Sample Technical Papers - Collection For Student Awarness

3rd International Conference on Renewable Energy Research and Applications Milwakuee, USA 19-22 Oct 2014
Determination of Iron Loss Considering Spatial

Harmonics and Tooth Pulsation Effects for Cage
Motor
Vinay Jaiswal, and N. K. Deshmukh

Global R&D Centre
Crompton Greaves Limited
Kanjur Marg, Mumbai, India
vinay.jaiswal@cgglobal.com
vinay_kjaiswal@hotmail.com
Abstract—The contribution of spatial harmonics and tooth motor efficiency. Lee et al. [3] presented iron loss analysis of
pulsation in iron loss of three phase squirrel cage induction three-phase IM used for industry applications. They used
motor has been calculated using lumped circuit approach. An transient FEM to analyze the magnetic flux density waveform
equivalent circuit of the three phase induction motor considering in order to find out the flux density according to varying time.
stator magnetomotive harmonics has been used for the Iron loss in each element was evaluated using iron loss curve
calculation of the same. An empirical formula has been developed tested by Epstein test apparatus. Their results showed the
and validated for determination of tooth pulsation loss. The distribution of iron loss density and partial thermal source.
usage of lumped parameter circuit considering magnetomotive Bangura et al. [4] studied the effects of rotor abnormalities
force harmonics for predicting the iron loss components has not
such as broken squirrel-cage bars, broken cage connectors and
appeared in the literature. Here iron loss due to fundamental,
airgap eccentricity on ohmic and iron losses of induction
spatial harmonics components and tooth pulsation for various
motors with different designs and polarities have been
motors. Enokizono et al. [5] examined the loss distribution by
determined and compared with experimental results. It is found the FEM considering the two-dimensional magnetic properties
that the iron losses due to the stator magnetomotive harmonics is to develop higher efficiency induction motors. Jamil et al. [6]
in same range as that due to fundamental component, the tooth used a combined finite-element/state-space (CFE-SS)
pulsation loss is in the range of 25% to 35% for 2 pole, and 20% modeling to compute the no-load iron losses in three-phase
to 30% of total iron loss for 4 pole induction motors. The induction motors.
percentage error between computed and average experimental
Earlier literature considering magnetomotive force
results for total iron loss are within acceptable limits (±5%)
without taking too much of computation time compared to the
(m.m.f.) harmonics [7]-[9] discussed the methodology.
numerical methods. Reference [8] compares the overall performance using the
methodology. However in this paper specifically, computation
Keywords—Iron loss; magnetomotive force harmonics; squirrel of iron loss considering all components of harmonics is
cage; three phase induction motor. targeted and the comparison with experimental results is
presented.
I. INTRODUCTION In certain cases, the iron loss is computed using empirical
In three phase induction motor (IM), iron loss accounts for formulae from the determined space and time variation of the
a large percentage of the total energy loss. It is desirable to flux density [10], [11]. Even if the formulae for estimating the
predict iron loss with satisfactory precision in the stage of iron loss were accurate, the obvious error is already present in
design or during analysis to realize high efficiency and good the exclusion of the iron loss from the field computation.
performance. Due to the complicated flux distribution, spatial Gyselinck et al. [12] presented a two-step algorithm for
harmonics, and the rotation of the flux vector, major predicting the iron loss in a squirrel cage induction motor. The
difficulties are encountered for the last several decades in method presented in their paper has produced losses that are
evaluation of iron loss in the induction motor. Kunihiro et al. 15% less than the measured losses for open rotor slots. Julius
[1] presented results of iron-loss evaluation of an induction Saitz [13] presented the application of the vector Preisach
motor model core by using the finite element method (FEM) model incorporated in 2D finite element analysis for the
considering two-dimensional vector magnetic properties. magnetic field computation and estimation of the iron loss of
Kwon et al. [2] computed the loss distribution of a three-phase an induction motor. The maximum relative error between the
IM and a BLDC motor considering operating point-of-core computed and measured iron losses is 12%. His numerical
material based on the FEM and experiments. Their presented procedure is stable convergent, however, it requires a lot of
results give guidelines to selecting core materials to improve computation time (about 20 hours for one period of the supply
ICRERA 2014 348

,(((
voltage consisting of 400 time steps). Lei Ma et al. [14] The triplen harmonics in three phase induction motors get
proposed a method to predict iron loss in rotating machines eliminated due to 120 degrees electrical displacement. So the
considering rotational variations of flux vectors as well as flux spatial harmonics from stator m.m.fs, can be given by
harmonics using the 2-D FEM. Using their proposed method, (6k±1)th, where k=1, 2, 3…i.e. 5th, 7th, 11th, 13th, 17th, 19th,....
the average prediction error of iron loss is 18% compared with till (Ns/p-1)th (Ns is number of stator slots, p is no. of pole
the experimental values. Stranges [15] described direct pairs) harmonic m.m.fs. are present.
approaches for coupling rotational iron loss measurements
with FEA results that yield the distribution of rotational flux in The equivalent circuit parameters have been calculated as
induction motor cores. Using those methods, the prediction of
follows [16]:
no-load iron loss can be made to include the effects of
rotational iron losses, but the components of iron losses cannot
be calculated. r1 = ρ (length of mean turn*turns per phase / area of
conductor),
The stator core of induction motor is subjected to flux that
rotates in the plane of the laminations. The iron loss in the Ts
2
§ 2 5 N sp ·
three phase induction motor can be divided into three x1 = 2πf℘ ¨¨ − 2 ¸
¸ + xe
(1)
components, i.e. fundamental, harmonic, tooth pulsation. The p ©q 4 q ¹
harmonic component of iron loss is primarily due to the
hysteresis and eddy currents set up in the tip region of the
stator and rotor teeth and to additional losses in the teeth from r1= stator resistance per phase
pulsations resulting from the passing flux harmonics. x1= stator reactance per phase
p= no. of pole pairs
II. MODELLING AND ANALYSIS Ts = no. of stator turns per phase
The motor is represented by the equivalent circuit shown q= no. of slots per pole per phase
in Fig. 1. This equivalent circuit represents the stator Nsp= no. of slots per pole – no. of slots spanned by Coil Pitch
magnetomotive force (m.m.f.) harmonics present in the
system.
xe = leakage reactance due to the end windings
There are a number of ways of estimating xe , we have
estimated this parameter as in [17].
Slot permeance,
§1 h d ·
℘ = "μ 0 ¨ + ¸ (2)
©3 w u ¹
h= slot height
w= slot width
d= slot depression height
u= slot depression width
"= slot length
k wn = k pn k dn is winding factor for the nth harmonics
k pn = sin( nxφ / 2) (3)
xφ = π*no. of slots spanned by Coil Pitch/(Ns*2p)
Ns= No. of stator slots.

Fig. 1: The equivalent circuit of three phase induction motor for the proposed
method.
sin( nqγ / 2)
k dn = (4)
q sin( nγ / 2)
γ = π / no. of stator slots per pole
ICRERA 2014 349

Magnetizing reactance for the nth harmonics, wd = rotor slot opening

3
Xm _ n = ω℘ag Ts k wn
2 2
2
(5) ωn angular speed for the nth harmonics= 2πfs _ n
Where,
s _ n is slip for the nth harmonics
ω = 2πf
℘ag air gap permeance for the nth harmonics = 4 μ0 Rl (ns _ n − nr )

π n2 p2 g s_n= (10)
nr
R = rotor radius
l = active length ns _ n = 60 f /( pn ) (11)
g = effective air-gap
where,
Rotor resistance for the nth harmonics,
2 2
ns _ n is synchronous speed for the nth harmonics
12lTs k wn
r2 _ n = Rslot (6)
NR nr is the rotor speed
NR = no. of rotor slots.

Determination of iron loss comprises the following steps:
For single cage rotors,
a) The back emf of the motor at fundamental frequency and
at (6k±1)th harmonics where k=1, 2, 3…i.e. 5th, 7th, 11th, 13th,
hs hs 17th, 19th, ....till (Ns/p-1)th,
sinh 2 + sin 2
1 δn δn
Rslot = (7) En = 2πk wn fTsφn
wsσδ n cosh 2 hs − cos 2 hs (12)
δn δn where,
ws = major slot width φn is the flux per pole for the nth harmonics
σ = conductivity of aluminum bars En is the back e.m.f. for the nth harmonics (voltage across
Xm_n)
δ n = skin depth for the nth harmonics
b) Calculation of iron loss in the stator teeth (Ps_teeth) of the
hs = trapeze height of the slot
motor at fundamental frequency and at (6k±1)th harmonics,
Rotor reactance for the nth harmonics, φn
B n m _ st = (π / 2) (13)
( N s / 2np ) ws L
2 2
12lTs k wn where,
x2 _ n = ω Lslot (8)
NR B n m _ st is the peak value of flux density in stator teeth for the
nth harmonics
hs hs
sinh 2 − sin 2 ws
hd 1 1 δn δn is width of stator teeth
Lslot = μ0 + (9)
wd ω n wsσδ n cosh 2 hs − cos 2 hs L is effective core length
δn δn
Resultant value of magnetic flux density (peak) in stator teeth
hd = slot lip height n=n (14)

Bm _ st = ¦ B n m _ st
n =1
ICRERA 2014 350

c) Calculation of the iron loss in the stator core (Ps_core) at r is the stator bore radius
fundamental frequency and at (6k±1)th harmonics,
(φn / 2) The amplitude of the magnetic field disturbance is:
B n sc = (15)
wc L
2 θD
where, B = BB sin (19)
π 2
B n sc is flux density in stator core for the nth harmonics and,
wc is effective core depth Ns
ωe = ω B (20)
p
Resultant value of magnetic flux density (peak) in stator core
Tooth pulsation loss in rotor,
n=n
Bsc = ¦ B n sc (16) Ppulsation = PdWrt (21)
n =1
where,
d) Calculation of the iron loss in the rotor teeth (tooth
pulsation losses) at fundamental frequency and at (6k±1)th W rt is the weight of rotor teeth; and
harmonics,
e) The total iron loss is calculated by summing Ps_teeth, Ps_core,
ef eb Ppulsation
§ω · § B ·
Pd = PB ¨¨ e ¸¸ ¨¨ ¸¸ (17)
III. RESULTS AND DISCUSSION
© ωB ¹ © BB ¹
The contributions from fundamental, spatial harmonic
components, and tooth pulsation losses in iron losses for
where, various three phase induction motors with different stamping
designs have been determined using proposed method, and the
ef= 1.2 for 2 Pole motor, total iron loss simulated has been compared with experimental
values. The loss in rotor bars can be easily calculated using
1.0 for 4 Pole motor and above formula based on one dimensional solution. However, the loss
due to pulsation in rotor teeth is considered here as distributed
eb= 2.5 throughout the rotor teeth. Since the rotor teeth (in
conventional outer stator motor) are saturated more than the
Pd is the tooth pulsation loss (watts/kg) in rotor stator teeth, hence it is correct to consider that the rotor teeth
flux density to be 1.8 Tesla on an average. At this level of
BB is the peak value of flux density (wb/m2) in the air-gap saturation, the relative permeability falls to be 150
(approximately) for M45 material. Hence the skin depth (į) is
PB is the iron loss (watts/kg) at BB about 4.0 mm. Considering penetration of the flux up to 6į,
almost the entire rotor tooth height has to be considered as
It is well known that the frequency of tooth pulsation is susceptible to iron loss.
Ns/p times the fundamental. Hence in above formula, we
No-load tests have been done to calculate iron losses for
substitute Ns/p in place of ω e / ω B all the ratings. The interception of the graph (no-load loss-
copper loss) vs. voltage2 on the y-axis is friction & windage
The stator slot openings ‘modulate’ the space fundamental loss. Simulations have been done for 215kW to 975kW, 3.3
magnetic flux density. A slot opening angle (relative to the kV to 11 kV, 2Pole and 4Pole, motors with materials M45 &
slot pitch) can be estimated by the following expression: M36, and having different stamping designs. TABLE I shows
the comparison of calculated and experimental values of iron
losses with all components for 2 pole motors. It can be seen
wd N s
θD = (18) from the table that the loss due to the fundamental component
r is in the range of 29% to 37% of total iron loss, and that due to
where, spatial harmonics is in the range of 34% to 39%. The
contribution from tooth pulsation loss is in the range of 27% to
35% of total iron loss.
wd is the stator slot opening
ICRERA 2014 351

TABLE I COMPARISON OF COMPUTED AND EXPERIMENTAL OF IRON LOSSES FOR 2 POLE THREE PHASE INDUCTION MOTORS WITH
DIFFERENT STAMPING DESIGNS
S kW Voltage Stator teeth and core loss (kW) Tooth Total Fundame- Harmonic Tooth Iron % error
. (V) pulsation iron loss ntal comp.*100 pulsation loss (E-D)*100/E
N loss (kW) (kW) comp.*100/ /total iron loss*100 (tested
. Fundame- 5th+7th+ Total (C) D=A+B+ total iron loss /total iron value)
ntal comp. 11th+13th+ (A+B) C loss (B*100/D) loss (kW)
(A) 17th+19th+. (A*100/D)) (C*100/D) (E)
.. (Ns/p-1)th
harmonic
comp.
(B)
1 250 6600 1.391 1.309 2.700 1.001 3.701 37.6 35.4 27.0 3.828 3.3
2 500 6600 1.679 2.246 3.925 1.723 5.648 29.7 39.8 30.5 5.447 -3.7
3 280 6600 1.311 1.363 2.674 1.15 3.823 34.3 35.7 30.1 3.845 0.6
4 225 6600 1.125 1.179 2.304 0.95 3.254 34.6 36.2 29.2 3.395 4.2
5 354 6600 1.600 1.650 3.250 1.398 4.648 34.4 35.5 30.1 4.394 -5.8
6 280 6600 1.311 1.363 2.674 1.15 3.823 34.3 35.7 30.1 3.686 -3.7
7 225 6600 0.917 1.126 2.043 1.106 3.148 29.1 35.8 35.1 3.3 4.6
8 450 6600 1.843 2.182 4.025 2.22 6.244 29.5 34.9 35.6 6.305 1.0
TABLE II COMPARISON OF COMPUTED AND EXPERIMENTAL IRON LOSSES FOR 4 POLE THREE PHASE INDUCTION MOTORS WITH DIFFERENT
STAMPING DESIGNS
S. kW Voltage Stator teeth and core loss (kW) Tooth Total Fundame- Harmonic Tooth Iron % error
N. (V) pulsation iron loss ntal comp. pulsation loss (E-D)*100/E
loss (kW) (kW) comp.*100/ *100/total loss*100 (tested
Fundame- 5th+7th+ Total (C) D=A+B total iron iron loss /total iron value)
ntal comp. 11th+13th+ (A+B) +C loss (B*100/D) loss (kW)
(A) 17th+19th+. (A*100/D) (C*100/D) (E)
...(Ns/p-1)th
harmonic
comp.
(B )
1 925 11000 4.293 4.198 8.491 3.030 11.521 37.3 36.4 26.3 10.938 -5.3
2 215 6600 1.162 1.130 2.292 0.661 2.953 39.3 38.3 22.4 2.95 -0.1
3 260 6600 1.353 1.376 2.729 0.841 3.570 37.9 38.5 23.6 3.48 -2.6
4 600 6600 2.850 2.777 5.627 1.665 7.292 39.1 38.1 22.8 6.9 -5.7
5 975 6600 3.111 3.550 6.661 1.692 8.352 37.2 42.5 20.3 8.188 -2.0
6 350 3300 1.885 1.659 3.544 1.371 4.914 38.4 33.8 27.9 5.09 3.5
7 450 3300 2.370 1.814 4.184 1.395 5.579 42.5 32.5 25.0 5.297 -5.3
8 250 3300 1.246 1.093 2.339 0.905 3.244 38.4 33.7 27.9 3.172 -2.3
9 750 3300 2.654 2.124 4.778 2.530 7.308 36.3 29.1 34.6 7.146 -2.3
10 225 3300 1.311 1.042 2.353 0.669 3.023 43.4 34.5 22.1 2.878 -5.0
11 315 6000 1.531 1.356 2.887 1.029 3.916 39.1 34.6 26.3 4.08 4.0
ICRERA 2014 352

TABLE II shows the results for 4 pole motors. The losses [4] Bangura, J.F. Demerdash, N.A., Effects of broken bars/end-ring
from fundamental, spatial harmonic components, and tooth connectors and airgap eccentricities on ohmic and core losses of
induction motors in ASDs using a coupled finite element-state space
pulsation, are in the range of 37% to 43%, 29% to 42%, and method, IEEE Trans.-Energy Conversion, vol. 15, no. 1, pp. 40-47,
20% to 34% of total iron loss, respectively. The percentage March 2000.
errors between computed and tested iron losses are within [5] Enokizono, M.; Shimoji, H.; Horibe, T., Effect of stator construction of
three-phase induction motors on core loss, IEEE Trans.-Magnetics, vol.
±5% for 2 Pole and 4 Pole motors.
39, no. 3, pp. 1484-1487, May 2003.
[6] Jamil, M.K.; Baldassari, P.; Demerdash, N.A.; No-load induction
IV. CONCLUSION motor core losses using a combined finite-element/state-space model,
Losses because of rotating flux differ from those observed IEEE Trans.-Magnetics, vol. 28, no. 5, pp. 2820-2822, Sept. 1992.
[7] P.L.Alger, The Nature of Polyphase Induction Machines, John Willy &
under alternating flux conditions. The components of iron loss Sons, INC, NEW York ChaPman & Hall, Ltd, London, 1951.
(fundamental, spatial harmonics, and tooth pulsation) of the [8] Williamson W., Laithwaite E.R., Generalised Harmonic Analysis for
three phase induction motor have been determined using the the steady state performance of sinusoidally excited cage induction
proposed equivalent circuit considering all stator motors, IEE Proceedings Electric Power Application, vol. 132, No 3,
May 1985, pp. 157-163, May 1985.
magnetomotive force harmonics. This method is very useful [9] Poloujadoff, M; The Theory of Three Phase Induction Squirrel Cage
for day to day design work because it is faster and more Motors, Electrical Machines & power Systems, 13, pp. 245-264, 1987.
accurate than numerical methods. It is concluded that the [10] G. Bertotti, A. Boglietti, M. Chiampi, D. Chairabaglio, F. Fiorillo, M.
harmonic component of iron loss is in addition to its Lazzari, An Improved Estimation of Iron Losses in Rotating Electrical
Machines, IEEE Trans.-Magnetics, vol. 27, no. 6, pp. 5007-5009, Nov.
fundamental component and may go up to the same value as 1991.
of latter in many ratings, and can be estimated accurately by [11] G. Bertotti; Canova A.; M. Chiampi; D. Chiarabaglio; F. Fiorillo; A.M.
considering spatial harmonics in the lumped circuit model. Rietto, Core Loss Prediction Combining Physical Models with
The tooth pulsation loss is in range of 20% to 35% of the total Numerical Field Analysis, Journal of Magnetism and Magnetic
Materials, vol. 133, pp. 647-650, 1994.
iron loss for 2 pole and 4 pole motors. It is necessary to [12] Johan J. C. Gyselinck, Luc R. L. Dupre, Lieven Vandevelde, and Jan A.
determine the iron loss due to the fundamental component, A. Melkebeek, Calculation of No-Load Induction Motor Core Losses
spatial harmonics, and tooth pulsation at the design stage itself Using the Rate-Dependent Preisach Model, IEEE Trans.-Magnetics, vol.
for the optimized and cost effective design of three phase 34, no. 6, pp. 3876-3881, November 1998.
[13] Julius Saitz, Computation of the Core Loss in an Induction Motor Using
induction motors. the Vector Preisach Hyteresis Model Incorporated in Finite Element
Analysis, IEEE Trans.-Magnetics, vol. 36, no. 4, pp. 769-773, July 2000.
REFERENCES [14] Lei Ma, Masayuki Sanada, Shigeo Morimoto, and Yoji Takeda,
[1] Kunihiro, N.; Todaka, T.; Enokizono, M., Loss Evaluation of an Prediction of Iron Loss in Rotating Machines With Rotational Loss
Induction Motor Model Core by Vector Magnetic Characteristic Included, IEEE Trans.-Magnetics, vol. 39, no. 4, pp. 2036-2041, July
Analysis, IEEE Trans.-Magnetics, vol. 47, no. 5, pp. 1098-1101, May 2003.
2011. [15] Stranges N., "Methods for predicting rotational iron losses in three phase
[2] Kwon, S.O. Lee, J.J. Lee, B.H. Kim, J.H. Ha, K.H. Hong, J.P. Hong, induction motor stators”, IEEE Trans.-Magnetics, vol. 36, no. 5, pp.
Loss Distribution of Three-Phase Induction Motor and BLDC Motor 3112-3114, September 2000.
According to Core Materials and Operating, IEEE Trans.-Magnetics, [16] H. Wayne Beaty, James L. Kirtley, Electric motor handbook, McGraw-
vol. 45, no. 10, pp. 4740-4743, Oct. 2009. Hill, 1998.
[3] Jeong-Jong Lee; Soon-O Kwon; Jung-Pyo Hong; Ji-Hyun Kim; [17] P. L. Alger, Induction Machines, Gordon and Breach, 1969.
Kyung-Ho Ha, Core loss distribution of three-phase induction motor
using numerical method, Proc. 31st International Telecommunications
Energy Conference (INTELEC), 18-22th Oct. 2009, pp. 1-4.
ICRERA 2014 353

Iranian Journal of Military Medicine Vol. 14, No. 3, Autumn 2012; 168-177
Plagiarism: Concepts, Factors and Solutions

Bahadori M.1 PhD, Izadi M1 MD, Hoseinpourfard M1* PhD
1
Health Management Research Center, Baqiyatallah University of Medical Science, Tehran, Iran
Downloaded from militarymedj.ir at 4:41 +0430 on Friday August 31st 2018
Abstract
The goal of knowledge production is the discovery of facts and improving the human situation, and as such,
plagiarism and using other unethical means are not compatible with this goal. Most academic scholars agree that
plagiarism is a serious violation of publishing ethics. In recent decades, the scientific community has become
really concerned about the fast growth of plagiarism. Although plagiarism is widespread, it isn’t consistent with
the principles of science.
Nowadays some media publish worrying news of plagiarism in scientific publications, including data
manipulation by well-known scientists. The prevalence rate of plagiarism has been reported in different studies
turns out to be different in various fields, countries, educational levels and times.
The goal of this study is to review the scientific concepts related to plagiarism, its factors and roots, its
prevalence in the world and methods of detecting it in order to improve the awareness of instructors and
students of plagiarism.
Keywords: Plagiarism, Solutions, Concepts, Causes
Introduction tional demand for better results and more publica-

The goal of research is to produce knowledge, and tions in scientifically accredited journals [4].
the aim of producing knowledge is to improve hu- Plagiarism is a controversial issue in higher edu-
man situation while doing research using unethical cation, and it is increasingly widespread among
or inappropriate means leads to scientific corrup- students. Some challenges in academic activities
tion, which is against scientific knowledge pro- are due to the increase in the number of students
duction [1]. No doubt, there is plagiarism in the [5]. Today, open access publications are not only
scientific community although it is against basic reasonable but also very vital to scientific innova-
scientific principles. Plagiarism is useless, mean- tions. Unlimited access to scientific ideas, meth-
ingless, unethical and thus forbidden [2].one of the ods, findings and results is not compatible with the
pathological componenets in the relationships be- restricting regulations of copyright, and this has
tween people is the legal culture in society [3]. made for more plagiarism [6].
Unethical issues are quickly increasing in the realm “If plagiarism turns into an ordinary and usual ac-
of science. In the future, such issues in gathering tivity, it will affect the security of scientific knowl-
data, cooperation between scientists and in publi- edge and destroy all social realms. In such a situ-
cations will most probably get more complicated ation, nobody will bother doing research; rather,
and more difficult to deal with. More than ever be- everybody will make use of ready-made knowl-
fore, postgraduate medical students should know edge produced by the past researchers and will de-
about methods, technologies and concepts of sci- stroy all knowledge. Such unreasonable behavior
ence. The global competition among the scientists will devastate the foundations of scientific prog-
of developing countries, especially Asian ones, is ress and everything else. And if a country loses its
a new reality for the western researchers who want firm scientific foundations, it will remain in past
to be the best in all areas of research. Researchers achievements and will not experience progress”
in developing countries are increasingly enjoying [7].
more research budgets, and this development has Plagiarism is one of the important issues of univer-
been accompanied by governmental and institu- sities in recent years. In the last two decades, the
*
Corresponding author: Hoseinpourfard M.J., Please, direct all correspondence at hpf.javad@gmail.com.
Bahadori M. et al
progress in computer technology, that is, running giarists to take advantage of the situation [11].
websites to provide university services, the copy- Ben Jonson was the first one to sue the term pla-
paste tool, and loads of pre-fabricated papers, has giarism in the early 17th century. It was hard for
made for an increase in plagiarism [8]. Nowadays authors to protect their writings before devising
some media publish worrying news of plagiarism copyright laws. But as plagiarism increased in the
in scientific publications, including data manipula- 18th century and copyright laws were consequent-
tion by well-known scientists. The ethics of scien- ly clearly defined and devised by the middle of the
tific publication is in direct connection to the con- century, plagiarists faced a change in the public
cepts of copyright in writing scientific papers and opinion and strong ethical viewpoints towards pla-
of plagiarism. Sometimes, journal editors take the giarism [12].
writers’ cunningness for their lack of familiarity In view of the prevalence of plagiarism in the sci-
with journal regulations or their lack of attention entific community and its devastating effects on
to a certain paper. As Kosovsky notes, “the road to scientific progress, this study aims at surveying
hell is paved with good intentions” and after that, the concepts, causes and solutions to the issue of
the writers make very serious ethical mistakes to plagiarism.
the end [9]. Terminology, Definitions and Idioms
The author of a book, paper, poem or a scientific According to the Persian dictionary of Dehkhoda,
passage, after hours of thinking and writing about the word “steal” means “taking away somebody’s
a subject, puts to paper the fruit of years of his or possession with deception and tricks” or “to take
her continuous efforts. As such, the plagiarist not hold of something without the right to do so” [13].
only steals the fruit of such efforts but also reg- Wilson Mizner states that “when we steal an idea
isters all that painstaking work to his or her own from one author, it will be called plagiarism, but
name [7]. when we do it from a few authors, it is called re-
Plagiarism is hundreds of years old, but, due to search” [14].The word plagiarism comes from the
the progress in information technology, it has ac- word “plagarius”, meaning kidnapper, robber, mis-
quired new and different methods compared to the leader, and literary thief” [15]. Plagiarism usually
past. Plagiarism was almost a rare phenomenon refers to stealing ideas or words that are higher
until 1990, but it has spread across the world in than the level of public knowledge [16].
recent years and has worried the academic com- In Webster’s Dictionary, a plagiarist is defined as
munity [10]. In the past, there were a few scien- “One who plagiarizes, or purloins the words, writ-
tists who produced knowledge and some of them ings, or ideas of another, and passes them off as his
would produce no more than a couple of papers in own; a literary thief” [17], and plagiarism as “tak-
their lifetime. In those times, strict reviewing prin- ing someone›s words or ideas as if they were your
ciples were at work, there just a few journals and own”[18]. The University of Liverpool defines
scientists had a hard time convincing the scientific plagiarism as the “use of materials from unac-
community to accept their ideas. In the 19th cen- knowledged sources or direct quotation of materi-
tury, the problem was stealing ideas, and that was als from documented references without acknowl-
why many discoveries and inventions were dis- edging that the words have been taken verbatim
puted. Today, however, the number of scientists, from those references” [19]. Payer sees plagiarism
students, journals and papers has really increased. as “taking others’ ideas, words or wok as if they
While there is no problem with the increase in the were your own” [20]. Or as Stebel man puts it,
number of papers, peer-reviewing of the papers is plagiarism consists of “claiming as your own the
the main problem. It is certainly expected of a re- writings and research papers that originally belong
viewer to have a good command of the subject of to others [21]. Vessal and Habibzadeh take pla-
a paper. But, given the large number of papers to giarism to be “ascribing others’ ideas, processes,
be reviewed, are there enough specialists to review results and words to oneself without due acknowl-
the papers? No scientist can claim that he or she edgement” [22]. Using sentences from published
has studied all specialist papers in his or her area of medical literature with little change in the words
knowledge, and this paves the way for some pla- without acknowledging the source is also an in-
Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

stance of plagiarism. Using unpublished images or the same data in two papers (on with a clinical fo-
pictures with the owners’ permission is also called cus and one with a theoretical focus), and publish-
plagiarism [23]. ing the same paper under two names, one being the
The Federal Government of the United States de- real author in his or her own country and the other
fines “research misconduct as fabrication, falsifi- being a foreign author.
cation or plagiarism in proposing, implementing or Republishing, which is done in a deceptive way,
reviewing of research projects or in reporting the is certainly unacceptable. If editors, reviewers and
results of research” [24]. end readers of data notice the overlap between pa-
Plagiarism is an unethical activity in scientific pers, they can make the right decision about it. Du-
writing. For something to be called plagiarism, plicate publication is, nevertheless, deceptive and
it needs to be a serious deviation from normally involves three problems: it is unethical, it wastes
accepted behavior of the relevant scientific com- resources and it has adverse impacts upon future
munity which is done consciously and deliberately clinical and research decisions. Editors and readers
and must be proved with solid evidence. Plagia- of a published report want to make sure that they
rism may occur in different forms: stealing ideas are dealing with new and important data, and may
and stealing parts of texts. Self-plagiarism happens wrongly be persuaded to think so, while this is not
when an author uses his or her own previously pub- the case. Duplicate or redundant publication mis-
lished work without acknowledging it [25]. Self- leads the readers and reduces the credibility of the
plagiarism is defined in three ways in the relevant journal as well as its ability to attract good papers.
literature: 1) publishing a paper which basically Duplicate publication makes for wasting resources
overlaps another paper without due acknowledge- by wasting the time which should be allocated to
ment; 2) breaking a large paper into a few smaller other papers [28].
papers and publishing them separately, called sa- “Most academic researchers agree that plagiarism
lami slicing and 3) republishing the same work. is a serious problem in the ethics of publication.
Copyright, on the other hand, means enhancing Plagiarism appears in different forms: stealing
knowledge and useful arts by providing limited- ideas and stealing texts (verbatim plagiarism).
time security for authors and inventions through Plagiarism is no doubt an instance of misconduct.
exclusive rights regarding their writings and in- Stealing part of text and rephrasing it is a severe
ventions. Authors of technical papers are usually problem in the humanities and literature where in-
asked to transfer the copyright of their work to the novation in phrasing and eloquence are essential.
journal or the publisher [26]. But in the realm of science, it is the scientific con-
Scientific integrity depends on honesty and trans- tent itself, not its eloquence, that matters” [29].
parency of the methods of producing and trans- The purpose of scientific journals is to some extent
ferring knowledge [26]. Republishing results is different from that of non-scientific ones. For in-
announcing the same results in two or more pa- stance, medical journals are published in order to
pers, multiple recalculations of the same results in improve the science of medicine and public health
meta-analyses and as a result in serious errors in by publishing the results of scientific research. In
research [27]. many areas such as literature and humanities, how-
Duplicate or redundant publication occurs when ever, different authors have different views. They
there is an overlap, without acknowledging it, try to reflect their own understanding and feelings
between two papers in terms of their hypotheses, of texts by means of a selection of good and suit-
data, arguments or results. This could include an able words. Thus, each and every word, along with
overlap with other authors, their results or their its immediate context, has a role in conveying the
samples. The most important cases involve lack meaning to the reader. But in a scientific writing,
of acknowledging the sources. The following are the writer’s audience consists of scholars who are
example cases of republishing: publishing data looking for facts based on solid evidence. There-
which has been published before, reusing tables fore, the writer is supposed to observe and report
and figures in later publications, publishing larger correctly. Unlike literary researchers, a scientific
papers using previous smaller papers, publishing paper author should follow a certain and well-es-

Bahadori M. et al
tablished scientific method and make sure that he copying materials of one or more texts and pro-
or she will not be become biased in his or judg- viding the right citations without using quotation
ments since this can endanger the truth or reliabili- marks to make the readers believe that they have
ty of the judgments. Thus, whether or not he or she paraphrased the materials not quoted them, and re-
is eloquent, as far as an author is a just observer wording sentences from other sources without giv-
who works based on accepted scientific methods, ing credit to them [31].
evidence and facts, he or she can publish his or Recognizing plagiarism faces a number of prob-
her findings and could be said to have followed a lems. One problem is recognizing the amount of
universally accepted method [29]. plagiarism because it can cover a wide scope. The
Plagiarism, in general, includes attributing some- second problem is the question as to how much
body else’s work to yourself without giving credit change in the original material can make for pla-
to the author, copying other’s ideas or words with- giarism [31]. Roig argues that many students
out giving credit to the source, not putting quota- struggle between rewording and summarizing be-
tions in quotation marks, giving the wrong infor- cause they cannot distinguish between them. The
mation about a reference, changing the words while third issue is that most authors believe that there
keeping the structure of a sentence from another is no need to reference common knowledge, but
source without acknowledging it, and copying a we may ask what common knowledge is and who
large number of words or ideas from other sources defines it? [32]
with or without due acknowledgement [30]. Plagiarism can be divided into two types with
Another definition of plagiarism numerates the regard to intentions. The first type is intentional
ways of plagiarizing in the following way: “‘co- plagiarism where the author is fully aware of the
py-past’ which means verbatim copying of words, plagiarism and is willing to do it. The second type
plagiarizing ideas, which consists of using a con- of is unintentional plagiarism where a person pla-
cept or idea which is not commonly known to oth- giarizes due to his or her unawareness and lack of
ers, rephrasing, which means changing the gram- skill in writing. The latter type could be prevented
matical structure, using synonyms, reordering the [33].
original sentences, or rewriting the same content in In another classification, plagiarism is divided into
different words, artistic plagiarism, which denotes four categories: 1) “casual plagiarism, which oc-
presenting others’ works using a different medium curs because of lack of awareness of plagiarism,
such as text, voice, or image, plagiarizing codes, or insufficient understanding of referencing or ci-
that is, using other programs’ codes, algorithms tation;” 2) unintentional plagiarism, where, due
and functions without the right permission or ref- to the wide amount of knowledge in the scientific
erencing, using expired or neglected links, adding area, a person may unknowingly present ideas sim-
quotation marks or other referencing signs without ilar to those of others;” 3) intentional plagiarism,
providing the right referencing information or up- where a person deliberately and knowingly copies
dating links to sources, inappropriate use of quota- part or all of somebody else’s work without giv-
tion marks, failure to recognize the quoted parts of ing credit to them; and 4) self-plagiarism, which
a text, incorrect referencing, i.e., adding incorrect consists of reusing one’s own published work in a
referencing information or references which do different form with acknowledging it” [30].
not exist and plagiarism in translation, which con- The Prevalence of Plagiarism
sists of translating a text without giving reference Researches show that plagiarism is an increasingly
to the original text” [30]. widespread practice in educational and research
The following are some instance of student pla- institutes. The rate of plagiarism is different in
giarism: stealing material from a source and pass- various areas of research. As reported, the rates of
ing it for as their own, for instance, by buying a prevalence of plagiarism are 78% in the students
preordered paper, copying an entire paper without of Organizational Studies and 63% in the students
acknowledging it, presenting another student’s of humanities. Also, there is a meaningful differ-
work without their knowledge, presenting some- ence between the behavior of American students
body else’s paper and passing it as your own, and that of Hungary in terms of plagiarism [31].

