝑊52𝑘 + 2𝑊53𝑘 + 2𝑊54𝑘
∴ 𝑦1 (𝑛) = (−1,0,3,2,2)
𝑦2 (𝑛) = 𝑥2 (𝑛) ⊛ ℎ1 (𝑛)
𝑌2 (𝑘) = 𝑋2 (𝑘)𝐻1 (𝑘 ) = (−1 + 𝑊52𝑘 + 3𝑊53𝑘 + 2𝑊54𝑘 )(1 + 𝑊5𝑘 + 𝑊52𝑘 )
= 4 + 𝑊5𝑘 + 4𝑊53𝑘 + 6𝑊54𝑘
∴ 𝑦2 (𝑛) = (4,1,0,4,6)
𝑦3 (𝑛) = 𝑥3 (𝑛) ⊛ ℎ1 (𝑛)
𝑌3 (𝑘) = 𝑋3 (𝑘)𝐻1 (𝑘 ) = (3 + 2𝑊5𝑘 + 𝑊53𝑘 + 2𝑊54𝑘 )(1 + 𝑊5𝑘 + 𝑊52𝑘 )
= 6 + 7𝑊5𝑘 + 5𝑊52𝑘 + 3𝑊53𝑘 + 3𝑊54𝑘
∴ 𝑦3 (𝑛) = (6, 7, 5, 3, 3)

17
 Prepared by Dr Harish M S

𝑌4 (𝑘) = 𝑋4 (𝑘)𝐻1 (𝑘) = (1 + 2𝑊5𝑘 + 𝑊52𝑘 )(1 + 𝑊5𝑘 + 𝑊52𝑘 )

= 1 + 3𝑊5𝑘 + 4𝑊52𝑘 + 3𝑊53𝑘 + 𝑊54𝑘
∴ 𝑦4 (𝑛) = (1, 3, 4, 3, 1)
𝑦(𝑛) = (3,2,2,0,4,6,5,3,3,4,3,1)
Problem: A long data sequence 𝑥(𝑛) is filtered through a filter with impulse response ℎ(𝑛) to
yield the output 𝑦(𝑛). If the input signal 𝑥(𝑛) = (1,1,1,1,1,3,1,1,1,4,2,1,1,3,1) and ℎ(𝑛) =
(1, −1). Compute 𝑦(𝑛) using overlap save method use only a 5-point circular convolution in
your approach.
Solution: Output 𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛), 𝑀=2
𝑁 =5 =𝐿+𝑀−1 =𝐿+2−1, 𝐿=4
Thus we have to perform 𝑁 = 5 point circular convolution. The signal 𝑥(𝑛) is subdivide
into 𝑥1 (𝑛) = (0,1,1,1,1 ), 𝑥2 (𝑛) = (1,1,3,1,1), 𝑥3 (𝑛) = (1,4,2,1,1) and 𝑥4 (𝑛) = (1,3,1,0,0),
and ℎ(𝑛) is modified to ℎ1 (𝑛) = (1, −1, 0, 0, 0).
𝑦1 (𝑛) = 𝑥1 (𝑛) ⊛ ℎ1 (𝑛)
𝑌1 (𝑘) = 𝑋1 (𝑘)𝐻1 (𝑘) = (𝑊5𝑘 + 𝑊52𝑘 + 𝑊53𝑘 + 𝑊54𝑘 )(1 − 𝑊5𝑘 ) = −1 + 𝑊5𝑘
𝑦1 (𝑛) = (1, −1,0,0,0)
𝑦2 (𝑛) = 𝑥2 (𝑛) ⊛ ℎ1 (𝑛)
𝑌2 (𝑘) = 𝑋2 (𝑘)𝐻1 (𝑘) = (1 + 𝑊5𝑘 + 3𝑊52𝑘 + 𝑊53𝑘 + 𝑊54𝑘 )(1 − 𝑊5𝑘 ) = 2𝑊52𝑘 − 2𝑊53𝑘
𝑦2 (𝑛) = (0,0,2, −2,0)
𝑦3 (𝑛) = 𝑥3 (𝑛) ⊛ ℎ1 (𝑛)
𝑌3 (𝑘) = 𝑋3 (𝑘)𝐻1 (𝑘) = (1 + 4𝑊5𝑘 + 2𝑊52𝑘 + 𝑊53𝑘 + 𝑊54𝑘 )(1 − 𝑊5𝑘 ) = 3𝑊5𝑘 − 2𝑊52𝑘 − 𝑊53𝑘
𝑦3 (𝑛) = (0,3, −2, −1,0)
𝑦4 (𝑛) = 𝑥3 (𝑛) ⊛ ℎ1 (𝑛)
𝑌4 (𝑘) = 𝑋4 (𝑘)𝐻1 (𝑘) = (1 + 3𝑊5𝑘 + 𝑊52𝑘 )(1 − 𝑊5𝑘 ) = 1 + 2𝑊5𝑘 − 2𝑊52𝑘 − 𝑊53𝑘
𝑦4 (𝑛) = (1,2, −2, −1,0)
𝑦(𝑛) = (1,0,0,0,0,2, −2,0,3, −2, −1,0,2, −2, −1,0)
Overlap-add method

18
 Prepared by Dr Harish M S

Let the length of FIR filter impulse response ℎ(𝑛) be M. The input long data sequence 𝑥(𝑛) is
segmented into number of blocks 𝑥1 (𝑛), 𝑥2 (𝑛), 𝑥3 (𝑛), … … each blocks consists of 𝑁 = 𝐿 +
𝑀 − 1 and 𝐿 >> 𝑀. The first L samples in 𝑥1 (𝑛)𝑥1 (𝑛), 𝑥2 (𝑛), 𝑥3 (𝑛), … … are from 𝑥(𝑛) and
last (𝑀 − 1) samples are zeros as shown in below figure.

