Latency Optimization and Power Efciency with

Higher Transmission Data Rate in OFDMA based

Wireless Distributed Computing System
Tangudu Ramji
School of Electrical Sciences
IIT Bhubaneswar
Bhubaneswar, 751013, India
Email: rt10@iitbbs.ac.in
AbstractCurrent wireless distributed computing (WDC) system
has one of the major factors is latency. There has a demand
for power efciency, reliable in its communication system. These
demands under fading channel conditions gives stupendous
challenges to WDC system over traditional distributed com-
puting system. In this paper, the objective is to minimize the
latency. Where constraints are considered as power efciency
and accuracy in huge data rate transmitting and receiving.
I propose OFDMA based WDC system, to achieve the zero
interference. This paper uses an evolutionary method, called
the particle swarm optimization (PSO). This algorithm gives
global optimum solution to above specied objective function
with particularized constraints. The performance of proposed
system with enumerated algorithm is analyzed using computer
simulation.
KeywordsWireless Distributed Computing system, OFDMA, la-
tency.
I. INTRODUCTION
Most of the wireless communications like military and
multimedia contains enormous tasks. These tasks are not able
to accomplish by single input single output (SISO) system,
in such type of cases wireless distributed computing system
(WDC) is introduced. In WDC system like single input multi
output (SIMO) and multi input multi output (MIMO), infor-
mation can share between multiple systems to complete the
enormous tasks with in less amount of time. But WDC sys-
tem facing colossal interference by fading and noise channel
conditions.
Due to this disadvantage of WDC system, orthogonal
frequency division multiplexing (OFDM) scheme deploys into
WDC system to achieve the zero interference. OFDM comes
under the physical layer. OFDM based WDC system employs
to accomplish the prodigious task in less amount of signal to
noise ratio (SNR) without interference. But this scheme does
not contain adaptivity in its subcarriers allocation, that means
it consider the static channel conditions and static capabilities
of receivers only.
To avoid the above specied problem in OFDM based
WDC system, orthogonal frequency division multiple access
(OFDMA) scheme is introduced. OFDMA is the small exten-
sion of OFDM scheme to include multiple access. Depends
on the fading channel conditions, receivers capabilities in
OFDMA based WDC system, OFDM transmits the huge data
rate with the help of better optimization algorithm. That means
adaptivity present in OFDM. Hence this technique or scheme
called the OFDMA.
Due to these advantages of OFDMA, it is now a part of
wireless standards in next generation like long term evolu-
tion (LTE), worldwide interoperability for microwave access
(WiMAX) [6]. In OFDMA based WDC system, latency, power
efciency and higher data rate transmission with huge accuracy
are major factors. This paper deploys an evolutionary algorithm
in base node or foremost node to achieve the global optimum
solution for required objective function and with proper con-
straints to be considered.
Generally resource and subcarriers allocation for OFDMA
based wireless distributed computing (WDC) are main issues.
These two allocations efciently achieved by an evolutionary
algorithm, called the particle swarm optimization (PSO), pre-
sented in [1]. In wireless communications, nomadic devices
and mobile devices share resource with high reliablity over
the wireless grids [2]. In WDC system, power is a major
factor and it mostly depends on the switching frequency and
computing density. In [3], work load distributed algorithm
gives performance between power, network size and clock
rate. The paper focused on computing power, dynamic voltage
scaling (DVS) and adaptive power controlling (APC) giving
high energy efciency. Performance of power consumption
(communication and computing power) and distributed com-
puting power ratio (DCPR) are major issues in WDC system
[4].
OFDMA based wireless comunication systems which are
usually applied in digital video broadcasting environments or
wireless metropoliton area networks (WMAN). [7] evaluated
the performance of cooperative communication for OFDMA
based wireless communication systems. The performance of
M-PSK and M-QAM for OFDMA based wireless communi-
cation systems are very useful in GSM-EDGE, LTE, WiMAX
technologies [8]. With the help of PSO, resource allocation
for OFDM based wireless communication system presented in
[9].
A survey on recent research challenges in WDC are pre-
sented in [5] and also in [5], the authors suggest that cognitive
radio (CR) and software dened radio (SDR) would be ideal
candidates for implementing WDC under tactical conditions.
Performance and evaluation on latency, occupancy and band-
width in DSM multiprocessors studied in [10]. Optimization of
energy and latency most essential things in cooperative com-
munication of wireless sensor networks (WSN). Performance
of optimization in latency-energy consumption in cooperative
transmission in duty-cycled wireless sensor networks presented
in [11]. However in all the above works, no one has been
performed on both latency and power minimizations in WDC
system with effective optimization algorithm.
In this paper, I proffer minimization of latency with power
efciency and reliable for OFDMA based WDC system. This
paper is organized as follows: In section II, the problem state-
ment is presented. In section III, the problem is formulated. In
section IV, present the proposed system model. In section V, I
present the simulation results. Finally, the paper is concluded
by the section VI.
II. PROBLEM STATEMENT
In this paper, minimization of latency is considered as an
objective. Power saving, accuracy and transmitting the higher
data rate are considered as constraints. These are depends on
the channel variances between foremost node to all menial
nodes, computational capabilities of menial nodes and num-
ber of operations happens at all menial nodes. Total power
consumption and average of total power consumption must
be satisfy the power constraints. Minimization of distributed
computing power ratio (DCPR) is considered to be momentous
performance.
III. PROBLEM FORMULATION
As mentioned earlier, the objective of WDC network is
to achieve the required quality of service (QoS) with limited
resource, limited power and trivial amount of latency. In
order to do so, one needs to mathematically formulate the
required objective function and useful constraints. In this
section, I present mathematical equations that relate various
WDC parameters.
One of the major parameters in a WDC network is the
latency.
t
d
= a C
L
V
cc
/(V
cc
V
th
)

