EEE 303 HW # 1 Solutions
EEE 303 HW # 1 Solutions
Problem 1.5
Since x(t) = 0 for t < 3, we have that x(t) = U (t 3)x(t).
a. x(1 t) = x(1 t)U (1 t 3) = x(1 t)U (t 2) which is zero for t > 2.
b. x(1 t) + x(2 t) = x(1 t)U (t 2) + x(2 t)U (t 1) which is zero for t > 1.
c. x(1 t)x(2 t) = x(1 t)U (t 2)x(2 t)U (t 1) = x(1 t)x(2 t)U (t 2) which is zero for t > 2.
d. x(3t) = x(3t)U (3t 3) which is zero for t < 1.
e. x(t=3) = x(t=3)U (t=3 3) which is zero for t < 9.
1
EEE 303 HW#2 Solutions
1.43
a. Let H denote a TI system such that y = H[x] and suppose that x is periodic. That is, x(t+T)=x(t),
for all t. Alternatively, in terms of the shift operator T, T-T[x] = x. Now, for y to be periodic, we should also
have T-T [y] = y. Substituting y = H[x], we get
T-T[y] = T-TH[x]
= HT-T[x] (TI system, commutes with shifts)
= H[x] (x is periodic)
= y
(The derivations hold in discrete time as well.)
b. The zero system (y = 0) is an easy but trivial example. A more interesting case is a low-pass filter
eliminating the high frequency component of, say, x(t) = sin t + sin 10t. (More details later.)
EEE 303 HW#3 Solutions
2.20
a. u (t ) cos(t )dt = (t ) cos(t )dt = (t ) cos(0)dt =1
0
5
b. sin( 2t ) (t + 3)dt = 0 , because (t+3) = 0 for t in [0,5].
0
5 d [cos(2t )]
c. 5
u1 (1 ) cos(2 )d =
dt
t =1 = 2 sin( 2 ) = 0 , since 1 is in [-5,5].
d ( t )
Alternatively, using integration by parts and u1 (t ) = dt , we have that u1 (1 ) = dd (1 ) .
Then,
d (1 )
5 5cos(2 )d =
5 u1 (1 ) cos(2 )d = 5 d
[ ]
(1 t ) cos(2t ) 5 + (1 )
5 5
5
d cos 2
d
d = sin( 2 1) = 0
where we used the fact that (1-5)= (1+5) = 0.
2.47
Let y0 = h0*x0. Then,
a. [ h0 (t )] * [ 2 x 0 (t )] = 2(h0 * x0 ) = 2 y 0 , (linearity of convolution)
b. [ h0 (t )] * [ x 0 (t ) x 0 (t 2)] = ( h0 * x 0 )(t ) [ h0 (t )] * [ x 0 (t 2)] = y 0 (t ) y 0 (t 2) ,
where the first equality follows from linearity and the second from time-invariance.
c. Using time-invariance, [ h0 (t + 1)] * [ x 0 (t 2)] = ([h0 (t )] * [ x 0 (t 2)])(t + 1) .
But,[h0 (t )] * [ x0 (t 2)] = ([h0 (t )] * [ x0 (t )])(t 2) = y 0 (t 2) .
Hence, [ h0 (t + 1)] * [ x 0 (t 2)] = y 0 (t + 1 2) = y 0 (t 1) .
Alternatively, using the definition of the shift operator T and its commutativity with convolution,
[h0 (t + 1)] * [ x0 (t 2)] = (T1 h0 ) * (T2 x0 ) = T1T2 (h0 * x 0 ) = T1 y 0 .
d. There is not enough information to determine the output in this case. To prove this statement we need to
find two pairs (h0,x0) that produce the same response y0 and such that their corresponding transformations
produce different responses. For this, consider h0 (t) = U(t) (a unit step) and x0(t) = {a pulse in [0,2] with
amplitude }. As a second pair, consider h0 (t) = y0(t), x0(t) = (t). For the first pair, it follows that
[h0(t)]*[x0(-t)] = y0(t+2). But for the second pair, [h0(t)]*[x0(-t)] = y0(t).
e. Reflections are not time-invariant operations so we need to work with the convolution integral itself.
Using a transformation of the integration variable = , we have that y0(-t) can be expressed as:
y 0 (t ) = h0 (t )x0 ( )d = h0 (t + )x0 ( )d = h(t )x( )d = (h * x)(t )
Notice that this is not a simple transformation of the independent variable. That is, we may not simply
state that since y0 (t) = h0(t)*x0(t), then y0 (-t) = h0(-t)*x0(-t). Even though the final relationship is true, this
statement is wrong as an implication! The error is caused by the bad notation (h(t)*x(t) is nonsense).