And

Now we have to perform N-point circular convolution 𝑥1 (𝑛) with ℎ1 (𝑛), 𝑥2 (𝑛) with
ℎ1 (𝑛)𝑥3 (𝑛) with ℎ1 (𝑛) and so on using DFT and IDFT method. We obtain the output 𝑦1 (𝑛) =
𝑥1 (𝑛) ⊛ ℎ1 (𝑛), 𝑦2 (𝑛) = 𝑥2 (𝑛) ⊛ ℎ1 (𝑛), 𝑦3 (𝑛) = 𝑥3 (𝑛) ⊛ ℎ1 (𝑛) and so on. To obtain the
final output 𝑦(𝑛) = 𝑥(𝑛) ⊛ ℎ(𝑛), we have to combine 𝑦1 (𝑛), 𝑦2 (𝑛), 𝑦3 (𝑛) and so on as
follows

19
 Prepared by Dr Harish M S

Problem: An FIR filter has a unit impulse response ℎ(𝑛) = (2, 2, 1). Determine the output
sequence in response to the input sequence 𝑥(𝑛) = (3,0,2,0,2,1,0, −2, −2,0) using over lap add
fast convolution technique. Use 8-point circular convolution.
Solution: Output 𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛), 𝑀=3
𝑁 = 8 = 𝐿 + 3 − 1, 𝐿 = 6
Thus we have to perform 𝑁 = 8 point circular convolution. The signal 𝑥(𝑛) is subdivide
into 𝑥1 (𝑛) = ( 3,0, 2,0,2,1,0,0 ) 𝑎𝑛𝑑 𝑥2 (𝑛) = (0, −2, −2,0,0,0,0,0) and ℎ(𝑛) is modified
to ℎ1 (𝑛) = (2,2,1,0,0,0,0,0).
𝑦1 (𝑛) = 𝑥1 (𝑛) ⊛ ℎ1 (𝑛)
𝑌1 (𝑘) = 𝑋1 (𝑘)𝐻1 (𝑘 ) = (3 + 2𝑊82𝑘 + 2𝑊84𝑘 + 𝑊85𝑘 )(2 + 2𝑊8𝑘 + 𝑊82𝑘 )
𝑌1 (𝑘) = 6 + 6𝑊8𝑘 + 7𝑊82𝑘 + 4𝑊83𝑘 + 6𝑊84𝑘 + 6𝑊85𝑘 + 4𝑊86𝑘 + 𝑊87𝑘
∴ 𝑦1 (𝑛) = (6,6,7,4,6,6,4,1)
𝑦2 (𝑛) = 𝑥2 (𝑛) ⊛ ℎ1 (𝑛)
𝑌2 (𝑘) = 𝑋2 (𝑘)𝐻1 (𝑘) = (−2𝑊8𝑘 − 2𝑊82𝑘 )(2 + 2𝑊8𝑘 + 𝑊82𝑘 )
𝑌2 (𝑘) = −4𝑊8𝑘 − 8𝑊82𝑘 − 6𝑊83𝑘 − 2𝑊84𝑘
∴ 𝑦2 (𝑛) = (0, −4, −8, −6, −2,0,0,0)

20
 Prepared by Dr Harish M S

Problem: Find the output of the filter whose impulse response ℎ(𝑛) = (1, −2) and the input
signal 𝑥(𝑛) = (3, −2,4,1,5,7,2, −9) using over lap add method. Use only 5-point circular
convolution.
Solution: Output 𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛), 𝑀=2
𝑁 = 5 = 𝐿 + 2 − 1, 𝐿 = 4
Thus we have to perform 𝑁 = 4 point circular convolution. The signal 𝑥(𝑛) is subdivide
into 𝑥1 (𝑛) = ( 3, −2,4,1,0 )𝑥2 (𝑛) = (5,7,2, −9,0), and ℎ(𝑛) is modified to ℎ1 (𝑛) =
(1, −2,0,0,0).
𝑦1 (𝑛) = 𝑥1 (𝑛) ⊛ ℎ1 (𝑛)
𝑌1 (𝑘) = 𝑋1 (𝑘)𝐻1 (𝑘) = (3 − 2𝑊5𝑘 + 4𝑊52𝑘 + 𝑊53𝑘 )(1 − 2𝑊8𝑘 )
𝑌1 (𝑘) = 3 − 8𝑊5𝑘 + 8𝑊52𝑘 − 7𝑊53𝑘 − 2𝑊54𝑘
∴ 𝑦1 (𝑛) = (3, −8,8, −7, −2)
𝑦2 (𝑛) = 𝑥2 (𝑛) ⊛ ℎ1 (𝑛)
𝑌2 (𝑘) = 𝑋2 (𝑘)𝐻1 (𝑘) = (5 + 7𝑊5𝑘 + 2𝑊52𝑘 − 9𝑊53𝑘 )(1 − 2𝑊8𝑘 )
𝑌2 (𝑘) = 5 − 3𝑊5𝑘 − 12𝑊52𝑘 − 13𝑊53𝑘 + 18𝑊54𝑘
∴ 𝑦2 (𝑛) = (5, −3, −12, −13,18)

Problem: Find the output 𝑦(𝑛) of the filter whose impulse response ℎ(𝑛) = {1,1,1} and the
input signal 𝑥(𝑛) = (3, −1, 0, 1, 3, 2, 0, 1, 2, 1) using overlap add method.
Solution: Output 𝑦(𝑛) = 𝑥(𝑛) ∗ ℎ(𝑛), 𝑀=3
𝑁 = 𝐿 + 𝑀 − 1, 𝐿 ≥ 𝑀, 𝐿=3
∴ 𝑁 = 3 + 3 − 1 = 5,
Thus we have to perform 𝑁 = 5 point circular convolution. The signal 𝑥(𝑛) is subdivide
into 𝑥1 (𝑛) = (3, −1,0,0,0), 𝑥2 (𝑛) = (1,3,2,0,0),𝑥3 (𝑛) = (0,1,2,0,0) and 𝑥4 (𝑛) = (1,0,0,0,0).
The ℎ(𝑛) is modified to ℎ1 (𝑛) = (1, 1, 1, 0, 0).
𝑦1 (𝑛) = 𝑥1 (𝑛) ⊛ ℎ1 (𝑛)
𝑌1 (𝑘) = 𝑋1 (𝑘)𝐻1 (𝑘) = (3 − 𝑊5𝑘 )(1 + 𝑊5𝑘 + 𝑊52𝑘 )