sec. (1)
t
com
= t
TX
+ t
RX
sec. (2)
t
td
= ((t
d
(f/2 + 1)) + (t
d
(f/2))) 2 2 sec. (3)
latency = (t
com
+ t
td
) N sec. (4)
where is the velocity saturation index and its typical range
from 1.30 to 1.50. t
TX
is the time taken for transmitting the
total data from foremost node to all menial nodes, t
RX
is the
time taken for receiving the total data from all menial nodes to
foremost node, t
com
indicates time taken for transreceiving
the data at foremost node, t
d
is the time taken for computing
the task at single menial node, t
td
indicates total time delay
for computing the task in above WDC network, f is the
number of cycles, N is the total number of menial nodes,
V
th
the threshold voltage for detection purpose and V
cc
is
the supply voltage .
One of the important resources in a WDC network is the
total computing power. It is dened as the total power required
for a menial node to perform a certain computational task.
In digital circuits computing power consumption is mainly
classied as static and dynamic power consumptions [3]. The
computing power consumed by the i
th
menial node is given
by [3].
P
(i)
co
= P
(i)
s
+ P
(i)
d
watts. (5)
where P
co
is the computing power, P
s
is the static power
and P
d
is the dynamic power.
Here static power is indicating that how much power is
leaking in processor. Dynamic power is about computing task.
It is expressing in below as [3], [4]:
P
(i)
d
= a C
L
V
2
cc
f watts. (6)
In this paper, above specied dynamic power consumption
is considerd in two locations. One is at the input side and other
one is at the output side. The dynamic power consumption
at the input side is called the transient dynamic power con-
sumption, at the ouput side is called the load dynamic power
consumption.
P
(i)
dtr
= (a f C
pd
V
2
cc
N
sw
) watts. (7)
P
(i)
dl
= (a f C
L
V
2
cc
N
sw
) watts. (8)
where P
dtr
is the computing power at the input side, P
dl
is
the computing power at the output side, P
s
is the static power
and P
d
is the dynamic power. a is the parameter relating to
computing density, C
pd
is the power dissipation capacitance,
C
L
is the loading capacitance, V
cc
is the supply voltage for
digital circuits. The parameter f indicates number of clock
cycles, N
sw
means number of bits changes are required, it
relates to the computational burden. Higher N
sw
means more
number of bits changes are required and hence capacitors are to
be charged and discharged. Therefore more energy is required.
The parameter a is also known as the processor parameter.
Higher a means, the processor has higher capacitance and
hence more energy is required to charge them. The total power
required for computation is [1]
P
co
=
K

i=1
P
(i)
co
watts. (9)
where K indicates total number of menial nodes.
Another important resource in WDC is total communi-
cation power consumed. As the name suggests, it is the
total power required for a foremost node to communicate the
computational task to the menial nodes. It depends on factors
like bandwidth, data rate required and channel variance.
The relationship between the above factors and communi-
cation power required is given by Shannons theorem
d = B log(1 + P
cm
/
2
) bits/sec. (10)
where B indicates the total bandwidth is available, d is
the maximum possible data rate, P
cm
is the communication
power required and
2
is noise power in the channel.
If the channel variance of the i
th
menial node is
2
i

