Hw09solns Rev
Hw09solns Rev
11, 2006
Finally, the dimensionality of the waveforms (or dimension of the signal set) is the number of functions in
the orthonormal basis for the given signal set with fewest functions. This is not so easy to deduce directly.
(The dimensionality would be obvious if we had applied Gram-Schmidt.) An equivalent definition is that
the dimension is the number of linearly independent signals in the signal set. Since signals are linearly
independent iff their corresponding vectors are linearly independent, the dimension is the number of
linearly independent vectors in the set {s1, s2, s3, s4}. One can see that all four vectors are all linearly
independent.Thus, the dimension of the signal set is 4.
Another way to check if an orthonormal basis has fewest functions is to check if each member of the basis
is a linear combination of the original signals. For the orthonormal basis above we see that
s1(t) + s 2(t) = -ψ4(t) so ψ4(t) is a linear combination of s1(t), s2(t), s3(t), s4(t)
1
s2(t) + s4(t) = -2ψ3(t) +2ψ4(t) so ψ3(t) = 2 (2ψ4(t) - s2(t) - s4(t) )
so ψ3(t) is a linear combination of s1(t),s2(t), s3(t), s4(t), because ψ4(t) is
s3(t) - s4(t) = ψ2(t) +3ψ3(t) - ψ4(t) so ψ2(t) = s3(t) - s4(t) - 3ψ3(t) + ψ4(t)
so ψ2(t) is a linear combination of s1(t),s2(t), s3(t), s4(t), because ψ3((t) and ψ4(t) are
s1(t) - s2(t) = 4ψ1(t) - 2ψ2(t) - 2ψ3(t) - ψ4(t)
so ψ1(t) is a linear combin. of s1(t),s2(t), s3(t), s4(t), because ψ1(t), ψ3((t) and ψ4(t) are
Since we have shown that each of the ψ(t)'s is a linear combination of the si(t)'s, they are a minimal
basis, and we have again shown that the dimension of the signal set is 4.
*** Part 3. The minimum distance between signals may be computed from the signals themselves or from
their vector representations. In either case, the minimum distance is
dmin = d(s1,s3) = √
5
2. (a) Problem 8.4, p. 457, Use the Gram-Schmidt procedure to find the orthonormal functions.
*** It is easiest to rearrange the signals and apply Gram-Schmidt to the signals in the order s2(t), s4(t),
s1(t), s4(t). Here is the G-S procedure applied to s2(t), s4(t), s1(t), s4(t)
s2(t) 1, 0<t<1
ψ 1(t) = =
||s2(t)|| 0, else
2, 1<t<2
φ2(t) = s4(t) - 〈s4(t),ψ 1(t)〉 ψ 1(t) = s4(t) - 2 ψ 1(t) =
0, else
1
φ 2 (t) 1, 1<t<2
ψ 2(t) = =
||φ 2 (t)|| 0, else
2, 2<t<3
φ3(t) = s1(t) - 〈s1(t),ψ1(t)〉 ψ1(t) - 〈s1(t),ψ2(t)〉 ψ2(t) = s1(t) - 2ψ1(t) - 2ψ2(t) =
0, else
φ 3 (t) 1, 2<t<3
ψ 3(t) = =
||φ 3 (t)|| 0, else
s1(t) = 2√
3 ψ 1(t), s2(t) = 2/√
3 ψ1(t) + 2√
6 ψ2(t),
s3(t) = -4/√
3 ψ1(t) +2√
2/3 ψ 2(t), s4(t) = 4/√
3 ψ1(t) + √
2/3 ψ2(t) + 1/√
2 ψ3(t).
Therefore,
s1 = (2√
3, 0, 0) = (3.46, 0,0), s2 = (2/√
3, 2√
6, 0) = (1.15, 4.90, 0)
s3 = ( -4/√
3, 2√
2/3, 0) = (-2.31, 1.63, 0), s4 = (4/√
3, √
2/3, 1/√
2) = (2.31, .82, .71)
*** (c) Make a plot of the signal vectors (as best you can).
With respect to the basis found by appyling G-S With respect to the basis found by appyling G-S
to the signals in order s2(t), s4(t), s1(t), s4(t), the to the signals in their original order, the signal
signal vectors are. vectors are
r2
r2
s1 5 s2
s4
2 r3
r3 s3
5
s4
-5
r1
-2 s2 s1 5
r1
2 -5
-2
-2
s3
2
3. Problem 8.9, p.458. When it says "sketch the output" it means "sketch the output when the input is the
signal to which the filter is matched.
T ,
t T-t, 0≤t≤T
1
0≤t≤T
(a) s1(t) = , h 1(t) = s1(T-t) = T =
0, else 0, else t
T
T/2
2, T/2≤t≤T
(c) s3(t) = , h 3(t) = s3(T-t) = 2
0, else
2, 0≤t≤T/2
=
0, else t
2T
3
4. (a) Problem 8.10, p. 459
P E = Q( √
2 E
No
) = Q( √
2 A 2T
No
) = Q( √
2×10-3A2)
^ 1, <s1(t)-s2(t),R(t)> < 0
I =
2, else
√
√
d2 A2T
P E = Q
2No = Q
6No
∞ T T T
A t - A (T-t) 2 dt = A A2
2 1
since d2 = ∫ (s1(t) - s2(t))2 dt = ∫ T T T2 ∫ ( 2t - T)2 dt = T2 ∫ u2 2du, u=2t-T
-∞ 0 0 -T
A2 T3 A 2T
= T2 3 = 3
√
A 2T Eb
E b = E1 = 3 . Hence, PE = Q 2N
o
4
√ √
Eb -5 Eb
Setting PE = Q = 10 , we find from the Q-function table that
2No 2No ≅ 4.25.
