You are on page 1of 5

Optimal Threshold Calculation for Improved

Impulsive Noise Mitigation in the Frequency


Domain
Israa Al-Neami*, Ghanim A. Al-Rubaye, Martin Johnston, Charalampos Tsimenidis
School of Electrical and Electronic Engineering, Newcastle University, Newcastle Upon Tyne, NE1 7RU, UK
Email: {I.Al-Shaikhli1, g.a.m.al-rubaye, martin.johnston, charalampos.tsimenidis}@newcastle.ac.uk

with extreme values are identified to detect the impulses in


Abstract—Impulsive noise (IN) can significantly degrade the the FD. Furthermore, in [6] pilot tones are used to identify
performance of multi-carrier based communication systems. In the position of noise impulses. Recent work in this topic is
order to improve the performance of such systems, conventional
IN suppression methods have been proposed to process the sig- described in [7], [8].
nal in the time domain. However, attention has recently focused The estimation process of the effects of IN are based on
on using methods implemented in the Frequency Domain. In preliminary decisions on the transmitted signals. Most work
general, the aim of IN frequency domain estimation schemes on this topic, which is based on FD signal processing, assume
is to consider noise samples whose amplitudes are below a a fixed threshold (FT). By using a FT value, the amplitude of
fixed threshold as Gaussian distributed noise and therefore
are blanked (nulled). In contrast to existing work, this paper IN sample exceeding a FT are considered to be an impulse
presents a novel optimal threshold calculation method to im- noise pulse, otherwise the amplitudes are assumed to be noise
prove the IN frequency domain mitigation methods. The system pulses and then blanked. A FT value does not always make
adopt in this paper is the G.fast standard which uses high order the correct decision for all assumed IN occurrences. Selecting
QAM constellations over copper channels. The derived optimum a very high value of FT may blank the true IN samples and
threshold is used in an IN compensation method based on the
relationship between two parameters,namely, impulsive Index a very low value of FT may not blank all the AWGN noise
A and the noise ratio Γ. In order to validate the effectiveness samples. Therefore, finding the optimal value of threshold is
of the proposed scheme, its performance is compared with that essential for reaching the best performance [9], [10].
of the conventional FD mitigation method which uses a fixed This paper proves that the performance of a mitigation
threshold. The impact of both the impulsive Index A and the noise system is threshold-dependent. Therefore, the main
noise ratio Γ on the DMT-based G.fast copper channel is also
discussed in detail. Results show that the proposed method contribution of this paper is to propose an optimal threshold
considerably outperforms conventional frequency domain IN calculation method to enhance the capability of IN cancella-
mitigation schemes. tion in the frequency domain. A direct relationship between
Index Terms—Optimal threshold, Middleton Class A impul- two parameters,namely, impulsive Index A and the noise
sive noise (MCA), Frequency Domain Mitigation Techniques, ratio Γ, is derived. The proposed method can be applied, for
G.fast over Copper channel Model.
example, to the G.fast standard and successive generations
I. I NTRODUCTION of Digital Subscriber Line (DSL) technology, which aim to
achieve data rates up to 1 Gb/s over very short copper loops
A major problem in several communication systems is up to 250 meters [11].
the presence of Impulse Noise (IN), which is a limiting The rest of this paper is organized as follows. Section
factor in many applications such as power lines, digital II describes the DMT G.fast coding system model over
subscriber line (DSL) and some wireless communication copper channels and proposed threshold-calculation method.
systems [1]. In multi-carrier systems, a traditional technique In Section III, the signal-to-noise (SNR) performance of an
to mitigate the effects of IN has been proposed in [2], [3]. OT is compared with a fixed threshold (FT) by simulations
The concept of time domain (TD) mitigation methods is to at different levels of IN. Finally, Section IV concludes the
identify high signal amplitudes and then apply clipping or paper.
nulling (Blanking) to these peaks when they are above a
fixed threshold value. However, the literature also proposes II. G. FAST IMPELEMENTATION AND CHANNEL
using IN cancellation depending on the frequency domain MODEL
(FD) signal processing after the multi-carrier demodulator A. DMT-based G.fast System Over Impulsive Noise Copper
[4], [5], [6]. The idea of IN frequency domain mitigation is Channel
to compensate for the IN samples and subtract them from
The block diagram for a DMT-based G.fast system over
the output of the equalizer. In [4], the authors described an
a copper channel with a blanking non-linear compensator is
iterative process where the information is exchanged between
illustrated in Fig. 1. It can be seen that the G.fast system
estimators that operate on the signal in the FD and TD.
model is constructed using a concatenated Reed-Solomon
Similar work also has appeared in [5] where sub-carriers
code and 4-Dimensional TCM code, which are separated by
*Israa Al-Neami is staff member with the University of Technology in an interleaver. The information data is first encoded by the
Iraq. outer (RS) code and then fed to the inner 4-Dimensional
where H is an M × 1 vector made up of the FFT of
the channel impulse response h, X is an M × 1 vector
which represents the DFT of the transmitted signal x, W
is the FFT of w and represents the background noise in the
frequency domain and I is the DFT of the impulsive noise
i, respectively.
In order to develop the receiver, we assume ideal channel
estimation (i.e. H̃ ≡ H). Now, the received signal R can be
equalized by the frequency-domain channel equalizer (FEQ)
Fig. 1. Block diagram of DMT based G.fast system over an IN copper and expressed as:
channel with an IN compensator at the receiver
Req = RHˆ−1 = X + W Hˆ−1 + I Hˆ−1 (7)
B. Middleton Class A Distribution
TCM encoder which comprises a 2/3 binary convolutional
encoder and 4-D 32 QAM signal mapper. Then the coded This paper assumes a Middleton Class A distribution [9],
sequence is passed to the DMT modulator. [14], [15] which is a well-known model used widely to
For a standard DMT system, the fast Fourier transform represent the effects of IN in communication systems. Based
(FFT) algorithm for modulation and demodulation is used. on this model, the probability density function (PDF) can be
The discrete time-domain signal generated by means of an expressed as:
inverse fast Fourier transform (IFFT) can be written as