This error becomes apparent when we consider scaling of the independent variable. That is, using the
previous derivation it follows that y0 (at) =|a| [h0(at)]*[x0(at)] (and not h0(at)*x0(at), as one might
expect!)
f. Using the property of the unit doublet u1, that x = u1*x, we have that
y = h * x = h' 0 *x' 0 = (u1 * h0 ) * (u1 * x0 ) = u1 * u1 * (h0 * x0 ) = y 0''
From the graph of y0, it follows that y(t) = (t) (t-2).
EEE 303 HW # 4 SOLUTIONS
Problem 2.23
Using the linearity and time invariance of convolution, we have that (T denotes the shift)
!
X
hx = h TkT
k
X
= h (TkT )
k
X
= TkT (h )
k
X
= TkT h
k
So, nally, X
y(t) = h(t kT )
k
T=4
T=2
T = 1.5
T=1
1
EEE 303 HW#4 Solutions
3.15 The Fourier series coefficients of the output y(t), say bk are related to the Fourier series coefficients of
the input (ak) by
bk = H(jkw0) ak
where w0 = 2 / T = 12. Since H(jw) = 0 for |w| > 100, it follows that bk = 0 for |12 k| > 100, or |k| >
100/12, or |k| 9. Since it is given that x(t) = y(t), then ak = bk (uniqueness of the FS expansion), implying
that ak must be zero for |k| 9.
EEE 303 HW Solutions
3.22.a.a
Consider the derivative of the given function:
dx
(t ) = 1 2 (t 2n 1)
dt n =
The train of deltas is periodic with period 2. Its Fourier Series expansion can be found either directly
(integrals of delta-times-a-function) or by using the tables and applying the shift property. We have,
e jkw0
2
a k = FS (t 2n 1) = 1
2 (t 1)e dt =
jkw t
0
n 0
2
where w0 = 2/2 = . The FS coefficients of the function 1 are 1 if k = 0 and 0 otherwise. Thus, the
coefficients of the derivative are
1 e jk if k = 0
bk = FS { dx
(t )} = jk
e
dt
otherwise
Next, the FS coefficients of x(t), say ck, are related to bk by bk = jkw0 ck. This can be solved for ck except
for k = 0. This can be found by applying the FS definition directly to x(t). Thus, we find
0 if k = 0
c k = FS {x(t )} = jk
e /( jk ) otherwise
Notice that the direct computation of c0 here is essential. Both x(t) and x(t)+const. have the same derivative
and their FS cannot be distinguished after differentiation.
| c k | = 2 2 = 2 2 2 = t 2 dt =
2
k 0 k k =1 k 2 1 3
This yields the well-known expression
1 2
k =1 k
2
=
6
3.22.a.b
Using the same technique as in a.a, let x0 be the square wave
1 | t |< 0.5
x0 (t ) = , x0 (t + 6) = x0 (t )
0 otherwise
whose Fourier Series coefficients are given in Table 4.2. Differentiating the given signal, we find
dx
(t ) = x0 (t + 1.5) x0 (t 1.5)
dt
Hence, applying the linearity and time-shift properties of the Fourier Series, we get
sin(k / 6) jk / 2 sin(k / 6)
FS {dx }
dt (t ) =
k
e ( )
e jk / 2 = 2 j
k
sin( k / 2)
Next, using the differentiation property, we find
sin( k / 6)
a k = FS{x(t )} = FS {dx }
dt (t ) /( jkw0 ) = 6 sin(k / 2), k 0
(k ) 2
while for k = 0, the FS coefficient is found by direct integration as a0 = (2+1/2+1/2)/6 = .
EEE 303 HW # 9 SOLUTIONS
Problem 3.16b
x2 [n] is composed of two periodic functions, so we can apply superposition:
Alternatively, one could write the Fourier series expansion for x2 and use the ltering property.
Problem 3.32
The set of equations for ak is
fnk gn;k fak gk = fx[n]gn
where f:gk denotes the vector or matrix indexed by k and = ej2=4 . Expanding the matrices, we get
2 0 32 3 2 3 2 3
0 0 0 a0 x(0) 1
6 0 1 2 3 7 6 a1 7 6 x(1) 7 6 0 7
6 0 76 7 6 7 6 7
4 2 4 6 5 4 a2 5 = 4 x(2) 5 = 4 2 5
0 3 6 9 a3 x(3) 1
| {z }
The matrix has a very special structure. Its columns are of the form [1; i ; 2i ; 3i ]> . Such matrices are called
Vandermonde matrices and they are nonsingular if the i 's are distinct. In our case 1 = 0 = 1, 2 = 1 = j,
3 = 2 = 1, 4 = 3 = j, which are indeed distinct.
