Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
madarJ
our
nal
I
SSN2313-
156X
f
orCommuni
cat
ions
,Inf
ormat
ionT
echnol
ogy
,andAppl
icat
ions
J
uly2
01,Vol
4 .0,No.0
1 1
Cogni
ti
veRadi
o
Pr
omi
singT
echnol
ogyf
ort
heConges
tedSpec
trum
Pu
bl
is
he
dbyR&DO ce
Al
mad
arAl
j
adi
dCo
.Li
by
a
www.
al
mada
r.l
y
Almadar Journal for Communications, Information
Technology, and Applications
Vol. 01, No. 01, July 2014
Editorial Page: Mobile Networks: The Platform for Smart Connected World
Mohammed S. Elmusrati, Page 1.
Interference Mitigation Using Optimal Successive Group Decoding for Interference Channels
Omar Abu−Ella and Mohammed S. Elmusrati, Page 37.
Advertisments in Arabic
Libyan Innovation Prize 2014 by National Authority for Research, Science, and Technology.
Introduction to Almadar Journal for Communications, Information Technology, and Applications, the
Arabic version.
1
EDITORIAL PAGE
Mobile Networks: The Platform for Smart Connected World
Fig. 2. Fixed and mobile subscription growth 2009-2018 [1]. Fig. 3. Global smartphone sales forecast by wholesale price tier. Smart-
phones below USD 190 are the biggest contributor to volume increases
(source:Strategy Analytics) [5].
need to ensure that high-value roaming subscribers can enjoy
a consistent MBB experience.
While LTE is often viewed as the key solution for handling A. Enter the affordable smartphone
high-capacity situations, WCDMA/HSPA is equally capable of
handling large numbers of smartphone users. What is more,
this capability is being dramatically extended over the coming At the centre of this more than fourfold predicted increase in
years, mainly through the implementation of software features. MBB subscriptions is the rising tide of affordable smartphones
and, to a lesser extent, tablets. According to the market analyst
Fundamentally, all operators whether they are voice-centric
firm Strategy Analytics [3], [4], 217 million smartphones were
GSM operators or MBB-centric operators who are deploying
sold in the fourth quarter of 2012 some 40 percent more than
LTE will benefit from having a strong WCDMA/HSPA net-
in the same quarter of 2011 overall. Within this figure, sales of
work, as HSPA is the only technology that will be used in all
Android smartphones grew close to 90 percent year-on-year.
smartphones for the foreseeable future.
And there is a similar story for tablets: fourth-quarter 2012
Operators deploying LTE in the same frequency band
sales were 45 million; 67 percent higher year-on-year overall
in which they already offer HSPA will benefit from
and 85 percent higher year-on-year for Android devices. The
WCDMA/HSPA functionality that allows effective spectrum
overall growth trend is set to continue. Total smartphone
refarming.
subscriptions are set to rise from 1.1 billion at the end of
2012 to about 3.3 billion in 2018. One of the most significant
II. T HE H IGH - GROWTH M ARKET factors behind this rapid growth in smartphone adoption will
be their significantly lower average selling price, driven by
In the past couple of years, MBB has become firmly
the availability of lower-cost chipsets, especially from Asian
established as people have grown accustomed to having high-
manufacturers.
speed-internet access wherever they go. According to market
analyst firm Wireless Intelligence [2], MBB technologies now We are already starting to see a number of sub-USD
account for about one-quarter of total global connections. 100 WCDMA/HSPA smartphones with 14.4Mbps, dual-band,
WCDMA/HSPA makes up the vast majority of MBB con- dual-core processor capabilities. There is also strong growth
nections and is the fastest-growing wireless technology so far. in the midrange smartphone market (USD 100-200), thanks
Now MBB is entering its next phase of expansion as to the arrival of much lower-cost chipsets for 42Mbps, quad-
the availability and affordability of devices grows, especially core, HD (1280x720) devices. According to market research
in developing markets. Ericsson’s estimates show that there firm Strategy Analytics [5], unit sales of entry and mid-range
were 6.6 billion mobile subscriptions (excluding machine-to- smartphones are set to grow at a compound annual rate of 45
machine subscriptions) at the end of 2012 [1]. Of these, about percent between 2011 and 2016, while unit sales of premium
1.5 billion were MBB subscriptions (including feature phones, and high-end smartphones will grow by a compound rate of
smartphones, mobile PCs, tablets, mobile routers and dongles). just 1.8 percent, as illustrated in Figure 3.
According to Wireless Intelligence, mobile subscriptions in This increase in entry and mid-range smartphones will
the developing world passed the 5 billion mark in the third power demand for MBB coverage, capacity and throughput,
quarter of 2012, and now comprise almost 80 percent of the especially in developing markets, where mobile devices will
world’s total. By the end of 2018, total mobile subscriptions provide many people’s first taste of high-speed internet con-
are expected to grow to 9.3 billion, and about 6.5 billion of nectivity. So, which networks will this growing band of MBB
these will be for MBB, as shown in Figure 2. subscribers be using?.
4 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
Fig. 5. How average monthly mobile-data usage will grow for mobile PCs,
tablets and smartphones [1]
their WCDMA/HSPA networks provide a comparable quality station portfolio for optimized coverage and capacity.
experience when users move outside LTE coverage. As de- • Excellent in-service performance built on scalable and
mand grows, MBB services are putting immense pressure on future-proof 3G platforms.
limited radio spectrum, and operators will need to find new • A clear evolution path to HSPA Evolution, which will
ways of using this spectrum ever more efficiently across 2G, provide speeds of more than 84Mbps in the downlink
3G and 4G technologies. So, how can operators ensure they and more than 12Mbps in the uplink.
have the WCDMA/HSPA coverage, capacity, performance and Behind these headline features, there are numerous improve-
business models they need to meet the rapid increases in MBB ments being developed and applied to WCDMA/HSPA that
uptake, usage and expectations?. will further boost its ability to deliver a better user experience
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 5
through improved smartphone capacity and higher uplink and the Global mobile Suppliers Association (GSA), 57 commer-
downlink bit rates. cial WCDMA 900MHz networks have been deployed in 39
Many newer smartphones on the market support features countries (as of December 2012) [6]. Other WCDMA/HSPA
that enable a substantial rise in overall smartphone capacity coverage-enhancing measures include four-way receiver diver-
(number of users per cell). For instance, multicarrier tech- sity (rather than two-way), lower speech rate for better voice
nology enables the WCDMA/HSPA system to use multiple coverage, and the capability for improved scaling of control
5MHz carriers for one user in both the uplink and downlink. and traffic channels.
From 3GPP Release 10 onwards, WCDMA/HSPA supports
multicarrier operation on up to four carriers in the downlink
B. Adapting data plans to market needs
(which can be spread across one or two frequency bands) and
up to two carriers in the uplink. Multicarrier technology pro- One important aspect of driving the uptake of MBB, and
vides both substantial capacity gains, as well as throughput and increasing ARPU overall, is to ensure that data plans meet
peak rate gains in the cell. The first step in WCDMA/HSPA subscriber needs, especially in markets where the ARPU has
multicarrier development (2x5MHz downlink) is available in traditionally been low. The MBB pricing models used in
smartphones today. Multicarrier in the uplink is expected to more established markets may not be appropriate for many
arrive in smartphones toward the end of 2013. developing markets where ARPU can be one-tenth that of a
Another function that boosts capacity is the 3GPP-specified developed market, for example. With the trend to bring your
Fast Dormancy Release 8. This enables a smartphone to own device (BYO D) in many mature markets, there are sev-
move to an energy-efficient state (Universal Terrestrial Ra- eral new plans that are designed to attract new user categories.
dio Access Network (UTRAN) Registration Area Forward Often, operations are run by a mobile virtual network operator
Access Channel (URA_FACH)) as soon as it has no data (MVNO) to differentiate from the major (owner) brand. These
to send, dramatically decreasing the time the smartphone is offerings tend to be characterized by a distinct internet flavour,
in the most resource-intensive state (Cell Dedicated Channel no operator subsidies for devices and less customer support.
(Cell_DCH)). Such approaches have proved successful in mature markets
The WCDMA/HSPA radio uplink is non-orthogonal by and have attracted new smartphone users.
nature, meaning that all users in a cell interfere with each Another trend in mature markets is “prepaid as postpaid”.
other on the radio interface. The most efficient way to counter Traditionally, before data buckets were introduced, postpaid
this interference is to eliminate unnecessary, or excessive, customers often represented a higher ARPU group than pre-
network chatter, such as control signalling. This is the purpose paid customers. Data buckets are often defined by a fixed
of a feature called Continuous Packet Connectivity (CPC), traffic amount for a fixed price, which has made the distinction
which has the effect of dramatically improving uplink capacity between prepaid and postpaid less important for operators.
by limiting interference. Many commercially available smart- Bucket allowances are typically not exceeded, which provides
phones already use fast dormancy and growing numbers are increased revenue opportunities for operators. With the in-
appearing with CPC. creasing popularity of tablets in developed markets, opera-
In addition, advances in base station signal processing tors have an opportunity to monetize this trend by offering
provide a clearer received uplink signal, which reduces the attractive tethering add-ons and sharing plans. Several oper-
cell’s total interference while sustaining uplink quality. Four- ators have already successfully transformed their subscriber
way (instead of two-way) receive diversity further amplifies plans from traditional voice minutes and SMS volume-based
the positive effects of advanced uplink receivers. charging to charging based on actual data use.
WCDMA/HSPA enables users that are sending and receiv- The increasing availability of low-priced smartphones will
ing only small bursts of data (as is typical with smartphones enable 2G/feature phone users to move to a highly capable
most of the time) to handle that data in a semi-active state WCDMA/HSPA smartphone next time they invest in a new
called CELL_Forward Access Channel (CELL_FACH). To device, for about the same price. As it is quite likely that
enable even more efficient CELL_FACH operation, 3GPP the WCDMA/HSPA smartphone will be the main broadband-
has specified High-Speed (HS)_FACH for the downlink and access device for many consumers, operators need to offer
Enhanced Uplink (EUL)_FACH for the uplink. HS_FACH- appropriate, affordable plans. One way to enable MBB on
enabled smartphones are already available and EUL_FACH these devices is to permit small payments such as pay-per-
capable smartphones are expected during 2013. hour or pay-per-day. Further, some operators have developed
plans that permit pay-per- (small) data volume, and here it is
important that the consumer is given control of the actual data-
A. Enhancing coverage traffic consumption. For instance, using the operator’s own
The ability to extend WCDMA/HSPA coverage efficiently portal would not consume any of the data allowance, while
is vital to turning the proliferation of low-cost HSPA smart- accessing Facebook would. With consumers in control, over
phones into increased revenue from MBB services. One key time they may be more willing to pay for a data bucket for
way of doing this is to re-farm the 900MHz spectrum from internet access, resulting in increased operator revenue.
GSM to WCDMA/HSPA. This spectrum typically gives a 6dB Yet another way to boost revenue is to allow consumers
link budget advantage over the 2100MHz spectrum, which time-limited free MBB access to any predefined internet ser-
translates into substantial coverage advantages. According to vices in return for being shown advertisements before access
6 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
IV. C ONCLUSION
Whether or not operators are rolling out LTE 4G networks,
they will need to focus their attention on the performance
of their WCDMA/HSPA networks if they are to deliver a
consistent, high-quality user experience throughout their cov-
erage areas. Fundamentally, it is vital that network technology
matches the capabilities and cost of the devices subscribers are
choosing to use to access MBB services. For the foreseeable
future, WCDMA/HSPA will be by far the biggest technol-
ogy by subscription numbers and by population coverage.
CDMA/HSPA already provides the backbone for most MBB
services, and is being continuously developed to efficiently
deliver a true broadband experience that is on a par with 4G
to any device, in any location. No other technology can make
that claim.
V. GLOSSARY
ARPU Average Revenue per User
BYOD Bring Your Own Device
CELL_DCH Cell Dedicated Channel
CELL_FACH Cell Forward Access Channel
CPC Continuous Packet Connectivity
EUL_FACH Enhanced Uplink Forward Access Channel
GSA Global mobile Suppliers Association
HS_FACH High-Speed Forward Access Channel
MBB mobile broadband
MVNO mobile virtual network operator
TD-SCDMA time division synchronous code division multiple access
URA_FACH UTRAN Registration Area Forward Access Channel
UTRAN Universal Terrestrial Radio Access Network
R EFERENCES
[1] Ericsson Mobility Report, November 2012, available at: http://www.
ericsson.com/ericsson-mobility-report
[2] Global cellular market trends and insights, Wireless Intelligence,
Q4, 2012, available at: https://wirelessintelligence.com/analysis/2012/01/
global-cellular-market-trends-andinsight-q4-2011/
[3] Android and Apple iOS Capture a Record 92 Percent Share of
Global Smartphone Shipments in Q4 2012, Strategy Analytics, January
2013, available at: http://www.strategyanalytics.com/default.aspx?mod=
reportabstractviewer&a0=8155
[4] Global Tablet OS Market Share: Q4 2012, Strategy Analytics, January
2013, available at: http://www.strategyanalytics.com/default.aspx?mod=
reportabstractviewer&a0=8147
[5] Global LTE Smartphone Shipments Will Reach 275 Million
Units in 2013, Strategy Analytics press release, December 2012,
available at: http://www.strategyanalytics.com/default.aspx?mod=
pressreleaseviewer&a0=5310
[6] UMTS900 Global Status Report, Global mobile Suppliers Associa-
tion, http://www.gsacom.com, November 2012, available at:http://www.
gsacom.com/php/access.php4
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 7
Abstract—There has been an increasing demand for high that have a lower priority on the spectrum and try to use it
data rate services, which necessitates the development of more opportunistically when it is not used by the licensees [5].
efficient schemes for using the scarce radio spectrum. Traditional
frequency allocation schemes are static and, thus, not capable
of accommodating the growing number of wireless users and II. C OGNITIVE R ADIO F UNCTIONS AND N ETWORK
services. Hence, it is required to have some form of spectrum A RCHITECUTE
sharing between existing and new users of the radio spectrum.
Cognitive Radio (CR) is a system that allows for sharing the CR system functions involve the following steps: spectrum
spectrum among users, which offers a highly flexible alternative sensing, analysis, reasoning, and adaptation [4]. Spectrum
to the traditional fixed frequency band assignment. In this paper,
sensing is needed in order to detect the unutilized portion of
we survey the recent progress along with some issues related to
cognitive radio technology. the spectrum as will be discussed in Section V. It then performs
the spectrum management and handoff so as to select the best
frequency band with a smooth transition and low latency.
I. I NTRODUCTION TO C OGNITIVE R ADIO Cognitive radio architecture involves a primary network,
which contains a set of primary users or license holders
restriction is imposed on the receivers. The Interference tem- 1) Hardware requirements: In traditional systems, the re-
perature is defined as “the temperature equivalent to the RF ceivers are tuned to receive signals over a limited frequency
power available at a receiving antenna per unit bandwidth” band and the problem of estimating the noise and interference
[4], that is is easier. However, cognitive radio terminals should be capable
PI (fc , B) of sensing signals over a wide bandwidth, which required wide
TI (fc , B) = (1)
KB band amplifiers and antennas. Furthermore, for the cognitive
radios to perform computationally complex functions with a
PI (fc ,B) in (1) is the average interference power that is
low delay, high speed Digital Signal Processing (DSP) units
measured in Watt and centered at frequency of fc with a
are required.
bandwidth of B measured in Hertz. K is the Boltzmann’s
Spectrum sensing can be performed based on either single-
constant (K = 1.38×10−23 ) with the unit of Joules per degree
radio or dual-radio architecture. The former is simple and
Kelvin.
cheap but has a specific time slot dedicated for the spectrum
The interference temperature limit is defined as a maximum
sensing with a bounded sensing duration and poor spectrum
amount of interference that can be tolerated for a particular
accuracy. Also, since a fraction of the time slot is used for
frequency band and location. By this constraint, the transmis-
sensing instead of transmitting data, the spectrum efficiency is
sion plus noise and interference of any unlicensed transmitter
low. The dual radio architecture allows sensing and data trans-
must not go beyond the interference temperature limit at the
mission at the same time and, hence, provides higher spectrum
receiver of the licensee. For example, if TL is the interference
efficiency and better sensing accuracy. However, it has a higher
temperature limit of the legitimate user for a given frequency
cost, more complexity and more power consumption.
band having a bandwidth of B, the average interference of the
2) Hidden primary user problem: In the case of severe
secondary transmitter should fall below KBTL .
multipath fading or shadowing, there is a possibility for the
In an overlay cognitive radio, the secondary user can coexist
primary user to be hidden to the secondary system. Thus, the
with the primary user over the same spectrum by knowing
primary user might not be detected and thus unwanted interfer-
its channel with the licensee along with its operation. For
ence is introduced to the legitimate user by the secondary user
instance, it is required to be aware of the primary user’s
transmission. The hidden primary user problem is illustrated
codebook that can help in decoding the legitimate user’s
in Figure 1. Cooperative spectrum sensing is used to combat
transmission [8].
this problem [12].
D. Cooperative Spectrum Sensing the cognitive radio traffic. In the case of large number of users,
The three main factors that limit the spectrum sensing a huge bandwidth is required to report the information. In
performance include noise, shadowing, and multi-path fading. order to decrease this bandwidth, CR devices with the same
When the SNR of the received primary signal is below a data can report their decisions to the central body, which
predetermined threshold, reliable detection is not possible requires censoring some of the data.
even with long sensing time. Therefore, the secondary user 2) Distributed Sensing: In this approach, cognitive nodes
cannot detect the presence of the primary user and would interchange information with each other, however, they make
interfere with it. As an example, the primary owner of the their own decisions about the portion of the spectrum that
spectrum can only be detected by one secondary user in a they use. Distributed Sensing does not require a backbone
scenario depicted in Figure 7. due to a deep shadowing effect. infrastructure, which implies less cost, and hence differs
Cooperative sensing can be used to overcome this problem. from the centralized sensing. However, the network overhead
Some advantages of cooperation include: needed for sharing the information among collaborating radios
will increase resulting in reduced spectrum efficiency.
3) External Sensing: In this technique, an external agent or
base station performs the spectrum sensing of primary users.
