DIGITAL SIGNAL PROCESSING

th
5 Semester, 2020-21

ANIRBAN BHATTACHARJEE
ASSISTANT PROFESSOR
DEPARTMENT OF ECE,NIT AGARTALA

9/17/2020 Anirban Bhattacharjee, Department of ECE, NIT Agartala

 Circular Convolution
Consider two finite length discrete time sequences 𝑥1 𝑛 𝑎𝑛𝑑 𝑥2 𝑛
𝑁 𝑁
𝑠𝑢𝑐ℎ 𝑡ℎ𝑎𝑡 𝑥1 𝑛 𝑋1 𝑘 and 𝑥2 𝑛 𝑋2 𝑘 .
Let
𝑋3 𝑘 = 𝑋1 𝑘 𝑋2 𝑘 , k = 0,1,2, … . , N − 1
Now taking IDFT
𝑁−1
1 2𝜋
𝑗 𝑘𝑚
𝑥3 𝑚 = 𝑋3 𝑘 𝑒 𝑁 , 𝑚 = 0,1,2, … , 𝑁 − 1
𝑁
𝑘=0

𝑁−1
1 2𝜋
𝑗 𝑁 𝑘𝑚
𝑥3 𝑚 = 𝑋1 𝑘 𝑋2 𝑘 𝑒
𝑁
𝑘=0

9/17/2020 Anirban Bhattacharjee, Department of ECE, NIT Agartala

 𝑁−1 𝑁−1 𝑁−1
1 2𝜋
−𝑗𝑘 𝑁 𝑛
2𝜋
−𝑗𝑘 𝑁 𝑙
2𝜋
𝑗 𝑁 𝑘𝑚
𝑥3 𝑚 = 𝑥1 𝑛 𝑒 𝑥2 𝑙 𝑒 𝑒
𝑁
𝑘=0 𝑛=0 𝑙=0

𝑁−1 𝑁−1 𝑁−1

1 2𝜋
𝑗 𝑁 𝑘(𝑚−𝑛−𝑙)
𝑥3 𝑚 = 𝑥1 𝑛 𝑥2 𝑙 𝑒
𝑁
𝑛=0 𝑙=0 𝑘=0

2𝜋
𝑗 𝑁 (𝑚−𝑛−𝑙)
Consider the inner sum and let 𝑎 = 𝑒
Therefore

𝑁−1
2𝜋
𝑁−1 𝑁 ,𝑎 = 1
𝑗 𝑁 𝑘(𝑚−𝑛−𝑙)
𝑒 = 𝑎𝑘 = 1 − 𝑎𝑁
,𝑎 ≠ 1
𝑘=0 𝑘=0 1−𝑎
9/17/2020
 When m−n−l is a multiple of N, a=1 and 𝑎𝑁 = 1 𝑓𝑜𝑟 𝑎 ≠ 0
Hence
𝑁−1
𝑁, 𝑙 = 𝑚 − 𝑛 + 𝑝𝑁 = (< 𝑚 − 𝑛 >𝑁 )
𝑎𝑘 =
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
𝑘=0
Therefore
𝑁−1 𝑁−1
1
𝑥3 𝑚 = 𝑥1 𝑛 𝑥2 𝑙 𝑁
𝑁
𝑛=0 𝑙=0

𝑁−1

𝑥3 𝑚 = 𝑥1 𝑛 𝑥2 (< 𝑚 − 𝑛 >𝑁 ), 𝑚 = 0,1,2, … , 𝑁 − 1

𝑛=0
The expression on the RHS represents the Circular Convolution of 𝑥1 𝑛 and 𝑥2 𝑛
represented as
𝑥3 𝑚 = 𝑥1 𝑛 ⊛ 𝑥1 𝑛

9/17/2020 Anirban Bhattacharjee, Department of ECE, NIT Agartala

Thus the circular convolution property may be stated as

𝑁 𝑁
𝐼𝑓 𝑥1 𝑛 𝑋1 𝑘 and 𝑥2 𝑛 𝑋2 𝑘 , 𝑡ℎ𝑒𝑛

𝑁
𝑥1 𝑛 ⊛ 𝑥1 𝑛 𝑋1 𝑘 𝑋2 𝑘

9/17/2020 Anirban Bhattacharjee, Department of ECE, NIT Agartala

 Example:
Determine the 4- point circular convolution for the following sequences
𝑥1 𝑛 = 2,2,1 𝑎𝑛𝑑𝑥2 𝑛 = 1,2,4,3