Similarly, studies carried out by Park in the United tion showed some signs of duplicate publication.
States, South Africa and Finland reveal that the According to Tramer et al, 17% of the reports and
rates of plagiarism are different for different ar- 28% of patient data were duplicated and the inclu-
eas of study [31]. According to some research, the sion of redundant data in a meta-analysis led to a
number of plagiarizing students in an institute in- 23% overestimation of the treatment effectiveness
creased from 11% in 1963 to 49% in 1993. These of an antiemetic agent. Redundant publication can
results include all forms of plagiarism, including undermine the results of studies which are based
copying material from encyclopedias, journals, on reliable evidence. It can exaggerate the signifi-
papers and the like [34]. Jude Carroll argues that cance of the results and mislead the reader [28].
unacknowledged copying of materials from books According to a met-analysis by Fanelli, medical
and journals are more common than from web scholars report more cases of scientific misconduct
sites [35]. that those of fields of study [23]. The University of
According to some research, 12% of the papers Sao Paolo has appeared in the media on the sus-
suspected of plagiarism belong to the students of picion of plagiarism in scientific publication and
Politics. According to another researcher, in an research. Journals are concerned about fabrication
American university, 16.5% students report to or making up of data in published papers or dupli-
have plagiarized, and 50% of the students believed cation production of data or text by other authors
that their classmates often copy-pasted material without proper citation or referencing or even du-
from the Internet without acknowledging it [36]. plication of the published research or texts in other
Satterthwaite argues that the rate of plagiarism in papers [43].
America is 30% [37]. One study shows that 94% Factors of Plagiarism
of students had misconducted in their research for According to Ashworth, the concept of plagiarism
at least once, and another study shows this rate to is not clear enough so much so that some students
be 91% [38]. Dordoy, who has studied plagiarism are afraid of unwitting plagiarism while putting to
in the students of an English university, claims that paper what they take to be their own ideas [44].
the rate of copying a paragraph from a book or a Researches show that students and teachers have
web site was 73.9% [39]. different understandings of plagiarism. For some
A study focusing on plagiarism reveals that 48% of teachers, some definitions are influenced by higher
the students were not aware of the methods and re- education values such as the copyright, personal
quirements of referencing [40]. The results of some effort and unity in the university [45]. The multi-
research on academic misconduct tell us that 76% plication of databases, with all its benefits, has also
of students had responded positively to cheating caused a rapid growth in plagiarism. Some factors
in high school or college [40]. Carroll holds that affecting student attitudes toward plagiarism are
because most students do not know what makes ignorance, lack of personal investment in their
for plagiarism, they do not commit it with the in- education, situational ethics, and lack of consistent
tention of deceiving others [41]. A study in 2009 styles among and within various disciplines [46].
indicates that 212 papers showed some potential According to Dordoy, the most important factors
signs of plagiarism. In these papers, the similar- influencing plagiarism include promotion, laziness
ity between the original paper and the republished or mismanagement of time, easy access to materi-
one was 86.2% while the average of shared sourc- als on the Internet, unawareness of rules and regu-
es was 73.1%. Of the 212 papers, only 47 (22%) lations and unwitting plagiarizing [39].
cited the original paper. Also, there were miscal- Some other factors causing plagiarism are low
culations, contradictory data and manipulation of commitment to the learning process and focus-
figures in 47% of the papers [42]. ing on getting an academic degree, the student life
Bloemenkamp et al. report that 20% of the papers style, family pressures, etc. make students try to
published in Holland’s Journal of General Medi- achieve the best results with the least efforts and in
cine had already been published elsewhere. Simi- the least time [47]. In the past, students had to go
larly, Schein and Paladugu reported that one sixth to libraries, retrieve information and retype it while
of the papers appearing in three journals of opera- today and with the rapid progress of the Internet,

Bahadori M. et al
this process has changed and most teachers believe seem to believe that ‘they should get grades based
that computers have made it easier to cheat and on effort rather than on achievement’.
plagiarize [48].Angellil-Carter claims that there is 3. Time management. There are many calls on
no transparency about factors influencing plagia- student’s time, including peer pressure for an ac-
rism all over a university [49]. Dickert claims that tive social life, commitment to college sports and
not only are Hong Kong university students not performance activities, family responsibilities and
familiar with plagiarism but also it is very hard to pressure to complete multiple work assignments in
detect plagiarism in this university [50]. Informa- short amounts of time. Little wonder that Silver-
tion is easily accessible through electronic media man (2002) concludes that ‘students’ overtaxed
and word processing applications can easily copy- lives leave them so vulnerable to the temptations
paste material [51]. of cheating’.
In some countries, there is a lot of pressure on re- 4. Personal values/attitudes. Some students see
searchers to publish so that if they do not publish no reason why they should not plagiarize or do it
in journals with high impact factors or other in- because of social pressure, because it makes them
ternationally indexed journals, they will not get feel good or because they regard short cuts as clev-
promoted even if they have high instructional er and acceptable.
skills. This situation represents the familiar saying 5. Defiance. To some students plagiarism is a tan-
“Publish or perish.” Therefore, some scholars may gible way of showing dissent and expressing a lack
make ethical mistakes under the pressure to make of respect for authority. They may also regard the
progress and to hurry up with publishing [9]. task set as neither important nor challenging.
Cultural issues are specially considered in the 6. Students’ attitudes towards teachers and class.
problem of plagiarism. Cheating and plagiarism is Some students cheat because they have negative
an acceptable practice among the teachers and stu- student attitudes towards assignments and tasks
dents of countries where there is little awareness of that teachers think have meaning but they don’t
copyright [52]. A study reveals that students with a (Howard, 2002). Burnett (2002) emphasizes the
stronger belief in detecting plagiarism commit this importance of a relationship of trust between stu-
less than others and turn out to have better writing dent and teacher, because ‘the classes in which
skills, self-confidence and creativity [53]. Robert [students] are more likely to cheat … are those
Harris takes students’ looking for short cuts, their where students believe their professor doesn’t
low interest in the research subject, their low plan- bother to read their papers or closely review their
ning skills, mismanagement of time, lack of skills work’.
in scientific writing and their interest in ignoring 7. Denial or neutralization. Some students deny to
regulations as some of the reasons why students themselves that they are cheating or find ways of
take to plagiarism [54]. legitimizing it by passing the blame on to others
Another study shows that the following are among 8. Temptation and opportunity. It is both easier and
the most important reasons why students plagia- more tempting for students to plagiarize as infor-
rize: mation becomes more accessible on the Internet
1. Genuine lack of understanding. Some students and web search tools make it easier and quicker to
plagiarize unintentionally, when they are not fa- find and copy.
miliar with proper ways of quoting, paraphrasing, 9. Lack of deterrence. To some students the bene-
citing and referencing and/or when they are un- fits of plagiarizing outweigh the risks, particularly
clear about the meaning of ‘common knowledge’ if they think there is little or no chance of getting
and the expression ‘in their own words’. caught and there is little or no punishment if they
2. Efficiency gain. Students plagiarize to get a bet- are caught [31].
ter grade and to save time. Some cheat because of Some of the perceived obstacles to changing the
what Straw (2002) calls ‘the GPA thing, so that management of plagiarism are:
cheating becomes ‘the price of an A’ (Whiteman a reluctance by staff to process a case of suspect-
& Gordon, 2001). Auer &Krupar (2001) identify a ed plagiarism due to the time and workload in-
strong consumer mentality amongst students, who volved in proving‟ the plagiarism;” a reluctance

to become the one who dares to differ where it has some of these applications are free and they are all
been somewhat common practice to “turn a blind good for English texts. There are methods, how-
eye” to some relatively minor cases of plagiarism; ever, that can be used in any language. The Glatt
a perception that the University is reluctant to act plagiarism service, for example, is a computer ap-
on suspected plagiarism and that therefore the ef- plication which does not depend on correlation
fort expended by individual staff is likely to be techniques. It deletes every fifth word in a paper
fruitless in terms of dissuading or punishing pla- suspected of plagiarism and the author of the paper
giarism; a fear of risking collegial relationships is then asked to fill in the missing words. If the
with students by seeming or becoming authoritar- student can’t fill in 77% of the missing words, the
ian through a focus on minimizing plagiarism; a work is most probably plagiarized [17]. Wcopy-
concern that following through with cases of re- find is a free application on the Internet which can
peated plagiarism that may lead to student expul- be used to detect plagiarism. This software exam-
sion might damage the international reputation of ines a group of document files to compare their
the faculty or university; and a further concern that content [57].
such damage to reputation may result in reduced There are other tools such as http://ithenticate.com,
international enrolments; fear of harassment from http://www.crossref.org) and http://turnitin.com to
the student(s) accused of plagiarism and/or from discover plagiarism, but these tools can examine
their friends (such harassment occurred previously the papers indexed in MEDLINE only [42]. This
in the faculty); fear of student complaints if accu- area of study has been attended to by Turnitin and
sations of plagiarism are made (this had been an Safe Assign in the last 10 years. Kohler and We-
issue for some sessional staff who were concerned ber-wulf carried out a study in 2010 on 47 systems
that a student complaint might mean the end of of direct plagiarism detection and concluded that
their employment) [55]. only 5 of them were to some extent useful [58].
Detecting Plagiarism There are three approaches to detecting plagia-
Detecting plagiarism is hard and this makes pla- rism. The most common approach is by comparing
giarism a threat to the health of scientific literature. the document against a number of other documents
Often plagiarism is recognized by learned review- on a word by word basis. The second approach is
ers who possess up-to-date knowledge in their own by taking a characteristic paragraph and just doing
specialist filed [23]. a search with a good search engine like Google.
The following include some of the methods that And the third is by style analysis, which is usu-
can be used by researchers to detect plagiarism. 1) ally called stylometry [30]. Computer applications
General sight overview: the academic staff should reports cannot be simply relied here and there will
assess the sentence structure, grammar and idioms be need for specialist interpretation in such cases
used in the students’ assignments. They should ex- [30]. Detecting plagiarism is sometimes very dif-
amine the work which is lower or higher than the ficult, especially when rephrasing has occurred,
student’s abilities can afford; 2) Search of online when non-electronic sources have been used and
bookstores: these stores help the academic staff to when there is shift of language between the origi-
decide whether the students mentioned the right nal document and the plagiarized one [30]. Al-
dates for publications or whether the sources used though comparing abstracts is a good way to de-
were appropriate to the subject in hand; 3) Search tect plagiarism, a comparison of the full texts will
of keywords: searching keywords in search engines render better results [23].
is another tool in the hands of academics to find Systems of retrieving data or plagiarism detect-
instances if plagiarism, although today’s searching applications are capable of finding plagiarism
ing technology makes it possible for us to search where a verbatim copy of words has happened, but
a whole text, too; 4) a use of plagiarism services: what happens when the order of words has been
there are many software applications and tools and changed but not the overall meaning of the sen-
web sites that can help us detect plagiarized texts tences? Naturally, in such cases, the software will
[56]. Most of these tools use correlation techniques not be able to detect plagiarism and the plagiarizer
to detect similarities between documents. Only will be able to deceive it. Therefore, these systems

Bahadori M. et al
may become useless in the short run [11]. they know well about it and not try to learn more.
Strategies to Tackle Pagiarism Similarly, the staff may also wrongly think that
As Delvin points out universities do not like to students are well aware, and thus lose the opportu-
endanger their reputations for the sake of plagia- nity to teach students to avoid plagiarism [40]. Al-
rizers [59]. One of the measures needed to assure though the preventing approach takes more time, it
the quality of universities is to make sure that their is more effective than other approaches [64]. Ex-
assessing policies and activities are useful enough ercises and activities encouraged by staff have led
that their assessment is effectively examined in to good results. These activities include methods
terms of its validity, reliability and fairness. Some of appropriate citation, quotation, paraphrasing,
of the plagiarism preventing measures recom- phrasing and presenting some instances of plagia-
mended by quality assuring organizations include rism [40].
providing a definition of academic misconduct The attempt by some Australian universities to
with regard to plagiarism, cheating, identity fraud hide their management of plagiarismhas made it
and using inappropriate content” [60]. In order to difficult to share the best practices in this field.
effectively and fairly fight against plagiarism, stu- Although it seems that common policies are in
dents and staff need to have the same definition of it practice with respect to plagiarism, there is noth-
[61]. A preventing approach on the part of the staff ing to indicate the success of these policies. Del-
can eradicate this sort of misconduct and make for vin reports that for some, “catch and persecute”
academic progress and consistency in freshmen. leads to a decrease in plagiarism [59] while there
Along with practical approaches based on skills, is little evidence to suggest the effectiveness of
interactive prevention can not only improve the such a measure [52]. Gallant similarly argues that
students’ skill in referencing and citing and avoid- traditional methods of preventing plagiarism, such
ing plagiarism but also increase their awareness of as persecution as a preventing measure, honor
and sensitivity to this matter [62]. code systems and instructor detection are not ef-
University authorities are responsible for prevent- fective today [8]. In the UK, in order to minimize
ing plagiarism in all departments of their universi- plagiarism, they makes use of special courses, as-
ties. The universities’ policies in this regard should sessment, giving awareness to students, teaching
be clearly defined and announced and disseminat- student the necessary skills, detecting plagiarism,
ed among students and staff, preferably published persecution and special policies, and instructional
on the universities’ web sites, in libraries, student programs [65].
deputy offices, research centers and dormitories. Conway and Groshek have shown that student eth-
Academic staff should always talk to students ics is flexible and can be shaped at any level of
about academic values and avoiding plagiarism. education. Students showed in this research that
Students, on the other hand, should try to improve they are concerned about ethical violations and ex-
their skills in writing papers, research method- pect that severe punishment will be considered for
ology, and organizing data. University teachers those students who plagiarize or fabricate materi-
should encourage academic honesty in students, als.” Repeating anti-plagiarism actions at any stage
clearly define plagiarism for them, and point out of education can empower students [66]. MIT has
to them that they should reference the accessed defined good methods and policies to manage aca-
materials. University policies can also help staff to demic misconduct. In this university, teachers are
decide how to deal with plagiarism [17]. rewarded for teaching the right academic behavior
Burke points out that universities should focus on to students [67]. The Online Writing and Commu-
teaching student as to how they should avoid pla- nication Centerof this university runs a program
giarism [63]. The results show that teaching stu- for improving students’ writing skills and explains
dents, especially in the first year, is more effective different aspects of plagiarism [68]. In some uni-
than other ways of preventing plagiarism. Landau versities, such as Berkeley, people are academical-
argues that taking an active approach to plagiarism ly sanctioned for plagiarism [69]. In Stanford Uni-
is very important because students who are not versity, students learn about the university policies
fully aware of such misconduct may suppose that to deal with academic misconduct, copyright and

fair use of materials. important question;

Although there is always need for good inspec- Competing submissions of coworkers who dis-
tions, the responsibility to keep research integrity agree on analysis and interpretation of the same
lies with the scientific community itself and aca- study;
demic staff should make sure that students learn Articles from different groups of authors who have
about this integrity. Authors should guarantee that analyzed the same data. This is often the case with
their reported work is new and correct. Scholars very large administrative data sets or large national
who agree to review papers should feel responsible surveys sponsored by government agencies;
for doing informed, thorough and conscientious Republication of an article in another language
reviewing. Journal editor, who are themselves dis- with cross-referencing. There are mixed thoughts
tinguished scholars, should assure the originality on the acceptability of this practice. Typically the
of the material they publish [42]. The ideas and two (or more) journals need to work together and
thoughts of different thinkers and authors are in- often permission to publish is needed. The Interna-
evitable connected. So, it is a great responsibility tional Council of Medical Journal Editors has pub-
of authors to make sure that no plagiarism occurs lished criteria for this practice. While publication
when they publish their results. This means that the of data in an uncommon language need not neces-
authors must do whatever they can to ensure that sarily prevent it being presented in English, sec-
the words of their papers are theirs. They should be ondary publication should follow the International
always sure that it is clear for their readers whether Council of Medical Journal Editors guideline in
the ideas presented in the papers are theirs or oth- the uniform requirements [9].
ers’ and this could be clear by citing earlier pub- Strategies to Avoid Plagiarism
lished sources [24]. 1. Read the instructions for authors provided by
The process of peer-reviewing is the best mech- the journal.
anism to ensure the high quality of publications. 2. Always acknowledge the contributions of others
But recent studies have shown that lack of appro- and the source of ideas and words, regardless of
priate standards can result in duplicate publication whether paraphrased or summarized.
as well as publication of papers which include 3. Use of verbatim text/material must be enclosed
plagiarism [42]. At present, plagiarism tackling in quotation marks.
approaches focus on instructions to students and 4. Acknowledge sources used in the writing.
making them aware of the related policies and pos- 5. When paraphrasing, understand the material
sible outcomes. For instance, students are taught to completely and use your own words.
utilize to access and use sources in the right way. 6. When in doubt about whether or not the concept
Also, developing scientific integrity and honor or fact is common knowledge, reference it.
code systems are among good approaches to pla- 7. Make sure to reference and cite references ac-
giarism [70]. Carroll argues that teachers should curately.
focus on prevention [71]. McCabe similarly thinks 8. If the results of a single complex study are best
that reducing the chances of plagiarism is an im- presented as a cohesive whole, they should not be
portant tool in reducing scientific misconduct [72]. sliced into multiple separate articles.
Authors should bear in mind that it is not accept- 9. When submitting a manuscript for publication
able to republish a paper which has already been containing research questions/hypotheses, meth-
published, but this rule has the following excep- ods, data, discussion points, or conclusions that
tions, if the right disclosure is made to the editors have already been published or disseminated in a
and reader: significantmanner (such as previously published as
Prior publication in abstract form only (generally anarticle in a separate journal or a report posted on
<400 words); theInternet), alert the editors and readers. Editorss-
A study is too large and/or complex to be reported hould be informed in the cover letter, and reader-
in one article. A proposed rule of thumb is an ex- sshould be alerted by highlighting and citing thee-
pansion of the original article by 50%. However, arlier published work.
each article should address a different distinct and 10. When submitting a manuscript for potential

Bahadori M. et al
publication, if there are any doubts or uncertainty in Science and Technology. 2010;4(3):1-16.
about duplication or redundancy of manuscripts 2.DeVoss D, Rosati AC. “It wasn’t me, was it?”
originating from the same study, the authors should Plagiarism and the Web. Computers and Composi-
alert the editors of the nature of the overlap anden- tion. 2002;19(2):191-203.
close the other manuscripts (published, in press/ 3. Pakjou A, Izadi M, Masoudipour Sh, Fazel M.
submitted, unpublished) that might be part of them “Pattern of travel medicine ethics in international
an uscript under consideration. Augmenting old cooperation programs.” Journal of Military Medi-
data that was previously published with new ad- cine. 2011;13(2):117-123.
ditional data and presenting it as a new study can 4.Stewart Jr CN. Research Ethics for Scientists: A
be an ethical breach and should be fully disclosed Companion for Students: Wiley; 2011.
to the editors. 5.Roberts DM, Toombs R. A scale to assess per-
11. Write effective cover letters to the editor, espe- ceptions of cheating in examination-related situ-
cially regarding the potential for overlap in publi- ations.Educational and Psychological Measure-
cation. The cover letter should detail the nature of ment. 1993;53(3):755-62.
the overlap and previous dissemination and ask for 6.Brown GO. Out of the way: how the next copy-
advice on the handling of the matter. right revolution can help the next scientific revolu-
12. Become familiar with the basic elements of tion. PLoS Biol. 2003 Oct;1(1):E9.
copyright law [28]. 7.Beheshti SM. Plagiarism. Tehran 2011 [Feb, 7];
Available from: http://www.tebyan.ne.
Conclusion 8.Gallant TB. Academic integrity in the twenty-
Ethical problems in science are quickly increas- first century: A teaching and learning imperative:
ing and have become controversial issues in uni- Jossey-Bass Inc Pub; 2008.
versities and educational research institutes. These 9.Miziara ID. Ethics in scientific publications: the
problems have also been reflected in media news double copyright problem. Braz J Otorhinolaryn-
recently. The growth of information technology, gol. 2010 Sep-Oct;76(5):543.
competition between countries, rapid growth of 10.Hart M, Friesner T. Plagiarism and poor aca-
knowledge, fast multiplication of scientific jour- demic practice–a threat to the extension of e-learn-
nals, lack of good explication of plagiarism and ing in higher education? Electronic Journal on e-
different understandings of it, lack of awareness, learning. 2004;2(1):89-96.
mismanagement of time, and low culture etc. have 11.Foudeh P. Performance evaluation of informa-
all contributed to the prevalence of plagiarism in tion retrieval methods in dealing with plagiarism.
the scientific community. This has worried scien- Scientific Communication. 2009;2(3):1-5.
tific institutes and has made them react to it. Some 12.Mallon T. Stolen words: Forays into the origins
institutes focus on detecting and persecuting while and ravages of plagiarism: Penguin Books; 1991.
others concentrate on preventions and teaching the 13.loghatnamehDehkhoda. Tehran2012 [cited
right behavior. Excessive stress on detection of 2012 May, 12]; Available from: http://www.jasjoo.
plagiarism has made for the development of data com/books/wordbook/dehkhoda.
retrieving systems in recent years, but these are 14.Bartlett J. Familiar quotations: a collection of
not effective enough, and even if they were, they passages, phrases, and proverbs traced to their
would not be the best solutions. Effective preven- sources in ancient and modern literature: Little,
tion through proper education at the right time, Brown, and company; 1904.
proper interaction between teachers and students 15.Barnhart RK, Steinmetz S. Chambers Diction-
and devising appropriate policies for this purpose ary of Etymology–the origins and development of
are possible means of tackling plagiarism. over 25,000 English words.Chambers. Edinburgh,
uk; 1988.
References 16.Fialkoff F. There’s no excuse for plagiarism.
1.Tavakol M, rad MN. Plagiarism with explana- Library Journal. 1993;118((17)):56.
tion of the sociology of science .Journal of Ethics 17.Austin MJ, Brown LD. Internet plagiarism: De-
veloping strategies to curb student academ48.Ben

ic dishonesty. Internet and Higher Education. 33.Park C. Rebels without a clause: towards an in-
1999;2:21-34. stitutional framework for dealing with plagiarism
18.Conners M. Cybercheating: the Internet could by students. Journal of further and higher educa-
become the newest battleground in academic tion. 2004;28(3):291-306.
fraud. The Muse. 1996. 34.University BGS, Administrators NAoSP, Af-
19.College LHU. Undergraduate modular scheme fairs AfSJ, Integrity CfA. Academic Integrity: The
(awards of the University of Liverpool)—appen- Truth of the Matter: Bowling Green State Univer-
dices. 2012 [cited 2012 May, 15]; Available from: sity; 1995.
www.hope.ac.uk/compass/. 35.Carroll J. A handbook for deterring plagiarism

20.Pyer H. Plagiarism. 2012 [cited 2012 May,16]; in higher education: Oxford Centre for Staff and
Available from: at: http://online.northumbria. Learning Development Oxford; 2007.
ac.uk. 36.Kellogg AP. Students plagiarize online less
21.Stebelman S. Cybercheating: Dishonesty goes than many think, a new study finds. Chronicle of
digital. American Libraries. 1998;29(8):48-50. Higher Education. 2002;48(23):44.
22.Vessal K, Habibzadeh F. Rules of the game of 37.Satterthwaite T. Catching the cut-and-pasters.
scientific writing: fair play and plagiarism. The 2003 [cited 2003 May,18]; Available from: http:///
Lancet. 2007;369(9562):641. www.here.ac.uk/inside_he/archive/catching_the_
23.Sharma BB, Singh V. Ethics in writing: Learn- cut_and_past4311.cfm?archive=yes.
ing to stay away from plagiarism and scientific 38.CMJ S. Prevalence of plagiarism among medi-
misconduct. Lung India. 2011 Apr;28(2):148-50. cal students. Croat Med J. 2005;46(1):126-31.
24.Mason PR. Plagiarism in scientific publica- 39.Dordoy A, editor. Cheating and plagiarism: stu-
tions. J Infect DevCtries. 2009;3(1):1-4. dent and staff perceptions at Northumbria 2002.
25.Dellavalle RP, Banks MA, Ellis JI. Frequently 40.Landau JD, Druen PB, Arcuri JA. Methods for
asked questions regarding self-plagiarism: How to helping students avoid plagiarism. Teaching of
avoid recycling fraud. J Am Acad Dermatol. 2007 Psychology. 2002;29(2):112-5.
Sep;57(3):527. 41.Carroll J. Suggestions for teaching international
26.Hayes N, Introna LD. Cultural values, plagia- students more effectively. Learning and Teaching
rism, and fairness: When plagiarism gets in the way Briefing Papers Series, Oxford Brookes University
of learning. Ethics & Behavior. 2005;15(3):213- Retrieved April. 2002;1:2005.
31. 42.Long TC, Errami M, George AC, Sun Z, Garner
27.Kravitz RL, Feldman MD. From the Editors HR. Responding to possible plagiarism. Science.
Desk: Self-Plagiarism and Other Editorial Crimes 2009;323(5919):1293-4.
and Misdemeanors. J Gen Intern Med. 2010 Nov 43.Rode Sde M. Plagiarism in scientific publica-
9. tion. Braz Oral Res. 2011 Apr;25(2):101.
28.Cicutto L. Plagiarism: avoiding the peril in sci- 44.Ashworth P, Bannister P, Thorne P, Unit
entific writing. Chest. 2008 Feb;133(2):579-81. SotQRMC. Guilty in whose eyes? University stu-
29.Habibzadeh F, Shashok K. Plagiarism in scien- dents perceptions of cheating and plagiarism in
tific writing: words or ideas? Croat Med J. 2011 academic work and assessment.Studies in Higher
Aug 15;52(4):576-7. Education. 1997;22(2):187-203.
30.Maurer H, Kappe F, Zaka B. Plagiarism-a 45.Flint A, Clegg S, Macdonald R. Exploring staff
survey. Journal of Universal Computer Science. perceptions of student plagiarism. Journal of fur-
2006;12(8):1050-84. ther and higher education. 2006;30(02):145-56.
31.Park C. In other (people›s) words: Plagiarism 46.Auer NJ, Krupar EM. Mouse click plagia-
by university students literature and lessons.As- rism: The role of technology in plagiarism and the
sessment & Evaluation in Higher Education. librarian›s role in combating it. Library Trends.
2003;28(5):471-88. 2001;49(3):415-32.
32.Roig M. Plagiarism and paraphrasing criteria of 47.Macdonald R. Why don’t we turn the tide of
college and university professors. Ethics & Behav- plagiarism to the learners’ advantage.Times High-
ior. 2001;11(3):307-23. er Educational Supplement. 2000;24.

Bahadori M. et al
48.Benning V. Higher learning, lower behavior- Prosper, but Do They Get College Degrees? Col-
students cheat by computer. The Seattle Times. lege Student Affairs Journal. 1995;15(1):87-93.
1998:A7. 66.Conway M, Groshek J. Ethics Gaps and Ethics
49.Angélil-Carter S. Stolen language? Recherche. Gains: Differences and Similarities in Mass Com-
2000;67:02. munication Students’ Perceptions of Plagiarism
50.Hyland F. Dealing with plagiarism when giving and Fabrication. Journalism & Mass Communica-
feedback. ELT Journal. 2001;55(4):375-81. tion Educator. 2008;63(2):127-45.
51.Grover D. Plagiarism and the internet: the use of 67.Massachusetts Institute of Technology Policies
correlation techniques to detect plagiarism. Com- and Procedures. 2012 [cited 2012 May, 23]; Avail-
puter Law and Security Report. 2003;19(10):6-8. able from: http://web.mit.edu/policies/10.0.html.
52.Hamilton M. Managing student plagiarism in 68.MIT Online Writing and Communication Cen-
large academic departments. 2002. ter. 2012 [cited 2012 May, 23]; Available from:
53.Martin DF. Plagiarism and technology: A tool http://web.mit.edu/writing.
for coping with plagiarism. The Journal of Educa- 69.Berkeley University of California, Student
tion for Business. 2005;80(3):149-52. Conduct, Sanctions. 2012 [cited 2012 May, 23];
54.Harris R. Anti-plagiarism strategies for research Available from: http://students.berkeley.edu/osl/
papers. Virtual salt. 2004;4:6. sja.asp?id=1004.
55.Devlin M, editor. Policy, preparation, preven- 70.Barrett R, Malcolm J. Embedding plagiarism
tion and punishment: one faculty’s holistic ap- education in the assessment process. International
proach to minimising plagiarism2011: Swinburne Journal for Educational Integrity. 2006;2(1):38-45.
University. 71.Carroll J. Institutional issues in deterring, de-
56.Ryan J. Student plagiarism in an online world. tecting and dealing with student plagiarism. Joint
ASEE Prism Magazine. 1998. Information Systems Committee (JISC). 2004.
57.Bloomfield L. WCopyfind 2.1 instructions. 72.McCabe DL, Treviño LK, Butterfield KD.
2012 [cited 2012 May, 13]; Available from: Cheating in academic institutions: A decade of re-
http://www.plagiarism.phys.virginia.edu/WCopy. search.Ethics &Behavior. 2001;11(3):219-32.
find%202.1.html.
58.Sheard J, Dick M. Directions and Dimensions
in Managing Cheating and Plagiarism of IT Stu-
dents. 2012.
59.Devlin M. Problem with plagiarism. Campus
review. 2003;12(44):4-5.
60.QAA. Code of practice for the assurance of ac-
ademic quality and standards in higher education.
2000 [cited 2012 May, 15]; Available from: www.
qaa.ac.uk/public/Cop/COPaosfinal/contents.htm.
61.Stefani LAJ, Carroll J, Centre LG. A briefing
on plagiarism: Learning and Teaching Support
Network; 2001.
62.Dunn K. Recommendations for an Interactive
Approach to Plagiarism Prevention. 2011.
63.Burke M. Deterring plagiarism: A new role for
librarians. Library Philosophy and Practice (e-
journal).2005:10.
64.Compton J, Pfau M. Inoculating against pro-
plagiarism justifications: Rational and affective
strategies. Journal of Applied Communication Re-
search. 2008;36(1):98-119.
65.Mitchell CM, Wisbey ME. Cheaters Never

2013 International Conference on Renewable Energy and Sustainable Energy [ICRESE’13]
Long-Range Forecasting of Electricity Demand and

Supply for Maharashtra
Rajesh V. Kale* Sanjay D. Pohekar
Rajiv Gandhi Institute of Technology Tolani Maritime Institute,
Andheri (W), Mumbai 400058, India Pune 410507, India
rajeshkale_rgit@yahoo.co.in
India. Because of 8.4 per cent growth in economy from 2002
Abstract— Long term forecasting of electricity demand is vital onwards, the demand for electricity has gone up in the state.
for any developing economy to arrive at suitable mix of supply Due to this, the demand and supply gap has widened and the
side options. In this paper, Long-range Energy Alternatives state is facing a peak power shortage of 5000 MW. In many
Planning system (LEAP) has been used to forecast the electricity places people are enduring a power cut of 6 – 14 hours. The
demand and corresponding supply mix for the target year 2030 burden of electricity bills is also increasing because of poor
for Maharashtra, a prominent state in India with a base line year distribution network and higher commercial losses of the order
2012. The paper also presents an analysis of demand and supply of 35 per cent [3].Mumbai is the capital of Maharashtra and is
prior to 2012 for a period of six years. Power plant capacities are
called as the financial capital of India. Most of the major
projected for various transformation processes for the target
year. It is found that the industrial demand of the state is going to corporate and financial institutions headquarters are located in
be 34.5 percent of total electricity requirement. Coal based Mumbai.
power plants will continue to play major role in the forecasted Regulatory structure of state electricity sector has
years. undergone a major policy changes after 1990. This has
enabled the state to adopt strategies like unbundling of
Index term- LEAP model, energy forecast, electricity generation, transmission and distribution operations, private
scenarios sector participation in generation and distribution, rationalised
tariffs and reduced political influence. This could enable the
I.INTRODUCTION AND STATE OVERVIEW state to attract financial support from World Bank, Asian
Development Bank etc. This led to beginning of a new era for
the state electricity sector.
Electricity is a critical input for economic development of In this paper, LEAP framework is used to forecast the
any country. Developing countries like India are facing severe electricity demand and corresponding supply mix for the target
electricity crisis. Its widespread shortage is a basic obstacle to year 2030. LEAP is a widely used software tool for energy
economic growth. Maharashtra is a prominent state amongst policy analysis and climate change mitigation assessment. The
twenty eight states and seven union territories of India. The base line year is assumed to be 2012. The paper also presents
population of the state is 115.2 million [1] with a geographical an analysis of demand and supply prior to 2012 for a period of
area of 3.08 lakh square km. The state occupies western and six years to arrive at appropriate growth rates for various
central part of the country and has a long coastline of nearly sectors of economy in the state. Projected input and output
720 kilometers along the Arabian Sea.It is a prominent state in energy for various energy transformation processes are
India having a vibrant industrial and service sector economy. presented and analyzed. An analysis of estimated emission
Both the sectors together contribute about 89 per cent of the from various energy transformation processes is also carried
state’s gross domestic product. Though 55 per cent of the out.
population is dependent on agriculture and allied sectors, they
contribute just 11 per cent to the state’s income. Maharashtra is II. ANALYSIS OF ENERGY DEMAND
the most industrialized state and is favored for investment by By analyzing the pattern of required energy and energy
domestic as well as foreign institutions. Special economic demand met, we can quantify the shortages. The state demand
zones have been set up to attract investment in the industry is categorized under domestic, commercial, industry, railway
sector. The state has 226.1 lakh hectares of land [2] under and agriculture. Table 1 shows sectoral consumption of
cultivation and energized about 32 lakh agricultural pump electricity. The domestic sector includes Low Tension below
(MAHAGENCO, 2013) sets which is the highest number by poverty line (LT BPL), LT domestic and group housing. The
any state. Out of the total 23,830,580 households in commercial sector consists of LT commercial, temporary
Maharashtra, 83.94 per cent of the households have been connections, hoardings/advertisements, and
electrified (Directorate of census operations, Maharashtra, crematorium/burial. The high Tension (HT) consumers are HT
2011). The state consumption of electricity is 12 per cent of
978-1-4799-2075-4/13/$31.00 ©2013 IEEE 5

commercial and HT commercial complex. There are three IV. APPLICATION OF LONG-RANGE ENERGY ALTERNATIVE
categories under commercial sector viz. 0-20 kW, > 20kW and PLANNING (LEAP)
<= 50 kW, > 50 kW. The industrial sector comprises LT LEAP is an integrated modeling tool which can be used to
industry, LT power loom, HT industry (express) HT industry track energy consumption, production and resource extraction
(non-express) HT seasonal, and HT temporary supply. The in all sectors of an economy. Many authors have used LEAP
agriculture sector includes LT and HT agriculture, LT and HT to forecast the demand and create energy models. Park et al.
poultry. (2013) analyzed the energy, environmental and economic
influences of electricity scenarios in Korea[11]. Yophy et al.
TABLE1. SECTORAL CONSUMPTION OF ELECTRICITY FOR MAHARASHTRA
(MILLION KWH) [4-7] (2011) developed LEAP model to check demand and supply
patterns for different scenarios and GHG emissions [12]. Wang
Year Domestic Commercial Industry Railway Agriculture Total et. al. (2011) studied the implications of changing energy and
2007 15381.3 6004.5 28084.8 2024.1 5011.1 56505.8 environmental policies in China [13]. Vashishtha and
2008 16583.1 6375.0 32640.8 2090.0 5752.0 63440.9 Ramachandran (2003) analyzed the possibility of applying
2009 18177.7 10410.9 31379.1 2123.1 6271.8 68362.8 demand side management (DSM) programs in Indian utilities
2010 19484.2 10706.0 35028.7 2198.9 7780.8 75198.6 for Rajasthan [14]. However, successful implementation of
2011 21123.7 11735.3 36915.9 4394.8 10662.3 84832.0 DSM requires very high involvement of customers. Supply
2012 22305.1 13180..2 38932.8 2269.6 22931.4 99619.1 side management (SSM) is supposed to be more reliable
approach to ensure the futuristic demand and reduced GHG
Table 1. shows the sectoral consumption of Maharashtra. emissions for a leading state like Maharashtra.
Mumbai, the state capital is a major consumer of electricity. In
the year 2011, for residential sector 6,674 million kWh of
electricity was used in Mumbai, while rest of Maharashtra
consumption was 14,449 million kWh. The consumption of
electricity in the industrial sector in Mumbai region was 2,966
million kWh, for rest of Maharashtra it was 33,949 million
kWh. In commercial sector, the consumption in Mumbai
region was 3,350 million kWh while for rest of Maharashtra
was 5,558 million kWh. There is a 33 per cent growth in the
electricity consumption during the period 2007 to 2011.
III. ANALYSIS OF ENERGY SUPPLY
Electricity supply for Maharashtra state is divided into two

zones; Mumbai and rest of Maharashtra. Mumbai’s electricity
supply is provided by Reliance Infrastructure Limited (RIL),
Tata Power Company Limited (TPCL) and Brihanmumbai Fig 1 Installed capacity of power plant in MW [4,8,9]
Electric Supply and Transport (BEST). Maharashtra State
Electricity Distribution Company Limited (MAHADISCOM) TABLE 2. ELECTRICITY GENERATION IN MILLION KWH [10]
and some captive power plants supply electricity to rest of the Year Thermal Hydraulic Natural Renewable Combined Total
state. There is a growing electricity demand in the state. Energy Energy Gas Energy Power Plant
Encouraging state government policies brought private sector 2009 48687.45 5957.34 10945.24 2662.79 - 71492.82
investments in the power sector. To reduce load shedding, 2010 51927.45 5815.49 16869.74 2709.70 1830.69 79153.07
concerted efforts are made by the state utilities to set up new 2011 39401.29 7899.66 13955.09 3214.13 8918.14 73388.31
plants and the expansion of their existing power plants. Figure 2012 49997.75 5590.01 3906.95 3804.32 15059.53 78358.56
1 gives situation of the installed power plant capacity in the
state in the year 2011. The state has the total installed capacity
of 19,307 MW. The aggregate thermal power plant capacity in V.DATA FRAMEWORK AND ASSUMPTIONS IN LEAP MODEL
the state is 11,260 MW, hydraulic power plant capacity is
3,098 MW, natural gas power plants having 2,639 MW LEAP employs an end use, demand driven approach. It has
capacity and renewable plant capacity is 2,310 MW. various modules such as key assumptions, demand,
Table 2 gives the statistics of the electricity generation transformation, resources etc. In the key assumption module,
million kWh. There is an increase of 9.6 per cent in the state’s data such as total population, urban and rural population,
electricity generation in the year 2012 over the year 2009. In households, GDP and similar data is inputted. Demand analysis
the year 2012 electricity generation from thermal plant is 63.8 is used for modeling the requirements for final energy
per cent, from hydraulic plant 7.1 per cent; combined power consumption. In a transformation module, simulation of
plant is 19.21 per cent, natural gas power plant 4.98 per cent conversion and transportation of energy from the source of
and renewable energy 4.85 per cent.V primary energy to the final fuel consumption is done. In order
978-1-4799-2075-4/13/$31.00 ©2013 IEEE 6

to develop this model, the electricity consumption values for per cent
base year 2012 are substituted. In the demand module, broad
sectors such as domestic sector, industrial sector, commercial Life time in
sector, agriculture sector and railways are created for which the years 40 40 40 25
statistics is given in Table 1.
VI.RESULTS AND DISCUSSIONS
FORECASTED DEMAND
Energy demand calculations in LEAP differ depending on
whether final demand or useful demand analysis is conducted
on a set of branches. In a final energy demand analysis,
demand is calculated as a product of total activity level and
energy intensity at all technology branches. In the useful
energy analysis, demand is calculated using the final energy
intensity of the category, fuel share and efficiency of each
technology branch.
Following the data of baseline year 2012 and growth rates
calculated earlier, projected energy demand is calculated for
target year 2030. Figure 3 shows the estimated values of sector
wise electricity demand for the state in million kWh. For the
year 2030, the projected electricity demand is 212.7 billion
Fig. 2 Proposed LEAP model [15] kWh. The maximum consumption of electricity is for industrial
The explicit values of electricity demand for intermediate sector which is 34.5 per cent of the total electricity
years up to the year 2030 are substituted in the module. These requirement. The electricity requirement for domestic sector is
values are obtained by statistical method. This is done after 22.6 per cent; for commercial sector 17.5 per cent, agriculture
monitoring the trend, auto correlation function (ACF) and sector 21.7 per cent of the total electricity demand
partial auto correlation function (PACF) plots in SPSS (
statistical Product and Service Solutions) and SAS (Statistical
analysis system) the forecasts were computed using Holt's
Exponential Smoothing method i.e. ARIMA (0,1,1) (Auto-
regressive integrated moving average) The forecasts were
validated as the best fit depending on Akaike information
criterion (AIC), Schwarz Information Criterion(SIC) and R-
Squared values (Montgomery et. al., 1994). Exponential
smoothing method can be applied to time series data to
produce smoothed data for presentation or to make forecasts.
The time series data are a sequence of observations. In simple
moving average the past observations are weighted equally,
exponential smoothing assigns exponentially decreasing
weights over time.
In the transformation module, the electricity supply side
data such as installed capacity in MW, historical production in
million kWh, process efficiency, life time of the power plant
etc. are substituted. Table 5 lists the assumptions made in the
Fig 3. Sector wise electricity demand
LEAP model. It is assumed that the electricity transmission and
distribution loss is 21.3 per cent [4,10]. The first simulation
year is set as 2013. The planning reserve margin is assumed as
30 per cent [10,14]
TABLE 3. ASSUMPTIONS USED IN LEAP [11, 12, 16]
Coal
Hydro electric Natural gas Renewable
based
plant plant energy
plant
Process
efficiency per
cent 30 100 35 100
Maximum
availability 47 32.5 64.9 15.9
978-1-4799-2075-4/13/$31.00 ©2013 IEEE 7

