Communications - Ankara Üniversitesi

Transcription

COMMUNICATIONS
DE LA FACULTE DES SCIENCES
DE L’UNIVERSITE D’ANKARA
FACULTY OF SCIENCES
ANKARA UNIVERSITY
Series A2-A3: Physics, Engineering Physics, Electronics/Computer Engineering,
Astronomy and Geophysics
VOLUME: 57
Number: 1
Faculy of Sciences, Ankara University
06100 BeĢevler, Ankara-Turkey
ISSN 1303-6009
YEAR: 2015
FACULTY OF SCIENCES
ANKARA UNIVERSITY
Series A2-A3: Physics, Engineering Physics, Electronics/Computer Engineering,
Astronomy and Geophysics
Owner
Hüseyin BEREKETOĞLU
Editor in Chief
Nuri ÖZALP
Managing Editor
A. Ulvi YILMAZER
Area Editors
Ali YAMAN (Physics)
Iman ASKERZADE(Askerbeyli)(Computer Engineering)
Tülay SERĠN (Engineering Physics)
Ziya TELATAR(Electronic Engineering)
H. Volkan ġENAVCI (Astronomy)
M. Emin CANDANSAYAR (Geophysical Engineering)
This Journal is published two issues in a year by the Faculty of Sciences, University of Ankara. Articles and
any other material published in this journal represent the opinions of the author(s) and should not be construed
to reflect the opinions of the Editor(s) and the Publisher(s).
Correspondence Address:
COMMUNICATIONS
DERGĠ BAġEDĠTÖRLÜĞÜ
Ankara Üniversitesi Fen Fakültesi,
06100 Tandoğan, ANKARA – TURKEY
Tel: (90) 312-212 67 20 Fax: (90) 312-223 23 95
Print:
Ankara Üniversitesi Press
ĠncitaĢ Sokak No:10 06510 BeĢevler
ANKARA – TURKEY
Tel: (90) 312-213 65 65
Publication Date::
e-mail: nozalp@science.ankara.edu.tr
©Ankara
University Press, Ankara 2015
COMMUNICATIONS
VOLUME: 57
FACULTY OF SCIENCES
ANKARA UNIVERSITY
Number: 1
YEAR: 2015
Series A2-A3: Physics, Engineering Physics, Electronics/Computer Engineering, Astronomy and
Geophysics
M. KÖSE, S. TAŞCIOĞLU, Selçuk, Z. TELATAR, Wireless device identification
using descriptive statistics ………………………….................................................
1
M. KÖSE, S. TAŞCIOĞLU, Selçuk, Z. TELATAR, Signal-to-noise ratio
estimation of noisy transient signals ……………………………………………........ 11
F. SARI, F. OZEK, Visibility based optical wireless availability assessment for
continental climate conditions ..................................................................................... 21
COMMUNICATIONS
VOLUME: 57
FACULTY OF SCIENCES
ANKARA UNIVERSITY
Number: 1
YEAR: 2015
Geophysics
Commun.Fac.Sc.Univ.Ank.Series A2-A3
Volume 57, Number 2, Pages 1-10 (2015)
DOI: 10.1501/commua1-2_0000000083
ISSN 1303-6009
WIRELESS DEVICE IDENTIFICATION USING DESCRIPTIVE STATISTICS
MEMDUH KÖSE, SELÇUK TAġCIOĞLU, ZĠYA TELATAR
Ahi Evran Univ, Computer Sciences Research and Application Center Kirsehir, Turkey
Ankara Univ, Faculty of Engineering, Electrical and Electronics Engineering Department Ankara,
Turkey
E-mail : memduh.kose@ahievran.edu.tr, {selcuk.tascioglu, ziya.telatar}@eng.ankara.edu.tr
(Received: March 13, 2015; Accepted: April 19, 2015 )
ABSTRACT
Physical layer identification systems classify the wireless devices by exploiting some distinctive
device characteristics which can be measured from transmissions. In this paper, characteristic
features based on descriptive statistics are extracted from the transient part of the received WiFi
signals for classification purpose. Classification performance of the features is evaluated through
experimental data. Feature sets consisting of combinations of descriptive statistics are tested in
order to evaluate the distinctiveness of statistics which measures the central tendency and dispersion
of data. Classification results for independent trials are presented in terms of mean and standard
deviation of classification accuracy as well as confusion matrix.
KEYWORDS: Wireless device identification, transient signal, descriptive statistics,
classification performance evaluation
1. INTRODUCTION
Wireless device identification process based on some unique characteristics caused by
hardware imperfections in transmitter circuitry is called physical layer identification.
Various feature extraction and classification methods based on physical layer
characteristics of devices have been proposed in order to identify wireless devices [1]-[6].
This identification approach has been shown to be useful to enhance the wireless network
security, e.g. in [1] security enhancement was achieved by identifying IEEE 802.11b
WiFi devices through experimental data. Danev et al. presented a survey of physical layer
identification systems in a systematic way [7].
© 2015 Ankara University
2
Distinctive features of devices can be measured from various parts of transmitted signals
such as transients, preambles and data. Accuracy for the detection of the signal part to be
used has a substantial effect on the identification system performance, especially for
transient based approaches since the detection of transients is a difficult task due to their
noise-like characteristics [1]. Several transient detection algorithms such as multifractal
segmentation method [8], Bayesian step detector [9] and Bayesian ramp detector [10]
have been proposed so far.
After detecting signal part, appropriate features are defined so as to represent the unique
characteristics of the device. For this purpose, descriptive statistics obtained from signal
characteristics such as amplitude frequency and phase responses have been employed. For
example, statistics such as standard deviation, variance, skewness, and kurtosis have been
used as features for classification of wireless devices [2]-[6]. Time domain and wavelet
domain statistical features were extracted from preamble part of 802.11a signals in [2].
Statistical features were generated from amplitude, phase and frequency characteristics of
transmitted signals. Classification of devices was performed through a multiple
discriminant analysis/maximum likelihood in a Bayesian framework. In addition to the
time domain and the wavelet domain schemes a feature extraction method in spectral
domain was proposed in [3]. Spectral domain features were obtained from the power
spectral density of emitted WiMax signals. In [4], classification performance of features
extracted from instantaneous phase responses of midmable and near-transient signal
regions of GSM signals were compared using the experimental data. It was shown that a
better classification performance was achieved with the features extracted from neartransient signal regions. Harmer et al. showed that statistical features obtained from time
domain preambles and corresponding spectral domain responses based on normalized
