Académique Documents
Professionnel Documents
Culture Documents
PII: S1874-4907(16)30244-0
DOI: http://dx.doi.org/10.1016/j.phycom.2017.07.004
Reference: PHYCOM 406
Please cite this article as: K.D. Kanellopoulou, K.P. Peppas, P.T. Mathiopoulos, On the SINR
statistics of a VFDM cognitive spectrum sharing system, Physical Communication (2017),
http://dx.doi.org/10.1016/j.phycom.2017.07.004
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
On the SINR Statistics of a VFDM Cognitive Spectrum
Sharing System
K. Denia Kanellopouloua,, Kostas P. Peppasb , P. Takis Mathiopoulosa
a
Department of Informatics and Telecommunications, National and Kapodistrian
University of Athens, 15784 Athens, Greece
b
Department of Informatics and Telecommunications, University of Peloponnese, 22100
Tripolis, Greece
Abstract
An analytical precise approximation of the SINR statistics of the two-tier
Vandermonde-subspace frequency division multiplexing (VFDM) cognitive
spectrum sharing systems over frequency-selective Rayleigh fading channels
is presented. It is shown that the gamma distribution provides the best
fitting accuracy and that its use leads to simple, yet accurate, closed-form
expressions for evaluating the ergodic capacity (EC) and average bit error
probability (ABEP) performance of such systems. In deriving these expres-
sions, the parameters of the gamma distribution have been obtained for var-
ious operating conditions of the considered VFDM system using distribution
fitting. Furthermore, regression analysis has been used to obtain approxi-
mate analytical expressions for these parameters in relation to the system
operating parameters. Performance evaluation results, obtained for various
system implementations, including the standardized IEEE 802.11 and 3GPP
LTE, are presented to demonstrate the validity of the proposed methodology.
Its accuracy has been verified by means of computer simulations.
Keywords: Cognitive radio, Two-tier spectrum sharing,
Vandermonde-subspace frequency division multiplexing (VFDM), SINR
statistics approximation, Ergodic capacity, Average bit-error probability
(ABEP)
Corresponding author.
Email addresses: konkanel@gmail.com (K. Denia Kanellopoulou), peppas@uop.gr
(Kostas P. Peppas), mathio@di.uoa.gr (P. Takis Mathiopoulos)
2
as functions of the system operating parameters.
The rest of this paper is organized as follows. After this Introduction, the
system model under study is presented in Section 2. In Section 3, different
distributions are tested for the analytical approximation of the statistics of
the SINR. By using the approximation with the gamma distribution, novel
closed-form expressions for the EC and ABEP are derived. In Section 4,
analytical and computer simulation results are presented for the performance
evaluation of different implementations of the system under consideration.
Finally, in Section 5, the concluding remarks of the paper can be found.
Mathematical Notations: A lower case italic symbol denotes a scalar
value, a lower case bold symbol denotes a vector and an upper case bold
symbol denotes a matrix. IN is the identity matrix of size N . X denotes
the pseudo-inverse and XH the transpose conjugate of the matrix X. Ehi
denotes expectation and f (x; a, b) denotes the PDF with parameters a and
b. () denotes the gamma function, G m,n p,q [] is the Meijers G-function, Q()
is the Q-function and 2 F1 (a, b; c; z) is the Gaussian hypergeometric function.
All vectors are columns, unless otherwise stated.
2. System Model
Let us consider the typical case of a cognitive interference channel model,
which is presented in [6] and depicted in Fig. 1. In particular, a primary com-
munication system (Tx1 Rx1 ) transmits over a licensed frequency band and
a secondary (cognitive) system (Tx2 Rx2 ) opportunistically communicates
over the same frequency band, without creating interference to the primary.
It is assumed that all transmitters and receivers have a single antenna, the
secondary transmitter and receiver have perfect channel state information
(CSI) and there is no cooperation between primary and secondary system.
In this context, the detailed system model under consideration is depicted
in Fig. 2, where the primary and secondary systems are presented in Figs. 2a
and 2b, respectively. Between transmitter Txi and receiver Rxj , where
i and j {1,2}1 , the (L + 1)-tap FS channel, denoted as h(ij) , consists of
unit-norm, independent and identically distributed (i.i.d.), complex circularly
symmetric and Gaussian CN (0, IL+1 /(L + 1)) channel entries. Both primary
1
From now on and unless otherwise indicated, all subscripts using the symbols i and j
will be assumed to take values from this alphabet.