[5] MERC BEST, 2012. Maharashtra Electricity Regulatory
Commission, MERC tariff order in case no. 95 of 2009 in the
matter of BEST dated September 12, 2012.
[6]MERC TPC, 2010.Maharashtra Electricity Regulatory
Commission, MERC tariff order in case no 98 of 2009 in the
matter of TPC dated September 12, 2010.
[7]MERC REL 2012.Maharashtra Electricity Regulatory
Commission, MERC tariff order in case no. 180 of 2011 in the
matter of REL’s seeking approval aggregate revenue
requirement and determination of tariff for FY 2011-12.
[8] MAHAHYDRO, 2013. Government of Maharashtra Water
Resources Department, Hydro-Electric Project, Office of the
Chief Engineer (Electrical) Mumbai.
http://mahahydro.org/priv/3.0%20UPDATED%20LIST.pdf
[Last visited on 24.06.2013]
[9] MEDA, 2013. Maharashtra Energy Development Agency
http://www.mahaurja.com/PDF/BP PG Renewable.pdf [Last
Fig. 4. Installed capacity of power plants visited on 24.06.2013]
[10] MAHASLDC, 2013. Maharashtra State Electricity Transmission
FORECASTED SUPPLY MIX Company Limited, Maharashtra. State Load Dispatch Center,
Kalwa. http://mahasldc.in/wp-content/reports/dr0.pdf [Last
The supply side data in terms of installed capacity is given
visited on 24.06.2013]
in figure 4. The aggregate of the installed capacity of the
[11] Park, N. Yun S. Jeon, E., “ An analysis of long-term scenarios
electricity generator in the year 2030 estimated by the model is
for the transition to renewable energy in the Korean electricity
52449.2 MW. The coal based plant tops with a share of 76 per sector”. Energy Policy 52, 288-296, 2013
cent, hydraulic power plant 9.8 per cent, renewable energy 9
[12] Yophy, H., Jeffrey, B., Yu, P., “The long term forecast of
per cent, gas based power plant 5 per cent. Taiwan’s energy supply and demand: LEAP model application”.
Energy Policy 39, 6790-6803, 2011
VII.CONCLUSIONS
[13] Wang, Y., Gu, A., Zhang, A., 2011. “Recent development of
energy supply and demand in China, and energy sector
In the preceding text of paper forecasting of electricity demand prospects through 2030”. Energy Policy 39, 6745-6759, 2011
is done for the year 2030 by following baseline scenario of [14] Vashishtha, S., Ramachandran, M., “Planning for demand-side
2012. Projected electricity demand is estimated to be 212.7 management programmes in Indian utilities: A case study of
billion KWh for the year 2030. To meet this demand, suitable Rajasthan”. Pacific and Asian Journal of Energy 13(1), 25-36,
supply side mix is proposed based on data input from various 2003
sources. The projected installed capacity approximately is [15] Heaps, C., Erickson, P., Kartha, S., Kemp-Benedict, E., 2009.
52,450 MW. It is also observed that coal based power plant Europe Share of the Climate Challenge; Domestic Actions and
will dominate with 76 per cent share followed by hydraulic International Obligations to Protect the Planet. Stockholm
power plant 9.8 per cent, renewable power plant with 9 per Environment Institute.
cent and natural gas power plant 5 per cent. [16] Dagher, L., Ruble, I., “ Modeling Lebanon’s electricity sector:
Alternative scenarios and their implications”. Energy 36, 4315-
4326, 2011
REFERENCES
[1] Government of India, Ministry of Home Affairs, Office of the
Registrar General and Census Commissioner, India.
http://www.censusindia.gov.in/2011census/PCA/PCA_Highlight
s/pca_highlights_file/India/Chapter-1.pdf [Last visited on
24.06.2013]
[2] Directorate of Economics and Statistics, Planning Department,
Government of Maharashtra, Mumbai. Economic Survey of
Maharashtra 2011-12.
[3] Kale, R., Pohekar, S.,” Electricity demand supply analysis:
Current status and future prospects for Maharashtra, India.”
Renewable and Sustainable Energy Reviews 16, 3960-3966,
2012
[4] MAHAGENCO, 2013. Maharashtra Power Generation Company
Limited. http://www.mahagenco.in/index.php/about-us [Last
visited on 24.06.2013].
978-1-4799-2075-4/13/$31.00 ©2013 IEEE 8

2015 International Conference on Nascent Technologies in the Engineering Field (ICNTE-2015)
Waste Sunflower Oil as an Alternative Fuel for Diesel

Engines
Abhishek Pavani, Akash Hebale, Veraj Poojary, Kapse Neeta

Shashank Parulekar and Chaudhari Kiran Dept. of Applied Sciences
Dept. of Mechanical Engineering MCT’s Rajiv Gandhi Institute of Technology
MCT’s Rajiv Gandhi Institute of Technology Mumbai, India.
Mumbai, India.
Abstract— Due to the depletion of fossil fuels and this oil needs to undergo refinement before it can be used in
increased awareness of environmental problems, the world is diesel engine.
looking to use alternative fuels. This paper is aimed at the M. Senthil Kumar, M. Jaikumar (2014) used waste cooking
conversion of waste edible oil into diesel fuel, which can then be oil (WCO) and its emulsion and tested it on a single cylinder
used in Compression Ignition (CI) engines. Production of edible water cooled engine. Base data was generated with diesel and
oil was 7.6 million tonnes in 2013-14. Sunflower oil is used in
majority of the houses and restaurants in India. So, the
neat WCO as fuels. Subsequently, WCO oil was converted
availability of waste sunflower oil in restaurants and houses is into its emulsion and tested. Neat WCO resulted in higher
more compared to any other edible oil. Therefore waste smoke, hydrocarbon and carbon monoxide emissions as
sunflower was considered as potential oil which could be used as compared to neat diesel. Significant reduction in all emission
an alternative to neat diesel fuel. The properties of waste was achieved with the WCO emulsion.
sunflower oil (WSO) were determined, which includes its specific Prashant Singare and P.V. Walke (2010) carried out the
gravity, kinematic viscosity, flash point, fire point, calorific value, performance analysis of soyabean and sunflower oils as C.I.
cloud point and pour point. The waste oil was initially refined engine fuel by reducing viscosity through Heat exchanger. No
using transesterification process and again the physical transesterification of oils was carried out. Results showed
properties of refined waste sunflower oil (RWSO) were
determined. An experimental study was conducted to evaluate
decrease in smoke opacity; decrease in BSFCand an increase
and compare the performance and emissions of different blends in brake thermal efficiency.
(B15,B20 and B25) of RWSO on a four stoke diesel engine. The Milan Tomic Lazar, Savin Radoslav, Micic Mirko
results indicate that blend B20 is an optimum fuel blend in terms Simikic, Timofej Furman (2014) carried out transesterification
of increased engine performance and reduced emissions process and the biodiesel obtained from sunflower oil was
compared to neat diesel fuel. blended with fossil diesel which did not have additionally
added additives for the improvement of lubrication properties.
The results indicated increased specific fuel consumption and
Keywords—Transesterification, waste sunflower oil, Biodiesel,
decreased thermal efficiency, decrease in CO2 and CO
Methyl Ester, Alternative fuels.
emissions and improved lubrication.
Jehad A. Yamin,Nina Sakhnini,Ahmad Sakhrieh,M.A.
I. INTRODUCTION Hamdan (2013) made an environmental and performance
The growing concern of the increasing demand for fossil study of a 4 stroke CI engine powered with waste groundnut
fuel and its increasing harmful effect on the environment has oil. Transesterification process was used for the refinement of
led to the search for a better environment friendly fuel. A lot of sunflower oil and then both used and unused oils were tested
research has been done in the past decade where various types on CI engine. Results showed lower thermal efficiency and
of vegetable oils have been used as an alternative diesel fuel. increase in CO2 emissions of biodiesel compared to pure
This research mainly aims at using Waste Sunflower Oil diesel.
(WSO) which is generally discarded after its use in restaurants K.A. Antonopoulos, D.C.Rakopoulos, D.T. Hountalas,
and houses as an alternative to diesel fuel. The sunflower E.G.Giakoumis (2006) made a comparative performance and
oilseed productivity in India was 706 kg/hectare in the year emissions study of a direct injection Diesel engine using
2012[1]. Though the overall sunflower oil production is less in blends of Diesel fuel with vegetable oils or bio-diesels of
India compared to the production of other oils, it is still used in various origins. Results showed that NOx and CO emission is
majority of the restaurants and houses as cooking oil. The used reduced .The smoke density was reduced with use of bio
sunflower oil is acidic in nature and also the viscosity of the oil diesel and smoke density was increased with the use of
is very high compared to conventional diesel fuel Table 1. So, vegetable oil. For higher load, engine shows high specific fuel
consumption.
978-1-4799-7263-0/15/$31.00 ©2015 IEEE

This study aims at the comparison of performance and II. FUEL CHARACTERIZATION AND
emissions of a 4 stroke diesel engine powered by different EXPERIMENTATION
blends (B15, B20 and B25) of waste sunflower oil obtained The properties of the WSO and refined waste sunflower oil
from various household and restaurant sources. (RWSO) were determined and it can be clearly seen from
Laforgia D., Ardito V., Biodiesel Fueled IDI Engines: Table 1 that the viscosity of WSO is almost 14 times of that of
Performances, Emissions and Heat Release Investigation, Bio- pure diesel. And after transesterification process, the viscosity
resource Technology (1995). Laforgia D., Ardito V., reported reduced considerably. Also it can be seen from the table that
that the higher viscosity of biodiesel enhances the lubricating the calorific value of RWSO is much less compared to pure
property and excess oxygen content results better combustion. diesel.
The higher oxygen content of biodiesel causes more carbon
dioxides at tailpipe which is a product of better combustion. Table 1 : Comparison of properties of pure diesel, WSO, RWSO.
The better combustion decreases the harmful CO and Properties Pure diesel WSO RWSO
hydrocarbon at the tailpipe. On the other hand, and oxides of Kinematic Viscosity
2.44 33.2 4.363
nitrogen emissions are increased with biodiesel fuel under @ 25 ◦C (cSt)
normal operating conditions. Flash point ( C) 71 139 143
Fire point( C) 103 152 149
M. Ziejewski, H. Goettler and G. L. Pratt, Comparative Cloud point( C) 3 -1 -2
analysis of the long-term performance of a diesel engine on Pour point( C) -5 -6 -8
vegetable oil based alternate fuels, SAE Paper (1986). M. Density(kg/m3) 835 919.2 887.3
Ziejewski, H. Goettler and G. L. Pratt, were evaluated that 25- Calorific
44000 37900 38100
75 blend (v/v) of alkali-refined sunflower oil and diesel fuel, a Value(kJ/kgK)
25-75 blend (v/v) of high oleic safflower oil and diesel fuel, a
non-ionic sunflower oil-aqueous ethanol micro emulsion, and The WSO couldn’t be directly used in the engine as
a methyl ester of sunflower oil as fuels in a direct injected, previous study showed that it led to gum formation and smoky
turbo-charged, intercooled, 4-cylinder Allis-Chalmers diesel exhaust due to improper combustion. So, the WSO had to be
engine during 200-hour EMA cycle laboratory screening transesterified, blended with pure diesel in various proportions
endurance tests. The results indicated that time had a and then tested on the engine.
significant effect only on exhaust temperature. In all other The biodiesel was made from WSO, which was initially
cases, time was not a factor. However, significant differences heated to about 120 C to remove the water content from the
in the intercepts of the prediction equations were found oil. After that a calculated exact quantity of methanol and KOH
between tested fuels. (lye) was added to the oil and the entire mixture was constantly
Perkins LA, Peterson CL, Auld DL. Durability testing of stirred with addition of heat. The temperature of the mixture is
transesterified winter rape oil (Brassica Napus L.) as fuel in kept to about 50 C. After stirring it for 1 hour, the mixture is
small bore, multi-cylinder, DI, CI engines. SAE paper No. allowed to settle for about 4-6 hours which separates out the
911764. Warrendale, PA: SAE, (1991). glycerol from the mixture and hence pure biodiesel is obtained.
Perkins LA, Peterson CL, Auld DL. Tests were conducted Fig. 2 shows the biodiesel process flow diagram.
to evaluate the performance and exhaust emissions of an Oils are made up of triglycerides which are esters of free
agricultural tractor engine when fueled with sunflower oil, fatty acids with trihydric alchol, glycerol. In transesterification
rapeseed oil, and cottonseed oil and their blends with diesel process, the triglyceride molecule is broken with the help of
fuel (20/80, 40/60 and 70/30 volumetrically). Tests were also catalyst (KOH) and methanol, which separates the free fatty
carried out with diesel fuel to be used as a reference point. acids and the glycerol. Methanol molecule then gets attached to
Engine power, torque, BSFC, thermal efficiency, NOx and the free fatty acid chains forming methyl esters. Fig.1
CO2 were recorded for each tested fuel. represents the exact chemical formula of the reaction.
Shirneshan A, Almassi M, Ghobadian B, Borghei A.M,
Najafi G.H. Effects of Biodiesel and Engine Load on Some
Emission Characteristics of a Direct Injection Diesel Engine,
September 17, (2012) Shirneshan A, Almassi M, Ghobadian
B, Borghei A.M, Najafi G.H. performed the experiments at the
rated torque speed of 1800 rev/min, and at 25%, 40%, 65%
and 80% engine loads. At each engine load, experiments were
Fig 1: Transesterification reaction process
carried out for diesel and each blended fuel. In this paper, the
effects of engine load and biodiesel on emissions included
HC, CO, CO2 and NOx were investigated. The quantity of methanol and KOH used in the process is
variable for different types of fatty acids; therefore it is
necessary to carry out titration process, which enables us to add
the exact amount of KOH required for transesterification of a
particular batch of biodiesel.
noted. Based on this parameters like brake specific fuel

consumption, indicated specific fuel consumption, brake
power, indicated power, mechanical efficiency, brake thermal
efficiency were calculated. An exhaust gas analyzer was used
to calculate the amount of exhaust gases coming out of the
engine at varying loads.
Table 2: Engine Specifications of 4 stroke CI engine
Rated engine speed 1500rpm
Stroke length 110mm
Bore diameter 87.5mm
Orifice diameter 25mm
Flow rate of cooling water 1.8lit/hr

in jacket
Fig.2: Biodiesel Process flow diagram
The following steps were used for the titration process:

1. 1 gm of (85% pure) KOH was added to 1 litre of
distilled water to prepare the titrating solution.
2. 10 ml of isopropyl alcohol were mixed with 1 ml of
WSO and thoroughly mixed.
3. The mixture was then titrated with KOH solution until
the color changed to dark pink, and remained for 15 s.
The quantity of KOH is then calculated and that exact

Fig. 3: Separated glycerol and biodiesel.
amount of KOH is added during the transesterification
reaction.
The biodiesel obtained from the reaction was then used in
various proportions with pure diesel and blends B15, B20 and
B25 were formed. These blends were then tested on a 4 stroke
CI engine operating at varying loads and constant speed (1500
RPM).
1 Engine 2 Base frame

3 Brake dynamometer 4 Fuel tank
5 Load gauge 6 Air box
7 Coupling 8 Manometer
9 Flywheel 10 Cooling water
11 Exhaust line 12 Gas analyzer
The schematic diagram of the engine setup is shown in
Fig.4. The engine specifications are mentioned in Table 2. The Fig. 4: Schematic diagram of engine
engine is coupled to a rope brake dynamometer which is used
to vary the load on the engine. The speed of the engine was
measured using a tachometer. During operation, the time taken
for 10 cm3 of fuel consumed and the manometric readings are
III. RESULTS AND DISCUSSION C. Effect on Mechanical efficiency
Performance analysis:
A. Effect on Brake thermal efficiency
Fig. 7: Mechanical efficiency vs Load
In general, as the load increases the mechanical efficiency

of each of the fuel increases. But when compared to neat
Fig. 5: Brake thermal efficiency vs Load
diesel, the mechanical efficiency of blends B15, B20 and B25
is very high. B25 and B15 have a higher mechanical efficiency
It can be clearly seen from Fig.5 that at lower loads, the compared to B20.
brake thermal efficiency (BTE) of pure diesel is higher than
other RWSO blends. But at higher loads, BTE of blends B20
and B25 are higher than that of pure diesel. As predicted, the
Emission Analysis:
BTE of biodiesel increases with increase in load.
A. Effect on Hydrocarbon emission
B. Effect on Brake Specific Fuel Consumption
Fig. 8: Hydrocarbon (HC) vs Load

Fig. 6: Brake Specific Fuel Consumption (BSFC) vs Load
The unburned hydrocarbon emission trends for all
It is observed in Fig. 6 that the BSFC of RWSO blends biodiesel blends and diesel are shown in Fig.8. It is observed
decreases with increase in load. Since BSFC is inversely that HC emissions increase as the load on the engine increases
proportional to calorific value, therefore BSFC of RWSO is for neat diesel as well as all RWSO blends. The HC emissions
less than that of diesel fuel at higher loads. Blend B20 can be of biodiesel in general are less as compared to neat diesel. B15
seen having the minimum BSFC at higher operating loads. The was found to have the lowest HC emission.
initial prediction of lower BSFC at higher loads was found to
be true.
B. Effect on carbon dioxide emission IV. CONCLUSIONS

The performance and emissions characteristics of a single
cylinder, 4-stroke diesel engine running at constant RPM
powered by neat diesel and RWSO blends in different
proportions were found and following are the conclusions
based on experimental results obtained:
• Brake thermal efficiency of B20 is superior to other
blends for all load conditions.
• BSFC for B20 is also lower as compared to other
blends at higher operating loads.
• Mechanical efficiency of B20 is superior to diesel and
other blends.
• Hydrocarbon and Carbon monoxide emission is lower
in case of B15, while it is higher in case of neat diesel.
Fig. 9: Carbon di-oxide (CO2) vs Load • Carbon dioxide emission is lower in case of B20 at all
load conditions.
The figure indicates the emission in percentage of CO2 As the properties of RWSO are found to be close to diesel
versus increasing load on the engine. It can be seen that the fuel it can used as substitute fuel for diesel engine and also we
CO2 emissions were slightly higher for neat diesel as found that B20 is an optimum fuel blend in terms of increased
compared to various biodiesel blends. B20 was found out to engine performance and reduced emissions compared to neat
have moderate CO2 emission level. diesel fuel.
C. Effect on carbon monoxide emission REFERENCES

[1] Dr. R.S.Paroda, “The Indian Oilseed Scenario: Challenges and
Opportunities”, August (2013).
[2] Prashant Singare & P.V. Walke., “Performance analysis of
soyabean and sunflower oils as C.I. engine fuel by reducing
viscosity through Heat exchanger”, Oct (2010).
[3] M. Senthil Kumar, M. Jaikumar., “A comprehensive study on
performance, emission and combustion behavior of a
compression ignition engine fuelled with WCO emulsion as
fuel”, May (2014).
[4] Milan Tomic Lazar, Savin Radoslav, Micic Mirko Simikic,
Timofej Furman, “Possibility of using biodiesel from sunflower
oil as an additive for the improvement of lubrication properties
of low-sulfur diesel fuel” ,Jun (2014).
[5] Jehad A. Yamin, Nina Sakhnini, Ahmad Sakhrieh, M.A.
Hamdan, “Environmental and performance study of a 4-
Fig. 10: Carbon Monoxide (CO) vs Load
Stroke CI engine powered with waste oil biodiesel”, March
(2013).
The variation of carbon monoxide for biodiesel blends and
[6] K.A. Antonopoulos, D.C.Rakopoulos, D.T. Hountalas,
neat diesel for different engine loads can be seen in Fig.10. E.G.Giakoumis, “Comparative performance and emissions study
The CO emission decreases with increase in load. The higher of a direct injection Diesel engine using blends of Diesel fuel
percentages of CO can be attributed to local rich regions and with vegetable oils or bio-diesels of various origins”, Mar
poor mixture formation in some locations of the combustion (2006).
chamber. The biodiesel blends have lower values of CO [7] M.S. Shehata, S.M. Abdel Razek, “Experimental investigation of
compared to that of pure diesel. diesel engine performance and emission characteristics using
jojoba/diesel blend and sunflower oil” ,July (2008).
[8] Laforgia D., Ardito V., Biodiesel Fueled IDI Engines:
Performances, Emissions and Heat Release Investigation, Bio-
resource Technology (1995).
[9] M. Ziejewski, H. Goettler and G. L. Pratt, Comparative analysis
of the long-term performance of a diesel engine on vegetable oil
based alternate fuels, SAE Paper (1986).
[10] Perkins LA, Peterson CL, Auld DL. Durability testing of
transesterified winter rape oil (Brassica Napus L.) as fuel in
small bore, multi-cylinder, DI, CI engines. SAE paper No.
911764. Warrendale, PA: SAE, (1991).
[11] Shirneshan A, Almassi M, Ghobadian B, Borghei A.M, Najafi

G.H. Effects of Biodiesel and Engine Load on Some Emission
Characteristics of a Direct Injection Diesel Engine, September
17, 2012
[12] K. Naima & A. Liazid “Waste oils as alternative fuel for diesel
engine”,March 2013.
[13] A.B.M.S. Hossain and A.N. Boyce “Biodiesel production from
waste sunflower cooking oil as an environmental recycling
process and renewable energy”, 2009.
[14] B. Rice, A. Fröhlich and R. Leonard”Bio-diesel production from
camelina oil,waste cooking oil and tallow”,September 1998.
[15] Prafulla D. Patil, Veera Gnaneswar Gude, Harvind K. Reddy,
Tapaswy Muppaneni, Shuguang Deng “Biodiesel production
from waste cooking oil using sulfuric acid and microwave
irradiation processes”,Nov 2011.
[16] Arjun B. Chhetri, K. Chris Watts and M. Rafiqul Islam "Waste
Cooking Oil as an alternate feedstock for biodiesel
production”,April 2010.
[17] Cholada Komintarachat and Sathaporn Chuepeng “Methanol-
based transesterification optimization of waste used cooking oil
over potassium hydroxide catalyst”2010
2017 International Conference on Emerging Trends & Innovation in ICT (ICEI)
Pune Institute of Computer Technology, Pune, India, Feb 3-5, 2017
TrafficIntel
Smart Traffic Management for Smart Cities
Anurag Saikar1 Mihir Parulekar1 Aditya Badve1

anuragsaikar100@gmail.com mihir19297@gmail.com Adityabadve123@gmail.com
Sagar Thakkar1 Aaradhana Deshmukh2

Sagarthakkar002@gmail.com aadeshmukh@sinhgad.edu
1
Student, Department of Computer Science, S.K.N.C.O.E, Pune
2
Ass. Professor, Department of Computer Science, S.K.N.C.O.E, Pune
Abstract— In India the unbalanced growth in the freight entirely inefficient in long term, there is need to utilize the
volume as compared to the road length has led to severe issues of existing infrastructure to its maximum capacity. This can be
traffic congestion, road safety and health hazards, which, as done by applying intelligence to the existing infrastructure, i.e.
estimated by recent studies will surely escalate very rapidly switching from static road traffic management system to a
within coming years. The use of a real time dynamic traffic
management system to intelligently navigate the vehicles can
more intelligent and dynamic real time traffic management
optimize the traffic flow on the roads, thus utilizing the system. considering the high dynamicity of the traffic flow
infrastructure efficiently and providing an environment along with polynomial increase of number of vehicles on the
conducive to solving gridlock in cities. The authors propose a real road networks makes management of the said traffic flow
time traffic management system (RTMS) consisting of real time through a RTMS much more complex [6]. Furthermore the
traffic monitoring system formed by small network of road side solution to a traffic congestion usually tends to be a series of
units, junction units and mobile units to dynamically decide the small intricately correlated solutions, applying only one of
time of traffic lights to discourage formation of gridlock, coupled which, while no doubt useful, generates insubstantial results.
with a web based application for vehicle drivers that will derive Here the authors propose a system consisting of real time
the data from real time traffic analysis to indicate the local traffic
traffic analysis, the data reported by which will further be
flow and suggest the incoming vehicles to make use of alternative
routes in order to further alleviate growth of the gridlock. derived by a web based application to prevent, contain and
disseminate the traffic congestion in real time. The system
Keywords — traffic congestion, real time traffic management consisting of a RSU (Road Side Unit), JU (Junction Unit) and
system, smart traffic lights MU (Mobile Unit) works on various levels to solve the
congestion. A part of the traffic analysis system i.e. the JU
I. INTRODUCTION will work at an intersection by communicating with the MU
(vehicles) to gauge the traffic density and dynamically set the
The road traffic, being versatile and economical is considered
time for traffic lights depending on it so as to maintain a
to be the most common mode of transportation in India.
constant flow of vehicles and if necessary, disperse the traffic
However, according to a study [18] by The Transport
quickly without letting it accumulate and form a congestion.
Corporation of India (TCIL) and IIM (Calcutta) on operational
While RSU’s will, by, operating on longer roads without
efficiencies of freight transportation by roads the traffic
intersections and communicating with local MU’s, try to
congestion has become such a dire situation that on some key
ascertain the traffic flow and formation of congestion. On
corridors such as Mumbai-Chennai, Delhi-Chennai and Delhi-
confirmation of a congestion the RSU’s will transfer this
Guwahati the vehicles crawl at an average speed of less than
information to central server via internet. The web application
20 kmph while it's only 21.35 kmph on Delhi-Mumbai stretch.
will, on suspicion and confirmation of a congestion by RSU’s
Consequently the Indian economy faces losses of around
in a particular area, inform the incoming vehicles and suggest
60,000 crore every year, which translates roughly to a
usage of another route thereby inhibiting the further growth of
staggering amount of $10.82 billion each year not accounting
the congestion, additionally the vehicles already within the
the subsequently engendered fatal environmental damages,
congestion can be notified to move synchronously to optimize
losses due to delay in receiving emergency services such as
the traffic flow by minimizing the progressive delay that
medical help etc. The study reports that while India's freight
occurs due to slower human reaction times when the stagnant
volume was increasing at a compounded annual growth rate of
traffic stream starts moving.
9.08% and vehicles were growing at 10.76%, the road length
In developing countries like India where the growth rate of
was increasing at only 4.01%. This has resulted paucity of vehicles outruns the growth of road length, the traffic
road space to accommodate vehicles and to increase the speed management is vital, as without it the increasing severity of the
ensuing the traffic congestions that lead to the huge gridlocks. congestion and its subsequent adverse effects are imminent. In
To overcome the gridlocks without modifying the existing this paper the authors have proposed a system as an attempt for
infrastructure, which proves to be cumbersome, expensive and overcoming the infrastructural and other issues to and
978-1-5090-3404-8/17/$31.00 ©2017 IEEE 46

providing the panacea to traffic congestion and its subsequent their proposed device is installed is monitored using a camera.
effects. The camera takes photos of traffic situations after set interval
of time and stores them in database for further viewing. The
II. LITERATURE REVIEW officer at that intersection can then take a look at these photos
and decide on the traffic light timing pattern from his
The recent increase in urbanization and motorization has been
smartphone.
perpetuating for the frequent traffic congestion and
development in the field of intelligent transportation system.
In [12] Petrovska, N., & Stevanovic, A. propose use of image
While a substantial amount of solutions have been proposed to
processing to determine traffic density. They then display the
assuage the issue, a lot of them put forth the solution in form
information on traffic congestion in various locations using
of static traffic management perspective [16], which proves to
their application for smartphones. For doing this they use
be capricious at times, considering the dynamicity of the
Google’s API for interfacing with the traffic layer of the
traffic flow. The real time traffic management model, while
Google maps. The traffic density in areas is then displayed
considerably more lucrative, the design of which, however
using color codes each denoting very high congestion to no
tends to be increasing complex on a large scale owing to the
congestion.
variable and fluid nature of the traffic stream [16]. For eg. In
[6] Kammoun, H. M., Kallel, I., Alimi, A. M., & Casillas, J.
In [3] a simple approach is used to detect traffic congestion in
propose a hybrid method that makes use of adaptive vehicular
any area in a developing region. Here, an application on the
guidance systems to analyze the road traffic network, the
user’s devices will pull their GPS and speed information
traffic is adjusted intelligently by suggesting alternative path
further sending this data to the server. Once the data reaches
to the destination based on ant colony behavior and
the servers, the data is analyzed and presence of congestion is
hierarchical fuzzy system, while in [13] Shashikiran, V.,
confirmed algorithmically. The user is then made aware of the
Kumar, T. S., Kumar, N. S., Venkateswaran, V., & Balaji, S.
congestion on the respective route, via the application, so that
propose a solution that makes use of Kruskal’s algorithm for
the user may choose to take a different route. This system is
suggesting optimal path by mapping the traffic junctions to
highly efficient and optimized as it requires very less
nodes and traffic flow as link weight by using Google maps.
infrastructure development and instead relies on
The selection of optimal path is based on various parameters
crowdsourcing for the data.
such as traffic rate, speed of the vehicle, shortest path etc. a
traffic in-charge personnel has to update the traffic
Sowmiya, A., & Ram, N. P. in [15] propose a tracking device
information periodically through their mobile in order to
placed every 250 meters apart on roads. A device placed
accomplish this.
within the vehicles, is capable of communicating the vehicle’s
speed and some other relevant information with the tracking
Nafi, N. S., Khan, R. H., Khan, J. Y., & Gregory, M. in [9]
device. The tracking device after receiving this information
propose a system using IEEE 802.11p based vehicle to
about vehicles within its range, will calculate custom speed for
infrastructure communications system (VANET) to predict the
every vehicle with which it should travel, in order to avoid all
future traffic intensities at intersections. The vehicles are
red lights on its route. Thus saving fuel as well as reducing
rerouted based on this prediction to reduce the traffic
congestion along various routes.
congestion and minimize the driving time of the individual.
In [13] a design for traffic management system has been
In [5] Jayapal, C., & Roy, S. S. suggest use of RSU (Road side
proposed for intersections, which makes use of long range
units) and mobile phones to detect traffic congestion in a
photoelectric sensors placed strategically to monitor traffic
certain area. The mobile phone and RSU communicate with
and decide the time of traffic lights accordingly. The increase
each other wirelessly via a vehicular ad hoc network
or decrease in time of traffic light is made only in case of
(VANET). The distance travelled by the vehicle in particular
suspicion of large number of vehicles i.e. emergency situation,
interval of time is detected. If this value is below a set
elsewise the system switches to the default system.
threshold, congestion is suspected. If other vehicles in the
same area show similar signs of congestion, then congestion is
confirmed. Once congestion is confirmed vehicles
Reference [9] proposes a system for cognitive traffic
approaching the congested route are notified about the same
management that primarily focuses on smart traffic lights. The
using the mobile app or via a display near the traffic signals.
dynamic timing of these smart traffic lights is based on the
conclusions drawn from the analysis of data obtained from
Reference [7] proposes a system which dynamically lets the
cars and road sensors, as well as alternative sources of user
officer in a particular intersection decide the traffic lights
data, such as social networks, the results of opinion polls and
timing pattern after he has taken a look at the traffic using his
others.
smartphone. The system also allows emergency vehicles to
claim their right of way by allowing the authorized personnel
From the systems discussed above it can be discerned that
inside the emergency vehicle to change the traffic light pattern
most of them, while making use of real time data, only suggest
and status. The traffic flow at a particular intersection where
47
alternate path on its basis, which has very little effect on traffic for further analysis of vehicular density for gauging the
traffic congestion, that too on a local level. Also if all the formation of congestion. The MU’s i.e. the tags placed inside
vehicles make use of the same guidance system it leads to them continuously broadcast RF beacons, once the MU arrives
concentration on other routes [17]. with the range the signal is sensed by the RSU.
The RSU’s are to be placed on the on the length of the roads
III. PROPOSED SYSTEM with an appreciable distance between them (about 200 meters)
The proposed system consists of three real time components [8]. The RSU (router) forwards this data to the appropriate
the RSU, the JU and the MU. The system makes use of active wireless coordinator, which relays the data to the server for
RFID (based on IEEE 802.15.4 protocol, 2.4 GHz ISM band), further analysis. Ideally the MU or the vehicle travels from
the specific RFID tags are to be kept in the vehicle, both these router to junction coordinator, i.e. the vehicle passes through a
coupled together form the MU i.e. the mobile unit. The RSU’s number of RSU’s towards the JU.
contain, a RFID reader together termed as a wireless router, The JU’s work at the intersections, acting partly as RFID
with an assigned gateway (i.e. a wireless coordinator) for a readers to measure the vehicle density at each side of the
particular set of RSUs. While the JU’s contain, within them, a intersection while simultaneously reading the data updated on
processor for the required computation, a controller for traffic the server by the RSU’s to estimate the incoming traffic along
lights and a RFID reader. The Tags emit radio signals that can each direction, this data, coupled with the wait time of the
be captured by devices like routers or coordinators. Routers vehicles at the junction and the distance of each of the vehicle
are capable of capturing tag data and relay the captured data to from that particular junction are the factors based on which the
either coordinator or another router in its range. Coordinators weight for a route of the road on the intersection will be
have a serial interface through which external GSM/ GPRS decided. These weights, calculated by the JU, for each
devices can be interfaced, thus enabling the further segment of the road on the intersection are measured against a
communication with the centrally located server. threshold to decide the time of the traffic light for the same.
The distance plays a major role in deciding the contribution of
an individual MU to the total effective weight of the segment
of the road, leading to incremental increase in contribution of
MU towards the effective weight of the road segment as it
nears the particular junction, this allows for easier
coordination between the JU’s to minimize the overall idle
time. Furthermore this enables the whole network of JU’s in
synchronicity to minimize, and in case it occurs, quickly
prevent the congestion from growing further.
The data updated by the RSU’s is also used to calculate the
lane velocity i.e. the average velocity of the vehicles on the
specific road segment. This can be done with using the
location of RSU’s and the timestamps for when the MU
arrives in its range, this helps in confirmation and later with
gauging the intensity of the congestion. This data can further
Figure 1 : General Architecture of the System be relayed by the web base application to the users to allow
selection optimal path towards the destination, and in case of
The figure 1 shows the general architecture of the system, congestion, the active users headed to the congested route can
depicting the data flow between the MU, the RSU, the JU and take an alternative route thereby reducing the glut of vehicles
further the server. The RSU's and the JU's sense the data contributing to the traffic congestion and disseminating the
associated with the specific RFID tags, identifying the MU's traffic over various available routes.
in their locality and further the vehicular density of the same. The system aims at synchronizing the traffic lights to
This data is further reported to the centrally locate server via minimize the possibility of formation of congestion, and in
internet, to be used by the web based application, amongst case of congestion, the system works on multiple levels to
other things. The data reported and updated by all the other rectify the traffic congestion. Firstly it tries to resolve the
RSUs and the JUs is read by a specific JU from the centrally congestion owing to the dynamicity of traffic lights time and
located server so as to ensure proper synchronization, thereby secondly prunes its further growth by alerting the incoming
minimizing the overall idle time of the vehicles and vehicles of the congestion and suggesting the possible
furthermore discouraging formation of congestion or alternative routes. Furthermore, it reduces the overall idle time
gridlocks. of the vehicles during the travel, thereby minimizing the fuel
wastage.
The RFID tags placed within the MU’s are specifically used to
identify the type of the vehicle. This helps in categorization of
48
V. CONCLUSION
Rapid urbanization in India has led to an increase in the
number of personal transport vehicles as well as public and
commercial transport vehicles while development of road
infrastructure however has not been commensurate. This fuels
an ever growing problem of traffic congestion, which despite
enforcement of new regulations about freight volume has been
a massive concern. This problem can be resolved by either
improving the road infrastructure or increasing the efficiency
of traffic on the current roads. The latter seems the apparent
choice as it requires comparatively negligible investment of
extraneous resources. Here the authors propose simple system
that, by making use of the RFID technology, aims at enabling
reduction in traffic congestion and also saving a lot of time and
fuel for the growing economy of India.
REFERENCES
[1] Burton, R. S., Thomas, N. C., & Wright, D. (2000). Wireless
Figure 2 : Schematic of the system communications to road users and within urban traffic management and
control systems. In Road Transport Information and Control, 2000.
The above image represents a schematic of the system spread Tenth International Conference on (Conf. Publ. No. 472) (pp. 152-157).
IET.
over several blocks. A JU is placed at each intersection which [2] Chorlton, C., & Fraser, I. (1998, April). Traffic monitoring on all-
senses the vehicles within their range, thereby monitoring the purpose roads for strategic network management. In Road Transport
traffic at the intersections. The overall traffic i.e. the traffic Information and Control, 1998. 9th International Conference on (Conf.
density on the roads and at the intersections can be monitored Publ. No. 454) (pp. 125-129). IET.
[3] Dhakad, R., & Jain, M. (2014, December). GPS based road traffic
and reported, with the help of network of RSUs and JUs, to congestion reporting system. In Computational Intelligence and
the centrally located server for further use by the web based Computing Research (ICCIC), 2014 IEEE International Conference
application and for further processing. Each JU consists its on (pp. 1-6). IEEE.
own processor for faster, efficient and real time computation, [4] Gregoriades, A. (2007, December). Towards a user-centred road safety
management method based on road traffic simulation. In 2007 Winter
furthermore local computation enables the system to compute Simulation Conference (pp. 1905-1914). IEEE.
in real time without introducing considerable lag. [5] Jayapal, C., & Roy, S. S. (2016, March). Road traffic congestion
management using VANET. In 2016 International Conference on
IV. RESULTS Advances in Human Machine Interaction (HMI) (pp. 1-7). IEEE.
[6] Kammoun, H. M., Kallel, I., Alimi, A. M., & Casillas, J. (2011, April).
Improvement of the road traffic management by an ant-hierarchical
fuzzy system. In Computational Intelligence in Vehicles and
Transportation Systems (CIVTS), 2011 IEEE Symposium on (pp. 38-
45). IEEE.
[7] Malla, A. M., & Sahu, R. K. (2013). A Review on Vehicle to Vehicle
Communication Protocols in VANETs. International Journal of
Advanced Research in Computer Science and Software
engineering, 3(2).
[8] Mandal, Koushik, et al. "Road traffic congestion monitoring and
measurement using active RFID and GSM technology." 2011 14th
International IEEE Conference on Intelligent Transportation Systems
T (ITSC). IEEE, 2011.
Figure 3 : Overview of the Web based Application he [9] Misbahuddin, S., Zubairi, J. A., Saggaf, A., Basuni, J., Sulaiman, A., &
The information updated by the network of JUs and RSUs in Al-Sofi, A. (2015, December). IoT based dynamic road traffic
the central server are reflected onto the web based application, management for smart cities. In 2015 12th International Conference on
relaying the status of the vehicular flow on the route to the High-capacity Optical Networks and Enabling/Emerging Technologies
(HONET) (pp. 1-5). IEEE.
users. This allows the vehicle drivers to choose an appropriate [10] Miz, V., & Hahanov, V. (2014, September). Smart traffic light in terms
path to the destination and in case of a significant congestions, of the Cognitive road traffic management system (CTMS) based on the
to make use of an alternate route thus helping not only to rein Internet of Things. In Design & Test Symposium (EWDTS), 2014 East-
in the growth of the congestion on the road in question but also West (pp. 1-5). IEEE
to curb wastage of resources such as fuel, while ensuring a [11] Nafi, N. S., Khan, R. H., Khan, J. Y., & Gregory, M. (2014, November).
A predictive road traffic management system based on vehicular ad-hoc
smooth and faster travel with less idle time. This allows for the network. In Telecommunication Networks and Applications Conference
traffic in already congested area to be disseminated fast. (ATNAC), 2014 Australasian (pp. 135-140). IEEE.
[12] Patel, H. N., Desai, K., & Panchal, T. (2015, December). An algorithm
for automatic license plate detection from video using corner features.
In 2015 International Conference on Information Processing (ICIP) (pp.
301-306). IEEE
49
[13] Petrovska, N., & Stevanovic, A. (2015, September). Traffic Congestion [19] Yu, S., Mabu, S., Mainali, M. K., Shimada, K., & Hirasawa, K. (2011,
Analysis Visualisation Tool. In 2015 IEEE 18th International October). Dynamic traffic management model for real world road
Conference on Intelligent Transportation Systems (pp. 1489-1494). networks. InSystems, Man, and Cybernetics (SMC), 2011 IEEE
IEEE. International Conference on (pp. 2197-2202). IEEE.
[14] Report on Operational Efficiency of Freight Transportation by Road in
India http://www.tcil.com/tcil/pdf/TCI%20IIM%20Study%20Report-
2011-12.pdf (last accessed on 15/9/2016)
[15] Salama, A. S., Saleh, B. K., & Eassa, M. M. (2010, November).
Intelligent cross road traffic management system (ICRTMS).
In Computer Technology and Development (ICCTD), 2010 2nd
International Conference on (pp. 27-31). IEEE.
[16] Shashikiran, V., Kumar, T. S., Kumar, N. S., Venkateswaran, V., &
Balaji, S. (2011, June). Dynamic road traffic management based on
krushkal's algorithm. In Recent Trends in Information Technology
(ICRTIT), 2011 International Conference on (pp. 200-204). IEEE.
[17] Sowmiya, A., & Ram, N. P. (2015, March). An intelligent approach for
effective road traffic management. In Innovations in Information,
Embedded and Communication Systems (ICIIECS), 2015 International
Conference on(pp. 1-4). IEEE.
[18] Vrancken, J. L., & Kruse, O. C. (2006). Intelligent control in networks:
The case of road traffic management. In 2006 IEEE International
Conference on Networking, Sensing and Control (pp. 308-311). IEEE.
50
SPECIAL SECTION ON RECENT ADVANCES IN SOFTWARE DEFINED
NETWORKING FOR 5G NETWORKS
Received July 11, 2015, accepted July 22, 2015, date of publication July 28, 2015, date of current version August 7, 2015.
Digital Object Identifier 10.1109/ACCESS.2015.2461602
A Survey of 5G Network: Architecture and

