Professional Documents
Culture Documents
Bahirdar University Bahirdar Institute of Technology (Bit) Faculty of Electrical and Computer Engineering MSC in Communication Systems Engineering
Bahirdar University Bahirdar Institute of Technology (Bit) Faculty of Electrical and Computer Engineering MSC in Communication Systems Engineering
Bahirdar University Bahirdar Institute of Technology (Bit) Faculty of Electrical and Computer Engineering MSC in Communication Systems Engineering
i
Lists of Figure
Figure 1 Five-term, K = 5, approximation of the Class A model ................................................... 3
Figure 2 Two-term, K = 2, approximation of the Class A model ................................................... 5
Figure 3 Three-term, K = 3, approximation of the Class A model ................................................. 6
Figure 4 Bit error rate results using the impulse noise model shown in Fig 1, with K = 2, 3 and 5.
BPSK was used for the modulation. ............................................................................................. 14
ii
1. Short review of the task of the paper
The paper discusses on impulse noise, its models and how it affects communications systems.
And discuss the different impulse noise models. The impulse noise models discussed are
memoryless (Middleton Class A and Bernoulli-Gaussian), and with memory (Markov Middleton
and Markov-Gaussian). Looking at their similarities and differences in communications systems.
And further give simulation results of Performance comparisons in terms of bit error rates for
some of the variants of impulse noise models.
In the simulation results our task to generate figure 4 a and b of the BER vs SNR at fixed
impulse index and different Gaussian to impulse noise power ratio. In our simulation we
consider the Middleton class A noise model.
1
2. Middleton noise model
Statistical noise models which catered for noise due to both man-made and natural phenomena.
The most famous of these noise models is Middleton Class A noise model. Used to model the
effects of impulse noise in communications systems.
𝐹𝑀 (𝑛𝑘 ) = ∑∞ 2
𝑚=0 𝑃𝑚 𝑁(𝑛𝑘 ; 0, 𝜎𝑚 ) --- (1)
2)
Where 𝑁(𝑛𝑘 ; 𝜇, 𝜎𝑚 represents a Gaussian PDF with mean 𝜇 and a variance 𝜎 2 . From which
the 𝑘 𝑡ℎ sample 𝑛𝑘 is taken
𝐴𝑚 𝑒 −𝐴
𝑃𝑚 =
𝑚!
And
𝑚 𝑚
2
𝜎𝑚 = 𝜎𝐼2 + 𝜎𝑔2 = 𝜎𝑔2 ( + 1)
𝐴 𝐴Γ
2
Where; 𝜎𝐼 the variance of the impulse noise and 𝜎𝑔2 is the variance of the background noise
𝜎𝑔2
(AWGN). Γ= The Gaussian to impulse noise power ratio. Here A represents the density
𝜎𝐼2
′
𝐹𝑀,𝐾 (𝑛𝑘 ) = ∑∞ 2
𝑚=0 𝑃𝑚 𝑁(𝑛𝑘 ; 0, 𝜎𝑚 ) (2)
𝑃𝑚
Where 𝑃𝑚 ′ = 𝐾−1
∑𝑚=0 𝑃𝑚
Here in the generations of figure 4 we have thought or understand that the above formula was
used. So the paper, gives some results showing the bit error rate (BER) versus SNR, when using
the model in (2) for different K values.
2
In the simulations, the paper shall use up to the first five terms of model (2), and such a model
is shown below the figure.
Where BPSK modulation is used and K = 2, 3 and 5. In each simulation result/figure, the paper
use a theoretical BER curve for BPSK, for M = 2, where M is the order of the PSK modulation
and Eb is the signal’s bit energy). It was used as a reference curve against which all curves are
compared.
𝜎𝑔2
𝑃0′
𝜎𝐼2
+ 𝜎𝑔2
𝐴
𝑃1′
̅𝑘
𝐷
𝐷𝑘 2𝜎𝐼2
+ 𝜎𝑔2
𝑃2′ 𝐴
3𝜎𝐼2
𝑃3′ + 𝜎𝑔2
𝐴
𝑃4′ 4𝜎𝐼2
+ 𝜎𝑔2
𝐴
Figure 1 Five-term, K = 5, approximation of the Class A model
Here is the model used to observe the simulation of effect of different values of A and Γ on the
model in terms of bit error rate (BER) vs signal to noise ratio (SNR) using the above
mathematical model.