3
Fig. 1. Cognitive interference channel model
and secondary systems are also corrupted by additive white Gaussian noise
(AWGN) noise, denoted as ni CN (0, 2 IN ).
The primary system consists of a primary user with a transmitter Tx1 ,
which utilizes the Orthogonal Frequency Division Multiplexing (OFDM) with
N subcarriers and a cyclic prefix of size L. Thus, the transmitted symbols
can be expressed as
x1 = AF1 s1 ,
where A is an (N +L)N cyclic prefix precoding matrix that appends the last
L elements of F1 s1 and s1 is a symbol vector of size N with unitary norm.
F1 is the inverse Fast Fourier Transform (FFT) matrix. The secondary
system consists of a secondary user with a transmitter Tx2 that uses the
VFDM technique. More specifically, the secondary user exploits the resources
created by the frequency selectivity of the channel and the cyclic prefix used
by the primary in order to transmit symbols of the form
x2 = Vs2 ,
where s2 is a symbol vector of size L with unitary norm. The secondary user
can transmit simultaneously with the primary, over the same frequency band,
by implementing an (N + L) N precoder V that eliminates its interference
towards the primary receiver Rx1 .
Let Hij denote the overall channel matrices defined in [4] and [6] as
4
(a)
(b)
Fig. 2. Block diagram of the two-tier cognitive VFDM system model: (a) Primary system
(b) Secondary (cognitive) system.
5
H12 = FT (h(12) )AF1 ,
T (h(21) )V = 0.
Then, the received signal at the primary receiver Rx1 can be expressed as
y1 = H11 s1 + 1
y2 = H22 s2 + H12 s1 + 2 .
In the above equations, i is the FFT of the noise vector ni , while the factor
[0, 1] is introduced (see also Fig. 2) in order to control the amount of
interference, H12 s1 , from the primary to the secondary system so that its
effects on the latters performance can be assessed.2
As the primary systems operation is not affected by the presence of the
secondary, it can be considered as a typical OFDM system, which has been
extensively studied in the past (e.g. see [8], [4], [6]). Therefore, the focus of
this paper will be on the performance of the secondary system. By including
2
For an overview of techniques that can be used to control this cross-interference the
reader may refer to [9].
6
, the SINR at the secondary receiver can be rewritten in a more general
form, as compared to [6, Eq. (11)], as
|gk h2k |2
k = , (1)
P
N P
L
2 |gk gkH |2 + 2 |2
|gk h1m + |gk h2n |2
m=1 n=1;n6=k
s02 = Gy2 .
In this work, the zero-forcing (ZF) and the minimum mean square error
(MMSE) equalization filters have been applied at the secondary receiver.
Thus G, denoted hereafter as GZF and GM M SE for the ZF and MMSE
equalization, respectively, is given by [6, Eq. (13)]
GZF = H22
3. Performance Analysis
Unfortunately for the considered system, the expression of the SINR in
(1) cannot be used to obtain analytical results. Furthermore, its evalua-
tion by means of computer simulations is a time-consuming task. Thus, an
alternative approach is proposed in this Section, which is based on an analyt-
ical accurate approximation. In particular, in Subsection 3.1 the analytical
approximation of the SINR statistics will be investigated. Then, in Sub-
section 3.2, the proposed approximation will be used to obtain closed-form
expressions for the EC and ABEP. Finally, in Subsection 3.3 the parameters
of the approximating distribution will be estimated in terms of the system
operating parameters.
7
3.1. SINR Statistics Approximation
The accurate approximation of the statistics of (1) with an appropriate
distribution can provide mathematically tractable formulas for evaluating
the system performance and thus facilitate further analysis. For this reason,
we have considered various Probability Density Functions (PDFs) of well-
known distributions that have been used in the past for channel modeling,
such as the gamma, Weibull, Nakagami, Rice, exponential and lognormal.
