RADIO SYSTEMS – ETIN15
Lecture no:
6
Demodulation,
bit-error probability
and
diversity arrangements
Anders J Johansson, Department of Electrical and Information Technology
[email protected]
5 April 2017 1
Contents
• Receiver noise calculations [Covered briefly in Chapter 3 of
textbook!]
• Optimal receiver and bit error probability
– Principle of maximum-likelihood receiver
– Error probabilities in non-fading channels
– Error probabilities in fading channels
• Diversity arrangements
– The diversity principle
– Types of diversity
– Spatial (antenna) diversity performance
5 April 2017 2
RECEIVER NOISE
5 April 2017 3
Receiver noise
Noise sources
The noise situation in a receiver depends on
several noise sources
Noise picked up
Wanted by the antenna
signal
Output signal
Analog
Detector with requirement
circuits
on quality
Thermal
noise
5 April 2017 4
Receiver noise
Equivalent noise source
To simplify the situation, we replace all noise sources
with a single equivalent noise source.
Wanted How do we determine
Noise free N from the other
signal
sources?
N
C Output signal
Analog
Detector with requirement
circuits
on quality
Noise free
Same “input quality”, signal-to-noise
ratio, C/N in the whole chain.
5 April 2017 5
Receiver noise
Examples
• Thermal noise is caused by random movements of electrons in
circuits. It is assumed to be Gaussian and the power is
proportional to the temperature of the material, in Kelvin.
• Atmospheric noise is caused by electrical activity in the
atmosphere, e.g. lightning. This noise is impulsive in its nature
and below 20 MHz it is a dominating.
• Cosmic noise is caused by radiation from space and the sun is a
major contributor.
• Artificial (man made) noise can be very strong and, e.g., light
switches and ignition systems can produce significant noise well
above 100 MHz.
5 April 2017 6
Receiver noise
Noise sources
The power spectral density of a noise source is usually given in one
of the following three ways:
This one is
1) Directly [W/Hz]: Ns often given in
dB and called
2) Noise temperature [Kelvin]: Ts noise figure.
3) Noise factor [1]: Fs
The relation between the three is
N s kTs kFsT0
where k is Boltzmann’s constant (1.38x10-23 W/Hz) and T0 is the,
so called, room temperature of 290 K (17O C).
5 April 2017 7
Receiver noise
Noise sources, cont.
Antenna example
Na
Model
Noise temperature Noise free
of antenna 1600 K antenna
Power spectral density of antenna noise is
N a =1.38×10−23 ×1600=2.21×10−20 W/Hz=−196.6 dB [W/Hz]
5 April 2017 8
Receiver noise
System noise
The noise factor or noise figure of a system component (with input
and output) is:
Nsys
System System
Model
component component
Noise factor F Noise free
Due to a definition of noise factor as the ratio of noise
powers on the output versus on the input, when a resistor in room
temperature (T0=290 K) generates the input noise, the PSD of the
equivalent noise source (placed at the input) becomes
N sys k F 1 T0 W/Hz
Don’t use dB value! Equivalent noise temperature
5 April 2017 9
Receiver noise
Several noise sources
A simple example
Ta
System 1 System 2
F1 F2
N a kTa
Noise
N1 k F1 1 T0
free
Na N1 N2 N 2 k F2 1 T0
System 1 System 2
Noise Noise
free free
5 April 2017 10
Receiver noise
Several noise sources, cont.
After extraction of the noise sources from each component, we need to
move them to one point.
When doing this, we must compensate for amplification and attenuation!
Amplifier:
N NG
G G
Attenuator:
N N/L
1/L 1/L
5 April 2017 11
Receiver noise
Pierce’s rule
A passive attenuator in room temperature, in this case a transmission line,
has a noise factor equal to its attenuation.
Nf
Lf Lf
Ff = Lf Noise free
N f k Ff 1 T0 k L f 1 T0
Remember to
convert from dB!
5 April 2017 12
Receiver noise
Remember ...
Antenna noise is usually given as a noise temperature!
Noise factors or noise figures of different system components
are determined by their implementation.
When adding noise from several sources, remember to
convert from the dB-scale noise figures that are usually given,
before starting your calculations.
A passive attenuator in (room temperature), like a transmission line, has
a noise figure/factor equal to its attenuation.
5 April 2017 13
Receiver noise
A final example
Let’s consider two (incomplete) receiver chains with equal
gain from point A to B:
Ta
1 Lf
G1 G2
A B Would there be any
F1 F2 reason to choose one
over the other?
