ICAI – Máster MIT

Sistemas de Comunicación I

Chapter 2 (1):
Channel characterization:
noise

Javier Matanza
W. Warzanskyj
Luis Cucala
1
Noise
• Noise is a natural limiter in communications, either wired or wireless
• Noise is captured by the receiver from the transmitter, the environment
or generated by the receiver
Interference

Signal Rx signal
Tx signal
Rx 𝑠! (𝑡)
⊕ 𝑟! 𝑡 = 𝑠! 𝑡 + 𝑛(𝑡)
Environment Noise from
electronics noise,
thermal noise
𝑛(𝑡)

• In communications models, the noise is treated as a random component

added to the received signal
• There are several noise types: white, impulsive, 1/f (pink), shot, . . .

2
1

AWGN
 AWGN (Additive White Gaussian Noise )
• White noise is the most common kind of noise in communications.
• AWGN noise is not band-limited; however, systems make it band-limited.

Power spectral density autocorrelation

𝑁,
𝐺()*+ 𝑓 = 𝑁,/2 𝑅()*+ 𝜏 = 𝛿(𝜏)
! !" 𝑁!
2
𝑁! 2
2
f Not band-limited τ

𝑓 𝑅()*+ 𝜏 = 𝑁,𝐵 · 𝑠𝑖𝑛𝑐(2𝐵𝜏)

𝐺()*+ 𝑓 = 𝑁,/2 2
2𝐵
𝑁!
! !" 𝑁!
2
2

f Band-limited τ
B
1/2B

4
 AWGN statistics
AWGN is a Gaussian random process with zero mean
• Gaussian probability density function:
!"
8 <
𝑓7 𝑛 = ·𝑒 ".$! "
9:;! "
mean 𝐸 𝑛 =0
variance 𝐸 𝑛% = σ%& (this is the Gaussian process power)

• Frequency-independent power spectral density (double sided):

"-
𝐺! 𝑓 =
#

• Autocorrelation; samples are uncorrelated:

"-
𝑅! 𝜏 = 𝐸 𝑛 𝑡 · 𝑛 𝑡 − 𝜏 = 𝐹 $% 𝐺! 𝑓 = 𝛿(𝜏)
#

• If bandlimited to bandwidth 𝐵 (single sided) its total power is

𝑁&
𝜎!# = . 2𝐵 = 𝑁& 𝐵
2 5
AWGN graphics
Intuitively, the higher the variance, the higher the chances of getting large
values from the PDF, i.e., larger power
0.4 0.4

0.35
𝜎% = 1 0.35 𝜎% = 2
0.3 0.3

0.25 0.25

PDF
PDF

0.2 0.2
0.15
0.15
0.1
0.1
0.05
0.05
0
-4 -2 0 2 4 -4 -2 0 2 4

AGWN with s2 = 1 AGWN with s2 = 2

4 4

3 3

2 2

1 1

0 0

-1 -1

-2 -2

-3 -3

-4 -4
0 200 400 600 800 1000 0 200 400 600 800 1000

6
 Thermal noise (AWGN). Expressions based in 𝑇!"
• Thermal noise is an AWGN whose single sided power spectral density is 𝑁' = 𝐾𝑇
*
𝐾 : Boltzmann constant, 1.380649 10'%( 𝑚% 𝐾𝑔 𝑠 '% 𝐾 ') (+)

• 𝑇 is the noise temperature. If coming from passive elements, its value is that of
ambient temperature, 𝑇& ≈ 300 (º𝐾𝑒𝑙𝑣𝑖𝑛)
• An electronic device generates an internal noise 𝑁(!) . Its value is associated with
an effective noise temperature 𝑇*+ → 𝑁(!) = 𝐺𝐾𝑇*+ 𝐵

- equivalent ideal
𝑁,& = 𝐾 𝑇,& + 𝑇./ 𝐵 𝑁012 = 𝐺𝐾 𝑇,& + 𝑇./ 𝐵
noiseless device
(G: gain)

𝑇./
Noise factor definition 𝐹 =1+
𝑇3

3 then 𝑇./ = (𝐹 − 1)𝑇3

" 4 12
Only If 𝑇,& = 𝑇3 𝐹 = "./0 = 3
12
4 ./0 7
Thermal noise (AWGN). Expressions based in F

𝑁,& = 𝐾𝑇,& 𝐵 Non-ideal noisy 𝑁012 = 𝐺𝑁,& + 𝑁,&2

device
G: gain
𝑁,&2 : internal noise

𝑁,&2 .51,678.&2 @ ,&:12 = 𝐾𝑇./ 𝐵 = 𝐾 𝐹 − 1 𝑇3 𝐵

𝑁,&2 = 𝐺. 𝑁,&2 .51,678.&2 @ ,&:12 = 𝐺𝐾𝑇./ 𝐵 = 𝐺𝐾 𝐹 − 1 𝑇3 𝐵

𝑁012 = 𝐺 𝑁,& + 𝑁,&2 .51,678.&2 @ ,&:12 = 𝐺𝑁,& + 𝑁,&2

8
 Thermal noise (AWGN) details if 𝑇#$ = 𝑇%
input noise if the device is
internal noise generated by the amplifier
noisy and 𝑇,& = 𝑇3