power spectral density can be employed for classification of WiFi and WiMax signals [5].
Rehman et al. proposed to use statistical features from the energy envelope of transient
signals for classification of Bluetooth devices [6]. Feature vector is defined to consist of
duration of the transient, maximum slope of the transient curve, area under the curve in
addition to statistical features based on the first four moments. It was shown that the
proposed features can be employed with a high accuracy rate at low sampling rates. Zhao
et al. proposed an identification method base on features obtained from the envelope of
the transient signal using the compressed sensing theory [11]. Transient envelop is
calculated by using complex analytical wavelet transform. It was showed that the
WIRELESS DEVICE IDENTIFICATION
3
identification performance of this method is better than that of the method proposed in [6]
for the case of high attenuation.
In this paper, a transient based classification system is considered. Transient detection is
carried out using Bayesian ramp detector proposed in [10]. Statistical features obtained
from instantaneous amplitude responses of transients are used as features. Classification
performance of descriptive statistics which measures the central tendency and the
dispersion of data have been evaluated separately and together.
The organization of the paper is as follows: In section 2, transient characteristics of WiFi
signals is presented. In section 3, features based on descriptive statistics are defined.
Performance evaluation of the classification method is given in Section 4. Section 5
concludes the paper.
2. TRANSIENT CHARACTERISTICS OF WIFI SIGNALS
A radio transmitter generates transient signal until a stable carrier is produced. Transient
signals have unique characteristics which can be exploited for identification of these
devices. An accurate transient detector increases the performance of transient based
identification systems [1], [7]. Transient signal behavior can be measured in terms of
instantaneous signal characteristics, such as instantaneous amplitude and phase [1]. In this
paper, we used instantaneous amplitude characteristics to extract unique characteristics of
transients. Instantaneous amplitudes of the real valued experimental signals were obtained
from the corresponding analytical signals calculated by using Hilbert transform [12].
Figure 1 shows an example for the instantaneous amplitude of a transmission from a WiFi
device. As shown in this figure, recorded transmission contain transient signal following
the channel noise and the steady state preamble part of the received signal. Therefore, the
first task of the classification system is to detect the transient part of instantaneous
amplitude signal. Transient starting point is estimated using a Bayesian ramp detector
which is proposed in [10]. In this method, transient signal detection problem was
formulated as a change-point detection problem and a ramp signal model was employed
to separate the transients from the channel noise in a Bayesian framework. Transient
duration for WiFi devices was found to be around 200 ns in [1], which corresponds 200
samples for the test signals in this work. After detecting transient signals, the unique
characteristics based on descriptive statistics were extracted from instantaneous
amplitudes of transients.
4
Figure 1. Instantaneous amplitude for a received signal from a WiFi device
3. FEATURES BASED ON DESCRIPTIVE STATISTICS
Instantaneous characteristics such as instantaneous amplitude and phase of received
signals from wireless devices have some distinctive properties caused by transmitters.
These properties have been used to identify wireless devices. For example in [1],
instantaneous amplitudes and dimension reduced form of these characteristics, obtained
by using principle component analysis, were used as features. In this paper, we
summarize the instantaneous amplitudes through the descriptive statistics in order to
reduce the feature dimension while keeping the ability to distinguish transmitters.
Characteristics of a number of the descriptive statistics can be visualized by means of
boxplot of data. For this purpose, average over fifty received signals was calculated for
each transmitter and statistical summary of these average signals is provided by a boxplot
in Figure 2. The notches of the boxes indicate the second quartile (Q2) values which is
also called the median. The edges of the boxes correspond to the first quartile (Q1), also
called the 25th percentile, and the third quartile (Q3), also called the 75th percentile,
values. The lengths of a boxes give interquartile range (IQR=Q3-Q1) values. The edges of
dashed lines represent the most extreme data points. As seen from this figure, there are
statistically significant differences between some of the statistics of different transmitters
while some of them are close to overlap. For example, Q3 values of Tx6 is far away from
5
those of other transmitters while Q2 values of Tx6 and Tx1 are closer to each other. Thus,
classification of these devices cannot be accomplished by using a simple threshold. In this
paper, a PNN (probabilistic neural network) classifier is used to carry out the
classification.
Figure 2. Boxplot of averaged data from each transmitter
Descriptive statistics can be categorized in two groups according to data characteristics to
be measured [13]. Statistics such as mean and median measure the central tendency,
average and location of data, while statistics such as standard deviation, skewness and
kurtosis measure the dispersion or variability of data. A list of statistics used as features in
this study is presented in Table 1.
4. CLASSIFICATION PERFORMANCE EVALUATION
Classification performance of the features based on descriptive statistics was evaluated
experimentally. Experimental data from six different IEEE 802.11b WiFi devices
operating in the 2.4 GHz ISM band were used to evaluate the classification performance.
Data set contains fifty transmissions from each wireless device. In the PNN classifier, ten
of fifty measurements were used as training vectors and remaining forty measurements
were employed as test vectors for a trial. The test results presented herein were given over
100 independent trails. In each trial, the train and the test vectors were selected randomly.
Four different feature set combination is generated from the descriptive statistics. Feature