21
 Prepared by Dr Harish M S

𝑌1 (𝑘) = 3 + 2𝑊5𝑘 + 2𝑊52𝑘 − 𝑊53𝑘

∴ 𝑦1 (𝑛) = (3,2,2, −1,0)
𝑦2 (𝑛) = 𝑥2 (𝑛) ⊛ ℎ1 (𝑛)
𝑌2 (𝑘) = 𝑋2 (𝑘)𝐻1 (𝑘) = (1 + 3𝑊5𝑘 + 2𝑊52𝑘 )(1 + 𝑊5𝑘 + 𝑊52𝑘 )
= 1 + 4𝑊5𝑘 + 6𝑊52𝑘 + 5𝑊53𝑘 + 2𝑊54𝑘
∴ 𝑦2 (𝑛) = (1,4,6,5,2)
𝑦3 (𝑛) = 𝑥3 (𝑛) ⊛ ℎ1 (𝑛)
𝑌3 (𝑘) = 𝑋3 (𝑘)𝐻1 (𝑘) = (𝑊5𝑘 + 2𝑊52𝑘 )(1 + 𝑊5𝑘 + 𝑊52𝑘 ) = 𝑊5𝑘 + 3𝑊52𝑘 + 3𝑊53𝑘 2𝑊54𝑘
∴ 𝑦3 (𝑛) = (0,1,3,3,2)
𝑌4 (𝑘) = 𝑋4 (𝑘)𝐻1 (𝑘) = (1 )(1 + 𝑊5𝑘 + 𝑊52𝑘 ) = (1 + 𝑊5𝑘 + 𝑊52𝑘 )
∴ 𝑦4 (𝑛) = (1, 1,1,0,0,0)

Efficient Computation of the DFT- FFT Algorithms:

Fast Fourier Transform (FFT) Algorithms

Efficient computation of DFT:
The direct computation of DFT involves large number of computations. Consider the direction
computation of N-point DFT of 𝑥(𝑛)

22
 Prepared by Dr Harish M S

𝑁−1

𝑋 (𝑘) = ∑ 𝑥(𝑛)𝑊𝑁𝑛𝑘 0 ≤ 𝑘 ≤ 𝑁 − 1
𝑛=0

The sequence 𝑥(𝑛) may be real or complex and 𝑊𝑁𝑘 is a complex number, thus direct
computation of DFT involves the complex multiplication and complex addition i.e.
(𝑁−𝑘)𝑘
𝑋(𝑘) = 𝑥(0)𝑊𝑁0 + 𝑥(1)𝑊𝑁𝑘 + 𝑥(2)𝑊𝑁2𝑘 + ⋯ . . +𝑥(𝑁 − 1)𝑊𝑁
Therefore, the N-point DFT computation involves (𝑁 − 1) complex addition and N complex
multiplication for one value of 𝑋(𝑘). For N values of 𝑋(𝑘) it needs 𝑁(𝑁 − 1) complex additions
and N× 𝑁 = 𝑁 2 complex multiplications.
In one complex addition, 2 real additions are required i.e. (𝑎 + 𝑗𝑏) + (𝑐 + 𝑗𝑑 ) = (𝑎 + 𝑐 ) +
𝑗(𝑏 + 𝑑).
In one complex multiplication 4 real multiplications and 2 real additions are required i. e
(𝑎 + 𝑗𝑏)(𝑐 + 𝑗𝑑 ) = (𝑎𝑐 − 𝑏𝑑 ) + 𝑗(𝑎𝑑 + 𝑏𝑐)
Therefore for N-point DFT computation
Numbers of real additions are 2𝑁(𝑁 − 1) + 2𝑁 2 = 4𝑁 2 − 2𝑁 = 2𝑁(2𝑁 − 1)
Numbers of real multiplications are 4𝑁 2
Thus in direct computation of N-point DFT we require
1. 𝑁 2 Complex multiplications.
2. 𝑁(𝑁 − 1) Complex additions.
3. 2N (2N-1) Real additions.
4. 4𝑁 2 Real multiplications.
The direct computation of the DFT is basically inefficient. The Fast Fourier Transform (FFT)
algorithm increases the computational efficiency using the symmetry and periodic properties of
twiddle factor
The efficient computation of DFT can be achieved using Fast Fourier Transform (FFT)
algorithm. The Fast Fourier Transform (FFT) algorithm increases the computational efficiency
using the symmetry and periodicity properties of twiddle factor 𝑊𝑁 . In particular, these two
properties are:
𝑁
𝑘+
Symmetry property: 𝑊𝑁 2
= −𝑊𝑁𝑘
Periodicity property: 𝑊𝑁𝑘+𝑁 = 𝑊𝑁𝑘

23
 Prepared by Dr Harish M S

For N-point DFT computation by using FFT algorithm, number of complex multiplications
𝑁
required are 2 log 2 𝑁 and number of complex additions required are 𝑁 log 2 𝑁. The table 3.1

shows the comparison of computational complexity of direct DFT and FFT algorithm
Table 3.1: Comparison between direct DFT computation and DFT using FFT algorithm
DFT Using direct computation DFT Using FFT algorithm
points Complex Complex Complex Complex
N multiplication 𝑁 2 addition multiplication addition Speed
𝑁(𝑁 − 1) 𝑁 𝑁 log10 𝑁 𝑁 log 2 𝑁 improvement
log 2 𝑁 = ( log )
2 2 22
𝑁2
𝑁
log2 𝑁
2

8 64 56 12 24 5.3
16 256 240 32 64 8
256 6553 65280 1024 1048 64
1024 1048576 1047552 5120 10240 204.8