Our purpose was to first obtain values of the SINR for different inputs and
then use the PDFs of known distributions to fit the PDF of the results. With
this in mind, we have first considered the system model described in Section 2
and have run extensive Monte Carlo simulations in Matlab in order to obtain
the SINR. For these simulations we have generated random Rayleigh channel
vectors and constructed the channel matrices and the precoder, as described
in Section 2. This procedure was repeated for various values of the system
operating parameters N and L, as well as for different operating conditions,
set by the input SNR and the cross-interference factor , for both receiver
structures employing ZF and MMSE equalizers. In this way, the SINR in
(1) was evaluated and its statistics were obtained. Then, the maximum
likelihood estimation (MLE) method [10] was used to match these results to
the statistics of these well-known distributions that are built-in in Matlab.
In order to evaluate their goodness-of-fit, the Kolmogorov-Smirnov (KS)
statistical test [11] has been employed, measuring the maximum values of the
absolute difference between the Cumulative Distribution Function (CDF) of
the simulated SINR and the CDF of the distributions that were fitted to
it. More specifically, the null hypothesis under which the observed data be-
long to the CDF of the approximating distribution at the 5% significance
level has been tested, resulting in 1 if the test is rejected and in 0 otherwise.
Typical comparative results can be found in Table 1, where the percentage
of the cases in which the KS goodness-of-fit statistical test succeeded is pre-
sented for each of the distributions tested, under the different combinations of
the simulation input parameters. The results have indicated that the gamma
distribution has provided the most accurate fitting for a wide range of the op-
erating conditions of the considered VFDM system, followed by the Weibull
distribution. However, as it will be shown in Subsection 3.2, the Weibull
distribution leads to very complex expressions for the evaluation of the EC
and ABEP performance of the considered VFDM cognitive spectrum sharing
system. On the contrary, the gamma distribution leads to simple closed-form
expressions, which can be very conveniently used for performance analysis
8
purposes.
Table 1. Percentage of cases the KS goodness-of-fit statistical test succeeded for the
considered distributions.
9
Another important performance metric is the ABEP, which can be ob-
tained for QPSK modulated symbols by
L
1X
Pe = E hQ( k )i .
L k=1
which can be obtained in closed-form for the gamma distribution with the
help of [15, Eq. (3.7.2.4)] as
r
e CF 1 (a + 21 ) b 1 1 3 b
Pe = 2 F1 ,a + ; ; . (5)
2 (a) 2 2 2 2 2
Note that the Meijers G-function in (3) and the Gaussian hypergeometric
function in (5) are both included in commercially available mathematical
software packages, e.g. Matlab. On the contrary, by using the PDF of the
Weibull distribution [16]
a x a1
f (x; a, b) = exp((x/b)a ),
b b
in (2) and (4), the integrals obtained can be expressed only for specific val-
ues of the Weibull shape and scale parameters (a, b)3 in terms of tabulated
functions. Note that although these integrals can be expressed in terms of
the H-function, their evaluation requires significantly more complex compu-
tations [19].
3
For example, (2) can be evaluated only for a = 1 using [17, Eq. (2.6.23.5)] and (4) for
a = 3/2 using [18, Eq. (2.8.5.6)].
10
analysis, we have examined the possibility of deriving expressions for these
two parameters in terms of independent controllable variables. A widely-used
statistical method that provides relationships between a dependent variable
and one or more independent variables is regression analysis [20]. However,
large sets of observational data are required in order to choose and apply
an appropriate regression technique. Those data sets were generated by the
Monte Carlo simulations that have been described in Subsection 3.1. The
simulations have been repeated for different values of the system operating
parameters (N, L) and different levels of cross-interference and SNR until a
sufficiently complete data set of SINR values was created. Then, the MLE
method was used to fit the gamma PDF to the PDF of the simulated SINR,
providing the estimates of the parameters (a, b) for each operating condition.
In this way, another set of data of the MLE parameters, denoted as (a, b), was
created. It is evident that the use of this distribution fitting method requires
repeating the simulations for every pair of the system operating parameters
(N, L) as well as for the different operating conditions. In order to obtain
analytical expressions for the parameters (a, b) as a function of the system
parameters (N, L), we tried linear and non-linear regression techniques. As
linear functions failed to fit the data, the non-linear regression (NLR) [21]
technique was applied. With this technique, successive approximations were
made in order to model the observational data with a heuristic function that
is a nonlinear combination of some independent variables. Thus, by applying
NLR in Matlab on the previously created data set of the MLE parameters,
the following approximate expressions were empirically derived for the pair
a, eb), in terms of (N, L)
(a, b), denoted in this case as (e
c1 c2 c4 c5
a
e + + c3 and eb + + c6 , (6)
N L N L
where e a and eb denote the parameters estimated with the NLR technique
and c1 -c6 are coefficients whose values depend on the cross-interference and
SNR levels. Note that these expressions are valid for both ZF and MMSE
equalization filters. The values for c1 -c6 with cross-interference factors =
{0, 0.1, 0.3, 1} and SNR values from -4 up to 20 dB (in steps of 2 dB) have
been obtained and can be found in tabulated form in [22]. Indicative values
for these coefficients, which have been also used for the performance evalua-
tion results presented in this paper, can be found in Table 2.