ANALYTICAL METHOD: 𝑥1 𝑛 = 2,2,1,0

3

𝑥3 𝑚 = 𝑥1 𝑛 𝑥2 (< 𝑚 − 𝑛 >4 ), 𝑚 = 0,1,2,3

𝑛=0
3

𝑥3 0 = 𝑥1 𝑛 𝑥2 (< −𝑛 >4 ) = 𝑥1 0 𝑥2 0 + 𝑥1 1 𝑥2 3 + 𝑥1 2 𝑥2 2 + 𝑥1 3 𝑥2 1 = 12
𝑛=0
3

𝑥3 1 = 𝑥1 𝑛 𝑥2 (< 1 − 𝑛 >4 ) = 𝑥1 0 𝑥2 1 + 𝑥1 1 𝑥2 0 + 𝑥1 2 𝑥2 3 + 𝑥1 3 𝑥2 2 = 9
𝑛=0

𝑥3 2 = 𝑥1 𝑛 𝑥2 (< 2 − 𝑛 >4 ) = 𝑥1 0 𝑥2 2 + 𝑥1 1 𝑥2 1 + 𝑥1 2 𝑥2 0 + 𝑥1 3 𝑥2 3 = 13
𝑛=0
3

𝑥3 3 = 𝑥1 𝑛 𝑥2 (< 3 − 𝑛 >4 ) = 𝑥1 0 𝑥2 3 + 𝑥1 1 𝑥2 2 + 𝑥1 2 𝑥2 1 + 𝑥1 3 𝑥2 0 = 16
𝑛=0
MATRIX METHOD:
𝑥3 0 𝑥2 0 𝑥2 3 𝑥2 2 𝑥2 1 𝑥1 0
𝑥3 1 𝑥2 1 𝑥2 0 𝑥2 3 𝑥2 2 𝑥1 1
=
𝑥3 2 𝑥2 2 𝑥2 1 𝑥2 0 𝑥2 3 𝑥1 2
𝑥3 3 𝑥2 3 𝑥2 2 𝑥2 1 𝑥2 0 𝑥1 3

𝑥3 0 1342 2 12
𝑥3 1 2134 2 = 9
=
𝑥3 2 4213 1 13
𝑥3 3 3421 0 16

9/17/2020 Anirban Bhattacharjee, Department of ECE, NIT Agartala

FREQUENCY DOMAIN METHOD:
3
2𝜋
−𝑗 𝑘𝑛
𝑋1 𝑘 = 𝑥1 𝑛 𝑒 4 ,𝑘 = 0,1,2,3
𝑛=0
3

𝑋1 𝑘 = 𝑥1 𝑛 𝑊4 𝑘𝑛
𝑛=0

𝑋1 (0) 1 11 12 13 𝑥1 (0)
1 11 12 13
𝑋1 (1) 1𝑊4 𝑊4 𝑊4 𝑥1 (1) 𝑊 𝑊 𝑊
= 2 4 6 ⇨ W4 = 1 4 2 4 4 4 6
𝑋1 (2) 1𝑊4 𝑊4 𝑊4 𝑥1 (2) 1𝑊4 𝑊4 𝑊4
𝑋1 (3) 1𝑊4 3 𝑊4 6 𝑊4 9 𝑥1 (3) 1𝑊4 3 𝑊4 6 𝑊4 9

𝑊41 = 𝑗, 𝑊4 2 = −1, 𝑊4 3 = 𝑗, 𝑊4 4 = 1, 𝑊4 6 = −1, 𝑊4 9 = −𝑗

9/17/2020 Anirban Bhattacharjee, Department of ECE, NIT Agartala

 𝑋1 (0) 11 1 1 2 5
𝑋1 (1) 1 −𝑗 −1 𝑗 2 = 1 − 𝑗2
=
𝑋1 (2) 1−1 1 −1 1 1
𝑋1 (3) 1 𝑗 −1 −𝑗 0 1 + 𝑗2
Similarly
𝑋2 (0) 11 1 1 1 10
𝑋2 (1) 1 −𝑗 −1 𝑗 2 = −3 + 𝑗
=
𝑋2 (2) 1−1 1 −1 4 0
𝑋2 (3) 1 𝑗 −1 −𝑗 3 −3 − 𝑗
Let
𝑋3 𝑘 = 𝑋1 (𝑘) 𝑋2 (𝑘)