Radix-2 Decimation-In-Time FFT (DIT-FFT) algorithms for computation of DFT

In this section we describe the most widely used radix-2 decimation-in-time FFT algorithm. In
this algorithm 𝑥(𝑛) is the given sequence, and we are finding 𝑁 = 2𝑣 point DFT, where v is any
integer. For example 𝑁 = 2𝑣 = 23 = 8 , then
𝑥(𝑛) = {𝑥(0), 𝑥(1), 𝑥(2), 𝑥(3), 𝑥 (4), 𝑥(5), 𝑥(6), 𝑥(7)}
𝑁
In the first decimation, the given sequence is 𝑥(𝑛) consists of N sequences is divided into 2

even and odd numbered sequences i.e. 𝑓1 (𝑛) = 𝑥(2𝑛) and 𝑓2 (𝑛) = 𝑥(2𝑛 + 1) respectively, 0 ≤
𝑁
𝑛≤ −1
2

𝑓1 (𝑛) = 𝑥(2𝑛) = 𝑥(0), 𝑥(2), 𝑥(4), 𝑥(6)

𝑓2 (𝑛) = 𝑥(2𝑛 + 1) = 𝑥(1), 𝑥(3), 𝑥(5), 𝑥(7)
The N-point DFT of 𝑥(𝑛) is 𝑋(𝑘) = ∑𝑁−1 𝑛𝑘
𝑛=0 𝑥(𝑛)𝑊𝑁 0 ≤𝑘 ≤ 𝑁−1
𝑁 𝑁
−1 −1
2 2
(2𝑛+1)𝑘
𝑋(𝑘) = ∑ 𝑥(2𝑛)𝑊𝑁2𝑛𝑘 + ∑ 𝑥(2𝑛 + 1)𝑊𝑁
𝑛=0 𝑛=0

24
 Prepared by Dr Harish M S

𝑁 𝑁
−1 −1
2 2

𝑋(𝑘) = ∑ 𝑓1 (𝑛)𝑊𝑁𝑛𝑘 + 𝑊8𝑘 ∑ 𝑓2 (𝑛)𝑊𝑁𝑛𝑘

2 2
𝑛=0 𝑛=0

Since 𝑁 = 8
3 3

𝑋(𝑘) = ∑ 𝑓1 (𝑛)𝑊4𝑛𝑘 + 𝑊8𝑘 ∑ 𝑓2 (𝑛)𝑊4𝑛𝑘 0≤𝑘 ≤𝑁−1

𝑛=0 𝑛=0

𝑋 (𝑘) = 𝐹1 (𝑘) + 𝑊8𝑘 𝐹2 (𝑘)

Where 𝐹1 (𝑘) is and 𝐹2 (𝑘) are 4-point DFT of 𝑓1 (𝑛) and 𝑓2 (𝑛) respectively.
𝑋(𝑘) = 𝐹1 (𝑘) + 𝑊8𝑘 𝐹2 (𝑘)0 ≤ 𝑘 ≤ 7 ………….. (1)
𝑋 (0) = 𝐹1 (0) + 𝑊80 𝐹2 (0)
𝑋(1) = 𝐹1 (1) + 𝑊81 𝐹2 (1)
𝑋 (2) = 𝐹1 (2) + 𝑊82 𝐹2 (2)
𝑋 (3) = 𝐹1 (3) + 𝑊83 𝐹2 (3)
𝑋 (4) = 𝐹1 (4) + 𝑊84 𝐹2 (4)
𝑋 (5) = 𝐹1 (5) + 𝑊85 𝐹2 (5)
𝑋 (6) = 𝐹1 (6) + 𝑊86 𝐹2 (6)
𝑋 (7) = 𝐹1 (7) + 𝑊87 𝐹2 (7)
We know that 𝑋(𝑘) is periodic with period N then
𝑋(𝑘) = 𝑋 (𝑘 + 𝑁) 0 ≤ 𝑘 ≤ 𝑁 − 1
𝑁
Similarly 𝐹1 (𝑘) and 𝐹2 (𝑘) are periodic with period 2
𝑁 𝑁
∴ 𝐹1 (𝑘) = 𝐹1 (𝑘 + ) 0 ≤ 𝑘 ≤ −1
2 2
𝑁 𝑁
𝐹2 (𝑘) = 𝐹2 (𝑘 + ) 0 ≤ 𝑘 ≤ −1
2 2
𝑁
𝑘+ 𝑁
And 𝑊𝑁 2
= −𝑊𝑁𝑘 0 ≤ 𝑘 ≤ −1
2

𝑋 (𝑘) = 𝐹1 (𝑘) + 𝑊8𝑘 𝐹2 (𝑘)0 ≤ 𝑘 ≤ 3

𝑋 (𝑘) = 𝐹1 (𝑘) − 𝑊8𝑘 𝐹2 (𝑘)0 ≤ 𝑘 ≤ 3
4 0
𝐹1 (0) = 𝐹1 (4) 𝐹2 (0) = 𝐹2 (4) 𝑊8 = −𝑊8
𝐹1 (1) = 𝐹1 (5) 𝐹2 (1) = 𝐹2 (5) 𝑊85 = −𝑊81
𝐹1 (2) = 𝐹1 (6) 𝐹2 (2) = 𝐹2 (6) 𝑊86 = −𝑊82
𝐹1 (3) = 𝐹1 (7) 𝐹2 (3) = 𝐹2 (7) 𝑊87 = −𝑊83

25
 Prepared by Dr Harish M S

𝑋(0) = 𝐹1 (0) + 𝑊80 𝐹2 (0)

𝑋 (1) = 𝐹1 (1) + 𝑊81 𝐹2 (1)
𝑋(2) = 𝐹1 (2) + 𝑊82 𝐹2 (2)
𝑋(3) = 𝐹1 (3) + 𝑊83 𝐹2 (3)
∴ ………………….. (2)
𝑋(4) = 𝐹1 (0) − 𝑊80 𝐹2 (0)
𝑋 (5) = 𝐹1 (1) − 𝑊81 𝐹2 (1)
𝑋(6) = 𝐹1 (2) − 𝑊82 𝐹2 (2)
{𝑋(7) = 𝐹1 (3) − 𝑊83 𝐹2 (3)
In general
𝑁
𝑋(𝑘) = 𝐹1 (𝑘) + 𝑊𝑁𝑘 𝐹2 (𝑘) 0 ≤ 𝑘 ≤ −1
2
𝑁 𝑁
𝑋 (𝑘 + ) = 𝐹1 (𝑘) − 𝑊𝑁𝑘 𝐹2 (𝑘) 0 ≤ 𝑘 ≤ − 1
2 2
𝑁
In the second decimation, the each sequence of 𝑓1 (𝑛) and 𝑓2 (𝑛) consists of sequences is
2
𝑁 𝑁
further divided into even and odd numbered sequences respectively 0 ≤ 𝑛 ≤ − 1.
4 4