The results about spectrum availability are then broadcast to
the cognitive radios. External sensing reduces the impact of
the hidden primary user problem and improves the power and
spectrum efficiencies for the CRs since the processing is done
by an external agent and CRs do not waste time or energy in
sensing the spectrum [3].
4) Recent Spectrum Sensing Schemes: There has been
some recent techniques for spectrum sensing using machine
learning schemes initially introduced in [14] for single carrier
modulation schemes and then extended to multicarrier Or-
thogonal Frequency Division Multiplexing (OFDM) schemes
in [15]. Those schemes train a polynomial classifier based
on different primary user pattern and then used the trained
classifier to decide upon the availability of the channel. These
schemes provide significant improvement in performance in
terms of improved detection probability without increasing the
false alarm probability, especially when used in a cooperative
architecture. Since the classifier training is done offline then
there is no major increase in complexity. Finally, a new CR
system using an overlay structure has been proposed in [16]
that alleviates the problem of spectrum sensing and allows for
simultaneous use of the radio channel.
VIII. C ONCLUSIONS
Cognitive radio is a new paradigm for improved radio
spectrum utilization and is expected to be a main driver
towards the goals of future wireless connectivity of billions
of radios. An overview of cognitive radio technologies and
issues along with standardization activities is presented in this
paper.
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 13
Abstract—From the literature we know that power allocation soft sensing information, and the capacity of the secondary
is of great importance in managing the interference in spectrum user was maximized under a peak power constraint at the
sharing networks, maximizing the spectrum reuse, increasing primary receiver. Power control for opportunistic spectrum
communication capacity, and making our living environment
greener. In this paper we reviewed the most important and up- access (OSA) in TV bands is investigated in [12] and [13],
to-date results of the power allocation approaches proposed in where the primary users transmit all the time and spatial
literature from an information-theoretic perspective. Therefore, (rather than temporal) spectrum opportunities are exploited by
we will take a look at the optimal power allocation strategies of secondary users. For the interference control of the secondary
the secondary users in order to maximize their ergodic capacity users over television white spaces, Koufos et al. in [14]
and effective capacity over fading channels. This survey improves
the understanding of ultimate performance limits of the cognitive proposed the power density and deployment based transmit
radios and the cognitive radio systems design. power control of the secondary users such that the quality of
the TV services is not violated by the aggregated interference.
Index Terms—Cognitive radio, ergodic capacity, effective ca-
pacity, optimum power allocation. Gastpar investigated the ergodic capacity of different non-
fading additive-white-Gaussian noise (AWGN) channels [15],
I. I NTRODUCTION [16]. The transmit power of the SU is regulated by the average
interference power received at a third-party receiver. The
Patial considerations for frequency reuse have been exten-
S sively studied in cellular systems. However, these systems
largely differ from the cognitive radio (CR) systems [1]. As
author illustrated that the received-signal constraints can lead
to substantially different results as compared to transmitted-
signal constraints. There are some important findings which
the command-and-control structure of frequency allocation
are different from a conventional point-to-point communica-
for traditional wireless communications, the within-system
tion. Without fading the author showed that in the point-to-
interference is the dominant interference to the users operating
point case, the transmitted- and received-power constraints are
with the same operator. This kind of interference can be well
largely equivalent. While in network cases, they can lead to
controlled through planning. For these systems, power control
quite different conclusions, for example, multiple access chan-
has been studied in SIR-based, e.g. [2], and information-
nels with dependent sources and feedback, and collaborative
theoretic contexts for fading and non-fading channels, for
communication scenarios. Ghasemi and Sousa, in [17], showed
instance [3]–[6]. However, in cognitive radio networks, the
that in many cases significant capacity gains may be achieved
interference is caused not only by the secondary users (SUs),
if the channels are varying due to fading and shadowing under
or cognitive users, sharing the same spectrum, but also by
either the average or the peak interference power constraint. In
the primary users (PUs), or licensed users, who share the
[18], the authors extended the work in [17] by investigating the
spectrum. Additionally, the secondary users should not cause
achievable capacity gains in asymmetric fading environments.
unacceptable interference to the primary users [7], [8].
In this paper we focus on the information-theoretic ap- Then, Musavian and Aı̈ssa in [19] studied the capacity gains
proaches, i.e., reviewing the optimal power allocation ap- offered by the spectrum-sharing approach in a Rayleigh fading
proaches for the SUs to maximize the achievable rate under environment subject to both average and peak received-power
certain constraints. The framework employed to evaluate the constraints at the primary receiver. In [20], Kang et al. studied
power allocation schemes and other performance matrices is the optimal power allocation strategies to achieve the ergodic,
mainly based on information theory [9]. There is a growing delay-limited, and outage capacities of a secondary fading
body of literature on power control/allocation in CR systems. channel subject to a diverse combinations of peak/average
In [10] and [11], power control for one pair of secondary transmit and/or peak/average interference power constraints.
users coexisting with one pair of primary users is considered. The authors observed that fading of the channel from sec-
In [10], the secondary transmitter adjusts its transmission ondary transmitter to primary receiver can be a good phe-
power to maximize its data rate without increasing the out- nomenon for maximizing the capacity of SU fading channel.
age probability at the primary receiver. The authors in [11] Zhang concluded in [21] that the average-interference-power
proposed the optimal power control schemes based on the (AIP) constraint can be more advantageous over the peak-
interference-power for minimizing the resultant capacity loss
The authors are with Communications and Systems Engineering Group, of the primary fading channel, and AIP should be used for
University of Vaasa, Finland. (emails: ruifeng.duan@ieee.org; {moel,
rvir}@uva.fi). the purposes of both protecting the PR communications as
The content of this paper is part of Duan’s thesis, and is reprinted with the well as maximizing the CR capacity. Therefore, in this paper
permission of the University of Vaasa. This work was supported in part by we review the channel model and the concepts of capacity,
the SMACIW (Statistical Modeling and Control of Aggregate Interference in
Wireless Systems, Decision no. 265077) project funded by the Academy of and then survey the main results of optimal power allocation
Finland. approaches for cognitive radios.
14 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
II. C HANNEL M ODEL AND C ONCEPTS OF C APACITY the natural logarithm operator, and Ex denotes the expectation
In this section we introduce the channel model, and review operator over x in this paper. The secondary user chooses the
two important concepts, i.e., ergodic capacity and effective optimal transmit power to maximize the achievable rate ac-
capacity. We consider independent and identically distributed cording to the instantaneous CSI of the two channels instead of
additive white Gaussian noise (AWGN) block-fading channels. only its own CSI as in the traditional wireless communications
Generally, we do not specify the fading distributions, but we systems. The maximization is over power allocation functions
mention them together with the associated results. The block- that are being discussed later in certain problems.
fading, or quasi-static, channel model was introduced in [22]
and has been commonly used in the literature for studying B. Effective Capacity
wireless communications systems over slowly-varying fading From literature, we know that the ergodic capacity has
channels [22], [23], through which a codeword spans only a no transmission delay limitation, while the outage capacity
certain number of fading blocks. During each fading block, does not allow any delay [31]. In order to study the delay
the channel gain remains constant while varying from block performance, the concept of effective capacity (EC) was
to block. We assumed that the primary user(s) are located developed in [32], [33] to define the maximum arrival data rate
far away from the secondary receiver so that there is no that can be supported by the channel subject to the required
significant interference to the secondary user [17], [19], [24]– communication delay. It is a link-layer channel model and
[26]. In addition, the interference from the primary user to the can be interpreted as the dual of effective bandwidth [34].
secondary user could be considered being absorbed into the The quality of service (QoS) is represented by a term, named
noise if the random Gaussian codebooks are applied at the QoS exponent θ ∈ R++ defined in Eqn.(3). The EC bridges
primary transmitters [27]–[29]. the ergodic capacity and the outage capacity. When the QoS
For imperfect channel information scenarios, we adopt exponent θ → 0, it means that there is no delay limitation,
the following channel estimation methods for measuring the and the EC equals the ergodic capacity. On the other hand, the
channel gain of ST-PR link, which has been widely used in link cannot tolerate any delay as θ → ∞. This concept has
literature, e.g. [19]. For Rayleigh fading channels, the complex received much attention in the point-to-point communication
channel gain from the secondary transmitter to the primary scenarios, e.g., [35], [36], as well as in cognitive radios, e.g.,
receiver, cps , is zero mean circularly symmetric complex [37], [38] and references therein. The effective capacity along
Gaussian distributed variable with the imaginary and real parts with energy efficiency was also investigated in [39].
having variances of 0.5. However, the CR transmitter is only Let q(x) be the queue length of a stationary ergodic arrival
provided with partial channel information of cps , namely c̃ps , and service process. The probability that q(x) exceeds a certain
where cps and c̃ps are jointly ergodic and stationary Gaus- threshold Tq decays exponentially as a function of Tq , and the
sian processes. The secondary user performs minimum mean delay QoS exponent is defined in [32] as
square error estimation (MMSE) of cps given c̃ps , such that
ĉps [n] = E {cps [n] | c̃ps [n], c̃ps [n − 1], ...}, where [n] denotes log(Pr {q(∞) > Tq })
θ = − lim . (3)
the time index. The MMSE estimation error can be presented Tq →∞ Tq
as c̆ps [n] = cps [n]−ĉps [n], and c̆ps [n] and ĉps [n] are zero mean It is worth noting that θ → 0 indicates that the system has
circularly symmetric complex Gaussian distributed variables no delay constraint, while θ → ∞ implies a stringent delay
2 2
with variances 1−σ 2 and σ2 respectively. So the associated constraint. The effective capacity is defined in [32, eqn. (12)]
2 2
channel power gain can be presented as g = |cps | , ĝ = |ĉps | , by
2
and the channel power gain estimation error by ğ = |c̆ps | . The 1 h Pt i
probability density function of estimated channel power gain, EC(θ) = − lim log E e−θ i=0 R[i] , t ≥ 0 (4)
t→∞ θt
ĝ, is characterized by [30]:
where {R[i], i = 1, 2, ...} denotes a discrete-time service pro-
1 ĝ cess of the maximum achievable instantaneous service rate of
fĝ (ĝ) = exp − , ĝ ≥ 0 (1)
1 − σ2 1 − σ2 time [i], which is assumed to be ergodic and stationary. For a
block fading channel, the EC can be reduced to [35],
A. Ergodic Capacity h i
1
This subsection reviews the ergodic capacity formulation EC(θ) = − log E e−θR[i] . (5)
θ
of the secondary user. With perfect channel state information
The maximum achievable instantaneous service rate R[i] of
(CSI) of the secondary link (ST-SR) and the secondary trans-
block i can be expressed as R[i] = T B log (1 + γ[i]), where T
mitter to the primary receiver (ST-PR), the ergodic capacity
denotes the block length duration, B is the channel bandwidth,
of the secondary user is given in [17] by
and γ[i] is the instantaneous SINR of block i.
ps (gss , gps )gss
maximize Egss ,gps log 1 + (2)
ps (gss ,gps )≥0 N1 B III. E RGODIC C APACITY
where ps (gss , gps ) is the transmit power of the secondary This section reviews the optimal power allocation policies
transmitter, gss and gps denote the channel power gains of of the secondary user in order to maximize its ergodic ca-
ST-SR and ST-PR, respectively. N1 represents the additive pacity (maximum achievable rate) under various constraints
white noise density at the secondary receiver. log(·) denotes categorized as short-term and long-term constraints. In the
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 15
transmitter. The instantaneous CSI can be fed back to the N1 is the additive white Gaussian noise density at the SR. log
secondary transmitter [20], [41]. denotes natural logarithm operation.
Figure 1 illustrates the achieved ergodic capacity of the The ergodic capacity of the SU with MVPA can be achieved
secondary user versus various peak transmit powers along through employing a frame-work presented by Zouheir Rezki
with different values of the peak interference power, where and Mohamed-Slim Alouini in [43]. According to [43], the
we assumed that all the mean values of the channel power interference outage probability constraint in the above opti-
gains are 1. We can observe that when the PIP constraint mization problem is equivalent to
is dominant, i.e. Pmax ≪ Qpk , the secondary user may Qpk
simply transmit at the maximum power to achieve its ergodic ps (gss , gps ) ≤ (11)
Fg−1
ps (1 − POth )
capacity. Additionally, under Pmax ≪ Qpk , the fades of gps
are not beneficial to the ergodic capacity of the SU. This is where Fg−1
ps
(1 − POth ) denotes the inverse c.d.f. of gps . For
because over Rayleigh fading the SU is not able to exploit Rayleigh fading scenarios, the probability density function of
16 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
0.8
Qpk = 5dB
0
10 Qpk = 1dB
0.7
0.5
−1
10 0.4
Pth=0.01, ρ = 1 0.3
O
Pth
O
=0.2, ρ = 1
0.2
Pth
O
=0.01, ρ = 2
th
PO =0.2, ρ = 2 0.1
−2
10
0
−10 −8 −6 −4 −2 0 2 4 6 8 10 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Qpk (dB) Pth
O
th
Fig. 2. Ergodic capacity of the SU versus Qpk for different values of PO th for different values of Q
Fig. 3. Ergodic capacity of the SU versus PO pk
and ρ = g ps /g ss . and ρ = g ps /g ss = 1.
the channel power gain is continuous and not null so that of the optimization problem is given by
Fg−1 (·) exists. This new transformed constraint is called a
ps ps (gss , gps )gss
variable peak transmit power constraint in [43]. In MVPA O2 : maximize E log 1 + (14)
ps (gss ,gps )≥0 N1 B
the secondary transmitter has the statistical information in
stead of the instantaneous ST-PR channel state information. In C12 : E {ps (gss , gps )gps } ≤ Qav (15)
addition, Fg−1ps
(1 − POth ) takes a fixed value [43]. This means C22 : E {ps (gss , gps )} ≤ Pav (16)
that the secondary user uses fixed transmit power which is
not variant with respect to gss . Based on the setting that where and C12
denote the AIP constraint and ATP con- C22
gps is exponentially distributed straint, respectively, associated to objective function O2 . Qav
with a mean of gps , i.e., represents the predefined average interference power caused by
Fg−1
ps
(1 − POth ) = g ps log P1th . Consequently, the fixed the SU at the primary receiver, and Pav denotes the average
O
transmit power for the secondary transmitter is transmit power.
Qpk Here we have to point out that besides the mentioned
ps (gss , g ps ) ≤ (12) long-term constraints above there is another constraint called
g ps log P1th primary capacity loss constraint (PCLC) proposed in [45].
O
This method was shown to be better than the common ones,
where we may use the notation ps (g ps ) rather than
e.g. the average and/or peak interference power constraints,
ps (gss , gps ). We can obtain the ergodic capacity of the SU
in terms of achievable ergodic capacities of both the primary
exploiting MVPA as following
and the secondary links. It protects the primary transmission
Z ∞
gss ps (g ps ) by ensuring that the maximum ergodic capacity loss of the
C= log 1 + fgss (gss ) dgss . primary link, due to the secondary transmission, is no greater
0 N1 B
Z ∞ than some predefined value. However, to enable the scheme,
gss ps (g ps ) 1 − ggss
= log 1 + e ss dgss . not only the CSI of the secondary fading channel and the
0 N1 B g ss
fading channel from the secondary transmitter to the primary
N1 B
N1 B receiver, but also the CSI of the primary direct link. For details
= −e gss ps (gps ) Ei − (13)
gss ps (g ps ) please refer to [45].
where in the last two 1) AIP constraint only with perfect CSI: In this scenario,
R x steps
t
we have the help of [44, 4.337-
the secondary user aims to maximize its ergodic capacity under
2], and Ei (x) = −∞ et dt, x < 0 denotes the exponential
integral function [44, 8.211-1]. This result also was shown in the average interference power (AIP) constraint predefined by
[42] using a different method of proof. The ergodic capacity the primary user. This problem is denoted as (O2 , C12 ). The
versus Qpk and POth are plotted in Figure 2 and in Figure optimal power allocation scheme is waterfilling, which is given
3, respectively. We have to point out that in the discussed in [17] by
environment if POth → 0, the secondary user needs to stop +
∗ 1 N1 B
transmission. ps (gss , gps ) = − , (17)
λgps gss
+
where [x] = max(x, 0), and gss and gps denote the channel
B. Long-Term Constraints power gains from the secondary transmitter to the secondary
We consider the long-term constraints are as follows: aver- receiver (ST-SR) and primary receiver (ST-PR), respectively.
age transmit power constraint (ATP) and average interference N1 is the noise density at the SR, B denotes the bandwidth,
power constraint (AIP) at the primary user. One formulation and λ ≥ 0 is the Lagrangian multiplier satisfying the AIP
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 17
0
0 10
10
AWGN
Rayleigh
2
AWGN Rayleigh, σe=0.01
−1
10 Lognormal,σ=4dB Rayleigh, σ2e=0.1
Lognormal,σ=8dB
2
Rayleigh Rayleigh, σe=0.4
Nakagami, m=2 −1
10
−15 −10 −5 0 5 10 15 20 −10 −8 −6 −4 −2 0 2 4 6 8 10
Qav/N1B (dB) Qav/(N0B) in dB
Fig. 4. Ergodic capacity of the SU with perfect CSI under AIP constraint in Fig. 5. The ergodic capacity of the SU under AIP constraint for AWGN, and
different fading scenarios. Rayleigh fading with/without estimation errors of ST-PR.
TABLE I
E RGODIC C APACITY OF THE SU WITH PERFECT SCI UNDER AIP CONSTRAINT.
* where γ0 = 1/λN1 B.
4.5
Pav= −10dB
4 Pav= −7.5dB
av
3.5
Pav= −2.5dB
4
3 Pav= 0dB
Pav= 2.5dB 3
2.5 Pav= 5dB
Pav= 7.5dB 2
2
Pav= 10dB
1
1.5 Pav= 12dB
Pav= 15dB 0
1 20
15
10 10
0.5 5
0 0
0 −5
−10 −5 0 5 10 15 Qav (in dB) −10 −10
Qav (in dB) Pav (in dB)
Fig. 6. The ergodic capacity of the SU under AIP and ATP constraints. Fig. 7. The ergodic capacity of the SU under AIP and ATP constraints.
