ELEC 464 (2019) Problem Set 4 Solutions to Problems 1-4

1. In the Enhanced Data Rates for GSM (EDGE) standard, 3π/8-shifted 8-PSK is used. That is, after
each transmitted symbol, the entire constellation shifts by an angle of 3π/8 radians in the signal space.
Sketch transitions in the signal space diagram for this modulation format. What is the ratio between
the average value and minimum value of the envelope? How is this an improvement over regular
(non-shifted) 8-PSK?
Solution:
The minimum value of the envelope is achieved when going from the point e j0 to the point e j7π/8 . This
same minimum value is also achieved by several other transitions. Trigonometric calculations give the
minimum distance to origin as 1 · sin(π/16) ≈ 0.195. Since the mean envelope is unity, the requested
ratio becomes 0.195.
In non-shifted 8-PSK, there are symbol transitions such as from e j0 to e j4π/8 , or e j2π/8 to e j6π/8 that
pass through the origin resulting in a transmitted envelope with a very large (∞) ratio of peak to mini-
mum power, complicating the amplifier design. That is why GSM EDGE adopted this rotating constel-
lation.

2. Both GSM and DECT use GMSK, but with different Gaussian filters (BG T = 0.3 in GSM, BG T = 0.5
in DECT). What are the advantages of having a larger bandwidth time product? Why is the lower one
used in GSM?
Solution:
The advantage of a smoother filter is that the spectral efficiency becomes higher. The drawback is that
the signal does not always reach the ideal MSK signal points, reducing Euclidean distance. Therefore
the smoother filter implies less power efficiency. For the same transmitted power, the smoother filter
(BG T = 0.3) has higher sensitivity to noise and other disturbances.

3. Calculate the nearest neighbour union bound for Gray-coded 16-QAM. Assuming that the BER must
not exceed 10−5 , what are the useful ranges of Eb /N0 for adaptive switching between the two modula-
tion schemes QPSK and 16-QAM for maximum achievable data rate?
Solution:
We start by calculating nearest neighbour union bound on BER for the 16-QAM modulation scheme,
expressed as a function of minimum distance dmin . Since the constellation is Gray coded only one
bit error occurs when a nearest neighbour constellation point is selected. Due to the symmetry of the
16-QAM constellation, there are three groups of constellation points, where the points in each group,
denoted by Ai , have the same distances to the origin and the same numbers of nearest neighbour con-
stellation points, ni j . We have
s !
1 M M di2j
pb ≈ ∑ ∑ Q 2N0 ni j
M log2 (M) i=1 j∈Ai
s !
2
dmin
1
= (4 · 2 + 8 · 3 + 4 · 4)Q
64 2N0
s !
2
dmin
3
= Q .
4 2N0

Next, replace dmin with the average bit energy. The average symbol energy (assuming equally probable
 symbols) is

1 M 2
Es = ∑ di
M i=1
!
1 1 9 5
= 4· +4· +8·
16 2 2 2
5 2
= dmin
2
which means that the average bit energy is E¯b = Ēs /4 = 58 dmin
2 . Thus,

s !
3 4E¯b
pb ≈ Q .
4 5N0

For the BER not to exceed 10−5 , the Eb /N0 must be above
!2
E¯b 5 4
= Q−1 ( · 10−5 )
N0 4 3
= 13.4 dB

in the case of 16-QAM, and above

in the case of QPSK as shown in class as well as in Molisch Eq. (12.34). Consequently, the useful
range of Eb /N0 for the QPSK mode of operation is between 9.6 and 13.4 dB and above 13.4 dB for
16-QAM. Many current systems in use today adapt their modulation according to channel conditions
based on estimating Eb /N0 by transmitting known training or pilot symbols and measuring received
power.

4. Consider a point-to-point radio link between two highly directional antennas in a stationary environ-
ment. Each has a gain of 30 dB, distance attenuation is 150 dB, and the RX has a noise figure of 7 dB.
The symbol rate is 20 Msymb/s and Nyquist signalling is used. It can be assumed that the radio link can
be treated as an AWGN channel without fading. How much transmit power is required (disregarding
power losses at TX and RX ends) for a maximum BER of 10−5 for the following scenarios:
(i) coherently detected BPSK,
(ii) coherently detected FSK,
(iii) differentially detected BPSK
(iv) noncoherently detected FSK?
Solution:
The parameters given in the problem statement are:
Antenna Gain Ga = 30 dB
Distance Attenuation LPL = 150 dB
Receiver noise figure Fsys = 7 dB
Symbol rate ds = 2 · 107 symb/s
Maximum BER pb,max = 10−5
Nyquist signalling
Using the receiver noise figure converted to linear scale, the power spectral density of the noise, N0 , is
calculated as

N0 = kB (Fsys − 1)T0
= 1.38 · 10−23 · (107/10 − 1) · 290
= 1.606 · 10−20 W /Hz.
The above error probabilities for scenarios (i)-(iv) can be found in Molisch (12.34), (12.29), (12.45)
and (12.37), respectively. Coherent BFSK is just a form of orthogonal signalling. Note that there is a
typo in Eq. (12.37) of Molisch. The BER for differential detection of binary orthogonal signals, which
would the same as that of noncoherent FSK should be:
1
BER = exp(−γb /2).
2
Unfortunately, there are a few other typos in Molisch in this section. Please check the Errata to the text
posted on the onQ site.