Set1 is defined as including all the considered statistics. Set2 is defined as including only
6
dispersion measurements, whereas Set3 is consist of central tendency, average and
location measurements. Set4 is a combination of statistics from both two groups, which is
used in order to visualize the features in the three dimensional feature space (Figure 3).
Classification performance of these feature sets is presented in Table 1. As seen from this
table, best classification performance with a correct classification rate of about 94 % is
achieved for Set1.
Table 1 Classification accuracy values for four different feature sets
Set1
X
X
X
X
X
Set2
X
X
X
X
X
Set3
Set4
X
X
X
Quartiles
X
Accuracy
93.95 %
86.28 %
X
X
Median
IQR
Kurt
Skew
Var
Std
Feature
set
Mean
Measures of central
tendency, average,
and location
Measures of
dispersion
X
X
X
85.64 %
90.44 %
Using only one of the measurement groups (Set2 or Set3) reduces the correct
classification rate. Classification accuracies are 86.28% and 85.64% for feature Set2 and
Set3, respectively. It is evident from the accuracy rates of Set1 and Set4 that classification
accuracy increases as the measurements from both groups are employed together.
Classification performances of feature Set1 and Set4 were also visualized through
confusion matrices in Table 2 and Table 3. From these tables correct classification rates
for all transmitters as well as misclassification rates can be seen. For example, for feature
Set1, correct classification rate for the test signals from Tx1 is 96.3% when it is
incorrectly labeled as Tx2 and Tx5 with a rate of 3.5% and 0.2%, respectively. It can be
deduced by comparing these tables that total classification performance of feature Set1 is
better than that of feature Set4, even though Set4 is more distinctive for a specific
transmitter, e.g. Tx3. The separability of feature Set4 can also be visually inspected from
7
three dimensional feature space in Figure 3. This figure shows that test signals from Tx6
can be easily separated from other transmitters while test signals from Tx5 can be
incorrectly labeled as Tx2 since features from these two transmitters overlap in the some
regions of feature space. This result is consisted with the results in Table 3.
Table 2 Confusion matrix for feature Set1
Actual
class
Predicted class
Tx1
Tx2
Tx3
Tx4
Tx5
Tx6
Tx1
96.3%
3.5%
0%
0%
0.2%
0%
Tx2
0.1%
97.2%
0%
0%
2.7%
0%
Tx3
0%
0%
88.2%
11.8%
0%
0%
Tx4
0%
0%
10.1%
89.9%
0%
0%
Tx5
1.7%
4.0%
0%
0%
93.4%
0.9%
Tx6
0%
0%
0%
0%
1.3%
98.7%
Table 3 Confusion matrix for feature Set4
Actual
class
Predicted class
Tx1
Tx2
Tx3
Tx4
Tx5
Tx6
Tx1
92.7%
7.3%
0%
0%
0%
0%
Tx2
0.7%
98.0%
0%
0%
1.3%
0%
Tx3
0%
0%
99.8%
0.2%
0%
0%
Tx4
0%
0%
10.7%
89.3%
0%
0%
Tx5
7.7%
24.1%
0%
0.7%
67.5%
0%
Tx6
0%
0%
3.0%
1.5%
0%
95.5%
8
Figure 4 shows mean and standard deviation of classification accuracy for 100
independent Monte Carlo trials. In each trial ten training vectors were randomly selected
from fifty transmissions and remaining forty transmissions were employed as test vectors.
Mean values of classification accuracy are represented as circles inside the boxes as the
length of the boxes represent standard deviation of classification accuracy. As seen from
this figure, best classification performance is achieved with Set1. Mean of classification
accuracy for Set1 is higher than those of
Figure 3 Visualization of feature Set4 for all measurements
9
Figure 4 Classification accuracy for four different feature sets.
other sets, as well as minimum standard deviation is obtained with Set1. From the same
figure it is evident that a slightly better classification performance is achieved with the
statistics of dispersion compared to the statistics of central tendency in terms of both
mean and standard deviation of classification accuracy. This figure also shows that Set4
has a better classification performance in terms of mean accuracy values compared to
Set2 and Set3, however standard deviations of classification accuracy for these three sets
are close to each other.
5. CONCLUSIONS
Features based on descriptive statistics are extracted from the transient part of
transmissions from WiFi devices in order to classify these devices. Experimental results
show that descriptive statistics can be used as features for classification. Experimental test
results demonstrated that combined use of descriptive statistics, which measure the
central tendency and dispersion of data, increases the classification performance.
ÖZET: Fiziksel tabaka tanımlama sistemleri, iletim sinyallerinden ölçülebilen ayırt edici cihaz
karakteristiklerini kullanarak kablosuz cihazları sınıflandırırlar. Bu makalede betimleyici
istatistiklere dayalı öznitelikler sınıflandırma amacıyla WiFi sinyallerinin geçici rejim kısımlarından
elde edilmektedir. Sınıflandırma baĢarımı deneysel verilerle değerlendirilmektedir. Betimleyici
istatistiklerin farklı kombinasyonlarından oluĢan öznitelik kümeleri kullanılarak merkezi eğilimi ve
dağılımı ölçen istatistiklerin ayırt ediciliği değerlendirilmektedir. Bağımsız deneyler için
sınıflandırma sonuçları, hata matrisinin yanı sıra sınıflandırma baĢarımının ortalama ve standart
sapması üzerinden verilmektedir.
REFERENCES
[1] O. Ureten and N. Serinken, Wireless security through RF fingerprinting, Canadian
Journal of Electrical and Computer Engineering, vol. 32, (2007), p.27-33.
[2] R.W. Klein, M.A. Temple, M.J. Mendenhall, Application of wavelet-based RF
fingerprinting to enhance wireless network security, Journal of Communications and
Networks, vol. 11, (2009), p.544-555.
[3] M.D. Williams, S.A. Munns, M.A. Temple, M.J. Mendenhall, RF-DNA fingerprinting
for airport WiMax communications security, 4th International Conference on Network
and System Security (NSS), (2010), p.32-39.
10
[4] D.R. Reising, M.A. Temple, M.J. Mendenhall, Improving intra-cellular security using
air monitoring with RF fingerprints, Wireless Communications and Networking
Conference ,(2010), p.1-6.
[5] P. Harmer, M.A. Temple, M. Buckner, E. Farquhar, 4G security using physical layer
RF-DNA with de-optimized LFS classification, Journal of Communications, vol. 6,
(2011), p.671-681.
[6] S.U. Rehman, K. Sowerby, C. Coghill, RF fingerprint extraction from the energy
envelope of an instantaneous transient signal, Australian Communications Theory
Workshop (AusCTW), (2012), p.90-95.
[7] B. Danev, D. Zanetti, S. Capkun, On physical-layer identification of wireless devices,