then relationship between communication power required and
data rate achieved is given by
d
i
= W
i
log(1 + P
(i)
cm
/
2
i
) bits/sec. (11)
where P
(i)
cm
is the communication power required to achieve a
maximum data rate of d
i
for the i
th
menial node. Therefore
the total power utilized for communication will [1]
P
cm
=
K

i=1
P
(i)
cm
watts. (12)
where K indicates total number of menial nodes.
The total power consumed for communication and compu-
tation is [1]
P
tc
= P
cm
+ P
co
watts. (13)
IV. PROPOSED SYSTEM MODEL
The objective of this paper is to: Minimize the latency
in OFDMA based WDC network. The number of subcarriers
allocate to a particular menial node in a tactical manner, it
indicates the communication power and computation burden
allocated to each menial node based on the QoS requirements
and resource constraints. The QoS for common WDC network
are less power consumption, assurance in reliable communi-
cation, high transmission data rate and minimum latency.
In order to achieve the above objective with specied QoS,
I propose the following optimization problem.
Minimize:

Latency (objective function)

Subject to:

P
tc
P
tc0
(constraints)
E(P
(i)
cm
+ P
(i)
co
) P
0
E(r
i
) E(d
i
)
97 precision
percentage
(r
i
) 100 i (14)
where objective function is optimizing the latency of our
OFDMA based WDC network.
First constraint is total power consume of our WDC should
not exceed total power constraint P
tc0
. Second constraint
is an average of total power consume should not exceed
average power constraint P
0
. In the third constraint, r
i
is the
transmission rate for the i
th
link.
The third constraint indicates that the transmission rate
must be less than the maximum data rate that can be achieved,
so then Shannons bound is satised. In fourth constraint,
precision
percentage
(r
i
) reects the reliability of the link i.
The reliability must be high, so that there is no data loss
and requirement for retransmission. Reliability is computed by
calculating the bit error rate (BER) at the receiver. The block
diagram of the proposed system is shown in Fig. 1. In the
proposed system, initially foremost node allocated to all menial
nodes with equal number of subcarriers. Here two control
blocks are used. These control blocks estimate the parameters
relate to latency from every menial node, computing power
from every menial node, channel characteristics and menial
Subcarriers
allocation
OFDM
transmitter
User1
User 2
User K
Optimization Algorithm
Control
channel
X(0)
X(1)
X(N-1)
Foremost Node
Latency
and
Computing
power
Control
Fig. 1. Block diagram of the Proposed system model
nodes capabilities. The foremost node collects the data from
control blocks, after then minimize the latency, adaptively
allocates the number of subcarriers to each menial node.
Hence computational burden and communication power are
adaptively allocate in each link by solving the optimization
problem (14). The optimization problem can be solved by
using any non linear optimization algorithm. In this paper, PSO
is used as non linear optimization algorithm and it gives better
global minimum solution [10].
V. SIMULATION RESULTS
In this section I analyze the performance of proposed
method using MATLAB simulation. latency, total power saving
and DCPR are considered as performance metrics.
For the simulation, I consider one foremost node and
number of menial nodes are varied from 20 to 200. The channel
between the foremost node and the particular menial node
is considered to be additive white Gaussian noise (AWGN),
whose SNR is taken from 20 to 35 in an uniform and random
manner. The total network power is considered as 20 watts.
The distance between the foremost node and the particular
menial node is taken from 2 to 20 k.m. in random manner.
The average power constraint on each menial node is taken
as 1 watt. The static power consumption, for the particular
menial node is taken from 0.01 to 0.10 watt in an uniform and
random manner. The maximum number of subcarriers in one
OFDMA block is taken as 1024.
Here Monte Carlo simulation is used, for this simulation
transmission bits are taken as 10
6
bits and its mapping scheme
as binary phase shift keying (BPSK). Maximum value of
computing density, switching frequency are denoted as a
M
,
N
swM
. In below simulations I study the effect of network
size, computing density and switching frequency on latency,
total power saved and DCPR.
In Fig. 2, I study the effect of number of nodes, xed
N
swM
(10
6
) and variation in computing density on the latency.
Here at each network size computing density varies from 0.15
to 1.05. When the network size increases, latency increases
gradually.
20 40 60 80 100 120 140 160 180 200
0
1
2
3
4
5
6
7
8
9
network size
la
t
e
n
c
y

(
s
e
c
o
n
d
s
)
latency versus network size


without channel variance
with channel variance
Fig. 2. Effect of network size, variation in computing density and xed
N
swM
(10
6
) on latency: with channel variance and without channel variance
conditions
20 40 60 80 100 120 140 160 180 200
0
2
4
6
8
10
12
network size
la
t
e
n
c
y