𝑁
𝑔1 (𝑛) = 𝑓1 (2𝑛) = 𝑥(4𝑛) 0 ≤ 𝑛 ≤ −1
4
𝑔1 (𝑛) = {𝑥(0), 𝑥(4)}
𝑁
𝑔2 (𝑛) = 𝑓1 (2𝑛 + 1) = 𝑥(2(2𝑛 + 1) = 𝑥(4𝑛 + 2) 0 ≤ 𝑛 ≤ −1
4
𝑔2 (𝑛) = {𝑥(2), 𝑥(6)}
𝑁
ℎ1 (𝑛) = 𝑓2 (2𝑛) = 𝑥(2(2𝑛) + 1)) = 𝑥(4𝑛 + 1) 0 ≤ 𝑛 ≤ −1
4
ℎ1 (𝑛) = {𝑥(1), 𝑥(5)}
𝑁
ℎ2 (𝑛) = 𝑓1 (2𝑛 + 1) = 𝑥( 2(2𝑛 + 1) + 1) = 𝑥(4𝑛 + 3) 0 ≤ 𝑛 ≤ −1
4
ℎ2 (𝑛) = {𝑥(3), 𝑥(7)}
𝑁 𝑁 𝑁
−1 −1 −1
2 4 4
2(2𝑛+1)𝑘 𝑁
𝐹1 (𝑘) = ∑ 𝑓1 (𝑛)𝑊𝑁2𝑛𝑘 = ∑ 𝑓1 (2𝑛)𝑊𝑁4𝑛𝑘 + ∑ 𝑓1 (2𝑛 + 1)𝑊𝑁 0≤𝑘≤ −1
2
𝑛=0 𝑛=0 𝑛=0

26
 Prepared by Dr Harish M S

𝑁 𝑁
−1 −1
4 4
2(2𝑛+1)𝑘
𝐹1 (𝑘) = ∑ 𝑓1 (2𝑛)𝑊𝑁4𝑛𝑘 + ∑ 𝑓1 (2𝑛 + 1)𝑊𝑁
𝑛=0 𝑛=0
𝑁 𝑁
−1 −1
4 4

= ∑ 𝑔1 (𝑛)𝑊𝑁4𝑛𝑘 + ∑ 𝑔2 (𝑛)𝑊𝑁2(2𝑛+1)𝑘
𝑛=0 𝑛=0
𝑁 𝑁
−1 −1
4 4

= ∑ 𝑔1 (𝑛)𝑊𝑁4𝑛𝑘 + 𝑊𝑁2𝑘 ∑ 𝑔2 (𝑛)𝑊𝑁4𝑛𝑘

𝑛=0 𝑛=0

𝑁
𝐹1 (𝑘) = 𝐺1 (𝑘) + 𝑊𝑁2𝑘 𝐺2 (𝑘) 0≤𝑘≤ −1
2
Using periodic property
𝑁
𝐹1 (𝑘) = 𝐺1 (𝑘) + 𝑊𝑁2𝑘 𝐺2 (𝑘) 0≤𝑘≤ −1
4
𝑁 𝑁
𝐹1 (𝑘 + ) = 𝐺1 (𝑘) − 𝑊𝑁2𝑘 𝐺2 (𝑘) 0≤𝑘 ≤ −1
4 4
𝑁
Similarly 𝐹2 (𝑘) = 𝐻1 (𝑘) + 𝑊𝑁2𝑘 𝐻2 (𝑘) 0≤𝑘≤ −1
4

𝑁 𝑁
𝐹2 (𝑘 + ) = 𝐻1 (𝑘) − 𝑊𝑁2𝑘 𝐻2 (𝑘) 0≤𝑘≤ − 1𝐼𝑓 𝑁 = 8
4 4
𝐹1 (0) = 𝐺1 (0) + 𝑊80 𝐺2 (0)
𝐹1 (1) = 𝐺1 (1) + 𝑊82 𝐺2 (1)
𝐹1 (2) = 𝐺1 (0) − 𝑊80 𝐺2 (0)
𝐹1 (3) = 𝐺1 (1) − 𝑊82 𝐺2 (1)
..……….. (3)
𝐹2 (0) = 𝐻1 (0) + 𝑊80 𝐻2 (0)
𝐹2 (1) = 𝐻1 (1) + 𝑊82 𝐻2 (1)
𝐹2 (2) = 𝐻1 (0) − 𝑊80 𝐻2 (0)
{ 𝐹2 (3) = 𝐻1 (1) − 𝑊82 𝐻2 (1)
𝑁
In the third decimation, the each sequence of 𝑔1 (𝑛), 𝑔2 (𝑛), ℎ1 (𝑛) and ℎ2 (𝑛) consists of 4

sequences, then
𝑁 𝑁
−1 −1
4 4
𝑁
𝐺1 (𝑘) = ∑ 𝑔1 (𝑛)𝑊𝑁4𝑛𝑘 = ∑ 𝑥(4𝑛)𝑊𝑁4𝑛𝑘 0 ≤ 𝑘 ≤ −1
4
𝑛=0 𝑛=0
1

