Rahul Amin, Dr. Jim Martin

qualtaghblurtingΚινητά – Ασύρματες Τεχνολογίες

12 Δεκ 2013 (πριν από 3 χρόνια και 10 μήνες)

93 εμφανίσεις


Rahul Amin,
Dr. Jim Martin

Clemson University, Clemson SC

Contact:
jim.martin@cs.clemson.edu

http://www.cs.clemson.edu/~jmarty



Dr. Ahmed
Eltawil
,
Amr

Hussien

University of California, Irvine CA

This project was sponsored in part by NSF through
contract
ECCS
-
0948132


Introduction


Problem Statement


Background


System Description


Simulation Methodology


Results and Discussion


Conclusions



Next Generation Wireless Networks (
HetNets
)


Made up of several Radio Access Technologies (RATs) such as 2G/3G/4G

and Wi
-
Fi


User devices are reconfigurable (or multi
-
modal) and support a multitude of RATs


Joint allocation of network
-
wide resources in this
hetnet

environment is shown to be
more efficient than ‘independent’ resource allocation by each RAT


Frameworks to support network
-
wide resource allocation process have been defined by
3GPP (CRRM, MRRM, JRRM) and IEEE (802.21, P1900.4) working groups



Our Previous Work


Studied the spectral efficiency vs. energy consumption tradeoffs while using
reconfigurable devices in a
hetnet

system


For a ‘balanced’ network topology, Random Waypoint mobility model, and an FPGA
-
based reconfigurable device, we showed an increase in spectral efficiency of 75% at the
cost of twice (2x) the energy consumption



The network topology, mobility model and hardware assumptions had a
significant impact on ‘moderate’ improvements (75%) in spectral efficiency
that were shown in our previous work



Analyze spectral efficiency vs. energy consumption
tradeoffs resulting from realistic
hetnet

assumptions
that show a ‘significant’ improvement in spectral
efficiency



Explore an ‘unbalanced’ network topology where resources of
one
c
ellular operator exceed those of another operator



Implement a clustered node movement pattern that changes a
highly unfavorable situation to a favorable situation when
network
-
wide resources are jointly allocated



Study the differences in energy consumption for several
reconfigurable radio hardware assumptions:


(
i
) ASIC
-
based radio


(ii) FPGA
-
based radio


(iii) Combination of ASIC and FPGA based radio




Towards widespread adoption of
HetNet

system concept


Need: Due to proliferation of smartphones, user data demand is
outpacing operator capacity


Steps taken: (
i
) ‘Wi
-
Fi offloading’ problem is being rigorously
studied by cellular operators where cellular systems interoperate
with 802.11 Wi
-
Fi networks
(ii) ‘
Femto
-
cell’ is utilized in practice today to increase spectral
efficiency by supplementing the macro
-
cell with an overlay of
smaller, co
-
operative networks




Energy
-
efficient reconfigurable device architectures (
MPSoC
)
are being investigated


Based on various hardware components such as ASICs, FPGAs and DSPs


Nodes (user devices) either have (ASIC
-
based) static radios
capable of operating in limited number of connectivity modes or
they have (ASIC+FPGA or FPGA
-
based) reconfigurable radios
capable of operating in any connectivity mode


Nodes can connect to
one or more Autonomous Wireless
Systems (AWS
) simultaneously


Each AWS has a controller that represents all nodes in the AWS
and that serves as gateway connecting the AWS with other
AWSs or external networks


A Global Resource Controller (GRC) implements a centralized
scheduler that
maps users to access technologies


Wireless Virtual Link Layer multiplexes/de
-
multiplexes data for
users associated to multiple AWSs


Vertical Handovers are initiated by the GRC


Reconfiguration handoff
-

Radio reconfigures itself to operate over a
different AWS


We formulate use cases that assume presence of 2
major cellular carriers in a given area


Use Case 1


No co
-
operation between the two carriers


Users use multiple static radios that can connect to its own
carrier

s access technologies


Use Case 2


Co
-
operation exists between the two carriers


Reconfigurable radios are used to support access
technologies implemented by the other carrier



Select trade
-
offs between two conflicting objectives


Maximize overall system throughput (Max sum
-
rate)


Maximize fairness amongst users (Max
-
Min Fair)



Comes up with user to access technology mappings
every 1 second



Scheduler properties


Wi
-
Fi resources are evenly distributed


Cellular resources are distributed using a two
-
step approach:


Allocate minimum required throughput


to each user using its best
cellular radios


Allocate unused resources to a window of


users with best connectivity
parameters in increments of



Any user assigned total throughput of 1 Mbps is not assigned any further
cellular resources



The parameters

,


and


help tune the fairness and overall
system throughput characteristics obtained by the scheduler


For the results presented in this study, we use (α,β,
ω
) = (100k,
100k, 10)


2 * 2 km
2

grid


Carrier 1


bad coverage area


2 EVDO
(3G
) base
-
stations at the edge of the grid


Carrier 2


good coverage area


1 HSPA
(3G
)

and 1 LTE
(4G
)

base
-
station at the center of the grid and 6 IEEE
802.11g (Wi
-
Fi
)
APs spread throughput the topology


Each
technology supports adaptive Modulation and Coding Scheme (MCS
)


MCS mapping for each user is determined based on distance of the user from the
Base
-
Station


100
nomadic users (50 subscribed to Carrier 1 and 50 to Carrier 2)