Therefore 𝑋3 (0) 50
𝑋3 (1) −1 + 𝑗7
=
𝑋3 (2) 0
𝑋3 (3) −1 − 𝑗7
9/17/2020
We know that
∗
1
x = W𝑁 X
𝑁
11 1 1
1 𝑗 −1 −𝑗
W4 ∗ =
1−1 1 −1
1 −𝑗 −1 𝑗
11 1 1 50 12
1 1 𝑗 −1 −𝑗 −1 + 𝑗7
𝑥3 = = 9
4 1−1 1 −1 0 13
1 −𝑗 −1 𝑗 −1 − 𝑗7 16

𝒙𝟑 (n)={12,9,13,16}
9/17/2020 Anirban Bhattacharjee, Department of ECE, NIT Agartala
GRAPHICAL METHOD:

𝑥3 (0)=12

𝑥3 (1)=9
 Linear Convolution using Circular Convolution
Consider two finite length discrete time causal sequences 𝑥1 𝑛 𝑎𝑛𝑑 𝑥2 𝑛
𝑠𝑢𝑐ℎ 𝑡ℎ𝑎𝑡
𝑁 𝑁
𝑥1 𝑛 𝑋1 𝑘 & 𝑥2 𝑛 𝑋2 𝑘 .

Let 𝑥1 𝑛 and 𝑥2 𝑛 have lengths 𝐿1 and 𝐿2 respectively and assume that 𝐿1 > 𝐿2

The linear convolution of 𝑥1 𝑛 𝑎𝑛𝑑 𝑥2 𝑛 given by

𝑦 𝑛 = 𝑥1 𝑙 𝑥2 𝑛 − 𝑙
𝑙=−∞
The length of 𝑦 𝑛 is 𝐿 = 𝐿1 + 𝐿2 −1 and is defined for 0 ≤ 𝑛 ≤ 𝐿 − 1.
The circular convolution of 𝑥1 𝑛 𝑎𝑛𝑑 𝑥2 𝑛 is defined as

𝐿1 −1

𝑥3 𝑚 = 𝑥1 𝑛 𝑥2 (< 𝑚 − 𝑛 > 𝐿1 ), 𝑚 = 0,1,2, … , 𝐿1 −1

𝑛=0
and yields a different result from 𝑦(𝑛).

In order to realize 𝑦 𝑛 from 𝑥3 (𝑚), we must have 𝑁 ≥ 𝐿 𝑖. 𝑒. 𝑁 ≥ 𝐿1 + 𝐿2 − 1.

Hence 𝑁 − 𝐿1 zeros must be appended to 𝑥1 (𝑛) & 𝑁 − 𝐿2 zeros must be appended

to 𝑥2 (𝑛) to make the sequences of length 𝐿 .
Once this is done the circular convolution of 𝑥1 (𝑛) & 𝑥2 (𝑛) results in a sequence
which is the linear convolution of the two sequences.
# 𝑥1 𝑛 = 2,5,0,4 𝑎𝑛𝑑 𝑥2 𝑛 = 4,1,3 . Determine the linear convolution of the
two sequences from the circular convolution.
Solution:
Here 𝐿1 = 4 and 𝐿2 = 3
Linear convolution of he two sequences will yield a sequence of length
𝐿 = 𝐿1 + 𝐿2 − 1 = 4 + 3 − 1 = 6
Consequently 𝑁 = 𝐿 = 6.
Hence 2 zeros must be appended to 𝑥1 𝑛 & 3 zeros must be appended to 𝑥2 𝑛
so that the sequences become equal to length 𝐿 = 6.
Next we determine N=6 point circular convolution using

