The Convolution Sum for DT LTI Systems The Convolution Sum for Discrete-Time LTI Systems Andrew W. H. House 01 June 2004 1 The Basi cs of the Convol ut ion Sum Consider a DT LTI system, L. x(n) −→ L −→ y(n) DT convolution is based on an earlier result where we showed that any signal x(n) can be expressed as a sum of impulses. x(n) = ∞ k=−∞ x(k)δ(n − k) So let us consider x(n) written in this form to be our input to the LTI system. y(n) = L [x(n)] = L ∞ k=−∞ x(k)δ (n − k) This looks like our general linear form with a scalar x(k) and a signal in n, δ (n − k). Recall that for an LTI system: • Linearity (L): ax 1 (n) + bx 2 (n) −→ L −→ ay 1 (n) + by 2 (n) • Time Invariance (TI): x(n − n o ) −→ L −→ y(n − n o ) We can use the property of linearity to distribute the system L over our input. y(n) = L ∞ k=−∞ x(k)δ(n − k) = ∞ k=−∞ x(k)L [δ(n − k)] So now we wonde r, what is L [δ(n − k)]? Well, we can figure it out. Suppose we know how L acts on one impulse δ(n), and we call it h(n) = L [δ(n)] ENGI 7824 – Discrete-Time Systems and Signals 1
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
This means that if we know one input-output pair for this system, namely
δ(n) −→ L −→ h(n)
then we can inferx(n) −→ L −→ y(n)
which gives us the following.
y(n) =∞
k=−∞x(k)h(n− k)
This is the convolution sum for DT LTI systems.
The convolution sum for x(n) and h(n) is usually written as shown here.
y(n) = x(n) ∗ h(n) =∞
k=−∞
x(k)h(n− k)
1.1 Comments on the DT Convolution Sum
1. A system’s impulse response, h(n), completely characterizes the behaviour of the sys-tem.
2. The impulse response h(n) can be generated directly from δ(n) through L, since δ(n) isan actual signal in DT. Thus, we can actually find the impulse response experimentally.
3. Compare convolution in DT and CT.
• DT convolution has an output variable n and a dummy variable k which causesa shift and flip.
DT y(n) = x(n) ∗ h(n) =∞
k=−∞
x(k)h(n − k)
• CT convolution has one signal in terms of the dummy variable, and the othershifted and flipped on the dummy variable, but centred on the output variable.
4. CT convolution is a model of behaviour of CT systems.
DT convolution is a model of behaviour of DT systems, but also an algorithm we canuse to implement the system, since it is computable. This is another reason why we
prefer DT signals and systems.
2 Examples of Convolution
Convolution is best understood when seen in action. Let’s look at a couple of examples,one using signals and impulse responses defined functionally, and another with an impulseresponse defined point-wise.
Example 2.1: DT Convolution: Step Response
Say we are given the following signal x(n) and system impulse response h(n).
x(n) = u(n) and h(n) =
1
2
n
u(n)
We wish to find the step response s(n) of the system (i.e. the response of thesystem to the unit step input x(n) = u(n). This is shown below.
s(n) = x(n) ∗ h(n) =∞
k=−∞
x(k)h(n− k)
Thus the step response is as follows, found by substituting our actual signals intothe general convolution sum.
s(n) =∞
k=−∞
u(k)
1
2
n−k
u(n− k)
Let’s look at this step response in smaller ranges to see what happens.
Here, s(n) = 0. This is because u(n − k) (and the associated exponential)will be starting at a point less than 0 in the k domain, and will extend to−∞, whereas u(k) starts at 0 and extends to +∞. We can visualize this,say for a value of n = −2.
Notice that there is no non-zero overlap of x(k) and h(n − k). Since theyare multiplied together, the zero part of one signal cancels out the non-zeropart of the other, and vice versa. Thus, s(n) = 0 for n < 0.
• The more interesting case is when n ≥ 0.