Emerging Technologies
AKHIL GUPTA, (Student Member, IEEE), AND RAKESH KUMAR JHA, (Senior Member, IEEE)
School of Electronics and Communication Engineering, Shri Mata Vaishno Devi University, Katra 182320, India
Corresponding author: A. Gupta (akhilgupta112001@gmail.com)
ABSTRACT In the near future, i.e., beyond 4G, some of the prime objectives or demands that need to
be addressed are increased capacity, improved data rate, decreased latency, and better quality of service.
To meet these demands, drastic improvements need to be made in cellular network architecture. This paper
presents the results of a detailed survey on the fifth generation (5G) cellular network architecture and some
of the key emerging technologies that are helpful in improving the architecture and meeting the demands of
users. In this detailed survey, the prime focus is on the 5G cellular network architecture, massive multiple
input multiple output technology, and device-to-device communication (D2D). Along with this, some of the
emerging technologies that are addressed in this paper include interference management, spectrum sharing
with cognitive radio, ultra-dense networks, multi-radio access technology association, full duplex radios,
millimeter wave solutions for 5G cellular networks, and cloud technologies for 5G radio access networks
and software defined networks. In this paper, a general probable 5G cellular network architecture is proposed,
which shows that D2D, small cell access points, network cloud, and the Internet of Things can be a part of
5G cellular network architecture. A detailed survey is included regarding current research projects being
conducted in different countries by research groups and institutions that are working on 5G technologies.
INDEX TERMS 5G, cloud, D2D, massive MIMO, mm-wave, relay, small-cell.
I. INTRODUCTION eruption of new applications which will be used in cases for

Today and in the recent future, to fulfill the presumptions mobile connectivity and a resultant exponential growth in
and challenges of the near future, the wireless based net- network traffic. This paper presents our view on the future of
works of today will have to advance in various ways. Recent wireless communication for 2020 and beyond. In this paper,
technology constituent like high-speed packet access (HSPA) we describe the key challenges that will be encountered by
and long-term evolution (LTE) will be launched as a future wireless communication while enabling the networked
segment of the advancement of current wireless based society. Along with this, some technology routes that may be
technologies. Nevertheless, auxiliary components may also taken to fulfill these challenges [1].
constitute future new wireless based technologies, which The imagination of our future is a networked society with
may adjunct the evolved technologies. Specimen of these unbounded access to information and sharing of data which
new technology components are different ways of accessing is accessible everywhere and every time for everyone and
spectrum and considerably higher frequency ranges, the everything. To realize this imagination, new technology com-
instigation of massive antenna configurations, direct device- ponents need to be examined for the evolution of existing
to-device communication, and ultra-dense deployments [1]. wireless based technologies. Present wireless based tech-
Since its initiation in the late 1970s, mobile wireless nologies, like the 3rd Generation Partnership Project (3GPP)
communication has come across from analog voice calls to LTE technology, HSPA and Wi-Fi, will be incorporating new
current modern technologies adept of providing high qual- technology components that will be helping to meet the needs
ity mobile broadband services with end-user data rates of of the future. Nevertheless, there may be certain scenarios that
several megabits per second over wide areas and tens, or cannot be adequately addressed along with the evolution of
even hundreds, of megabits per second locally. The extensive ongoing existing technologies. The instigation of completely
improvements in terms of potentiality of mobile communica- new wireless based technologies will complement the current
tion networks, along with the initiation of new types of mobile technologies which are needed for the long term realization
devices such as smart phones and tablets, have produced an of the networked society [2].
2169-3536
2015 IEEE. Translations and content mining are permitted for academic research only.
1206 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 3, 2015
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
A. Gupta, R. K. Jha: Survey of 5G Network: Architecture and Emerging Technologies
The remainder of the paper is organized as follows: and with no security, since voice calls were stored and played
In Section II, we present the evolution of wireless in radio towers due to which vulnerability of these calls from
technologies. Section III gives the detailed description of unwanted eavesdropping by third party increases [7].
the proposed general 5G cellular network architecture.
Section IV comprises of the detailed explanation of the B. 2G
emerging technologies for 5G wireless networks. We con- The 2nd generation was introduced in late 1990’s.
clude our paper in Section V. A list of current research Digital technology is used in 2nd generation mobile tele-
projects based on 5G technologies is shown in the appendix. phones. Global Systems for Mobile communications (GSM)
was the first 2nd generation system, chiefly used for voice
II. EVOLUTION OF WIRELESS TECHNOLOGIES communication and having a data rate up to 64kbps.
G. Marconi, an Italian inventor, unlocks the path of 2G mobile handset battery lasts longer because of the radio
recent day wireless communications by communicating the signals having low power. It also provides services like Short
letter ‘S’ along a distance of 3Km in the form of three dot Message Service (SMS) and e-mail. Vital eminent technolo-
Morse code with the help of electromagnetic waves. After gies were GSM, Code Division Multiple Access (CDMA),
this inception, wireless communications have become an and IS-95 [3], [7].
important part of present day society. Since satellite com-
munication, television and radio transmission has advanced C. 2.5G
to pervasive mobile telephone, wireless communications has It generally subscribes a 2nd generation cellular system
transformed the style in which society runs. The evolution merged with General Packet Radio Services (GPRS) and
of wireless begins here [2] and is shown in Fig. 1. It shows other amenities doesn’t commonly endow in 2G or 1G
the evolving generations of wireless technologies in terms of networks. A 2.5G system generally uses 2G system
data rate, mobility, coverage and spectral efficiency. As the frameworks, but it applies packet switching along with
wireless technologies are growing, the data rate, mobility, circuit switching. It can assist data rate up to 144kbps. The
coverage and spectral efficiency increases. It also shows main 2.5G technologies were GPRS, Enhanced Data Rate
that the 1G and 2G technologies use circuit switching while for GSM Evolution (EDGE), and Code Division Multiple
2.5G and 3G uses both circuit and packet switching and Access (CDMA) 2000 [3], [7].
the next generations from 3.5G to now i.e. 5G are using
packet switching. Along with these factors, it also differ- D. 3G
entiate between licensed spectrum and unlicensed spectrum. The 3rd generation was established in late 2000. It imparts
All the evolving generations use the licensed spectrum while transmission rate up to 2Mbps. Third generation (3G)
the WiFi, Bluetooth and WiMAX are using the unlicensed systems merge high speed mobile access to services based
spectrum. An overview about the evolving wireless on Internet Protocol (IP). Aside from transmission rate,
technologies is below: unconventional improvement was made for maintaining QoS.
Additional amenities like global roaming and improved voice
quality made 3G as a remarkable generation. The major
disadvantage for 3G handsets is that, they require more
power than most 2G models. Along with this 3G network
plans are more expensive than 2G [3], [7]. Since
3G involves the introduction and utilization of Wideband
Code Division Multiple Access (WCDMA), Universal
Mobile Telecommunications Systems (UMTS) and Code
Division Multiple Access (CDMA) 2000 technologies, the
evolving technologies like High Speed Uplink/Downlink
Packet Access (HSUPA/HSDPA) and Evolution-Data
Optimized (EVDO) has made an intermediate wireless
generation between 3G and 4G named as 3.5G with improved
data rate of 5-30 Mbps [3].
FIGURE 1. Evolution of wireless technologies.
E. 3.75G
A. 1G Long-Term Evolution technology (LTE) and Fixed
The 1st generation was announced in initial 1980’s. Worldwide Interoperability for Microwave Access (WIMAX)
It has a data rate up to 2.4kbps. Major subscribers were is the future of mobile data services. LTE and Fixed WIMAX
Advanced Mobile Phone System (AMPS), Nordic Mobile has the potential to supplement the capacity of the network
Telephone (NMT), and Total Access Communication and provides a substantial number of users the facility to
System (TACS). It has a lot of disadvantages like below access a broad range of high speed services like on demand
par capacity, reckless handoff, inferior voice associations, video, peer to peer file sharing and composite Web services.
VOLUME 3, 2015 1207

Along with this, a supplementary spectrum is accessible the wireless setup which had come about from 1G to 4G.
which accredit operators manage their network very compli- Alternatively, there could be only the addition of an appli-
antly and offers better coverage with improved performance cation or amelioration done at the fundamental network to
for less cost [4]–[7]. please user requirements. This will provoke the package
providers to drift for a 5G network as early as 4G is com-
F. 4G mercially set up [8]. To meet the demands of the user and
4G is generally referred as the descendant of the 3G and 2G to overcome the challenges that has been put forward in the
standards. 3rd Generation Partnership Project (3GPP) 5G system, a drastic change in the strategy of designing
is presently standardizing Long Term Evolution (LTE) the 5G wireless cellular architecture is needed. A general
Advanced as forthcoming 4G standard along with Mobile observation of the researchers has shown in [14] that most of
Worldwide Interoperability for Microwave Access (WIMAX). the wireless users stay inside for approximately 80 percent of
A 4G system improves the prevailing communication time and outside for approximately 20 percent of the time.
networks by imparting a complete and reliable solution based In present wireless cellular architecture, for a mobile user
on IP. Amenities like voice, data and multimedia will be to communicate whether inside or outside, an outside base
imparted to subscribers on every time and everywhere basis station present in the middle of a cell helps in communication.
and at quite higher data rates as related to earlier generations. So for inside users to communicate with the outside base
Applications that are being made to use a 4G network are station, the signals will have to travel through the walls of
Multimedia Messaging Service (MMS), Digital Video the indoors, and this will result in very high penetration loss,
Broadcasting (DVB), and video chat, High Definition TV which correspondingly costs with reduced spectral efficiency,
content and mobile TV [2], [4]–[6]. data rate, and energy efficiency of wireless communications.
To overcome this challenge, a new idea or designing tech-
G. 5G nique that has come in to existence for scheming the
With an exponential increase in the demand of the users, 5G cellular architecture is to distinct outside and inside
4G will now be easily replaced with 5G with an setups [8]. With this designing technique, the penetration loss
advanced access technology named Beam Division Multiple through the walls of the building will be slightly reduced.
Access (BDMA) and Non- and quasi-orthogonal or Filter This idea will be supported with the help of massive MIMO
Bank multi carrier (FBMC) multiple access. The concept technology [15], in which geographically dispersed array
behind BDMA technique is explained by considering the case of antenna’s are deployed which have tens or hundreds of
of the base station communicating with the mobile stations. antenna units. Since present MIMO systems are using either
In this communication, an orthogonal beam is allocated to two or four antennas, but the idea of massive MIMO systems
each mobile station and BDMA technique will divide that has come up with the idea of utilizing the advantages of large
antenna beam according to locations of the mobile stations array antenna elements in terms of huge capacity gains.
for giving multiple accesses to the mobile stations, which To build or construct a large massive MIMO network,
correspondingly increase the capacity of the system [8]. firstly the outside base stations will be fitted with large
An idea to shift towards 5G is based on current drifts, it is antenna arrays and among them some are dispersed around
commonly assumed that 5G cellular networks must address the hexagonal cell and linked to the base station through
six challenges that are not effectively addressed by 4G i.e. optical fiber cables, aided with massive MIMO technologies.
higher capacity, higher data rate, lower End to End latency, The mobile users present outside are usually fitted with a
massive device connectivity, reduced cost and consistent certain number of antenna units but with cooperation a large
Quality of Experience provisioning [22], [23]. These virtual antenna array can be constructed, which together with
challenges are concisely shown in Fig. 2 along with antenna arrays of base station form virtual massive MIMO
some potential facilitators to address them. An overview links. Secondly, every building will be installed with large
of the challenges, facilitators, and corresponding design antenna arrays from outside, to communicate with outdoor
fundamentals for 5G is shown in Fig. 2 [20]. Recently base stations with the help of line of sight components.
introduced IEEE 802.11ac, 802.11ad and 802.11af standards The wireless access points inside the building are connected
are very helpful and act as a building blocks in the road with the large antenna arrays through cables for communi-
towards 5G [9]–[13]. The technical comparison between these cating with indoor users. This will significantly improves
standards is shown in table 1 and the detailed comparison of the energy efficiency, cell average throughput, data rate, and
wireless generations is shown in table 2. spectral efficiency of the cellular system but at the expense
of increased infrastructure cost. With the introduction of
III. 5G CELLULAR NETWORK ARCHITECTURE such an architecture, the inside users will only have to
To contemplate 5G network in the market now, it is evident connect or communicate with inside wireless access points
that the multiple access techniques in the network are while larger antenna arrays remained installed outside the
almost at a still and requires sudden improvement. Current buildings [8]. For indoor communication, certain technolo-
technologies like OFDMA will work at least for next gies like WiFi, Small cell, ultra wideband, millimeter wave
50 years. Moreover, there is no need to have a change in communications [16], and visible light communications [17]
1208 VOLUME 3, 2015

FIGURE 2. 5G challenge, facilitators, and design fundamental [20].
are useful for small range communications having large data conventionally used for cellular communications. But it is
rates. But technologies like millimeter wave and visible light not an efficient idea to use these high frequency waves for
communication are utilizing higher frequencies which are not outside and long distance applications because these waves
VOLUME 3, 2015 1209

TABLE 1. Technical comparison between recent 802.11 standards.
will not infiltrate from dense materials efficiently and can Control plane, respectively. Special network functionality as
easily be dispersed by rain droplets, gases, and flora. Though, a service (XaaS) will provide service as per need, resource
millimeter waves and visible light communications technolo- pooling is one of the examples. XaaS is the connection
gies can enhance the transmission data rate for indoor setups between a radio network and a network cloud [20].
because they have come up with large bandwidth. Along The 5G cellular network architecture is explained
with the introduction of new spectrum, which is not being in [8] and [20]. It has equal importance in terms of front
conventionally used for wireless communication, there is one end and backhaul network respectively. In this paper, a
more method to solve the spectrum shortage problem by general 5G cellular network architecture has been proposed
improving the spectrum utilization of current radio spectra as shown in Fig. 3. It describes the interconnectivity
through cognitive radio (CR) networks [18]. among the different emerging technologies like Massive
Since the 5G cellular architecture is heterogeneous, so it MIMO network, Cognitive Radio network, mobile and
must include macrocells, microcells, small cells, and relays. static small-cell networks. This proposed architecture also
A mobile small cell concept is an integral part of 5G wireless explains the role of network function virtualization (NFV)
cellular network and partially comprises of mobile relay and cloud in the 5G cellular network architecture. The concept
small cell concepts [19]. It is being introduced to put up of Device to Device (D2D) communication, small cell access
high mobility users, which are inside the automobiles and points and Internet of things (IoT) has also been incorporated
high speed trains. Mobile small cells are positioned inside the in this proposed 5G cellular network architecture. In general,
moving automobiles to communicate with the users inside this proposed 5G cellular network architecture may provide
the automobile, while the massive MIMO unit consisting a good platform for future 5G standardization network.
of large antenna arrays is placed outside the automobile to But there are several issues that need to be addressed in
communicate with the outside base station. According to order to realize the wireless network architecture in partic-
user’s opinion, a mobile small cell is realized as a regular ular, and 5G networks in general. Some of these issues are
base station and its allied users are all observed as a single summarized in Table. 3 [20].
unit to the base station which proves the above idea of
splitting indoor and outdoor setups. Mobile small cell IV. EMERGING TECHNOLOGIES FOR
users [19] have a high data rate for data rate services with 5G WIRELESS NETWORKS
considerably reduced signaling overhead, as shown in [8]. It is expected that mobile and wireless traffic volume will
As the 5G wireless cellular network architecture consists increase a thousand-fold over the next decade which will
of only two logical layers: a radio network and a network be driven by the expected 50 billion connected devices con-
cloud. Different types of components performing different nected to the cloud by 2020 and all need to access and share
functions are constituting the radio network. The network data, anywhere and anytime. With a rapid increase in the num-
function virtualization (NFV) cloud consists of a User plane ber of connected devices, some challenges appear which will
entity (UPE) and a Control plane entity (CPE) that per- be responded by increasing capacity and by improving energy
form higher layer functionalities related to the User and efficiency, cost and spectrum utilization as well as providing
1210 VOLUME 3, 2015

TABLE 2. Evolution of wireless technologies.
better scalability for handling the increasing number of the of applications and requirements of the user. To provide a
connected devices. For the vision of all-communicating world common connected platform for a variety of applications and
relative to today’s network, the overall technical aim is to requirements for 5G, we will research the below technology
provide a system idea that supports [21]: components [21]:
• 1000 times increased data volume per area • Radio-links, includes the development of new transmis-
• 10 to 100 times increased number of connected devices sion waveforms and new approaches of multiple access
• 10 to 100 times increased typical user data rate control and radio resource management.
• 10 times extended battery life for low power Massive • Multi-node and multi-antenna transmissions, includes
Machine Communication (MMC) devices designing of multi-antenna transmission/reception tech-
• 5 times reduced End-to-End (E2E) latency nologies based on massive antenna configurations and
In this paper, we will cover a wide area of technologies developing advanced inter-node coordination schemes
with a lot of technical challenges arises due to a variety and multi-hop technologies.
VOLUME 3, 2015 1211

FIGURE 3. A general 5G cellular network architecture.
• Network dimension, includes considering the demand, In this section, we identify several technologies, ranked in
traffic and mobility management, and novel approaches perceived importance, which will be crucial in future wireless
for efficient interference management in complex standards.
heterogeneous deployments.
• Spectrum usage, includes considering extended A. MASSIVE MIMO
spectrum band of operation, as well as operation in new Massive MIMO is an evolving technology that has been
spectrum regimes to provide a complete system concept upgraded from the current MIMO technology. The Massive
for new spectrum regimes that carefully addresses the MIMO system uses arrays of antenna containing few hundred
needs of each usage scenario. antennas which are at the same time in one time, frequency
Now the topics which will integrate a subset of the slot serving many tens of user terminals. The main objective
technology components and provides the solution of some of of Massive MIMO technology is to extract all the benefits
the goals which are identified earlier are [21]: of MIMO but on a larger scale. In general, massive MIMO
• Device-to-Device (D2D) communications refers to is an evolving technology of Next generation networks,
direct communication between devices allowing local which is energy efficient, robust, and secure and spectrum
exchange of user plane traffic without going through a efficient [24].
network infrastructure. Massive MIMO depends on spatial multiplexing, which
• Massive Machine Communications (MMC) will form further depends on the base station to have channel state
the basis of the Internet of Things with a wide range information, both on the uplink as well as on the downlink.
of application fields including the automotive industry, In case of downlink, it is not easy, but in case of uplink,
public safety, emergency services and medical it is easy, as the terminals send pilots. On the basis of
field. pilots, the channel response of each terminal is estimated.
• Moving Networks (MN) will enhance and extend In conventional MIMO systems, the base station sends the
linking together potentially large populations of jointly pilot waveforms to the terminals and based on these, the
moving communication devices. terminal estimate the channel, quantize it and feedback them
• Ultra-dense Networks (UDN) will be the main driver to the base station. This process is not viable for mas-
whose goals are to increase capacity, increase energy sive MIMO systems, especially in high mobility conditions
efficiency of radio links, and enable better exploitation because of two reasons. Firstly the downlink pilots from the
of under-utilized spectrum. base station must be orthogonal among the antennas, due
• Ultra-reliable Networks (URN) will enable high to which the requirement of time, frequency slots for the
degrees of availability. downlink pilots increases with the increase in the number
1212 VOLUME 3, 2015

TABLE 3. Small cell setup options and concern [20].
of antennas. So Massive MIMO systems would now require play by confirming that all the wave fronts that have been
a large number of similar slots as compared to the con- emitted from the antennas possibly will add constructively at
ventional MIMO system. Secondly, as the number of base the intended terminal’s locations and destructively elsewhere.
station antennas increases the number of the channel esti- Zero forcing is used to suppress the remaining interfer-
mates also increases for each terminal which in turn needed ence between the terminals, but at the expense of increased
hundred times more uplink slots to feedback the channel transmitted power [24].
responses to the base station. A general solution to this prob- The desirability of maximum ratio combining (MRC) is
lem is to work in Time Division Duplexing (TDD) mode more as related to Zero forcing (ZF) because of its com-
and depend on the reciprocity amid the uplink and downlink putational ease i.e. received signals are multiplied by their
channels [25]. conjugate channel responses and due to the reason that it is
Massive MIMO technology depends on phase coherent executed in a dispersed mode, autonomously at every antenna
signals from all the antennas at the base station, but the element. Though ZF also works equally well for an orthodox
computational processing of these signals is simple. Below MIMO system which MRC normally does not. The main
are certain positives of a massive MIMO system [24]: reason behind the efficient use of the MRC with massive
MIMO involving large number of base station antennas, the
1) MASSIVE MIMO HAS THE CAPABILITY THAT IT CAN channel responses allied with different terminals tend to be
IMPROVE THE RADIATED ENERGY EFFICIENCY BY almost orthogonal.
100 TIMES AND AT THE SAME TIME, INCREASES With the use of MRC receiver, we are operating in a
THE CAPACITY OF THE ORDER OF 10 OR MORE noise restricted system. MRC in Massive MIMO system
The positive of increase in capacity is because of the will scale down the power to an extent possible deprived of
spatial multiplexing technique used in Massive really upsetting the overall spectral efficiency and multiuser
MIMO systems. Regarding the improvement in the radiated interference, but the effects of hardware deficiencies are
energy efficiency, it is because of the increase in the number likely to be overcome by the thermal noise. But the intention
of antennas, the energy can now be concentrated in small behind the overall 10 times higher spectral efficiency as
regions in the space. It is based on the principle of coherent compared to conventional MIMO is because 10 times more
superposition of wave fronts. After transmitting the shaped terminals are served concurrently in the same time frequency
signals from the antennas, the base station has no role to resource [26].
VOLUME 3, 2015 1213
2) MASSIVE MIMO SYSTEMS CAN BE PUT TOGETHER it is helpful to deploy base stations to the places where
WITH THE HELP OF LOW POWER AND electricity is not available. Along with this, the increased
LESS COSTLY COMPONENTS concerns of electromagnetic exposure will be considerably
Massive MIMO has come up with a change with less.
respect to concept, schemes and execution. Massive
MIMO systems use hundreds of less expensive amplifiers in 3) MASSIVE MIMO PERMITS A SUBSTANTIAL DECREASE
respect to expensive ultra-linear 50 Watt amplifiers because IN LATENCY ON THE AIR INTERFACE
earlier are having an output power in the milliwatt range, Latency is the prime area of concern in the next generation
which is much better than the latter which are generally networks. In wireless communication, the main cause of
being used in conventional systems. It is dissimilar to con- latency is fading. This phenomenon occurs amid the base
ventional array schemes, as it will use only a little antenna’s station and terminal, i.e. when the signal is transmitted from
that are being fed from high power amplifiers but having a the base station, it travels through different multiple paths
notable impact. The most significant improvement is about because of the phenomenon’s like scattering, reflection and
the removal of a large number of expensive and massive items diffraction before it reaches the terminal. When the signal
like large coaxial cables [24]. through these multiple paths reaches the terminal it will inter-
With the use of a large number of antennas in massive fere either constructively or destructively, and the case when
MIMO technology the noise, fading and hardware deficits following waves from these multiple paths interfere destruc-
will be averaged because signals from a large number of tively, the received signal strength reduces to a considerable
antennas are combined together in the free space. It condenses low point. If the terminal is caught in a fading dip, then it has
the limits on precision and linearity of every single amplifier to wait for the transmission channel to change until any data
and radio frequency chain and altogether what matters is can be received. Massive MIMO, due to a large number of
their collective action. This will increase the robustness of antennas and with the idea of beam forming can avoid fading
massive MIMO against fading and failure of one of the dips and now latency cannot be further decreased [24].
antenna elements.
A massive MIMO system has degrees of freedom in excess. 4) MASSIVE MIMO MAKES THE MULTIPLE
For example, with 100 antennas, 10 terminals are showing ACCESS LAYER SIMPLE
presence while the remaining 90 degrees of freedom are With the arrival of Massive MIMO, the channel strength-
still available. These available degrees of freedom can be ens and now frequency domain scheduling is not enough.
exploited by using them for signal shaping which will be OFDM provides, each subcarrier in a massive MIMO system
hardware friendly. Specifically, each antenna with the use of with considerably the same channel gain due to which each
very cheap and power proficient radio frequency amplifiers and every terminal can be provided with complete bandwidth,
can transmit signals having small peak to average ratio [27] which reduces most of the physical layer control signaling
and constant envelope [28] at a modest price of increased total terminated [24].
radiated power. With the help of constant envelope multiuser
precoding, the signals transmitted from each antenna are 5) MASSIVE MIMO INCREASES THE STRENGTH EQUALLY
neither being formed in terms of beam nor by weighing of AGAINST UNINTENDED MAN MADE INTERFERENCE
a symbol. Rather, a wave field is created and sampled with AND INTENDED JAMMING
respect to the location of the terminals and they can see Jamming of the wireless systems of the civilian is a
precisely the signals what we intended to make them see. prime area of concern and poses a serious threat to cyber
Massive MIMO has a vital property which makes it possible. security. Owing to limited bandwidth, the distribution of
The massive MIMO channel has large null spaces in which information over frequency just is not possible. Massive
nearly everything can be engaged without disturbing the MIMO offers the methods of improving robustness of
terminals. Precisely modules can be placed into this null wireless communications with the help of multiple antennas.
space that makes the transmitted waveforms fulfill the It provides with an excess of degrees of freedom that can
preferred envelope restraints. Nevertheless, the operative be useful for canceling the signals from intended jammers.
channels amid the base station and every terminal, can be If massive MIMO systems use joint channel estimation and
proceeded without the involvement of PSK type modulation decoding instead of uplink pilots for channel estimation,
and can take any signal constellation as input [24]. then the problem from the intended jammers is considerably
The considerable improvement in the energy efficiency reduced [24].
facilitates massive MIMO systems to work two steps of lower The advantages of massive MIMO systems can be
magnitude than with existing technology on the total output reviewed from an information theoretic point of view.
RF power. This is important because the cellular base stations Massive MIMO systems can obtain the promising multi-
are consuming a lot of power and it is an area of concern. plexing gain of massive point to point MIMO systems,
In addition, if base stations that consume less power could be while eliminating problems due to unfavorable propagation
driven by renewable resources like solar or wind and therefore environments [29].
1214 VOLUME 3, 2015

Let us study a massive MIMO system having L cells, where An exhaustive debate about this result can be seen in [31].
every cell has K attended single antenna users and one base Centered on the result in (6), the overall achievable rate of all
station with N antennas. hi,k,l,n represent the channel coeffi- users come to be
cient from the k-th user in the l-th cell to the n-th antenna of
C = log2 det(I + ρu H H H )
the i-th base station, which is equivalent to a complex small
scale fading factor time an amplitude factor that interprets for ≈ log2 det (I + N ρu D)
K bits
geometric attenuation and large-scale fading: X
= log2 (1 + N ρu dk ) s (7)
p Hz
hi,k,l,n = gi,k,l,n di,k,l (1) k=1
Capacity in (7) can be achieved at the base station by
Where gi,k,l,n and di,k,l represent complex small scale fad- simple MF processing. When MF processing is used, the
ing and large scale fading coefficients, respectively. The small base station processes the signal vector by multiplying the
scale fading coefficients are implicit to be diverse for diverse conjugate transpose of the channel, as
users or for diverse antennas at every base station though √
H H yu = H H ρu Hxu + nu

the large scale fading coefficients are the same for diverse
√
antennas at the same base station, but are user dependent. ≈ N ρu Dxu + H H nu (8)
Then, the channel matrix from all K users in the l-th cell to
the i-th base station can be expressed as where (6) is used. Note that the channel vectors are asymp-
  totically orthogonal when the number of antennas at the base
hi,1,l,1 · · · hi,K ,l,1 station grows to infinity. So, H H does not shade the noise.
Hi,l =  ... .. .. 1/2 Since D is a diagonal matrix, the MF processing splits the
. .  = Gi,l Di,l (2)
 
signals from diverse users into diverse streams and there is
hi,1,l,N ··· hi,K ,l,N
asymptotically no inter user interference. So now the signal
Where transmission can be treated as a SISO channel transmission
  for each user. From (8), the signal to noise ratio (SNR) for the
gi,1,l,1 ··· gi,K ,l,1
k-th user is N ρu dk . Subsequently, the attainable rate by using
Gi,l =  ... ..
.
.. 
.  (3)

MF is similar as the limit in (7), which indicates that simple
gi,1,l,N ··· gi.K ,l,N MF processing at the base station is best when the number of

di,1,l ··· ...
 antennas at the base station, N, grows to infinity.
 .. .. .. 
Di,c = . . .  (4) b: DOWNLINK
... ··· di.K ,l yd ∈ C K ∗1 can be denoted as the received signal vector at
all K users. Massive MIMO works properly in time division
Let us study a single cell (L = 1) massive MIMO system duplexing (TDD) mode as discussed in [29], where the down-
with K singled antenna users and a base station with link channel is the transpose of the uplink channel matrix.
N antennas. For ease, the cell and the base station indices are Then, the received signal vector can be expressed as
plunged when single cell systems are deliberated [29]. √
yd = ρd H T xd + nd (9)
a: UPLINK where xd ∈ C N ∗1 is the signal vector transmitted by the
The received signal vector at a single base station for uplink base station, nd ∈ C K ∗1 is an additive noise and ρd is the
signal transmission is denoted as yu ∈ C N ∗1 , can be stated as: transmit power of the downlink. Let us assume, E[|xd |2 ] = 1
√ for normalizing transmitting power.
yu = ρu Hxu + nu (5)
As discussed in [29], the base station usually has channel
where xu ∈ C K ∗1 is the signal vector from all users, state information equivalent to all users based on uplink pilot
H ∈ C N ∗K is the uplink channel matrix defined in (2) by transmission. So, it is likely for the base station to do power
reducing the cell and the base station indices, nu ∈ C N ∗1 is allocation for maximizing the sum transmission rate. The sum
a zero mean noise vector with complex Gaussian distribution capacity of the system with power allocation is [32]
and identity covariance matrix, and ρu is the uplink transmit C = max log2 det(IN + ρd HPH H )
power. The transmitted symbol from the k-th user, xku , is the p
bits
k-th element of xu = [x1u , . . . ., xKu ]T with [|xku |2 ] = 1. ≈ max log2 det (IK + ρd NPD) s
(10)
The column channel vectors from diverse users are asymp- p Hz
totically orthogonal as the number of antennas at the base where (6) is used and P is a positive diagonal matrix with
station, N, grows to infinity by supposing that the small the power allocations (p1 , . . . .., pk ) as its diagonal elements
scale fading coefficients for diverse users is independent [30]. and K
P
k=1 pk = 1
Then, we have If the MF precoder is used, the transmitted signal vector is
H H H = D1/2 GH GD1/2 ≈ ND1/2 IK D1/2 = ND (6) xd = H ∗ D−1/2 P1/2 sd (11)
VOLUME 3, 2015 1215

where sd ∈ C K ∗1 is the source information vector. Then,

the received signal vector at all K users is
√
yd = ρd H T H ∗D−1/2 P1/2 sd + nd
√
≈ ρd ND1/2 P1/2 sd + nd (12)
where the second line of (12) is for the case when the
number of antennas at the base station, N, grows to infinity,
and (6) is used. Since P and D are both diagonal matrices so
the signal transmission from the base station to every user can
be treated as if initiating from a SISO transmission which thus
inhibited the inter user interference. The overall attainable
data rate in (12) can be maximized by proper choice of the
power allocation as in (10), which validates that the capacity
can be attained using the simple MF precoder.
According to the auspicious propagation assumption of (6),
FIGURE 4. Average total power consumption in the scenario containing
the simple MF precoder or detector can attain the capacity of small cell access points.
a massive MIMO system when the number of antennas at the
base station, N, is much larger than the number of users, K,
and grows to infinity, i.e., N K and N → ∞. Another However, there are saturation points where extra hardware
scenario assumption is that both the number of antennas at will not decrease the total power anymore.
the base station and the number of users grows large while With the introduction of the concept of small cell access
their ratio is bounded, i.e., N/K = c as N, K → ∞, where c point, it will fulfill the need of self organizing network (SON)
is a constant, are different [35]. technology for minimizing human intervention in the
The main area of concern in today’s wireless cellular networking processes as given in [36] and [37]. While a brief
network is on energy efficiency and power optimization. summary of the work done on the massive MIMO technology
So a lot of researchers are working on to increase the energy to increase the energy efficiency and optimizing the power of
efficiency and optimizing the power. The work done on the wireless cellular network is shown in Table 4.
power optimization in [33] has been realized and shown
in Fig. 4. B. INTERFERENCE MANAGEMENT
Fig. 4 clearly shows that if we increase the number of For efficient utilization of limited resources, reuse is one
antennas at the base station as well as on the small cell access of the concept that is being used by many specifications
point, the total power per subcarrier decreases to 10 fold as of cellular wireless communication systems. Along with
compare to the case of no antenna at small cell access point. this, for improved traffic capacity and user throughput
TABLE 4. Effect of massive MIMO technology on energy efficiency of the wireless cellular network.
1216 VOLUME 3, 2015

densification of the network is one of the key aspect. So with the network side need to be stated in detail in the 5G systems,
the introduction of reuse and densification concept, there without separating it entirely as an employment issue. For
will be an additional enhancement in terms of efficient load attracting maximum coordination, the user equipment and
sharing between macro cells and local access networks. But network side, advanced interference management must be
all these advantages have come up with a problem that the deliberated instantaneously [38].
density and load of the network have increased considerably
C. SPECTRUM SHARING
and correspondingly receiver terminals in the network suffer
To apprehend the performance targets of future mobile
from increased co-channel interference, mainly at the bound-
broadband systems [22], [39], there is a need of considerably
aries of cells. Thus co-channel interference poses a threat
more spectrum and wider bandwidths as compared to the
which is inhibiting the further improvement of 4G cellular
current available spectrum for realizing the performance.
systems. Hence the need for efficient interference
So to overcome this difficulty, spectrum will be made avail-
management schemes is vital. Below are the two interference
able under horizontal or vertical spectrum sharing systems.
management techniques [38]:
The significance of spectrum sharing is probable to
1) ADVANCED RECEIVER
increase, dedicated licensed spectrum access is expected to
remain the baseline approach for mobile broadband which
Modern day and growing cellular system, interference grow
provides reliability and investment certainty for cellular
as a big threat, so to mitigate or manage interference, an
mobile broadband systems. Network components using joint
appropriate interference management technique is the need of
spectrum are likely to play a balancing role [40].
the hour. Advanced interference management at the receiver,
There are mainly two spectrum sharing techniques
or an advanced receiver is the technique which will somewhat
that enable mobile broadband systems to share spectrum
help in interference management. It will detect and even
and are classified as distributed solutions and
try to decode the symbols of the interference signal within
centralized solutions [40]. In a distributed solution the
the modulation constellation, coding scheme, channel, and
systems coordinate amid each other on an equal basis while
resource allocation. Then based on the detector output, the
in a centralized solution each system coordinates discretely
interference signals can be reconstructed and cancelled from
with a central unit and the systems do not directly interact
the received signal so as to improve the anticipated signal
with each other.
decoding performance [38].
Advanced receivers not only limits to inter cell interference 1) DISTRIBUTED SPECTRUM SHARING TECHNIQUES
at the cell boundaries, but also intra cell interference as in Distributed spectrum sharing techniques is more efficient as
the case of massive MIMO. According to LTE-Advanced it can take place in a local framework. Its principle is to
Release 10, every base station transmitter has been equipped only manage those transmissions that really create interfer-
with up to eight antennas which will call for intra cell ence amid systems. Distributed coordination can be entirely
interference, as the number of antenna’s increases. [38]. included into standards and thus they can work without the
need for commercial contracts between operators [40].
2) JOINT SCHEDULING The management of horizontal spectrum sharing
In LTE standard, Releases 8 and 9, interference random- happens through the clear exchange of messages unswerv-
ization through scrambling of transmitting signals is the ingly between the sharing systems through a distinct interface
only interference management strategies that were con- in a peer to peer coexistence protocol. This protocol describes
sidered and there were no advanced co-channel interfer- the performance of the nodes on the receiving of certain
ence management strategies. But in 3GPP LTE-Advanced, messages or taking place of certain events. An example of
Release 10 and 11, through probability readings, it was this is explained in [41].
realized that there was a space for additional performance The systems frequently transmit generally understood
improvement at the cell edges with the help of synchronized signals that will show presence, activity factor and the time
transmission among multiple transmitters dispersed over when they will transmit in a coexistence beacon based solu-
different cell sites [38]. tions. The information that is available openly can be used by
For calibrating the development, some typical coordinated the other systems to adjust their spectrum access performance
multipoint schemes, like to coordinate scheduling, for providing fair spectrum sharing. Coexistence beacons
coordinated beam forming, dynamic point selection, and joint are possibly the solution for both, horizontal and vertical
transmission, were normally conferred [38]. sharing setups. An example of its implementation is the
In the article [38], joint scheduling is broadly used to refer 802.22.1 standard [42].
advanced interference management of cellular systems and MAC behavior based schemes uses a MAC protocol which
link variation from the network side. But as in coordinated is designed to allow horizontal spectrum sharing. Bluetooth
multipoint schemes, the transmission rates and schemes of using frequency hopping and WLAN systems using request
multiple cells are not autonomously determined. In the case to send/clear to send functionality are some of the exam-
of fast network distribution and interoperability, advanced ples. For an even horizontal coexistence with Wi-Fi systems,
interference management schemes by joint scheduling from a Wi-Fi coexistence mode is adapted. The MAC protocol may
VOLUME 3, 2015 1217

leave silent periods for Wi-Fi systems to operate and use a It is an area of concern that the academic researchers and the
listen before talk method which allows Wi-Fi systems to gain industry in this area has reached a point of fading returns.
access to the channel. Its future will now depend on the multi institutional research
In Spectrum sensing and dynamic frequency selection, teams that are working on a new approach with real world
operating frequency range is dynamically selected on the experimental deployments of cognitive radio networks [46].
basis of measurement results like energy detection or feature
detection. To detect the aforementioned coexistence beacons, D. DEVICE TO DEVICE COMMUNICATION SYSTEM
feature detection is highly useful. Due to a hidden node Device to Device Communication system can be explained
problem, this method is not considered as a very dependable by visualizing a two level 5G cellular network and named
method [40]. them as macro cell level and device level. The macro cell level
comprises of the base station to device communications as in
2) CENTRALIZED SPECTRUM SHARING TECHNIQUES an orthodox cellular system. The device level comprises of
The Centralized spectrum sharing technique is useful for the device to device communications. If a device links the cellular
systems that have granularity of spectrum sharing on a higher network through a base station, then it will be operating in
level than the actual radio resource allocation granularity. the macro cell level and if a device links directly to another
This technique has some restraints, as it is conservative and device or apprehends its transmission through the support of
possibly separate users on orthogonal resources without com- other devices, then it will be on the device level. In these types
plete information on whether they would actually interfere or of systems, the base stations will persist to attend the devices
not. While the benefits are in terms of reliability, certainty and as usual. But in the congested areas and at the cell edges, an
control. ad hoc mesh network is created and devices will be permitted
Geo-location database method is an example of a to communicate with each other [47].
centralized sharing technique which involves the querying of In the insight of device level communications, the base
a database to obtain information about the resources available station either have full or partial control over the resource
at a particular location [43]. This is the required classical allocation amid source, destination, and relaying devices,
vertical sharing solution for accessing the locally unused or not have any control. Thus, we can describe the
TV bands [44]. subsequent four main types of device-level communications
The spectrum broker approach is one of the example of a (Figs. 5-8) [47]:
centralized sharing technique in which horizontally sharing
systems negotiate with a central resource management unit
for getting short term grants to use spectrum resources on a
limited basis [45].
Both the Geo-location database and the spectrum
broker approach may additionally support horizontal sharing
between unlicensed systems [40].
However, along with the two above spectrum sharing
techniques most easily usable spectrum bands have also been
allocated, but various studies have revealed that these bands
are significantly underutilized. These concerns have driven
the researchers to innovate a new radio technology which
will encounter with the upcoming demands both in terms of
spectrum efficiency and performance of certain applica-
tions. To encounter the demand of the future, a disruptive
technology revolution that will empower the future wireless
world is Cognitive Radio. Cognitive radios are completely
programmable wireless devices and has an extensive adap-
tation property for achieving better network and application
performance. It can sense the environment and dynamically
performs adaptation in the networking protocols, spectrum
utilization methods, channel access methods and transmis-
FIGURE 5. Device relaying communication with base station controlled
sion waveform used. It is expected that cognitive radio tech- link formation.
nology will soon arise as a general purpose programmable
radio. Similar to the role of microprocessors in the computa-
tion, cognitive radio will also serve as a universal platform for 1) DEVICE RELAYING WITH BASE STATION
wireless system expansion. But the task of successfully build- CONTROLLED LINK FORMATION
ing and large scale deployment of cognitive radio networks This type of communication is applicable for a device which
to dynamically improve spectrum use is an intricate task. is at the edge of a cell, i.e. in the coverage area which
1218 VOLUME 3, 2015