𝑥3 𝑚 = 𝑥1 𝑛 𝑥2 (< 𝑚 − 𝑛 >6 ), 𝑚 = 0,1,2, … , 5

𝑛=0
 6 Point Circular Convolution
𝑥3 0 𝑥2 0 𝑥2 5 𝑥2 4 𝑥2 3 𝑥2 2 𝑥2 1 𝑥1 0
𝑥3 1 𝑥2 1 𝑥2 0 𝑥2 5 𝑥2 4 𝑥2 3 𝑥2 2 𝑥1 1
𝑥3 2 𝑥 2 𝑥2 1 𝑥2 0 𝑥2 5 𝑥2 4 𝑥2 3 𝑥1 2
= 2
𝑥3 3 𝑥2 3 𝑥2 2 𝑥2 1 𝑥2 0 𝑥2 5 𝑥2 4 𝑥1 3
𝑥3 4 𝑥2 4 𝑥2 3 𝑥2 2 𝑥2 1 𝑥2 0 𝑥2 5 𝑥1 4
𝑥3 5 𝑥2 5 𝑥2 4 𝑥2 3 𝑥2 2 𝑥2 1 𝑥2 0 𝑥1 5

𝑥3 0 4 0 0 0 3 1 2 8
𝑥3 1 1 4 0 0 0 3 5 22
𝑥3 2 3 1 4 0 0 0 0 = 11
=
𝑥3 3 0 3 1 4 0 0 4 31
𝑥3 4 0 0 3 1 4 0 0 4
𝑥3 5 0 0 0 3 1 4 0 12

𝑦 𝑛 = {𝟖, 𝟐𝟐, 𝟏𝟏, 𝟑𝟏, 𝟒, 𝟏𝟐}

# 𝑥1 𝑛 = 1,1,1,1 𝑎𝑛𝑑 𝑥2 𝑛 = 1,1,1 . Determine the circular convolution of
the two sequences for N=6 , N=5 and N=4 and compare the result in each case with
the linear convolution sequence 𝒚 𝒏 = {𝟏, 𝟐, 𝟑, 𝟑, 𝟐, 𝟏}

Solution:
N=6

𝑥3 0 1 0 0 1 1 1 1 1
𝑥3 1 1 1 0 0 1 1 1 2
𝑥3 2 1 1 1 0 0 1 1 3
= =
𝑥3 3 1 1 1 1 0 0 0 3
𝑥3 4 0 1 1 1 1 0 0 2
𝑥3 5 0 0 1 1 1 1 0 1

𝐱𝟑 𝐧 = 𝟏, 𝟐, 𝟑, 𝟑, 𝟐, 𝟏 = 𝐲(𝐧)
N=5
𝑥3 0 1 0 1 1 1 2
𝑥3 1 1 1 0 1 1 2
𝑥3 2 = 1 1 1 0 1 = 3
𝑥3 3 1 1 1 1 0 3
𝑥3 4 0 1 1 1 0 2
𝐱𝟑 𝐧 = {𝟐, 𝟐, 𝟑, 𝟑, 𝟐}
Comparing the circular convolution sequence with the linear convolution
sequence, 𝑦 n = {1,2,3,3,2,1}, we observe that first sample of the circular
convolution sequence is dissimilar.
𝑦 n = {1,2,3,3,2,1},
1 2 3 3 2
Wrap around 1
Time Aliasing = {2, 2, 3, 3, 2}
N=4
𝑥3 0 1 1 1 1 1 3
𝑥3 1 1 1 1 1 1 = 3
=
𝑥3 2 1 1 1 1 1 3
𝑥3 3 1 1 1 1 0 3

𝐱 𝟑 𝐧 = {𝟑, 𝟑, 𝟑, 𝟑}
Comparing the circular convolution sequence with the linear convolution
sequence, 𝑦 n = {1,2,3, 3,2,1}, we observe that first three samples of the
circular convolution sequence are dissimilar.
𝑦 n = {1,2,3, 3,2,1},
1 2 3 3
Wrap around 2 1