Ta
Let’s calculate the
2 Lf equivalent noise
G1 G2 at point A for both!
A B
F1 F2
5 April 2017 14
Receiver noise
A final example
Ta 1 2
Ta
Lf Lf
G1 G2 G1 G2
A B A B
F1 F2 F1 F2
Equivalent noise sources at point A for the two cases would have the
power spectral densities:
1 N 0 kTa k F 1 L 1 / G F 1 L
1 f 1 2 f / G1 T0
2 N 0 kTa k L 1 F 1 L F2 1 L f / G1 T0
f 1 f
Two of the noise contributions are equal and two are larger in (2),
which makes (1) a better arrangement.
This is why we want a low-noise amplifier (LNA) close to the antenna.
5 April 2017 15
Receiver noise
Noise power
We have discussed noise in terms of power spectral density N0 [W/Hz].
For a certain receiver bandwidth B [Hz], we can calculate the equivalent
noise power:
N = B× N 0 [W]
N ∣dB= B∣dB N 0∣dB [dBW]
This is the version
we will use in our
link budget.
5 April 2017 16
Receiver noise
The link budget
”POWER” [dB]
P TX
Lf, TX G
a, TX
Lp
G a, RX Lf, RX
C
F [dB] is the noise figure of the equivalent C/N
noise source at the reference point and N
B [dBHz] the system bandwidth . The receiver
B
noise calculations
N0
show up here.
Noise reference level = kT0 = -204 dB[W/Hz] F
Transmitter Receiver
In this version the reference point is here
5 April 2017 17
OPTIMAL RECEIVER
AND
BIT ERROR PROBABILITY
5 April 2017 18
Optimal receiver
What do we mean by optimal?
Every receiver is optimal according to some criterion!
We would like to use optimal in the sense that we achieve a
minimal probability of error.
In all calculations, we will assume that the noise is white and
Gaussian – unless otherwise stated.
5 April 2017 19
Optimal receiver
Transmitted and received signal
Transmitted signals Channel Received (noisy) signals
s1(t) r(t)
1:
t n(t) t
s (t ) r(t)
s0(t) r(t )
0:
t t
5 April 2017 20
Optimal receiver
A first “intuitive” approach
“Look” at the received signal and compare it to the possible received
noise free signals. Select the one with the best “fit”.
Assume that the following Comparing it to the two possible
signal is received: noise free received signals:
r(t), s1(t)
r(t ) 1: This seems to be
t the best “fit”. We
assume that “0”
t was the
r(t), s2(t) transmitted bit.
0:
t
5 April 2017 21
Optimal receiver
Let’s make it more measurable
To be able to better measure the “fit” we look at the energy of the
residual (difference) between received and the possible noise free signals:
s1(t) - r(t)
r(t), s1(t)
1: 2
e 1 =∫∣s1 t −r t ∣ dt
t t
s0(t) - r(t)
r(t), s0(t)
2
0: e 0=∫∣s0 t −r t ∣ dt
t t
This residual energy is much
smaller. We assume that “0” was
transmitted.
5 April 2017 22
Optimal receiver
The AWGN channel
The additive white Gaussian noise (AWGN) channel
n t
s t r t s t n t
s t - transmitted signal In our digital transmission
system, the transmitted
- channel attenuation signal s(t) would be one of,
n t - white Gaussian noise let’s say M, different alternatives
s0(t), s1(t), ... , sM-1(t).
r t - received signal
5 April 2017 23
Optimal receiver
The AWGN channel, cont.
It can be shown that finding the minimal residual energy (as we
did before) is the optimal way of deciding which of s0(t), s1(t), ... , sM-1(t)
was transmitted over the AWGN channel (if they are equally probable).
For a received r(t), the residual energy ei for each possible transmitted
alternative si(t) is calculated as
2 *
ei =∫∣r t− s i t ∣ dt=∫ r t − s i t r t − si t dt
2 * * 2 2
=∫∣r t ∣ dt−2 Re { ∫ r t s t dt }∣∣ ∫∣s i t ∣ dt
i
Same for all i Same for all i,
if the transmitted
The residual energy is minimized by signals are of
maximizing this part of the expression. equal energy.
5 April 2017 24
Optimal receiver
The AWGN channel, cont.