It is underlined that, as it will be shown next, the main advantage of
our approach is that it allows, by using (6) in connection with the closed-
11
Table 2. The coefficient values obtained for different equalization filters, cross-interference
and SNR levels.
c1 c2 c3 c4 c5 c6
ZF (=0, SNR=10dB) 19.4130 46.2444 1.3516 -0.5481 0.8821 -0.0059
ZF (=0.1, SNR=10dB) -16.3806 56.3218 1.2986 -0.1324 0.6963 -0.0038
ZF (=0.1, SNR=12dB) -16.2428 54.8538 1.3386 -0.0782 0.4647 -0.0031
MMSE (=0, SNR=10dB) 25.2595 57.7680 5.6632 -0.6919 -0.0164 0.2590
MMSE (=0.1, SNR=10dB) -5.8618 68.4794 5.6899 -0.3083 -0.2102 0.2471
MMSE (=0.1, SNR=12dB) -7.5849 82.9164 5.8843 -0.2207 -0.3626 0.2135
form expressions in (3) and (5), the accurate and computational efficient
evaluation of the EC and ABEP of different (N, L) VFDM systems. As
important case studies, we have considered the IEEE 802.11 and the 3GPP
LTE telecommunication standards in obtaining the performance evaluation
results, which will be presented next.
12
form expression (3) always coincide with the values of the integral expression
(2) for the EC. It is further noted that, while the results of the approximate
EC, i.e. when the parameters (a, b) are obtained using the NLR technique, are
precise for a receiver employing a ZF equalization, they are less accurate when
an MMSE equalizer is employed. This happens because of the more complex
nature of the MMSE. Moreover, it can be observed from the results of Figs. 3
and 4 that the EC performance of the secondary user is significantly affected
by the interference from the primary user. By comparing these results, it
is also evident that, as expected, the MMSE equalization outperforms the
ZF, since its operation is the maximization of the SINR in the presence of
interference.
4
Simulation (ZF)
0
4 8 12 16 20 24
SNR [dB]
Fig. 3. EC vs. SNR performance for the IEEE 802.11 a/g standard with ZF equalization,
for the various methods of evaluation under consideration and different values of .
13
4
Simulation (MMSE)
0
4 8 12 16 20 24
SNR [dB]
Fig. 4. EC vs. SNR performance for the IEEE 802.11 a/g standard with MMSE equal-
ization, for the various methods of evaluation under consideration and different values of
.
14
MMSE equalization has been employed and cross-interference factor values
= {0, 0.1, 0.3} have been considered. The results are illustrated in Figs. 5
and 6 for the WLAN and LTE standards, respectively, again showing a fine
agreement with the simulated performance results. It can be also observed
that the values of the closed-form expression (5) always coincide with the
values of the integral expression (4) for the ABEP performance. Note that,
by using the analytical approximation, the results are very precise when the
gamma distribution parameters (a, b) were obtained by the MLE method. In
contrast, when the NLR technique was used to obtain (e a, eb), the results are
less accurate, especially for high SNR and low cross-interference values. This
happens because the MLE method was applied after obtaining results for
the SINR of the specific, i.e. (N = 128, L = 32), system, whereas the NLR
technique was applied directly using the values of (N, L) in (6). Therefore,
there is a difference in accuracy between the values of (ea, eb) and (a, b) that
decreases as the SNR decreases and/or the cross-interference increases. In
the same figures, the dependence of the ABEP performance on the amount of
interference from the primary system is also illustrated. Comparing the two
case studies, it can be concluded that for the considered system model, the
WLAN standard outperforms the LTE, since it uses fewer subcarriers and
taps. This difference in performance is more pronounced at the high-SNR
and low cross-interference regimes.