In general, Vandermonde matrices are numerically ill-conditioned except when the i 's are the n-th roots of
unity, which is our case. When this happens,
P the columnsPof the matrix are orthogonal to each other and the
matrix is easy to invert. (Observe that k (kn1 ) kn2 = k k(n2 n1 ) which is 0 if n2 6 = n1 . This property is
precisely the reason why we can write a simple expression for the Fourier series coecients ak .)
Regardless of any simplications, we can use Matlab (or any other numerical package) to solve the above
system of equations. Substituting the numerical values we have
2 32 3 2 3
1 1 1 1 a0 1
6 1 j 1 j 7 6 7 6 7
6 7 6 a1 7 = 6 0 7
4 1 1 1 1 5 4 a2 5 4 2 5
1 j 1 j a3 1
1
EEE 303 HW # 5 SOLUTIONS
Problem 4.34
a. From the denition of the transfer function as the Fourier transform of the impulse response, we have
Y (jw) (jw) + 4
= )
X(jw) (jw)2 + 5(jw) + 6
[(jw)2 + 5(jw) + 6]Y (jw) = [(jw) + 4]X(jw)
Taking the inverse Fourier transform of both sides, we recover the dierential equation that describes the
input-output relationship for the system
d2 d d
y(t) + 5 y(t) + 6y(t) = x(t) + 4x(t)
dt dt dt
b. We have that h(t) = F 1 fH(jw)g; using a partial fraction expansion, the roots of the denominator are
2; 3 and
A B
H(jw) = +
jw + 2 jw + 3
where A = [(jw + 2)H(jw)]jjw=2 = 2 and B = [(jw + 3)H(jw)]jjw=3 = 1. From the tables,
1 1 2 1 1
F fH(jw)g = F +F = 2e2t U (t) e3t U (t)
jw + 2 jw + 3
c. The response of the system to the input x(t) = e4t U (t) te4t U (t) can be found by either a direct
evaluation of the convolution of the impulse response with the input or by Fourier transform methods. Using
the latter approach, we nd the Fourier transform of x(t) as
1 1
X(jw) =
jw + 4 (jw + 4)2
1
b. The impulse response of the system is the inverse Fourier transform of its transfer function:
1 1 a + a a jw 1 2a
h(t) = F fH(jw)g = F =F 1 = 2aeat U (t) (t)
a + jw a + jw
can be evaluated through Fourier transforms. X(jw) will contain 6 impulses and so will the product X(jw)H(jw).
Then the inverse transform will contain 6 complex exponentials, but the simplication of that formula is quite
tedious.
Alternatively, using the sinusoid response formula and for a = 1,1
p p p p
y(t) = cos(t= 3 2 tan1 (1= 3)) + cos(t 2 tan1 (1)) + cos( 3t 2 tan1 ( 3))
p p
= cos(t= 3 =3) + cos(t =2) + cos( 3t 2=3)
The following gure shows the response of the all-pass lter to the three cosines. For comparison, the
response of the lter to the input is x(t)U (t) is also included. In the latter case, notice that the appearance of
the (t) in the impulse response causes the initial output to be y(0+) = x(0+) = 3.
output response to 3 cosines Transient vs. steadystate response detail
3 3
2 2
1 1
0 0
1 1
2 2
3 3
20 0 20 40 60 80 100 5 0 5 10
Figure 1: Left: Output response to the three cosines. Right: Detail showing the dierence between the steady-
state response and the transient response. (Steady-state response: solid line.)
2
EEE 303 HW # 9 SOLUTIONS
Problem 5.10
P1
Let A = 0 n 21n and dene the sequence x[n] = 1
2n u[n]. Then X(ejw ) = 1
1ejw =2 . The transform of the
jw jw
sequence nx[n] is j dX(e
dw
)
= e =2
(1ejw =2)2 . Then, A = Ffnx[n]gjej0 = 1=2
(11=2)2 =2
Problem 5.19
By inspection, the transfer function is
1
H(z) =
1 16 z 1 16 z 2
where z = ejw .