The associated optimal power allocation scheme for the sec- ST-PR channel state information under the AIP constraint are
ondary user to maximize the ergodic capacity is given in [20] presented in detail in [50].
by From previous analysis, we can see that under the long-
+
1 N1 B term constraints the optimal power allocation approaches are
p∗s (gss , gps ) = − , (22) (modified) waterfilling, and the water-level is jointly decided
µ + λgps gss
by the long-term constraints. In the following, we will take
where µ and λ are the nonnegative Lagrangian variables a look at how the combined constraints, long-term and short-
associated with the average transmit power constraint in (16) term, influence the power allocation and the ergodic capacity.
and average interference power constraint in (15), respectively.
We can see that this scheme is also waterfilling. However,
the water lever is related to not only the interference channel C. Combined Long-term and Short-term Constraints
condition, but also the average transmit power. Intuitively, even This section reviews the optimal power allocation schemes
having enough power budget for transmission and the CR link and the ergodic capacity of the secondary user under the
has a very good condition, the secondary user may not able to combined long-term and short-term constraints, which is pretty
transmit if the interference channel in a very good condition. different from the long-term constraints cases [53]. The opti-
To solve this problem (O2 , C12 , C22 ) we used ellipsoid method mization problem may be formulated as
[51], [52], shown in Table II.
gss ps (gss , gps )
We show the simulation results in the Figure 6 and in Figure O4 : maximize E log 1 + (23)
ps (gss ,gps )≥0 N1 B
7, where we illustrate the ergodic capacity of the SU using
bits/s/Hz instead of nats/s/Hz only for a purpose of comparison C14 : ps (gss , gps )gps ≤ Qpk (24)
with the original results shown in [20]. We can see from the C24 : ps (gss , gps ) ≤ Ppk (25)
figures that at low Pav case the ergodic capacity is mainly C34 : E {ps (gss , gps )gps } ≤ Qav (26)
affected by the average transmit power constraint, in other
C44 : E {ps (gss , gps )} ≤ Pav (27)
words, ATP dominates AIP. On the other hand, at high Pav
regime, AIP dominates ATP. where (24) represents the peak interference power (PIP)
4) AIP constraint with imperfect CSI and receive MRC: constraint, (25) is the peak transmit power (PTP) constraint
The optimal power allocation schemes and the associated indicating the maximal transmit power of the SU, (26) and
ergodic capacity of the SU with receive MRC and imperfect (27) are the average interference power (AIP) constraint and
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 19
TABLE II
E LLIPSOID M ETHOD : P SEUDOCODE .
bisection method.
1) PIP and AIP Constraints: Under the PIP and AIP 0
10
3 1.8
Ppk=0dB
Ppk=5dB 1.6
Q = 5 dB
Ergodic Capacity of the SU (nats/sec/Hz)
2 Ppk=10dB 1.2
P =0dB, no interf. constr.
pk
no transmit power constr. 1
1.5
Q = 0 dB
0.8 pk
1 0.6
0.4 Q = −5 dB
pk
0.5
0.2 Without ATP
With ATP
0 0
−10 −8 −6 −4 −2 0 2 4 6 8 10 −10 −5 0 5 10 15
Qav (dB) Pav (dB)
Fig. 9. Ergodic capacity of the SU under different values of PTP and AIP Fig. 10. Ergodic capacity of the SU under different values of ATP and PIP
constraints over Rayleigh fading. Constraints over Rayleigh fading.
some opportunities to transmit. This corresponds to the AIP- schemes are combined channel inversion and water filling or
dominant regime. On the other hand, if the PTP is dominant, two-dimensional water filling. As we know that in the analysis
the ergodic capacity is unbounded by the ergodic capacity with of ergodic capacity, the delay limit is not considered which
the ATP constraint. This is because that the SU has a lot of means that it can be approaching to infinity. In the successive
chances to transmit, but the PTP limits the ergodic capacity. section, we review the effective capacity which takes the delay
3) ATP and PIP Constraints: Under the average transmit into consideration.
power constraint and peak interference power constraint, the
optimization problem is formulated by (O4 , C14 , C44 ) in (23),
IV. E FFECTIVE C APACITY
(24), and (27). The resultant optimal power allocation for the
secondary user holds in [20] as The concept of effective capacity (EC) has been reviewed
Qpk Q in subsection II-B. This section reviews some results of
, gps ≥ 1 pk N1 B , gss > λN1 B
optimal power allocation strategies and effective capacity of
gps λ − gss
p∗s (gss , gps ) = λ1 − Ng1ssB , gps < 1 QpkN1 B , gss > λN1 B
the secondary user over block fading channels. For simplicity,
λ − gss we in this section omit the parameters for ps (θ, gss , gps ), i.e.
0, gss ≤ λN1 B the transmit power of the SU is denoted as ps . The objective
(31) function and possible constraints are listed in the following,
It is intuitive that the power allocation scheme is a combination
of channel inverse and waterfilling. The waterfilling reflects 1 −θT B log 1+ pNs gss
O5 : maximize − log E e 1B (32)
the average transmit power constraint and the channel inverse ps (θ,gss ,gps )≥0 θ
reflects the peak interference power constraint. The simulation Constraints:
results are shownin Figure 10. The ergodic capacity of the SU C15 : ps gps ≤ Qpk (33)
Qpk
is capped by log 1 + gps N1 B gss . In the low-ATP regime, the
C25 : ps ≤ Ppk (34)
ergodic capacity is dominated by the ATP constraint, while in
the high-ATP regime the ergodic capacity is limited by the PIP. C35 : E {ps gps } ≤ Qav (35)
These can be explained as follows: In the low-ATP regime, the C45 : E {ps } ≤ Pav (36)
power allocation scheme is mainly the water-filling, and in in
the high-ATP regime the power allocation scheme is performed where T denotes the block length duration, B is the channel
as the channel inversion. bandwidth, θ is the delay exponent, Ppk denotes the maximum
allowed peak transmit power, Pav denotes the average transmit
power constraint, and Qpk and Qav represent the peak and
D. Summary average interference power threshold, respectively. We can
We can see that the strategies how the secondary user use Lagrangian method to solve the optimization problems
allocates the optimal transmit power to maximize the ergodic with different combinations of constraints. Without loss of
capacity are decided by the types of constraints. When the generality we assume that T B = 1 and N1 B = 1 in following
constraint is the peak interference constraint, the OPA is the simulations.
channel inversion with respect to the interference channel
from the secondary transmitter to the primary receiver (ST-
PR). If the constraint is the average interference constraint, A. Short-term Constraint
the OPA is the water-filling scheme, where the water level is The same as in the previous sections that the short-term
decided by the interference channel, ST-PR, power gain. For constraints include the peak transmit power constraint and the
the combined long-term and/or short-term constraints, the OPA peak interference power constraint.
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 21
1) PTP and PIP Constraints: The optimization problem is power constraint becomes to dominate. In the two figures, we
given by (O5 , C15 , C35 ) in (32), (33), and (34). The power also show the upper bounds given by Eq. (39), i.e. no peak
allocation strategy is straightforward
obtained that the SU transmit power constraint, for Qpk = −5dB and Qpk = 5dB.
Qpk
transmits using the power of min Ppk , gps . Then the ef-
fective capacity can be obtained as
−θT B 0
10
ρ=0.001
Qpk ρ=0.01
1 min P ,
pk gps g ss ρ=0.1
h
f (h) = 2 (38)
h+h
ρ=0.001
( " −θT B #)
0
10 ρ=0.01
ρ=0.1
1 Qpk gss
Effective capacity of the SU (nats/s/Hz)
EC ub = − log E 1 + ρ=1
ρ=10
θ gps N1 B ρ=100
ECup, Q = 5dB
−1
pk
" −θT B #
10 ECup, Qpk = −5dB
1 Qpk h
= − log E 1 +
θ
N1 B
}
| {z −2
C1 10
where
Z ∞ −θT B
Qpk h h
C1 =
−2 −1 0 1 2
1+ 2 dh 10 10 10
θ
10 10
0 N 1 B h+h
Qh Fig. 12. Effective capacity of the SU under different values of θ over Rayleigh
= B(1, 1 + θT B) 2 F1 θT B, 1; θT B + 2, 1 − fading with Ppk = 5dB, where ρ = Qpk /Ppk .
N1 B
where, in the last step, we have used [44, 3.197-1],
2 F1 (a, b; c, d) is hypergeometric function [44, 9.14], and
B(a, b) denotes the beta function [44, 8.38]. B. Long-term Constraints
Then we have In the following, we review the optimal power allocation
1 strategies and the simulation results of the effective capacity
EC ub = − log {B(1, 1 + θT B) of the SU under long-term constraints.
θ (39) 1) AIP Constraint: Under average interference power con-
Qh
× 2 F1 θT B, 1; θT B + 2, 1 − straint and secondary QoS constraint, the optimization problem
N1 B
is given by (O5 , C35 ) in (32) and (34). The resultant optimal
Figures 11 and 12 illustrate the effective capacity of the power allocation for the secondary user to maximize effective
SU versus different values of the delay component along with capacity holds in [38] by
the different ratios of PIP and PTP over Rayleigh fading. In " #+
the simulation we assume that the mean value of the channel
1
β 1+α
N1 B − g1ss , gps ≤ βgss
power gains are 1, and the AWGN power at the receiver is p∗s = 1
1+α 1+α
gps gss
α
(40)
1. First, it is intuitive that when the value of Qpk decreases, 0, otherwise
i.e. ρ decreases, the effective capacity of the SU decreases.
Second, when the value of ρ is bigger than 1 , for instance, where [x]+ = max(0, x), α = θT B, β = λNα1 B , and λ is the
ρ = 1 and ρ = 100, the effective capacity of the SU increases non-negative Lagrangian variable associated with the average
slowly and will converge. This is because the peak transmit interference power constraint. Based on the above optimal
22 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
1.8 1.4
Qav=−5dB, Pav=−5dB
θ=0.001
1.6 θ=0.01 Qav=5dB, Pav=−5dB
1.2
θ=0.1 Q =−5dB, P =5dB
Effective capacity of the SU (nats/s/Hz)
1.2 θ=20
θ=50
1 θ=100 0.8
0.8 0.6
0.6
0.4
0.4
0.2
0.2
0 0 −2 −1 0 1
−5 −4 −3 −2 −1 0 1 2 3 4 5 10 10 10 10
Qav (in dB) θ
Fig. 13. Effective capacity of the SU under different values of θ over Rayleigh Fig. 14. Effective capacity of the SU over Rayleigh fading.
fading.
1.4 1.4
Qpk= −5dB, Pav= −5dB Qav= −5dB, Ppk= −5dB
Qpk= −5dB, Pav= 5dB Qav= −5dB, Ppk= 5dB
1.2 1.2
Q = 5dB, P = −5dB Q = 5dB, P = −5dB
Effective capacity of the SU (nats/s/Hz)
0.8 0.8
Pav= 5dB
0.6 0.6
0.4 0.4
Fig. 15. Effective capacity of the SU over Rayleigh fading under PIP and Fig. 16. Effective capacity of the SU over Rayleigh fading under AIP and
ATP constraints. PTP constraints.
and average interference power constraints, and delay con- power constraint is applied, i.e. short-term constraint.
straint, the optimization problem is given by (O5 , C25 , C35 ) in • Two-dimensional waterfilling: when the average transmit
(32), (34) and (35). Using Lagrangian method, the resultant and/or interference power constraints are applied, i.e.
optimal power allocation for the secondary user to maximize long-term constraints.
effective capacity is obtained as • Capped two-dimensional waterfilling: when the aver-
" #+ age/peak transmit power constraint and peak/average in-
1 terference power constraint are considered, i.e. combined
min P , N B β 1+α
− 1 g
, gps ≤β
pk 1
p∗s =
1 α
1+α 1+α
gps gss
gss ss long- and short-constraints.
The ergodic capacity is mainly influenced by the interfer-
0, otherwise
ence channel from the secondary transmitter to the primary
(43)
receiver. In addition, the effective capacity is affected by the
where α = θT B, β = λNα1 B , and λ is the non-negative
delay component besides the interference channel. Especially,
Lagrangian variable associated with the average interference
the short-term constraints, i.e. peak transmit power and peak
power constraint. This power allocation scheme is capped by
interference power constraints, have different influences on
the peak transmit power. Thus the water level of the water
the effective capacity over lower and higher delay component
filling algorithm is different from the AIP-only case and the
regimes.
ATP-PIP scenario that the water level is changing from block
to block. To the best of our knowledge, there is no closed-form
expression for the effective capacity for this case. In Figure 16 R EFERENCES
we show the simulations for i.i.d. Rayleigh fading channels. [1] N. Hoven and A. Sahai, “Power scaling for cognitive radio,” in Proc.
From Figure 16 we can see that when the average inter- Int’l Conf. Wireless Netw., Commun. and Mobile Computing, Vol. 1,
ference power constraint is low, Qav = −5dB, higher peak pp. 250–255, Jun. 2005.
[2] M. Chiang, P. Hande, T. Lan, and C. W. Tan, “Power control in wireless
transmit power threshold is not an advantage at the higher cellular networks,” Foundations and Trends
inR Networking, Vol. 2,
Ppk regime. In addition, the effective capacity of the case of No. 4, pp. 381–533, 2008.
Qav = 5dB and Pav = −5dB is supreme over the case of [3] O. Kaya and S. Ulukus, “Optimum power control for cdma with
deterministic sequences in fading channels,” IEEE Trans. Inf. Theory,
Qav = −5dB and Pav = 5dB at the stringent delay regime. Vol. 50, No. 10, pp. 2449–2462, Oct. 2004.
This can be explained as that when θ is large, the secondary [4] R. Knopp and P. Humblet, “Information capacity and power control
user needs to maintain the rate as constant as possible, and in single-cell multiuser communications,” in Proc. IEEE ICC, Vol. 1,
pp. 331–335, Jun. 1995.
in lower Qav and higher Ppk case the SU has much less [5] R. Yates, “A framework for uplink power control in cellular radio
opportunities to transmit than the case of higher Qav and systems,” IEEE J. Sel. Areas Commun., Vol. 13, No. 7, pp. 1341–1347,
lower Ppk . Moreover, when the PTP constraint is lower and Sep. 1995.
[6] A. Goldsmith and P. Varaiya, “Capacity of fading channels with channel
not bigger than Qav , Ppk = −5dB, the AIP constraint can be side information,” IEEE Trans. Inf. Theory, Vol. 43, No. 6, pp. 1986–
ignored. 1992, Nov. 1997.
24 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
[7] A. Goldsmith, S. A. Jafar, I. Maric, and S. Srinivasa, “Breaking spectrum [31] D. Tse and S. Hanly, “Multiaccess fading channels. I. polymatroid
gridlock with cognitive radios: An information theoretic perspective,” structure, optimal resource allocation and throughput capacities,” IEEE
Proc. IEEE, Vol. 97, No. 5, pp. 894–914, May 2009. Trans. Inf. Theory, Vol. 44, No. 7, pp. 2796–2815, Nov. 1998.
[8] S. Haykin, “Cognitive radio: brain-empowered wireless communica- [32] D. Wu and R. Negi, “Effective capacity: a wireless link model for
tions,” IEEE J. Sel. Areas Commun., Vol. 23, No. 2, pp. 201–220, support of quality of service,” IEEE Trans. Wireless Commun., Vol. 2,
Feb. 2005. No. 4, pp. 630–643, Jul. 2003.
[9] T. M. Cover and J. A. Thomas, Elements of Information Theory, 2nd ed. [33] ——, “Downlink scheduling in a cellular network for quality-of-service
John Wiley & Sons, Inc., 2006. assurance,” IEEE Trans. Veh. Technol., Vol. 53, No. 5, pp. 1547–1557,
[10] Y. Chen, G. Yu, Z. Zhang, H. hwa Chen, and P. Qiu, “On cognitive radio Sep. 2004.
networks with opportunistic power control strategies in fading channels,” [34] C.-S. Chang and J. Thomas, “Effective bandwidth in high-speed digital
IEEE Trans. Wireless Commun., Vol. 7, No. 7, pp. 2752–2761, Jul. 2008. networks,” IEEE J. Sel. Areas Commun., Vol. 13, No. 6, pp. 1091–1100,
[11] S. Srinivasa and S. A. Jafar, “Soft sensing and optimal power control Aug. 1995.
for cognitive radio,” IEEE Trans. Wireless Commun., Vol. 9, No. 12, [35] J. Tang and X. Zhang, “Quality-of-service driven power and rate
pp. 3638–3649, Dec. 2010. adaptation over wireless links,” IEEE Trans. Wireless Commun., Vol. 6,
[12] H. Islam, Y. chang Liang, and A. Hoang, “Joint power control and beam- No. 8, pp. 3058–3068, Aug. 2007.
forming for cognitive radio networks,” IEEE Trans. Wireless Commun., [36] ——, “Quality-of-service driven power and rate adaptation for mul-
Vol. 7, No. 7, pp. 2415–2419, Jul. 2008. tichannel communications over wireless links,” IEEE Trans. Wireless
[13] L. Qian, X. Li, J. Attia, and Z. Gajic, “Power control for cognitive Commun., Vol. 6, No. 12, pp. 4349–4360, Dec. 2007.
radio ad hoc networks,” in Proc. IEEE LANMAN Workshop, pp. 7–12, [37] S. Akin and M. Gursoy, “Effective capacity analysis of cognitive radio
Jun. 2007. channels for quality of service provisioning,” IEEE Trans. Wireless
[14] K. Koufos, K. Ruttik, and R. Jäntti, “Controlling the interference from Commun., Vol. 9, No. 11, pp. 3354–3364, Nov. 2010.
multiple secondary systems at the tv cell border,” in PIMRC, pp. 645– [38] L. Musavian and S. Aissa, “Effective capacity of delay-constrained
649, 2011. cognitive radio in Nakagami fading channels,” IEEE Trans. Wireless
[15] M. Gastpar, “On capacity under received-signal constraints,” in Commun., Vol. 9, No. 3, pp. 1054–1062, Mar. 2010.