2N −1 e−A As 1 x2
1 n p(x) = .√ e(− 2σ2 ) (8)
sn = √ Sk ej 2πk , n = 0, 1, ..., N − 1 (1) s! 2
2πσs
2N k=0 2N s=0

where
where Sk (n = 1, 2, ..., N −1) are complex QAM symbols
[12], [13]. ( s )
In order to prevent intersymbol interference (ISI), a cyclic +Γ σ 2W
σs2 =σ 2 A
, σ 2 = σW
2
+ σI2 and Γ = (9)
prefix is inserted to the parallel to serial converted symbols, 1+Γ σ 2I
resulting in [13]. Where A is referred to as the impulsive index and is
inversely proportional to the impulsive behavior. The noise
s̃ = [sN −µ , sN −µ+1 , ..., s0 , s1 , sN −1 ]T (2) becomes Gaussian distributed when A → ∞. The parameters
2
σW and σI2 represent the Gaussian noise power and impulsive
where µ is the length of the cyclic prefix, which should noise power, respectively. The parameter σ 2 is the total noise
obey the constraint µ∆t = τmax , power and the ratio between the mean power of the Gaussian
The DMT signal s̃ is transmitted over an IN Copper and the mean power of IN component is represented by Γ.
channel modeled by a Middleton Class A (MCA) distribution
[9], [14], [15] and the received signal r (in TD) is given by: C. Frequency-domain Methods
r = hs̃ + i + w (3) Impulsive noise mitigation methods in the frequency do-
main have been presented in [7], [8], [18]. The basic principle
where the received signal r is a (N + µ) × 1 vector. The behind this method is to estimate the positions and amplitudes
impulse response channel h is an (N + µ) × 1 vector, while of IN in the frequency domain after the OFDM demodulator
the IN and AWGN are (N + µ) × 1 vectors. As shown in and the channel equalizer output, then subtract them from
previous research, various cable models have been described the received signal. This method is very effective for higher-
for DSL copper channels, such as the BT0H model, Chen’s order QAM constellations, such as those used in the G.fast
model and KPN1. The twisted pair copper channel model system.
used in this paper is Chen’s model which is suitable in The received OFDM signal over a copper channel in the
frequencies up to 200 MHz and causal impulse responses as presence of IN can be expressed as [7]:
defined in [16], [17] as:

Definition (Chen): The propagation constant, γ = α + jβ Rk = Hk Xk + Wk + Ik , k = 0, 1, . . . , N − 1, (10)


is modelled by where Wk is the DFT of the AWGN samples and Ik is the

α = K1 f + K2 f (4) DFT of the IN samples. Based on the algorithm presented in
[7], the received signal at the zero forcing equalizer (ZFE)
β = K3 f (5)
output can be expressed as
where α is the attenuation constant and β is the phase Rk Wk Ik
(eq)
constant. Rk = = Xk + + , k = 0, 1, . . . , N − 1,
Hk Hk Hk
At the receiver, the cyclic prefix is removed, the received (11)
symbols after a FFT can be recovered as:
The main idea is based on the estimation of the equalized
R = HX + I + W (6) Ik
IN samples H and then to subtract this from the equalized
k
Fig. 2. Block diagram of IN compensator

