Numerical Convolution

EE 327

Numerical Convolution
Numerical evaluation of the convolution
integral
For difficult-to-solve convolution problems
We will focus on the Matlab conv function

Discretizing the Time

Convolution Integral

y (t ) = h( )x(t )d

Discretize the Time

t = nT
n
sample value
T
sampling period

y (nT ) = h( )x(nT )d

Break the Integral into Pieces

(T-second intervals)
T

2T

y (nT ) = L + h( )x(nT )d + h( )x(nT )d

iT +T

h( )x(nT )d + L

iT
iT +T

h( )x(nT )d
i = iT

Let T be sufficiently small that h( )x(nT )

is relatively constant over T
Then

h( ) = h(iT )

for iT < iT + T
x(nT ) = x(nT iT )

y (nT ) =

iT +T

h(iT )x(nT iT )d
i = iT

iT +T

i =

iT

h(iT )x(nT iT ) d

h(iT )x(nT iT )(iT + T iT )

i =

T h(iT )x(nT iT )
i =

y (nT ) =

T h(iT )x(nT iT )
i =

In discrete-time notation

y[n] =

T h[i ]x[n i]
i =

Numerical Convolution in Matlab

Use conv function
yy = conv(T*hh,xx);
yy = conv(hh,T*xx);

Keys to Numerical Convolution

Convert to discrete time
The smaller the sampling period, T, the
more exact the solution
Tradeoff
computation time vs.
exactness of solution
Remember to account for T in the
convolution
ex. T*hh