5. Conclusions
This paper has proposed the approximation of the SINR statistics of a
VFDM cognitive system by the gamma distribution. Although other dis-
tributions have been considered, the gamma distribution provided the best
fitting results. Moreover, it has been shown that using this approximation,
closed-form expressions for the EC and the ABEP can be derived, facilitating
the performance evaluation. Furthermore, analytical expressions have been
provided for obtaining the parameters of the gamma distribution as functions
of the system operating characteristics. The accuracy of the proposed anal-
ysis was validated with analytical and complementary computer simulation
performance evaluation results for such VFDM cognitive spectrum sharing
systems employing the IEEE 802.11 a/g and 3GPP LTE standards.
15
0.15
0.1
Average Bit Error Probability (ABEP)
0.01
Simulation (W LAN)
1E-3
4 8 12 16 20 24
SNR [dB]
Fig. 5. ABEP vs. SNR performance for the IEEE 802.11 a/g standard with MMSE
equalization, for the various methods of evaluation under consideration and different values
of .
16
0.15
0.1
Average Bit Error Probability (ABEP)
0.01
Simulation (LTE)
1E-3
4 8 12 16 20 24
SNR [dB]
Fig. 6. ABEP vs. SNR performance for the 3GPP LTE standard with MMSE equal-
ization, for the various methods of evaluation under consideration and different values of
.
17
Acknowledgements
The authors would like to thank the anonymous reviewers as well as the
reviewing Editor for their comments, which have helped us in improving the
presentation of our research.
References
[1] E. Hossain, M. Rasti, H. Tabassum, A. Abdelnasser, Evolution toward
5G multi-tier cellular wireless networks: An interference management
perspective, IEEE Wireless Commun. 21 (3) (2014) 118127.
18
[10] J.G. Proakis, M. Salehi, Digital Communications, 5th Edition, McGraw-
Hill, 2007.
[13] I.S. Ansari, F. Yilmaz, M.S.Alouini, O. Kucur, New results on the sum
of Gamma random variates with application to the performance of wire-
less communication systems over Nakagami-m fading channels, Trans.
Emerging Tel. Tech., 28 (1) (2017).
[14] N.C. Sagias, G.S. Tombras, G.K. Karagiannidis, New results for the
shannon channel capacity in generalized fading channels, IEEE Com-
mun. Lett. 9 (2) (2005) 9799.
[15] A.P. Prudnikov, Yu.A. Brychkov, O.I. Marichev, Integrals and series,
Vol. 4: Direct Laplace transforms, Gordon and Breach, 1992.
[17] A.P. Prudnikov, Yu.A. Brychkov, O.I. Marichev, Integrals and series,
Vol. 1: Elementary Functions, Nauka, 1981.
[18] A.P. Prudnikov, Yu.A. Brychkov, O.I. Marichev, Integrals and series,
Vol. 2: Special Functions, Nauka, 1983.
[21] G.A.F. Seber, C.J. Wild, Nonlinear Regression, Hoboken, NJ: Wiley-
Interscience, 2003.
19
[22] K.D. Kanellopoulou, K.P. Peppas, P.T. Mathiopoulos, On
the SINR Statistics of a VFDM Cognitive Spectrum Shar-
ing System: equation (6) coefficients, Data set available at:
http://doi.org/10.5281/zenodo.573234, accessed 9/5/2017.
[24] Physical layer aspects for evolved UTRA, 3GPP TR 25.814 v.2.0.0, 2006.
20
that, he was with the Institute for Space Applications
and Remote Sensing (ISARS) of the National Obser-
vatory of Athens (NOA), first as his Director (2001 -
2005) and then as Director of Research (2006 - 2014).
From 1989 - 2003, he was a faculty member in the De-
partment of Electrical and Computer Engineering, the
University of British Columbia (UBC), where he was a Professor from 2000 -
2003. He has been an ASI and Killam Fellow and has served as an Editor for
several IEEE and IET journals. His research activities are in the general area
of wireless communications for terrestrial and satellite based applications, as
well as in the remote sensing field with emphasis on LiDAR systems and
photogrammetry. He has published more then 100 journal papers and 120
conference papers in these fields and has received two best paper awards.
21