FIGURE 6. Direct device to device communication with base station FIGURE 8. Direct device to device communication with device controlled
controlled link formation. link formation.
2) DIRECT DEVICE TO DEVICE COMMUNICATION WITH

BASE STATION CONTROLLED LINK FORMATION
In this type of communication, the source and destination
devices are exchanging data with each other without the
involvement of a base station, but they are supported by the
base station for link formation.
3) DEVICE RELAYING WITH DEVICE

CONTROLLED LINK FORMATION
In this type of communication, a base station is neither
involved in link formation nor for communication purpose.
So, source and destination devices are totally responsible for
synchronizing communication using relays amid each other.
4) DIRECT DEVICE TO DEVICE COMMUNICATION

WITH DEVICE CONTROLLED LINK FORMATION
In this type of communication, the source and destination
devices have direct communication with each other and the
link formation is controlled itself by the devices without any
assistance from the base station. Hence, the resource should
be utilized by the source and destination devices in a way
FIGURE 7. Device relaying communication with device controlled link to certify limited interference with other devices in the same
formation. level and the macro cell level.
For a substantial advancement in excess of traditional
cellular system architecture, a dualistic cellular system should
have poor signal strength. In this type of communication, the be designed. For introducing the concept of device to device
devices will communicate with the base station by relaying communication, some technical issues needs to be addressed
their information through other devices. like security and interference management issues [47].
This type of communication will be helpful for the As in device to device communication, the routing of user
device to attain a higher quality of service and respective data is through the devices of the other users, so the main
increased battery life. For partial or full control link area of concern is about security because the privacy need
formation, the base station communicates with the relaying to be maintained. Closed access will ensure their security
devices. for the devices that want to operate in the device level.
VOLUME 3, 2015 1219

In closed access, a device has a list of certain reliable devices, amid devices. Basically the base station has complete control
like the users in the close vicinity or office to whom you are over the device to device connections, like connection setup
familiar with, otherwise the users that have been legitimated and maintenance, and resource allocation. Since device to
through a reliable party like an association, can unswervingly device connections share the cellular licensed band in the
communicate with each other, sustaining a level of discretion, device level with the regular cellular connections in the macro
whereas the devices not on this list need to use the macro cell level. So for assigning resources to every device to device
cell level to communicate with it. Also to prevent divulging connection, the network can either assign resources in an
of their information to other devices in a group, one can identical manner as a regular cellular connection or in the
set an appropriate encryption amongst one another. Instead form of a dedicated resource pool to all devices to device
of this, in open access, each device can turn in to relay for connections [47].
other devices deprived of any limits. Meanwhile, in such In device relaying communication with device controller
an instance security is an open research problem. Security and direct device to device communication with device con-
problems in device to device communication contain the troller, there is no base station to control the communication
empathy of possible attacks, threats, and weakness of the amid devices. As shown in Figs. 7 and 8, several devices are
system. To discourse security problems in open access device communicating with each other by using supportive or non-
to device, the research on the security problems of machine supportive communication by playing the role of relays for
to machine communication [48]–[52] can be utilized. the other devices. Since there is no centralized supervision of
Second technical issue of a dualistic system that need to be the relaying, so distributed methods will be used for processes
addressed is of interference management. In device relaying like connection setup, interference management, and resource
communication with the base station controller and direct allocation. In this type of communication, two devices need
device to device communication with base station controlled, to find each other and the neighboring relays first by periodi-
the base station can execute the resource allocation and call cally broadcasting their identity information. This will aware
setup process. So, the base station, to a certain degree can the other devices of their presence and then they will decide
ease the problem of interference management by using cen- whether or not to start a device to device direct or device
tralized methods. But in device relaying communication with relaying communication [53].
device controller and direct device to device communication Now to know the effect of relay’s, let us study a system
with device controller, resource allocation between devices model for relay aided device to device communication [58]
will not be supervised by the centralized unit. Devices will as shown in Fig. 9. For studying it, let us consider that
unavoidably effect macro cell users because they are working the cellular user equipment eNodeB links are unfavorable
in the same licensed band. So to confirm the nominal effect for direct communication and need the assistance of relays.
on the performance of prevailing macro cell base stations, a The device to device user equipment’s are also supported by
dualistic network needs to be considered that involves differ- the relay nodes due to long distance or poor link condition
ent interference management techniques and resource alloca- between peers.
tion schemes. In addition to the interference amid the macro
cell and device levels, interference amid users at the device
level is also of prime concern. For performing the resource
allocation in this type of communication, different algorithms
as shown in table 4 and methods like resource pooling [53],
non-cooperative game [54] or bargaining game, admission
control and power allocation [55], cluster partitioning, and
relay selection [56] can be engaged.
In device relaying communication with the base station
controller, as shown in Fig. 5, since the base station is one
of the communicating units, so the aforementioned chal-
lenges can be addressed with the help of the base station
like authenticating the relaying devices through encryption
for maintaining adequate privacy of the information of the
devices [57]. The challenge of spectrum allocation amid the
relaying devices to prevent them from interfering with other
devices will also be managed by the base station.
In direct device to device communication with base station
controlled, shown in Fig. 6, the devices communicate directly FIGURE 9. A single cell with multiple relay nodes.
with each other, but the base station controls the formation of
links between them. Precisely, the work of the base station is a: NETWORK MODEL
to authenticate the access, control the connection formation, Let us consider a device to device enabled cellular network
resource allocation, and also deals with financial interaction with multiple relays as shown in Fig. 9. A relay node in
1220 VOLUME 3, 2015

4G (LTE-Advanced) is connected between the radio access given by

network and both the cellular and devices to the device user (n)
equipment’s through a donor eNodeB with a wireless connec- (n) hul ,l
γul ,l,1 =P (15)
tion. Let L = {1, 2, . . . , L} represents the set of fixed location (n) (n)
∀uj ∈Uj ,j6=l,j∈L Puj ,j guj ,l + σ2
relays [57] in the network. The system bandwidth is divided
into N resource blocks denoted by N = {1, 2, . . . ., N }. Relay The unit power SINR for the link between relay l and
node can be used for scheduling and resource allocation for eNodeB for cellular user equipment ul (i.e.,ul ∈ {C ∩ Ul )
the device to device user equipment’s, when the link condition in the second hop is as follows:
between two devices to device user equipment’s is too poor (n)
hl,eNodeB
(n)
for direct communication. In addition, the direct communica- γl,ul ,2 = P (n) (n)
(16)
tion between two devices to the device user equipment’s also ∀uj ∈{D∩Uj },j6=l,j∈L Pj,uj gj,eNodeB + σ2
requires the aid of a relay node. Both cellular and device to
In the same way, the unit power SINR for the link between
device user equipment’s assisted by relay ` are denoted by u` .
relay l and receiving device to device user equipment for the
The set of user equipment’s S assisted by relay ` is U T` such that device to device user equipment’s ul (i.e., ul ∈{D ∩ Ul ) in the
U` ⊆ {C ∪ D}, ∀` ∈ L, ` U` = {C ∪ D}, and ` U` = ∅.
second hop can be written as
In the second step of communication, there could be mul-
tiple relays communicating to their related device to device (n)
(n) hl,ul
user equipment’s. According to our assumed system model, γl,ul ,2 = P (n) (n)
(17)
relays are useful for scheduling and resource allocation for ∀uj ∈Uj ,j6=l,j∈L Pj,uj gj,ul + σ2
the user equipment’s to reduce the computational load at the (n)
eNodeB [58]. In (15)–(17), Pi,j is the transmit power in the link between
i and j over resource block n, σ 2 = N0 BRB , where BRB is
b: RADIO PROPAGATION MODEL bandwidth of an resource block, and N0 denote thermal noise.
(n) (n)
For realizing and exhibiting the propagation channel, hl,eNodeB is the gain in the relay and eNodeB link and hl,ul is
distance dependent path loss and shadow fading are consid- the gain in the link between relay l and receiving device to
ered and assumed that the channel is experiencing Rayleigh device user equipment corresponding to the device to device
fading. Particularly, 3GPP propagation environment pre- transmitter user equipment’s ul .
sented in [60] is considered. For example, link between user The attainable data rate for ul in the first hop can be
equipment and relay or between relays and device to device expressed as
follows the following path loss equation (n) (n) (n)
rul ,1 = BRB log2 (1 + Pul ,l γul ,l,1 )
PLul ,l (l)[dB] = 103.8 + 20.9 log (l) + Lsu + 10 log (ζ ) (13) In the same way, the attainable data rate in the second hop
is given by
Where l is the distance between user equipment and relay (n) (n) (n)
in kilometer, Lsu is interpreted as shadow fading and is rul ,2 = BRB log2 (1 + Pl,ul γl,ul ,2 )
demonstrated as a log normal random variable, and ζ is an Since we are considering a two hop communication
exponentially distributed random variable which denotes the approach, the end to end data rate for ul on resource block n
rayleigh fading channel power gain. In the same way, the path is the half of the minimum attainable data rate over two
loss equation for the relay and eNodeB link is expressed as hops, i.e.,
PLl.eNodeB (l)[dB] = 100.7 + 23.5 log (l) + Lsr + 10 log (ζ ) 1 (n) (n)
min{rul ,1 , rul ,2
R(n)
ul = (18)
(14) 2
The ongoing problem in device to device communication
Where Lsr is a log normal random variable accounting for is about Resource allocation. So a lot of researchers are
shadow fading. Hence, given the distance l, the link gain working on to propose an optimal resource allocation
between any pair of network nodes i, j can be calculated algorithm. Table 5 will provide a brief summary on the
as 10−(PLi,j (l)/10) . proposed algorithms.
c: REALIZABLE DATA RATE E. ULTRA DENSE NETWORKS

(n)
hi,j can be denoted as the direct link gain between node To meet the increasing traffic demands due to the increased
i and j over resource block n. The interference link gain number of users, densification of the infrastructure will be
between relay (user equipment) i and a user equipment (relay) the prior aspect of 5G communications. But for achieving
(n)
j over resource block n is denoted by gi,j where user equip- ultra-dense, heterogeneous networks will play an important
ment (relay) j is not associated with relay (user equipment) i. role. With the introduction of moving networks and ad-hoc
The unit power SINR for the link between user equipment social networks, the heterogeneous networks are becoming
ul ∈ Ul and relay l using resource block n in the first hop is more dynamic. Though dense and dynamic heterogeneous
VOLUME 3, 2015 1221

TABLE 5. Summary of proposed algorithms for optimal resource allocation in device to device communication.
networks will give rise to new challenges in terms of With the introduction of smart wireless devices, the
interference, mobility and backhauling. To overcome these interaction between these devices and with the environment
challenges, there arises a requirement of designing new are destined to increase. To meet the challenges that have
network layer functionalities for maximizing the performance arisen because of the increasing density of nodes and inter-
farther from the design of the existing physical layer. changing connectivity options, there arises a need of the
In present networks like Long Term Evolution (LTE), user independent algorithms. So future smart devices are
there exists interference mitigation techniques like enhanced designed in such a way that with the help of the context
Inter-Cell Interference Coordination and autonomous com- information, they will learn and decide how to manage the
ponent carrier selection. But these techniques are applica- connectivity. Contextual information possibly will be the
ble only to nomadic and dense small cell deployments and approaching service profile, battery position of a device or a
have limited flexibility. So for 5G networks, the interference complete data acquired through either in built sensors, cloud
mitigation techniques should be more flexible and open to servers or serving base station. For example, to enable faster
the variations as changes in the traffic and deployment are initialization of direct Device-to-Device communications and
expected to occur more rapidly than existing networks [66]. native multicast group making, context information about the
1222 VOLUME 3, 2015

social networking will be very helpful as it will decrease the G. FULL DUPLEX RADIOS
signaling overhead in the network. Context information can For a long duration of communication period, it is assumed in
also provide sustenance for the network to decrease energy the wireless system design that radios have to operate in half
consumption in base stations because of the switching of cells duplex mode. It means that it will not transmit and receive
by improving the mobility and traffic management simultaneously on the same channel. Many scholars, aca-
procedures and local handover strictures [66]. demics and researchers at different universities and research
In short, future smart devices and small cell networks groups have tried to undermine this assumption by proposing
will be capable of providing the best wireless connectivity many designs to build in-band full-duplex radios.
with minimum interference and less power consumption. But the realization to build full duplex radio has a lot of
Along with this, they should be rapidly adaptable to implications. The cellular networks will have to reduce their
the changing requirements of devices and radio access spectrum demands to half as only a single channel is used
network. for achieving the same performance. As in LTE, for both
uplink and downlink, it uses equal width separate channels
F. MULTI RADIO ACCESS TECHNOLOGY ASSOCIATION for empowering radios to realize full duplex.
As we are heading towards 5G, the networks are becoming For communicating in the full duplex mode, the self-
more heterogeneous. The main aspect that has attracted many, interference results from its own transmission to the received
is the integration among different radio access technologies. signal has to be completely removed. Let us consider the case
A distinctive 5G aided device should be manufactured whose of WiFi signals which are transmitting at 20dBm (100mW)
radios not only support a new 5G standard like millimeter average power with the noise floor of around −90dBm.
wave frequencies, but also 3G, various releases of 4G LTE, So the transmit self-interference need to be canceled by
numerous types of WiFi, and possibly direct device to device 110dB (20dBm-(−90dBm)) to achieve the similar level as of
communication, all across the different spectral bands [67]. the noise floor and reduce it to insignificant. If any residual
So, defining of standards and utilization of spectrum to which self-interference is not completely canceled, then it will acts
base station or users will be a really intricate job for the as noise to the received signal, which in turn reduces SNR
network [68]. and subsequently throughput [82].
Defining of the optimal user association is the prime
area of concern which depends on the signal to interfer-
ence and noise ratio from every single user to every single H. A MILLIMETER WAVE SOLUTION
base station, the selections of other users in the network, FOR 5G CELLULAR NETWORK
the load at every single base station, and the prerequisite The Wireless industry has been growing day by day and in
to apply the same base station and standard in both uplink spite of the efforts by the industrial researchers for creating
and downlink for simplifying the operation of control the proficient wireless technologies, the wireless industry
channels for resource allocation and feedback [69], [70]. continuously facing the overpowering capacity demands from
So, certain procedures must be implemented to overcome its current technologies. Recent innovations in computing and
these issues. communications and the arrival of smart handsets along with
To increase edge rates by as much as 500%, a simple, the need to access the internet poses new anxieties in front
apparently highly suboptimal association method centered of the wireless industry. These demands and anxieties will
on aggressive but static biasing towards small cells and grow in the approaching years for 4G LTE and indicates that
blanking about half of the macrocell transmissions has been at some point around 2020, there will arise a problem of con-
shown in [71]. The combined problem of user association and gestion in wireless networks. It will be must for the research
resource allocation in two tier heterogeneous networks, with industry to implement new technologies and architectures for
adaptive tuning of the biasing and blanking in each cell, is meeting the increasing demands of the users. The ongoing
considered in [69], [70], and [72]–[77]. A model of hotspot work plans a wireless future in which data rates increase to
traffic shows that the optimal cell association is done by rate the multi gigabit per second range. These high data rates
ratio bias, instead of power level bias [73]–[75]. An active can be attainable with the help of steerable antennas and the
model of cell range extension as shown in [79], the traffic millimeter wave spectrum and at the same time will support
arrives as a Poisson process in time and at the possible arrival mobile communications and backhaul networks [83].
rates, for which a steadying scheduling policy subsists. With Recent researches have put forward that mm-wave
massive MIMO at the base stations, user association and frequencies of 2.6 GHz radio spectrum possibly will sup-
load balancing in a heterogeneous networks, is considered plement the presently saturated 700 MHz band for wireless
in [79]. An exciting game theoretic approach is used in [80] communications [84]. Feasibility of millimeter wave wireless
for the problem of radio access technology selection, in which communications is supported by the fact that the use of high
union to Nash equilibria and the Pareto-efficiency of these gain, steerable antennas at the mobile and base station and
equilibria are deliberated [67]. cost effective CMOS technology can now operate well into
In conclusion, there is a vast scope for modeling, exploring the millimeter wave frequency bands [85]–[87]. Additionally,
and optimizing base station-user associations in 5G [81]. with the use of millimeter wave carrier frequencies,
VOLUME 3, 2015 1223

larger bandwidth allocations will come up with higher data because of unfriendly channel conditions like path loss effect,
transfer rates and service providers that are presently using absorption due to atmosphere and rain, small diffraction and
20 MHz channels for 4G customers will now significantly penetration about obstacles and through objects respectively.
expand the channel bandwidths [87]. With the increase in There is one more reason of unsuitability is due to strong
bandwidth, capacity will also get increased, while the latency phase noise and excessive apparatus costs. But the prevailing
will get decreased, which give rise to better internet based reason is that the large unlicensed band around 60 GHz [89],
access and applications like real time streaming. Since the were appropriate primarily for very short range transmis-
wavelength of millimeter wave frequencies are very small, sion [90]. So, the emphasis had been given to both fixed
so it will utilize polarization and different spatial processing wireless applications in the 28, 38, 71–76 and 81–86 GHz
techniques like massive MIMO and adaptive beam- and WiFi with the 802.11ad standard in the 60 GHz band.
forming [15]. With the significant increase in bandwidth, Semiconductors are also evolving, as their costs and power
the data links to densely populated areas will now handle consumption values are decreasing rapidly due to the growth
greater capacity than present 4G networks. Likewise the base of the abovementioned short range standards. The main prop-
stations are constantly reducing the coverage areas of the cell agation issues regarding millimeter wave propagation for 5G
for spatial reuse, cooperative MIMO, relays and interference cellular communication are [67]:
mitigation between base stations. Since the base stations are
abundant and more densely dispersed in urban areas, which 1) PATH LOSS
will reduce the cost per base station. Spectrum distributions The free space path loss is dependent on the carrier frequency,
of over 1 GHz of bandwidth are currently being utilized in as the size of the antennas is kept constant which is measured
the 28 GHz and 38 GHz bands. by the wavelength λ = c/fc , where fc is the carrier frequency.
By far as for the concern of building a prototype, the Now as the carrier frequency increases, the size of the anten-
antenna is essentially being positioned in very close vicinity nas got reduced and their effective aperture increases with
to the 28 GHz Radio Frequency Integrated Circuit and the the factor of 4πλ2
, while the free space path loss between a
front end module because there will be high signal attenua- transmitter and a receiver antenna grows with fc2 . So, if we
tion at 28 GHz. Realizing the antenna array directly on the increase the carrier frequency fc from 3 to 30 GHz, it will
printed circuit board of the 5G cellular device will minimize correspondingly add 20 dB of power loss irrespective of the
the insertion loss between the antenna and Radio Frequency transmitter-receiver distance. But for increased frequency, if
Integrated Circuit. This infers that an employment of the the antenna aperture at one end of the link is kept constant,
Radio Frequency blocks in the 5G architecture before the then the free-space path loss remains unchanged. Addition-
intermediate frequency stage will be reliant on the placement ally, if both the transmitter and receiver antenna apertures
of the 28 GHz antenna array in the cellular phone. Taking are kept constant, then the free space path loss decreases
this concept into a thought, a minimum set of two 28 GHz with fc2 [67].
antenna arrays is proposed for millimeter wave 5G cellular
applications in [88], the two antenna arrays are employed
2) BLOCKING
in the top and bottom part of the cellular device. The
28 GHz antenna array configuration for 5G cellular mobile Microwave signals are less prone to blockages but it dete-
terminals and its comparison with the 4G standard is given riorates due to diffraction. In the contrary, millimeter wave
in table 6. signals suffer less diffraction than the microwave signals and
exhibit specular propagation, which makes them much more
TABLE 6. 28 GHz antenna array configuration for 5G cellular mobile vulnerable to blockages. This will fallout as nearly bimodal
terminals and its comparison with the 4G standard. channel subject to the existence or lack of Line of Sight.
Recent studies in [84] and [91] reveals that, with the increase
in the transmitter and receiver distance the path loss increases
to 20 dB/decade under Line of sight propagation, but descents
to 40 dB/decade plus an added blocking loss of 15–40 dB for
non-line of sight [67].
So due to the presence of blockages, the set connection will
promptly shift from usable to unusable which will results in
large scale impediments that cannot be avoided with typical
small scale diversity countermeasures.
3) ATMOSPHERIC AND RAIN ABSORPTION

Within the unlicensed 60-GHz band, the absorption due to
The millimeter wave spectrum is under-utilized and is left rain and air particularly the 15 dB/km oxygen absorption are
idle until present years. The main reason behind the under- more perceptible. But these absorptions are insignificant for
utilization is its unsuitability for cellular communications the urban cellular deployments, where base station spacing’s
1224 VOLUME 3, 2015

might be on the order of 200 m. But actually, these type

of absorptions are useful as it will efficiently increase the
segregation of each cell by further attenuating the background
interference from more distant base stations [67].
So from the above explanation, it is inferred that the propa-
gation losses for millimeter wave frequencies are resolvable,
but only by steering the beam energy with the help of large
antenna arrays and then collect it coherently. But for practical
viability, the concept of narrow beam communication is fresh
for cellular communications and poses problems like:
a: LINK ACQUISITION
FIGURE 10. Virtual resource cloud made up of mobile devices in the
The main problem that the narrow beams are facing is in vicinity.
establishing links amid users and base stations for both initial
access and handoff. The user and base stations will have
to locate each other by scanning lots of angular positions
where the possibility of a narrow beam is high. This problem
poses an important research challenge predominantly in the
perspective of high mobility [67].
b: NEED OF NEW TRANSCEIVER ARCHITECTURES

Wireless millimeter wave systems have gone through
significant improvement but still there are some hardware
issues which will affect the designing of the communication FIGURE 11. A cloudlet enabling mobile devices to bypass latency and
systems. The analog to digital and digital to analog con- bandwidth issues while benefitting from its resources.
verters needed for large bandwidths are the prime cause of

power consumption. A prime reason of power consumption is
because of the use of large antenna arrays. Along with these,
high receiver sensitivities are needed to deal with the path loss
because it is not feasible that each antenna will be provided
with normal fully digital beam formers [67].
I. CLOUD TECHNOLOGIES FOR FLEXIBLE

5G RADIO ACCESS NETWORKS
1) MOBILE CLOUD COMPUTING
In the recent years, mobile cloud computing has earned a
lot of admiration as it is a coalition of many computing
fields. It offers computing, storage, services, and applications
over the Internet. It also reduces cost, disconnect services
from the existing technology, and offers flexibility in terms
of resource provisioning. So mobile cloud computing can
be defined as an incorporation of cloud computing tech-
nology with mobile devices. This integration will make the
mobile devices resource full in terms of computational power,
memory, storage, energy, and context awareness [92]. Mobile
cloud computing can also be explained with different con-
cepts of the mobile cloud [93].
In the first method, let us consider that the other mobile
devices will also act as resource providers as in [95]. FIGURE 12. Flexible functional split [103].
So the combined resources of the numerous mobile devices
and other available stationary devices in the local area will in Fig. 11, where a local cloudlet encompassed by numerous
be exploited as shown in Fig. 10. This method supports multi core computers with connectivity to the remote cloud
user mobility and identifies the potential of mobile clouds to servers is used by the mobile device to relieve from its work-
perform collective sensing. load. Plug Computers having form factor, diversity and low
The cloudlet concept proposed in [96] is the second power consumption can be considered as good contenders for
method of mobile cloud computing. This method is explained cloudlet servers. But these computers are ideal for small scale
VOLUME 3, 2015 1225

FIGURE 15. Machine to Machine traffic to increase 40-fold from

2010 to 2015.
FIGURE 13. Characteristics of a radio access network as a service
implementation.
FIGURE 14. Number of Machine to Machine (M2M) connections in

mobile [105].
FIGURE 16. Cell spectral efficiency in 5G networks [105].
servers installed in the public organization because they have

the similar general architecture as a normal computer and are
less powerful, smaller, and less costly. Hence, these cloudlets
should be installed in public areas like restaurants so that
mobile devices can connect directly with the cloudlet instead
of a remote cloud server to remove latency and bandwidth
problems [93].
Mobile cloud computing follows the basic concepts of
cloud computing. There are some specific requirements that
need to be encountered in a cloud like adaptability, scalability,
availability and self-awareness as discussed in [94].
So mobile cloud computing should also fulfill these FIGURE 17. Demand to delay in control and user planes for
requirements. For example, a mobile computing cloud 4G/5G networks [105].
should be cognizant of its availability and dynamically plug
themselves in, depending on the requirements and workload. 2) RADIO ACCESS NETWORK AS A SERVICE
An appropriate technique of self pretentious one’s own qual- Centralization is the prime objective of 5G mobile networks
ity is desirable for mobile users to proficiently take advantage because processing and management will need to be flexible
of the cloud, as the internal status and the external environ- and adapted to the actual service requirements. This will
ment is subject to change. Others facets like mobility, low lead to a compromise between the decentralized today’s net-
connectivity and limited source of power also needed to be work and fully centralized cloud radio access network. This
considered [93]. compromise is addressed by the radio access network as a
1226 VOLUME 3, 2015

TABLE 7. 5G related activities in Europe [109].
service concept, which partly centralizes functionalities of the service does not fully centralize all radio access network
radio access network depending on the needs and character- functionalities [103].
istics of the network. The Radio access network as a service Functional split realization poses a serious challenge for
is an application of the software as a service paradigm [97], the radio access network. Theoretically, the functional split
so every function may be packed and distributed in the occur on every protocol layer or on the interface amid each
form of a service within a cloud platform. This will cause layer. Present architecture involves restraints on the func-
increased data storage and processing capabilities, as pro- tions between discrete protocol layers. So with a restrained
vided by a cloud platform accommodated in data centers. The backhaul, most of the radio protocol stack and radio resource
design of radio access network as a service based on cloud management will accomplished locally, while functions with
enables flexibility and adaptability from different percep- less restrained requirements like bearer management and load
tions. Recent advances in Cloud radio access network is given balancing are placed in the radio access network as a service
in [98]–[102]. platform. So when a high capacity backhaul is available,
There is a flexible functional split of the radio protocol lower-layer functions like PHY and MAC are shifted for a
stacks as shown in Fig. 12 is present in the central element higher degree of centralization into the radio access network
of radio access network as a service between the central as a service platform [103].
radio access network as a service platform and the local radio The following list as shown in Fig. 13 condenses major
access points. With the introduction of this functional split, characteristics of a radio access network as a service
degrees of freedom increases. implementation similar to the basic characteristics of a cloud
The left side demonstrates a traditional LTE employment computing platform and is explained in [103].
in which all functionalities up to admission/congestion con-
trol are locally employed at the base station. The right side 3) JOINT RADIO ACCESS NETWORK BACKHAUL OPERATION
illustrates the cloud radio access network approach in which The main reliability factor of 5G wireless networks is densely
only the radio front-end is locally employed, and all the rest spread small cell layer which necessitates to be connected to
functionality is centralized. But radio access network as a the radio access network as a service platform. Though,
VOLUME 3, 2015 1227

TABLE 8. 5G related activities in America [109].
the need of deployment of small cells is in the places where critical part of the infrastructure. In particular, there is a need
the line of sight centered microwave solutions are either hard of flexible centralization for dynamic adaptation of network
or too costly to deploy for backhaul. Hence, the need to con- routes. The degree of radio access network centralization
nect small cells at diverse locations made backhaul network a depends on available backhaul resources.
1228 VOLUME 3, 2015

So there is a need of a refined transport network design demands with the same demands to 4G networks shows the
that can convey the data headed towards the central unit free progress of spectral efficiency by 3-5 times [105].
of the degree of centralization. This is an important necessity Assessment of demands to delay in control and user
for maximum flexibility when the introduction of the new planes for signaling traffic and user traffic respectively is
functionalities to the network is taking place. shown in Fig. 17. This figure depicts that the demands to
But the complications increases in routing and 5G networks will be twice more firm for traffic in the
classification of data packets according to their quality user plane and 10 times more firm in the subscriber traffic
of service. On the other hand, software defined network plane [106].
provides quicker reaction to link/node letdowns, higher uti-
lization of the accessible resources, and faster deployment V. CONCLUSION
of new updates with ease. These advantages have come up In this paper, a detailed survey has been done on the per-
with a centralized control example, which streamlines the formance requirements of 5G wireless cellular communica-
arrangement and management, but with increased computation systems that have been defined in terms of capacity,
tional efforts, as algorithmic complexity increases [103]. Also data rate, spectral efficiency, latency, energy efficiency, and
for spectrum utilization, software defined radio (SDR) and Quality of service. A 5G wireless network architecture has
software defined networks (SDN) are the optimum solution been explained in this paper with massive MIMO technology,
and the study in [104] revealed that the co-existence of SDR network function virtualization (NFV) cloud and device to
and SDN is essential, and the optimal results can be attained device communication. Certain short range communication
only by co-existence and joint compliments. technologies, like WiFi, Small cell, Visible light communica-
tion, and millimeter wave communication technologies, has
J. TRENDS AND QUALITY OF SERVICE been explained, which provides a promising future in terms
MANAGEMENT IN 5G of better quality and increased data rate for inside users and
5G technologies are likely to appear in the market in 2020. at the equivalent time reduces the pressure from the outside
It is expected to significantly improve customers Quality of base stations. Some key emerging technologies have also
Service in the context of increasing growth of data volume been discussed that can be used in 5G wireless systems to
in mobile networks and the growth of wireless devices with fulfill the probable performance desires, like massive MIMO
variety of services provided. Some general trends related and Device to Device communication in particular and inter-
to 5G can be explained in terms of machine to machine ference management, spectrum sharing with cognitive radio,
traffic and number of machine to machine connections in ultra dense networks, multi radio access technology, full
mobile [105]. duplex radios, millimeter wave communication and Cloud
Based on the projections as shown in Fig. 14, in 2018 the Technologies in general with radio access networks and soft-
number of machine to machine (M2M) connections in the ware defined networks. This paper may be giving a good
networks of mobile operators will surpass 15 billion [108], platform to motivate the researchers for better outcome of
which is 2 times more than the present rate, and in 2022 different types of problems in next generation networks.
mobile operators will have more than 26 billion machine to
machine connections. APPENDIX
At the same time the stake of machine to machine con- A list of current research projects based on 5G technologies
nections of the total number of connections in the mobile are given in Table 7, 8 and 9.
operator’s networks will rise from the present 5% to 15%
in 2018 and to 22% in 2022 [105]. TABLE 9. 5G related activities in Asia [109].
A key trend relates to mobility, as broadband mobile usage,

with more than 2.4 billion users globally (as of June 2012)
is expected to be dominant over the coming years. For data
traffic and machine to machine communications, an expected
40-fold increase between 2010 and 2015 is shown in Fig. 15
and a 1000 fold increase is predicted over a decade. This
level of growth force the network operators to provide global
broadband access to all types of heterogeneous and modified
Internet based services and applications [105].
While the Quality of Service management in 5G can be REFERENCES
realized in terms of cell spectral efficiency and latency. [1] R. Baldemair et al., ‘‘Evolving wireless communications: Addressing the
challenges and expectations of the future,’’ IEEE Veh. Technol. Mag.,
Demand to the cell’s spectral efficiency in 5G networks for vol. 8, no. 1, pp. 24–30, Mar. 2013.
diverse transmission channels are shown in Fig. 16. Increased [2] T. Rappaport, Wireless Communications: Principles and Practice.
spectral efficiency of 5G networks can be attained using Englewood Cliffs, NJ, USA: Prentice-Hall, 1996.
[3] T. Halonen, J. Romero, and J. Melero, Eds., GSM, GPRS and EDGE
non-orthogonal access methods in radio access networks and Performance: Evolution Towards 3G/UMTS. New York, NY, USA: Wiley,
by using non-orthogonal signals [107]. Comparison of these 2003.
VOLUME 3, 2015 1229