Proc.42nd Annual Allerton Conf. on Commun., Control and Comp., [39] M. Gursoy, D. Qiao, and S. Velipasalar, “Analysis of energy efficiency in
(Monticello, IL, USA), Sept. 29 - Oct. 1 2004. fading channels under qos constraints,” IEEE Trans. Wireless Commun.,
[16] ——, “On capacity under receive and spatial spectrum-sharing con- Vol. 8, No. 8, pp. 4252–4263, Aug. 2009.
straints,” IEEE Trans. Inf. Theory, Vol. 53, No. 2, pp. 471–487, [40] C.-X. Wang, X. Hong, H.-H. Chen, and J. Thompson, “On capacity
Feb. 2007. of cognitive radio networks with average interference power con-
[17] A. Ghasemi and E. S. Sousa, “Fundamental limits of spectrum-sharing straints,” IEEE Trans. Wireless Commun., Vol. 8, No. 4, pp. 1620–1625,
in fading environments,” IEEE Trans. Wireless Commun., Vol. 6, No. 2, Apr. 2009.
pp. 649–658, Feb. 2007. [41] J. Peha, “Sharing spectrum through spectrum policy reform and cogni-
tive radio,” Proc. IEEE, Vol. 97, No. 4, pp. 708–719, Apr. 2009.
[18] H. Suraweera, P. Smith, and M. Shafi, “Capacity limits and performance
[42] S. Lim, H. Wang, H. Kim, and D. Hong, “Mean value-based power
analysis of cognitive radio with imperfect channel knowledge,” IEEE
allocation without instantaneous CSI feedback in spectrum sharing
Trans. Veh. Technol., Vol. 59, No. 4, pp. 1811–1822, May 2010.
systems,” IEEE Trans. Wireless Commun., Vol. 11, No. 3, pp. 874–879,
[19] L. Musavian and S. Aissa, “Fundamental capacity limits of cognitive
Mar. 2012.
radio in fading environments with imperfect channel information,” IEEE
[43] Z. Rezki and M. Alouini, “Ergodic capacity of cognitive radio under
Trans. Commun., Vol. 57, No. 11, pp. 3472–3480, Nov. 2009.
imperfect channel-state information,” IEEE Trans. Veh. Technol., Vol. 61,
[20] X. Kang, Y.-C. Liang, A. Nallanathan, H. Garg, and R. Zhang, “Opti- No. 5, pp. 2108–2119, Jun. 2012.
mal power allocation for fading channels in cognitive radio networks: [44] I. Gradshteyn and I. Ryzhik, Table of Integrals, Series, and Products,
Ergodic capacity and outage capacity,” IEEE Trans. Wireless Commun., Seventh Edition, 7th ed. Academic Press, 2007.
Vol. 8, No. 2, pp. 940–950, Feb. 2009. [45] R. Zhang, “Optimal power control over fading cognitive radio channel by
[21] R. Zhang, “On peak versus average interference power constraints for exploiting primary user CSI,” in Proc. IEEE GLOBECOM, Nov. 2008.
protecting primary users in cognitive radio networks,” IEEE Trans. [46] D. Brennan, “Linear diversity combining techniques,” Proc. IEEE,
Wireless Commun., Vol. 8, No. 4, pp. 2112–2120, Apr. 2009. Vol. 91, No. 2, pp. 331–356, Feb. 2003.
[22] L. Ozarow, S. Shamai, and A. Wyner, “Information theoretic consider- [47] M.-S. Alouini and A. Goldsmith, “Capacity of rayleigh fading channels
ations for cellular mobile radio,” IEEE Trans. Veh. Technol., Vol. 43, under different adaptive transmission and diversity-combining tech-
No. 2, pp. 359–378, May 1994. niques,” IEEE Trans. Veh. Technol., Vol. 48, No. 4, pp. 1165–1181,
[23] E. Biglieri, J. Proakis, and S. Shamai, “Fading channels: information- Jul. 1999.
theoretic and communications aspects,” IEEE Trans. Inf. Theory, Vol. 44, [48] I. E. Telatar, “Capacity of multi-antenna gaussian channels,” Eur. Trans.
No. 6, pp. 2619–2692, Oct. 1998. Telecomm., Vol. 10, No. 6, pp. 585–595, 1999.
[24] H. Ding, J. Ge, D. Benevides da Costa, and Z. Jiang, “Asymptotic [49] R. Duan, M. Elmusrati, R. Jäntti, and R. Virrankoski, “Capacity for
analysis of cooperative diversity systems with relay selection in a spectrum sharing cognitive radios with MRC diversity at the secondary
spectrum-sharing scenario,” IEEE Trans. Veh. Technol., Vol. 60, No. 2, receiver under asymmetric fading,” in Proc. IEEE Global Telecommu-
pp. 457–472, Feb. 2011. nications Conference, pp. 1–5, Dec. 2010.
[25] J. Lee, H. Wang, J. G. Andrews, and D. Hong, “Outage probability [50] R. Duan, R. Jäntti, M. Elmusrati, and R. Virrankoski, “Capacity for
of cognitive relay networks with interference constraints,” IEEE Trans. spectrum sharing cognitive radios with MRC diversity and imperfect
Wireless Commun., Vol. 10, No. 2, pp. 390–395, Feb. 2011. channel information from primary user,” in Proc. IEEE Global Telecom-
[26] C. Zhong, T. Ratnarajah, and K.-K. Wong, “Outage analysis of decode- munications Conference, pp. 1–5, Dec. 2010.
and-forward cognitive dual-hop systems with the interference constraint [51] R. G. Bland, D. Goldfarb, and M. J. Todd, “The ellipsoid method: A
in nakagami- fading channels,” IEEE Transactions on Vehicular Tech- survey,” Operations Research, Vol. 29, No. 6, pp. 1039–1091, Dec. 1981,
nology, Vol. 60, No. 6, pp. 2875 –2879, Jul. 2011. http://www.jstor.org/stable/170362.
[27] T. Duong, P. L. Yeoh, V. N. Q. Bao, M. Elkashlan, and N. Yang, [52] S. Boyd and L. Vandenberghe, Convex Optimization. New York, NY,
“Cognitive relay networks with multiple primary transceivers under USA: Cambridge University Press, 2004.
spectrum-sharing,” IEEE Signal Process. Lett., Vol. 19, No. 11, pp. 741– [53] M. Khojastepour and B. Aazhang, “The capacity of average and peak
744, 2012. power constrained fading channels with channel side information,” in
[28] K. J. Kim, T. Duong, H. Poor, and L. Shu, “Performance analysis of Proc. IEEE WCNC, Vol. 1, pp. 77–82, Mar. 2004.
cyclic prefixed single-carrier spectrum sharing relay systems in primary [54] A. Papoulis and S. U. Pillai, Probability, Random Variables and Stochas-
user interference,” IEEE Trans. Signal Processing, Vol. 60, No. 12, tic Processes, 4th ed. McGraw-Hill, 2002.
pp. 6729–6734, Dec. 2012.
[29] R. Etkin, A. Parekh, and D. Tse, “Spectrum sharing for unlicensed
bands,” IEEE J. Sel. Areas Commun., Vol. 25, No. 3, pp. 517–528,
Apr. 2007.
[30] L. Musavian and S. Aissa, “Capacity and power allocation for spectrum-
sharing communications in fading channels,” IEEE Trans. Wireless
Commun., Vol. 8, No. 1, pp. 148–156, Jan. 2009.
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 25
Abstract—We investigate the effective capacity of a cognitive- studied in the literature for traditional wireless communication
shared channel with implementing transmit antenna selection at systems, e.g. [8]–[10]. In [8], [9], the authors investigated the
the secondary transmitter and maximal ratio combining at the outage probability and bit error rate of TAS/MRC in Rayleigh
secondary receiver under different transmit antenna selection
schemes, minimum interference selection, maximum secondary and Nakagami-m fading channels, respectively. In [10], the
composite channel gain selection, and the maximum channel ratio authors proposed the expressions of the outage probability of
selection. Closed-form expressions for the effective capacity are multiuser diversity for a TAS/MRC system in independent
presented and validated through simulations. and identically distributed Nakagami-m channels. The full
Index Terms—Cognitive radio, generalized selection combin- diversity is achieved at high SNR regime.
ing, ergodic capacity, symbol error probability. Recently, the authors in [11] studied the ergodic capacity
of a spectrum sharing secondary user link with TAS/MRC, in
I. I NTRODUCTION the Rayleigh fading environment. The closed-form expression
of the ergodic capacity of the secondary user with peak
(i)
corresponding p.d.f. and c.d.f. of gss are given as follows [21]:
hL−1 exp (−h/gss )
fg(i) (h) = , h ≥ 0, 1 ≤ i ≤ M (1)
ss gL
ss (L − 1)!
1 h
Fg(i) (h)= γ L,
ss (L − 1)! gss
L−1 l
h X1 h
= 1 − exp − (2)
gss l! gss
l=0
and Using order theorem [23], the c.d.f. of the maximum channel
(k)
gain, gss , can be represented as
M−1
g exp −g/gps Fg(k) (x)
fg(k) (g)= M exp − ss
ps gps g ps M
Y M
Y
M 1 x
M g = Fhl (x) = γ L, (12)
= exp − (6) (L − 1)! gss
l=1 l=1
g ps gps " #
YM
− gx
X 1 x kl
L−1
= 1 − e ss
kl ! g ss
Now we derive the probability distribution of the ratio, l=1 kl =0
" #nl
(k) (k) (k) (k) (k)
gss /gps . Let z = gss /gps and w = gps . The Jacobian X Y M
− gx
X 1 x kl
L−1
nl
is given by = (−1) e ss
kl ! g ss
n∈θM l=1 kl =0
M
" L−1 kl #nl
w z X Y x X 1 x
Jz,w = det =w (7) = (−1)nl e− gss nl
0 1 kl ! g ss
n∈θM l=1 kl =0
X − x PM n Y M X Lnl −1 x kl nl
L−1
l=1 l nl
The joint p.d.f. of z and w. = e gss
(−1)
(kl !)nl g ss
n∈θM l=1 kl =0
X L−1
X L−1
X − gx
PM
nl
PM
kl nl
fz,w = fg(k) ,g(k) (h = zw, g = w) w (8) = ··· e ss l=1
x l=1
ss ps
M n∈θM k1 =0 kM =0
M w (zw)L−1 e−zw/gss M nl
= exp − w Y −1 Lnl −1
g ps gps gL
ss (L − 1)! ×
l=1
kl ! gkssl nl
The c.d.f. of z. The above proof follows the Lemma in [25] and [26]. The
p.d.f. can be obtained as following
Z z Z z d
M LρxL−1 fg(k) (x) =
Fh (x)
Fz (z)= fz (x) dx = dx ss dx max
0 0 (x + M ρ)L+1
L X L−1
X L−1
X
z = ··· Kn,k e−xBn xAn,k −1 [An,k − xBn ]
= (10)
z + Mρ n∈θM k1 =0 kM =0
(16)
(k) (k) (k)
Let z = gss /gps and w = gps . The Jacobian is given by
w z
B. Maximum secondary composite channel gain selection Jz,w = det =w (17)
0 1
The transmit antenna which has maximal channel gain of The joint p.d.f. of z and w can be obtained by
ST-SR is selected, i.e., the channel from the selected antenna
fz,w = fg(k) ,g(k) (h = zw, g = w) w
to the secondary receiver has the maximal composite channel ss ps
L−1 L−1
gain among all antennas. 1 −w/gps X X X
= e ··· Kn,k e−zwBn
g ps
n o n∈θM k1 =0 kM =0
(i)
k = arg∀i max gss , 1 ≤ i ≤ M. (11) ×(zw)An,k −1 [An,k − zwBn ] w (18)
28 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
M=1, L=4
M=2, L=2
following theorem.
1
Theorem 2: The effective capacity of the cognitive-shared
0.8 channel of Figure 1 under peak interference power constraint
over Rayleigh fading with maximum secondary composite
0.6 channel gain selection is given by
( )
0.4
1 X L−1 X L−1
X Kn,k
EC(θ) = − log ··· A
[C1 + C2 ]
0.2
θ Bn n,k n∈θM k1 =0 kM =0
(29)
0 −2 −1 0 1 2
10 10 10 10 10
θ where C1 for Bn 6= 0 and C2 for Bn = 0 are given by
Fig. 2. Effective capacity of the SU versus QoS exponent under PIP constraint C1 = An,k Γ (An,k + 1) B(An,k , α + 1)
and minimum interference selection, where the mean channel power gain ratio
ρ = 1, N1 B = 1, and Qpk = −5dB. Qpk
× 2 F1 α, An,k ; An,k + α + 1; 1 −
N1 Bg ps Bn
(30)
over Rayleigh fading with minimum interference selection
scheme is given by and
1
EC(θ)= − log [L B(L, 1 + α) C2 = −Γ (An,k + 2) B(An,k + 1, α + 1)
θ
Qpk M ρ
× 2 F1 α, L; L + 1 + α; 1 − (28) Qpk
N1 B × 2 F1 α, An,k + 1; An,k + α + 2; 1 −
N1 Bg ps Bn
Proof: (31)
1 zQ
−α log 1+ N pk Proof: see Appendix A for details.
EC(θ)= − log Ezk e 1 B
θ
"Z −α #
∞
1 zQpk M Lρz L−1 1.4
= − log 1+ dz M=1, L=1
θ 0 N1 B (z + M ρ)L+1 M=2, L=1
M=4, L=1
1.2
M=1, L=2
1
Effective capacity of the SU (nats/s/Hz)
M=1, L=4
= − log [L B(L, 1 + α) M=2, L=2
θ
1
Qpk M ρ
× 2 F1 a, L; L + 1 + α; 1 − 0.8
N1 B
0.6
where in the last step, we have used Eqn. (3.197-1) in [24].
0.4
In Fig. 2 we show the effective capacity of the secondary
user when the SU chooses the transmit antenna element having 0.2
θ
degrees of diversity. Without loss of generality, N1 B = 1.
First, it is obvious that the TAS/MRC improves the achievable
Fig. 3. Effective capacity of the SU versus QoS exponent under PIP constraint
effective capacity over a large range of the QoS exponent. and maximum MRC channel gain, where the mean channel power gain ratio
Second, the performance of using multiple receive antennas ρ = 1, N1 B = 1, and Qpk = −5dB.
surpasses the one using multiple transmit antennas, e.g. the
performance of M = 1, L = 2 is better than the one of In Fig. 3 we show the effective capacity of the secondary
M = 2, L = 1, and M = 1, L = 4 is better than the one user when the SU chooses the transmit antenna element having
of M = 4, L = 1. Third, in this case using M = 2 and the maximum ST-SR channel gain over different transmit
L = 2 can not achieve the full diversity, because we only and receive degrees of diversity. First, it is obvious that the
utilize partial channel information, i.e. only the interference TAS/MRC improves the achievable effective capacity over a
channel information. Forth, at high θ regime, M = 1, L = 2 large range of the QoS exponent. Second, in this case using
is slightly better than M = 4, L = 1. This is because the more receiving antennas always (over the simulated range of θ)
MRC is the optimal linear combing technique [22], and the is superior using more transmit antennas given the same value
benefit of using TAS is not able to compensate the capacity of M × L. This is different from the case of Sel (1) scheme at
loss caused by the channel deep fading. the high θ regime. Third, the larger value of M × L, the better
30 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
M=1, L=4 N1 B
1
M=2, L=2 maximize − E e (33)
θ
0.8
subject to
h i
(k) (k) (k)
0.6 E gps ps (θ, gps , gss ) ≤ Qav (34)
(k) (k)
0.4 ps (θ, gps , gss ) ≥0 (35)
Fig. 4. Effective capacity of the SU versus QoS exponent under PIP constraint subject to (34), (35)
and maximum channel radio selection, where the mean channel power gain
ratio ρ = 1, N1 B = 1, and Qpk = −5dB. The power allocation can be obtained through using La-
grangian method as
+
1
performance the SU has. However, the performance of the β 1+α 1
, gss ) = N1 B
(k) (k)
ps (θ, gps − (k) (37)
scenario of M = 1, L = 4 surpasses the one of M = 2, L = 2. α 1
(k) 1+α (k) 1+α gss
This is because the former case can use the full diversity while gss gps
the later does not according to the antenna selection method. where α = θT B, β = λNα1 B , and λ is the Lagrangian
multiplier associated to (34).
Fig. 5. Effective capacity of the SU versus QoS exponent under AIP constraint 1.2 M=4, L=1
and minimum interference selection, where the mean channel power gain ratio M=1, L=2
ρ = 1, N1 B = 1, and Qav = −5dB. 1
M=1, L=4
M=2, L=2
(41) 0.2
0.4
Proof:
0.3
h i
(k) (k) (k)
Qav = E gps ps (θ, gps , gss ) 0.2
Z ∞ " 1 #+ 0.1
β 1+α 1
= N1 B α − fzk (z) dz 0 −2
0 z 1+α z 10 10
−1 0
10 10
1
10
2
" 1 # θ
Z ∞
β 1+α 1 M Lρz ML−1
= N1 B α − dz Fig. 8. Effective capacity comparison under PIP constraint and different TAS
1/β z 1+α z (z + ρ)ML+1 schemes, where, M × L = 2, the mean channel power gain ratio ρ = 1, and
" Z ∞ ML− 1+α α Qpk = −5dB.
−1
1 z
= N1 BM Lρ β 1+α
ML+1
dz
1/β (z + ρ)
Z ∞ #
z ML−2 1.4
− ML+1
dz
1/β (z + ρ)
Effective capacity of the SU (nats/s/Hz)
1.2
θ
versus θ under peak interference power (PIP) constraint with
diversity order M × L = 2 and M × L = 4, respectively.
Fig. 9. Effective capacity comparison under PIP constraint and different TAS
From Fig. 8 we can see that Sel (1) scheme is superior to Sel schemes, where, M × L = 4, the mean channel power gain ratio ρ = 1, and
(2) at the low range of delay component θ which represents Qpk = −5dB.
delay-insensitive regime. However, along with increasing the
value of θ, Sel (2) becomes better than Sel (1). Moreover,
The effective capacity is capped by using full diversity order. conclude that at the stringent case, i.e. lager values of θ, with
Fig. 9 shows the similar phenomenon as in Fig. 8. We may the same diversity order the receiving diversity is superior to
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 33
1 1.4
M=1, L=1
0.9 Sel (1), M=2, L=1
Effective capacity of the SU (nats/s/Hz)
Fig. 10. Effective capacity comparison under AIP constraint and different Fig. 11. Effective capacity comparison under AIP constraint and different
TAS schemes, where, M × L = 2, the mean channel power gain ratio ρ = 1, TAS schemes, where, M × L = 4, the mean channel power gain ratio ρ = 1,
and Qav = −5dB. and Qav = −5dB.