(eq)
received signal samples Rk . This operation can provide the This model can be expressed as mixture of two Gaussian
estimated DFT of the transmitted signal X̂k by utilizing the PDFs, each with zero mean but with different variances as
Maximum Likelihood (ML) criterion based on the minimum presented by the Spaulding and Middleton model in [20]
Euclidean distance computation. The total noise samples
Dk = Wk + Ik in the frequency domain can be estimated as pA (dˆℜ
n) = e
−A
N (dˆℜ
n , 0, σw ) + (1 − e
2 −A
)N (dˆℜ 2
n , 0, σI ),

(eq) pA (dˆℑ
n) = e
−A
N (dˆℑ
n , 0, σw ) + (1 − e
2 −A
)N (dˆℑ 2
n , 0, σI ).
D̂k = Hk (Rk − X̂k ), k = 0, 1, . . . , N − 1, (12)
(16)
Therefore, we need to estimate the IN samples Iˆk from the
For complex noise, the magnitude of the simplified IN form
estimated total noise samples D̂k . That can be achieved by
in (16) exhibits a mixture of two Rayleigh PDFs as
transforming D̂k into the time domain dˆk by utilizing an
e−A |dˆn | − |2σ (1 − e−A )|dˆn | − |d̂2σn2|
2
IDFT as d̂n |2
{ } p(|dˆn |) = e 2
w + e I , (17)

dˆn = F −1 Hk (Rk − X̂k ) , n = 0, 1, . . . , N − 1,


(eq) 2
σw σI2
(13) where (1 − e−A ) is the probability of an impulse occurring,
( 1
) σ2
Then the IN samples in the TD can be estimated using the σI2 = σw
2
1 + AΓ and Γ = σw2 . The decision rule based on
I
following condition: the ML criterion [21] can be expressed as

{ 

|d̂ |2
− n2
dˆn , if |dˆn | > Th σ̂ 2 
î ∈ impact impulsive, if
|d̂n |
e 2σw
, n = 0, 1, · · · , N − 1, (14)
2
≥1
σ
în = n
w
|d̂ |2
0, otherwise , ML = − n2 .


|d̂n |
e 2σ
I

where Th is the Fixed Threshold (FT). This is a crucial 



σ2
I
în ∈ impulsive free, elsewhere
factor in impulsive noise compensation process as it is re-
(18)
sponsible for deciding which amplitude |dˆn | is to be blanked
[19] (i.e. whether the amplitude |dˆn | belongs to the AWGN The optimal threshold TM opt
L based on the ML criterion
noise or to the IN). threshold must satisfy the following condition
However, one FT value is not suitable for all the SNR
opt 2 opt 2
|T |
values for different probabilities of IN. Therefore, an optimal |TMopt opt
L | − 2σw |TM L|
|T |
ML − ML
2 2σ 2
estimation threshold exists which is derived in the next 2
e = 2 e I , (19)
σw σI
section.
We can also simplify (19) as
D. The Proposed Threshold-Calculation Method for fre-
opt 2 opt 2
quency domain IN mitigation method σI2 −
|T
ML
|
+
|T
ML
|
2σ 2 2
2σw
We propose that instead of fixing a threshold, blanking will 2
=e I ,
σw
( )
take place at higher SNRs if it is implemented at an OT value σ 2 −σ 2
σI2 |TMopt 2
L|
I
2σ 2 σw2
w
determined by the distribution characteristics of the IN.The =e I ,
σw2
block diagram of the proposed impulsive noise compensation ( 2)
method is shown in Fig. 2. The proposed calculation method opt 2 2σI2 σw 2
σI
|TM | = ln ,
for OT is shown as follows L
σI2 − σw 2 2
σw
√ ( 2)
The real and imaginary distributions of the MCAIN model
opt 2σI2 σw2 σI
can be expressed as |TM L | = ln
( ) σI − σw
2 2 σw 2

∑ e−A Aℓ 1 −
(d̂ℜ
n)
2 √ ( )
pA (dˆℜ
n) = √ e 2σ 2
ℓ , 2 (1 + AΓ) ln 1 +
1
ℓ! 2πσℓ2 = 2σw . (20)
ℓ=0
( ) AΓ
∑∞ (d̂ℑ 2
e−A Aℓ 1 − n)
pA (dˆℑ
n) = √ e 2σ 2
ℓ , (15) Subsequently, the estimated samples are transformed to the
ℓ! 2πσℓ2
ℓ=0 FD Iˆk and then divided over the channel frequency response
Hk . The signal after the frequency domain IN mitigation 100

method can be expressed as


(out) (eq) Iˆk
Rk = Rk − . (21) 10-1
Hk
(out)
where the compensated signal Rk is an N × 1 vector
10-2
represented in the FD by means of the FFT.