The central part of the comparison of different signal alternatives
is a correlation, that can be implemented as a correlator:
r t
∫T s
*
si t * The real part of
the output from
or a matched filter either of these
is sampled at t = Ts
r t *
si Ts t
*
where Ts is the symbol time (duration).
5 April 2017 25
Optimal receiver
Antipodal signals
In antipodal signaling, the alternatives (for “0” and “1”) are
s0 t t
s1 t t
This means that we only need ONE correlation in the receiver
for simplicity:
If the real part
r t at T=Ts is
∫T s
>0 decide “0”
<0 decide “1”
* t *
5 April 2017 26
Optimal receiver
Orthogonal signals
In binary orthogonal signaling, with equal energy alternatives s0(t) and s1(t)
(for “0” and “1”) we require the property:
*
〈 so t , s 1 t 〉=∫ s 0 t s1 t dt =0
The approach here is to use two correlators:
∫T s Compare real
*
* part at t=Ts
s0 t
r t and decide in
favor of the
∫T s
larger.
s1
*
t *
(Only one correlator is needed, if we correlate with (s0(t) - s1(t))*.)
5 April 2017 27
Optimal receiver
Interpretation in signal space
The correlations performed on the previous slides can be seen as
inner products between the received signal and a set of basis functions
for a signal space.
The resulting values are coordinates of the received signal in the
signal space.
Antipodal signals Orthogonal signals
s1 t
Decision
boundaries “1”
“1” “0” “0”
t s0 t
5 April 2017 28
Optimal receiver
The noise contribution
Assume a 2-dimensional signal space, here viewed as the complex plane
Im
Noise-free
positions
si
Noise pdf.
Es
sj This normalization of
axes implies that the
Re noise centered around
each alternative is
complex Gaussian
Es
N 0, 2 jN 0, 2
Fundamental question: What is the probability with variance σ2 = N0/2
that we end up on the wrong side of the decision in each direction.
boundary?
5 April 2017 29
Optimal receiver
Pair-wise symbol error probability
What is the probability of deciding si if sj was transmitted?
Im
We need the distance
between the two symbols.
si d ji In this orthogonal case:
Es 2 2
d ji Es Es 2 Es
sj
Re The probability of the noise
pushing us across the boundary
at distance dji / 2 is
Es
Pr s j si =Q
d ji / 2
N 0 /2
=Q
Es
N0
The book uses erfc()
instead of Q().
1
= erfc
2 Es
2 N0
5 April 2017 30
Optimal receiver
The union bound
Calculation of symbol error probability is simple for two signals!
When we have many signal alternatives, it may be impossible to
calculate an exact symbol error rate.
When s0 is the transmitted
s2
signal, an error occurs when
s1 the received signal is outside
this polygon.
s0
s3
s6 The UNION BOUND is the sum
of all pair-wise error probabilities,
and constitutes an upper bound
on the symbol error probability.
s5
s4
s7
The higher the SNR, the better
the approximation!
5 April 2017 31
Optimal receiver
Symbol- and bit-error rates
The calculations so far have discussed the probabilities of selecting
the incorrect signal alternative (symbol), i.e. the symbol-error rate.
When each symbol carries K bits, we need 2K symbols.
Gray coding is used to assigning bits so that the nearest neighbors only
differ in one of the K bits. This minimizes the bit-error rate.
011
010 001
000
Gray-coded 8PSK
110
111 100
101
5 April 2017 32
Optimal receiver
Bit-error rates (BER)
EXAMPLES: 2PAM 4QAM 8PSK 16QAM
Bits/symbol 1 2 3 4
Symbol energy Eb 2Eb 3Eb 4Eb
BER Q
2 Eb
N0
Q
2 Eb
N0
2
~ Q
3 0.87
Eb
N0 3
~ Q
2 E b , max
2.25 N 0
Gray coding is used when calculating these BER.
5 April 2017 33
Optimal receiver
Bit-error rates (BER), cont.
0
10
-1
2PAM/4QAM
10
8PSK
16QAM
Bit-error rate (BER)
-2
10
-3
10
-4
10
-5
10
-6
10
0 2 4 6 8 10 12 14 16 18 20
Eb / N 0 [dB]
5 April 2017 34
Optimal receiver
Where do we get Eb and N0?
Where do those magic numbers Eb and N0 come from?
The noise power spectral density N0 is calculated according to
N 0=k T 0 F 0 ⇔ N 0∣dB=−204F 0∣dB
where F0 is the noise factor of the “equivalent” receiver noise source.