(
s
e
c
o
n
d
s
)
latency versus network size


without channel variance
with channel variance
Fig. 3. Effect of network size, variation in switching frequency and xed
a
M
(1.50) on latency: with channel variance and without channel variance
conditions
In Fig. 3, I study the effect of number of nodes, xed a
M
(1.50) and variation in switching frequency on the latency.
Here at each network size switching frequency varies from
10
5
to 10
6
. When the network size increases, latency increases
gradually.
From Figs. 2 and 3, latency in with channel variance
condition is less than in without channel variance condition at
every network size. Why because, when the channel variance
presents, more supply voltage is used. So, it decreses the time
delay factor and it reects to the latency (1). Compared to
in Fig. 2, in Fig. 3 at every network size latency is more.
Hence compared to switching frequency, computing density
gives more effect on latency.
In Fig. 4, I study the effect of number of menial nodes
on total power saving ratio or power efciency. I also study
the effect of channel variance and computing density on the
performance (power saving ratio). Here switching frequency
is xed N
swM
(10
6
). It can be seen that when network size
increases, power efciency or power saving ratio gradually
goes to negative value. After network size is 100, necessity of
power supply is increases. Therefore I exceed the total power
constraint of 20 watts. This is because, increase in computing
density means increase in the number of capacitors or load in
receivers.
In Fig. 5, the performance slightly differ from Fig. 4, that
I vary the switching frequency and xed a
M
(1.50). It can
be seen that I get positive power saving upto network size is
90 only. However power efciency decreases when switching
frequency increases. This is because, increase in switching
frequency means increase in the computational burden.
20 40 60 80 100 120 140 160 180 200
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
network size
p
o
w
e
r

s
a
v
in
g

r
a
t
io

(
n
o

u
n
it
s
)
power saving ratio versus network size


without channel variance
with channel variance
Fig. 4. Effect of network size, variation in computing density and xed
N
swM
(10
6
) on power saving ratio: with channel variance and without
channel variance conditions
20 40 60 80 100 120 140 160 180 200
1.5
1
0.5
0
0.5
1
network size
p
o
w
e
r

s
a
v
i
n
g

r
a
t
i
o

(
n
o

u
n
i
t
s
)
power saving ratio versus network size


without channel variance
with channel variance
Fig. 5. Effect of network size, variation in switching frequency and xed
a
M
(1.50) on power saving ratio: with channel variance and without channel
variance conditions
From Figs. 4 and 5, I observed that, for each network
size the total power consumptions of with channel variance
condition and without channel variance conditions are more
pronounced in Fig. 5 as compared to the Fig. 4. Hence
computing density has more effect on total power consumption
and power efciency.
20 40 60 80 100 120 140 160 180 200
0.16
0.17
0.18
0.19
0.2
0.21
0.22
0.23
0.24
0.25
network size
e
x
p
e
c
t
a
t
io
n

o
f

t
o
t
a
l
p
o
w
e
r

c
o
n
s
u
m
e

(
w
a
t
t
s
)
expectation of total power consume versus network size


without channel variance
with channel variance
Fig. 6. Effect of network size, variation in computing density and xed
N
swM
(10
6
) on expectation of total power consumption: with channel
variance and without channel variance condition
In Figs. 6 and 7, I study the effect on average power
consumed when number of menial nodes are increased. Here
I vary the both switching frequency and computing density. It
can be seen that the average power consumed almost constant
when number of menial nodes are increased. I observed that,
for every network size the average of total power consump-
tions of with channel variance condition and without channel
20 40 60 80 100 120 140 160 180 200
0.16
0.18
0.2
0.22
0.24
0.26
0.28
0.3
network size
e
x
p
e
c
t
a
t
io
n

o
f

t
o
t
a
l
p
o
w
e
r

c
o
n
s
u
m
e

(
w
a
t
t
s
)
expectation of total power consume versus network size


without channel variance
with channel variance
Fig. 7. Effect of network size, variation in switching frequency and xed
a
M
(1.50) on expectation of total power consumption: with channel variance
and without channel variance conditions
variance conditions is more pronounced in Fig. 7 as compared
to the Fig. 6. Hence computing density has more effect on the
expectation of total power consumption.
From Figs. 8 and 9, I study the effect of number of menial
nodes, computing density and switching frequency on DCPR.
Here DCPR is almost constant as I vary computing density
and switching frequency at every network size. In Fig. 8, at
all network sizes, switching frequency is xed with maximum
value as 10
6
. In Fig. 9, at all network sizes, computing density
is xed with maximum value as 1.50.
The DCPR is dened by the ratio of dynamic network
computing power consumption to the available network power
to complete this task [5].
DCPR =
P
d
(P
cm
+ P
d
+ P
s
)
(15)
20 40 60 80 100 120 140 160 180 200
0.6
0.62
0.64
0.66
0.68
0.7
0.72
0.74
0.76
0.78
network size
D
C
P
R
DCPR