3
3.1. The detailed mathematical model/formula for the simulations
The paper uses the model in (2), which is given by:
′
𝐹𝑀,𝐾 (𝑛𝑘 ) = ∑∞ 2
𝑚=0 𝑃𝑚 𝑁 (𝑛𝑘 ; 0, 𝜎𝑚 )
′ 𝑃𝑚 𝐴𝑚 𝑒 −𝐴
𝑃𝑚 = ∑𝐾−1
, 𝑃𝑚 =
𝑚=0 𝑃𝑚 𝑚!
It uses the approximate of the occurrence probability of noise given by Poisson 𝑃𝑚 ′ based
on the model of the above figure the paper uses the formula for probability of error for BPSK
modulation for different values of K. It uses when K=2, 3, and 5.
3.1.1. Theoretical BER for AWGN
1
𝑃𝑒,𝐴𝑊𝐺𝑁_𝑇ℎ𝑒𝑜𝑟𝑦 = (𝑒𝑟𝑓𝑐(√𝑆𝑁𝑅))
2
𝑀−1 𝐸 𝑀−1 𝐸𝑏
𝑃𝑒,𝑀𝑃𝑆𝐾_𝑇ℎ = (1-A) 𝑒𝑟𝑓𝑐 (√ 𝑏2 ) + (A) 𝑒𝑟𝑓𝑐 (√ )
𝑀 𝜎𝑔 𝑀 𝜎𝑔2 (1+1⁄𝐴Γ)
1-A, and A = the probability occurring for AWGN noise and Impulsive noise respectively which
is given by the Poisson distribution.
4
1 1 𝑆𝑁𝑅
𝑃𝑒,𝑀𝑃𝑆𝐾_𝑇ℎ𝑒𝑜𝑟𝑦 = (1-A) 𝑒𝑟𝑓𝑐(√𝑆𝑁𝑅) + (A) 𝑒𝑟𝑓𝑐 (√ )
2 2 1+1⁄𝐴Γ
𝜎𝑔2
𝑃0′
𝑫𝒌 ̅𝒌
𝑫
𝜎𝐼2
𝑃1′ + 𝜎𝑔2
𝐴
𝐴𝑚 𝑒 −𝐴 ′ 𝑃𝑚
.𝑃𝑚 = And 𝑃𝑚 =
𝑚! ∑𝐾−1
𝑚=0 𝑃𝑚
𝑃0
𝑃0 ′ =
∑𝐾−1
𝑚=0 𝑃𝑚
𝐴1 𝑒 −𝐴
𝑃1 = = 𝐴𝑒 −𝐴
1!
𝑃1
𝑃1 ′ =
∑𝐾−1
𝑚=0 𝑃𝑚
Therefore,
1 1 𝑆𝑁𝑅
𝑃𝑒,𝐾=2 = (𝑃0 ′ ) 𝑒𝑟𝑓𝑐(√𝑆𝑁𝑅) + (𝑃1 ′ ) 𝑒𝑟𝑓𝑐 (√ )
2 2 1+1⁄𝐴Γ
5
BER for AWGN +Impulsive for K=3
′
For K=3, m runs from 0 to 2, hence we need to determine 𝑃2 . The 𝑃0 ′ and 𝑃1 ′ has already
determined in the above.
𝜎𝑔2
𝑃0′
2
𝑫𝒌 𝜎𝑫
𝐼̅ 𝒌
+ 𝜎𝑔2 ̅𝒌
𝑫
𝑃1′ 𝐴
𝑃2′ 2𝜎𝐼2
+ 𝜎𝑔2
𝐴
Figure 3 Three-term, K = 3, approximation of the Class A model
𝐴2 𝑒 −𝐴
𝑃2 =
2!
𝑃2
𝑃2 ′ =
∑𝐾−1
𝑚=0 𝑃𝑚
Therefore,
1 1 𝑆𝑁𝑅
𝑃𝑒,𝐾=3 = (𝑃0 ′ ) 𝑒𝑟𝑓𝑐(√𝑆𝑁𝑅) + (𝑃1 ′ ) 𝑒𝑟𝑓𝑐 (√ ) +
2 2 1+1⁄𝐴Γ
1 𝑆𝑁𝑅
(𝑃2 ′ ) 𝑒𝑟𝑓𝑐 (√ )
2 1+1⁄𝐴Γ
1 𝑆𝑁𝑅
𝑃𝑒,𝐾=3 = 𝑃𝑒,𝐾=2 + (𝑃2 ′ ) 𝑒𝑟𝑓𝑐 (√ )
2 1+2⁄𝐴Γ
6
𝐴4 𝑒 −𝐴
𝑃4 =
4!