Time Aliasing ={3, 3, 3, 3}

 Block Processing of long data sequences
Long data sequences can be thought to be infinite in length.
Determining the system output using any of the known methods is
impractical considering the size of the input sequence.
The long data sequence is divided into smaller sequences i.e. blocks
and each block is processed using DFT & IDFT to produce a block of
output data. The output blocks are joined together to generate the
sequence which represents the linear convolution sequence i.e. the
system output.
Two different methods are used: (i) Overlap-save method.
(ii) Overlap-add method.
 Overlap-Save Method

𝒙(𝒏) 𝒉(𝒏) 𝒚(𝒏)

Consider a long data sequence , 𝑥(𝑛) as input to an LTI system with impulse response
ℎ(𝑛) of length 𝐿.
 Divide 𝑥(𝑛) into blocks, 𝑥𝑏 𝑛 , of length 𝑁 with each block overlapping the previous
block by 𝐿 − 1 samples as shown below. The first 𝐿 − 1 samples of the first block is set
to zero.
 Assume 𝑁 >> 𝐿.
𝑵 𝑵 ……….
𝑵
Input Data Sequence , 𝒙(𝒏)
𝐿−1
zeros 𝒙𝟎 (𝒏)
𝐿−1
data
𝒙𝟏 (𝒏)
𝐿−1
data
𝒙𝟐 (𝒏) ……….
The blocks can be represented as
𝑥𝑏 𝑛 = 𝑥 𝑛 + 𝑏 𝑁 − 𝐿 + 1 − 𝐿 + 1 , 𝑏 = 0,1,2, … . .
Let
𝑦𝑏 𝑛 = 𝑥𝑏 𝑛 ⊛ ℎ(𝑛)

 The first 𝐿 − 1 samples of each output block is discarded and the remaining 𝑁 − 𝐿 + 1 samples
are saved and concatenated with the 𝑁 − 𝐿 + 1 samples from other blocks to obtain the final
output.

𝑳−𝟏
data
𝒚𝟎 (𝒏)

𝑳−𝟏
data
𝒚𝟏 (𝒏)

𝑳−𝟏
𝒚𝟐 (𝒏) ……….
data

Output Sequence,𝒚(𝒏)
 EXAMPLE
# Let 𝑥 𝑛 = 𝑛 + 1 , 0 ≤ 𝑛 ≤ 9 and ℎ 𝑛 = 1,0,1 . Compute the linear convolution
𝑦 𝑛 = 𝑥 𝑛 ∗ ℎ(𝑛) using the overlap save method.

Solution: The length of h(n) , L=3. Let the number of blocks be 3.

𝑡𝑜𝑡𝑎𝑙 𝑙𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑡ℎ𝑒 𝑖𝑛𝑝𝑢𝑡 𝑠𝑒𝑞𝑢𝑒𝑛𝑐𝑒

Number of blocks required=
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑛𝑒𝑤 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 𝑖𝑛 𝑒𝑎𝑐ℎ 𝑏𝑙𝑜𝑐𝑘
𝐿𝑥 10
𝐻𝑒𝑛𝑐𝑒 3 = =
𝑁−𝐿+1 𝑁−2

⇨ 𝑁 − 2 = 4 𝑖. 𝑒. 𝑁 = 6
The following are the blocks:
𝑥0 𝑛 = {0,0,1,2,3,4}
𝑥1 𝑛 = {3,4,5,6,7,8}
𝑥2 𝑛 = {7,8,9,10,0,0}

Next we compute the 6 point circular convolution

𝑦0 𝑛 = 𝑥0 𝑛 ⊛ ℎ 𝑛 = {−3, −4,1,2,2,2}

𝑦1 𝑛 = 𝑥1 𝑛 ⊛ ℎ 𝑛 = {−4, −4,2,2,2,2}

𝑦2 𝑛 = 𝑥2 𝑛 ⊛ ℎ 𝑛 = {7,8,2,2,2, −9, −10}

 𝑥 𝑛 ⇨

𝐿 − 1 𝑧𝑒𝑟𝑜𝑠
𝑁 − 𝐿 + 1 𝑑𝑎𝑡𝑎 ℎ 𝑛
𝑥0 𝑛 ⇨

𝑥1 𝑛 ⇨ 𝑁 𝑝𝑜𝑖𝑛𝑡 𝐶𝑖𝑟𝑐𝑢𝑙𝑎𝑟 𝐶𝑜𝑛𝑣𝑜𝑙𝑢𝑡𝑖𝑜𝑛

𝑥2 𝑛 ⇨
𝐿 − 1 𝑜𝑣𝑒𝑟𝑙𝑎𝑝𝑝𝑒𝑑 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 Linear Convolution :
𝑦′ 0 = {0,0,1,2,2,2, −3, −4}
𝑦 ′ 1 = {3,4,2,2,2,2, −7, −8}
𝑦 ′ 2 = {7,8,2,2, −9, −10,0,0}
𝑦0 𝑛 ⇨