without channel variance
with channel variance
Fig. 8. Effect of network size, variation in computing density and xed
N
swM
(10
6
) on DCPR: with channel variance and without channel variance
conditions
From Figs. 8 and 9, I observed that the DCPR of with chan-
nel variance condition and without channel variance conditions
are more pronounced in Fig. 9 . Hence compared to computing
density, switching frequency has less effect on DCPR.
VI. CONCLUSION
In this paper, I proposed a new scheme for OFDMA
based WDC system. The proposed scheme minimizes the
latency, DCPR in an immense amount. It offers better resource
and power allocation for the above specied system. The
performance of the proposed scheme is studied using computer
simulations. The simulation results show that DCPR is almost
20 40 60 80 100 120 140 160 180 200
0.64
0.66
0.68
0.7
0.72
0.74
0.76
0.78
network size
D
C
P
R
DCPR


without channel variance
with channel variance
Fig. 9. Effect of network size, variation in switching frequency and xed
a
M
(1.50) on DCPR: with channel variance and without channel variance
conditions
constant, when the number of nodes is increased. Finally
I conclude that, compared to computing density, switching
frequency gives less effect on latency, DCPR and power
consumption.
REFERENCES
[1] T. Ramji, Barathram. R, and M. Sabarimalai Manikandan, Resource and
subcarriers allocation for OFDMA based WDC system, 4th IEEE IACC,
pp. 338-342, Feb. 2014.
[2] L. W. McKnight, J. Howison, and S. Bradner, Wireless grids-distributed
resource sharing by mobile, nomadic, and x devices, IEEE Internet
Comput., vol. 8, no. 4, pp. 24-31, July 2004.
[3] X. Chen, T. R. Newman, D. Dinesh, T. Bose, and J. H. Reed, The impact
of channel variances on wireless distributed computing networks, IEEE
Globecom, Honolulu, HI, USA Dec. 2009.
[4] X. Chen, S. M. Hasan, T. Bose, and J. H. Reed, Cross-layer resource
allocation for wireless distributed computing networks, IEEE Radio and
Wireless Symp., pp. 605-608, Jan. 2010.
[5] D. Datla, X. Chen, T. Tsou, S. Raghunandan, S. M. Shajedul Hasan, J.
H. Reed, C. B. Dietrich, and T. Bose, Wireless distributed computing:
A survey of research challenges, IEEE, Communications Magazine, vol.
50, no.1, pp. 144-152, Jan. 2012 .
[6] H. Marcelino, Z. Hua-shen, and G. Yanbin, Performance analysis of
OFDMA system in next generation wireless communication networks,
ICCSIT 3rd IEEE International Conference, vol. 8, PP. 335-339, IEEE,
2010.
[7] J. S. Yoon, H. M. Lim, H. J. Park, and H. K. Song, Performance
evaluation of co-operative communication for OFDMA based wireless
communication systems, IEEE ICCT, pp. 780-783, 2010.
[8] A. Mraz, and L. Pap, General performance analysis of M-PSK and
M-QAM wireless communications applied to OFDMA interference, in
Wireless Telecommunications Symposium (WTS), pp. 1-7, 2010.
[9] S. C. K. Lye, H. T. Yew, B. L. Chua, R. K. Y. Chin, and K. T. K.
Teo, Paricle swarm optimization based resource allocation in orthogonal
frequency division multiplexing, in 7th Asia Model Symposium (AMS),
pp. 303-308, IEEE, 2013.
[10] M. Chaudhuri, M. Heinrich, C. Holt, J. Pal Singh, E. Rothberg, and J.
Hennessy, Latency, occupancy and bandwidth in DSM multiprocessors:
A performance evaluation, IEEE Trans. on comput., Vol. 52, no. 7, pp.
862-880, IEEE, July 2003.
[11] Y. Wang, and S. Guo, Optimized energy-latency cooperative trans-
mission in duty-cycled wireless sensor networks, in proc. of IEEE
International conference on Mechatronics and Automation, IEEE, pp no.
185-190, Aug. 2013.