The poles (roots of denominator) are z = 1=2 and 1=3. Taking the Partial Fraction Expansion we have
3 2
1 5 5
H(z) = 1 1 1 1 = 1 1 + 1 1
(1 2z )(1 + 3z ) 1 2z 1+ 3z
Note: The impulse response can also be computed directly from the recursion. First, since the system is
linear, its response to the zero input sequence is zero. Furthermore since it is causal, then its response to [n]
must be identical to the zero-input response up to the index n = 0. So, y[n] = 0, for n < 0. Then,
y[0] = 0 + 0 + x[0] = 1
y[1] = 1=6 0 + 0 = 1=6
y[2] = (1=6)(1=6) + 1=6 + 0 = 7=36
..
.
1
EEE 303 HW # 6 SOLUTIONS
Problem 6.9
The lter is stable so the steady-state response is well-dened. The steady-state part of the step is the
2
constant function 1. The lter transfer function is H(jw) = jw+5 which at w = 0 is 2/5. Therefore, the
steady-state part of the step response is the constant function (2/5)(1) = 2/5, which is also the nal value of
ys (t); t ! 1.
2 1
The step response is Ys (s) = s+5 s . Taking the inverse Laplace via partial fraction expansion we nd:
2 1 2=5 2=5 2 2 2
ys (t) = L1 = L1 + = u(t) e5t u(t) = (1 e5t )u(t)
s+5s s s+5 5 5 5
The same result is obtained by using the Fourier transform, but notice the dierence in the transform of the
step function:
2 1 2 2 1
Ys (jw) = + (w) = (w) +
jw + 5 jw 5 jw + 5 jw
2 2 1 2 2=5 2=5
ys (t) = F 1 fYs (jw)g = F 1 (w) + = F 1 (w) + +
5 jw + 5 jw 5 jw jw + 5
2 1 2=5 2 2
= F 1 + (w) + = u(t) e5t u(t)
5 jw jw + 5 5 5
As t ! 1, this expression yields ys (1) = 2=5, as found in the rst part. Then, the time t0 at which
ys (t0 ) = ys (1)(1 e2 ) is t0 = 2=5.
Problem 6.12
H(jw)
The transfer function of the cascade connection is H(jw) = H1 (jw)H2 (jw). Hence, H2 (jw) = H1 (jw) .
Taking the magnitude and log of both sides,
log10 jH2 (jw)j = log10 jH(jw)j log10 jH1 (jw)j ) jH2 (jw)jdB = jH(jw)jdB jH1 (jw)jdB
for < 1w < 8: straight line with slope (0)-(20) = -20 dB/dec;
for 8 < w < 40: straight line with slope (-40)-(0) = -40 dB/dec;
for 40 < w: straight line with slope (-40)-(-20) = -20 dB/dec;
1
EEE 303 HW # 6 SOLUTIONS
Problem 6.27
(a.) Using the dierentiation property of the Fourier transform (or Laplace) the transfer function of the
system is:
1
H(jw) =
jw + 2
or, H(s) = 1=(s + 2). The Laplace version is more advantageous since in the subsequent derivations it is
convenient to keep jw as one argument.
With s = jw, the frequency response of this transfer function has the following magnitude and phase:
1
jH(jw)j = p
w 2 + 22
6H(jw) = tan1 (w=2)
These are easy to plot; in MATLAB, the relevant command is bode([1],[1 2]); (see also help bode).
(b.) For the group delay, dierentiate tan1 (w=2) with respect to w.
(c. & d.) When x(t) = et U(t), F fxg = 1=(s + 1), with s = jw. Using the convolution property,
1
Y (jw) =
(s + 2)(s + 1) s=jw
s+1 A1 A2
Y (s) = 2
= +
(s + 2) (s + 2) (s + 2)2
Converting the PFE back to one fraction we have that
A1 s + 2A1 + A2 = s + 1 ) A1 = 1; A2 = 1
1
EEE 303 HW # 6 SOLUTIONS
Problem 7.4
The Nyquist rate of x(t) is w0 , so x(t) is bandlimited in (w0 =2; w0 =2). In other words, the maximum
frequency where x(t) has energy is wM = w0 =2.
1. F fx(t 1)g = ejw F fx(t)g is also bandlimited in (wM ; wM ). Hence, F fx(t 1) + x(t)g is bandlim-
M; wM ), and the Nyquist rate of x(t 1) + x(t) is 2wm = w0 .
ited in (w
2. F dx dt = jwF fx(t)g which is again bandlimited in (wM ; wM ) and, hence, the Nyquist rate of dx=dt
is w0 .