𝐿 − 1 𝑑𝑖𝑠𝑐𝑎𝑟𝑑𝑒𝑑 𝑠𝑎𝑚𝑝𝑙𝑒𝑠
𝑦1 𝑛 ⇨

𝐷𝑖𝑠𝑐𝑎𝑟𝑑𝑖𝑛𝑔 𝐿 − 1 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 𝑎𝑛𝑑 𝑦2 𝑛 ⇨

𝑐𝑜𝑛𝑐𝑎𝑡𝑒𝑛𝑎𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 𝑟𝑒𝑚𝑎𝑖𝑛𝑖𝑛𝑔
𝑁 − 𝐿 + 1 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 𝑜𝑓 𝑒𝑎𝑐ℎ 𝑏𝑙𝑜𝑐𝑘

𝑦 𝑛
 Overlap-Add Method
∞

𝑥 𝑛 = 𝑥𝑏 (𝑛 − 𝑏𝑀)
𝑏=0

𝑥 𝑛 + 𝑏𝑀 , 0 ≤ 𝑛 ≤ 𝑀 − 1
𝑥𝑏 𝑛 =
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
∞ ∞

System Output: 𝑦 𝑛 =ℎ 𝑛 ∗ 𝑥𝑏 𝑛 − 𝑏𝑀 = 𝑦𝑏 (𝑛 − 𝑏𝑀) 𝑦𝑏 𝑛 = ℎ 𝑛 ∗ 𝑥𝑏 (𝑛)

𝑏=0 𝑏=0

Length of 𝑦𝑏 𝑛 is 𝑀 + 𝐿 − 1
𝑁 =𝑀+𝐿−1 𝑦0 𝑛 = ℎ 𝑛 ∗ 𝑥0 𝑛 , 0 ≤ 𝑛 ≤ 𝑀 + 𝐿 − 2
𝑦1 𝑛 = ℎ 𝑛 ∗ 𝑥1 𝑛 , 𝑀 ≤ 𝑛 ≤ 2𝑀 + 𝐿 − 2
𝑦2 𝑛 = ℎ 𝑛 ∗ 𝑥2 𝑛 , 2𝑀 ≤ 𝑛 ≤ 3𝑀 + 𝐿 − 2 𝑎𝑛𝑑 𝑠𝑜 𝑜𝑛
𝑥 𝑛 𝑦 𝑛

∗ 𝒚𝟎 (𝒏) {1,2,2,2, −3, −4}

∗ 𝒚𝟏 (𝒏) {5,6,2,2, −7, −8}

𝒚𝟐 (𝒏) {9,10, −9, −10,0,0}

∗
The linear convolution is of length M+L-1 which can be obtained by N=M+L-1 point circular convolution

Input Data Sequence , 𝒙(𝒏)

Each of the block is extended in length to N

by appending L-1 zeros
𝑳−𝟏
𝒙𝟎 (𝒏) zeros
𝑳−𝟏
𝒙𝟏 (𝒏) zeros

𝑀 𝑑𝑎𝑡𝑎 𝑳−𝟏
…. 𝒙𝒃 (𝒏) zeros
𝑀 𝑑𝑎𝑡𝑎 ……..

