ECT303 DIGITAL SIGNAL

PROCESSING
MODULE 1- PART III

Ms. Neethu Radha Gopan, Asst. Professor, Dept. of ECE, RSET

Circular Convolution

➢ Defined as the convolution of two periodic functions.

➢ The length of the output samples is N= Max(L,M).
𝑁−1

𝑦 𝑛 = 𝑥1 𝑛 𝑥2 𝑛 = ෍ 𝑥1 (𝑘) 𝑥2 ( 𝑛 − 𝑘 )𝑁
𝑘=0
Concentric Circle Method

➢ Given 2 sequences 𝑥1 (𝑛) & 𝑥2 (𝑛) , the circular convolution is obtained as follows:
1. Graph N samples of 𝑥1 (𝑛) as equally spaced points around an outer circle in the
counter clockwise direction.
2. Starting at the same point as 𝑥1 (𝑛), graph N samples of 𝑥2 (𝑛)as equally spaced
points around the inner circle in the clockwise direction.
3. Multiply the corresponding samples on the 2 circles & sum the products to produce
output.
4. Rotate the inner circle one sample at a time in the counter clockwise direction and
repeat step 3 to obtain the next output sample.
5. Repeat step 4 till the inner circle samples line up with the first sample of the exterior
circle once again.
 Q. Find the circular convolution of 2 finite duration sequences 𝑥1 (𝑛) =
1, −1, −2,3, −1 & 𝑥2 (𝑛) = {1,2,3} using concentric circle method

Soln:
Both sequences must be of same length N= (max[5,3]) =5. Hence , append 2 zeros to 𝑥2 (𝑛).
𝑥1 𝑛 = 1, −1, −2,3, −1 𝑥2 (𝑛) = {1,2,3,0,0}

𝑦(0) = 1 1 + 0 −1 + 0 −2 + 𝑦(1) = 2 + −1 + 0 + 0 + −3 𝑦(2) = 3 − 2 − 2 + 0 + 0

3 3 + 2 −1 = 8 = -2 = -1
𝑦(3) = 0 − 3 − 4 + 3 + 0 𝑦(4) = 0 + 0 − 6 + 6 − 1
= -4 = -1

𝑦(𝑛) = {8, −2, −1, −4, −1}

Matrix Multiplication Method

➢ In this method, the circular convolution of 𝑥1 𝑛 & 𝑥2 (𝑛), given by 𝑥3 𝑛 =

𝑥1 (𝑛) 𝑥2 (𝑛) can be obtained as follows:
𝑥2 (0) 𝑥2 (𝑁 − 1) 𝑥2 𝑁 − 2 ⋯ 𝑥2 (1) 𝑥1 (0) 𝑥3 (0)
𝑥2 (1) 𝑥2 (0) 𝑥2 𝑁 − 1 ⋯ 𝑥2 (2) 𝑥1 (1) 𝑥3 (1)
𝑥2 (2) 𝑥2 (1) 𝑥2 0 ⋯ 𝑥2 (3) 𝑥1 (2) 𝑥3 (2)
. =
⋮ ⋮ ⋮ ⋮ ⋮ ⋮
𝑥2 (𝑁 − 2) 𝑥2 (𝑁 − 3) 𝑥2 𝑁 − 4 ⋯ 𝑥2 𝑁 − 1 𝑥1 (𝑁 − 2) 𝑥3 (𝑁 − 2)
𝑥2 (𝑁 − 1) 𝑥2 (𝑁 − 2) 𝑥2 𝑁 − 3 ⋯ 𝑥2 (0) 𝑥1 (𝑁 − 1) 𝑥3 (𝑁 − 1)
➢ The sequence 𝑥2 (𝑛)is repeated via circular shift of samples (till the 𝑁 − 1𝑡ℎ
circular shift) and represented as an N×N matrix.
➢ The sequence 𝑥1 (𝑛) is represented as a column matrix . The multiplication of
these 2 matices results in 𝑥3 (𝑛)
Q. Find the circular convolution of 2 sequences 𝑥1 (𝑛) = 1, −1, −2,3, −1 &
𝑥2 (n) = {1,2,3} using Matrix method.

Soln:
To find circular convolution , both sequences must be of same length N= max(5,3)=5.
Hence , append 2 zeros to 𝑥2 (𝑛).
𝑥1 (𝑛) = 1, −1, −2,3, −1 𝑥2 (n) = {1,2,3,0,0}

1 0 0 3 2 1 8
2 1 0 0 3 −1 −2
3 2 1 0 0 . −2 = −1
0 3 2 1 0 3 −4
0 0 3 2 1 −1 −1

y(n)={8, -2, -1, -4, -1}

 Linear Filtering Using DFT

➢ Circular convolution cannot be used to determine the output of a linear filter.