ACM Computing Surveys, vol. 45, (2012), p.1–29.
[8] D. Shaw and W. Kinsner, Multifractal modeling of radio transmitter transients for
clasification, Proceedings of the IEEE Conference on Communications, Power and
Computing, (1997), p.306-312.
[9] O. Ureten and N. Serinken, Detection of radio transmitter turn-on transients,
Electronics Letters, vol. 35, (1999), p.1996-1997.
[10] O. Ureten and N. Serinken, Bayesian detection of Wi-Fi transmitter RF fingerprints,
[11] C. Zhao, X. Wu, L. Huang, Y. Yao, Y.C. Chang , Compressed sensing based
fingerprint identification for wireless transmitters, The Scientific World Journal, vol.
2014, (2014), p.1-9.
[12] R.G. Lyons, Understanding Digital Signal Processing, (Prentice Hall, New Jersey,
2004), p.688.
[13] S. Bernstein, R. Bernstein, Schaum's Outline of Theory and Problems of Elements of
Statistics I: Descriptive Statistics and Probability, (The McGraw-Hill Companies, Inc.
1999), p.354.
DOI: 10.1501/commua1-2_0000000084
ISSN 1303-6009
SIGNAL-TO-NOISE RATIO ESTIMATION OF NOISY TRANSIENT SIGNALS
MEMDUH KÖSE 1, SELÇUK TAġCIOĞLU 2, ZĠYA TELATAR 2
1
Ahi Evran Univ, Computer Sciences Research and Application Center Kırşehir, Turkey
2
Ankara Univ, Faculty of Engineering, Electrical and Electronics Engineering Department
Ankara, Turkey
E-mail: memduh.kose@ahievran.edu.tr, {selcuk.tascioglu, ziya.telatar}@eng.ankara.edu.tr
ABSTRACT
In this paper, the problem of signal-to-noise ratio (SNR) estimation of noisy transient signals is
considered. The SNR estimation is performed based on the average energy of the transient and
noise signals. The performance of the SNR estimator is evaluated in terms of normalized mean
square error and normalized bias. It is shown by simulations that the SNR estimation method
exhibits good estimation performance for the SNRs encountered in many practical applications.
Besides, the effect of the estimation of the transient starting point on the performance of the SNR
estimation is investigated
KEYWORDS: SNR estimation, transient signal detection, normalized bias, normalized
mean square error
1. INTRODUCTION
The knowledge of signal-to-noise ratio enhances the performance of various wireless
communication systems, which motivates the development of efficient SNR estimation
techniques. Channel quality is measured using SNR estimation for many operations in
wireless communication systems such as data rate adaptation, mobile assisted handoff,
power control, and decoding. In wireless OFDM systems, various methods have been
proposed for providing power and bandwidth efficiency by adapting the transmission
parameters with respect to channel conditions in terms of SNR. For example, estimates of
SNR were obtained for this purpose by utilizing the periodic structure of the OFDM
preamble in [1]. Balachandran et al. proposed an adaptation method, which switches
between the different modulation schemes depending on the SNR, in order to improve
data throughput over fading channels [2]. In cognitive radio networks, SNR has been used
as one of the parameters which can be employed by secondary users when sensing the
12
spectrum occupancy for the aim of opportunistic access to unused frequency bands
allocated to primary users [3]. SNR is also required for iterative decoding algorithms.
Summers and Wilson addressed the effect of SNR estimation error on the performance of
iterative turbo code decoder [4]. In [5], SNR was estimated based on Kalman filter to
monitor the link quality in wireless sensor networks.
The SNR estimation approaches can be considered in two categories: data-aided and non
data-aided. Wiesel et al. addressed both data-aided and non data-aided SNR estimators
for phase-shift keying communication systems in time-varying fading channels [6]. A
maximum likelihood based non-data aided algorithm was proposed for the SNR
estimation purposes in the presence of constant deterministic interference [7]. For the
applications where the bandwidth efficiency is an important parameter, non data-aided
estimators may be preferred since throughput is reduced for data-aided estimators by
inserting known symbols into the data stream [8]. For a detailed comparison of several
SNR estimation methods for the additive white Gaussian noise channel, see [9].
When a wireless transceiver starts to transmit, carrier signal shows a transient behavior
[10]. Transient signals are generated at the beginning of transmission and convey some
characteristic information identifying the transceiver due to hardware imperfections of
physical layer components. Transient signal analysis has been employed for many signal
processing and communication applications. For example, in [10]-[11] some distinctive
features based on detected transients have been extracted to classify the wireless devices.
It was reported that increased accuracy of the transient signal detector increases the ability
to obtain whole characteristic information [10].
SNR estimate of transient signals was employed to improve the performance of a wireless
device identification system for noisy channels in [11]. This method was based on
injection of noise into the neural network training set. The amount of noise to be added to
the training set was determined by the SNR estimate of test transient. Thus, SNR
estimation accuracy of an unknown transient has become an important issue for device
identification systems based on transient analysis. In this paper, we investigate a non dataaided SNR estimation technique for noisy transients based on average energy ratio of
transient and noise signals as well as the effect of transient detection on the performance
of the SNR estimation.
2. SNR ESTIMATION BASED ON AVERAGE ENERGY
The discrete time transient signal at the receiver can be given as
SIGNAL-TO-NOISE RATIO ESTIMATION
y (k )
13
(1)
s ( k ) n( k )
where s ( k ) and n( k ) denotes transient and noise signals, respectively. Let average
energy of noisy transient be
ESN
1
K
K
y2 k
(2)
k 1
where K represents the analysis window length. By using the fact that the noise is
independent of the signal, Eq. (2) can be written in terms of signal and noise terms as
ESN
1
K
K
s2 k
k 1
1
K
K
n2 k
(3)
k 1
since the statistical average of s (k )n(k ) equals to zero for the independent case [12].
The SNR for transient signal is defined as the ratio of the average energy of the transient
signal to the average energy of noise signal. The SNR estimation in dB is calculated as