[4] J. G. Andrews, A. Ghosh, and R. Muhamed, Fundamentals of WiMAX. [28] S. K. Mohammed and E. G. Larsson, ‘‘Per-antenna constant envelope
Englewood Cliffs, NJ, USA: Prentice-Hall, 2007. precoding for large multi-user MIMO systems,’’ IEEE Trans. Commun.,
[5] B. Furht and S. A. Ahson, Eds., Long Term Evolution: 3GPP LTE Radio vol. 61, no. 3, pp. 1059–1071, Mar. 2013.
and Cellular Technology. Boca Raton, FL, USA: CRC Press, 2009, ch. 12, [29] L. Lu, G. Y. Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, ‘‘An
pp. 441–443. overview of massive MIMO: Benefits and challenges,’’ IEEE J. Sel.
[6] S. Sesia, I. Toufik, and M. Baker, Eds., LTE: The UMTS Long Term Topics Signal Process., vol. 8, no. 5, pp. 742–758, Oct. 2014.
Evolution. New York, NY, USA: Wiley, 2009. [30] T. L. Marzetta, ‘‘Noncooperative cellular wireless with unlimited num-
[7] K. R. Santhi, V. K. Srivastava, G. SenthilKumaran, and A. Butare, ‘‘Goals bers of base station antennas,’’ IEEE Trans. Wireless Commun., vol. 9,
of true broad band’s wireless next wave (4G–5G),’’ in Proc. IEEE 58th no. 11, pp. 3590–3600, Nov. 2010.
Veh. Technol. Conf., vol. 4. Oct. 2003, pp. 2317–2321. [31] M. Matthaiou, M. R. McKay, P. J. Smith, and J. A. Nossek, ‘‘On the
[8] C.-X. Wang et al., ‘‘Cellular architecture and key technologies for 5G condition number distribution of complex Wishart matrices,’’ IEEE Trans.
wireless communication networks,’’ IEEE Commun. Mag., vol. 52, no. 2, Commun., vol. 58, no. 6, pp. 1705–1717, Jun. 2010.
pp. 122–130, Feb. 2014. [32] S. Vishwanath, N. Jindal, and A. Goldsmith, ‘‘Duality, achievable rates,
and sum-rate capacity of Gaussian MIMO broadcast channels,’’ IEEE
[9] E. Perahia and R. Stacey, Next Generation Wireless LANs:
Trans. Inf. Theory, vol. 49, no. 10, pp. 2658–2668, Oct. 2003.
Throughput, Robustness, and Reliability in 802.11n. Cambridge,
[33] E. Björnson, M. Kountouris, and M. Debbah, ‘‘Massive MIMO and small
U.K.: Cambridge Univ. Press, 2008.
cells: Improving energy efficiency by optimal soft-cell coordination,’’ in
[10] E. H. Ong, J. Kneckt, O. Alanen, Z. Chang, T. Huovinen, and T. Nihtila,
Proc. 20th Int. Conf. Telecommun. (ICT), May 2013, pp. 1–5.
‘‘IEEE 802.11ac: Enhancements for very high throughput WLANs,’’
[34] E. Björnson, L. Sanguinetti, J. Hoydis, and M. Debbah, ‘‘Optimal design
in Proc. IEEE 22nd Pers. Indoor Mobile Radio Commun., Sep. 2011,
of energy-efficient multi-user MIMO systems: Is massive MIMO the
pp. 849–853.
answer?’’ IEEE Trans. Wireless Commun., vol. 14, no. 6, pp. 3059–3075,
[11] E. Perahia and M. X. Gong, ‘‘Gigabit wireless LANs: An overview Jun. 2015. [Online]. Available: http://arxiv.org/abs/1403.6150
of IEEE 802.11ac and 802.11ad,’’ ACM SIGMOBILE Mobile Comput. [35] Y. Xu, G. Yue, and S. Mao, ‘‘User grouping for massive MIMO in FDD
Commun. Rev., vol. 15, no. 3, pp. 23–33, Jul. 2011. systems: New design methods and analysis,’’ IEEE Access, vol. 2, no. 1,
[12] E. Perahia, C. Cordeiro, M. Park, and L. L. Yang, ‘‘IEEE 802.11ad: pp. 947–959, Sep. 2014.
Defining the next generation multi-Gbps Wi-Fi,’’ in Proc. 7th IEEE [36] M. Peng, D. Liang, Y. Wei, J. Li, and H.-H. Chen, ‘‘Self-configuration
Consum. Commun. Netw. Conf., Jan. 2010, pp. 1–5. and self-optimization in LTE-advanced heterogeneous networks,’’ IEEE
[13] A. B. Flores, R. Guerra, E. W. Knightly, P. Ecclesine, and S. Pandey, Commun. Mag., vol. 51, no. 5, pp. 36–45, May 2013.
‘‘IEEE 802.11af: A standard for TV white space spectrum sharing,’’ IEEE [37] M. Peng, C. Wang, J. Li, H. Xiang, and V. Lau, ‘‘Recent advances in
Commun. Mag., vol. 51, no. 10, pp. 92–100, Oct. 2013. underlay heterogeneous networks: Interference control, resource alloca-
[14] V. Chandrasekhar, J. G. Andrews, and A. Gatherer, ‘‘Femtocell networks: tion, and self-organization,’’ IEEE Commun. Surveys Tuts., vol. 17, no. 2,
A survey,’’ IEEE Commun. Mag., vol. 46, no. 9, pp. 59–67, Sep. 2008. pp. 700–729, Secondquarter 2015.
[15] F. Rusek et al., ‘‘Scaling up MIMO: Opportunities and challenges with [38] W. Nam, D. Bai, J. Lee, and I. Kang, ‘‘Advanced interference manage-
very large arrays,’’ IEEE Signal Process. Mag., vol. 30, no. 1, pp. 40–60, ment for 5G cellular networks,’’ IEEE Commun. Mag., vol. 52, no. 5,
Jan. 2013. pp. 52–60, May 2014.
[16] A. Bleicher, ‘‘Millimeter waves may be the future of 5G phones,’’ [39] 5G Radio Access—Research and Vision, Ericsson, Stockholm, Sweden,
Samsung’s millimeter-wave transceiver technology could enable ultrafast Jun. 2013.
mobile broadband by 2020, Jun. 2013 [40] T. Irnich, J. Kronander, Y. Selen, and G. Li, ‘‘Spectrum sharing scenarios
[17] H. Haas. (Aug. 2011). Wireless Data From Every Light Bulb. [Online]. and resulting technical requirements for 5G systems,’’ in Proc. IEEE 24th
Available: http://bit.ly/tedvlc PIMRC, London, U.K., Sep. 2013, pp. 127–132.
[18] X. Hong, C.-X. Wang, H.-H. Chen, and Y. Zhang, ‘‘Secondary spectrum [41] M. Goldhammer, 802.16h Main Concepts, document IEEE
access networks,’’ IEEE Veh. Technol. Mag., vol. 4, no. 2, pp. 36–43, 802.19-05/0051r0, 2005.
Jun. 2009. [42] G. J. Buchwald, S. L. Kuffner, L. M. Ecklund, M. Brown, and
[19] F. Haider et al., ‘‘Spectral efficiency analysis of mobile Femtocell E. H. Callaway, ‘‘The design and operation of the IEEE 802.22.1 dis-
based cellular systems,’’ in Proc. IEEE ICCT, Jinan, China, Sep. 2011, abling beacon for the protection of TV whitespace incumbents,’’ in Proc.
pp. 347–351. 3rd IEEE DySPAN, Oct. 2008, pp. 1–6.
[20] P. Agyapong, M. Iwamura, D. Staehle, W. Kiess, and A. Benjebbour, [43] ECC REPORT 159, ‘‘Technical and Operational Requirements for the
‘‘Design considerations for a 5G network architecture,’’ IEEE Commun. Possible Operation of Cognitive Radio Systems in the ‘White Spaces’
Mag., vol. 52, no. 11, pp. 65–75, Nov. 2014. of the Frequency Band 470–790 MHz,’’ CEPT, Cardiff, Wales, Tech.
Rep. 159, Jan. 2011.
[21] A. Osseiran et al., ‘‘Scenarios for 5G mobile and wireless communica-
[44] FCC. FCC12-36 Third Memorandum Opinion and Order in the Matter
tions: The vision of the METIS project,’’ IEEE Commun. Mag., vol. 52,
of Unlicensed Operation in the TV Broadcast Bands (ET Docket No. 04-
no. 5, pp. 26–35, May 2014.
186) and Additional Spectrum for Unlicensed Devices Below 900 MHz
[22] M. Fallgren et al., Scenarios, Requirements and KPIs for 5G Mobile and and in the 3 GHz Band (ET Docket No. 02-380). Washington, DC, USA:
Wireless System, document ICT-317669-METIS/D1.1, Apr. 2013. Federal Communications Commission, 2012.
[23] Industry Proposal for a Public Private Partnership (PPP) in [45] J. Huang, R. A. Berry, and M. L. Honig, ‘‘Auction-based spectrum
Horizon 2020 (Draft Version 2.1), Horizon 2020 Advanced 5G sharing,’’ Mobile Netw. Appl., vol. 11, no. 3, pp. 405–418, Jun. 2006.
Network Infrastructure for the Future Internet PPP. [Online]. [46] P. Steenkiste, D. Sicker, G. Minden, and D. Raychaudhuri, ‘‘Future
Available: http://www.networks-etp-eu/fileadmin/user_upload/Home/ directions in cognitive radio network research,’’ in Proc. NSF Workshop
draft-PPP-proposal.pdf Rep., Mar. 2007.
[24] E. G. Larsson, F. Tufvesson, O. Edfors, and T. L. Marzetta, ‘‘Mas- [47] M. N. Tehrani, M. Uysal, and H. Yanikomeroglu, ‘‘Device-to-device
sive MIMO for next generation wireless systems,’’ IEEE Commun. communication in 5G cellular networks: Challenges, solutions, and future
Mag., vol. 52, no. 2, pp. 186–195, Feb. 2014. [Online]. Available: directions,’’ IEEE Commun. Mag., vol. 52, no. 5, pp. 86–92, May 2014.
http://arxiv.org/pdf/1304.6690v1.pdf [48] I. Cha, Y. Shah, A. U. Schmidt, A. Leicher, and M. V. Meyerstein,
[25] J. Nam, J.-Y. Ahn, A. Adhikary, and G. Caire, ‘‘Joint spatial division and ‘‘Trust in M2M communication,’’ IEEE Veh. Technol. Mag., vol. 4, no. 3,
multiplexing: Realizing massive MIMO gains with limited channel state pp. 69–75, Sep. 2009.
information,’’ in Proc. IEEE 46th Annu. Conf. Inf. Sci. Syst., Mar. 2012, [49] J. Yue, C. Ma, H. Yu, and W. Zhou, ‘‘Secrecy-based access control for
pp. 1–6. device-to-device communication underlaying cellular networks,’’ IEEE
[26] A. Pitarokoilis, S. K. Mohammed, and E. G. Larsson, ‘‘On the optimal- Commun. Lett., vol. 17, no. 11, pp. 2068–2071, Nov. 2013.
ity of single-carrier transmission in large-scale antenna systems,’’ IEEE [50] A. Perrig, J. Stankovic, and D. Wagner, ‘‘Security in wireless sensor
Wireless Commun. Lett., vol. 1, no. 4, pp. 276–279, Aug. 2012. networks,’’ Commun. ACM, vol. 47, no. 6, pp. 53–57, Jun. 2004.
[27] C. Studer and E. G. Larsson, ‘‘PAR-aware large-scale multi-user MIMO- [51] Y. Zhou, Y. Fang, and Y. Zhang, ‘‘Securing wireless sensor networks:
OFDM downlink,’’ IEEE J. Sel. Areas Commun., vol. 31, no. 2, A survey,’’ IEEE Commun. Surveys Tuts., vol. 10, no. 3, pp. 6–28,
pp. 303–313, Feb. 2013. Third Quarter 2008.
1230 VOLUME 3, 2015

[52] R. Muraleedharan and L. Osadciw, ‘‘Jamming attack detection and coun- [77] S. Deb, P. Monogioudis, J. Miernik, and J. P. Seymour, ‘‘Algorithms for
termeasures in wireless sensor network using ant system,’’ in Proc. Symp. enhanced inter-cell interference coordination (eICIC) in LTE HetNets,’’
Defense Secur., Orlando, FL, USA, Apr. 2006. IEEE/ACM Trans. Netw., vol. 22, no. 1, pp. 137–150, Feb. 2014.
[53] G. Fodor et al., ‘‘Design aspects of network assisted device-to-device [78] B. Soret, K. I. Pedersen, T. Kolding, H. Kroener, and I. Maniatis, ‘‘Fast
communications,’’ IEEE Commun. Mag., vol. 50, no. 3, pp. 170–177, muting adaptation for LTE-A HetNets with remote radio heads,’’ in Proc.
Mar. 2012. IEEE GLOBECOM, Dec. 2013, pp. 3790–3795.
[54] E. Hossain, D. Niyato, and Z. Han, Dynamic Spectrum Access [79] S. V. Hanly and P. A. Whiting, ‘‘On the capacity of HetNets,’’ in Proc.
and Management in Cognitive Radio Networks. Cambridge, U.K.: ITA, Feb. 2014, pp. 1–9.
Cambridge Univ. Press, 2009. [80] D. Bethanabhotla, O. Y. Bursalioglu, H. C. Papadopoulos, and G. Caire,
[55] D. Feng, L. Lu, Y. Yuan-Wu, G. Y. Li, G. Feng, and S. Li, ‘‘Device- ‘‘User association and load balancing for cellular massive MIMO,’’ in
to-device communications underlaying cellular networks,’’ IEEE Trans. Proc. ITA, Feb. 2014, pp. 1–10.
Commun., vol. 61, no. 8, pp. 3541–3551, Aug. 2013. [81] T. A. Levanen, J. Pirskanen, T. Koskela, J. Talvitie, and M. Valkama,
[56] B. Zhou, H. Hu, S.-Q. Huang, and H.-H. Chen, ‘‘Intracluster device-to- ‘‘Radio interface evolution towards 5G and enhanced local area commu-
device relay algorithm with optimal resource utilization,’’ IEEE Trans. nications,’’ IEEE Access, vol. 2, pp. 1005–1029, 2014.
Veh. Technol., vol. 62, no. 5, pp. 2315–2326, Jun. 2013. [82] D. Bharadia, E. McMilin, and S. Katti, ‘‘Full duplex radios,’’ in Proc.
[57] L. Lei, Z. Zhong, C. Lin, and X. Shen, ‘‘Operator controlled device- ACM SIGCOMM Conf. SIGCOMM (SIGCOMM), 2013, pp. 375–386.
to-device communications in LTE-advanced networks,’’ IEEE Wireless [83] T. S. Rappaport et al., ‘‘Millimeter wave mobile communications for 5G
Commun., vol. 19, no. 3, pp. 96–104, Jun. 2012. cellular: It will work!’’ IEEE Access, vol. 1, pp. 335–349, 2013.
[84] Z. Pi and F. Khan, ‘‘An introduction to millimeter-wave mobile broad-
[58] Q. Ye, M. Al-Shalash, C. Caramanis, and J. G. Andrews, ‘‘Distributed
band systems,’’ IEEE Commun. Mag., vol. 49, no. 6, pp. 101–107,
resource allocation in device-to-device enhanced cellular networks,’’
Jun. 2011.
IEEE Trans. Commun., vol. 63, no. 2, pp. 441–454, Feb. 2015.
[85] F. Gutierrez, S. Agarwal, K. Parrish, and T. S. Rappaport,
[59] D. I. Kim, W. Choi, H. Seo, and B.-H. Kim, ‘‘Partial information relaying
‘‘On-chip integrated antenna structures in CMOS for 60 GHz WPAN
and relaying in 3GPP LTE,’’ in Cooperative Cellular Wireless Networks.
systems,’’ IEEE J. Sel. Areas Commun., vol. 27, no. 8, pp. 1367–1378,
Cambridge, U.K.: Cambridge Univ. Press, 2011.
Oct. 2009.
[60] Y. Yuan, ‘‘LTE-A relay scenarios and evaluation methodology,’’ in [86] T. S. Rappaport, E. Ben-Dor, J. N. Murdock, and Y. Qiao, ‘‘38 GHz
LTE-Advanced Relay Technology and Standardization. Berlin, Germany: and 60 GHz angle-dependent propagation for cellular & peer-to-peer
Springer-Verlag, 2013, pp. 9–38. wireless communications,’’ in Proc. IEEE Int. Conf. Commun., Jun. 2012,
[61] G. Yu, L. Xu, D. Feng, R. Yin, G. Y. Li, and Y. Jiang, ‘‘Joint mode pp. 4568–4573.
selection and resource allocation for device-to-device communications,’’ [87] T. S. Rappaport, J. N. Murdock, and F. Gutierrez, ‘‘State of the art in
IEEE Trans. Commun., vol. 62, no. 11, pp. 3814–3824, Nov. 2014. 60-GHz integrated circuits and systems for wireless communications,’’
[62] L. Wang and H. Wu, ‘‘Fast pairing of device-to-device link underlay Proc. IEEE, vol. 99, no. 8, pp. 1390–1436, Aug. 2011.
for spectrum sharing with cellular users,’’ IEEE Commun. Lett., vol. 18, [88] W. Hong, K.-H. Baek, Y. Lee, Y. Kim, and S.-T. Ko, ‘‘Study and proto-
no. 10, pp. 1803–1806, Oct. 2014. typing of practically large-scale mmWave antenna systems for 5G cellular
[63] Z. Zhou, M. Dong, K. Ota, J. Wu, and T. Sato. (2014). ‘‘Energy efficiency devices,’’ IEEE Commun. Mag., vol. 52, no. 9, pp. 63–69, Sep. 2014.
and spectral efficiency tradeoff in device-to-device (D2D) communica- [89] Estimated Spectrum Bandwidth Requirements for the Future Develop-
tions.’’ [Online]. Available: http://arxiv.org/abs/1407.1556 ment of IMT-2000 and IMT-Advanced, document Rep. M.2078, ITU-R,
[64] N. Naderializadeh and A. S. Avestimehr, ‘‘ITLinQ: A new approach for Geneva, Switzerland, 2006.
spectrum sharing in device-to-device communication systems,’’ IEEE J. [90] T. Baykas et al., ‘‘IEEE 802.15.3c: The first IEEE wireless standard for
Sel. Areas Commun., vol. 32, no. 6, pp. 1139–1151, Jun. 2014. data rates over 1 Gb/s,’’ IEEE Commun. Mag., vol. 49, no. 7, pp. 114–121,
[65] H. Chen, D. Wu, and Y. Cai, ‘‘Coalition formation game for green Jul. 2011.
resource management in D2D communications,’’ IEEE Commun. Lett., [91] W. Roh et al., ‘‘Millimeter-wave beamforming as an enabling technology
vol. 18, no. 8, pp. 1395–1398, Aug. 2014. for 5G cellular communications: Theoretical feasibility and prototype
[66] T. Ihalainen et al., ‘‘Flexible scalable solutions for dense small cell results,’’ IEEE Commun. Mag., vol. 52, no. 2, pp. 106–113, Feb. 2014.
networks,’’ in Proc. WWRF, vol. 30. Oulu, Finland, Apr. 2013. [92] A. R. Khan, M. Othman, S. A. Madani, and S. U. Khan, ‘‘A survey of
[67] J. G. Andrews et al., ‘‘What will 5G be?’’ IEEE J. Sel. Areas Commun., mobile cloud computing application models,’’ IEEE Commun. Surveys
vol. 32, no. 6, pp. 1065–1082, Jun. 2014. Tuts., vol. 16, no. 1, pp. 393–413, First Quarter 2014.
[68] O. Galinina et al., ‘‘Capturing spatial randomness of heterogeneous [93] N. Fernando, S. W. Loke, and W. Rahayu, ‘‘Mobile cloud computing:
cellular/WLAN deployments with dynamic traffic,’’ IEEE J. Sel. Areas A survey,’’ Future Generat. Comput. Syst., vol. 29, no. 1, pp. 84–106,
Commun., vol. 32, no. 6, pp. 1083–1099, Jun. 2014. 2012.
[69] Q. Ye, B. Rong, Y. Chen, M. Al-Shalash, C. Caramanis, and [94] L. Mei, W. K. Chan, and T. H. Tse, ‘‘A tale of clouds: Paradigm com-
J. G. Andrews, ‘‘User association for load balancing in heterogeneous parisons and some thoughts on research issues,’’ in Proc. Asia-Pacific
cellular networks,’’ IEEE Trans. Wireless Commun., vol. 12, no. 6, Services Comput. Conf. (APSCC), Dec. 2008, pp. 464–469.
pp. 2706–2716, Jun. 2013. [95] E. E. Marinelli, ‘‘Hyrax: Cloud computing on mobile devices using
MapReduce,’’ M.S. thesis, School Comput. Sci., Carnegie Mellon Univ.,
[70] S. Corroy, L. Falconetti, and R. Mathar, ‘‘Cell association in small het-
Pittsburgh, PA, USA, 2009.
erogeneous networks: Downlink sum rate and min rate maximization,’’ in
[96] M. Satyanarayanan, P. Bahl, R. Caceres, and N. Davies, ‘‘The case for
Proc. IEEE Wireless Netw. Commun. Conf., Apr. 2012, pp. 888–892.
VM-based cloudlets in mobile computing,’’ IEEE Pervasive Comput.,
[71] S. Singh and J. G. Andrews, ‘‘Joint resource partitioning and offloading vol. 8, no. 4, pp. 14–23, Oct./Dec. 2009.
in heterogeneous cellular networks,’’ IEEE Trans. Wireless Commun.,
[97] P. Mell and T. Grance. (Sep. 2011). The NIST definition of cloud
vol. 13, no. 2, pp. 888–901, Feb. 2014.
computing. NIST. [Online]. Available: http://csrc.nist.gov/publications/
[72] Q. Ye, M. Al-Shalash, C. Caramanis, and J. G. Andrews, ‘‘On/off macro- nistpubs/800-145/SP800-145.pdf, accessed Aug. 2013.
cells and load balancing in heterogeneous cellular networks,’’ in Proc. [98] M. Peng, Y. Li, Z. Zhao, and C. Wang, ‘‘System architecture and key
IEEE GLOBECOM, Atlanta, GA, USA, Dec. 2013, pp. 3814–3819. technologies for 5G heterogeneous cloud radio access networks,’’ IEEE
[73] D. Fooladivanda and C. Rosenberg, ‘‘Joint resource allocation and user Netw., vol. 29, no. 2, pp. 6–14, Mar./Apr. 2015.
association for heterogeneous wireless cellular networks,’’ IEEE Trans. [99] M. Peng, X. Xie, Q. Hu, J. Zhang, and H. V. Poor, ‘‘Contract-based
Commun., vol. 12, no. 1, pp. 248–257, Jan. 2013. interference coordination in heterogeneous cloud radio access networks,’’
[74] A. Bedekar and R. Agrawal, ‘‘Optimal muting and load balancing for IEEE J. Sel. Areas Commun., vol. 33, no. 6, pp. 1140–1153, Jun. 2015.
eICIC,’’ in Proc. Int. Symp. Modeling Optim. Mobile, Ad Hoc, Wireless [100] M. Peng, C. Wang, V. Lau, and H. V. Poor, ‘‘Fronthaul-constrained
Netw. (WiOpt), May 2013, pp. 280–287. cloud radio access networks: Insights and challenges,’’ IEEE Wireless
[75] S. Borst, S. Hanly, and P. Whiting, ‘‘Throughput utility optimization in Commun., vol. 22, no. 2, pp. 152–160, Apr. 2015.
HetNets,’’ in Proc. IEEE Veh. Technol. Conf., Jun. 2013, pp. 1–5. [101] M. Peng, C. I, C. Tan, and C. Huang, ‘‘IEEE ACCESS special section
[76] S. Borst, S. Hanly, and P. Whiting, ‘‘Optimal resource allocation in editorial: Recent advances in cloud radio access networks,’’ IEEE Access,
HetNets,’’ in Proc. IEEE Int. Conf. Commun., Jun. 2013, pp. 5437–5441. vol. 2, pp. 1683–1685, Dec. 2014.
VOLUME 3, 2015 1231

[102] M. Peng, Y. Li, J. Jiang, J. Li, and C. Wang, ‘‘Heterogeneous cloud radio RAKESH KUMAR JHA (S’10–M’13–SM–15)
access networks: A new perspective for enhancing spectral and energy received the B.Tech. degree in electronics and
efficiencies,’’ IEEE Wireless Commun., vol. 21, no. 6, pp. 126–135, communication engineering in Bhopal, India, the
Dec. 2014. M.Tech. degree from NIT Jalandhar, India, and the
[103] P. Rost et al., ‘‘Cloud technologies for flexible 5G radio access networks,’’ Ph.D. degree from NIT Surat, India, in 2013.
IEEE Commun. Mag., vol. 52, no. 5, pp. 68–76, May 2014. He is currently an Assistant Professor with the
[104] H.-H. Cho, C.-F. Lai, T. K. Shih, and H.-C. Chao, ‘‘Integration of SDR School of Electronics and Communication Engi-
and SDN for 5G,’’ IEEE Access, vol. 2, pp. 1196–1204, 2014.
neering, Shri Mata Vaishno Devi University,
[105] V. Tikhvinskiy and G. Bochechka, ‘‘Perspectives and quality of service
Jammu and Kashmir, India. He is carrying out
requirements in 5G networks,’’ J. Telecommun. Inf. Technol., no. 1,
pp. 23–26, 2015. his research on wireless communication, power
[106] Y. Park, 5G Vision and Requirements, 5G Forum, Korea, Feb. 2014. optimizations, wireless security issues, and optical communications.
[107] 5GNOW, (5th Generation Non-Orthogonal Waveforms for Asynchronous He has authored over 30 international journal papers and more than
Signaling) is a European collaborative Research Project Supported by 20 international conference papers. His area of interest is wireless communi-
the European Commission Within FP7 ICT Call 8, 2015. cation, optical fiber communication, computer networks, and security issues.
[108] M. Hatton, The Global M2M Market in 2013. London, U.K.: Machina Dr. Jha’s concept related to router of wireless communication has been
Research White Paper, Jan. 2013 accepted by the International Telecommunication Union (ITU) in 2010.
[109] 4G America’s Summary of Global 5G Initiatives, A One-Time Overview He received the Young Scientist Author Award from ITU in 2010, the APAN
of Global 5G Initiatives as of the First Quarter of 2014, Jun. 2014. Fellowship in 2011 and 2012, and the Student Travel Grant from COMSNET
[110] G. Wunder et al., ‘‘5GNOW: Non-orthogonal, asynchronous waveforms in 2012. He is a Senior Member of the Global ICT Standardization Forum for
for future mobile applications,’’ IEEE Commun. Mag., vol. 52, no. 2, India, Society for Industrial and Applied Mathematics, the International
pp. 97–105, Feb. 2014. Association of Engineers, and the Advance Computing and Communication
[111] G. Wunder et al., ‘‘5GNOW: Challenging the LTE design paradigms Society.
of orthogonality and synchronicity,’’ in Proc. IEEE 77th Veh. Technol.
Conf. (VTC Spring), Jun. 2013, pp. 1–5.
[112] K. Samdanis, A. Ripke, and D. Kutscher, ‘‘Enhancing user QoE in
multi-carrier LTE dense networks via multi-path support,’’ in Proc. Eur.
Wireless, 20th Eur. Wireless Conf., May 2014, pp. 1–5.
[113] H. Tullberg et al., ‘‘Towards the METIS 5G concept: First view on hor-
izontal topics concepts,’’ in Proc. Eur. Conf. Netw. Commun. (EuCNC),
Jun. 2014, pp. 1–5.
[114] J. F. Monserrat et al., ‘‘Rethinking the mobile and wireless network
architecture: The METIS research into 5G,’’ in Proc. Eur. Conf. Netw.
Commun. (EuCNC), Jun. 2014, pp. 1–5.
[115] J.-C. Belfiore, ‘‘Codes for wireless wiretap channels,’’ in Proc. IEEE Inf.
Theory Workshop (ITW), Nov. 2014, pp. 307–308.
[116] M. Olsson, C. Cavdar, P. Frenger, S. Tombaz, D. Sabella, and
R. Jantti, ‘‘5GrEEn: Towards Green 5G mobile networks,’’ in Proc. 9th
Int. Conf. Wireless Mobile Comput., Netw. Commun. (WiMob), Oct. 2013,
pp. 212–216.
AKHIL GUPTA (S’15) received the

B.E. degree in electronics and communication
engineering from Jammu University, Jammu and
Kashmir, India, in 2010, and the M.Tech. degree in
electronics and communication engineering from
the Jaypee University of Information Technology,
Waknaghat, India, in 2013. He is currently pursu-
ing the Ph.D. degree in electronics and communi-
cation engineering with Shri Mata Vaishno Devi
University, Jammu and Kashmir.
He is currently involved in research work on massive MIMO and device-
to-device communication. He is also working on the security issues of next-
generation networks, and OPNET simulation, MATLAB, and NS3 tools
for wireless communication. His research interests include the emerging
technologies of 5G wireless communication network.
Mr. Gupta received the Teaching Assistantship at the Ministry of Human
Resource Development from 2011 to 2013. He is a member of the Interna-
tional Association of Engineers and the Universal Association of Computer
and Electronics Engineers.
1232 VOLUME 3, 2015

2018 ASEE Southeastern Section Conference
A Course Assessment Tool for A Mechanical Engineering Design

Class
Yucheng Liu and Francie Baker
Department of Mechanical Engineering, Mississippi State University
Abstract
A course assessment questionnaire was designed and used as an assessment tool for evaluation of
a design-based and team-oriented mechanical engineering senior design course. The student
feedback to the evaluation questionnaire was collected and analyzed to gain a better understanding
of how well this course met the learning needs of students and addressed the goal of developing
their career skills, as well as its impact on ABET educational objectives, upon which a plan to
improve this senior design course would be formulated. The designed assessment questionnaire is
a good supplement tool to the regular student evaluation form as a means to gather more insightful
and valuable information from students for this specially designed course, through which the
additive value of the unique industry-tied and team-oriented education mode implemented in that
course can be correctly evaluated.
Keywords
Course assessment, questionnaire, mechanical systems design, ABET
1. Introduction
An industry-tied and team-oriented mechanical systems design course had been previously
developed and offered to senior students at Mississippi State University (MSU) (Liu and Dou
2015, Liu 2017). In that course, design projects provided and sponsored by industrial partners,
research centers, and state agencies were assigned to student teams and used as an effective device
to improve student capacity of solving real-world engineering problems and develop their career
skills in a multidisciplinary environment. In order to assess and improve teaching approaches,
learning materials, and education model implemented in that course, an effective student
evaluation instrument was needed. However, the currently used course evaluation questionnaire is
too simple to allow the instructor to collect all useful information from the students to evaluate
that if the specific aims of this course have been achieved and to determine what specific
modifications need to be made to improve the effectiveness of that course. A powerful evaluation
questionnaire which enable the instructor to obtain a comprehensive understanding of advantages
and shortcomings of the renovated design course needs to be constructed. In particular, the
evaluation questionnaire should be able to effectively assess the unique industry-tied and team-
oriented education mode implemented in that course.
In this study, we designed and constructed an evaluation questionnaire to effectively evaluate the
quality of the mechanical systems design course, especially measure the industry-tied and team-
oriented education mode and determine the effectiveness and efficiency of the group design project
in improving students’ skills in problem solving and multidisciplinary team working.
2. Design of A New Assessment Instrument
© American Society for Engineering Education, 2018

An improved assessment instrument was designed, which includes four sessions. The first session
is about the achievement of the course goals, including 10 closed-format questions and two open
format questions. In the session the students are asked to evaluate their growths on following skills,
organization, teamwork, communication, leadership, management, and problem solving as well as
the knowledge development on topics of solid mechanics, fluid mechanics, manufacturing, and
CAD/CAE. The second session uses four open format questions to obtain more information about
the nature of the project that each student team worked on. The acquired information can provide
additional details related to their motivations, knowledge gains, and skill performances of this
course. The third session with its 10 closed format questions on a 6-point Likert scale aims at
finding out student perception of industry importance on the skills and knowledge covered in this
course. The collected results can be used to assess their possible motivation for mastering different
skills and topics associated with employability and industry standards.
A fourth session was added to discuss the impact of the renovated course on ABET educational
goals. Student perceptions of course performance for ABET criteria would enhance the assessment
of the course by expanding the prior formative and summative assessments to detail specific ABET
measures. Thus, the ABET assessment part of this survey was designed to examine the rationale
and motivation for learning gains, student perceptions of preparation and awareness of the ABET
criteria inclusion within this design course. 11 closed format questions on a 5-point Liker scale
were added to assess the student outcomes according to the 11 ABET (a-k) educational objectives.
Finally, each student is asked to write down their additional comments on this course.
This survey was conducted in the fall semester of 2016 and 44 senior students enrolled in that class
participated in the survey. All the questions, student evaluation results for the closed format
questions are listed in the next section.
3. Assessment Results
3.1 Course Goals

In the Before Course column rank your level for each skill and knowledge of each topic before
the course and on the After Course column rank how you think you are now that you have
completed the course. The number 1 represents the lowest ranking and 5 represents the highest
ranking for each skill and topic. The numbers in the cell represent the number of student who chose
that point. For example the “13” in the row of “Organization” means that 13 students selected “3”
for that question.
Table 1. Assessment and results on course goals

Before Course Over After Course Over Δ
August 2016 all November 2016 all
Skills 1 2 3 4 5 5 1 2 3 4 5 5
Organization 1 13 21 9 3.86 6 25 13 4.16 0.3
Teamwork 1 6 25 12 4.16 3 19 22 4.43 0.27
Communication 1 1 12 18 12 3.89 3 22 19 4.36 0.47
Leadership 1 3 15 14 11 3.70 9 20 15 4.14 0.44

Project
1 3 14 18 8 3.66 1 4 20 19 4.30 0.64
Management
Problem Solving 1 10 24 9 3.93 3 21 20 4.39 0.46
Topics 1 2 3 4 5 1 2 3 4 5
Solid Mechanics 3 19 19 3 3.50 1 4 28 11 4.11 0.61
Fluid Mechanics 10 22 6 6 3.18 7 21 11 5 3.32 0.14
Manufacturing 4 17 18 5 3.55 7 28 9 4.05 0.5
CAD/CAE 3 17 17 7 3.64 1 8 25 10 4.00 0.36
 Which, if any, skills or topics did you note an increase? What do you think lead to your increase
in that area?
 Which, if any, skills or topics did you note a decrease? What do you think lead to your decrease
in that area?
3.2 Industry Importance

Use the following table to rank the skills and topics in order of importance for industry according
to your experiences and perceptions. The number 1 is the most important and number 6 is the least
important. Use each number 1, 2, 3, 4, 5, 6 only once for each set. The numbers in parentheses
represent the number of students that selected the scale before the parentheses. For example, 1(6)
means that there were six student choosing “1” for a particular question.
Table 2. Assessment and results on industry importance

Skills Skill Ranking (1 – 6) Overall scores
Organization 1(6), 2(7), 3(7), 4(6), 5(5), 6(13) 3.82/6
Teamwork 1(12), 2(8), 3(6), 4(8), 5(2), 6(8) 3.09/6
Communication 1(9), 2(13), 3(6), 4(2), 5(4), 6(10) 3.20/6
Leadership 1(4), 2(4), 3(5), 4(9), 5(11), 6(11) 4.23/6
Project Management 1(6), 2(10), 3(5), 4(6), 5(11), 6(6) 3.55/6
Problem Solving 1(6), 2(7), 3(9), 4(6), 5(7), 6(9) 3.64/6
Topics Topic Ranking (1 – 6)
Solid Mechanics 1(9), 2(13), 3(8), 4(5), 5(6), 6(3) 2.89/6
Fluid Mechanics 1(4), 2(3), 3(9), 4(5), 5(12), 6(11) 4.16/6
Manufacturing 1(13), 2(7), 3(6), 4(10), 5(4), 6(4) 2.93/6
CAD/CAE 1(4), 2(3), 3(6), 4(6), 5(6), 6(19) 4.45/6
3.3 Student Outcomes

Please mark the cell that best describes your response to the following statements about how this
course met each of the following ABET criteria.
Table 3. Assessment results of the course’s impact on ABET outcomes

Strongly Strongly Overall

This course has provided me with Agree Agree Neutral Disagree Disagree scores
A. An ability to apply knowledge of
mathematics, science, and
21 18 4 1 4.30
engineering
B. An ability to design and conduct
experiments, as well as to analyze
20 19 4 1 4.30
and interpret data
C. An ability to design a system,
component, or process to meet
desired needs within realistic
constraints such as economic,
environmental, social, political,
ethical, health and safety, 22 18 1 2 1 4.32
manufacturability, and
sustainability
D. An ability to function on 13 22 6 2 1 4
multidisciplinary teams
E. An ability to identify, formulate, 26 16 1 1 4.48
and solve engineering problems
F. An understanding of professional 20 20 3 1 4.32
and ethical responsibility
G. An ability to communicate 19 20 4 1 4.27
effectively
H. The broad education necessary to
understand the impact of
engineering solutions in a global,
economic, environmental, and 17 14 10 2 1 4
societal context
I. A recognition of the need for, and
an ability to engage in life-long
24 14 4 1 1 4.34
learning
J. A knowledge of contemporary 13 16 12 2 1 3.82
issues
K. An ability to use the techniques,
skills, and modern engineering
tools necessary for engineering 25 17 1 1 4.45
practice
4. Analyses and Discussions
4.1 Effectiveness on Skill Growth and Knowledge Development

Based on Table 1 it can be found that all the listed skills and knowledge of the participating
students were improved through this course. The students identified that they had achieved biggest
improvements on the project management skill and their solid mechanics knowledge. This can be
attributed to the implementation of the team-based design projects and the fact that this systems
design course focused more on the solid mechanics, including mechanical components design and
analysis. The least improvements they made in this class include the fluid mechanics knowledge
and the teamwork skill. This is because that the course syllabus did not cover many fluid mechanics
topics and few design projects include the design of fluid systems. In addition, the results also
suggest that the organization of the group design projects needs to be improved to better develop
the students’ teamwork skill.
Table 2 reveals the student perceptions of the industry importance of the listed skills and topics.
From that table it can be seen that the students highly valued the influence of knowledge base of
solid mechanics and manufacturing on their careers and ranked the leadership skill and CAD/CAE
experience with the least industry importance. These results will be brought to our industry
partners to find out the most wanted employability skills, according to which the course structure
will be further modified.
Table 3 links the student evaluation results on the course goals (Table 1) with the industry
importance (Table 2) to measure student satisfaction in this design course, in which the topics and
skills are reordered according to their ranks in the industry importance survey. From that table it
can be seen that most skills and knowledge of topics that the students considered important were
evidently improved through this course. The only exception is the teamwork skill. The students
considered that skill the third most important in industry but the growth of that skill only ranked
9th after completing the course. It is once again suggested that further measures need to be taken
to fully develop the students’ teamwork skill.
Table 4. Ranking of student perceptions on industry importance of skills and knowledge and their
corresponding growths
Ranking Skills/Topics Growth
1 Solid mechanics (2.89) 0.61(2nd)
2 Manufacturing (2.93) 0.5 (3rd)
3 Teamwork (3.09) 0.27 (9th)
4 Communication (3.20) 0.47 (4th)
5 Management (3.55) 0.64 (1st)
6 Problem solving (3.64) 0.46 (5th)
7 Organization (3.82) 0.3 (8th)
8 Fluid mechanics (4.16) 0.14 (10th)
9 Leadership (4.23) 0.44 (6th)
10 CAD/CAE (4.45) 0.36 (7th)
5.2 Impact on ABET
This course has a broad impact on ABET educational objectives, which has also been verified
through this survey. As can be seen from Table 4, the students agreed that this design course helped
them to achieve 10 out of the 11 ABET outcomes (with an overall score above 4 out of 5). The
only ABET outcome that received a score below to 4 is the criteria j): a knowledge of

contemporary issues. This fact that the students did not feel much improvement on their knowledge
of contemporary issues suggests the teacher to introduce more contemporary issues in engineering
and design into the class to fill this theory-practice gap.
The overall final average for the fall 2016 class was 79.72%, with 27 out of 44 students scoring at
or above 75% for the semester average. The overall performance breakdown for this class
consisted of 23% A, 23% B, 43% C, and 11% D grades. Homework problems were used to
evaluate ABET goals (a) and (e), and 42 out of 44 students scored at or above 75% for both goals.
The team design project was used to evaluate ABET goals (b-d), and (g). Students’ performance
showed that all the 44 students passed the performance criteria (score ≥ 75%) for goals (b), (c),
and (g), and 43 students passed the performance criteria for (d). Final exam was used to evaluate
the ABET goal (k) and we have 30 students scored at or above 75% in the final exam. A quiz on
engineering ethics conducted in classroom was used to evaluate the ABET goal (f) and all the
students passed it with a score ≥ 75%. Finally, the presented questionnaire was used to evaluate
ABET goals (h) and (j), we have 41 students passed the criteria for (h) and all 44 students passed
that for (j). Results from the direct measures of the student performance agreed very well with the
results obtained from this questionnaire (Tables 1-4) and once again confirmed that most ABET
goals as well as the course goals were achieved through the renovation of this course.
6. Conclusions
A course assessment questionnaire was designed and used for assessing teaching and course
quality of the Mechanical Systems Design course at MSU, in which a unique industry-tied and
team-oriented education mode was implemented. The questionnaire comprehensively investigate
the student perceptions on achievement of course goals, student outcomes, and impact on the
ABET outcomes, based on which the effectiveness and efficiency of the teaching and education
approach and the satisfaction level of the students on this course can be deduced. The student
feedback were collected and studied. The results showed that most course goals were achieved and
overall the students were satisfied with the renovated course and recognized the effectiveness of
the team design projects in growing their knowledge bases and developing their employability
skills. Adoption of this new survey design also advanced the course assessment to ascertain
motivation, experience, and understanding of ABET outcomes. From the results, it is also
suggested that the organization of the design projects should be improved to enable the students to
function more effectively in their teams. More contemporary issues in engineering design should
also be brought into the course syllabus.
References
1 Liu, Y.-C., and Y.-Q. Dou, 2015, “Design of an industry-tied and team-oriented course for mechanical
engineering seniors”, Proceedings of ASEE SE Section Annual Conference, University of Florida,
Gainesville, USA.
2 Liu, Y.-C., 2017, “Renovation of a Mechanical Engineering Senior Design Class to an Industry-Tied and
Team-Oriented Course”, Accepted by European Journal of Engineering Education, In Press.
Dr. Yucheng Liu is Associate Professor and Graduate Coordinator in the Department of
Mechanical Engineering at MSU.
Ms. Francie Baker is a PhD student in the Department of Mechanical Engineering at MSU.