EC(θ)
"Z
EC(θ) 1 ∞ h 1 1
i+ −α X
−→
"Z −α = − log 1+ β 1+α z 1+α −1 Kn,k g An,k
1 ∞
Qpk 1 X−→ K θ 0
ps
n,k
= − log 1+ z
θ N1 B gps An,k +1
0 Bn
An,k Γ(An,k + 1)z An,k −1 Bn g ps Γ(An,k + 2)z An,k
× A +1 − A +2 dz
An,k Γ(An,k + 1)z An,k −1 Γ(An,k + 2)z An,k
× 1 + Bn g ps z n,k 1 + Bn g ps z n,k
An,k +1 − An,k +2 dz | {z }
1 1
z + Bn g z + Bn g
G
ps ps
" Z ∞ −α
1 (Qpk /N1 B)
−α X−→ Kn,k N1 B
X −→ Z 1/β
= − log + z 1
A +1 = − log Kn,k g ps An,k G dz
θ g ps Bn n,k 0 Qpk θ
0
| {z }
I1
An,k Γ(An,k + 1)z An,k −1 Γ(An,k + 2)z An,k
× An,k +1 − An,k +2 dz Z ∞
z + Bn1g z + Bn1g
α
− 1+α α
− 1+α
+ β z G dz
ps ps
1/β
| {z }
I2
then we have the result in (29) by implying [24, Eqn. (3.197-
1)] to the above integration. where for An,k 6= 0
Γ(An,k + 1) g ps Bn
I1 = 2 F1 A n,k + 1, A n,k ; A n,k + 2; −
β An,k β
B. Proof of Theorem 3 Bn Γ(An,k + 1)g ps
−
β An,k +1
g ps Bn
× 2 F1 An,k + 2, An,k + 1; An,k + 2; −
1 Q
−α log 1+ N pk z β
− log Ezk e 1B
Γ(An,k + 1)β (1 + α)
θ I2 = A +1
"Z −α #
∞ g ps Bn n,k (1 + 2α)
1 Qpk z ML−1
= − log 1+ z M Lρ dz α α β
θ 0 N1 B (z + ρ)ML+1 × An,k 2 F1 An,k + 1, 1 + ;2 + ;−
1+α 1+α g ps Bn
1
= − log [M L B(M L, α + 1) α α β
θ − (An,k + 1) 2 F1 An,k + 2, 1 + ;2 + ;−
1+α 1+α g ps Bn
Qpk ρ
× 2 F1 α, M L; α + n + 1; 1 − and for An,k = 0
N1 B
Bn g g ps Bn
I1 = − 2 F1 2, 1; 2; −
where in the last step, we have used [24, Eqn. (3.197-1)]. β β
β (1 + α)
I2 = −
g ps Bn (1 + 2α)
C. Proof of Theorem 4 α α β
×2 F1 2, 1 + ;2 + ;−
1+α 1+α g ps Bn
in the above we have used [24, Eqn. (3.194-1) and (3.194-2)]
1 h i
− log Ezk e−α log(1+ps gss /N1 B )
(k)
and Γ(1 + x) = xΓ(x).
θ "Z
1 ∞ h 1 i+ −α
= − log 1+ β 1+α z
1
1+α −1 E. Proof of the expression of Qav given by (40)
θ 0 The proof is on the next page.
M Lρz L−1
× dz
(z + M ρ)L+1 F. Proof of Theorem 6
"Z
1/β
1 M Lρz L−1 The proof is on the next page.
= − log dz
θ 0 (z + M ρ)L+1
Z ∞ # R EFERENCES
α
− 1+α α
− 1+α M Lρz L−1
+ β z dz [1] D. Wu and R. Negi, “Effective capacity: a wireless link model for
1/β (z + M ρ)L+1 support of quality of service,” IEEE Trans. Wireless Commun., Vol. 2,
No. 4, pp. 630–643, Jul. 2003.
[2] C.-S. Chang and J. Thomas, “Effective bandwidth in high-speed digital
We then have the result (38) by using [24, Eqn. (3.194-1) and networks,” IEEE J. Sel. Areas Commun., Vol. 13, No. 6, pp. 1091–1100,
(3.194-2)] to the integrals. Aug. 1995.
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 35
Proof of theorem 6.
EC(θ)
" !# "Z #
1
(k)
ps gss
−α log 1+ N 1 ∞ h 1 1
i+ −α M Lρz ML−1
= − log Ezk e 1B
= − log 1 + β 1+α z 1+α − 1 × ML+1
dz
θ θ 0 (z + ρ)
"Z Z ∞ #
1/β
1 M Lρz ML−1 α
− 1+α α
− 1+α M Lρz ML−1
= − log dz + β z dz
θ 0 (z + ρ)ML+1 1/β (z + ρ)ML+1
1 ML α α 1
= − log × 2 F1 M L + 1, M L + ; ML + 1 + ;− +M Lρβ 2 F1 (M L + 1, 1; 2; −ρβ)
θ (ρβ) ML ML + α 1 + α 1 + α ρβ
1+α
We have used [24, Eqn. (3.194-1) and (3.194-2)] to the above integrals.
36 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
Abstract—In this study we aim to assess the optimal successive output terminals, where each input communicates through a
group decoder (OSGD) considering different scenarios and in a common medium with its respective outputs. Also, authors of
wide range of testing metrics. We investigate the OSGD in the [3] presented the interference channel as a model for studying
K−user interference channel (IC) and evaluate its capability
to mitigate interference. We inspect OSGD performance in networks with two or more (source-destination) pairs where
terms of its (ergodic and effective) capacity. Also, we evaluate the signals of the sources interfere with each other at the
its minimum required energy per bit, bit error rate (BER), destination.
and outage probability, under different quality of service (QoS)
constraints. This study considers both the spatially correlated
Now, we proceed to introduce one of the most crucial
and uncorrelated (Rayleigh and Rician) flat fading channels. principles of the interference channel systems, which is how to
In addition, it explores performance of the OSGD in different interpret the interference effect, traditional wireless transceiver
SNR and SIR environments, where we consider both the power- designers commonly view interference as an augmentation to
limited and bandwidth-limited regimes with various cross-talk the additive Gaussian noise. This assumption is not consis-
values, taking in account several transmit-receive multiple-input
multiple-output (MIMO) antenna configurations. The obtained
tently true; it is valid only if the employed detectors do not
numerical results in this work, show that OSGD technique ex- take into consideration the interference formation; however, in
hibits very efficient performance to cope with interference in the fact, practical systems have some level of knowledge about
investigated scenarios, proving by that its competency comparing interference; because the signals emitted by the interferers
to the most developed interference cancellation approaches.1 This belong to discrete constellations [4]. In the light of this
makes OSGD a favorable technique for interference reduction;
especially, if we bear in mind that OSGD is formed on receive-
fact and because of interference management is essential to
side processing only. This confirmed also if we contemplate attain higher spectral efficiency and thereupon higher peak
the reduction of its arithmetic complexity, as a result of its bit rates, researchers have evolved and enhanced abundant of
innate complexity controlling characteristic, comparing to the interference-aware mitigation techniques. One of those meth-
large computational complexity of the other optimal interference ods has lately grabbed much research concentration, known
cancellation schemes, such as, the maximum likelihood multi-
user detection (ML-MUD) or the other iterative interference
as (interference alignment) [5]–[7]. Interference alignment
alignment schemes. is fundamentally established on the concept of designing
transmitter and receiver to align interfering signals to each
Index Terms—Interference Mitigation, Optimal Successive,
Group Decoding, Interference Channels, MIMO, Interference other at receiver side [8]. More precisely, vector interference
Alignment, Effective Capacity. alignment divides the dimensions of the receiver observation
space to two subspaces, one of the subspaces is occupied
by the desired signal and all the undesired interference are
I. I NTRODUCTION
aligned to the other subspace. From the theoretical point of
He accelerated expansion of wireless communication is
T foreseen to carry on due to the quick move towards the
next generation technologies. This inspires researchers to make
view, interference alignment is claimed to establish optimality
to approach Shannon capacity of interference network at
high SNR [9]. However, the existing interference alignment
promising development to realize peak bit rates as high as 1 schemes in fact are facing very challenging problems when it
Gbit/s and more. Achievement of such rate is predicted to be comes to practical implementation. Here we state some of the
facilitated by the deployment of distributed broadband wire- challenges that considered as strain in practical implementa-
less communications (BWC) systems. This indicates that the tion of interference alignment algorithms:
essential importance in the prospect wireless communication
systems is to devise wireless transceivers with the capability • Generally, analytical solution for the interference align-
to communicate in a reliable manner in the existence of ment is difficult to obtain. The existing closed form
interferers [1]. Therefore; studying the interference channel solutions are only been founded for certain cases with
(IC) and its adopted technologies becomes an indispensable very limited number of users.
to keep up and to cope with the demands of such emerging • When the closed form expression exists, global channel
wireless networks. knowledge is required to obtain such a closed form
Therefore, as a starting point we define the interference solution, which is of course requires an overwhelming
channel as in [2] by the channel of multi pairs of input- feedback overhead. In practice, restricted feedback con-
ditions, can lead to imperfect channel state information at
1 Part of the results in this article has been published by Omar Abu-Ella, and
the transmitter (CSIT), which severely affects the system
Mohammed Elmusrati, on Optimal Successive Group Decoding to Mitigate
Interference in Wireless Systems. In Proc. 10th IEEE International Conference efficiency and performance.
on Distributed Computing in Sensor Systems (DCOSS ’14). • The required signaling dimension of interference align-
38 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
ment scheme grows exponentially with the total number different levels of complexity. They can be implemented with
of users in the system; consequently, a question rises up as low complexity as the conventional successive single-user
about the impracticality of implementing such a system. decoder to the high-complexity of the maximum likelihood
• Although, the distributed interference alignment algo- decoder. Therefore, by imposing a constraint on the decoder
rithms require only local channel knowledge at each node, complexity, we can adopt the decoder with an adequate
it is yet require an extensive computational complex- complexity that each receiver can sustain. In this stream of
ity due to the iterative manner of finding its optimum research, a constrained partial group decoding (CPGD) scheme
solution. In addition, the final values and the conver- was proposed in [15], [16]. In this technique, each receiver
gence speed of the kernel iterative algorithm, which the employs a constrained partial group decoder to decode its
distributed interference alignment schemes rely on to desired message conjointly with a part of the interference.
optimize their objective function, is very sensitive to the In other words, this decoder exploits the knowledge about
initialization conditions. the interference to determine which interfering signals (with
• To avoid feedback some of interference alignment a constraint on their group size) should jointly decoded along
schemes are build on the assumption of reciprocity of with the desired signal, while treating the remaining interfering
the wireless network which does not always hold. signals as Gaussian noise.
To overcome all the previously stated challenges facing in- The bottom line contribution of this work is to explore
terference alignment strategy, we have to look for (no or the efficiency of using the OSGD in the interference channel
very limited) feedback system with tolerable complexity using systems, by extensively inspecting its performance in terms of
non-iterative technique. We try to avoid feedback because achieved (ergodic and effective) capacity. In addition to that,
the exhibited overall achieved capacity of the feedback-based we evaluate its minimum required energy per bit, BER, and
regimes, for instance, in the case of interference alignment outage probability, under different quality of service (QoS)
schemes, can be misleading when it is compared to non- constraints. For generality purposes, this study considers both
feedback system capacity. In other words, feedback in some the spatially correlated and uncorrelated (Rayleigh and Rician)
cases can hurt the system more than helping it, by decreasing flat fading channels. Moreover, it investigates the operation of
the pure throughput of the system. Consequently, our goal is the OSGD in different SNR and SIR environments, where
to have a better strategy to mitigate interference with limited we consider both the power-limited and bandwidth-limited
feedback. So, we should avoid all interference mitigation pro- regimes with different cross-talk values, as well as consid-
cesses at the transmitter side, and consider only the approaches ering different transmit-receive multiple-input multiple-output
with the operations at the receiver side, at the same time, we (MIMO) antenna configurations. To make our results more
want to accomplish this task within a feasible computational sensible, we contrast the aforementioned performance with
complexity. those of other well known interference mitigation approaches,
Looking for another track of research to cope with inter- such as, maximum likelihood multi-user detection (ML-MUD)
ference, we encounter an alternative traditional methodology, and interference alignment technique.
which is based on designing decoders in presence of inter- The remainder of this work is organized as follows: Sec-
ference. This approach employs one of the two following tion II defines the used notation in this work. Followed by
techniques: successive interference cancelation for small SIR Section III describing the model of interference channel which
regimes [10], or, on the other hand, treating interference like applied in this study. A brief description of the addressed
noise for larger SIR regimes. This can be seen in power allo- concepts and interference mitigation techniques is presented
cation systems for frequency selective Gaussian interference in Section IV. Complexity issues related to the presented
channels [11], and in CDMA cellular communication system schemes are discussed in Section V. Numerical results of the
design [12]. However, there is still some range of SIR when different scenarios in terms of achieved ergodic and effective
both of these techniques are suffering from an error floor [13]. capacity, minimum required energy per bit, in addition to the
Motivated by the necessary need for an optimal interference bit error rate and outage probability performance evaluation
mitigation technique considers all ranges of interference levels, are demonstrated in Section VI. Finally, Section VII concludes
a novel interference mitigation scheme proposed in [14], where this study.
the authors assumed that each receiver uses a successive
group decoder (SGD) which is considered as an extension
II. N OTATION
of the conventional successive decoder, however, instead of
decoding only one user at each decoding stage, a subgroup Scalars represented in this study with lowercase italics, vec-
of users are jointly decoded. Considering an interference tors, and matrices denoted respectively by lowercase boldface
channel system where a fixed power allocated to all users and and uppercase boldface; superscripts ·T and ·H symbolize
assuming predetermined rates for all transmitters, the authors the transpose and Hermitian (complex-conjugate) transpose;
of [14] obtained the decoding strategy minimizes the outage [·]i,j stands for the (i, j)th element of a matrix; tr(A) implies
probability at every receiver and generates the optimal subset the trace ofPmatrix A; abs(·) designates the absolute value;
Nr PNt H
of interferers that must be decoded along with the desired user kHk2 = i=1
2
j=1 |[H]i,j | = tr(H H) is the squared
under an imposed complexity constraint. Frobenius norm of H; E{·} expresses the statistical average; ∼
It is worth to mention that successive group decoders have denotes the distribution equivalence between the left and right
a substantial feature, that they can be implemented with random variables; ≈ means approximately equals; , means
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 39
if it presumes that will assist the decoding of its intended Hr = Rr2 Hw Rt2 (4)
user. The received signal of the kth receiver at the nth
where Hw is an (Nr × Nt ) normalized complex Gaussian
symbol interval through a K-user interference channel can be
random matrix, Rt , Rr , are the deterministic correlation ma-
expressed by
trices at the transmit and receive end respectively. Because of
√ √ K
X its simplicity and relative accuracy to simulate the realistic
yk [n] = P Hkk xk [n] + αP Hki xi [n] + zk [n], antenna inter-element correlation, we adopted the exponential
i=1,i6=k correlation model to generate the transmit and receive corre-
1≤k≤K (1) lation matrices, where they can be constructed using a single
correlation factor ρ ∈ C and |ρ| ≤ 1 as follows
where yk [n] is the Nrk × 1 received signal vector. further, abs(j−i)
in this system we assume a quasi static flat fading scenario, ρ ,i 6 j
Rij = ∗ (5)
where the channel gain experienced by the kth receiver from ρabs(j−i) ,i > j
the ith transmitter is described by the Nrk × Nti matrix Hki
consists as in [17], [18] of two components, deterministic (i.e., In this work, Hki is assumed to be perfectly known to the
mean) denoted by Hd and random component represented by kth receiver, but it is unknown to any of the transmitters and
Hr r r any other receiver. zk [n] is the Ntk × 1 complex additive
K 1 white Gaussian noise (AWGN) vector at the receiver k. It
H= Hd + Hr (2) is assumed to have independent and identically distributed
1+K 1+K
q (i.i.d.)∼ CN (0, 1) elements and it is temporally uncorrelated.
K
where 1+K is the LOS component of the channel and P represents the average transmit power used by the ith user.
q
1 α denotes the cross-talk factor between users, i.e., it represents
1+K is the fading component, assuming uncorrelated flat
the relative propagation path loss of the interference channel.
fading. K is the Ricean K-factor of the channel and is defined
xi [n] represents a unit power symbol vector transmitted from
as the ratio of the power in the LOS component of the channel
the ith user during the nth time interval. Hereafter, for the
to the power in the fading component.
sake of simplicity, we omit the symbol interval index n.
kHd k2
K= (3)
E{kHr k2 }
IV. OVERVIEW FOR THE A DDRESSED C ONCEPTS AND
It is clear that for K = 0, the MIMO channel Hki has a pure S CHEMES
Rayleigh fading, also, if 0 < K < ∞, then, Hki has a Rician
fading; while, the case of K = ∞ corresponds to a non-fading This section gives a brief review for the concepts and the
channel scenario. different systems which we address in this study. This section
In this study, we use the Kronecker correlation model is concluded by detailed description of the main idea of the
expressed in (4) to simulate the correlation effect in the OSGD and its employed algorithms.
40 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
A. Effective Capacity where B stands for the system bandwidth, P is the allocated
Modeling the wireless channel regarding to the connection power. The stochastic service process in a multiple-input
quality of service (QoS) metrics such as delay, ratio of packet multiple-out (MIMO) channel system is given by
loss and bandwidth is crucial to promote the QoS in the
P
future wireless networks. However, the commonly adopted B log2 det I + HQx HH =
conventional physical-layer wireless models do not precisely BN0
distinguish these QoS metrics. Intelligently, authors in [20], B log2 det I + Nr SNRHQx HH bps (10)
proposed and develop a new approach to model the channel
QoS, they introduced a new terminology called (effective and the signal-to-noise ratio (SNR) defined as
capacity, EC) analogous to the (effective bandwidth), where
they assessed the wireless channel by two EC parameters, E kxk2 P
SNR = = . (11)
the probability of non-empty buffer, and the connection QoS E {knk }2 Nr BN0
exponent.