Carrier 1 users grouped in 2 clusters; one cluster located at left edge of grid and the other at
the right edge of the grid


Carrier 2 users are grouped in 1 cluster located at the center of the grid


Movement of each user follows Random Waypoint Model and is restricted to an area of 500 *
500 m
2

from its initial starting position


Experimental Parameters


Network
Outage: Varied between 0
-
25% in increments of 5%


Impact of
Reconfiguration: Increased Energy Consumption and Communications Downtime
multiplied by a scalar


[0,1]


Simulation run for 10,000 seconds


Carrier 2:
WiFi

Carrier 1:
WiFi

Carrier 1:
WiFi

Carrier 1:
WiFi

Carrier 2:
WiFi

Carrier 2:
WiFi

Carrier 1:
WiMAX

Carrier 2:
LTE

Carrier 2:
HSPA

Carrier 1:
EVDO


Worst Case: For no network outage
and impact of reconfiguration equal to
1, the
spectral efficiency gain for use
case 2 (
1.43
bits/sec/Hz) when
compared to use case 1
(0.34
bits/sec/Hz) is
314.30
%.


Best Case: For 25% network outage
and no impact of reconfiguration, the
spectral
efficiency gain for use case 2
(2.73
bits/sec/Hz) when compared to
use case 1
(1.79
bits/sec/Hz) is
553.70%.


Due to an unbalanced network
topology and clustered movement
pattern, the spectral efficiency
increase which is in the range [314.3%,
553.7] is ‘significantly’ higher
compared to [14.3%, 75%] obtained
for our previous study


314.3%

553.7%


Energy Consumption Model:










where:



P
dyn,FPGA

represents run
-
time energy consumption of FPGA
-
based hardware


P
dyn,ASIC

represents run
-
time energy consumption of
ASIC
-
based
hardware



P
rec,FPGA

represents energy consumption of FPGA
-
based hardware during reconfiguration


P
rec,ASIC

represents energy consumption of
ASIC
-
based hardware
when switching from
‘off’



to an ‘on’ mode




run

represents percentage of time system operates in regular mode




rec


represents percentage of time system operates in reconfiguration mode



β percentage of hardware built using FGPA components



1


β percentage of hardware built using ASIC components



λ

impact of reconfiguration



The ratio of
P
dyn,
FPGA
:
P
dyn,ASIC


is 12:1


P
rec,
FPGA

and
P
rec,
ASIC

values are the same

P
t
o
t
a
l


r
u
n

.
P
d
y
n
,
F
P
G
A

1




.
P
d
y
n
,
A
S
I
C






r
e
c

.
P
r
e
c
,
F
P
G
A

1




.
P
r
e
c
,
A
S
I
C





Hardware Setting 1: Use Case 1
(Made
up of completely ASIC
components, i.e.
β
=
0) vs. Use Case 2
(
Made up of completely FPGA
components, i.e.
β
= 1
)



For increase in spectral efficiency of
314.30% shown earlier, for these
hardware assumptions, the increase in
energy consumption is 104.90%



For increase in spectral efficiency of
553.70% also shown earlier, for these
hardware assumptions, the increase in
energy consumption is 614.9%



Most pessimistic hardware
implementation for a reconfigurable
radio in practice. Almost linear
tradeoff for worst case increase in
energy consumption

104.9%

614.9%


Hardware Setting 2: Use Case 1
(Made
up of completely ASIC
components, i.e.
β
=
0) vs. Use Case 2
(
Made up of
50% ASIC, 50% FPGA
components, i.e.
β
=
0.5 )



For increase in spectral efficiency of
314.30% shown earlier, for these
hardware assumptions, the increase in
energy consumption is 70.0%



For increase in spectral efficiency of
553.70% also shown earlier, for these
hardware assumptions, the increase in
energy consumption is 355.40%



Most likely hardware implementation
for a reconfigurable radio in practice.
Increase in spectral efficiency is
greater than increase in energy
consumption

70.0%

355.4%


Hardware Setting 3: Use Case 1
(Made
up of completely ASIC
components, i.e.
β
=
0) vs. Use Case 2
(
Made up of
completely ASIC
components, i.e.
β
=
0 )



For increase in spectral efficiency of
314.30% shown earlier, for these
hardware assumptions, the increase in
energy consumption is 35.10%



For increase in spectral efficiency of
553.70% also shown earlier, for these
hardware assumptions, the increase in
energy consumption is 98.80%



Hardware
setting 3 is
not really possible
in practice. But gives an
estimate of
increase in energy consumption if the
only difference between two use cases is
the number of ‘reconfiguration handoffs’
experienced by
devices

35.1%

98.8
%


The gains in spectral efficiency are
much greater (increase from
75.5% to
553.7%)

for an unbalanced network
topology and clustered node
movement pattern when
compared to a balanced network
topology and
random waypoint movement pattern studied in our previous work


Based on the hardware choices, the increase in energy consumption can
range from
98.80%
to
614.90%
for the corresponding increase in spectral
efficiency of 553.7
%


Depending on the number of possible modalities supported
by user
devices, it might be possible to attain a tradeoff in terms of lower energy
consuming ASIC radios at the cost of decreased reconfigurable options


In the worst case, our results show a more or less linear trade
-
off
between spectral efficiency and power
consumption


This result is an artifact of our workload assumptions that assume traffic
flows are always backlogged. In future work, we will explore more
realistic scenarios that involve on/off traffic
flows