Effects of Data Center Vibration on Compute System Performance
Julian Turner
CTO - Q Associates
Julian.Turner@QAssociates.com
Abstract were tested. In general, it was shown that 3.5”

consumer grade hard drives were moderately to
From September 21, 2009 until October 2nd, 2009, significantly impacted by vibration with with 2.5”
Q Associates performed a series of tests and enterprise grade typically being much less susceptible.
benchmarks to determine the effect of data center and Since 2005, hard drives have become faster and much
system vibration on the I/O performance of an end-to- higher capacity [8]. In addition, the combination of
end compute environment. These tests revealed that hard drives into large arrays of tens or hundreds of
ambient vibration inherent in a world-class, raised TeraBytes has become much more common [9] [10].
floor data center caused performance degradation of The first demonstration of the impact of vibration on
up to 246% for random reads and up to 88% one of these high capacity arrays was done by Brendan
degradation for random writes for an enterprise class Gregg of Sun Microsystems in a YouTube video
storage system. This loss of performance due to commonly referred to as the “Yell Test” [11]. In this
environmental vibration results in a commensurate video, he demonstrated the direct impact on
increase in energy usage for equivalent work to be performance caused by yelling at a running storage
performed. A prototype anti-vibration rack was tested system. By using the real-time graphical analytic tool
within the same environment and shown to built into the array, he visually showed the reduction in
significantly reduce or eliminate the detrimental I/O during the period in which he yelled. It was this
vibration effects resulting in significantly increased video that led to questions about the potential impact of
performance. A "latent performance effect" was also typical data center vibration on storage system
discovered and analyzed associated with the testing. performance.
This effect is a potential source of traditional
benchmark error and would likely not be detected by 2. Compute Environment and Testing
normal benchmark procedures and tests. The study of Procedures
this effect further collaborates the impact that
vibration has on overall system performance. The intent of the testing was to determine the impact
of “typical” data center vibration on the performance of
1. Introduction a current generation storage array. A secondary goal
was to determine if any performance degradation found
The reduction of vibration within the IT could be reduced or eliminated within the data center
environment has long been considered an important environment using a specifically designed anti-
attribute both at the macro-data center level [1] and the vibration rack by Green Platform Corporation [12].
micro-component level [2]. At the macro data center The test plan called for running a number of freely
level, the reduction of vibration primarily centers on available benchmarks within two well defined
personal comfort and safety [3] and impact on the environments: a specially designed sound room having
physical components such as cable connectors or virtually no external vibration as well as a production
system attachment. On the micro level, there are raised-floor data center. Within the data center, tests
countless methods for reducing vibration associated were to be performed in an industry standard metal
with individual hard drives [4]. These methods and rack enclosure as well as a specialized rack that
commercially available products are primarily aimed at incorporated full-frequency anti-vibration modules.
reducing disc drive failure or reducing perceived The storage array selected for the tests was a Sun
audible noise [5] [6]. None of the researched products 7110 array with 16 x 300GB, 10k 2.5” Seagate
imply increased performance as a primary benefit. enterprise class SAS hard drives. The Sun array was
The first relevant reference on the potential impact selected due to its manageable size and the inclusion of
of vibration on hard drive performance was an IEEE the real-time GUI D-Trace Analytics software that
paper by Ruwart and Lu in 2005 [7]. In this paper, a comes with the system [13]. The benchmark source
number of consumer and enterprise-grade hard drives system was a Sun X4440 16 core system with 64GB of
RAM. These two systems were connected using four pedestrian sources of vibration (traffic, etc.) within ¼
Gigabit Ethernet connections running NFSV4 via a mile of the location. Equipment was placed on a rubber
non-blocking switch. With this configuration, network pad directly on a concrete foundation.
performance of a sustained 280 MBytes/sec (2.2 The second test environment was a Tier 1 data
Gbits/sec) could be maintained with temporary spikes center located in Houston, Texas. This Tier 1 data
in excess of 320 MBytes/sec (2.6 Gbits/sec) recorded. center is a world-class raised floor data center
Additional optimization was not performed since it was specializing in hosted systems. This environment is
felt that this bandwidth was more than sufficient for representative of most large enterprise data centers and
testing purposes. included vibration sources from other compute
For all tests, the Sun Analytics tool was used to systems, air handling/AC equipment, and UPS units.
capture 12 different system characteristics including Ambient noise was measured at approximately 82dB at
disk I/O broken down by disk, disk I/O broken down a location 2 feet from the rack being tested. Vibration
by latency, and network interface bytes per second was measured at several locations both on servers and
broken down by interface. Information on CPU racks through-out the data center and typically were .3
utilization, memory utilization, and cache misses was m/s2 RMS (.03 GRMS) or below for any single
also recorded. vibration axis. The Tier 1 data center furnished a
Two configurations of the 7110 were tested. For standard CPI (APC style) metal stand-alone rack with
configuration #1, the 7110 utilized a mirrored file 40 Amps of power. The general layout for the
system across all 16 drives. Two drives were used for environment is shown in Figure 1.
OS and the remaining 14 drives for data. For this
phase, all drives were originally blank and data was
deleted after each test. For configuration #2, 10 drives
were taken off-line. The remaining 6 drives were in a
mirrored configuration with 2 drives for OS and the
remaining 4 for data. In this configuration, there was a
total of 570GB available for data storage. The entire
570GB was filled with files ranging in size from 1k to
75GB. Approximately 400 files were semi-randomly
deleted to give back 37 GB of storage for use in Sun X7110
testing. The smallest file size deleted was 50k with the GigE Switch
largest being 1GB. Approximately ten 1GB files were (Shelf)
deleted, thirty 250MB or 500MB files, and the
remainder were 100MB or less.
The use of configuration #2 was viewed as justified
as a means to better approximate “real world” (Slide
Sun X4440
conditions where there would be a much closer match Rail)
of Host devices to Target hard drives within a typical
NAS or HPC environment. In addition, it was
concluded that very seldom would their be a “blank (Shelf)
disk” condition in a “real world” environment so PDU
adding data to the drives in a random manner would be
a better approximation of actual working conditions. Metal Anti-Vibration
The expectation was also that a smaller number of hard 42RU Rack 20RU Rack
drives would make it easier to see performance
anomalies via the GUI if the anomalies existed. Figure #1: Representative Physical Layout
3. Physical Environments 4. Test Environment #1

The first test environment was a specially Since the time available at the Tier 1 data center
constructed sound room in Bastrop, Texas. Ambient was limited and it was critical to have a non-vibration
noise was less than 40dB and there were no baseline, initial configuration, performance testing, and
commercial sources of vibration within 2 miles or baseline benchmark testing were performed in a
controlled environment. This facility allowed all SingleStreamRead1m, SingleStreamReadDirect1m,
outside vibration and sound influences to be greatly SingleStreamWrite1m, and
reduced/eliminated. The ambient sound pressure with SingleStreamWriteDirect1m workloads. Due to target
no systems on was measured at 38.2dB. At a distance disk space limitations, multi-stream reads and writes
of 1 foot from the front of the Sun 7110, the pressure were not performed. Prior testing had shown that multi-
was 84.7dB at initial system start-up with all fans stream reads and writes were roughly equivalent to
working at maximum, and 68.5dB at typical system single stream reads and writes for this system so
idle. Temperature in the room ranged from 80-85 removal of them from the script was viewed as
degrees through-out the testing. acceptable.
The first step was to perform an initial baseline and
determine the optimal configuration for testing. The 5. Test Environment #2
Sun X4440 was configured as an NFS4 Client with all
4 GigE ports trunked together to form a single virtual After completion of the initial baseline testing at
connection. The Sun 7110 was configured as a NFS4 Environment #1, all systems were moves to the Tier 1
server with each port assigned a separate IP address. data center in Houston. Due to a shipping error, only
With four separate mount points on the X4440 client, one slide-rail kit was available for attaching either the
separate 10GigaByte files could be sent to each mount X4440 or 7110 to the rack. This added an additional
point. The Sun Analytics feature of the 7110 was then variable to the testing but ultimately worked out well
used to view the NFS throughput coming into each since it allowed the top system to be moved from one
7110 port as well as CPU utilization and raw disk rack to the other without being turned off.
throughput and utilization. For the first round of testing, the slide-rail kit was
With sustained average throughput for serial transfer attached to the 7110 and mounted in the metal CPI
of over 280Mbytes/sec via the NFSV4 ports, the CPU rack. The X4440 was placed on top of the 7110 with a
utilization of the 7110 remained in the 75%-85% range rubber no-slip pad between them. The D-Link switch
and never went above 85%. With the data being spread was mounted directly to the metal rack and was not
over 16 hard drives, actual hard drive throughput for moved for the duration of the testing.
any individual hard drive was extremely low (<5% The amount of measured vibration in the Tier 1 data
utilization on average) and very spiky in nature due to center was actually lower than expected. Even though
RAM caching on the 7110. Maximum disk drive you could physically feel considerable vibration of the
latency was typically less than 120 milliseconds and metal racks, the actual measured vibration was
less than 10 milliseconds was typical. Due to the low typically .2 m/s2 RMS (.02 GRMS) or less in all three
disk utilization and the initial focus on serial versus axis. Since the Reed vibration meter only measured up
random I/O, the storage system was then reconfigured to 1kHz, the belief is that there is an additional
as explained in Section 2. vibration component above 1kHz that was not
IOZone had originally been selected as the primary captured. During power-up, both the 7110 and X4400
benchmarking tool. However, after several tests it was were measured at 1.6 m/s2 RMS (.16 GRMS) along a
evident via Analytics that IOZone was not sufficiently single vertical axis. At steady state, both servers were
stressing the hard drives. As such, Solaris FileBench less than .4 m/s2 RMS (.04 GRMS) depending on
was then selected since it had both Micro and Macro where the measurement was taken.
benchmark capabilities and could be easily modified if Testing with the anti-vibration rack was done in one
required. of two manners. The 7110 was either sat on a
Prior to moving the system to Tier 1 data center, plexiglass shelf that was supported by an anti-vibration
the following FileBench scripts were run: module or was supported by a composite slide-rail that
slid on an anti-vibration module. Since the anti-
FileMacro: Consists of File Server, Varmail, Web vibration module was roughly equivalent in both cases,
Proxy, and Web Server workloads. Large database it was assumed that the anti-vibration characteristics
OLTP workloads were also available but were not run would be roughly equivalent. Testing indicated that the
due to an error in the sub-script causing system lock- properties were not exactly the same but close enough
up. to demonstrate the objectives of these tests. Since the
rack was a prototype, testing the exact performance
FileIO: Consists of RandomRead2k, difference between the two methods of placement was
RandomRead8k, RandomRead1m, RandomWrite2k, not warranted.
RandomWrite8k, RandomWrite1m,
C o n tro l M etal M etal M etal AV R AV R AV R M etal M etal M etal AV R % D iff
C onc rete 7110 in 7110 in 7110 in Sitting on C om po s ite C om pos ite 7110 on 7 110 on 7110 on Sitting on M etal vs
F l oor s li de-ra il s l ide-rai l s li de-rai l top s hel f s li de-rail s lide-rai l X4440 X4440 X 4440 top s helf AV R
W ork load
rando mread1m 440 130 131 129 627 522 514 187 156 163 405 246%
rando mread2k 1438 962 793 782 1664 1488 1526 1138 1040 1114 1391 56%
rando mread8k 1399 960 782 782 1712 1513 1518 1141 1037 1070 1461 61%
rando mwrite1m 20 22 20 17 26 25 26 19 19 20 28 35%
rando mwrite2k 189 96 102 100 193 189 158 186 115 101 171 52%
rando mwrite8k 433 156 115 139 269 242 317 202 118 152 281 89%
s in gles treamread1m 64 62 60 59 71 71 68 62 66 65 74 14%
s in gles treamreadd irect1m 58 51 51 51 62 61 55 56 58 55 64 13%
s in gles treamwrite1m 96 92 95 102 98 100 103 97 101 93 101 4%
s in gles treamwritedirect1m 16 19 19 16 22 21 21 20 19 20 22 14%
F ig u re #2: Th ro u gh p u t Op eratio n s p er S econ d
FileBench FileIO Test surprising. Upon observing the much higher impact that
The FileBench FileIO test is a publicly available vibration was having on the random I/O, test scripts
multi-part benchmark consisting of several sub micro- were modified to remove the sequential portion. This
benchmarks. Ten separate runs were performed in significantly reduced the amount of time required for
Environment #2 with the results shown in Figure 2 each test run and allowed other trends to become more
above. visible.
For this data-set, single stream read performance
was better in the anti-vibration rack by approximately Latent Performance Effect
13%. Single stream write was better in the anti- Prior to analyzing the data, one of the most puzzling
vibration rack by approximately 4% which corresponds aspects of testing was that the performance of the “next
and collaborates the testing done in other test cases. run” was never what was predicted. A trending was
However, there was a shocking 246% performance noticed so test scripts were optimized to look
difference for the RandomRead1m test and a 56% and specifically at this trending. For these tests, the 7110
61% difference for the Random Read 2k and 8k tests was not turned off or rebooted between moves.
respectively. The performance difference of Random As shown in the numbers within Figure 3, there
Writes was similarly compelling with 52% difference exists a “latent performance effect” by which once a
for the 2k case, 88% difference for the 8k case, and system is moved, the performance remains roughly
34% difference for 1m test. While the number of runs equivalent to the previous system for some amount of
comprising the data-set in Figure 2 was relatively small time. Rebooting the system seems to erase the “bad
by statistical standards, this data is loosely correlated performance effect” and thus allow the system to work
and confirmed by over 500 runs performed during the 2 at optimal speed for a short period of time.
weeks of testing. Conservatively, an average
performance increase of Sequential Write = 2%+,
Sequential Read = 7%+, Random Read = 80%+, and
Random Write = 50%+ can be easily supported and
defended by the full test data.
Since the original focus had been on sequential data
transfer, the obtained random I/O results were quite
AVR Metal Metal Metal Metal AVR AVR
Sitting on Sitting on Sitting on Sitting on Sitting on Sitting on Sitting on
top shelf X4440 X4440 X4440 X4440 top shelf top shelf
(moved without turning off immediately after previous test)
randomread1m 504 344 203 160 171 166 575
randomread2k 1538 1350 1384 1057 1115 1053 1684
randomread8k 1524 1409 1164 1057 1088 1065 1626
randomwrite1m 26 26 23 22 20 21 28
randomwrite2k 182 165 141 126 126 132 183
randomwrite8k 273 264 138 133 119 119 255
Figure #3: Modified FileIO Operations per Second
RandomWrite 2K Blocks
(AVR runs in black; Metal rack runs in light gray)
250
200
150
Ops /s ec
100
50
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Consecutive Run Number
Figure #4: Consecutive System Runs

The graph in Figure 5 below is a good example of
Looking at this graphically over a long series of runs that oscillating slope phenomenon. Run #5 was a
as shown in Figure 4, it is easy to see the trend. The steady-state run by which the disk latency was at
graph above shows 19 consecutive runs for Random equilibrium for being in the vibrating metal rack. When
Write with 2k block size. In this figure, the runs on the the 7110 was then moved to the anti-vibration rack, the
anti-vibration rack are shown in black (runs 1 and 6- system kept the same reduced performance but
11) and the runs using the standard metal rack are experienced much lower disk I/O latency. Thus, for
shown in light gray. Run #7 the system was allowed to go “full throttle” but
A review of the raw data in the Analytics tool seems again experienced latency and this was throttled back
to indicate that the system uses some pre-set algorithm for Run #8. Had the system been allowed to stay in the
to determine the appropriate system speed relative to anti-vibration rack, it would have likely reached
experienced disk latency. A fast run that has significant equilibrium close to the performance of Run #12. Run
disk I/O latency will be followed by a slower run with #12 had very good performance but saw significant
much lower latency. If the latency is below some latency above 200ms. As such, the disk I/O was
threshold, the disk I/O speed is increased by the system significantly throttled back for run #13 where it
by some factor dependent on the latency of the reached equilibrium at a much lower performance than
previous run. This explains why there is not a gradual on the anti-vibration rack.
up or down slope but a very ragged, oscillating slope as A key point is that if this latency effect is not taken
the system finds equilibrium. into account, it could be a source of significant error in
standard benchmarks. Worse yet, if someone did not
know to be looking for the effect, it likely would not be
caught by standard benchmark test cases.
Performance Trend Oscillation for Random Read of 1M Blocks

(AVR runs in black; Metal rack runs in light gray)
700
600
500
400
OPs/sec
300
200
100
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Consecutive Run Number
Figure 5: Consecutive Runs with Oscillating Performance

6. Dissipation versus Transmittal performance. Within the FileBench FileMacro
benchmark, there is a steady-state performance of 300-
One observation is that performance was 350 disk I/O ops/sec with regular bursts of 700-725
consistently worse in the situation where the X4440 ops/sec. During this state, CPU usage is less than 3%,
was sitting on the 7110 versus the other way around. NFSV4 latency is less than 19 milliseconds, and
One possible explanation is that the X4440 transferred memory cache usage is less than 2% for both high
vibration directly to the 7110 when it was sitting on top performance and low performance runs. The
of it but transmitted some of the this vibration to the differentiator between high and low performance runs
rails when it was in the slide-rail kit and the 7110 was is a small percentage of disk I/O latency above 200
sitting on top. In a full rack, this could potentially mean milliseconds. For a high performance run,
significant vibration being transferred to the rails and approximately 11% of I/O will have a value between
thus other systems within the same rack. Another 100-190 milliseconds. As long as the maximum disk
possible explanation is that the rack transfers vibration I/O remains below 200 milliseconds, no stoppage in
directly to the 7110 when it is attached directly via the network I/O is seen. The low performance runs look
slide-rails. fairly similar to the high performance runs with
To test the theory on the effect of server placement, approximately 12%-15% of the I/O over 100
additional tests were conducted in the control milliseconds (similar to high performance case) and
environment. The FileIO benchmark was run a 4%-6% being over 200 milliseconds (average of
minimum of 4 times for three separate cases: approximately 234 milliseconds). For each case when a
disk I/O latency greater than 200 milliseconds is seen,
1) 7110 sitting on an unpowered server on a the network I/O goes to zero for a brief period of time.
concrete floor with X4440 sitting next to it Therefore, an average increase of disk I/O latency of
2) 7110 sitting on an unpowered server on a less than 5% causes an overall system performance
concrete floor with X4440 sitting on top of the decrease of 40%+.
7110
3) 7110 sitting on an power server on a concrete 8. So Why is this Relevant to “Me”?
floor with X4440 sitting on top of the 7110
The amount of system compute time required to
A small but measurable performance reduction perform a set amount of work is relevant whether you
could be seen in both transitions. A roughly 2%-3% are a system administrator, application developer, or a
performance reduction could be seen when the X4440 data center designer. While the ability to finish a “job”
was placed on the 71110 and an additional 1%-2% was by x% faster is an obvious but subjective benefit of
seen when the bottom server was turned on. While vibration reduction, the impact it has on energy
there were some fluctuations from run to run, the utilization can be a double-edged sword. With Power
average of all runs was roughly equivalent to the Utilization Efficiency (PUE) ranging anywhere from
performance of the 7110 in the anti-vibration rack 1.19 claimed by Google [14] to 2.5 or greater for
while at the Tier 1 data center. This is significant since legacy raised floor data centers [15], every watt saved
it provides further proof that it is the data center or generated has a multiplied effect.
environment and the interaction of vibration via the In many Corporate enterprise environments where
metal rack that causes the most significant performance set jobs are run each day, decreasing the amount of
degradation. This also shows that a “simulated” time it takes for a job to run effectively reduces the
vibration environment using other servers as vibration energy usage. This amount is effectively: time saved *
source is not sufficient for adequately testing difference in idle system energy versus running system
performance degradation associated with external usage * PUE. An additional computational component
vibration. associated with system fan speed may also be required
depending on the system being analyzed. Based on a
7. Analytics Assessment theoretical 15% performance increase, the overall
energy savings for a job having a 1 hour run on a 1kW
The use of the Sun D-trace based Analytics tool system in a 2.2 PUE data center would be
provided essential clues as to the detrimental impact approximately 138 Watts or roughly 7% energy savings
that vibration was causing. Analysis of the Analytics [16].
output shows that only a small difference in disk Within a typical HPC environment, system
latency can cause a huge difference in overall system utilization run-time can be close to 100%. In this case,
as soon as one job finishes the next job runs. In this significantly more seek head movement in the random
type of environment, energy utilization may actually read and write cases, vibration takes a greater toll on
increase as system performance improves since the performance as the seek head tries to lock onto the data
workload of the server CPU goes up since there is bits. Whether this is true or if there a totally unrelated
much less wait-time on storage. While the total power cause for the performance degradation needs to be
increase may be marginal on a per-server basis, it could determined.
be quite significant in an HPC environment with An analysis of the cause of the latent performance
thousands of servers. effect also needs to be performed. Discussions with
For those that perform benchmarking for a living, engineers at both hard drive and complete systems
the discovered results could be equally compelling for companies have not resolved the issue as to whether the
an entirely different reason. In many companies, effect is due to settings within the hard drive BIOS or
benchmarking and performance testing is done on the System OS. Understanding what is leading to the
development or test systems that are physically latent performance effect would be significant in
separated from production systems. Thus, performance building benchmarks that reduce or mitigate the effect.
numbers obtained in a “quite” test lab with only the An understanding of the frequency and amplitude
tested system being run might be quite different than required to negatively affect system performance would
what would be obtained if the tests were conducted in a be of great benefit to the community. It is obvious that
“noisy” raised floor data center. Even if all testing is an “impact” or “shock” vibration such as dropping or
done in the data center, the time of day might influence yelling at a server would have a negative effect on
the test results because of more or less vibration due to performance. However, this effect would be temporary
natural usage patterns of IT systems. and is not what would normally be experienced in a
An additional concern to a benchmarking data center environment. The ability to feel the
professional would be the latent performance effect. vibration was also observed to be a poor indicator of
Testing seemed to indicate that the characteristics of the effect it had on the system performance.
the previous run had an effect on the subsequent run. Finally, spread the word that vibration “might” be a
Therefore, a test run that had a tendency to cause high performance issue within any given environment. The
latency would cause the subsequent test to have slower more people working on this, the faster a resolution
performance than if the first test had a lower latency will be found. There are likely several ways to resolve
profile. the performance degradation “symptom” and even
more to resolve the vibration source “problem”. Which
of these is more cost effective will have to be
9. Community Next Steps researched and analyzed.
Given limited time, budget, and resources, only the 10. Summary
tip of the iceberg has been discovered in relation to the
effect of vibration on system performance. The testing The test results clearly demonstrate that random
performed answered many questions but created even reads and writes are significantly impacted by vibration
more. What can be said with certainty is that vibration in the data center environment. Performance
in a typical data center does impact performance. improvements for random reads ranged from 56% to
Whether that impact is relevant depends on the 246% while improvements for random writes ranged
environment as well as the profile of the application. from 34% to 88% for a defined set of industry
So what are the next steps? First, the community benchmarks. Streaming sequential reads and writes had
needs to perform independent tests to validate the a much smaller performance improvement but were
results in their own environment. This may be as still measurable and potentially relevant to some
simple as running random read/write benchmark 5 environments.
times on a standalone server in an office and then By reviewing the raw data within the Analytics tool,
running the benchmark another 5 times with the server it is clear that it only takes a small increase in disk
mounted in a rack in a production data center. latency caused by vibration to have a cascading effect
Second, the true root cause for the performance that disproportionately effects overall system
degradation needs to be analyzed and understood. It is performance. Additional testing performed in the
theorized that the difference in performance between control environment further indicates that vibration has
the sequential and random cases are due to increased a cumulative effect and the data center environment
head movement within the hard drive. Since there is
itself (CRAC units, air movement, populated metal [5] www.jab-tech.com, Vibration Dampening Hard
racks, etc.) is a primary contributor. Drive Screws
A final finding that was not part of the initial goals
was the discovery of the “latent performance effect” [6] www.silentpcreview.com, NoiseMagic's No-Vibes
associated with vibration and disk I/O latency. This Hard Drive Suspension Kit, 2005
effect seems to be a performance or reliability
characteristic of the system and is clearly visible when [7] T.M. Ruwart, Y. Lu, Performance Impact of
viewed across a long series of runs. More importantly, External Vibration on Consumer-Grade and
this effect shows that point-in-time benchmarks can be Enterprise-Class Disk Drives, IEEE, 2005
misleading or blatantly wrong depending on the
amount of time the system has been active and [8] en.wikipedia.org/wiki/Hard_disk_drive
potentially the I/O characteristics of runs performed
prior to the benchmark. [9] www.sun.com/storage, 7410 Unified Storage Tech
All of these support the very simple premise that Spec
data center vibration affects system performance and
should be actively mitigated as part of any next-gen [10] www.datadirectnetworks.com, SFA 10000
data center or system deployment. In the case of Overview
random reads and writes, this performance difference is
significant and mitigation of vibration would have a [11] blogs.sun.com/brendan, Shouting in the Data
significant positive effect on total energy usage in a Center, Dec 31, 2008
data center.
[12] www.greenplatformcorp.com, AVR-1000
11. Author Biography
[13] www.sun.com, Sun Storage 7000 Unified Storage
Julian Turner is currently serving as the Chief Systems
Technology Officer for Q Associates out of Houston,
Texas. Mr. Turner's past employment consisted of five [14] www.google.com, Data Center Efficiency
years as the Chief Architect of the Southern US for Sun Measurements, October 2009
Microsystems, three years as an independent multi-
national consultant, and nine years with Andersen [15] M Szalkus, What is Power Usage Effectiveness?,
Consulting's/Accenture's Advanced Technology ecmweb.com, Dec 1, 2008
practice both in the US and Asia. He can be reached at
Julian.Turner@QAssociates.com [16] www.sun.com, X86 Power Calculator
Additional thanks to Sun Microsystems for
providing server and storage hardware for this test and
Green Platform Corporation for providing the AVR
1000 anti-vibration rack.
12. References
[1] ASHRE, Structural and Vibration Guidelines for
Datacom Equipment Centers, ASHRAE Publications,
2008
[2] sdd.toshiba.com, Toshiba Hard Drive

Environmental Specifications,
[3] European Parliament and the Council, Directive

2002/44/EC
[4] www.silentpc.com/forums, HDD Vibration and

Noise Reducing Methods – Ranked, 2003-2009
International Conference on Communication and Signal Processing, April 6-8, 2017, India
SURF based Video Summarization and its

Optimization
Dipti Jadhav and Udhav Bhosle

Abstract—Video Summarization (VS) is a technique of This paper proposes an efficient static video summarization
extracting keyframes from the video based on the video content. method using SURF features of the frames. The video
It provides the user with a concise representation of the video summary obtained is optimized using Graph Theory with the
content from which the video is semantically understood. This objective function that the graph generated by the optimal
paper proposes a video summarization technique based on
video summary is a simple graph with a simple walk. The aim
Speeded Up Robust Features (SURF). Authors further propose a
Graph Theory based approach to optimize the number of is to generate a graph such that the corresponding video
keyframes based on the objective function that the graph summary does not contain redundancy and retains only those
generated by the optimized video summary is a simple graph with keyframes that essentially add a semantic understanding to the
a simple walk. The proposed algorithm is tested on two different video summary.
videos from OpenVideo database. The video summarization
results and its optimization obtained shows non-redundancy and Various techniques using colour, texture, motion, object
improvement in the semantic understanding of the video extraction, mpeg feature extraction and attention model are
summary. also proposed in the literature. Xinding Sun et al. [1] proposed
content-based adaptive clustering (CBAC) using R-Sequences.
An innovative approach for video summarization using color
Index Terms—Video Summarization, SURF, Keyframes, histogram, wavelet statistics and an edge direction histogram is
optimization, objective function, graph theory, simple walk, reported by authors in [2]. Sandra E. F. de Avila et al. [3]
simple graph, semantic understanding. proposed VSUMM to generate video summary and a
quantitative evaluation. ZHAO Guang-Sheng [4] presented a
technique for video summarization using shot boundary
I. INTRODUCTION detection and keyframe extraction. Xiaohua He and
In the current digital era there is unexceptional growth in the JianLing[5] proposed a clustered block based color histogram
digital contents in the form of videos. Video Summarization keyframe extraction method TMOF. The main limitation of the
(VS) aims at developing the techniques that extracts keyframes video summarization techniques using low-level features is
from the video based on the video content. The semantic that it fails to capture the semantic understanding of the video
meaning of the video is understood from the keyframes. There [3, 4, 5].
are two forms of video summarization: 1. Static Video Changick Kim and Jenq-Neng Hwang [6] proposed a video
Summarization 2. Dynamic Video Summarization. Static video abstraction technique for video surveillance systems. Yael
summarization represents the summary as a collection of static Pritch, Alex Rav-Acha et al.[7] proposed a Non-Chronological
images which depicts the main contents of the video. Dynamic Video Synopsis and Indexing method for surveillance videos.
video summarization represents the original video as a short An efficient method to generate clustered summaries which
video that captures the main contents of the video [1]. In this can depict simultaneously multiple objects having similar
scenario, techniques for video browsing based on video motion is presented in [8]. Zhong Ji, Yuting Su, Rongrong
contents such as color, texture, shape, motion are proposed. Qian, Jintao Ma [9] proposed a technique for video
summarization based on moving object detection and
trajectory extraction. Zhiqiang Tian et al.[10] proposed key
Dipti Jadhav , Research scholar, Department of Computer
object-based video summarization (KOBVS) scheme for both
Engineering, Ramrao Adik Institute of Technology, Navi Mumbai,
University of Mumbai. India (Phone: 91-022-27709574; Email: static and dynamic background, efficiently depicting semantic
dipti.jadhav@rait.ac.in ) representative actions. Zhiqiang Tian, Jianru Xue et al.[11]
proposed a video summarization technique for moving and
Udhav Bhosle, Principal and Professor Department of Electronics static camera.
and Telecommunication Engineering, Rajiv Gandhi Institute of
Technology, Mumbai. University of Mumbai, India. (Phone: 91-022-
26707025; Fax: 022-26707026;Email: udhav.bhosle@mctrgit.ac.in)
978-1-5090-3800-8/17/$31.00 ©2017 IEEE
1252
Aditya Khosla et al.[12] proposed a novel video If the number of matched points and matched features are less
summarization and an evaluation technique of user generated than the threshold T it means that the two consecutive frames
videos using web-images as a prior. The presented techniques are not similar and hence frame R should be considered as a
are based on object extraction which helps in detecting and keyframe for video summarization. The frame I+1 is now the
tracking objects, detection of key actions. Feng Wang and reference frame R for further comparison.
Chong Wah[13] proposed to use Hierarchical Hidden Markov In the next step, SURF features for frame I+2 are extracted and
Model(HHMM) for summarizing videos. Wafee Sabbar and number of matched points and matched features between
et al. [14] proposed a technique using shot segmentation and frame R=I+1 and frame I+2 are calculated and the procedure
local motion estimation for video summarization. for keyframe extraction is repeated as described above. This
process is repeated for the next consecutive frames in the
This paper proposes a novel video summarization technique
video and the SURF based video summary S-VS is generated.
based on SURF features of the frames. The SURF features are
scale and rotation invariant and have fast computation. This
paper also proposes an optimization technique based on graph
theory and an objective function. The optimization technique
helps to eliminate the redundancy in the generated video
summary and thus improves the semantic understanding of the
video summary.
This paper is organized as under: Section II presents the
proposed SURF based video summarization technique. Section
III presents the proposed graph theory and objective function
based optimization technique. Section IV presents the
experimental results of the proposed techniques and its
performance analysis. Section V presents the conclusion.
II. NOVEL VIDEO SUMMARIZATION TECHNIQUE BASED ON

SURF
In the area of computer vision and image processing, Speeded
Up Robust Features (SURF) is used as interest point or
keypoint detector and descriptor for image matching, image
recognition and image registration. SURF is very robust
against different geometric and photometric transformations.
SURF works on integral images to reduce the computation
time[15]. In a video numerous transformations or changes
occur and thus the above mentioned properties of the SURF
can be used to capture keyframes from the video.
To detect image interest points, SURF uses determinant of
Hessian matrix. It detects major interest points in scale-space
using non-maximal suppression on scaled interest points map.
To achieve rotational invariance, SURF uses Haar Transforms
to find the direction of change in intensity. The feature vector Fig. 1. Block Diagram of the Proposed Video Summarization Technique S-
VS and Optimization Technique OVS-GT
or the descriptor is a 64 dimensional vector which describes
the distribution of Haar-wavelet responses around the interest
point[15].
III. PROPOSED VIDEO S U M M A RY
The block diagram of the proposed system is as shown in Fig
OPTIMIZATION TECHNIQUE BASED ON GRAPH
1.The input video is separated into individual frames. The first
T HEORY AND OBJECTIVE FUNCT I ON
frame in the video is considered as reference frame R. The
SURF features of two consecutive frames I=R and I+1 are The proposed optimization technique is based on the concept
extracted. The number of matched points and matched features of graph theory and objective function. The graph theory is
are extracted for the frames R and I+1. The number of used to generate a graph G from the video summary S-VS. The
matched points between frame R and I+1 are compared with a graph G generated helps to identify the redundancy, if any in
threshold T calculated empirically. If the number of matched the video summary.
points and matched features are more than the threshold T, it The objective function is such that the graph G generated by
means that both the frames R and I+1 are similar and hence the video summary should be a simple graph with simple walk.
frame I+1 is rejected and frame R still remains reference frame This objective function helps to remove any redundancy in the
for further comparison. summary and retains the essential keyframes.
1253
A. Optimization algorithm OVS-GT and the second keyframe kfi+1. The number of matched points
Algorithm: OVS-GT. and matched features between keyframe kfi and kfi+1 are
Find optimized set of keyframes O-KF from the video calculated. If the number does exceeds an empirical threshold
summary S-VS based on the objective function and graph Tg calculated, then the keyframes kfi and kfi+1 are similar to
theory. each other and hence a loop ei with weight Wei=1 is generated
Input: A set of keyframes KF of the form KF: kf1,kf2,..,kfn around the vertex Vi and the weight of the vertex Vi is also
where kfi is the keyframe from S-VS. incremented by 1.This loop helps to detect the redundancy.
Output: A set of optimized keyframes O-KF without If the number of matched points and matched features for the
redundancy. keyframe kfi and kfi+1 does not exceeds the threshold Tg, then
Method: both the keyframes are not similar and hence a new vertex
V <=Set of vertices generated. Vi+1 with weight i+1 is created and an edge is generated from
E <=Set of Edges or Loops generated. Vi to Vi+1. This edge represents the chronological order of the
Wvi <= weights of vertices. images that should be present in the optimized video summary.
Wei<=weights of edges or loops. Further the keyframe kfi+2 in the video summary S-VS is
Tg <= empirical threshold. considered and its SURF features are calculated. The number
cnt1 <=counter1. of matched points and matched features between keyframe
cnt2 <= counter2. kfi+1 and kfi+2 are calculated. If the matched features does
wt <= weight allocated to vertex and edge. exceeds the threshold Tg then keyframe kfi+1 and keyframe
SF <= SURF features of keyframe kfi. kfi+2 are similar and hence a loop with weight Wei+2 is again
generated around vertex Vi and the weight of the vertex Vi is
for each keyframe kfi in KF do incremented by 1. Thus the weight of the vertex is an
indication of the number of adjacent keyframes that are similar
SFi <=SURF features extracted for kfi and hence helps to detect redundancy. If the number of
SFi+1<= SURF features extracted for kfi+1 matched points and matched features does not exceeds
threshold Tg, then the two keyframes are dissimilar and a new
Calculate the number of matched points and matched SURF vertex Vi+2 with weight i+2 is generated and an edge ei is
features between the keyframes kfi and kfi+1 generated between vertex Vi+1 and Vi+2. This process is
repeated for all the keyframes kf in the keyframe set KF.
If matched points > Tq Thus a graph G(V,E) with vertices V that represents the non-
redundant or dissimilar keyframes, edges E that represents the
cnt2(i) <=cnt2+1 chronological order between two keyframes that should follow
else each other and loops that represent redundant keyframes in the
cnt2(i) <= 0 video summary is generated.
endIf In the next step, the graph G is checked for the objective
endFor function. At the first iteration, the generated graph G is
cnt2(i+1) <= 0 evaluated for objective function that checks if the graph is
for each i do simple with a simple walk. If the graph G contains loops, then
if cnt2(i)==cnt2(i+1) and cnt2(i+1) ==1 the first vertex Vi is considered and the ranking is assigned to
then Wvi <= wt+1 and Wei <= wt+1 every loop around Vi as equation (1).
else
Wvi <= wt and Wei <= 0 Rank Ei = Wei / Wvi (1)
for each i do The loop around Vi with the lowest rank is deleted which
if Wvi == Wvi+1 and Wei == 1 actually leads to the deletion of a redundant keyframe from the
then cnt <= cnt+1andWvi <= cnt S-VS video summary. The weight of the vertex Vi is
decremented by 1. Rankings are reassigned to the loops around
for each i do vertex Vi with new weight of Vi as equation (1).
In the next iteration, the graph G is again evaluated for the
Wei <= Wei/cnt objective function and if the objective function is not satisfied
then the first vertex Vi in the graph G with loop is considered
D <= maximum of Wei again. The loop with lowest rank is deleted. The rankings are
for each i do reassigned to the loops. The graph is again checked for the
if Wei == D objective function.
then delete keyframe kf from set KF. The optimization continues until the objective function is
reached. The algorithm converges when there are no loops in
A vertex Vi is generated for the first keyframe kfi from the the graph G. The ranking assigned helps assigning a factor of
summarized output S-VS and a weight Wvi=1 is allocated to importance to the keyframes. The number of loops generated
it. The SURF features are extracted for the first keyframe kfi around a vertex denotes the number of adjacent similar
1254
keyframes in the video summary. The edge between the ver- The video summary of video v37 obtained in Fig.4 is
tices establishes a chronological order of the non-redundant or optimized using proposed graph theory based objective
dissimilar keyframes in the optimized video summary. The function algorithm OVS-GT. The redundant keyframes which
proposed algorithm preserves the semantic meaning of the are deleted at different iterations is as shown below.
summary by assigning independent vertices to every non 1st iteration: Frame 10 deleted. Fig 5
redundant and semantically meaningful keyframe in the video 2nd iteration: Frame 13 deleted. Fig 6
summary. 3rd iteration: Frame 13 deleted. Fig 7
4th iteration: Frame 15 deleted. Fig 8
IV. EXPERIMENTAL RESULTS AND PERFORMANCE ANALYSIS 5th iteration: No change
The proposed SURF based video summarization and The optimization algorithm for video v37 converges after 4th
optimization algorithm is tested on videos v37 and v24 taken iteration. As observed above, in every iteration the graph
from the OpenVideo database. theory approach and the objective function helps to delete only
those keyframes that are redundant. It retains the keyframes
that are important and improves the semantic understanding of
A. Experimental results for video v37 the video summary. The video summary of STIMO[16] and
VSUMM[3] for video v37 is shown in Fig. 9 and Fig. 10
respectively.
Fig. 2. Sample Video Frames extracted from the video 37
Fig 5. Iteration 1: Frame 10 deleted
Fig. 3. Sample SURF feature vector for an image I under consideration