𝑀 𝑑𝑎𝑡𝑎
 Overlap:
𝑀+𝐿−2 − 𝑀−1 = 𝐿 − 1 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑦0 𝑛 𝑎𝑛𝑑 𝑦1 (𝑛)
2𝑀 + 𝐿 − 2 − 2𝑀 − 1 = 𝐿 − 1 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑦1 𝑛 𝑎𝑛𝑑 𝑦2 (𝑛)
and so on

Overlapped 𝐿 − 1 samples of each are added to the first 𝐿 − 1 samples of the succeeding block

𝑳−𝟏
𝒚𝟎 (𝒏) data

𝐿 − 1 𝑎𝑑𝑑𝑒𝑑 𝑳−𝟏
𝒚𝟏 (𝒏)
𝑳−𝟏
data data

𝑳−𝟏 𝑳−𝟏
𝐿 − 1 𝑎𝑑𝑑𝑒𝑑 data
𝒚𝟐 (𝒏) data

𝐿 − 1 𝑎𝑑𝑑𝑒𝑑 𝑳−𝟏 𝑳−𝟏

……..
data 𝒚𝟑 (𝒏) data

Output Sequence,𝒚(𝒏)
 # Let 𝑥 𝑛 = 𝑛 + 1 , 0 ≤ 𝑛 ≤ 9 and ℎ 𝑛 = 1,0,1 . Compute the linear convolution 𝑦 𝑛 = 𝑥 𝑛 ∗ ℎ(𝑛)
using the overlap add method.

𝑥 𝑛 ⇨
𝑡𝑜𝑡𝑎𝑙 𝑙𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑡ℎ𝑒 𝑖𝑛𝑝𝑢𝑡 𝑠𝑒𝑞𝑢𝑒𝑛𝑐𝑒 10
Number of blocks required= = =3
𝑙𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑒𝑎𝑐ℎ 𝑑𝑎𝑡𝑎 𝑏𝑙𝑜𝑐𝑘 4
𝑥0 𝑛 ⇨
𝑁 𝑝𝑜𝑖𝑛𝑡 𝐶𝑖𝑟𝑐𝑢𝑙𝑎𝑟 𝐶𝑜𝑛𝑣𝑜𝑙𝑢𝑡𝑖𝑜𝑛
𝑥1 𝑛 ⇨

𝑥2 𝑛 ⇨
ℎ 𝑛

𝑦0 𝑛 ⇨

𝑦1 𝑛 ⇨ 𝐿 − 1 𝑎𝑑𝑑𝑒𝑑
𝐿 − 1 𝑎𝑑𝑑𝑒𝑑
𝑦2 𝑛 ⇨

𝑦 𝑛 ⇨
 Multiplication Property
𝑁 𝑁
𝐼𝑓 𝑥1 𝑛 𝑋1 𝑘 and 𝑥2 𝑛 𝑋2 𝑘 , 𝑡ℎ𝑒𝑛
𝑁 1
𝑥1 𝑛 𝑥1 𝑛 [𝑋1 𝑘 ⊛ 𝑋2 𝑘 ]
𝑁

Parseval’s Relation
𝑁−1 𝑁−1
𝑁 𝑁 1
𝐼𝑓 𝑥 𝑛 𝑋 𝑘 and 𝑦 𝑛 𝑌 𝑘 , 𝑡ℎ𝑒𝑛 𝑥 𝑛 𝑦∗ 𝑛 = 𝑋(𝑘)𝑌 ∗ (𝑘)
𝑁
𝑛=0 𝑘=0

∗
𝑁−1 𝑁−1 𝑁−1 𝑁−1 𝑁−1
1 1
𝑥 𝑛 𝑦∗ 𝑛 = 𝑥 𝑛 𝑌(𝑘)𝑊𝑁 𝑘𝑛 = 𝑥 𝑛 𝑌 ∗ (𝑘)𝑊𝑁 −𝑘𝑛
𝑁 𝑁
𝑛=0 𝑛=0 𝑘=0 𝑛=0 𝑘=0

𝑁−1 𝑁−1 𝑁−1

1 1
= 𝑌 ∗ (𝑘) 𝑥(𝑛 𝑘 𝑊𝑁 −𝑘𝑛 = 𝑋(𝑘)𝑌 ∗ (𝑘)
𝑁 𝑁
𝑘=0 𝑛=0 𝑘=0
𝑊ℎ𝑒𝑛 𝑦(𝑛) = 𝑥(𝑛)