, 𝑁(!)
𝑁(! = 𝐾𝑇& 𝐵 𝑁(! = 𝑁(! + = 𝐾𝑇&, 𝐵 = 𝐾 𝑇& + 𝑇*+ 𝐵 → 𝑁(!) = 𝐾𝑇*+ 𝐵𝐺
𝐺
Equivalent input noise if the device were noiseless
by noise factor definition
𝑆
𝑁 ,& 𝐺𝑁,& + 𝑁,&2 𝑁,&2 𝐾𝑇./ 𝐵𝐺 𝑇./
= =1+ =1+ =1+ =𝐹
𝑆 𝐺𝑁,& 𝐺𝑁,& 𝐾𝑇3 𝐵𝐺 𝑇3
𝑁 012
!
" #$
then, 𝐹 = ! only valid for thermal noise input power, 𝑁,& = 𝐾𝑇3 𝐵
" %&'

- equivalent ideal
𝑁,& = 𝐾 𝑇3 + 𝑇./ 𝐵 𝑁012 = 𝐺𝐾 𝑇3 + 𝑇./ 𝐵
noiseless device
(G: gain)

9
2

Impulsive noise
 Impulsive Noise
• Less common than AWGN noise,
present in AC lines used for
telecom (PLC)
• Consists of high-power short
pulses with an “on/off” behavior
• Depends on the transmission
medium
• Might be caused by switching
processes, surface degradation,
high power sources, etc.
• We’ll focus on:
• Poisson-Gauss Model: simple and somewhat unrealistic, to be used
when there is no other option
• Middleton Model: realistic, well suited for Power Line Communications
(PLC)

11
Poisson-Gauss Model (gated noise)
• Noise is seen as a combined random process 𝑖- = 𝑏- ⋅ 𝑔-
[𝒊] Vector of impulsive noise samples
[𝒃] Impulse rate. Vector of Poisson process (b 𝑘 = 0 𝑜𝑟 1) with l impulses
per second (pulse width is ignored, pulses treated as deltas)
[𝒈] Amplitude. Vector of white Gaussian process with 𝜎 % variance.
• Occurrence is modeled as a Poisson Process
';2 ;< (
𝑃 𝑘, 𝑇 = 𝑒 k occurrences in the time interval T
=!
0.35 1
l=2
0.3 l=5
l=8 0.8
0.25

0.2 PDF 0.6 CDF

P(x<=k)
P(x=k)

0.15
0.4

0.1
0.2
0.05

0 0
0 5 10 15 20 0 5 10 15 20
k k

12
Middleton Model
• Middleton Class A Noise: summation of weighted Gaussian noises.

White Gaussian noise

C &)
%𝑚
A
𝐴 1 '
%D)
%
𝜎A = 𝜎?% + 𝜎,
𝑝@ (𝑛) = O 𝑒 '@ · 𝑒 *
𝐴
𝑚! 2πσ%A
AB3

weight factor

• A º Impulsive Index
Average number of impulses per second x Mean impulse duration. 𝐴 < 1
• Γ º Gaussian-to-Impulsive variance ratio (GIR), Γ = 𝜎?% ⁄𝜎,%
• 𝜎.# (or 𝜎&# ) is a background gaussian noise
• 𝜎(# is the impulsive noise variance
• Only the first 3-4 components of the expression are meaningful

13
Some graphs . . .

14
3

Other noise phenomena

 Flicker noise
• 1/𝑓, pink noise or Flicker noise:
• 1/𝑓 behavior below an 𝑓/ (corner freq.)
• Produced by slow variations in material
properties
• Impairs baseband communications, and
produces phase noise in oscillators,
which degrades passband comm.
a PLL converts it to a
higher frequency

flicker noise degrades

low freq. oscillator phase modulations
spectrum with phase are then degraded by
noise the phase noise

https://www.electronics-tutorials.ws/oscillator/crystal.html, https://www.analog.com/en/analog-dialogue/articles/phase-locked-loop-pll-fundamentals.html
https://www.digikey.gr/en/articles/converting-oscillator-phase-noise-to-time-jitter, https://dsplog.com/2012/07/09/evm-phase-noise/
16
Interference noise in Wireless communication
Wireless communications are usually limited by the
interference noise, rather than thermal noise. Some examples:

In 4G/5g urban deployments, eNB’s are separated 300-400

m, interference is ALWAYS the dominant noise
In WiFi @ 2.4 GHz there are only three 20 MHz non-
overlapping channels, interference is ALWAYS the dominant 17
Appendix
 Middleton Model - the Γ parameter
In practical terms:

C &)
'@
𝐴A 1 '
%D)
𝑝@ 𝑛 = O 𝑒 · ·𝑒 * ≈
𝑚! 2πσ%A
AB3
&) &)
1 '
%D) 𝐴 ' )
≈ 𝑒 '@ · ·𝑒 + + · 𝑒 %D, + 𝑜𝑡ℎ𝑒𝑟 𝑡𝑒𝑟𝑚𝑠
2πσ?% 2πσ)%

𝐵𝑎𝑐𝑘𝑔𝑟𝑜𝑢𝑛𝑑 𝑛𝑜𝑖𝑠𝑒 𝑝𝑜𝑤𝑒𝑟

≈
𝐼𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒 𝑛𝑜𝑖𝑠𝑒 𝑝𝑜𝑤𝑒𝑟 𝜎,% ≫ 𝜎?%

𝜎?% · 𝑃𝑟𝑜𝑏 𝑚 = 0 𝑐𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡 𝜎56 𝜎56 𝜎56

≈ = = ≈ = Γ
𝜎)% · 𝑃𝑟𝑜𝑏 𝑚 = 1 𝑐𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡 𝜎76 · 𝐴 6 𝜎!6 𝜎!6
𝐴 · 𝜎5 +
𝐴
𝑚 1
"
𝜎! = 𝜎#" + 𝜎$" → 𝜎%" = 𝜎#" + 𝜎$"
𝐴 𝐴 19
Middleton model with noise in dBm
• The noise voltage (n) pdf plot provides little information. But the pdf plot
when the noise is expressed in dBm, 𝑝𝑑𝑓 𝑛EFA , 𝑛EFA = 20𝑙𝑜𝑔)3 (𝑛EFA )
provides visually useful information (see Appendix)

• The abscissa of the first maximum of the 𝑝𝑑𝑓 𝑛EFA function is 𝜎?% , and
that of the second maximum is very approximately 𝜎)%

More powerful sources of noise are less probable

20
Middleton example
• Given the following model for Middleton’s impulsive noise, find:
a) Value of 𝜎.# and 𝜎(#
b) Value for Γ

𝜎?% = −20 𝑑𝐵𝑚 = 10'G 𝑊

𝜎,%
𝜎)% = 𝜎?% + = 20 𝑑𝐵𝑚 = 0.1 𝑊 →
𝐴
→ 𝜎,% = 𝐴 𝜎)% − 𝜎?% ≈ 𝐴𝜎)% = 0.02 𝑊

𝜎?%
10'G 𝑊
Γ= %= = 5. 10'H = −33 𝑑𝐵
𝜎, 0.02 𝑊

21
𝑝𝑑𝑓 𝑛&'(
20
𝑛89: = 20 · 𝑙𝑜𝑔7, 𝑛 + 30 = 𝑎 = = 𝑎 · ln 𝑛 + 30
𝑙𝑛10
;, =&'(
𝑛(𝑛89: ) = exp − <
· exp( <
)
𝑑𝑛 30 1 𝑛89:
𝑛> (𝑛89: ) = = exp − · · exp( )
𝑑𝑛89: 𝑎 𝑎 𝑎

@ =)(=&'()
𝐴: 1 A
6B)(
𝑝𝑑𝑓 𝑛89: = 2 · 𝑛> 𝑛89: ·P 𝑒 A( · 𝑒
𝑚! 2πσ6:
:?,

One power value comes

from two voltage values

22
How to generate noise samples from a customized PDF (1)

1. Calculate cdf (cumulative) from pdf

23
How to generate noise samples from a customized PDF (2)

2. Theory
• Let Y be a random variable in [0,1]. Let X be a random variable in range, say,
[a,b]. Let cdf(x) = y. What distribution does y follow?

1 Y = cdf(X)
y(n)

0 x
a x(n) b
𝑐𝑑𝑓(𝑥) = 𝑃 𝑋 < 𝑥 = 𝑃 𝑐𝑑𝑓 A7(𝑌) ≤ 𝑥 = 𝑃 𝑌 ≤ 𝑐𝑑𝑓(𝑥)

But 𝑃 𝑌 ≤ 𝑐𝑑𝑓(𝑥) = 𝑐𝑑𝑓 𝑋 <==> Y is a uniform distribution in [0,1]

• This means that if we want to generate random numbers that follow a
given cdf, say Middleton´s cdf, we generate random numbers y[n] with a
uniform distribution in [0,1], and calculate the desired numbers as
cdf -1(y[n])

24
How to generate noise samples from a customized PDF (3)

3. Last, map from uniform into Middleton distribution, with

random occurrence of sign

cdf = cdf/max(cdf);
Uniform xq = rand(1,N);
for ii=1:N
Distribution jj = min(find(cdf>=xq(ii)));
[0,1] end
xdbm(ii) = ndbm(jj);

xM = 10.^( (xdbm-30)/20 );
xM = xM.*(randi([0 1], 1,N)*2 - 1);

Middleton distribution [-60,+10] dBm

25