BER
III. SIMULATION RESULTS
10-3
In this section, simulation results for a DMT-based G.fast
model with RS+TCM error-correction coding over a copper
channel affected by Middletons class A IN are presented. 10-4

Note that 32-QAM has been utilized for all simulations. A=0.001

A=0.1

A=1
-5
10
0 5 10 15 20 25 30 35
SNR (dB)
100
Fig. 4. BER performance comparison in the presence of IN of our proposed
method over 32QAM G.fast copper channel system with different values of
No Cancelation
impulsive noise index A = 0.001, A = 0.1 and A = 1
10-1

10-2 100
BER

No Cancelation
10-3
(FD methods with fixed Th) 10-1
Optimum Th

Fixed Th1=8.5
Proposed method
Fixed Th2=8.9 (Used Optimal Th )
10-4
Fixed Th3=9.5 10-2

Fixed Th4=10.5
BER

Fixed Th5=11.3
10-5
0 5 10 15 20 25 30 35 10-3
SNR (dB)

Fig. 3. BER performance comparison in the presence of IN between the


proposed optimal threshold method and FD mitigation method presented in 10-4

[7] over 32QAM G.fast copper channel system Gamma=1


Gamma=0.1
Gamma=0.01
-5
10
To begin with, Fig. 3 shows a BER performance com- 0 5 10 15 20 25 30 35 40 45
SNR (dB)
parison between the nonlinearity FD method which uses a Fig. 5. BER performance in the presence of IN of our proposed method with
fixed threshold value as presented in [7] and our proposed three different values of Gamma parameter,simulated results for 32QAM
G.fast copper channel system
method of calculating the optimal threshold obtained from
(20). The method presented in [7] depends on a FT value
for the whole range of SNR values. Accordingly, it has
been tested for the range of all possible threshold values. IV. CONCLUSION
The closest three BER performance curves to AWGN are An optimal threshold expression is derived in this paper
observed at a FT=[Th1=8.5, Th2=8.9, Th3=9.5, Th4=10.5, to achieve the optimal performance of IN cancellation meth-
Th5=11.3] as depicted in Fig.1. It is obvious that the BER ods based in the frequency domain. A direct relationship
performance results from our optimal threshold calculation between the parameters impulsive index A, the noise ratio
method outperforms all BER performance results obtained Γ and optimal threshold is derived in this work. Moreover,
using a fixed threshold value. the performance of this FD mitigation scheme with OT is
Additionally, the effect of varying the impulsive index compared with FT in terms BER using the MCA IN model.
parameter A and Γ on the BER performance is shown in The effect of changing A and Γ was investigated and it was
Fig.4 and Fig.5 respectively. The BER performance is tested found that varying A and Γ for three different values does
at three different values of the A for (A = 0.001, 0.1 and cause a significant change in the BER performance. Results
1). It is evident that as A increases the BER performance also showed that the proposed scheme is more superior than
improves with increasing SNR as shown in Fig. 4. other existing IN mitigation techniques.
The last set of results is shown in Fig. 5 which illustrates
R EFERENCES
how the parameter Γ affects the BER performance of the
proposed method obtained at three different values of Γ (Γ = [1] J. W. Cook, R. H. Kirkby, M. G. Booth, K. T. Foster, D. E. Clarke,
and G. Young, “The noise and crosstalk environment for adsl and vdsl
0.01, 0.1 and 1). It is clearly shown that the BER performance systems,” IEEE Communications Magazine, vol. 37, no. 5, pp. 73–78,
is improved significantly with higher Γ values. 1999.
[2] J. Armstrong and H. A. Suraweera, “Impulse noise mitigation for ofdm
using decision directed noise estimation,” in Spread Spectrum Tech-
niques and Applications, 2004 IEEE Eighth International Symposium
on. IEEE, 2004, pp. 174–178.
[3] S. V. Zhidkov, “Analysis and comparison of several simple impulsive
noise mitigation schemes for ofdm receivers,” IEEE Transactions on
