Scribd
Upload
47 views2 pages

Courses Circuits and Systems 2009 Convolution

This document discusses discrete time convolution in Matlab. It explains that convolution is defined as the sum of one signal multiplied by time-shifted versions of another. It then shows several mathematical manipulations to rewrite the convolution equation in a form that can be directly computed in Matlab. Specifically, it demonstrates how to shift and window the signals such that their convolution results in the first non-zero value corresponding to the desired output of the unshifted convolution.

Uploaded by

Bizura Saruma
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
47 views2 pages

Courses Circuits and Systems 2009 Convolution

This document discusses discrete time convolution in Matlab. It explains that convolution is defined as the sum of one signal multiplied by time-shifted versions of another. It then shows several mathematical manipulations to rewrite the convolution equation in a form that can be directly computed in Matlab. Specifically, it demonstrates how to shift and window the signals such that their convolution results in the first non-zero value corresponding to the desired output of the unshifted convolution.

Uploaded by

Bizura Saruma
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Discrete time convolution in Matlab

Using Matlab to compute convolution


Convolution is defined as:
+∞
X
y[n] = x[n] ∗ h[n] = x[i] · h[n − i]
i=−∞

If the signal x[n] has non zero taps only in the region [n1, n2] convolution can be rewritten as:

n2
X
y[n] = x[i] · h[n − i]
i=n1

This happens because all other terms of the sum are zero.
We can also do a change of variable and rewrite the sum as:

X1 +1
n2 −n
y[n] = x[i + n1 − 1] · h[n − (i + n1 − 1)]
i=1

Now observe that x ′[i] = x[i + n1 − 1] is a shifted version of x[n] with the first non zero tap at
i = 1.
X1 +1
n2 −n
y[n] = x ′[i] · h[(n − n1 + 1) − i]
i=1

We can do now yet another manipulation and obtain:

X1 +1
n2 −n
y[n − n1] = x ′[i] · h[n + 1 − i]
i=1

Assume now that the signal h[m] is non zero in the interval [m1, m2]. At any given n the pre-
vious multiplication will use taps from [n + 1 − n2 + n1 − 1, n + 1 − 1] = [n − (n2 − n1), n]. There-
fore for any n < m1 and n − (n2 − n1) > m2 the signal y[n + n1] is equal to 0.
We can further modify the formula using h ′[k] = h[k + m1 − 1], a shifted version of h[k] that has
its first non zero tap at k = 1.

X1 +1
n2 −n
y[n − n1] = x ′[i] · h ′[n + 1 − i − m1 + 1]
i=1

We can further rewrite the expression as:

X1 +1
n2 −n
y ′[n] = y[n − n1 − m1 + 1] = x ′[i] · h ′[n + 1 − i]
i=1

Since h ′[n] is non zero only in the [1, m2 − m1 + 1] we see that only for values of n > 1 it can be
non zero. This implies that y[n] has its first non zero tap at n1 + m1.

1
The last summation is exactely what matlab computes (you can see this in the documentation).
Therefore y ′[n] = conv(x ′[n], h ′[n]) and we can obtain y[n] by shifting conv(x ′[n], h ′[n]) by n1 +
m1 + 1 to the left. The matlab output contains the non zero taps of y ′[n], wich we computed
start at 1, so the first entry of the vector returned by matlab is corresponds to y[m1 + n1].

Computing convolution of signal with infinite support


Clearly if both signals we are convolving have infinite support we cannot compute y[n] with
Matlab. If only one of them has infinite support we can compute part of the convolution by
taking a window of the signal. The values of the convolution for all taps for which the finite sup-
port signal “fits” in the window of the other signal will correct.
Now we will compute for which values exactely the convolution is correct. Assume we have a
signal x[n] which support is contained in [n1, n2] and we take a window hw[n] of the infinite sup-
port signal h[n] in the interval [w1, w2], where:

h[n] ∀n ∈ [w1, w2]
hw[n] =
0 otherwise

We know that the convolution can be rewritten as:


n2
X
x[n] ∗ h[n] = x[i] · h[n − i]
i=n1

So as long as n − i ∈ [w1, w2] using h[n] or hw[n] doesn’t matter. In other words to compute the
tap of the convolution at time n we use taps from n − n2 to n − n1 of h[n]. So if n − n2 > w1 and
n − n1 6 w2, i.e. n ∈ [w1 + n2, w2 + n1], we have that hw[n] ∗ x[n] = h[n] ∗ x[n].
If the infinite support signal is non zero only for taps n > m1 then if we take a window of the
signal such that w1 6 m1 all values of the convolution computed for n 6 w2 + n1 are correct. This
because as long as n 6 w2 + n1 the summation takes taps of hw[n] wich are equal to h[n].

You might also like