Notice that this result hinges on the assumption that X(jw) is zero for jwj > wM . This can be deceptive in
a practical situation where the Nyquist rate is dened in terms of the signal \bandwidth," i.e., the frequency
beyond which the energy content of the signal is small. In such a case, dierentiation of the signal will increase
the magnitude (and energy) of the signal at high frequencies and will change the eective Nyquist rate.
1
3. Let y(t) = x2 (t). Then F fyg = 2 F fxg F fxg. Since F fxg is bandlimited in (wM ; wM ), we have
that F fyg is bandlimited in (2wM ; 2wM ); hence the Nyquist rate of y(t) is 2w0 .
1
4. Let y(t) = x(t) cos(w0 t). Then F fyg = 2 F fxg F fcos w0 tg = 12 [X(j(w w0 )) + X(j(w + w0 ))]. Hence,
F fyg is bandlimited in (w0 wM ; w0 + wM ) = (3wM ; 3wM ) and the Nyquist rate of y(t) is 6wM = 3w0 .
Problem 7.22
Let x1 (t) be bandlimited in (1000; 1000) and x2 (t) be bandlimited in (2000; 2000).
a. Let y(t) = (x1 x2 )(t). Then F fyg = F fx1 g F fx2 g which is bandlimited in the intersection of the
two frequency intervals, that is (1000; 1000). Hence the Nyquist rate of y(t) is 2000 and the maximum
sampling time to allow perfect reconstruction is T = 2=(2000) = 1=1000.
b. Let y(t) = 2x1 (t)x2 (t). Then F fyg = F fx1 g F fx2 g which is bandlimited in the frequency interval
(1000 2000; 1000 + 2000). Hence the Nyquist rate of y(t) is 6000 and the maximum sampling time to
allow perfect reconstruction is T = 2=(6000) = 1=3000.
1
EEE 303 HW # 7 SOLUTIONS
Problem 8.4
Since v(t) = g(t) sin 400t is passed through an ideal lowpass lter and this operation is best described in
the frequency domain, we need to compute V (jw). This can be achieved by either trigonometric identities (for
this specic problem) or by performing the various frequency-domain convolutions.
Let w0 = 200. Then,
2 2
X(jw) = (w w0 ) (w + w0 ) + (w 2w0 ) (w + 2w0 )
j j j j
1
G(jw) = X(jw) (w 2w0 ) (w + 2w0 )
2 j j
1 1
= X(j(w 2w0 )) X(j(w + 2w0 ))
2j 2j
1
V (jw) = G(jw) (w 2w0 ) (w + 2w0 )
2 j j
1 1
= G(j(w 2w0 )) G(j(w + 2w0 ))
2j 2j
1 1
= X(j(w 4w0 )) X(jw)
4 4
1 1
X(jw) + X(j(w + 4w0 ))
4 4
1 1 1
= X(j(w 4w0 )) + X(jw) X(j(w + 4w0 ))
4 2 4
1 2 2
= (w 5w0 ) (w 3w0 ) + (w 6w0 ) (w 2w0 )
4 j j j j
1 2 2
+ (w w0 ) (w + w0 ) + (w 2w0 ) (w + 2w0 )
2 j j j j
1 2 2
(w + 3w0 ) (w + 5w0 ) + (w + 2w0 ) (w + 6w0 )
4 j j j j
3
= (w 6w0 ) (w 5w0 ) + (w 3w0 ) + (w 2w0 ) + (w w0 )
2j 4j 4j 2j 2j
3
(w + w0 ) (w + 2w0 ) (w + 3w0 ) + (w + 5w0 ) + (w + 6w0 )
2j 2j 4j 4j 2j
Filtering v(t) with a lowpass lter with cuto frequency 2w0 and passband gain of 2 we get an output, say y(t),
that has frequency content in jwj 2w0 . This operation is ambiguous in that there is an impulse exactly at the
cuto frequency. Formally, the ideal lowpass lter will only allow half of that impulse to go through.