➢ Hence we perform linear filtering using DFT (circular convolution).
➢ The length of y(n) after linear convolution is L+M-1.
➢ Hence, increasing the length of x(n) & h(n) by zero padding to L+M-1 and then
circularly convolving the sequences, gives the same result as that of linear
convolution.
Q. Find the output response of a filter using circular convolution if
impulse response h(n)= {1,1,1} and x(n)= {1,2,3,1}

Soln:
Length of h(n)=3, length of x(n)=4, length of linear convolution N= L+M-1=6
Hence zero pad such that x(n) and h(n) are of length 6
x(n) = {1,2,3,1,0,0} h(n) = {1,1,1,0,0,0}. Perform circular convolution.

1 0 0 0 1 1 1 1
1 1 0 0 0 1 2 3 𝑦(𝑛) = {1,3,6,6,4,1}
1 1 1 0 0 0 3 6
. =
0 1 1 1 0 0 1 6
0 0 1 1 1 0 0 4
0 0 0 1 1 1 0 1
Linear Filtering of Long Data Sequences

➢ Practical applications involve filtering of long data sequences.

➢ The block size of the data should be limited in size due to memory constraints of
digital processors.
➢ Hence, the input sequence is segmented to fixed size blocks prior processing.

➢ Successive blocks can be processed one at a time and the output blocks are
concatenated together to form the final response.
➢ There are 2 methods of filtering long data sequences:

1. Overlap Add Method

2. Overlap Save Method
Overlap Add Method

➢ The input data is divided into blocks of size L.

➢ Let M be the length of the impulse response of the filter.
➢ By appending L−1 zeros, the impulse response of FIR filter is increased in length to
N=L+M-1.
➢ M-1 zeros are padded to the L new data points to form a data sequence of length
N=L+M-1.
➢ Perform circular convolution of h(n) and 𝑥𝑘 (𝑛), where k denotes the number of
blocks.
➢ Form y(n) by overlapping the last M − 1 samples of 𝑦𝑘 𝑛 with the first M − 1
samples of 𝑦𝑘+1 (𝑛) and adding the result.
Q. Find the output y(n) of a filter whose impulse response is h(n)={1,1,1} and
input signal x(n)={3,-1,0,1,3,2,0,1,2,1} using overlap add method .

Soln:
Length of impulse response M=3. Let L=3. The data should be divided into blocks of 3.
N=L+M-1=5.
x(n)= { 3 , -1 , 0 , 1 , 3 , 2 , 0 ,1 , 2 , 1 }.
Each block should have M-1=2 zeros padded to the L=3 new data samples. Total length =N=5
𝑥1 𝑛 = { 3, −1, 0, 0, 0 }
𝑥2 (𝑛) = {1 , 3 , 2, 0, 0}
𝑥3 (𝑛) = {0 , 1 , 2, 0, 0}
𝑥4 (𝑛) = {1 , 0, 0 , 0 , 0}

ℎ(𝑛) = {1, 1, 1, 0, 0}. (L-1=2 zeros appended to make length equal to N=5)
➢ Now perform circular convolution of all data sequences with h[n]
𝑦1 (𝑛) = 𝑥1 (𝑛) ℎ(𝑛) = {3 , 2 , 2, −1, 0}
𝑦2 (𝑛) = 𝑥2 𝑛 ℎ(𝑛) = {1 , 4, 6, 5, 2}
𝑦 (𝑛) = 𝑥3 𝑛 ℎ 𝑛 = {0, 1, 3 , 3, 2}
3
𝑦4 (𝑛) = 𝑥4 𝑛 ℎ(𝑛) = {1, 1, 1, 0, 0}
3 2 2 −1 0
add
1 4 6 5 2
add
𝑦(𝑛) = {3, 2, 2, 0, 4, 6, 5, 3, 3, 4, 3, 1} 0 1 3 3 2
add
1 1 1 0 0
Overlap Save Method

➢ The input data is divided into blocks of size L.

➢ Let M be the length of the impulse response of the filter.
➢ By appending L−1 zeros, the impulse response of FIR filter is increased in length to
N=L+M-1.
➢ In order to avoid loss of data due to aliasing, the last M-1 points of each data block
are saved and these points become the first M-1 data points of the subsequent block
followed by L new data points to form a data sequence of length N=L+M-1.
➢ For the first block of data, M-1 zeros are appended at the beginning.
➢ Perform circular convolution of h(n) and each block 𝑥𝑘 (𝑛), k denotes the number of
blocks.
➢ Find y[n] by discarding the first M-1 samples of each output block 𝑦𝑘 (𝑛) and then
concatenating the rest samples.
Q. Find the output y(n) of a filter whose impulse response is h(n)={1,1,1} and
input signal x(n)={3,-1,0,1,3,2,0,1,2,1} using overlap save method .