E
Therefore, Nb ≅ 36.1 or 15.6 dB
o
√
2
1 M di,j
P E ≤ M ∑ ∑ Q 2No ,
where M=3 is the number of signals, di,j = ||si-sj||
i=1 j≠i
We find d1,2 = d2,1 = d1,3 = d3,1 = √
2T and d2,3 = d3,2 = 2√
T. Therefore,
1
√
√
√ √2T
T 2T 4 T 2
P E ≤ 3 4Q
No + 2Q
N o = 3 Q
N o + 3Q
No
(b) In class we also proposed approximating this upper bound by the dominate terms corresponding to the
signal separated by dmin = d1,2 = √
2T. In this case we have
√
2
~ K
√
dmin 4 T
PE ≤ M Q = Q
2No 3 N o
Notice that this is the first and dominate term in answer (a).
(c) Finally, the book upper bounds K by K ≤ M(M-1), in which case
√
√
2 2
~ K
√
dmin dmin T
PE ≤ M Q
2No ≤ (M-1) Q 2No = 2 Q N o
5
7. (a) Problem 8.28, p. 464.
From the description in Chapter 8, biorthogonal signals have equal energy (let E denote the energy of a
signal), and for each signal there is one signal that is the antipodal of it, and all other signals are
orthogonal to it. Thus if there are 8 signals, then for signal si(t), the antipodal signal sj(t) has
di,j = ||si(t)-sj(t)|| = ||2si(t)|| = 2||si(t)|| = 2√
E.
On the other hand if sj(t) is one of the 6 signals orthogonal to si(t), then
di,j = ||si(t)-sj(t)|| = √
2E.
M
1 dm,m'
The union bound is P(E) ≤ M ∑ ∑ Q(√
2No
)
m=1 m'≠i
√
) + 6 Q( √
Nob)
6Eb 3E
P E ≤ Q( No
~
√
√
E 3Eb
P E ≤ 6 Q( No ) = 6 Q( No ).
Or one can use the upper bound stated in the book
√
2
dmin
~
√
√
E 3Eb
2No = 7 Q(
PE ≤ (M-1) Q No ) = 7 Q( No ).
(b) Find the energy efficiency for M = 8.
Equating 6 Q( √
3Eb
No ) = 10-5, we find from the Q-function table that √
3Eb
2No ≅ 4.25.
E
Therefore, Nb ≅ 12.0 or 10.8 dB
o
(c) Compare M-ary biorthogonal and M-ary orthogonal signal sets on the basis of energy efficient. Which
is more efficient? Is there a simple way to quantitatively compare them?
For M-ary biorthogonal signals, there are M-2 signals at distance dmin = √
2E = √
2 (log2 M) Eb from each
signal, so the approximate union bound gives
~
P E ≤ (M-2) Q( √
E
No ) = (M-2) Q( √
( ) log2 M Eb
No ).
For M-ary orthogonal signals, there are M-1 signals at distance dmin = √
2E = √
2 (log2 M) Eb from each
signal, so the approximate union bound gives
~
P E ≤ (M-1) Q( √
E
No ) = (M-1) Q( √
( ) log2 M Eb
No ).
M-2
We see that biorthogonal is better in that the upper bound for biorthogonal is fraction M-1 less. This is
rather significant if M=2, somewhat significant if M=4, and not significant if M≥8.
6
8. A binary modulator uses the two signals
1, 0≤t≤T/2 1, T/2≤t≤T
s1(t) = s2(t) =
0, else 0, else
Design an optimum receiver that has as its first stage one and only one matched filter, and no other analog
filters and no analog correlators.
An orthonormal basis for the given signals is:
√(2/T) √( 2/T)
ψ 1(t) = and ψ2(t) =
T/2 T T/2 T
With the respect to this basis, the signal vectors are s1 = (A√
T/2,0) and s2 = (0,A T/2).
√
A block diagram of one potential optimal receiver is:
sample at T
R1
matched
R(t) filter h1 (t) ^I
sample at T decision
R2 device
matched
filter h2 (t)
where h1(t) = ψ1(T-t) = ψ2(t), h2(t) = ψ2(T-t) = ψ1(t) and the decision device chooses ^I = 1 if
||(R1,R2) - s1|| ≤ ||(R1,R2) - s2||, which simplifies to ^I = 1 if R1 > R2, and ^I = 2 if R2 > R1.
Here are two acceptable answers:
(a) The following receiver, with only one matched filter, but with a sampler that samples twice produces
exactly the same decisions as the one above (because R1 and R2 are always the same as in the above).
Therefore, it too is an optimum receiver.
R(t) sample at T/2 and T
matched R1 R2 I
filter h2(t) decision
device
The matched filter is h2(t) = ψ1(t) and the decision device is the same as in part 2).
(b) Alternatively, we see that the original receiver with two matched filters decides ^I = 1 if R1 > R2,
and ^I = 2 if R2 > R1. Equivalently, it decides ^I = 1 if R1 - R2 > 0, and ^I = 2 if R1 - R2 < 0.
Moreover, a sample at time T of a single filter with impulse response h(t) = h1(t) - h2(t) will be R1-R2.
Therefore, one can use the system shown below:
R(t)
sample
sample at T/2atand
T T
R = R1 -R2 ^
matched
filter I
filterh(th2(t) decision
device