The first metric, i.e., the non-empty buffer probability is where N0 is the power spectral density of the noise.
equivalent in concept to the outage probability of the system, Now, we can obtain effective capacity of the closed and
where the later means that the signal-to-noise ratio (SNR) open loop MIMO systems. First, we take into account the
at the receive falls below a specific threshold. However, the case of closed loop MIMO system, i.e., the system with the
probability of non-empty buffer does not equal the outage channel state information available at the transmitter (CSIT).
probability; because, the former considers the packet accumu- Therefore, the transmitter can adapt its transmitted power and
lation effect on the system, while, the later dose not. Therefore, consequently, Qx according to the channel fading. Hence,
the outage probability is less than the probability of non-empty using (8) we can formulate the effective capacity after normal-
buffer [20]. izing it by the receiver dimensionality Nr and the bandwidth
Generally, effective capacity is defined as in [21], [22] by B as in (12) in the top of next page.
the maximum fixed rate of arrival, that can be supported by The trace of Qx conditioned to be less than or equal 1
some service process, such that, the requirements specified by to maintain the total allocated power condition. From the
the QoS exponent are statistically guaranteed. More explicitly, formula, one can notice that when θ → 0, QoS constraints
if we define Q as the stationary length of a queue, then, the turn to be loose, and the effective capacity reaches the ergodic
decaying rate of the tail of Q distribution, which denoted by capacity as in (13).
θ is given by However, for θ > 0 the ergodic capacity is generally greater
than the effective capacity. One can readily see that in (14)
1 by applying Jensen’s inequality after interchanging the the
θ = − lim log Pr(Q ≥ q) (6)
q→∞ q expectation and the logarithm terms in (12).
It is important to notice that θ → 0 refers to a system that has Second, we consider the case of the open loop MIMO
no delay constraint. On the other hand, θ → ∞ corresponds system, i.e., the channel state information (CSI) is not available
to a system with a rigorous delay constraint. In other words, at the transmitter side. Then, in practice, it is preferred to
larger θ implies more restriction on QoS, while smaller θ refers allocate power uniformly across the MIMO antennas, and
to looser QoS guarantees. consequently Qx = N1t I. Therefore, the effective capacity is
Equivalently, if we define the buffering delay of the packet given by (15).
at the steady-state by D, then for a large dmax , Pr(D ≥ where, the subscript (id) refers to the identical distribution
dmax ) ≈ e−θδdmax , where δ is a constant depends on the of the power across the antenna elements.
service and the arrival processes. The effective capacity is
defined in [23]–[25]
n o B. Water-filling Power Allocation
Λ(−θ) 1
− = − lim loge E e−θS[t] (7) As described in [26] for single-user MIMO channel which
θ t→∞ θt
Pt can be decomposed to a set of non-interfering parallel sub-
where S[t] = i=1 R[i] is the time-accumulated service channels, each of them is corrupted by an independent noise,
process, and {R[i], i = 1, 2, . . .} refers to a discrete time sta- with a constrained on the total allocated power, such that,
tionary ergodic stochastic service process. Assuming a block
fading scenario with a frame duration T , the formula of the N
X
effective capacity simplifies to PT = Pn , (16)
n=1
Λ(−θ) 1 n o
− =− loge E e−θT R[i] (8) where N is the number of parallel sub-channels, and Pn is the
θ θT
allocated power to the nth sub-channel. Then, the maximum
with a short-time constraint on the total allocated power and rate of reliable communication using this scheme is
using a normalized input covariance matrix Qx defined as
N
X
E xxH Pn |hn |2
Qx = (9) log 1 + (17)
P/B n=1
σ2
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 41
1 H
CE (SNR, θ) = − loge E exp −θT B max log2 det I + Nr SNRHQx H bps/Hz/dimension (12)
θT BNr Qx 0,tr(Qx )≤1
1
lim CE (SNR, θ) = E max log2 det I + Nr SNRHQx HH bps/Hz/dimension (13)
θ→0 Nr Qx 0,tr(Qx )≤1
1
CE (SNR, θ) = − loge E exp −θT B max log2 det I + Nr SNRHQx HH
θT BNr Qx 0,tr(Qx )≤1
1 H
≤− E loge exp −θT B max log2 det I + Nr SNRHQx H
θT BNr Qx 0,tr(Qx )≤1
1 H
= E max log2 det I + Nr SNRHQx H (14)
Nr Qx 0,tr(Qx )≤1
1 Nr H
CE,id (SNR, θ) = − loge E exp −θT B log2 det I + SNRHH bps/Hz/dimension (15)
θT BNr Nt
It is easy to notice that the sum rate given by (17) can be levels. This is because the function f (SNR) = log2 (1 + SNR)
maximized by choosing the (optimal power allocation) as is a concave function, and it can be approximated using
XN log2 (1 + x) ≈ x, x→0 (23)
Pn |hn |2
CN := max log 1 + (18) log2 (1 + x) ≈ log2 (x), x≫1 (24)
P1 ,... ,PN
n=1
σ2
Consequently, the system attains a decaying marginal capacity
under the constraint of
gain by adding more power to the the sub-channels with higher
N
X SNR. On the other hand, capacity increases linearly with
Pn = PT , Pn > 0, n = 1, . . . , N. (19) power in the low SNR levels. Therefore, providing some power
n=1 to weaker sub-channels can increase the total sum capacity
Using the Lagrangian method we can solve the problem of the [27]. However, water-filling scheme does not allocate power
concave objective function in (18) as follows to the sub-channels with too low SNR; because, transmitting
information through such sub-channels is waste of power.
XN XN
Pn |hn |2 Then, capacity of water-filling power allocation algorithm is
L(λ, P1 , . . . , PN ) := log 1 + − λ Pn ,
σ2 given by
n=1 n=1
XN
(20) Pn∗ |hn |2
C= log2 1 + . (25)
where λ is the Lagrange multiplier. The allocation power σ2
n=1
optimality condition is given by
The above argument also can be seen from other point of
∂L = 0 if Pn > 0 view, for instance assuming that the system can concentrate its
=
∂Pn 6 0 if Pn = 0 resources, i.e., the transmitted power in this case, in the times
when the channel has high SNR, then the system can achieve
By defining x+ := max(x, 0), then the optimal power alloca- a huge capacity gain. By reflecting this insight into a multi-
tion which satisfies the optimality condition can be expressed user system scenario with large number of users; we can claim
as + that it is most likely to have at any time instance a sub-group
∗ 1 σ2 of users whose channels are in good condition, then, using
Pn = − , (21)
λ |hn |2 a proper selection scheme the multi-user system capacity can
Then, a numerical algorithm can be employed to compute the be achieved. This form of diversity is called (opportunistic
Lagrange multiplier λ to satisfy the total power constraint beamforming) [28].
Finally, if the SNR of all sub-channels are equal, or the
XN + transmitter does not know the channel state information, water-
1 σ2
− = PT . (22) filling algorithm transformed to be an equal power allocation
n=1
λ |hn |2
scheme, i.e., Pn = PT /N and capacity in this case is
Given the channel information, noise variance and using (21) XN
PT
transmitter can determine how to optimally allocate power C= log2 1 + . (26)
N σ2
across the sub-channels. n=1
It is worth to notice that water-filling does not allocate all As one can observe, the slope of the log function is less
the power to the channels with the highest SNR, instead, it is than N ; then, the sum capacity is significantly larger than
still provides some power to the the channels with weaker SNR the capacity of SISO system.
42 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
K
X Thus, the total interference power due to the transmitter k is
rk = wH
k Hkj vj xj + wH
k zk . (27) given by
j=1,j6=k Ĺk = tr{vH
k Ŕk vk }. (36)
When the interference
PK signals are aligned, we need to find a where
K
X
wk such that j=1,j6=k wk Hk,j vj = 0. A reasonable choice
Ŕk = Pk [wH H H
j Hjk ] [wj Hjk ]. (37)
for wk is one that minimize the interference leakage power
j=1,j6=k
at receiver k. There is also a constraint on the dimensionality
of the desired signal: rank(wk Hkk vk ) = dk , where dk is the The objective function here is to choose vk to mini-
degree of freedom (DOF) assigned to the kth transmitter. Thus mize Ĺk , subject to the desired signal dimension constraint,
given the channel realization Hkk and the precoding vectors i.e, rank(wHk Hkk vk ) = dk . Including the regularization term
vk , the optimal receive filter wk is designed to minimize the the objective function is modified as
cost function
Ĺk = tr{vH H
k Ŕk vk + vk vk }. (38)
Jk , tr(wH k Qk wk ) (28)
Thus, the constrained optimization is given by
such that
wH min Ĺk = tr{vH
k Hkk vk }, (39)
k Hkk vk = βIdk (29) vk
D. Optimal Maximum Likelihood Multi-User Detection receiver i, the user of interest will be user i, and given the users
(i) √ (i) √ (i)
Assuming that all of the data symbol vectors are equally rates R, channel realization H̃ = [ P1 H1 , · · · , PK HK ],
likely, ML scheme is optimal in the sense of minimizing the M , {1, · · · , K}, and any disjoint subsets A, B of M, It says
probability of error of the detected signal [30]. For our system that an ordered partition G = {G1 , · · · , Gp } of any subset of
model, the ML detector is given by M , {1, · · · , K} for any p ≥ 1 is valid, if:
• Gk 6= φ
x̂ML = arg min{krk − yk k2 } (44)
x∈D • fi (Gk ) = 1, 1≤k≤p
where • i ∈ Gp
√ √ K
X
yk = P Hkk xk + αP Hki xi (45) where, fi (·) is a bounding function whose purpose is to impose
i=1,i6=k
decoding complexity constraint given as
Each x represents an Nt × 1 transmit vector, where x , [x1 , · · 1 |J | ≤ µi
fi (J ) =
·, xNt ]T , the mth data symbol xm is a complex valued, drawn 0 otherwise
from constellation alphabet A, all elements of the vector x are for a specified integer µi ≥ 1.
assumed to be independent and have zero mean unit variance. By defining a rate outage as an event where in a decod-
Also, set D includes all possible transmitted data vectors x, ing stage the rates of the signals to be decoded falls out
and its cardinality |D| = |A|KNt , with an exponential growth of the corresponding achievable rate region, and Ri to be
with KNt . One disadvantage of ML detection is related to the transmission rate of the signal on transmitter i, also,
its optimization problem, where D is in fact not a convex R , [Ri ]1≤i≤M . Then, the rate margin for decoding A while
set. Therefore, the often used numerical convex optimization treating B as noise for two disjoint subsets A, B ⊆ M is
methods are not suitable for such a scheme [30]. Because defined as follows
the conventional exhaustive search method to find the optimal
solution of (44) by evaluating the krk − yk k2 has a complexity
(i) (i)
O(|A|KNt ), it is common to consider the maximum likelihood ε(H̃ , A, B, R) , min {△ (H̃ , D, B, RD )}, A 6= φ
D⊆A,D6=φ
multiuser detection infeasible for most current communica- (46)
tion applications [31]. Recently, authors in [32] introduced (i)
with ε(H̃ , φ, B, R) = 0 and where
a novel promising approach of designing a low-complexity
maximum likelihood multi-user detector employing quantum (i)
△ (H̃ , D, B, RD ) , (47)
search algorithms (QSA) for potential application in wireless
(i)H (i) (i)H −1 (i) X
communication systems. Their presented results show that log I + H̃D I + H̃B H̃B H̃D − Rj
the employed quantum based MUD search algorithm ties the j∈D
performance of the conventional optimal ML-MUD. However,
(i)
this performance achieved with a significant computational Note that RA ∈ C(H̃ , A, B) if and only if
(i)
complexity reduction compared to the conventional ML-MUD. ε(H̃ , A, B, R) ≥ 0. Now, for any valid ordered partition
This can be seen as an opening for using the joint and group G = {G1 , · · · , Gp } ∈ Qi we define
detection (GD) techniques that can support larger number n (i) o
(i)
of users and higher order modulation schemes in the next ε(H̃ , G, R) , min ε H̃ , Gk , M\ ∪kj=1 Gj , R
generation of the communication systems. 16k6p
(48)
Now, let Qi to be the set of all valid ordered partitions of
E. Optimal Successive Group Decoding all subsets of M which contain i, then, the outage event O(i)
Successive group decoder (SGD) was introduced as an occurs for the valid ordered partition G = {G1 , · · · , Gp } ∈ Qi
extension of the standard successive decoder in which at each if and only if
stage a subset of users is jointly decoded after treating the (i)
ε(H̃ , G, R) < 0. (49)
transmissions of the remaining users as a Gaussian interfer-
ence. Therefore, an OSGD using the partition G ∈ Qi will attempt
In this system, each receiver employs a successive decoding (i)
to decode user i if and only if ε(H̃ , G, R) ≥ 0.
procedure, in each stage a subset of users are jointly decoded,
To perform this decoding procedure each receiver i where
after subtracting the already decoded users from the received
1 ≤ i ≤ K employs a greedy algorithm (Algorithm 1) that
signal, and by treating the remaining users as AWGN. OSGD
either declare an outage or yields an optimal valid partition of
limits the number of users being jointly decoded at each stage
the OSGD. The involved steps of the successive decoding for
to be at most µ to control the complexity of the decoder.
the OSGD in the ith receiver are as follows.
In the remainder of this section, for the sake of the reader (i)
convenience, we reproduce the basic concepts of the OSGD. 1) Initialize with input H̃ , R
Assuming that the power and the rate of all users are pre- 2) Receiver i runs Algorithm 1
specified and are not dependent on the channel realizations. 3) If there is no outage, Algorithm 1 outputs an optimal
Then, we employ the OSGD proposed in [14], which min- partition, G opt = {G1 , · · · , Gp }, Then
imizes the outage probability of each user. Considering the For 1 ≤ k ≤ p
44 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
+
• Compute the noise covariance matrix: O |S|min{(|S|−µi ) ,µi } , where (x)+ = max(0, x). Also, the
(i) P (i) iH
Φ Gk = I + j∈M\∪k G H̃ j H̃j and loop consists the steps 4 to 9 is repeated at most |G| times.
j=1 j
jointly decode the users in Gk using the ML Therefore, since Algorithm 1 will invoke Algorithm 2 no more
− 21 J than K times, the complexity of Algorithm 1 for fixed group
(i) (i) (i)
rule on r̃ [n] = ΦGk y [n] size µi is O K min{K−µi +1,µi +1} , which is polynomial in
n=1
assuming the model K.
− 12 (i) On the other hand, the MIL iterative interference alignment
(i)
r̃(i) [n] = Φ Gk H̃Gk xGk [n] + scheme requires a huge computational complexity to compute
z̃(i) [n], 1 ≤ n ≤ J the transmit precoding vectors and receiving filters. For in-
with z̃(i) [n] ∼ CN (0, I). stance, in each iteration, the kernel algorithm of this scheme
n oJ 2 2
, requires K(K − 1)(Nr Nt + Nr ) and K(K − 1)(Nt Nr + Nt )
(i) (i) (i) (i)
• Update y [n] ←− y [n] − H̃Gk x̂G [n]
k
n=1 complex multiplications, only to compute the covariance ma-
(i)
where x̂Gk [n] is the decision made trix in the transmitter and the receiver respectively. Also, since
corresponding to the nth symbol interval, it is an iterative based algorithm, it needs (depending on the
through the coherence interval of J symbols, initial conditions) a considerable number of iterations L to
in the codeword of the users in the set Gk , converge.
which are re-encoded and modulated by Finally, the maximum likelihood multi-user detection is the
receiver i post-processing. scheme with the highest computational complexity, where it
End For uses the exhaustive search to find the optimal solution of its
4) Otherwise, Algorithm 1 declares an outage for the objective function as shown in Subsection IV-D. This means
intended user i. the receiver performs its search over all combination of all
possible transmitted vectors. Therefore, overall complexity of
this scheme grows exponentially with the number of users
Algorithm 1 Greedy Partitioning for Fixed Rates and the number of transmit antennas. Thus, if we denote
(i)
1 Initialize S = M, G opt = φ the searching set by D, then, its size is given by |A|KNt ,
2 Determine a group G ∗ ⊆ S using Algorithm 2 after initializing it where |A| is the constellation size of the employed modulation
with user set S and rates RS .
(i) ∗ ∗
scheme.
3 If ε(H̃ , G , S\G , R) < 0 then
4 Declare an outage and Stop.
5 else VI. S IMULATION R ESULTS
(i) (i)
6 Update S ←− S\G ∗ and G opt ←− {G opt , G ∗ }
7 If i ∈ G ∗ In this section we present the numerical results carried out
(i)
8 Output G opt . by Monte Carlo simulation of K-user interference channel
9 Stop. MIMO system. Intentionally, we limit the number of users
10 else
11 Go to Step 2
in the simulated scenarios; because of the fact that in many
12 End if realistic scenarios the number of interfering users which
13 End if significantly affecting the wireless connection between the
access point and its desired user is usually limited. As in the
case of downlink cellular system, where there are only limited
Algorithm 2 Selecting an Optimal Group
1 Initialize user set S and rates RS number of co-channel interfering base stations disturbing the
2 Form S , {G ⊆ S : G 6= φ, |G| = µi or G = S} and set S1 = φ, δ = −∞. connection between the base station of interest and its desired
3 For each G ∈ S
4 Repeat user. Also, in uplink system, assuming that there is no intra-
5 Update S1 ←− {S1 , G}. cell interference, which is a reasonable assumption for the
6 Determine
(i)
ξ = minW⊆G,W6=φ ∆(H̃ , W, S\G, RW )
most of the modern cellular systems, the only source of
and let Ŵ be the minimizing set which among interference comes from the co-channel users in the other
all minimizers has the smallest cardinality neighboring cells and those users must be close enough to the
7 If δ < ξ then set A = G and δ = ξ.
8 Update G ←− G\Ŵ base station of interest to be considered as interferers, thus, it is
9 Until G = φ or G ∈ S1 reasonable to assume that they are limited too. Therefore, in all
10 End For
11 Output G ∗ = A, ε(H̃
(i)
, G ∗ , S\G ∗ , R) = δ and stop.
of the following simulation setups of the interference channel
systems we assume the number of users K is 3 transmit-
receive pairs. One user represented as a desired client and
the other two as interfering sources.