The sample of SURF features extracted from an image I of

video v37 under consideration is as shown in Fig 3. It shows
the 64 dimensional SURF feature vector f1 extracted from the
142 detected interest points of an image I. These detected
interest points and its feature vector is matched with the
interest points and feature vector f2 of image I+1. The number
of matched points between two images I and I+1 is calculated.
Fig. 4. The video summary S-VS generated by the SURF parameters for the
video v37 (OpenVideo database)
Fig. 8. Iteration 4: Frame 15 deleted
The experimental results of the proposed SURF based video
summarization S-VS as discussed in Section 2 is shown in Fig
4. As observed in Fig. 4 the robustness of SURF generates a
satisfactory video summary that captures salient actions in the
video alongwith various intermediate activities of the video.
1255
Fig. 9. Keyframes extracted by STIMO[16] for video v37
Fig 13. The video summary S-VS generated by the SURF parameters for the
video v24 (OpenVideo database)
Fig. 10. Keyframes extracted by VSUMM[3] for video v37
B. Experimental results for video v24
Fig 14. First iteration: Frame 3 deleted
Fig 11. Sample Video frames extracted from the video v24
The sample of SURF features extracted from an image I of
video v24 under consideration is as shown in Fig 12. It shows
the 64 dimensional feature vector or descriptor f1 for the 24
detected interest points for image I. These detected points and
its vector f1 is matched with the detected interest points and
feature vector f2 of image I+1 and matched points are
calculated.
The SURF based video summary S-VS for the video v24 is as Fig 15. Frames obtained after 2nd, 3rd, 4th, 5th, 6th, 7th iteration
shown in Fig. 13. The iteration wise output for the
optimization of the video summary of video v24(OpenVideo)
from the proposed optimization algorithm OVS-GT is as
shown in Fig. 14, Fig. 15, Fig. 16.
First iteration: Frame 3 deleted. Fig.14.
Frames obtained after 2nd, 3rd, 4th, 5th, 6th, 7th iterations.
Fig. 15.
8th iteration: Frame 25 deleted. Fig.16
The algorithm converges after 8 iterations. Fig 16. 8th iteration: Frame 25 deleted
The video summary of STIMO [16] and VSUMM [3] for
video v24 is as shown in Fig. 17 and Fig. 18 respectively.
Fig 17. Video summary of STIMO[16] for video v24
Fig 12. Sample SURF features extracted from a frame I of video v24
Fig 18. Video summary of VSUMM[3] for video v24
The comparison of the video summary of proposed algorithm

S-VS and the final iteration output of the optimization
1256
algorithm OVS-GT for videos v37 and v24 is performed with V. CONCLUSION
the video summary of VSUMM [3] and STIMO [16]. The robustness of the SURF features is used to generate a
It is observed that the robustness of SURF’s detector and video summary that can capture salient frames and actions in
descriptors helps to capture much finer details and actions in the video. The proposed optimization algorithm with its
the video. Thus the video summary generated by SURF based objective function helps to retain keyframes that helps to
algorithm S-VS is more semantically meaningful. The finer improve the semantic meaning of the video summary. The
details captured by SURF can lead to some amount of proposed algorithms can be used across various genre of
redundancy as observed in Figure 4 and Figure 13. videos. The future scope is to develop an objective evaluation
The optimization algorithm OVS-GT when applied to S-VS technique to evaluate the performance of the algorithm used to
helps to retain the essential keyframes that add semantic generate a video summary.
meaning to the video summary and removes the redundant
keyframes as illustrated in Figure 8 and Figure 16. REFERENCES
In the field of Video Summarization the performance analysis
[1] Xinding Sun and Mohan S. Kankanhalli, “Video Summarization using
is done subjectively by displaying the video summary to testers R-Sequences”, Real Time Imaging 6, 449-459, 2000 Academic Press.
from various categories like students, researchers, etc. Three [2] Ciocca Gianluigi, Schettini Raimondo, “An Innovative Algorithm for
scaling grades are allocated: 3.Good 2. Acceptable 1.Bad Key-frame extraction in Video Summarization”, Journal of Real Time
based on whether the semantic meaning of the video is Image Proc, 1:69-88, Springer-Verlag, 2006.
[3] Sandra E.F. de Avila, Antonio da Luz Jr. Arnaldo A Araujo, Matthieu
understood from the video summary. The testers are first given Cord, “VSUMM: An Approach for Automatic Video Summarization
the database videos and then presented with the video and Quantitative Evaluation”, XXI Brazilian Symposium on Computer
summary. The subjective evaluation of the video summary Graphics and Image Processing, 2008 IEEE.
generated after optimization is shown in the Table I. The [4] Zhao Guang-Sheng, “A Novel Approach for Shot Boundary Detection
and Key-Frames Extraction”, 2008 International Conference on
number of frames column represents the total number of Multimedia and Information Technology, IEEE 2008.
frames in the video. The number of keyframes column [5] Xiaohua He and Jian Ling, “A Video Summarization Method Based on
represents the number of keyframes generated by the Key-frames Extracted by TMOF”, IEEE 2012.
mentioned techniques. The tester’s score in columns 2, 3 and 4 [6] Changick Kim, Jenq-Neng Hwang, “Object- Based Video Abstraction
for Video Surveillance Systems”, IEEE Transactions on Circuits and
are given for the proposed optimized video summary.
Systems for Video Technology, Vol 12, No. 12, December 2002.
[7] Yael Pritch, Alex Rav-Acha, Shmuel Peleg, “Non-Chronological
TABLE I Video Synopsis and Indexing”, IEEE Transactions on Pattern Analysis
PERFORMANCE ANALYSIS AND SUBJECTIVE EVALUATION and Machine Intelligence,2008.
[8] Yael Pritch, Sarit Ratovitch, Avishai Hendel, Shmuel Peleg,
“Clustered Synopsis of Surveillance Video”, IEEE 2009, Advanced
Video and Signal Based Surveillance.
[9] Zhong Ji, Yuting Su, Rongrong Qian, Jintao Ma, “Surveillance Video
Number of Key Frames Extracted Summarization Based on Moving Object Detection and Trajectory
Extraction “, IEEE 2010, 2nd International Conference on Signal
Video G A B NF OVS- VSUMM STIMO Processing Systems(ICSPS).
GT [3] [16] [10] Zhiqiang Tian, Jianru Xue, Xuguang Lan, Ce Li, Nanning Zheng, “Key
Object-based Static Video Summarization”, MM 2011, Proceedings of
the 19th ACM International Conference on Multimedia pages, 1301-
1304, ACM, NY, USA.
V24 85% 15% 0 1850 22 11 10 [11] Zhiqiang Tian, Jianru Xue, Xuguang Lan, Ce Li, Nanning Zheng,
“Object Segmentation and Key-pose Based Summarization for Motion
Video”, Multimedia Tools Applications, Springer, May 2013.
[12] Aditya Khosla, Raffay Hamid, Chih-Jen Lin, Neel Sundaresan,
V37 85% 15% 0 2900 16 05 11
“Large Scale Video Summarization using Web Image Priors”, IEEE,
Proceedings of Computer Vision and Pattern Recognition, 2013.
[13] Feng Wang, Chong Wah, “Summarizing Rushes Videos by Motion,
Object and Event Understanding”, IEEE Transactions on Multimedia,
G= Good, A=Acceptable, B=Bad, NF= Number of frames in the video. Vol.14, No.1, February 2012.
[14] Wafae Sabbar, Adil Chergui, Abdelkrim Bekkhoucha, “Video
As observed in Table I the testers score shows that the Summarization using Shot Segmentation and Local Colour
proposed techniques give satisfactory results. The number of Estimation”, Computing Technology (INTECH), 2012, pp.190, and 18-
20 Sept 2012.
keyframes retained by OVS-GT is more than the keyframes [15] Herbert Bay, Tinne Tuytelaars, and Luc Van Gool, “SURF: Speeded Up
generated by VSUMM [3] and STIMO [16].This is because Robust Features”, A. Leonardis, H. Bichof, and A. Pinz (Eds.): ECCV
the robustness of SURF features captures minor variations and 2006, Part I, LNCS 3951, pp. 404–417, Springer-Verlag, Berlin,
transitions in the video. The proposed optimization algorithm Heidelberg 2006.
[16] Marco Furini, Filippo Geraci, Manuela Montangero, Marco Pellegrini,
and ranking mechanism retains only those keyframes that "STIMO: STIll and MOving Video Storyboard for the WebScenario",
depicts the chronological actions in the video. This improves Multimed Tools Appl (2010) 46: 47. doi:10.1007/s11042-009-0307-7.
the understandability of the video summary.
1257
International Conference on Signal Processing and Communication (ICSPC’17) – 28th & 29th July 2017
Mammography Classification Using Modified

Hybrid SVM-KNN
Poonam Sonar 1 and Udhav Bhosle 2 Chandrajit Choudhury3
3
Department of Electronics & Telecom Engineering Department of Electronics & Communication Engineering
3
Rajiv Gandhi Institute of Technology, Mumbai. National Institute of Technology Silchar.
3
University of Mumbai, India chandrajit.choudhary@gmail.com
{poonam.sonar 1, udhav.bhosle 2}@mctrgit.ac.in
Abstract- Today leading cause of cancer deaths for women is the for mammogram analysis and classification. Mass, change in
Breast cancer. For early and accurate detection of breast cancer, shape, color and dimensions of breast are some of the
mammography is found to be the most reliable and effective symptoms of cancerous breast. These variations can be studied
technique. In this context, computer aided diagnosis of breast
by extracting the textual features of mammogram. In this
cancer from mammograms is gaining high importance and
priority for many researchers. In this paper machine learning context, representation of mammograms by significant and
based mammogram classification using modified hybrid SVM- most distinctive features and use of these feature sets for
KNN is proposed. The idea is to map the feature points to kernel classification is most important [3].
space using kernel and find the K nearest neighbors among the Numbers of techniques for breast cancer detection has been
training dataset for a given test data point. Doing this we narrow reported in literature. Mammogram classification using
down the search for support vectors. Mammogram images are association rules and neural network classifier is reported in
preprocessed and region of interest is extracted using Fuzzy C [4]. Authors in [5] used image mining technique based on
Means clustering and Active Counter technique. GLCM (grey statistical and GLCM features to classify mammogram images
level covariance matrix) based texture features are extracted
into normal, benign and malignant class. Authors in [6] extract
from segmented ROI. These features are used to train modified
hybrid SVM-KNN classifier proposed by authors. The trained texture features from mammogram ROI. These features are
classifier is used to classify breast tissues in normal/abnormal ranked and highest rank features are used for classification
classes and further abnormal class into benign/malignant. using KNN. Hassanienet al [7] used maximum difference
Proposed technique is experimented on two standard MIAS and method for feature selection and proposed ensemble
DDSM databases. The proposed classifier reports classification supervised algorithms for classification. Masses detection on
accuracy of 100 % for DDSM and 94% for MIAS database for digitized mammograms and further classification using SVM
benign and malignant class. Results are compared with SVM, (Support Vector Machine) is reported by Martains [8]. Author
KNN and Random Forest classifiers. uses texture feature extraction and K-means algorithm for
image segmentation. Authors in [9] used KNN, PCA and
Keywords—Mammograms, Texture feature ,SVM,KNN,Hybrid
SVM combination for mammogram classification.
SVM_KNN,Random Forest
In this paper, post preprocessing of mammograms, author use
I. INTRODUCTION fuzzy C means clustering and active contour technique for
segmentation. From segmented ROI, low level local texture
The most frequently diagnosed cancer among women in the features are extracted. Using these features, mammograms are
world is the breast cancer [1]. Early stage breast cancer classified into normal, benign and malignant class by SVM,
detection is helpful for reducing patient mortality [2]. Patient KNN, and Random Forest classifier followed by proposed
survival rate has been improved by early detection through modified hybrid KNN_SVM classifier. Experimentation is
screening. The mammography is the most reliable and carried out on standard MIAS and DDSM databases. The rest
effective technique for early and accurate detection of breast of the paper is structured as follows. Section II discusses
cancer. Thus, everyday large numbers of mammograms are proposed method. Section III proposed modified hybrid
generated in hospitals. Therefore, physicians require an SVM_KNN classifier and Random Forest is discussed. The
accurate and automatic reading of digital mammograms due to simulations results are presented in Section IV and conclusion
its huge volume. At the same time, analysis and diagnosis of and scope for future work is presented in Section V.
mammograms is very critical and difficult. Hence, there is a
need of development of efficient CAD system. Duo to this II. PROPOSED METHOD
breast cancer detection using CAD system has gained a lot of
research attention in the past few years. Fig. 1 shows the block diagram of proposed scheme. It starts
Attempts are being made to develop an efficient CAD system with mammogram segmentation to extract ROI using active
for accurate interpretation of mammograms and further to contour and snake method. From the extracted ROI, GLCM
enhance the classification accuracy. CAD system reduces based texture feature are extracted to form feature vector
variability of judgments among the radiologists. Mammogram representation for each mammogram image. The feature
images show breast tissue density variations, which are used vector is given as input to classifier, to classify input
305
ECE Programme, Karunya University
D. Classification of Mammogram images
(a) Support Vector Machine (SVM):
Support Vector Machine is a machine learning technique [12]

used to analyze data, recognize patterns and classify data into
different classes by constructing a hyper plane in an infinite
dimensional space. In SVM training data samples are
projected on to higher dimension space to find an optimal
hyper plane. The data samples along the hyper planes near the
class boundary are called support vectors. The distance
between support vectors and separating hyper planes is called
margin.
A classification technique consists of training and testing data
along with some data instances. Each instance in the training
set contains one “target value” (class labels) and several
“attributes” (features) [13].
Fig.1 Block diagram of proposed system Consider a set of n data samples of two classes ( , ),
( , )...., ( , ), i=1,2,…,n, mapped to a higher
mammogram into abnormal and normal class and further into dimensional space and = ±1..The separating hyper plane
benign and malignant class should be optimal for correct classification of class [13]. Here
the optimization problem is to find optimal separating hyper
A. Mammogram Preprocessing plane to separate positive and negative classes defined by
following equations (2) and (3)
Preprocessing of digital mammograms involves suppression of
noise and radiopaque artifacts. It is necessary to improve the w. +b=0 (1)
mammogram image quality to make feature extraction phase (w. + b) ≥ 1(for = 1) (2)
more reliable. Median filtering is used as preprocessing step (w. + b) ≤ -1 (for = -1) (3)
to remove digitization noise introduced during image
acquisition process. Image enhancement is achieved by SVM tries to maximize the margin between two classes by
Contrast Limited Adaptive Histogram equalization (CLAHE) finding a weight vector w and bias weight b by
and thresholding. This step removes mammographic artifacts
such as labels, markers and wedges. minimizing || ||
In SVM different kernel functions are used to make non-
B. Segmentation and ROI Extraction separable problem into separable and to map data in better
representation space. Here kernel functions are used to map
Artifacts and background noise are prominent in feature values from finite space to infinite dimension space. In
mammograms images. It also contains unwanted Pectoral SVM, different kernel functions are used. Some of the
muscles in the object area. So, full mammogram images are commonly used kernel functions are below [13].
not required for texture feature extraction and classification.
Therefore, for effective texture analysis, mammogram images Linear Kernel Function: K( , ) = 1 + (4)
are segmented to remove pectoral muscles and to extract the Polynomial of the power p: K( , ) = (1 + ) (5)
region of interest (ROI). For segmentation, fuzzy C means Radial-Basis Function (Gaussian):
clustering and active contour techniques are used. ROI ⃦ ⃦
contains the abnormalities in the texture variation and K( , )=exp − (6)
eliminates unwanted portions of the mammogram image. Sigmoid: K( , ) = tanh ( + ) (7)
Here, in our experimentation, linear kernel function is used.
C. Feature Extraction
SVM is trained with the feature vectors of mammograms.
GLCM is a statistical method of examining texture based on
(b) K-Nearest Neighbor algorithm (KNN):
spatial relationship of pixels and is used for texture features
extraction [10]. GLCM characterizes texture of an image by
KNN is supervised learning algorithm used for statistical
calculating its Co-occurrence matrix. Co-occurrence matrices
estimation and pattern recognition. It classifies objects into
are calculated for 00 and 450 directions and for distances 1& 2,
separate classes based on closest training examples in feature
Thus, each mammogram image is represented by feature
vector space. It is used to estimate the class of an unknown
vector of size [88 X 1]T [11].
instance from its neighboring instances. It stores all possible
cases and classifies new cases based on distance metric and
306
returns the most common value. All training samples are average for regression). Here, we have used DDSM database
stored in n dimensional pattern space. The class of an for experimentation. We have generated 500 decision tresses
unknown sample is decided by a majority vote of its for classification of mammogram into appropriate class.
neighboring samples in the training pattern space. There are Gini Index:
many distance measures that can be used in KNN In the step (b) in the above list of steps for training and testing
classification such as Euclidian distance, Manhattan distance, random forest we have mentioned the splitting criteria in each
Minkowski distance, Mahalanobis distance etc. [14]. However decision node of a tree. We shall further discuss that here.
Euclidean distance is the most commonly used distance There are many methods for splitting at a decision node in a
metric. tree, namely Gini Index, Information Gain, Chi-Square etc. In
The KNN classifier can perform efficiently, if the optimal our experiment, we have used Gini Index as this yielded the
value of ‘k’ is used in the classifier design. In the present best result. Gini Index of a node (t) can be calculated as:
work, after experimentation, the optimal value of ‘k’ is
decided as 5. GINI (t) = 1 - ∑ (8)
Let k denotes desired number of nearest neighbors and
F: Pattern Space: set of training samples {F1 ,F2,F3,….FN,}: Where M is the number of classes of the variable and is the
The set of feature vector of training mammograms in the form proportion or probability of class in the data. And Gini
of Fi = {Xi,Ci} Where Xi: feature vector of the point Fi and Index of a split is calculated as [18]:
Ci is the class that belongs to.
F' = {X'i,C'i} :Test mammogram feature vector
( , )= ( )− ( )− ( ) (9)
For a test mammogram, following steps are followed to make
prediction: Where s: split
(i) Compute the Euclidian distance d (X', Xi) between test t : node
mammogram and all training mammograms. : Proportion of observation in left node after split s.
(ii) Sort all points Fi according to distance ( ) : Gini index of left node after split s.
(iii) Select the first K points from the sorted list. These are k : Proportion of observation in right node after split s.
closest samples to F. ( ) : Gini index of left node after split s.
(iv) Choose the closest label.
(v) Return Majority label and assign class to test mammogram III. HYBRID SVM KNN SUPERVIED CLASSIFIER
F' based on majority vote
The limitation of KNN classifier is false classification of test
(c) Random Forest image when majority of the nearest neighbors have closely
matched features. To overcome this problem, KNN could be
Random forest is a versatile machine learning technique that combined with another classifier [15, 16]. Here we combine
can be used for classification as well as regression. It is an KNN with SVM and resultant hybrid supervised classifier is
ensemble method where a group of weak classifiers are used for mammogram classification for improving the
combined to form a more powerful classifier. Here we grow performance of the classifier and accuracy. The idea is to map
multiple decision trees from the training dataset unlike a single the feature points to kernel space using kernel and find the K
Classification and Regression tree (CART) model. When we nearest neighbors among the training dataset for a given test
feed a new data point or case to the model each of the trees data point. Doing this we narrow down the search for support
classifies the data, we say each of the tree votes for a “class”. vectors to the nearest neighbors or the more relevant data
The class with maximum number of votes is considered the points. SVM classifier is trained over these nearest data points
winner and we say the model classified the data into that class. and used to classify test data point.
The training and testing steps of a random forest for
classification are stated below [17]: Proposed Hybrid KNN-SVM Classifier:
If there are N data points in the training dataset, then they are
randomly sampled with replacement and that forms the (1) Divide given database into training and testing class data.
training dataset for a tree in the forest (for each tree random (2) Segment ROI, extract GLCM texture features and
sampling is done). represent training and testing images by feature vectors.
If there are K variables in each data point, then a number k < (3) Apply K nearest neighbor algorithm using Euclidean
K is selected such that k variables are randomly taken and are distance. Different values of k are experimented, but
used for splitting at each node in the tree. The best split optimum results are obtained at k=30, hence it is used.
(according to whatever splitting criteria is chosen) out of these (4) If all the k neighbors have the same label, classification
k variables is selected to split that node. The value of k is kept process is terminated and the test image is given as the
constant across the forest. Each tree is grown to the deepest label of the neighbors.
extent possible. To predict a new data-point the output of all (5) If the test image does not identify in previous step, then
the trees are aggregated (maximum vote for classification and compute the pair wise distance matrix between the
307
neighbors. Convert distance matrix into kernel matrix
using kernel trick.
(6) Using kernel matrix, train SVM on the K nearest TABLE 1: IMAGE FEATURE VECTOR
neighbors in the training data and classify test image
using the trained SVM classifier. Label
Image Image Feature Values from 1 to 88
Numb
1 2 . . . 86 87 88
IV RESULTS AND DISCUSTION er
1 1111 18.34986 18.32752 . . . 0.997142 0.997599 0.997033
2 1111 20.82134 20.79414 . . . 0.998902 0.999104 0.998823
To validate the proposed algorithm, simulations are carried in 3

4
1111
1111
31.29402
33.22682
31.28536
33.22527
.
.
.
.
.
.
0.999055
0.999189
0.999373
0.999411
0.999238
0.999214
the MATLAB environment. Mammographic images from two . . . . . . . . . .
standard databases Mammographic Image Analysis Society .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
(MIAS) and Digital Database for Screening Mammogram 137 2222 60.47571 60.22955 . . . 0.989071 0.990012 0.989046
(DDSM) are used for experimental work. 138

139
2222
2222
61.83373
60.74252
61.50633
60.28726
.
.
.
.
.
.
0.983929
0.980215
0.985562
0.98165
0.98394
0.980209
140 2222 61.83209 61.47212 . . . 0.983782 0.984861 0.983758
(a) DDSM Database: From this database, 120 images are

selected for experimentation. Number of images used for Fig.4 (a) shows original mammogram, initialization rectangle
testing and training are shown in Table3. Few Sample of size [200*200] is placed at the center, 200 iterations are
mammogram images are shown in Fig.2. performed using snake method. Final extracted ROI is shown
in fig 4(d).From the extracted ROI, GLCM features are
calculated in two directions and two distances. So, it forms a
feature vector of size [88 x 1]T. We use 140 images for
training, so it generates a feature vector of size [140*88] as
shown in Table 1. Similarly, GLCM features and feature
vector is calculated for test images [11].This Feature vector
along with its appropriate class is given as input to SVM
Fig.2: Few Sample images from DDSM database [26] classifier to SVM to form support vectors.
(b) MIAS Database:

Fig. 5 shows some sample images in MIAS database [27].
Fig. 6 shows preprocessing using median filtering and
Contrast enhanced image.
(a) (b) (c) (d)
Fig.3 Mammogram preprocessing (a) Original Image(b) Median
filtering (c) Thresholding (d) Contrast Enhanced
Fig.5: Few Sample images from MIAS database [27]
(a) (b) (c) (d)

Fig.4: Mammogram Segmentation: (a) original image (b) rectangular
initialization (c) output of active contour method (d) extracted ROI
In Fig.3 shows mammogram preprocessing. Original

mammogram is shown in 3(a). Median filtered mammogram
image as shown in Fig.(3(b). Contrast Limited Adaptive (a) (b) (c)
Histogram equalization (CLAHE) (3(d)).Active contour Fig 6: Mammogram Preprocessing: (a) Original(b) Median filtered
(c) Contrast Enhanced
segmentation using snake approach is used for segmentation.
Edge based active contours is used to locate the contour C that
minimizes total energy E with certain set of weight. A set of
snake points residing on image plane are defined. The next
location of those snake points are determined by the local
minimum E. The associated form of those snake points is
considered as the contour. Fig.4 shows segmentation process. (a) (b) (c)
Fig.7: Mammogram Segmentation (a) Fuzzy c means clustering
output (b) Hough Transform output (c) Extracted ROI
308
Fuzzy C means clustering is used to segment the normal and abnormal class. Table 4 shows result of
mammogram, for removal of radiopaque artifacts such as normal/abnormal classification. Table 5 shows number of
labels, markers and wedges. Hough transform is used to images for benign/malignant classification. Results of
remove pectoral muscle removal.Fig.7(C) shows final ROI benign/malignant classification is shown in Table 6.
extracted. From extracted ROI, GLCM features are calculated TABLE 5: IMAGES FOR BENIGN AND MALIGNANT CLASSIFICATION.
to generate feature vector for test images from DDSM
database as shown in Table 2. Training Images Testing Images
Database Benign Malignant Benign Malignant
TABLE 2: IMAGE FEATURE VECTOR MIAS 45 45 15 15
DDSM 40 50 21 25
Image Image Feature Values from 1 to 88
number 1 2 . . . 86 87 88
1 55.93361 55.91095 . . . 0.997142 0.997599 0.997033 TABLE 6: CLASSIFICATION ACCURACY OF BENIGN/MALIGNANT
2 53.94997 53.65203 . . . 0.998902 0.999104 0.998823
3 42.45662 42.09956 . . . 0.999055 0.999373 0.999238
4 47.44773 47.04855 . . . 0.999189 0.999411 0.999214 Data Classifier Accuracy
. . . . . . . . .
. . . . . . . . .
Set Benign Malignant Avg Sensitivity Specificity
. . . . . . . . .
73 59.27305 58.66695 . . . 0.989931 0.990901 0.989918
MIAS KNN 93.33 93.33 93.33 93.33 93.33
74 62.67675 62.49319 . . . 0.996633 0.996936 0.996635
75 62.95228 62.89389 . . . 0.978394 0.980028 0.978368
76 60.44935 59.99166 . . . 0.989771 0.99089 0.989731
MIAS SVM 93.33 93.33 93.33 93.33 93.33
TABLE 3: IMAGES FOR NORMAL/ABNORMAL CLASSIFICATION MIAS Hybrid 95.77 95.77 95.77 95.77 95.77
KNN-SVM
Training Images Testing Images
Database DDSM KNN 96 95.23 95.65 96 95.23
Normal Abnormal Normal Abnormal
MIAS 45 70 15 30
DDSM 50 90 25 46 DDSM SVM 95.55 95.55 95.55 95.55 95.55
DDSM Random 95.55 95.55 95.55 95.55 95.55

TABLE 4: RESULT FOR NORMAL/ABNORMAL CLASSIFICATION Forest
DDSM Hybrid 100 100 100 100 100
KNN-SVM
Data Accuracy
Classifier
Set Normal Abnormal Avg Sensitivity Specificity TABLE 7: COMPARISON OF THE PROPOSED METHOD WITH OTHER CLASSIFIERS
MIAS KNN 86.66 93.33 91.11 86.66 93.33

Investigator Features Classifier Accuracy
W. Borges Shape and
MIAS SVM 86.66 93.33 91.11 86.66 100 SVM 80
Sampaio, et al. [20] Texture
GLCM, Structured
Hybrid Wang [21] 91.4
MIAS 86.66 93.33 91.11 86.66 100 Gabor SVM
KNN-SVM Shape
Y.Ireaneus Anna
Feature SVM 82
Rejani, et al. [22]
DDSM KNN 96 95.65 95.77 96 95.652 based DWT
Shearlet
DDSM SVM 100 86.95 91.54 100 86.95 transform
Kanchanamani[25] SVM 92.5
and statistical
Random features
DDSM 100 95.65 97.18 100 95.65
Forest Shape and
Maria-Luiza[4] BPNN 92
Hybrid GLCM
DDSM 100 95.65 97.18 100 95.65 Statistical Rough set
KNN-SVM
features and Theory, Joining
Jiang Yun et el[19] 77.48
rough set Associative
Experimentation is performed using KNN, SVM, Hybrid theory Classifier (JAC)
KNN-SVM classifier and Random forest classifier. Initially M. ArfanJaffar[23] DCT SVM 98.3
mammogram images are classified into two classes normal Deepa
Texture Associative 94.66
and abnormal and then it is further classified into benign and Despande[24]
Statistical
malignant class. The proposed hybrid KNN-SVM classifier Vipul Sharma[6]
Ranked
KNN 97.22
uses KNN with K = 30 and for SVM Radial bias function with 100 :Benign-
σ = 1000. First KNN classifier is used, since it requires less Proposed
GLCM
Hybrid KNN- Malignant
computation time, and when it combines with SVM, Approach SVM 97.18.:Normal
Abnormal
performance of the classifier is enhancing. Table 3 shows
number of images used from DDSM and MIAS database for
309
MIAS DDSM
Category Predicted Class Category Predicted Class
Malignant Benign Malignant Benign

Actual Malignant TP = 13 FN = 02 Actual Malignant TP = 24 FN = 01
Class Benign FP = 02 TN = 28 Class Benign FP = 01 TN = 20
Table 8 : Confusion Matrix – KNN
MIAS DDSM

Table 9 : Confusion Matrix – SVM
MIAS DDSM

Table 10 : Confusion Matrix – Hybrid SVM KNN
Comparison results with other classification technique are experimentation with more classifiers is required. There is
represented in Table 7. Table 8, 9, 10 shows confusion matrix further scope for reducing the complexity of the proposed
for benign/malignant classification using KNN, SVM and classifier by reducing the number of features for training the
Hybrid KNN-SVM classification respectively. classifier. For that feature optimization technique using
transformed domain approach can be used.
V. CONCLUSION AND FUTURE SCOPE Acknowledgement
Authors present mammogram classification modified hybrid The authors express thanks to Dr. J. Suckling and co-authors
KNN-SVM classifier to improve the classification accuracy. for providing access to the dataset entitled “Mammographic
The experimental results are validated on two standard Image Analysis Society (MIAS) [27]. The authors also would
databases MIAS and DDSM. The results are compared with like to thank Dr. Thomas Deserno and co-authors, Department
KNN, SVM, and Random Forest classifier. The proposed of Medical Informatics, RWTH Aachen, Germany, for
method gives accuracy up to 97.18 for normal/abnormal providing access to the dataset entitled, “IRMA VERSION OF
mammogram classification and 100% for benign malignant DDSM LJPEG DATA” [26].
classification. The extracted feature set seems to be stable as it
yields same accuracy in hybrid SVM-KNN and Random
Forest Classifier. But at this stage, we cannot conclusively say
anything about the stability of the features as exhaustive
310
[14] http://machinelearningmastery.com/k-nearest-neighbors-for-machine-
References learning/
[15] Hao Zhang Alexander C. Berg, “ SVM-KNN: Discriminative Nearest
[1] Population Based Cancer Report, Neighbor Classification for Visual Category Recognition”
http://ncrpindia.org/Annual_Reports.aspx [Online].
[16] Lam Hong,Lee ,Chin Heng, “Overview of Nearest neighbor support
[2] American Cancer Society (1999). Breast Cancer Facts &Figures 1997- vector machine hybrid classifier”, Journal of Applied sciences 10(17)
1998, 1845-1858,2010
[3] Nithya R, Santhi B.,” Classification of normal and abnormal patterns in [17] https://www.analyticsvidhya.com/blog/2016/04/complete-tutorial- tree-
digital mammograms for diagnosis of breast cancer”, Int J Comput Appl based-modeling-scratch-in-python/
T. 2011;28(6):21–5.
[18] http://dni-institute.in/blogs/gini-index-work-out-example/
[4] Maria-Luiza Antonie, Osmar R. Zaiane, Alexandru Coman,
[19] Jiang Yun; Li Zhanhuai; Wang Yong; Zhang Longbo, "Joining
“Application of Data Mining Techniques for Medical Image
associative classifier for medical images," Hybrid Intelligent Systems,
Classification”, in Proceedings of Second International Workshop on
2005. HIS '05
Multimedia Data Mining (MDM/KDD’2001) in conjunction with ACM
SIGKDD conference, SanFrancisco, USA, pp. 94-101, Aug 26, 2001. [20] Sampaio WB, Diniz EM, Silva AC, de Paiva AC, Gattass M.” Detection
of masses in mammogram images using CNN, geostatistical functions
[5] Aswini Kumar Mohanty, Sukanta Kumar Swain, Pratap Kumar
and SVM”, Journal of Computers in Biology and Medicine 2011; 41:
Champati, Saroj Kumar Lenka, “Image Mining for Mammogram
653-664
Classification by Association Rule Using Statistical and GLCM
features”, International Journal of Computer Science Issue, 2011. [21] Wang D, Shi L, Ann Heng P, “Automatic detection of breast cancers in
mammograms using structured support vector machines”,
[6] Vipul Sharma and Sukhwinder Singh “Automated Classification of Fatty
Neurocomputing2009; 72: 3296-3302.
and Dense Mammograms” Journal of Medical Imaging and Health
Informatics Vol. 5, 520–526, 2015. [22] Rejani YIA, Selvi ST., “Early Detection of Breast Cancer using SVM
Classifier Technique”, International Journal of Computer Science
[7] Hossein Yousefi, Banaem, Alireza ,Mehri Dehnavi,, “Ensemble
Engineering 2009; 1: 127-130.
Supervised Classification Method Using the Regions of Interest and
Grey Level Co-Occurrence Matrices Features for Mammograms Data” [23] M. Arfan Jaffar, “Dct Features based Malignancy and abnormality type
Iranian journal of radiology ISSN: 1735-1065 detection method for mammograms “International Journal of Innovative
Pages: e11656 Computing, Information and Control ICIC ISSN 1349-4198Volume 7,
Number 9, September 2011
[8] Leonardo de Oliveira Martins, Geraldo Braz Junior,'” Detection of
Masses in Digital Mammograms using K-means and Support Vector [24] Deepa Deshpande, Archana Rajuskar, “Texture based associative
Machine”, Electronic Letters on Computer Vision and Image Analysis classifier-An application of data mining for mammogram classification”,
8(2):39-50, 2009 Computational Intelligence in Data Mining 1,Springer
[9] Kirti and Jitendra Virmani,” Comparison of CAD systems for three class [25] Kanchanamani M, Varalakshmi Perumal, “Performance evaluation and
Breast tissue classification using mammographic images”, Medical comparative analysis of various machine learning techniques for
imaging in clinical applications, Springer International Publishers, diagnosis of breast cancer”, proceedings of biomedical research 2016.
Switzerland,2016 [26] http://irma-project.org/index_en.php
[10] Haralick. R. M. et al.,” Textural features for image classification”, IEEE [27] http://peipa.essex.ac.uk/info/mias.html
Trans Syst. Man. Cybern., Vol. SMC-3. Ppc 610-621, 1973.
[11] Poonam Sonar, Dipti Jadhav, Udhav Bhosle, “Optimized association
rules using objective function for mammography image
classification",2016 International Conference on Communication and
Signal Processing
[12] Vapnik VN. The nature of statistical learning theory. 2nd ed. NY (USA):
Springer-Verlag;, p. 131,2000.
[13] Baljit Singh Khehra, Amar Pratap Singh Pharwaha b,” Classification of
Clustered Microcalcifications using MLFFBP-ANN and SVM,’’
Egyptian Informatics Journal 17, 11–20,(2016).
311

Sample Technical Papers - Collection For Student Awarness

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sample Technical Papers - Collection For Student Awarness

Uploaded by

Copyright:

Available Formats

3rd International Conference on Renewable Energy Research and Applications Milwakuee, USA 19-22 Oct 2014

Determination of Iron Loss Considering Spatial

Vinay Jaiswal, and N. K. Deshmukh

ICRERA 2014 348

k wn = k pn k dn is winding factor for the nth harmonics

k pn = sin( nxφ / 2) (3)

xφ = π*no. of slots spanned by Coil Pitch/(Ns*2p)

Ns= No. of stator slots.

γ = π / no. of stator slots per pole

ICRERA 2014 349

Magnetizing reactance for the nth harmonics, wd = rotor slot opening

℘ag air gap permeance for the nth harmonics = 4 μ0 Rl (ns _ n − nr )

NR = no. of rotor slots.

hd = slot lip height n=n (14)

ICRERA 2014 350

ICRERA 2014 351

ICRERA 2014 352

ICRERA 2014 353

Plagiarism: Concepts, Factors and Solutions

Keywords: Plagiarism, Solutions, Concepts, Causes

Introduction tional demand for better results and more publica-

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

fair use of materials. important question;

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

www.hope.ac.uk/compass/. 35.Carroll J. A handbook for deterring plagiarism

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Ir J Military Medicine Vol. 14, No. 3, Autumn 2012

Long-Range Forecasting of Electricity Demand and

978-1-4799-2075-4/13/$31.00 ©2013 IEEE 5

Electricity supply for Maharashtra state is divided into two

978-1-4799-2075-4/13/$31.00 ©2013 IEEE 6

TABLE 3. ASSUMPTIONS USED IN LEAP [11, 12, 16]

978-1-4799-2075-4/13/$31.00 ©2013 IEEE 7

978-1-4799-2075-4/13/$31.00 ©2013 IEEE 8

Waste Sunflower Oil as an Alternative Fuel for Diesel

Abhishek Pavani, Akash Hebale, Veraj Poojary, Kapse Neeta

978-1-4799-7263-0/15/$31.00 ©2015 IEEE

noted. Based on this parameters like brake specific fuel

Table 2: Engine Specifications of 4 stroke CI engine

Rated engine speed 1500rpm

Stroke length 110mm

Bore diameter 87.5mm

Orifice diameter 25mm

Flow rate of cooling water 1.8lit/hr

Fig.2: Biodiesel Process flow diagram

The following steps were used for the titration process:

The quantity of KOH is then calculated and that exact

1 Engine 2 Base frame

III. RESULTS AND DISCUSSION C. Effect on Mechanical efficiency

A. Effect on Brake thermal efficiency

Fig. 7: Mechanical efficiency vs Load

In general, as the load increases the mechanical efficiency

Fig. 8: Hydrocarbon (HC) vs Load

B. Effect on carbon dioxide emission IV. CONCLUSIONS

C. Effect on carbon monoxide emission REFERENCES

[11] Shirneshan A, Almassi M, Ghobadian B, Borghei A.M, Najafi

xφ = πno. of slots spanned by Coil Pitch/(Ns2p)