Soln:
Length of impulse response M=3. Let L=3. The data should be divided into blocks of 3.
N=L+M-1=5.
x(n)= { 3 , -1 , 0 , 1 , 3 , 2 , 0 ,1 , 2 , 1 }.
Each block should have M-1=2 data sequences from the previous block followed by
L=3 new data samples. Total length =N=5
𝑥1 (𝑛) = { 0 , 0, 3, −1, 0 }
𝑥2 (𝑛) = {−1, 0 , 1 , 3 , 2}
𝑥3 (𝑛) = {3 , 2 , 0 , 1 , 2}
𝑥4 (𝑛) = {1 , 2 , 1 , 0 , 0}
ℎ(𝑛) = {1, 1, 1, 0, 0}. (L-1=2 zeros appended to make length equal to N=5)
➢ Now perform circular convolution of all data sequences with h(n)
𝑦1 𝑛 = 𝑥1 𝑛 ℎ(𝑛) = {−1, 0, 3 , 2 , 2}

𝑦2 𝑛 = 𝑥2 𝑛 ℎ(𝑛) = {4 , 1, 0 , 4, 6}

𝑦3 𝑛 = 𝑥3 𝑛 ℎ(𝑛) = {6 , 7, 5 , 3, 3}

𝑦4 𝑛 = 𝑥4 𝑛 ℎ(𝑛) = {1 , 3, 4 , 3, 1}

The first M-1=2 data points are discarded from each output.
𝑦(𝑛) = {3, 2, 2, 0, 4, 6, 5, 3, 3, 4, 3, 1}
HWQ. Find the output y(n) of a filter whose impulse response is h(n) ={1,2}
and input x(n) ={1, 2, -1, 2, 3, -2,-3,-1, 1, 1, 2,-1} using overlap save & overlap
add method .
Soln:
Length of impulse response M=2. Let L=3. The data should be divided into blocks of 3.
N=L+M-1=4.
x(n)= {1, 2, -1 , 2, 3, -2 , -3,-1, 1 , 1, 2,-1}.
Overlap save method
𝑥1 (𝑛) = { 0, 1, 2, −1}
𝑥2 (𝑛) = {−1 , 2 , 3, −2, }
𝑥3 (𝑛) = {−2 , −3 , −1, 1} ℎ(𝑛) = {1, 2, 0, 0}. (L-1=2 zeros appended to
𝑥4 (𝑛) = {1 , 1, 2 , −1} make length equal to N=4)
𝑥5 (𝑛) = −1 , 0, 0, 0
➢ Now perform circular convolution of all data sequences with h(n)
𝑦1 𝑛 = 𝑥1 𝑛 ℎ(𝑛) = {−2, 1 , 4 , 3}

𝑦2 𝑛 = 𝑥2 𝑛 ℎ(𝑛) = { −5, 0 , 7, 4}

𝑦3 𝑛 = 𝑥3 𝑛 ℎ(𝑛) = { 0 , − 7, −7 , −1}

𝑦4 𝑛 = 𝑥4 𝑛 ℎ(𝑛) = {−1 , 3, 4 , 3}

𝑦5 𝑛 = 𝑥5 𝑛 ℎ(𝑛) = {−1 , −2, 0 , 0}

The first M-1=1 data points are discarded from each output.
𝑦(𝑛) = {1, 4, 3, 0, 7, 4, −7, −7, −1,3, 4, 3, −2}
HWQ. Given two sequences x[n] and y[n] such that x[n]=[2 2 0 2] and
the 4 point circular convolution between x[n] and y[n] results in g[n]=[10
10 8 8] . Determine the sequence y[n].

Convolution in time domain property:

𝑦 𝑛 = 𝐼𝐷𝐹𝑇 𝑌 𝑘 = {2,2,1,1}
DFT: 𝑋 𝑘 . 𝑌 𝑘 = 𝐺(𝑘)

𝐺 𝑘
𝑌 𝑘 =
𝑋 𝑘
𝑦 𝑛 = 𝐼𝐷𝐹𝑇{𝑌 𝑘 }
𝑋 𝑘 = 6,2, −2,2
𝐺 𝑘 = 36, 2 − 2𝑗, 0, 2 + 2𝑗

𝐺 𝑘
𝑌 𝑘 = = {6,1 − 𝑗, 0,1 + 𝑗}
𝑋 𝑘
 REFERENCES
23

1. Proakis J. G. and Manolakis D. G., Digital Signal Processing, 3/e,

Prentice –Hall India , 2005.
2. P. Ramesh Babu, Digital Signal Processing, Scitech Publications (India)
Pvt Ltd.
24 END of PART -III

THANK YOU!