V. C OMPLEXITY I SSUES In this work we aim to investigate the OSGD performance in
In this section, we summarize the computational complexity different scenarios and using a wide range of testing metrics.
issues related to the different used techniques in this study. We inspect the OSGD in the interference channel in terms
Starting with few remarks on the complexity of the OSGD of its achieved (ergodic, effective and outage) capacity, also,
and its employed algorithms. Given the users set S and we assess its minimum required energy per bit, as well, we
the group size constraint µi , we find that the number of evaluate its BER and outage probability performance. This
sub-groups that Algorithm 2 is examining will be at most study considers both the spatially uncorrelated and correlated
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 45
18 15
OGD, µ = 3, θhat = 0 OGD, µ = 3, θhat = 0
16 OGD, µ = 3, θhat = 5 OGD, µ = 3, θhat = 5
OGD, µ = 3, θhat = 10 OGD, µ = 3, θhat = 10
14 OGD, µ = 2, θhat = 0 OGD, µ = 2, θhat = 0
OGD, µ = 2, θhat = 5 OGD, µ = 2, θhat = 5
12 OGD, µ = 2, θhat = 10 10 OGD, µ = 2, θhat = 10
Rate per user (bps/Hz)
6 5
0 0
−30 −20 −10 0 10 20 30 40 −50 −40 −30 −20 −10 0 10 20 30 40
SNR (dB) SNR (dB)
Fig. 2: OSGD-µ = 2, 3 and MMSE effective rate versus SNR Fig. 3: OSGD-µ = 2, 3 and MMSE effective rate versus SNR
(dB), α = 1, Nt = 3, Nr = 3, ρ = 0 and the normalized QoS (dB), α = 1, Nt = 3, Nr = 3, ρ = 0.9 and the normalized
exponent θ̂ = 0, 5 and 10. QoS exponent θ̂ = 0, 5 and 10.
35
(with various inter-element correlation factors) flat fading OGD, µ = 3, θhat = 0
OGD, µ = 3, θhat = 5
channels, under different QoS constraint values. Also, this 30
OGD, µ = 3, θhat = 10
work explores the OSGD efficiency in different SNR and SIR OGD, µ = 2, θhat = 0
25 OGD, µ = 2, θhat = 5
environments, (where we consider both the power-limited and Rate per user (bps/Hz) OGD, µ = 2, θhat = 10
bandwidth-limited regimes with different cross-talk values). 20
MMSE, θhat = 0
MMSE, θhat = 5
All of aforementioned investigation performed with numerous MMSE, θhat = 10
transmit-receive antenna configurations. 15
10
A. Achieved Capacity 5
effective capacity results of MMSE receiver and OSGD with OGD, µ = 2, θhat = 0
OGD, µ = 2, θhat = 5
group-size (µ) equals 2 and 3 respectively. Clearly, it can be 20 OGD, µ = 2, θhat = 10
Rate per user (bps/Hz)
seen from the figures that OSGD has the highest achieved MMSE, θhat = 0
MMSE, θhat = 5
ergodic and effective capacity in all cases. Also, we can notice 15 MMSE, θhat = 10
25 20
OGD 4×4 OGD 4×4
OGD 3×3 18 OGD 3×3
MIL 4×4 MIL 4×4
20 MIL 3×3 16 MIL 3×3
14
Rate per user (bps/Hz)
10
10 8
5 4
0 0
−30 −20 −10 0 10 20 30 40 −30 −20 −10 0 10 20 30 40
SNR (dB) SNR (dB)
Fig. 6: OSGD and MIL effective rate versus SNR (dB), α = 1, Fig. 7: OSGD and MIL effective rate versus SNR (dB), α = 1,
(3 × 3, 4 × 4)-MIMO, ρ = 0 and the normalized QoS exponent (3 × 3, 4 × 4)-MIMO, ρ = 0.9 and the normalized QoS
θ̂ = 3. exponent θ̂ = 3.
50
B. Interference Mitigation Capability OGD 8×8
45 OGD 6×6
In the second subsection, we study the OSGD capability to MIL 8×8
40 MIL 6×6
mitigate and suppress interference, so we compare its achieved
effective rate per user measured by (bps/Hz) with the achieved Rate per user (bps/Hz)
35
From figures (6, 7, 8 and 9), we notice that in all cases OSGD 0
−30 −20 −10 0 10 20 30 40
has achieved a higher effective capacity comparing with the SNR (dB)
25
20
C. Minimum Required Energy Per Bit
15
Now, with the same aforementioned simulation setup in the
previous subsection, we study the minimum required energy 10
per bit (Eb /N0 )min for the OSGD and compare it with 5
that for the MIL interference alignment algorithm. Where,
Eb /N0 = RESN R
(SN R) , and RE (SN R) is the effective rate of
0
−30 −20 −10 0 10 20 30 40
the system at the given SN R. Results of this comparison are SNR (dB)
2 2
10 10
1 1
10 10
Effective rate per user (bps/Hz)
−1 −1
10 10
−2 −2
10 MIL 3×3 10 MIL 6×6
MIL 4×4 MIL 8×8
OGD 3×3 OGD 6×6
OGD 4×4 OGD 8×8
−3 −3
10 10
−15 −10 −5 0 5 −25 −20 −15 −10 −5 0 5
Eb/No (dB) Eb/No (dB)
Fig. 10: OSGD and MIL effective rate versus Eb /No (dB), Fig. 12: OSGD and MIL effective rate versus Eb /No (dB),
α = 1, (3 × 3, 4 × 4)-MIMO, ρ = 0 and the normalized QoS α = 1, (6 × 6, 8 × 8)-MIMO, ρ = 0 and the normalized QoS
exponent θ̂ = 3. exponent θ̂ = 3.
1
10
2
10
0 1
10 10
Effective rate per user (bps/Hz)
0
10
−1
10
−1
10
−2
10
MIL 3×3
−2
MIL 4×4 10 MIL 6×6
OGD 3×3 MIL 8×8
OGD 4×4 OGD 6×6
−3
10 OGD 8×8
−15 −10 −5 0 5 −3
Eb/No (dB) 10
−25 −20 −15 −10 −5 0 5
Eb/No (dB)
Fig. 11: OSGD and MIL effective rate versus Eb /No (dB),
α = 1, (3 × 3, 4 × 4)-MIMO, ρ = 0.9 and the normalized QoS Fig. 13: OSGD and MIL effective rate versus Eb /No (dB),
exponent θ̂ = 3. α = 1, (6 × 6, 8 × 8)-MIMO, ρ = 0.9 and the normalized QoS
exponent θ̂ = 3.
15 3
SU−MIMO−WF, perfect CSI SU−MIMO−WF, perfect CSI
SU−MIMO −equal allocated power SU−MIMO−WF, quantized CSI , (8 bits)
OGD SU−MIMO−EP
2.5 OGD, strong interference
10 2
Rate per user (bps/Hz)
5 1
0.5
0 0
−50 −40 −30 −20 −10 0 10 20 30 40 −50−45 −40 −35 −30 −25 −20 −15 −10 −5
Power SNR (dB)
Fig. 14: Ergodic rate per user versus SNR (dB) for OSGD, Wa- Fig. 17: Ergodic rate per user versus SNR (dB) for OSGD,
ter filing and Equal power allocation schemes, Nt = 1, Nr = Water filing and Equal power allocation schemes in power-
2, and α = 1, ρ = 0. limited regeim, Nt = 3, Nr = 3, and α = 1, ρ = 0.
30
SU−MIMO−WF, perfect CSI E. Effect of Cross-Talk Factor
SU−MIMO −equal allocated power
25
OGD, with strong interference In this subsection we discuss the effect of changing the
cross talk α on the capacity of the OSGD scheme. It is clear
20 that the cross-talk factor plays a major role in determining
Rate per user (bps/Hz)
1
is simply because in the very low cross-talk levels OSGD
tends to decode its desired user alone, treating the other weak
0.8
interfering signals as Gaussian noise, i.e., the OSGD converts
0.6 to MMSE decoder.
0.4
40 8
SNR = 30 dB 10×10
SNR = 20 dB 8×8
35 SNR = 10 dB 7
6×6
SNR = 0 dB
4×4
SNR = −10 dB
30 6 2×2
1×1
20 4
15 3
10 2
5 1
0 0
0 −5 −10 −15 −20 −25 −30 −35 −40 −20 −15 −10 −5 0 5 10 15 20 25 30
Cross talk factor (α) in (dB) SNR (dB)
Fig. 18: MMSE ergodic capacity versus cross-talk factor in Fig. 21: MMSE ergodic capacity versus SNR using different
different SNR levels, Nt = 4, Nr = 4, ρ = 0. sizes of MIMO configuration, ρ = 0.
40 12
SNR = 30 dB 10×10
SNR = 20 dB 8×8
35 SNR = 10 dB 6×6
10
SNR = 0 dB 4×4
SNR = −10 dB
30 2×2
1×1
20 6
15
4
10
2
5
0 0
0 −5 −10 −15 −20 −25 −30 −35 −40 −20 −15 −10 −5 0 5 10 15 20 25 30
Cross talk factor (α) in (dB) SNR (dB)
Fig. 19: OSGD with µ = 2 ergodic capacity versus cross-talk Fig. 22: OSGD with µ = 2 ergodic capacity versus SNR using
factor in different SNR levels, Nt = 4, Nr = 4, ρ = 0. different sizes of MIMO configuration, ρ = 0.
40 it is still far away from the efficiency of the OSGD with group
SNR = 30 dB
SNR = 20 dB size of 3. It is also clear that MMSE is an interference-limited
35 SNR = 10 dB
SNR = 0 dB
scheme no matter what the MIMO size is. Differently, OSGD
30
SNR = −10 dB
with µ = 3 exhibits a noise-limited behavior, revealing its
capability to suppress interference. In between, we perceive
25
Capacity (bps/Hz)
0
0 −5 −10 −15 −20 −25 −30 −35 −40 G. Effect of Antenna Inter-Element Correlation
Cross talk factor (α) in (dB)
In this part, we aim our attention to study the relationship
Fig. 20: OSGD with µ = 3 ergodic capacity versus cross-talk between the antenna inter-element correlation factor ρ and
factor in different SNR levels, Nt = 4, Nr = 4, ρ = 0. the achieved capacity. In the following scenario, we employ
(4 × 4)-MIMO channel for every transmit-receive pair, with
cross-talk parameter α = 1 and θ̂ = 0. From the results in
(21, 22, and 23). From the outcomes, one can see that although figures (24, 25 and 26), one can notice that for a specific
the OSGD with group size of 2 is more efficient in exploiting SNR level achieved capacity of OSGD with µ = 3 decreases
the increase of MIMO antenna size than the MMSE receiver, monotonically with increasing the correlation factor. However,
50 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
50 7
10×10 γ = 30 dB
45 8×8 γ = 20 dB
6×6 6 γ = 10 dB
40 4×4 γ = 3 dB
2×2 γ = 0 dB
35 1×1 5 γ = −3 dB
Effective Capacity (bps/Hz)
γ = −10 dB
Capacity (bps/Hz)
30 γ = −20 dB
4
25
3
20
15 2
10
1
5
0 0
−20 −15 −10 −5 0 5 10 15 20 25 30 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
SNR (dB) Inter−element Correlation factor (ρ)
Fig. 23: OSGD with µ = 3 ergodic capacity versus SNR using Fig. 25: OSGD with µ = 2 ergodic capacity versus antenna
different sizes of MIMO configuration, ρ = 0. inter-element correlation factor in different SNR levels, using
4 × 4 MIMO for each transmit-receive pair.
6
γ = 20 dB
γ = 10 dB 18
5 γ = 3 dB γ = 30 dB
γ = 0 dB 16 γ = 20 dB
γ = −3 dB γ = 10 dB
γ = −10 dB γ = 3 dB
4 14
γ = −20 dB γ = 0 dB
Capacity (bps/Hz)
γ = −3 dB
12
γ = −10 dB
Capacity (bps/Hz)
3 γ = −20 dB
10
2 8
6
1
4
0 2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Inter−element Correlation factor (ρ)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Inter−element Correlation factor (ρ)
Fig. 24: MMSE ergodic capacity versus antenna inter-element
correlation factor in different SNR levels, using 4 × 4 MIMO Fig. 26: OSGD with µ = 3 ergodic capacity versus antenna
for each transmit-receive pair. inter-element correlation factor in different SNR levels, using
4 × 4 MIMO for each transmit-receive pair.
this is not the case for the OSGD with µ = 2 or the MMSE
receiver. They exhibit a monotonic decrease of their attained
antennas. The modulation scheme used here is the BPSK, and
capacity with increasing the correlation factor only at low
information decoded in the receiving side symbol by symbol.
levels of SNR, but for higher SNR levels, they are interestingly
The results are depicted in Fig. 27, from which it is obvious
revealing a sudden increase in their capacities in an overshoot- Eb
that BER decreasing linearly with the logarithmic scale of N .
like behavior for short range of correlation values, before 0
Also, we can readily notice the increase of the diversity gain
they return to steeply decreasing their capacity for higher
(which can be inferred from the increase of the absolute slope
correlation. However, apparently, we can recognize that for
of the BER curves) by increasing the number of the receive
all correlation and SNR combination the OSGD is the scheme
antennas. Now, to get a deeper insight on the OSGD efficiency,
with the highest achieved capacity.
we choose to compare its performance with the ML-MUD
scheme, where all the users are jointly decoded in the receiver
H. Bit Error Rate side. Fig. 28 presents the BER performance of the OSGD and
This subsection presents the results of BER evaluation for ML-MUD in a strong uncorrelated interference environment,
the OSGD scheme in different scenarios, where we conduct but in this case, all of the transmitters and receivers are
this assessment using various SIMO and MIMO configura- equipped with multiple antennas. At each transmitter the
tions, using several correlation factors. In the first scenario, information bits are modulated and transmitted simultaneously
we assume a strong interference environment α = 1 with un- over all antenna elements of that user, this is of course will
correlated Rayleigh flat fading, where, single transmit antenna introduce a transmit diversity gain over the previous single
users communicate with their intended receivers which employ antenna case. The results of this scenarios show that at high
OSGD and are equipped in each case with (2, 3, 4, or 5) SNR levels BER performance of the OSGD is slightly deviates
ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014 51
0
10
−1
10
−2
10
BER
−3
10
0 1 2 3 4 5 6 7 8 9 10
Eb/No (dB)
Fig. 27: OSGD BER performance versus SNR for Fig. 29: OSGD and ML-MUD BER performance versus SNR
1 × (2, 3, 4, 5) SIMO configurations, with cross-talk fac- for 4 × 4 MIMO configuration, with cross-talk factor α = 1
tor α = 1 and receive antenna inter-element correlation ρ = 0. and antenna inter-element correlation ρ = 0, 0.5 and 0.9.
0
10
0
10
−1
10
−1
10
−2
10
−2
10
BER
−3
10
BER
−3
10
−4
10
OGD, 2×2 OGD, 5×5, ρ = 0.9
−4
10 ML−MUD, 2×2 ML−MUD, 5×5, ρ = 0.9
OGD, 3×3 −5 OGD, 5×5, ρ = 0.5
10
ML−MUD, 3×3 ML−MUD, 5×5, ρ = 0.5
−5 OGD, 4×4 OGD, 5×5, ρ = 0
10
ML−MUD, 4×4 ML−MUD, 5×5, ρ = 0
−6
OGD, 5×5 10
0 1 2 3 4 5 6 7 8
ML−MUD, 5×5
−6 Eb/No (dB)
10
0 1 2 3 4 5 6 7 8 9 10
Eb/No (dB)
Fig. 30: OSGD and ML-MUD BER performance versus SNR
Fig. 28: OSGD and ML-MUD BER performance versus for 5 × 5 MIMO configuration, with cross-talk factor α = 1
SNR for different MIMO configurations, with cross-talk fac- and antenna inter-element correlation ρ = 0, 0.5 and 0.9.
tor α = 1 and antenna inter-element correlation ρ = 0.
I. Outage Probability
In this section we present the outage probability results
from the ML-MUD performance; however, in low SNR levels of the OSGD scheme for several transmit data rates, using
OSGD exhibits very enhanced capability to decode its desired different MIMO configurations, with various correlation set-
users and matches with the ML-MUD scheme. The reason for tings, including the (uncorrelated, semi-correlated and fully
the slight performance drift is that OSGD tends to be a single correlated) scenarios, also for the heterogeneous (Rayleigh and
decoder at higher SNR levels. On the other hand, Figures (29 Racian) flat fading environments. All the results shown in this
and 30) demonstrate the OSGD BER performance for the part are based on a three-user interference system channel,
(4 × 4 and 5 × 5) MIMO configuration in a strong interference with cross-talk factor α = 1, where, simulation results of each
environment, in this scenario we assume correlation exists setup are carried out using 10000 channel realizations.
between the elements of each antenna. Results in both figures Results depicted in figures (31, 32 and 33) show the outage
present the BER performance of OSGD and ML-MUD for two probability against the decibel SNR level, using (2 × 2, 3 × 3
different correlation levels, where (ρ = 0.5 and 0.9) along with and 4×4) Rayleigh MIMO configuration, for the transmit data
the no correlation case, i.e., ρ = 0. All of the shown results rate of (2, 3, and 4) bps/Hz respectively, in this case we assume
clearly prove that BER performance of OSGD is significantly no correlation. From the figures we can perceive the increase
matches the performance of the ML-MUD scheme, especially of the negative slope of the outage probability (which can be
for the tight specification environments, i.e., for low SNR and appreciated as a diversity gain) as the MIMO size increase.
higher correlation levels. demonstrating by that the OSGD Also, from the multiplexing gain point of view, it can be
capability to cope with interference. seen that OSGD efficiently exploits the MIMO dimension to
52 ALMADAR JOURNAL FOR COMMUNICATIONS, INFORMATION TECHNOLOGY, AND APPLICATIONS, VOL. 01, NO. 01, JULY 2014
0 0
10 10
2×2, ρ = 0 2×2, ρ = 0.9
3×3, ρ = 0 3×3, ρ = 0.9
4×4, ρ = 0 4×4, ρ = 0.9
Outage probability for a rate of: (4) bps/Hz
−2 −2
10 10
−3 −3
10 10
−4 −2 0 2 4 6 8 10 12 14 16 0 5 10 15 20 25
SNR (dB) SNR (dB)
Fig. 33: OSGD outage probability performance versus SNR, Fig. 35: OSGD outage probability performance versus SNR,
using rate of 4 bps/Hz, for different MIMO configurations, using rate of 4 bps/Hz, for different MIMO configurations,
with cross-talk factor α = 1 and antenna inter-element with cross-talk factor α = 1 and antenna inter-element
correlation ρ = 0. correlation ρ = 0.9.