3 3
Y (jw) = H(jw)V (jw) = + (w 2w0 ) + (w w0 ) (w + w0 ) (w + 2w0 )
2j j j 2j
Hence, converting back to time domain
3
y(t) = F 1 fY (jw)g = sin 200t + sin 400t
2
Problem 8.34
Let us rst dene the intermediate signals v(t) = x(t) + cos wc t, z(t) = v 2 (t). Then
V (jw) = X(jw) + (w wc ) + (w + wc )
1
Z(jw) = V (jw) V (jw)
2
1
1
= X(j(w wc )) + (w 2wc ) + (w)
2 2 2
1
+ X(j(w + wc )) + (w) + (w + 2wc )
2 2 2
1 1 1
+ X(j(w wc )) + X(j(w + wc )) + X(jw) X(jw)
2 2 2
1
= X(j(w wc )) + X(j(w + wc )) + (w 2wc ) + (w + 2wc ) + (w) + X(jw) X(jw)
2 2 2
The exact expression of X(jw) X(jw) is irrelevant here; we only need the fact that, since X(jw) is bandlimited
in (wM ; wM ), then X(jw) X(jw) is bandlimited in (2wM ; 2wM ).
The same result is obtained by squaring v(t) in time-domain and then taking its Fourier transform
2
EEE 303 HW # 8 SOLUTIONS
Problem 1.54 P PN
N 1
a. Let SN = n=0 an . Then, aSN = n=1 an . Therefore, (1 a)SN = a0 aN , from which we get
N
X 1
1 aN
SN = an =
n=0
1a
PN 1
provided that a 6
= 1.
P1 If a = 1, then simply SN = n=0 1n = N .
n N
b. Let S1 = n=0 a = limN !1 SN , provided that the limit exists. For this, the limit of a should
N j N N jN
exist as N ! 1 which occurs if jaj < 1. In the complex case, a = (jaje ) = jaj e . So, if jaj < 1,
limN !1 aN = 0, implying
X1
1
S1 = an =
n=0
1 a
c. Observe that for jaj < 1,
X1 X1
dS1 dan d 1 1
a =a = nan = a =
da n=0
da n=0
da 1 a (1 a)2
Notice that the exchange of summation and dierentiation makes sense since the series are absolutely convergent.
d. Again for jaj < 1,
1
X X1 k1
X 1 1 ak ak
an = an an = =
n=0 n=0
1a 1a 1a
n=k
Problem 2.21
a. Let x(n) = an U (n), h(n) = bn U (n). Then
1
X
y(n) = (h x)(n) = bnk U (n k)ak U (k)
k=1
n
X
= bnk ak
k=0
n
X a k
= bn ; (assuming b 6
= a)
b
k=0
(a=b)n+1 1
= bn
(a=b) 1
n+1
a bn+1
=
ab
Obviously, if a = b, then y(n) = (n + 1)bn .
b. Let x(n) = U (n) U (n 5), h(n) = h0 (n 2) + h0 (n 11), where h0 (n) = U (n) U (n 6). Then,
x(n) h(n) = y0 (n 2) + y0 (n 11), where y0 (n) = x(n) h0 (n). The last sequence is easy to compute:
n -1 0 1 2 3 4 5 6 7 8 9 10 11
y0 (n) 0 1 2 3 4 5 5 4 3 2 1 0 0
From this, we get
n -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 . . .
y0 (n 2) 0 0 0 1 2 3 4 5 5 4 3 2 1 0 0 0 0 0 ...
y0 (n 11) 0 0 0 0 0 0 0 0 0 0 0 0 1 2 3 4 5 5 ...
x(n) h(n) 0 0 0 1 2 3 4 5 5 4 3 2 2 2 3 4 5 5 ...
Problem 2.31
1
Consider the dierence equation
..
. (y(n) = 0; for n < 3)
y(3) = 2y(4) + x(3) + 2x(5) = 0
y(2) = 2y(3) + x(2) + 2x(4) = 1
y(1) = 2y(2) + x(1) + 2x(3) = 0
y(0) = 2y(1) + x(0) + 2x(2) = 5
y(1) = 2y(0) + x(1) + 2x(1) = 4
y(2) = 2y(1) + x(2) + 2x(0) = 16
y(3) = 2y(2) + x(3) + 2x(1) = 27
y(4) = 2y(3) + x(4) + 2x(2) = 58
y(5) = 2y(4) + x(5) + 2x(3) = 114
y(6) = 2y(5) + x(6) + 2x(4) = 2 114
y(7) = 2y(6) + x(7) + 2x(5) = 4 114
y(8) = 2y(7) + x(8) + 2x(6) = 8 114
..
. (y(n) = 114(2)n5 ; for n 5)
1
z =0
2
then the solution for the homogeneous dierence equation is
n
1
h[n] = C0 U [n]
2
h[0] = 1
so C0 = 1 then n
1
h[n] = U [n]
2