Recall the convolution sum we are using to determin s(n).
s(n) =∞
k=−∞u(k)1
2n−k
u(n− k)
Note that u(k) means we know the summation will be 0 for all values of k < 0, so we can change the lower limit of the summation to 0. Similarly,the u(n − k) term means that the summation for all values of k > n willbe 0, since that unit step is flipped and extends toward −∞. So, we canchange the upper limit of the summation to n. In the range 0 ≤ k ≤ n,both of the unit steps will have a value of 1. This is shown below.
Now we have a form consistent with a geometric series. We can use that tosolve.
Recalln
k=0
2k =1− 2n+1
1− 2= 2n+1
− 1
So we have s(n) as follows.
s(n) =
1
2
n 2n+1
− 1
=
1
2
n
(2 · 2n− 1)
=
1
2
n
2 ·
1
2
−n
− 1
= 2 ·
1
2−n
1
2n
− 1 ·
1
2n
s(n) = 2−
12
n
We can visualize this, say for n = 2, as shown below. Note how the systemoutput comes from the overlap of the input signal and the shifted and flippedimpulse response.
The u(n) comes from our first case above since s(n) = 0 for n < 0, and obviouslythe other part comes from the expression found in the second case above.
Now consider some variations on this first example.
Example 2.2: Variations on the Step Response
1. How would the system in the previous example react if the input was x(n) =u(n)− u(n− 4)?
We could work this through mathematically, using the convolution sum,but that is not necessary in this case. Remember, our system is LTI. (If thesystem is not LTI, there is no valid impulse response). Since the system isLTI, we can break down its response.
These variations have shown how we can deal with more complex systems ascombinations of simpler systems that are often already known.
Now let’s consider an example which is not so nicely mathematically defined.
Example 2.3: Graphical Convolution
Even though the convolution sum is nicely defined, sometimes it can’t be nicelyworked out. Thus, it is also useful to understand the concept of how the convo-lution sum works.
In the convolution sum, the impulse response is written as h(n−k), meaning thatin the k domain, the impulse response is shifted by n and flipped around thatpoint. We can visualize the convolution operation as that shifted-and-flippedimpulse response sliding along the k axis from −∞ to ∞ as the summationoccurs. Whenever there is some non-zero overlap between this shifted-and-flippedimpulse response and the input signal, the system output will be non-zero (unlessthe non-zero overlaps cancel each other).
but we have no convenient functional representation of h(n) to allow us to solve.
Consider the solution in a piece-wise fashion.
For n < 4, y(n) = 0 since there is only zero overlap between the two signals.This is illustrated below.
For n ≥ 4, we need to visualize what is happening in order to determine thevalue of y(n). In the figure below, we can see how h(n − k) slides along the k
x(n) ∗ [h(n) + g(n)] = x(n) ∗ h(n) + x(n) ∗ g(n)This is significant to all parallel connections because it means the following two arrangementsare equivalent.
x(n) −→ h(n) + g(n) −→ y(n)
is the same as
3.4 Identity
We have previously established that δ(n) is the identity with respect to discrete-time convo-lution.
Recall x(n) =
∞k=−∞
x(k)δ(n− k) = x(n) ∗ δ(n)
So x(n) ∗ δ(n) = x(n).
This concept is quite easily extended, so x(n) ∗ δ(n − no) = x(n − no) for no ∈ Z andx(n− no) ∗ δ(n− n1) = x(n− (no + n1)) for no, n1 ∈ Z.
Example 3.1: Convolution Properties
1. Consider the following interconnection of systems.
2. Consider the following interconnection of systems.
For the above systems, we can apply the same principles as in the first partof the example. Once we have found the overall impulse response, we canfind the output just as with any other impulse response (albeit with a lotmore tedium).
As we have seen, it is possible to build more complex systems from interconnec-tions of simpler ones, and still have an overall impulse response to represent thesystem.