0
0 10
10
2×2, κ = 10 dB, ρ = 0 2×2, ρTx = 0.9, ρRx = 0.9
3×3, κ = 10 dB, ρ = 0 2×2, ρTx = 0.9, ρRx = 0
4×4, κ = 10 dB, ρ = 0 2×2, ρTx = 0, ρRx = 0.9
Outage probability for a rate of: (4) bps/Hz
Outage probability for a rate of: (4) bps/Hz
−2
−2 10
10
−3
−3 10
10
−4 −2 0 2 4 6 8 10 12 14 16 8 10 12 14 16 18 20 22
SNR (dB) SNR (dB)
Fig. 34: OSGD outage probability performance versus SNR, Fig. 36: OSGD outage probability performance versus SNR,
using rate of 4 bps/Hz, for different MIMO configurations, using rate of 4 bps/Hz, for 2 × 2 MIMO configuration, with
with cross-talk factor α = 1 , antenna inter-element corre- cross-talk factor α = 1 and different antenna inter-element
lation ρ = 0, for the Racian-Rayleigh fading scenario with correlation ρ = 0 and 0.9.
κ − factor = 10 dB.
0
support of quality of service,” IEEETransactions on Wireless Commu-
10
nications, Vol. 2, No. 4, pp. 630–643, 2003.
8×8, ρTx = 0.9, ρRx = 0.9
[21] M. Gursoy, “MIMO wireless communications under statistical queueing
8×8, ρTx = 0.9, ρRx = 0
constraints,” in Control, and Computing, 2009. Allerton 2009. 47th
8×8, ρTx = 0, ρRx = 0.9
Outage probability for a rate of: (16) bps/Hz
مشكلة التشويش بسبب أجهزة التشويش على الهاتف المحمول شركة المدار الجديد هي إحدى شركات االتصاالت التابعة للشركة
الموجودة بالمساجد (Interference Due to Mobile الليبية لالتصاالت وتقنية المعلومات القابضة .تأسست شركة المدار
)Jammers in Mosques الجديد سنة ،6991وبدأت في تقديم خدماتها للجمهور سنة .6991
تعتبر شركة المدار الجديد أول مشغل للهاتف المحمول في ليبيا.
مشكلة صناديق الشفرات ()SIM Boxes قريبا ً ستشهد شبكة المدار الجديد تطوراً كبيراً ،حيث سيتم تحديث
مكوناتها لدعم تقنيات الجيل الثالث المتطور ( )3.7Gوالجاهز لتقبل
تقنيات الجيل الرابع ( .)4Gمزيد من التفاصيل حول هذا المشروع
الطموح تظهر في اإلصدار القادم من هذه المجلة.
4
مجلة المدار لالتصاالت وتقنية المعلومات وتطبيقاتهما
عمله المستقبلي حتى يتم نشر هذه الورقة وبعد ذلك يستخدمها يجوز لهيئة التحرير إصدار عدداً استثنائيا ً إذا تراكمت
كمرجع. المشاركات أو تطلبت أهمية االبحاث سرعة نشرها .كما يجوز
يتجنب المراجع مراجعة الورقة إذا كان هناك تضارب في إصدار عدداً خاصا ً يتناول األبحاث التي تعالج موضوعا ً بعينه.
المصالح مع أحد الباحثين أو أن موضوع الورقة ال يكون من
ضمن تخصص المراجع. البحث المنشور في المجلة يصبح ملكا ً لها ويؤول إليها حق النشر،
قراءة الورقة بالتفصيل ،والتحقق من المعادالت والخوارزميات لكن يحق للباحث إعادة نشر بحثه في كتاب وفي هذه الحالة ال بد أن
واألشكال ،وتقييم المساهمات ،والتحقق من قائمة المراجع، يشير إلى المصدر األصلي للنشر.
ونسق ووحدة الورقة ،والتحقق من لغة وتنظيم الورقة.
إعداد الورقات
كتابة تقرير مفصل عن الورقة ،ويوصي المراجع بأن تكون
نتيجة تقييم الورقة أحد الخيارات التالية: يمكن للسادة الراغبين في نشر أعمالهم بالمجلة إرسال أعمالهم إلى
oقبول. هيئة التحرير على شكل ملف في صيغة ،pdfوإذا تم قبول العمل
oقبول بشروط. للنشر يجب على الباحث إعادة إرساله في الصيغة التي تطلبها هيئة
oطلب تعديالت جوهرية. التحرير مع االلتزام بالمواصفات الواردة بموقع المجلة.
oرفض.
آلية التقديم
حيث يبرر المراجع قراره لكل من هيئة التحرير والباحث في تقرير
مفصل. تكون جميع المراسالت بين الباحثين وهيئة تحرير المجلة عبر البريد
اإللكتروني .ajcita@almadar.ly
ذكر جميع النقاط اإليجابية والسلبية للورقة في تقرير المراجع،
كما يتم اإلشارة إلى االحتيال المشبوه أو االنتهاك لحقوق النشر
اختيار الورقات المناسبة للنشر
في تقرير خاص بهيئة التحرير.
يتم الفحص األولي لألوراق المقدمة للنشر في المجلة من قبل هيئة
لهيئة التحرير الحق في رفض تقرير المراجع إذا لم يقدم تبريرات
التحرير للتأكد من أنها تتبع المتطلبات الرئيسية للمجلة من حيث
وتعليقات مقنعة في تقريره.
المواضيع واألسلوب ،وفي حالة تحقق المتطلبات الرئيسية يتم إرسالها
تستكمل عملية المراجعة خالل 6أسابيع من تاريخ استالم طلب التقييم إلى ثالثة مراجعين على األقل لتقييم الورقة ومراجعتها بعد طمس
والمراجعة ،ويمكن تمديد هذه المدة أسبوعين إضافيين كحد أقصى إذا أسماء المؤلفين.
طلب المراجع ذلك ،وسيتم رفض تقرير المراجع عند وصوله بعد
فترة تزيد عن 8أسابيع.
رفض الورقات
3
مجلة المدار لالتصاالت وتقنية المعلومات وتطبيقاتهما
ينبغي أن يستند ترتيب قائمة أسماء الباحثين على نسبة العمل المنجز، عمل هيئة تحرير المجلة
وذلك بأن يكون االسم األول للباحث األكثر مساهمة في الورقة ،ثم
الباحث الذي يليه وهكذا ،وال يتم قبول ورقة تشتمل على أسماء في تجتمع هيئة التحرير مرة واحدة كل ثالثة أشهر على األقل أو كلما
قائمة الباحثين يثبت أنه ليس لديهم مساهمة في الورقة. دعت الحاجة إلى ذلك بدعوة من رئيس التحرير أو سكرتيره.
يجب ذكر مصادر تمويل البحوث المقدمة للنشر في المجلة (شركة، يعفى أي عضو من أعضاء هيئة التحرير من عضوية الهيئة إذا تخلف
جامعة ،منحة التمويل ،الخ) وذلك باإلشارة لها داخل الورقة كنوع من عن حضور ثالثة اجتماعات متتالية أو خمسة اجتماعات متفرقة على
الشكر والتقدير. مدار السنة بدون عذر مقبول.
يجب أال تكون الورقات المقدمة للنشر في المجلة قد سبق أن نشرت تنشر أسماء أعضاء هيئة التحرير والسادة المشاركين في التقييم
في أي مجلة أو مؤتمر آخر ،وأال تكون تحت المراجعة الحالية من والمراجعة في صدارة كل عدد يصدر من المجلة أو حسب ما تقتضيه
قبل مؤتمر أو مجلة أخرى ،كما أنه سيتم توقيع اتفاق نقل حقوق النشر طبيعة اإلخراج.
لألوراق المقبولة قبل نشرها.
تمارس هيئة تحرير المجلة االختصاصات التالية:
يجوز تقديم ورقات نشرت جزئيا ً من قبل في مجالت أو مؤتمرات
أخرى ،بحيث تكون هناك مساهمة جديدة واضحة في األوراق تسلم الدراسات والبحوث المعروضة للنشر وإقرار تحويل ما
المقدمة. يصلح منها للتقييم والمراجعة.
رسم سياسة المجلة وإقرار المحاور التي تهتم بها.
الباحثون في الورقة المقدمة مسئولون عن جميع القضايا والدعاوى تحديد سعر المجلة ورسوم االشتراك وطرق توزيعها وإهدائها و
المتعلقة بحقوق النشر والملكية الفكرية التي تثار من الغير ضد المجلة كل ما من شأنه تسيير أمورها.
ويتحملون كافة التبعات القانونية المترتبة بالخصوص دون أدنى تطوير المجلة ومراجعة سياسات التحكيم والنشر وتقييم
مسئولية من المجلة. المستوى العلمي واألكاديمي لها.
يجب على الباحثين احترام أخالقيات الكتابة والنشر وأي احتيال يمارس رئيس هيئة التحرير االختصاصات التالية:
جزئي أو كلي لعمل قام به باحثون آخرون سوف يعتبر جريمة علمية
وفي هذه الحالة سوف يتم اتخاذ اإلجراءات التالية: رئاسة اجتماعات هيئة التحرير.
تمثيل المجلة أمام الهيئات الرسمية والقضائية.
رفض نشر هذه الورقة. اإلشراف على األمور التحريرية بالمجلة.
يجب على الباحثين المشاركين في هذه الورقة تقديم تقرير تلقي المراسالت التي ترد إلى المجلة والرد عليها.
توضيحي حول هذه الحالة لمعرفة المسؤول عن هذا العمل. مراجعة كافة مواضيع المجلة قبل تحويلها إلى الطباعة.
حرمان الباحثين المسؤولين عن عملية االحتيال من التعاون
واالشتراك مرة أخرى في المجلة. يمارس سكرتير التحرير االختصاصات التالية:
يحق لرئيس تحرير المجلة إبالغ الجمعيات والمؤسسات الوطنية
والدولية وهيئات تحرير الدوريات العلمية األخرى حول هذه إعداد جدول اجتماعات هيئة التحرير.
الجريمة. الدعوة الجتماعات هيئة التحرير وتحرير المحاضر وتسجيلها.
إذا تم اكتشاف احتيال في وقت الحق بعد نشر العمل ،سوف يتم مساعدة رئيس التحرير في أعماله والنيابة عنه في حالة غيابه
سحب الورقة ونشر مذكرة االحتيال. واإلشراف على متابعة قرارات هيئة التحرير.
يجوز للباحثين االعتراض على أي إجراء عند هيئة تحرير متابعة طباعة المجلة وال سيما ما يتعلق منها بترتيب المواضيع
المجلة عن طريق تقديم اعتراض كتابي موجه إلى رئيس هيئة ومواضيع الغالف.
التحرير ،وفي هذه الحالة فإنه يجوز لهيئة التحرير استشارة قواعد النشر
خبراء في هذا المجال للحكم في المسألة. تعتمد الموافقة على نشر المقاالت البحثية على الثقة ،حيث أنَّ القراء
يراعى في أولوية النشر ما يلي: يثقون في إجراءات محرري المجلة لتحديد الورقات األكثر أهمية
تاريخ استالم البحث وتعطى األسبقية للبحوث التي طلب إجراء ذات المستوى الرفيع ،كما أنَّ هيئة التحرير تعطي الثقة للمراجعين
تعديالت عليها. بأنهم سيعطون الوقت والجهد للتأكد من جميع المعادالت واألشكال
تنوع األبحاث والباحثين لتحقيق التوازن بحيث تنشر المجلة في كل الورقات ،وكذلك الباحثون يثقون في صدق المراجعين
ألكبر عدد من الكتاب ومن أكبر عدد ممكن من األقطار في وعدلهم في اختيار الورقات التي سيتم نشرها.
العدد الواحد وبأوسع مدى من التنوع. يجب على الباحثين أن يقدموا أفضل ملخص لعملهم األصلي دون
المواضيع التي تخص الشأن الليبي أو اهتمام الشركة. وجود أي احتيال أو انتهاك لحقوق الطباعة والنشر.
2
مجلة المدار لالتصاالت وتقنية المعلومات وتطبيقاتهما
التعريف بالمجلة
مجاالت اهتمام المجلة أهداف المجلة
تغطي المجلة مجموعة واسعة من المواضيع المتعلقة بهندسة مجلة المدار لالتصاالت وتقنية المعلومات وتطبيقاتهما ( ISSN
االتصاالت وتقنية المعلومات ،وتطبيقاتهما ،فعلى سبيل المثال ال ( )2313-156Xرقم إيداع بدار الكتب الوطنية )4102\963هي
الحصر تقع المواضيع التالية ضمن اهتمام المجلة :تقنيات االتصاالت مجلة علمية محكمة متخصصة في مجالي االتصاالت وتقنية
الالسلكية ونمذجة القناة ،الشبكات الالسلكية ،االتصاالت المتنقلة، المعلومات وتطبيقاتهما ،تصدر نصف سنويا ً عن مكتب البحث
الشبكات المتنقلة ذات النطاق العريض ،الهوائيات ،معالجة اإلشارات، والتطوير بشركة المدار الجديد بدعم الشركة ورعايتها .تطمح هذه
شبكات الحاسوب ،األتمتة الالسلكية ،تطبيقات تقنية المعلومات على المجلة إلى تحقيق األهداف التالية:
شبكات المحمول ،شبكات الحماية ،اتصاالت األلياف البصرية،
شبكات األقمار االصطناعية ،تطبيقات تقنية المعلومات واالتصاالت تشجيع ونشر ثقافة البحث العلمي في مجاالت عمل الشركة.
في نظم الطاقة ،المدن الذكية ،تحديد المواقع باستخدام شبكات الهاتف توفير مرجع علمي رصين ذو موثوقية عالية يخدم الطالب
المحمول ،شبكات االستشعار ،وغيرها. الدارسين والمهندسين العاملين والباحثين واألكاديميين
بالجامعات والمراكز البحثية والصناعية والمؤسسات ذات
هيئة تحرير المجلة العالقة داخل ليبيا وخارجها.
المساهمة في تنمية العلوم الهندسية والتقنية وتطبيقاتها من خالل
المشرف العام نشر البحوث النظرية والتطبيقية مع التأكيد على الجودة العالية
لهذه البحوث وارتباطها بالواقع حاضراً ومستقبالً.
عبدهللا علي أعبودة ،مدير مكتب البحث والتطوير بشركة المدار إتاحة الفرصة للمبدعين والباحثين في المجاالت المستهدفة بنشر
الجديد ،ليبيا ،البريد اإللكترونيa.abouda@almadar.ly : نتاج أنشطتهم العلمية والبحثية.
المساهمة في إقامة شبكة تعاون علمي بحثي أكاديمي.
رئيس التحرير
متابعة النشاطات العلمية في مجال هندسة االتصاالت وتقنية
محمد سالم المصراتي ،أستاذ ورئيس مجموعة االتصاالت والنظم المعلومات في العالم عامة وفي ليبيا خاصة.
الهندسية بجامعة فازا ،فنلندا ،وكذلك أستاذ بقسم الهندسة الكهربائية
واإللكترونية ،جامعة بنغازي ،ليبيا ،البريد اإللكتروني:
محتويات المجلة
mohammed.elmusrati@uwasa.fi
يتم طباعة نسخ ورقية بكمية محدودة لكل إصدار من المجلة مع توفير
النسخة اإللكترونية منه عبر الموقع اإللكتروني للشركة .وتشمل
أعضاء هيئة التحرير
محتويات المجلة ما يلي:
محمد جمعة الترهوني ،أستاذ ورئيس قسم الهندسة الكهربائية،
ورقات علمية محكمة من قبل نخبة من أهل االختصاص.
الجامعة األمريكية في الشارقة الشارقة ،اإلمارات العربية المتحدة،
مقاالت علمية مدعوة لخبراء في مجالي االتصاالت وتقنية
البريد اإللكترونيmtarhuni@aus.edu :
المعلومات وتطبيقاتهما.
علي أحمد قنون ،أستاذ مساعد بقسم الهندسة الكهربائية واإللكترونية، أخبار ومستجدات قطاع االتصاالت وتقنية المعلومات في ليبيا.
جامعة طرابلس ،ليبيا ،البريد اإللكتروني: إعالنات تجارية ذات طابع تقني يقع ضمن اهتمام المجلة من
مختلف المؤسسات والشركات العامة والخاصة.
ali.ganoun@ee.edu.ly إعالنات ودعايات لنشاطات مختلفة في المجاالت ذات العالقة
باهتمامات المجلة كالمؤتمرات وورش العمل والمسابقات.
مجدي علي الشيباني ،أستاذ مشارك بكلية التقنية الصناعية ،ليبيا،
البريد اإللكترونيmashibani@lptic.ly : لغة النشر في المجلة
محمد مرعي العماري ،أستاذ مشارك ورئيس قسم هندسة البرمجيات،
تعتمد المجلة لغتان للنشر هما اللغة اإلنجليزية واللغة العربية .يفترض
جامعة بنغازي ،ليبيا ،البريد اإللكتروني:
في المقاالت المقدمة للنشر باللغة اإلنجليزية أن تقدم مشاركة جديدة
mohamed.elammari@uob.edu.ly في مجال العمل ،أما المقاالت المقدمة للنشر باللغة العربية فيمكن أن
تكون عبارة عن مقاالت تعريفية بتقنيات حديثة ال يتوفر عنها مصادر
عبدالقادر الصادق عكي ،أستاذ بقسم الهندسة الكهربائية واإللكترونية، باللغة العربية.
جامعة طرابلس ،ليبيا ،البريد اإللكترونيasakki@lttnet.net :
1