ˆ 10log10
ESN
EN
1
(4)
where EN stand for the average energy of the noise segment. EN can be calculated
during no transmission. This approach has been used for SNR estimation of wireless
communication signals [11] as well as of speech signals [13]. In [13], SNR estimation
was performed using the noise power obtained from the non-speech segments which were
detected by using different voice activity detectors. In the same work, the effect of these
detectors on the performance of the SNR estimation methods were investigated.
3. TRANSIENT SIGNAL DETECTION
Detection of transients for wireless devices is a difficult problem, since this task requires
estimating the starting point of noise-like behavior transient signals. One of the methods
for transient detection is to formulate the problem as a change-point detection problem
[14]-[15]. In [15], a method based on a ramp signal model was proposed to detect the
starting point of transients in a Bayesian framework. The performance of this method was
compared to a previous approach in [14], which was based on a piecewise constant signal
14
model containing a step change. It was experimentally verified that the transient detection
performance of the ramp detector is better than that of the abrupt change detector for turnon transients of Wi-Fi transmitters. It was also reported that the method can also be used
for the transient detection of the Bluetooth devices. In this paper, we employ both of these
Bayesian approaches to detect noise corrupted transients at different SNR levels.
4. PERFORMANCE METRICS
Mean Square Error (MSE) is a widely used metric to evaluate the performance of the
SNR estimator since the variance and bias of the estimator are both included in the MSE
[7]-[9]. MSE of the SNR estimator is defined as
MSE ˆ
1
T
T
ˆt
2
(5)
t 1
where T denotes the number of independent Monte Carlo trials. In Eq. (5) the error terms
can be normalized with respect to true SNR in which case the quantity is called
normalized MSE (NMSE) and given by
NMSE ˆ
1
T
T
ˆt
2
(6)
t 1
NMSE demonstrates the asymptotic behavior of the estimator as the SNR increases [9].
The results are also given in terms of normalized bias which is defined as
NB ˆt
1
T
T
ˆt
(7)
t 1
5. SIMULATION RESULTS
IEEE 802.11b standard defines the transmit power-on ramp by a smoothly increasing
function in order to avoid adjacent channel interference [16]. Considering this definition,
transient signals were simulated using four different signal models: the rising parts of the
cosine, raised cosine, Gaussian and exponential functions. These functions are shown in
Figure 1. The start of the transient was chosen to be sample 1000 following channel noise
with a length of 1000 samples. Therefore total analysis signal window was consisted of
2000 samples. Simulated Gaussian noise was added to transient signal and the SNR level
15
of the detected noisy transient was estimated. 500 Monte Carlo simulations were
performed for each signal model.
Figure 1. Simulated transient signals based on four different models
Figure 2. Simulated noisy transient signal generated from the rising part of the cosine at
the SNR value of 25 dB
16
Normalized bias and normalized MSE of the SNR estimator are plotted in Figure 3 and
Figure 4, respectively. In these figures, estimated SNR values are given for the transients
which are detected by using two different transient detection methods. Figure 3 shows
that the SNR estimator is biased and overestimates SNR for both cases. Also, it is
observed from this figure that as the SNR increases, normalized bias values decreases.
When the transient is detected using Bayesian ramp detector, normalized bias of the SNR
estimator is small and does not change significantly as
Figure 3. Normalized bias of the SNR estimator
Figure 4. Normalized MSE of the SNR estimator
17
the SNR changes. On the other hand for the Bayesian step detector, the bias of the SNR
estimator increases as the SNR decreases. Figure 4 demonstrates that the normalized MSE
of the SNR estimator decreases with increasing SNR for both cases.
The effect of transient detection accuracy on the performance of the SNR estimator can
be observed from Figure 3 and Figure 4. When the transient is detected using ramp
detector, a better SNR estimation performance is achieved over the entire range of SNR
values. The performance improvement of the ramp detector increases with decreasing
SNR. It is evident from these two figures that the SNR estimator in Eq. (4) can be used
for SNR estimation with a reasonable error as long as the transient detection is performed
accurately.
It was reported in [11] that the SNR of the transient signals is expected to vary in the
range of 10-15 dB for many practical applications. It is evident from Figure 4 and Figure
5 that the estimation performance is good for these SNR values as the transients are
detected by using a ramp detector. The histograms of SNR estimation error are given in
Figure 5 for two transient detection methods at the actual SNR level of 10 dB. The mean
values of SNR estimation error were found to be 1.0 dB and 3.3 dB for ramp and step
detectors, respectively. These results are consistent with the results in Figure 3.
Figure 5. Error histogram of the SNR estimator at the actual SNR of 10 dB for two
different transient detection method
18
6. CONCLUSIONS
In this paper, SNR estimation of noisy transient signals is investigated. In order to
evaluate the estimation performance, normalized MSE and normalized bias metrics are
defined. Simulation results show that the performance of the SNR estimator is good for
the SNRs encountered in many practical applications. In addition, the SNR estimator can
be used with a reasonable error for lower SNR levels from 2 dB to 10 dB as long as the
transient detection performance is high.
ÖZET: Bu makalede gürültülü geçici rejim iĢaretlerinin sinyal-gürültü oranı kestirimi problemi ele
alınmaktadır. SNR kestirimi, geçici rejim ve gürültü sinyallerinin ortalama enerji değerlerine dayalı
olarak hesaplanmaktadır. SNR kestirici baĢarımı, düzgelenmiĢ hata kareleri ortalaması ve
düzgelenmiĢ yanlılık değerleri üzerinden değerlendirilmektedir. Ġncelenen kestiricinin çoğu pratik
uygulamada karĢılaĢılan SNR değerleri için iyi baĢarıma sahip olduğu benzetim sonuçlarıyla
gösterilmektedir. Ayrıca, geçici rejim sinyali baĢlangıç noktası kestiriminin SNR kestirim baĢarımı
üzerindeki etkisi de incelenmektedir.
