Вы находитесь на странице: 1из 2

Idiots Guide to Discrete-Time Convolution

Convolution can be complicated and messy. You need to be very careful to make your answers
come out correctly. When you are learning the technique, it pays to draw accurate diagrams and
follow a step-by-step procedure to eliminate (or at least reduce) errors. If you follow the steps
below, it should help considerably.


Consider the following problem of convolving ] [n x with ] [n h :

=
=
k
k n h k x n y ] [ ] [ ] [


1. (Optional) Check to see whether ] [n x or ] [n h is simpler. Exchange the order of ] [n x and
] [n h so that ] [n h is the simpler function. (Recall this is valid since convolution is
commutative).

2. Write expressions for ] [k x and ] [ k n h .

3. Sketch the function ] [k x . Note that this is the same as ] [n x except that you are using k
instead of n . Make sure to label the axis as a function of k . It is easy to get confused
between n and k .

4. Sketch the function ] [k h . Note that this is the same as ] [n h except that you are using k
instead of n .

5. Sketch the function ] [ k h . This is just the original function ] [k h reflected about the y-
axis. Also note that this is the function ] [ k n h when 0 = n . Finally, note that, since
the function is plotted with respect to k , n acts like a constant. Changing the value of
n will simply shift the function around in the usual way that constants shift functions.

6. Now sketch the function ] [ k n h . Note that positive values of n will make the function
shift to the right. This is because this plot has been rotated about the y-axis. You can
now draw the general graph of ] [ k n h by using the graph of ] [ k h and adding n to all
the values along the k-axis. I sometimes include a note to myself indicating with an
arrow that increasing n shifts the plot to the right.

7. You now need to think about multiplying the function you drew in step 3 with the
function you drew in step 6 (one value of k at a time). Of course, when you multiply two
functions and one of them is zero over a particular region, the result will be zero over this
same region. Hence, the function that goes to zero first will be the controlling function
that defines the edge of the product.

8. Slide the function in step 6 to the left over the function in step 3 until there is no overlap
(if possible). Determine how much you needed to slide this function (what value of n
shifts the function sufficiently). For all values of n that are smaller than this, the result is
zero (since the overlap is zero). Of course, if this never occurs, that means that the
convolution has values that extend to negative infinity.

9. Now shift the function in part 6 so that overlap occurs. Write the summation that
describes the product of the two functions, paying special attention to the starting and
ending points of the sum. You can simply substitute the equations you wrote down in
step 2 into the summation. Note that you can leave out all the expressions that correspond
to step functions, since they simply control the limits of the sum, and are equal to unity
everywhere within the sum. If the edge of the function in step 3 defines the edge of the
product, the limit will be a constant (given by the location of the edge in step 3).
However, if the edge of the product is given by the function in part 6 (the function that is
shifting), then the limit will be given by the limit on this function, which is a function of
n . This is the tricky part. Making sure you have these limits correct is not always easy.

10. Once you have the summation established with the proper limits, it is a simple matter to
sum the result. There are two things to note. First, all factors of n can be pulled out of
the sum since n is a constant over the summation variable k . Second, the general
summation formula is very useful in evaluating these summations:

1 | | ,
1
1
<

=
+
=

a
a
a a
a
N M N
M k
k


Of course, you may need to do some substituting of variables if your expression is not in
this form. Also, if M or N is infinite, its corresponding power in the equation becomes
zero.

11. Finally, you can often simplify the expression by consolidating terms, realizing that
n
n
a
a
= |
.
|

\
|

1
, etc.

12. You need to repeat this procedure for different sections of the overlapping functions if
shifting by n makes the controlling edge change.

13. When you are all finished, you should be able to stitch the sections together to make a
continuous function. You can check to see if you are correct by ensuring that the values
calculated for the right and left sides of each section match with their corresponding
sections.

Вам также может понравиться