𝐺1 (𝑘) = ∑ 𝑥 (4𝑛)𝑊84𝑛𝑘 = 𝑥(0)𝑊80 + 𝑥(4)𝑊84𝑘 0 ≤ 𝑘 ≤ 1

𝑛=0

27
 Prepared by Dr Harish M S

𝐺1 (0) = 𝑥(0)𝑊80 + 𝑥(4)𝑊80

𝐺1 (1) = 𝑥(0)𝑊80 + 𝑥(4)𝑊84 = 𝑥(0)𝑊80 − 𝑥(4)𝑊80
Similarly
𝑁 𝑁
−1 −1
4 4
𝑁
𝐺2 (𝑘) = ∑ 𝑔2 (𝑛)𝑊𝑁4𝑛𝑘 = ∑ 𝑥(4𝑛 + 2)𝑊𝑁4𝑛𝑘 0 ≤ 𝑘 ≤ −1
4
𝑛=0 𝑛=0
1

𝐺2 (𝑘) = ∑ 𝑥(4𝑛 + 2)𝑊84𝑛𝑘 = 𝑥(2)𝑊80 + 𝑥(6)𝑊84𝑘 0≤𝑘≤ 1

𝑛=0

𝐺2 (0) = 𝑥(2)𝑊80 + 𝑥(6)𝑊80

𝐺2 (1) = 𝑥 (2)𝑊80 + 𝑥(6)𝑊84 = 𝑥(2)𝑊80 − 𝑥(6)𝑊80
And
𝐻1 (0) = 𝑥(1)𝑊80 + 𝑥 (5)𝑊80
𝐻1 (1) = 𝑥(1)𝑊80 − 𝑥 (5)𝑊80
𝐻2 (0) = 𝑥(3)𝑊80 + 𝑥(7)𝑊80
𝐻2 (1) = 𝑥(3)𝑊80 − 𝑥(7)𝑊80
𝐺1 (0) = 𝑥 (0)𝑊80 + 𝑥(4)𝑊80
𝐺1 (1) = 𝑥(0)𝑊80 − 𝑥(4)𝑊80
𝐺2 (0) = 𝑥 (2)𝑊80 + 𝑥(6)𝑊80
𝐺2 (1) = 𝑥(2)𝑊80 − 𝑥(6)𝑊80
∴ …………. (4)
𝐻1 (0) = 𝑥(1)𝑊80 + 𝑥(5)𝑊80
𝐻1 (1) = 𝑥(1)𝑊80 − 𝑥(5)𝑊80
𝐻2 (0) = 𝑥(3)𝑊80 + 𝑥(7)𝑊80
{ 𝐻2 (1) = 𝑥(3)𝑊80 − 𝑥(7)𝑊80
Using set of equations 4, 3, 2 the signal flow graph for DIT-FFT algorithm for 𝑁 = 8 as shown
in figure 3.7.

28
 Prepared by Dr Harish M S

Important features of DIT-FFT algorithm

1. It requires log2 𝑁 stages of computations. If 𝑁 = 8, number of computation stages are
log 2 8 = 3.
2. By observing the above flow graph, at every computation it takes two inputs say a and b.
multiply b by 𝑊𝑁𝑟 and then add 𝑎 + 𝑏𝑊𝑁𝑟 gives a new number 𝑎 + 𝑏𝑊𝑁𝑟 = 𝐴 and 𝑎 −
𝑏𝑊𝑁𝑟 = 𝐵 as shown in the figure 3.8. This diagram is also known as Butterfly diagram,
because the graph resembles like Butterfly.

29
 Prepared by Dr Harish M S

𝑁
3. In each stage there are Butterflies if N=8, then 4 Butterflies in each stages and total
2
𝑁
number of Butterflies 2 log 2 𝑁 = 4 log2 8 = 12 Butterflyies.

4. In the signal flow graph, the output sequence 𝑋(𝑘) is in the normal order and the input
sequence 𝑥(𝑛) is not in the normal order. The input and output sequences are bit
reversed.
x(n) X(k)
0 (000) 0 (000)
4 (100) 1 (001)
2 (010) 2 (010)
6 (110) 3 (011)
1 (001) 4 (100)
5 (101) 5 (101)
3 (011) 6 (110)
7 (111) 7 (111)
Figure 3.9: Table shows sequences of 𝑥(𝑛) and 𝑋(𝑘) are bit reversed.
Problem: Using DIT-FFT algorithm, find the 8-point DFT of the sequence
𝑥 (𝑛 ) = 𝑛 0≤𝑛≤7.
Solution: 𝑥(𝑛) = (0,1,2,3, 4 ,5, 6, 7)
First stage computation:
𝐺1 (0) = 𝑥(0)𝑊80 + 𝑥(4)𝑊80 = 0 + 4 = 4
𝐺1 (1) = 𝑥(0)𝑊80 − 𝑥(4)𝑊80 = 0 − 4 = −4
𝐺2 (0) = 𝑥 (2)𝑊80 + 𝑥(6)𝑊80 = 2 + 6 = 8
𝐺2 (1) = 𝑥(2)𝑊80 − 𝑥(6)𝑊80 = 2 − 6 = −4
𝐻1 (0) = 𝑥(1)𝑊80 + 𝑥(5)𝑊80 = 1 + 5 = 6
𝐻1 (1) = 𝑥(1)𝑊80 − 𝑥(5)𝑊80 = 1 − 5 = −4
𝐻2 (0) = 𝑥(3)𝑊80 + 𝑥 (7)𝑊80 = 3 + 7 = 10
{ 𝐻2 (1) = 𝑥(3)𝑊80 − 𝑥(7)𝑊80 = 3 − 7 = −4
Second stage computation:

30
 Prepared by Dr Harish M S

𝐹1 (0) = 𝐺1 (0) + 𝑊80 𝐺2 (0) = 4 + 8 = 12

2
𝐹1 (1) = 𝐺1 (1) + 𝑊8 𝐺2 (1) = −4 + (−𝑗)(−4) = −4 + 4𝑗
𝐹1 (2) = 𝐺1 (0) − 𝑊80 𝐺2 (0) = 4 − (8) = −4
𝐹1 (3) = 𝐺1 (1) − 𝑊82 𝐺2 (1) = −4 − (𝑗)(−4) = 4 + 4𝑗
𝐹2 (0) = 𝐻1 (0) + 𝑊80 𝐻2 (0) = 6 + 10 = 16
𝐹2 (1) = 𝐻1 (1) + 𝑊82 𝐻2 (1) = −4 + (−𝑗)(−4) = −4 + 4𝑗
𝐹2 (2) = 𝐻1 (0) − 𝑊80 𝐻2 (0) = 6 − 10 = −4
{ 𝐹2 (3) = 𝐻1 (1) − 𝑊82 𝐻2 (1) = −4 − (−𝑗)(−4) = −4 − 4𝑗
Third stage computation:
𝑋(0) = 𝐹1 (0) + 𝑊80 𝐹2 (0) = 12 + 16 = 28
1 1
𝑋(1) = 𝐹1 (1) + 𝑊81 𝐹2 (1) = (−4 + 4𝑗) + ( − 𝑗 ) (−4 + 4𝑗) = −4 + 𝑗9.65
√2 √2
2
𝑋 (2) = 𝐹1 (2) + 𝑊8 𝐹2 (2) = −4 + (−𝑗)(−4) = −4 + 4𝑗
1 1
𝑋(3) = 𝐹1 (3) + 𝑊83 𝐹2 (3) = (−4 − 4𝑗) + (− − 𝑗 ) (−4 − 4𝑗) = −4 + 1.65𝑗
√2 √2
0
𝑋(4) = 𝐹1 (0) − 𝑊8 𝐹2 (0) = 12 − 16 = −4
1 1
𝑋 (5) = 𝐹1 (1) − 𝑊81 𝐹2 (1) = (−4 + 4𝑗) − ( − 𝑗 ) (−4 + 4𝑗) = −4 − 1.65𝑗
√2 √2
2
𝑋 (6) = 𝐹1 (2) − 𝑊8 𝐹2 (2) = −4 − (−𝑗)(−4) = −4 − 4𝑗
1 1
𝑋(7) = 𝐹1 (3) − 𝑊83 𝐹2 (3) = (−4 − 4𝑗) − (− − 𝑗 ) (−4 − 4𝑗) = −4 − j9.65
{ √2 √2
∴ 𝑋(𝑘) = [28, −4 + 𝑗9.65, −4 + 4𝑗, −4 + 1.65, −4, −4 − 𝑗1.64, −4 − 4𝑗, −4 − j9.65].

31
 Prepared by Dr Harish M S

Problem: Using DIT-FFT algorithm, find the IDFT of the sequence

𝑋(𝑘) = [28, −4 + 𝑗9.65, −4 + 4𝑗, −4 + 1.65𝑗, −4, −4 − 𝑗1.64, −4 − 4𝑗, −4 − j9.65].
Solution: 𝑋 (0) = 28, 𝑋(1) = −4 + 𝑗9.65, 𝑋(2) = −4 + 4𝑗, 𝑋(3) = −4 + 1.65𝑗, 𝑋 (4) = −4
𝑋(5) = −4 − 𝑗1.64, 𝑋(6) = −4 − 4𝑗, and (7) = −4 − j9.65

First stage computation:

𝑔1 (0) = 𝑋(0)𝑊80 + 𝑋 (4)𝑊80 = 28 − 4 = 24
𝑔1 (1) = 𝑋 (0)𝑊80 − 𝑋(4)𝑊80 = 28 + 4 = 32
𝑔2 (0) = 𝑋(2)𝑊80 + 𝑋 (6)𝑊80 = (−4 + 4𝑗) + (−4 − 4𝑗) = −8
𝑔2 (1) = 𝑋(2)𝑊80 − 𝑋 (6)𝑊80 = (−4 + 4𝑗) − (−4 − 4𝑗) = 8𝑗
ℎ1 (0) = 𝑋 (1)𝑊80 + 𝑋(5)𝑊80 = (−4 + 𝑗9.65) + (−4 − 𝑗1.65) = −8 + 8𝑗
ℎ1 (1) = 𝑋 (1)𝑊80 − 𝑋(5)𝑊80 = (−4 + 𝑗9.65) − (−4 − 𝑗1.65) = 11.3𝑗
ℎ2 (0) = 𝑋(3)𝑊80 + 𝑋 (7)𝑊80 = (−4 + 1.65𝑗) + (−4 − 9.65𝑗) = −8 − 8𝑗
{ ℎ2 (1) = 𝑋(3)𝑊80 − 𝑋 (7)𝑊80 = (−4 + 1.65𝑗) − (−4 − 9.65𝑗) = 11.3𝑗
Second stage computation:

32
 Prepared by Dr Harish M S

𝑓1 (0) = 𝑔1 (0) + 𝑊80 𝑔2 (0) = 24 − 8 = 16

𝑓1 (1) = 𝑔1 (1) + 𝑊8−2 𝑔2 (1) = 32 + (𝑗)(8𝑗) = 24
𝑓1 (2) = 𝑔1 (0) − 𝑊80 𝑔2 (0) = 24 − (−8) = 32
𝑓1 (3) = 𝑔1 (1) − 𝑊8−2 𝑔2 (1) = 32 − (𝑗)(8𝑗) = 40
𝑓2 (0) = ℎ1 (0) + 𝑊80 ℎ2 (0) = (−8 + 8𝑗) + (−8 − 8𝑗) = −16
𝑓2 (1) = ℎ1 (1) + 𝑊8−2 ℎ2 (1) = (11.3𝑗) + (𝑗)((11.3𝑗) = −11.3 + 11.3𝑗
𝑓2 (2) = ℎ1 (0) − 𝑊80 ℎ2 (0) = (−8 + 8𝑗) − (−8 − 8𝑗) = 16𝑗
{ 𝑓2 (3) = ℎ1 (1) − 𝑊8−2 ℎ2 (1) = (11.3𝑗) − (𝑗)((11.3𝑗) = 11.3 + 11.3𝑗
Third stage computation:
1 1
𝑥(0) = (𝑓1 (0) + 𝑊80 𝑓2 (0)) = (16 − 16) = 0
8 8
1 1 1 𝑗
𝑥(1) = [𝑓1 (1) + 𝑊8−1 𝑓2 (1)] = [24 + ( + ) (−11.3 + 11.3𝑗)] = 1
8 8 √2 √2
1 1 1
𝑥(2) = [𝑓1 (2) + 𝑊8−2 𝑓2 (2)] = [32 + (𝑗)(16𝑗)] = [32 − 16] = 2
8 8 8
1 1 1 𝑗
𝑥(3) = [𝑓1 (3) + 𝑊8−3 𝑓2 (3)] = [(40) + (− + ) (11.3 + 11.3𝑗)] = 3
8 8 √2 √2
1 1
𝑥 (4) = [𝑓1 (0) − 𝑊80 𝑓2 (0)] = [16 + 16] = 4
8 8
1 1 1 𝑗
𝑥(5) = [𝑓1 (1) − 𝑊8−1 𝑓2 (1)] = [24 − ( + ) (−11.3 + 11.3𝑗)] = 5
8 8 √2 √2
1 1
𝑥(6) = [𝑓1 (2) − 𝑊8−2 𝑓2 (2)] = [32 − (𝑗)(16𝑗)] = 6
8 8
1 1 1 𝑗
𝑥(7) = [𝑓1 (3) − 𝑊8−3 𝑓2 (3)] = [(40) − (− + ) (11.3 + 11.3𝑗)] = 7
{ 8 8 √2 √2
∴ 𝑥(𝑛) = [0,1,2, 3,4,5,6,7
Problem: Use DIT-FFT algorithm to find 𝑦(𝑛) = 𝑥1 (𝑛) ⊛ 𝑥2 (𝑛), where 𝑥1 (𝑛) = (1,1,2,2) and
𝑥2 (𝑛) = (1,2,3,4)
Solution: 𝑦(𝑛) = 𝑥1 (𝑛) ⊛ 𝑥2 (𝑛)
𝑌(𝑘) = 𝑋1 (𝑘)𝑋2 (𝑘)
3
1
𝑦(𝑛) = ∑ 𝑌 (𝑘)𝑊𝑁−𝑛𝑘 0≤𝑛≤3
4
𝑘=0

If we apply DFT and IDFT algorithm we get 𝑦(𝑛)

33
 Prepared by Dr Harish M S

Computation of 𝑋1 (k)
𝐹1 (0) = 𝑥1 (0) + 𝑥1 (2)𝑊𝑁0 = 1 + 2 = 3
𝐹1 (1) = 𝑥1 (0) − 𝑥1 (2)𝑊𝑁0 = 1 − 2 = −1
𝐹2 (0) = 𝑥1 (1) + 𝑥1 (3)𝑊𝑁0 = 1 + 2 = 3
𝐹2 (1) = 𝑥1 (1) − 𝑥1 (3)𝑊𝑁0 = 1 − 2 = −1
𝑋1 (0) = 𝐹1 (0) + 𝐹2 (0)𝑊40 = 3 + 3 = 6
𝑋1 (1) = 𝐹1 (1) + 𝐹2 (1)𝑊41 = −1 + (−1)(−𝑗) = −1 + 𝑗
𝑋1 (2) = 𝐹1 (0) − 𝐹2 (0)𝑊40 = 3 − 3 = 0
𝑋1 (3) = 𝐹1 (1) − 𝐹2 (1)𝑊41 = −1 − (−1)(−𝑗) = −1 − 𝑗
∴ 𝑋1 (k) = [6, −1 + j, 0, −1 − j]
Computation of 𝑋2 (𝑘)
𝐹12 (0) = 𝑥2 (0) + 𝑥2 (2)𝑊𝑁0 = 1 + 3 = 4
𝐹12 (1) = 𝑥2 (0) − 𝑥2 (2)𝑊𝑁0 = 1 − 3 = −2
𝐹22 (0) = 𝑥2 (1) + 𝑥2 (3)𝑊𝑁0 = 2 + 4 = 6
𝐹22 (1) = 𝑥1 (1) − 𝑥2 (3)𝑊𝑁0 = 1 − 4 = −2
𝑋2 (0) = 𝐹12 (0) + 𝐹22 (0)𝑊40 = 4 + 6 = 10
𝑋2 (1) = 𝐹12 (1) + 𝐹22 (1)𝑊41 = −2 + (−2)(−𝑗) = −2 + 2𝑗

34
 Prepared by Dr Harish M S

𝑋2 (2) = 𝐹12 (0) − 𝐹22 (0)𝑊40 = 4 − 6 = −2

𝑋2 (3) = 𝐹12 (1) − 𝐹22 (1)𝑊41 = −4 − (−2)(−𝑗) = −2 − 𝑗2
∴ 𝑋2 (k) = [10, −2 + 2j, −2, −2 − 2j]
𝑌(𝑘) = 𝑋1 (𝑘)𝑋2 (𝑘) = (60, −4𝑗, 0,4𝑗)
𝑓13 (0) = 𝑌(0) + 𝑌(2)𝑊40 = 60 + 0 = 60
𝑓13 (1) = 𝑌(0) − 𝑌(2)𝑊40 = 60 − 0 = 60
𝑓23 (0) = 𝑌(1) + 𝑌(3)𝑊40 = −4𝑗 + 4𝑗 = 0
𝑓23 (1) = 𝑌(1) − 𝑌(3)𝑊40 = −4𝑗 − 4𝑗 = −8𝑗
1 1
𝑦(0) = (𝑓13 (0) + 𝑓23 (0)𝑊4−0 ) = (60 + 0) = 15
4 4
1 1
𝑦(1) = (𝑓13 (1) + 𝑓23 (1)𝑊4−1 ) = (60 + (−8𝑗)(𝑗) = 17
4 4
1 1
𝑦(2) = (𝑓13 (0) − 𝑓23 (0)𝑊4−0 ) = (60 − 0) = 15
4 4
1 1
𝑦(3) = (𝑓13 (1) − 𝑓23 (1)𝑊4−1 ) = (60 − (−8𝑗)(−𝑗) = 13
4 4
∴ 𝑦(𝑛) = {15,17,15,13}

35