The bit energy Eb can be calculated from the received
power C (at the same reference point as N0). Given a certain
data-rate db [bits per second], we have the relation
E b=C / d b ⇔ E b∣dB=C∣dB−d b∣dB
THESE ARE THE EQUATIONS THAT RELATE DETECTOR
PERFORMANCE ANALYSIS TO LINK BUDGET CALCULATIONS!
5 April 2017 35
Optimal receiver
What about fading channels?
We have (or can calculate) BER expressions for non-fading AWGN
channels.
If the channel is Rayleigh-fading, then Eb/N0 will have an
exponential distribution (N0 is assumed to be constant)
b -- Eb/N0
1 − /
pdf b = e b b
b
b -- average Eb/N0
The BER for the Rayleigh fading channel is obtained by averaging:
∞
BERRayleigh b =∫ BERAWGN b × pdf b d b
0
5 April 2017 36
Optimal receiver
What about fading channels?
THIS IS A SERIOUS PROBLEM!
Bit error rate (4QAM)
0
10
-1 10 dB Rayleigh fading
10
10 x
-2
10
-3
10
-4
10
-5
10 No fading
-6
10
0 2 4 6 8 10 12 14 16 18 20
Eb/N0 [dB]
5 April 2017 37
DIVERSITY ARRANGEMENTS
5 April 2017 38
Diversity arrangements
Let’s have a look at fading again
Illustration of interference pattern from above
Received power [log scale]
Movement
A B
Position
Transmitter
A B
Reflector
Having TWO separated antennas in this case may increase
the probability of receiving a strong signal on at least one of them.
5 April 2017 39
Diversity arrangements
The diversity principle
The principle of diversity is to transmit the same information on
M statistically independent channels.
By doing this, we increase the chance that the information will
be received properly.
The example given on the previous slide is one such arrangement:
antenna diversity.
5 April 2017 40
Diversity arrangements
General improvement trend
Bit error rate (4PSK)
0
10
-1 10 dB Rayleigh fading
10
No diversity
10 x
-2
10
10 dB
-3
10 Rayleigh fading
M:th order diversity
-4
10
10M x
-5
10 No fading
-6
10
0 2 4 6 8 10 12 14 16 18 20
Eb/N0 [dB]
5 April 2017 41
Diversity arrangements
Some techniques
Spatial (antenna) diversity
We will focus on this
...
one today!
TX Signal combiner
Frequency diversity
D D D
TX Signal combiner
Temporal diversity
Inter- De-inter-
Coding De-coding
leaving leaving
(We also have angular and polarization diversity)
5 April 2017 42
Spatial (antenna) diversity
Fading correlation on antennas
With several Isotropic
antennas, we uncorrelated
want the fading on scattering.
them to be as
independent
as possible.
E.g.: An antenna
spacing of about 0.4
wavelength gives
zero correlation.
5 April 2017 43
Spatial (antenna) diversity
Selection diversity
RSSI = received
signal strength
indicator
5 April 2017 44
Spatial (antenna) diversity
Selection diversity, cont.
By measuring BER instead of RSSI, we have a better guarantee
that we obtain a low BER.
5 April 2017 45
Spatial (antenna) diversity
Maximum ratio combining
This is the optimal way (SNR sense) of combining antennas.
5 April 2017 46
Spatial (antenna) diversity
Simpler than MRC, but almost the same performance.
5 April 2017 47
Spatial (antenna) diversity
Performance comparison
Cumulative distribution of SNR
MRC
Comparison of
SNR distribution RSSI selection
for different number
of antennas M and
two different diversity
techniques.
[Fig. 13.10]
These curves can be used to calculate fading margins.
5 April 2017 48
Spatial (antenna) diversity
Performance comparison, cont.
MRC
Comparison of RSSI selection
2ASK/2PSK BER
for different number
of antennas M and
two different diversity
techniques.
[Fig. 13.11]
5 April 2017 49
Summary
• Optimal (maximum likelihood) receiver in AWGN channels
• Interpretation of received signal as a point in a signal space
• Euclidean distances between symbols determine the probability of
symbol error
• Bit error rate (BER) calculations for some signal constellations
• Union bound (better at high SNRs) can be to derive approximate BER
expressions
• Fading leads to serious BER problems
• Diversity is used to combat fading
• Focus on spatial (antenna) diversity
• Performance comparisons for RSSI selection and maximum ratio
combining diversity.
5 April 2017 50