𝑁−1 𝑁−1
2
1
|𝑥(𝑛)| = |𝑋(𝑘)|2
𝑁
𝑛=0 𝑘=0
 Conjugate and Conjugate Symmetry Property
𝑁
𝐼𝑓 𝑥 𝑛 𝑋 𝑘 𝑡ℎ𝑒𝑛
𝑁
∗
𝑥 𝑛 𝑋 ∗ < −𝑘 >𝑁 = 𝑋 ∗ 𝑁 − 𝑘 𝑎𝑛𝑑
𝑁
∗ ∗
𝑥 𝑁 − 𝑛 = 𝑥 < −𝑛 >𝑁 𝑋∗ 𝑘
Proof: ∗
𝑁−1 𝑁−1
−𝑗2𝜋𝑘𝑛 2𝜋
−𝑗 (−𝑘)𝑛
𝐷𝐹𝑇[𝑥 ∗ 𝑛 ]= 𝑥∗ 𝑛 𝑒 𝑁 = 𝑥 𝑛 𝑒 𝑁
𝑛=0 𝑛=0
𝑁
∗ ∗
𝑥 𝑛 𝑋(< −𝑘 >𝑁 = 𝑋 ∗ < −𝑘 >𝑁 = 𝑋 ∗ 𝑁 − 𝑘

𝐼𝐷𝐹𝑇 𝑋 ∗ 𝑘 ∗
𝑁−1 𝑁−1
1 𝑗2𝜋𝑘𝑛 1 2𝜋
+𝑗 𝑁 (−𝑘)𝑛
= 𝑋∗ 𝑘 𝑒 𝑁 = 𝑋 𝑘 𝑒 = 𝑥(< −𝑛 >𝑁 ) ∗
𝑁 𝑁
𝑛=0 𝑛=0
= 𝑥 ∗ < −𝑛 >𝑁
Case I: x(n) is real
𝑁
Since 𝑥∗ 𝑛 𝑋 ∗ < −𝑘 >𝑁
Hence
𝑋 𝑘 = 𝑋 ∗ < −𝑘 >𝑁 = 𝑋 ∗ 𝑁 − 𝑘
𝑋 ∗ 𝑘 = 𝑋 < −𝑘 >𝑁 = 𝑋(𝑁 − 𝑘) ⇨ Conjugate Symmetric

Case II: x(n) is real and circularly even

𝑥 𝑛 = 𝑥 ∗ 𝑛 = 𝑥 < −𝑛 >𝑁
Hence
𝑋 𝑘 = 𝑋 ∗ < −𝑘 >𝑁 = 𝑋 < −𝑘 >𝑁
i.e.
𝑋 𝑘 = 𝑋 ∗ 𝑁 − 𝑘 = 𝑋(𝑁 − 𝑘) ⇨ Real & Circularly Even
Case III: x(n) is real and circularly odd
𝑥 𝑛 = 𝑥 ∗ 𝑛 = −𝑥 < −𝑛 >𝑁
Hence
𝑋 𝑘 = 𝑋 ∗ < −𝑘 >𝑁 = −𝑋 < −𝑘 >𝑁
i.e.
𝑋 𝑘 = 𝑋 ∗ 𝑁 − 𝑘 = −𝑋(𝑁 − 𝑘) ⇨ Imaginary & Circularly Odd

Case IV:
𝑁
𝐶𝑜𝑛𝑠𝑖𝑑𝑒𝑟 𝑥 𝑛 𝑋 𝑘
1
𝑥𝑐𝑒 𝑛 = 𝑥 𝑛 + 𝑥 < −𝑛 >𝑁
2
1 1
𝐷𝐹𝑇 𝑥𝑐𝑒 𝑛 = 𝐷𝐹𝑇 𝑥 𝑛 + 𝑥 < −𝑛 >𝑁 = 𝑋 𝑘 + 𝑋∗ 𝑘 = 𝑋𝑅 𝑘
2 2
1
𝑥𝑐𝑜 𝑛 = 𝑥 𝑛 − 𝑥 < −𝑛 >𝑁
2
1 1
𝐷𝐹𝑇 𝑥𝑐𝑜 𝑛 = 𝐷𝐹𝑇 𝑥 𝑛 − 𝑥 < −𝑛 >𝑁 = 𝑋 𝑘 − 𝑋 ∗ 𝑘 = 𝑗𝑋𝐼 𝑘
2 2