Communications, vol. 56, no. 1, 2008.
[4] J. Haring and A. H. Vinck, “Iterative decoding of codes over complex
numbers for impulsive noise channels,” IEEE Transactions on Infor-
mation Theory, vol. 49, no. 5, pp. 1251–1260, 2003.
[5] M. Sliskovic, “Signal processing algorithm for ofdm channel with
impulse noise,” in Electronics, Circuits and Systems, 2000. ICECS
2000. The 7th IEEE International Conference on, vol. 1. IEEE, 2000,
pp. 222–225.
[6] F. Abdelkefi, P. Duhamel, and F. Alberge, “On the use of pilot tones for
impulse noise cancellation in hiperlan2,” in Signal Processing and its
Applications, Sixth International, Symposium on. 2001, vol. 2. IEEE,
2001, pp. 591–594.
[7] S. V. Zhidkov, “Impulsive noise suppression in ofdm-based communi-
cation systems,” IEEE Transactions on Consumer Electronics, vol. 49,
no. 4, pp. 944–948, 2003.
[8] J. Armstrong and H. A. Suraweera, “Impulse noise mitigation for ofdm
using decision directed noise estimation,” in Eighth IEEE International
Symposium on Spread Spectrum Techniques and Applications - Pro-
gramme and Book of Abstracts (IEEE Cat. No.04TH8738), Aug 2004,
pp. 174–178.
[9] E. Alsusa and K. M. Rabie, “Dynamic peak-based threshold estimation
method for mitigating impulsive noise in power-line communication
systems,” IEEE Transactions on Power Delivery, vol. 28, no. 4, pp.
2201–2208, 2013.
[10] K. Rabie, E. Alsusa, A. Familua, and L. Cheng, “Constant envelope
ofdm transmission over impulsive noise power-line communication
channels,” in Power Line Communications and its Applications (IS-
PLC), 2015 International Symposium on. IEEE, 2015, pp. 13–18.
[11] J. Maes and C. J. Nuzman, “The past, present, and future of copper
access,” Bell Labs Technical Journal, vol. 20, pp. 1–10, 2015.
[12] E. Martos-Naya, J. López-Fernández, L. D. del Rio, M. C. Aguayo-
Torres, and J. E. Munoz, “Optimized interpolator filters for timing
error correction in dmt systems for xdsl applications,” IEEE Journal
on Selected Areas in Communications, vol. 19, no. 12, pp. 2477–2485,
2001.
[13] T. Bai, H. Zhang, R. Zhang, L.-L. Yang, A. F. Al Rawi, J. Zhang, and
L. Hanzo, “Discrete multi-tone digital subscriber loop performance in
the face of impulsive noise,” IEEE Access, vol. 5, pp. 10 478–10 495,
2017.
[14] W. Abd-Alaziz, Z. Mei, M. Johnston, and S. L. Goff, “Non-binary
turbo-coded ofdm-plc system in the presence of impulsive noise,” in
2017 25th European Signal Processing Conference (EUSIPCO), Aug
2017, pp. 2576–2580.
[15] D. Middleton, “Statistical-physical models of electromagnetic interfer-
ence,” IEEE Transactions on Electromagnetic Compatibility, no. 3, pp.
106–127, 1977.
[16] D. Acatauassu, S. Host, C. Lu, M. Berg, A. Klautau, and P. O.
Borjesson, “Simple and causal twisted-pair channel models for g. fast
systems,” in Global Communications Conference (GLOBECOM), 2013
IEEE. IEEE, 2013, pp. 2834–2839.
[17] T. Magesacher, M. Jakovljevic, M. Loiola, P. Odling, P. Borjesson
et al., “Measurement and modeling of short copper cables for ultra-
wideband communication,” in Proceedings of SPIE-The International
Society for Optical Engineering, 2006.
[18] H. Matsuo, D. Umehara, M. Kawai, and Y. Morihiro, “An iterative
detection for ofdm over impulsive noise channel,” in Proceedings of
the 6th International Symposium on Power-Line Communications and
its Applications (ISPLC), 2002, pp. 27–29.
[19] Z. Ali, F. Ayaz, and C.-S. Park, “Optimized threshold calculation
for blanking nonlinearity at ofdm receivers based on impulsive noise
estimation,” EURASIP Journal on Wireless Communications and Net-
working, vol. 2015, no. 1, p. 191, 2015.
[20] A. Spaulding and D. Middleton, “Optimum reception in an impulsive
interference environment–part i: coherent detection,” IEEE Trans. Inf.
Theory, vol. 25, no. 9, pp. 910–923, 1977.
[21] J. G. Proakis, “Digital communications. 2008,” McGraw-Hill, New
York.

You might also like