REFERENCES
[1] M. Zivkovic and R. Mathar, Preamble-based SNR estimation in frequency selective
channels for wireless OFDM systems, In Proceeding of IEEE Vehicular Technology
Conference, (2009), p.1-5.
[2] K. Balachandran, S. R. Kadaba, S. Nanda, Channel quality estimation and rate
adaptation for cellular mobile radio, vol. 17, (1999), p.1244-1256.
[3] J. Vartiainen, H. Saarnisaari, J.J. Lehtomaki, M. Juntti, A Blind Signal Localization
and SNR Estimation Method, In Proc. IEEE Military Communications Conference
(MILCOM), (2006), p.1-7.
[4] T.A. Summers and S.G. Wilson, SNR mismatch and online estimation in turbo
decoding, IEEE Transactions on Communications, vol. 46, (1998), p.421-423.
[5] F. Qin, X. Dai, J.E. Mitchell, Effective-SNR estimation for wireless sensor network
using Kalman filter, Ad Hoc Networks, vol. 11, (2013), p.944-958.
19
[6] A. Wiesel, J. Goldberg, H. Messer-Yaron, SNR estimation in time-varying fading
channels, IEEE Transactions on Communications, vol. 54, (2006), p.841-848.
[7] F. Chen, Y. Kang, H. Yu, F. Ji, Non-data-aided ML SNR estimation for AWGN
channels with deterministic interference, EURASIP Journal on Wireless Communications
and Networking, (2014).
[8] C. Gong, B. Zhang, A. Liu, D. Guo, A Highly accurate and low bias SNR estimator:
Algorithm and implementation , Radioengineering, Proceedings of Czech and Slovak
Technical Universities and URSI Committees, vol. 20, (2011), p.976-981.
[9] D.R. Pauluzzi and N.C. Beaulieu, A comparison of SNR estimation techniques for the
AWGN channel, IEEE Transactions on Communications, vol. 48, (2000), p.1681-1691.
[10] O. Ureten and N. Serinken, Wireless security through RF fingerprinting, Canadian
Journal of Electrical and Computer Engineering, vol. 32, (2007), p.27-33.
[11] O.H. Tekbas, O. Ureten, N. Serinken, Improvement of transmitter identification
system for low SNR transients, Electronics Letters, vol. 40, (2004), p.182-183.
[12] A.B. Carlson and P.B. Crilly, Communication Systems: An Introduction to Signals
and Noise in Electrical Communication (McGraw –Hill, New York, 2010) p.924.
[13] M. Vondrasek and P. Pollak, Methods for speech SNR estimation: Evaluation tool
and analysis of VAD dependency, Radioengineering, Proceedings of Czech and Slovak
Technical Universities and URSI Committees, vol. 14, (2005), p.6-11.
[14] O. Ureten and N. Serinken, Detection of radio transmitter turn-on transients,
[15] O. Ureten and N. Serinken, Bayesian detection of Wi-Fi transmitter RF fingerprints,
[16] IEEE Std 802.11b-1999, Part 11, Wireless LAN medium access control (MAC) and
physical layer (PHY) specifications: Higher-speed physical layer extension in the 2.4
GHz band, (IEEE Inc., New York, 1999), p.90.
20
Commun.Fac.Sci.Univ.Ank.Series A2-A3
DOI: 10.1501/commua1-2_0000000085
ISSN 1303-6009
VISIBILITY BASED OPTICAL WIRELESS AVAILABILITY
ASSESSMENT FOR CONTINENTAL CLIMATE CONDITIONS
Filiz SARI1 and Faruk OZEK2
1
Aksaray University, Faculty of Engineering, Electrical and Electronics Eng. Dept.,
68100, Aksaray, Turkey E mail: filizsari@aksaray.edu.tr
2
E-mail:ozek@eng.ankara.edu.tr
(Received: October 04, 2014; Accepted: June 10, 2015)
ABSTRACT
Potential application of the cumulative distribution function of the airport visibility data,
a statistical technique suggested and tested for various regions in Europe, is verified for
optical wireless availability and range assessment in the city of Ankara, the regional center of
continental climate mid – Anatolia. Results, obtained by computations with reference to the
system parameters of the enterprise class Ankara University optical wireless link have been
presented.
KEYWORDS: Airport visibility data, Availability, Cumulative Distribution
Function, Link range, Optical Wireless Communication
1
INTRODUCTION
Optical wireless communication, OWC, is identified to offer the well – known
advantages over radio waves [1, 2], such as the transmission of higher bandwidth
data rates for distances up to about 4 km. However, OWC links use the air as the
transmission medium where the adverse weather conditions cause shorter visibility
occurrences. Reduced visibility, on the other hand, means increased laser signal
power loss. Consequently, historical airport visibility data can be utilized to assess
the link availability and range by power loss evaluation.
A literature survey has indicated that the link availability assessment is possible
by the statistical analysis of the airport recorded visibility values, alone [3]. The
suggested statistical analysis is based on the cumulative distribution function (CDF)
of the visibility and has yielded positive results in order to estimate the OWC
availability for the mid-north and especially the Mediterranean regions of Europe.
The main objective of the present study is to verify the applicability of the CDF
approach to the airport visibility data recorded in the mid-Anatolian city of Ankara,
the Asian part of Turkey, where the typical continental climate prevails. In the
computations, the technical parameters of the enterprise class Ankara University
OWC system are considered [4, 5] to confirm the statistical model.
22
Filiz SARI and Faruk OZEK
2
POWER LOSSES
As summarized in Figure 1, the laser power losses can be divided mainly into
three different groups: (a) system loss, (b) geometric loss, and (c) atmospheric
attenuation.
The system loss, loss(system) is the sum of pointing error, loss(pnt) and optical
losses, loss(opt), and is constant for a given OW link. The optical loss is due to
power decrease at the lenses and optical filters.
Figure 1 Schematic summary of power losses due to degrading factors in a general
optical wireless link
The geometric loss, loss(geo), is a consequence of the laser beam spread and can
be computed using Eq. (1), [6].
loss geo =S beam S rec =20log R
D
(dB)
(1)
where S(beam): beam cross-section area at range R (m2), S(rec): receiver lens area
(m2), θ: beam divergence (mrad), D: receiver lens diameter (m).
Atmospheric effects: Molecular absorption is negligible, because the generally
used laser wavelengths 785, 850 and 1550 nm coincide with the atmospheric
transmission windows [7]. The effect of the scintillation due to turbulence is almost
constant, for ranges up to 4 km [6]. The scattering (Mie) of the laser rays by fog
droplets is therefore the dominant atmospheric effect and is quantified by the
attenuation coefficient, σ, Eq. (2), [8].
23
VISIBILITY BASED ON OPTICAL WIRELESS
q
= 3.91
(km-1)