𝑃4
𝑃4 ′ =
∑𝐾−1
𝑚=0 𝑃𝑚
Therefore,
1 1 𝑆𝑁𝑅
𝑃𝑒,𝐾=5 = (𝑃0 ′ ) 𝑒𝑟𝑓𝑐(√𝑆𝑁𝑅) + (𝑃1 ′ ) 𝑒𝑟𝑓𝑐 (√ ) +
2 2 1+1⁄𝐴Γ
1 𝑆𝑁𝑅 1 𝑆𝑁𝑅
(𝑃2 ′ ) 𝑒𝑟𝑓𝑐 (√ ) + (𝑃4 ′ ) 𝑒𝑟𝑓𝑐 (√ )
2 1+1⁄𝐴Γ 2 1+4⁄𝐴Γ
1 𝑆𝑁𝑅
𝑃𝑒,𝐾=5 = 𝑃𝑒,𝐾=3 + (𝑃4 ′ ) 𝑒𝑟𝑓𝑐 (√ )
2 1+4⁄𝐴Γ
%on impulse noise and its model. Simulations of Bit Error rate
over signal
%to noise ratio for Middleton class A noise models where BPSK
modulation is used
% for each simulation the theoretical BER is used
%Case 1:
%%%%%%%%%%%%%%% A = 0.01, Γ = 0.1%%%%%%%%%%%%%%%
r=0.1;
A=0.01;
SNRdB=0:1:40; %Signal to Noise Ratio (in dB)
SNR=10.^(SNRdB/10);
% Ber of AWGN Theory;
BER_AWGN_THEORY=0.5*erfc(sqrt(SNR));
semilogy(SNRdB, BER_AWGN_THEORY,'k-','linewidth',2)
hold on;
7
% theoretical simulation of AWGN+Impulsive noise
BER_Theory=(1-
A)*0.5*erfc(sqrt(SNR))+A*0.5*erfc(sqrt((SNR)/(1+1/(A*r))));
semilogy(SNRdB,BER_Theory,'c-','linewidth',2)
hold on
syms m
%%%%%%%%%%%%%%%%% K=2 %%%%%%%%%%%%%%%%%%%%%%%%
K=2;
Mmax=K-1
Pm=(A.^m*exp(-A))/(factorial(m));
Pms=symsum(Pm,m,0,K-1);
P_appr=Pm/Pms;
P0=exp(-A);
P1=A.^1*exp(-A);
P0_app=P0/Pms;
P1_App=P1/Pms;
% simulation of AWGN+Impulsive noise, K=2
BER_K2=P0_app*0.5*erfc(sqrt(SNR))+P1_App*0.5*erfc(sqrt((SNR)/(1+
1/(A*r))));
semilogy(SNRdB,BER_K2,'b*-','linewidth',2)
hold on
%%%%%%%%%%%%%%%%% K=3 %%%%%%%%%%%%%%%%%%%%%%%%
K=3;
Mmax=K-1;
Pm=(A.^m*exp(-A))/(factorial(m));
Pms=symsum(Pm,m,0,K-1);
P_appr=Pm/Pms;
P2=(A.^2*exp(-A))/(factorial(2));
P2_App=P2/Pms;
BER_K3=P0_app*0.5*erfc(sqrt(SNR))+P1_App*0.5*erfc(sqrt((SNR)/(1+
1/(A*r))))+...
P2_App*0.5*erfc(sqrt((SNR)/(1+2/(A*r))));
semilogy(SNRdB,BER_K3,'m*-','linewidth',2)
hold on
%%%%%%%%%%%%%%%%% K=5 %%%%%%%%%%%%%%%%%%%%%%%%
K=5;
Mmax=K-1;
Pm=(A.^m*exp(-A))/(factorial(m));
Pms=symsum(Pm,m,0,K-1);
8
P_appr=Pm/Pms;
P4=(A.^3*exp(-A))/(factorial(3));
P4_App=P4/Pms;
BER_K5=P0_app*0.5*erfc(sqrt(SNR))+P1_App*0.5*erfc(sqrt((SNR)/(1+
1/(A*r))))+...
P2_App*0.5*erfc(sqrt((SNR)/(1+2/(A*r))))+...
P4_App*0.5*erfc(sqrt((SNR)/(1+2/(A*r))));
semilogy(SNRdB,BER_K3,'k*','linewidth',2)
axis([min(SNRdB) max(SNRdB) 10^(-5) 1]);
grid on
xlabel(' Eb/2σ^2')
ylabel('Bit Error Rate')
title('A = 0.01, Γ = 0.1')
legend('AWGN,theory','AWGN+impulsiv,theory','AWGN+impulsive,K=2'
,.... 'AWGN+impulsive,K=3','AWGN+impulsive,K=5')
%Case 2:
%%%%%%%%%%%%%%%%%% A = 0.01, Γ = 0.01%%%%%%%%%%%%%%%%%%
r=0.01;
A=0.01;
SNRdB=0:1:50; %Signal to Noise Ratio (in dB)
SNR=10.^(SNRdB/10);
BER_AWGN_THEORY=0.5*erfc(sqrt(SNR));
figure;
semilogy(SNRdB, BER_AWGN_THEORY,'k-','linewidth',2)
hold on
BER_Theory=
(1-A)*0.5*erfc(sqrt(SNR))+A*0.5*erfc(sqrt((SNR)/(1+1/(A*r))));
semilogy(SNRdB,BER_Theory,'c-','linewidth',2)
hold on
syms m
9
P1_App=P1/Pms;
BER_K2=P0_app*0.5*erfc(sqrt(SNR))+P1_App*0.5*erfc(sqrt((SNR)/(1+
1/(A*r))));
semilogy(SNRdB,BER_K2,'b*-','linewidth',2)
hold on
%%%%%%%%%%%%%%%%% K=3 %%%%%%%%%%%%%%%%%%%%%%%%
K=3;
Mmax=K-1;
Pm=(A.^m*exp(-A))/(factorial(m));
Pms=symsum(Pm,m,0,K-1);
P_appr=Pm/Pms;
P2=(A.^2*exp(-A))/(factorial(2));
P2_App=P2/Pms;
BER_K3=P0_app*0.5*erfc(sqrt(SNR))+P1_App*0.5*erfc(sqrt((SNR)/(1+
1/(A*r))))+...
P2_App*0.5*erfc(sqrt((SNR)/(1+2/(A*r))));
semilogy(SNRdB,BER_K3,'m*-','linewidth',2)
hold on;
%%%%%%%%%%%%%%%%% K=5 %%%%%%%%%%%%%%%%%%%%%%%%
K=5;
Mmax=K-1;
Pm=(A.^m*exp(-A))/(factorial(m));
Pms=symsum(Pm,m,0,K-1);
P_appr=Pm/Pms;
P4=(A.^3*exp(-A))/(factorial(3));
P4_App=P4/Pms;
BER_K5=P0_app*0.5*erfc(sqrt(SNR))+P1_App*0.5*erfc(sqrt((SNR)/(1+
1/(A*r))))+...
P2_App*0.5*erfc(sqrt((SNR)/(1+2/(A*r))))+...
P4_App*0.5*erfc(sqrt((SNR)/(1+2/(A*r))));
semilogy(SNRdB,BER_K3,'k*','linewidth',2)
xlabel(' Eb/2σ^2')
ylabel('Bit Error Rate')
title('A = 0.01, Γ = 0.01')
axis([min(SNRdB) max(SNRdB) 10^(-5) 1]);
grid on
legend('AWGN,theory','AWGN+impulsiv,theory','AWGN+impulsive,K=2'
,....'AWGN+impulsive,K=3','AWGN+impulsive,K=5')
10
4.2. Results and Observation
11
For visibility we can give annotation for the graph like this
12
(b) A = 0.01, Γ = 0.01
13
For visibility we can give annotation for the graph like this
Figure 4 Bit error rate results using the impulse noise model shown in Fig 1, with K = 2, 3 and 5.
BPSK was used for the modulation.
Observations:
It can be observed from the above figure (simulation) that the model in (2) approximates the
Class A model, better than the model in (1) for low values of A. because the mathematical model
given in (2) has the approximated Poisson , i.e. 𝑃𝑚′ .
When we calculated the value of 𝑃𝑚′ for different values of K=2, 3 and 5, it approximated to the
impulse index as 𝑷𝟎 ′ ≈ (𝟏 − 𝑨) and the rest (𝑷𝟏 ′ , 𝑷𝟐 ′ and 𝑷𝟒 ′ )≈ 𝑨. Thus this results the BER
versus signal to noise ratio for the given values of K and constant values of A, gives Avery good
approximations to the theoretical BER curve, almost approximated as we have seen the result.
Finally as Γ decrease BER increases or it will more impulsive.
14