V
(2)
550nm
where V: visibility (km), λ laser wavelength (nm), q is the particle size distribution
coefficient given in [8] and varies with V,
1,3
for 6,0 < V < 50,0 km
0,16.V+0,34 for 1,0 < V < 6,0 km
q=
V-0,5
for 0,5 < V < 1,0 km
0
for
V < 0,5 km
Atmospheric attenuation in terms of dBloss/ km, Eq. (4), can be derived from
Eq. (3) for R = 1 km [1 – 3].
P rec =P0 .e-
R
(3)
where P(rec): power received at distance R, P 0:laser transmit power.
Attenuation:dB/km
= 4,343 .
(4)
dBloss/km data, for λ=1550 nm laser wavelength, in Table 1 are computed
sequentially from Eq. (2) and (4). The visibility limits corresponding to various
weather conditions are adapted from International Visibility Code [9].
Table 1 Power Loss as a Function of Visibility
WEATHER
CONDITIONS
DENSE FOG
VISIBILITY
dBloss/km
(λ=1550 nm)
0 m
50 m
340,0
200 m
84,9
500 m
33,0
770 m
17,9
THICK FOG
MODERATE
FOG
LIGHT
FOG
24
1,0 km
10,2
1,9 km
2,0 km
4,6
4,3
2,8 km
2,7
4,0 km
1,5
5,9 km
0,8
10,0 km
0,4
THIN
FOG
HAZE
LIGHT
HAZE
CLEAR
It is clear that the visibility data suffice to determine the value of σ and
dBloss/km.
3
VISIBILITY DATA
The visibility data, recorded over three years, at Ankara Etimesgut airport [10]
are based on the 2% transmission (17 dB) criterion for visibility with reference to
METAR [11]. Attenuation is then expressed as in Eq. (5).
Attenuation= 17
4
q
V
550nm
(dB/km)
(5)
LM, AVAILABILITY
The link margin, LM, is the power remaining at a distance R, to counter the
atmospheric attenuation, Eq. (6)
LM=[P0 – P(sen) – loss(constant)] – loss(geo) (dB)
(6)
where P(sen):receiver detector sensitivity,
loss (constant) =loss(pnt)+loss(opt)+loss(scn)
Consequently, the link will be available provided that LM is greater than the
atmospheric attenuation as indicated Eq. (7) which is arranged from Eq.(5) and (6).
V
17.R
LM
q
550nm
(km)
(7)
25
From the preceding, the availability can be expressed as a function of the
minimum required visibility, Vmin, as defined in Eq. (8).
Availability=Probability LM loss atm
=Probability V
1 F Vmin R
Vmin R
(8)
where F is the cumulative distribution function, CDF, obtained from the probability
of airport visibility data, PDF, [3].
5
RESULTS
5.1
SYSTEM PARAMETERS
For all the computations, starting from LM, availability, then the range
assessment, the parameters of an existing system, namely 2,9 km enterprise class
Ankara University OWC link are used [4, 5] , Table 2.
Table 2 Ankara University OWC system parameters
Transmitter
P0
7 - 640 mW
9 - 28 dBm
,
1550 nm, 2,8 mrad
Receiver
P(sen)
D, lens
loss(pnt)
loss(opt)
loss(scn)
loss(constant)
- 36 dBm (PIN)
0,2 m
Losses
3 dB
3 dB
5 dB
11 dB
It should be noted that the transmitter laser power is adaptive from 9 dBm for
clear to 28 dBm for unfavorable weather conditions, respectively.
26
From Eq. (6), LM=f(R) function is plotted in Fig. 2, for weather condition
dependent limits of the variable transmit laser power, P0. Figure 3 is the attenuation,
in terms of dB/km as derived from the data of Figure 2. The combination of the
Figure 3 data with that of Table 1 is the relationship between the visibility V and
link range, R, Figure 4.
70
Po = 28 dBm
60
Po = 9 dBm
LM, dB
50
40
30
20
10
0
0
0.5
1
1.5
R, km
2
2.5
3
Figure 2 Variation of link margin with range for minimum and maximum laser
transmit power: Eq. (6).
50
Po = 28 dBm
Po = 9 dBm
Attenuation, dB/km
40
30
20
10
0
0
0.5
1
1.5
R, km
Figure 3 Variations of attenuation with link range
2
2.5
3
27
8
2.9 km
Po = 28 dBm
7
Po = 9 dBm
6
V, km
5
LIGHT HAZE
4
3
HAZE
2
THIN FOG
1
0
LIGHT FOG
THICK FOG
0
0.5
1
1.5
R, km
2
2.5
3
Figure 4 Variations of visibility with range for different weather conditions, from
data of Figure 3 and Table 1
5.2
CUMULATIVE DISTRIBUTION FUNCTION
An initial analysis of the airport visibility data indicated that over the months
from April to October inclusively, mostly summery, no link unavailability is
expected. However, the remaining months, November to March, are critical from the
point of high link cut – off likelihood.
CDF vs. visibility, Figure 5, is drawn through the following steps: (a) From the
number of observations for each V, probability density function PDF data, (b) then
added to give cumulative distribution function CDF data, (c) the resulting data are
fitted approximately to a third – order polynomial [3], the solid line in Figure 5.
28
0
10
-1
PDF, CDF
10
-2
10
-3
10
PDF data
CDF data
CDF fitted
-4
10
0
1
2
3
4
5
6
7
8
9
10
V, km
Figure 5 PDF, CDF and approximated CDF vs. visibility for critical months
Equation (8) is the basis for the availability computation, at that V is converted
to R via Figure 4 data. Considering the regional critical months, November to
March:151 days in total, and the Ankara University OWC system parameters, Figure
6 displays the variation of the availability with the link range.
Figure 6 Variations of availability with range for the critical months
6
29
CONCLUSION
In this work, the initial elimination of the favourable months is based on the
threshold criterion of visibility which is 6 km or greater, i.e. dBloss/km = 0,8, Table
1. For an OWC link of R=2,9 km, the total power loss is 2,3 dB or less, and is
therefore negligible. Apart from shortening the airport visibility data treatment, the
consideration of the unfavourable months alone, November to March, inclusively,
means a more realistic availability assessment.
The CDF based Figure 6 indicates that for the range R=2,9 km, the estimated
availability is 96,7% which corresponds to downtime of 4,98 days, taken over the
critical months, 151 days. For the enterprise systems, on the other hand, the
availability requirement is 99%, [9], the downtime is therefore 3,65 days which is
approximately and practically equal to the estimated value, so as to confirm the
possible application of the CDF approach to the geographical region of interest.
It has been decided, finally, that considering the positive findings reported in the
literature for the Mediterranean region of Europe [3] and experimental evaluation of
the Ankara University OW system [4, 5], the results of the present study, the CDF
approach can be used in the construction of a countrywide optical wireless
availability map for Turkey, similar to that published such as Brazil [12].
REFERENCES
[1] F. Ghassemlooy., W. Popoola, S. Rajbhandari. Optical Wireless
Communications: System and Channel Modeling with Matlab. CRC Press, New
York, 2012.
[2] A. K. Majumdar and J. C. Ricklin. Free – space laser communications:
Principles and advances. Springer, New York, 2008.
[3] A. Proakes and V. Skorpil. Estimation of free space optics systems availability
based on meteorological visibility. IEEE LATINCOM’09, Colombia, 2009, pp.
1-4.
[4] A.Akbulut, M. Efe, A. M. Ceylan, F. Ari, Z. Telatar, H. G. Ilk. An experimental
hybrid FSO/RF communication systems. In Proc. Of the IASTED International
Conference on Communication System and Networks, CSN 2003,pp. 406 – 411,
2003.
[5] Akbulut, H. G. Ilk, F. Ari. Design, availability and reliability analysis on an
experimental outdoor FSO/RF communication system. Transparent Optical
Networks, 2005, Proceedings of 2005 7th International Conference , 1, pp.403
- 406 , 2005.
[6] I. I. Kim, R. Stieger, J. A. Koontz, C. Moursund, M. Barclay, P. Adhikari, J.
Schuster, E. Korevaar, R. Ruigrok, C. DeCusatis. Wireless optical transmission
of fast Ethernet, FDDI, ATM, and ESCON protocol data using the TerraLink
laser communication system. Optical Engineering, 17, 12(1998), pp. 3143-3155.
30
[7] H. Willebrand and B. S. Ghuman. Free-space optics: Enabling optical
connectivity in today’s networks. Sums Publishing, Indianapolis, 2001.
[8] I. I. Kim, B. McArthur, E. Korevaar. Comparison of laser beam propagation at
785 nm and1550 nm in fog and haze for optical wireless communications. Proc.
SPIE, 4214, 26(2001), pp. 26-37.
[9] I. I. Kim and E. Korevaar. Availability of free space optics (FSO) and hybrid
FSO/RF systems. Proc. SPIE, 4530, 84(2001), pp. 84-95
[10] Weather Underground [online] www.wunderground.com
[11] Office of the Federal Coordinator for Meteorological Services and Supporting
Research. Federal Meteorological Handbook No. 1: Surface Weather
Observations and Reports. FCM-H1-2005, Washington, D.C., September 2005.
[12] P. B. Harboe and J. R. Souza. Free space optical communication systems: A
feasibility study for deployment in Brazil. Journal of Microwaves and
Optoelectronics, 3, 4(2004), pp. 58 -66.
INSTRUCTIONS TO CONTRIBUTORS
Communications Faculty of Sciences University of Ankara Series A2-A3: Physics, Engineering Physics,
Electronics/Computer Engineering, Astronomy and Geophysics a peer reviewed journal which has been
published since 1948 by Ankara University, accepts original research articles written in English in the fields
of Physics, Engineering Physics, Electronics/Computer Engineering, Astronomy and Geology/Geophysics.
Review articles written by eminent scientists can also be invited by the Editor.
Manuscripts should be submitted as a single PDF file attached to an e-mail with a covering letter. In the
covering letter, authors should nominate three potential reviewers and e-mailed the file to the most
appropriate Area Editor of the research. The editorial office may not use these nominations, but this may
help to speed up the selection of appropriate reviewers.
Manuscripts should be typeset using the LATEX typesetting system. Authors should prepare the article using
the COMMUNICATIONS style before submission by e-mail. Manuscripts written in DOC form are also
acceptable.
A
template
of
manuscript
can
be
reviewed
in
http://communications.science.ankara.edu.tr/index.php?series=A2A3&link=300. After the acceptance of
manuscripts for publication, we will ask you to submit the TeX or Doc form of the manuscript prepared in
accordance with the style of the Journal.
Acknowledgements should be given as short as possible at the end of the text. Formulas should be numbered
consecutively in parentheses ( ). Footnotes should be avoided if possible, but when necessary, should be short
and never contain any important part of the work and should be numbered consecutively by superscripts. All
illustrations not including tables (photographs and other films, drawings, graphs, etc) must be labeled as
“Figure”.
The proper position of each table and figure must be clearly indicated in the paper.
All tables and figures must have a number (Table 1, Figure 1) and a caption or legend. References including
comments must be numbered consecutively in order of first appearance in the text. The reference number
should be put in brackets [] where referred to in the text. References should be listed at the end of the
manuscript in the numbered order in which they appear in the text as follows:
[1] Bairamov, E, Ozalp N. “Uniform convergence and numerical computation of the Hubbell radiation
rectangular source integral”, Radiation Physics and Chemistry, 80 (2011) 1312–1315.
[2] Kelley, J. L., General Topology, Van Nostrand, 1970.
It is a fundamental condition that articles submitted to COMMUNICATIONS have not been previously
published and will not be simultaneously submitted or published elsewhere. After the manuscript has been
accepted for publication, the author will not be permitted to make any new additions to the manuscript.
Before publication the galley proof is always sent to the author for correction. Thus it is solely the author's
responsibility for any typographical mistakes which occur in their article as it appears in the Journal. The
contents of the manuscript published in the COMMUNICATIONS are the sole responsibility of the authors.
The PDF copies of accepted papers are free of charges, but hard copies of the paper, if required, are due to
be charged for the amount of which is determined by the administration each year.
Editor in Chief
Commun. Fac. Sci. Univ. Ank. Ser. A2-A3.
Ankara University, Faculty of Sciences
06100, Besevler - ANKARA TURKEY
COMMUNICATIONS
VOLUME: 57
FACULTY OF SCIENCES
ANKARA UNIVERSITY
Number: 1
YEAR: 2015
Geophysics
M. KÖSE, S. TAŞCIOĞLU, Selçuk, Z. TELATAR, Wireless device identification
using descriptive statistics ………………………….................................................
1
M. KÖSE, S. TAŞCIOĞLU, Selçuk, Z. TELATAR, Signal-to-noise ratio
estimation of noisy transient signals ……………………………………………........ 11
F. SARI, F. OZEK, Visibility based optical wireless availability assessment for
continental climate conditions ..................................................................................... 21

Communications - Ankara Üniversitesi

Transcription

Similar documents

get protected! - Pine Cellular

en-open_belgian_championchip_2015

- Saurashtra University

PDF Document

SignalPro® - EDX Wireless

Baguio Apartelle and Transient House accommodtion

SCENERA RESEARCH LAUNCHES NEW WEBSITE

Gerringong Public School Weekly Bulletin

Security Safe Company

Comparison of SONY ILX511B and Hamamatsu